Disclosure of Invention
The invention provides a power adjustment method and a power adjustment system for parallel modules of flexible equipment, which solve the problem that the existing control algorithm is difficult to accurately adjust the power of each parallel module in the flexible equipment.
In order to solve the above technical problems, a first aspect of the present invention provides a power adjustment method for a parallel module of a flexible device, including:
Acquiring real-time power data of each parallel module in the flexible equipment, and analyzing and processing the real-time power data to obtain power regulation target values and power curves of each parallel module;
Evaluating the power state of each parallel module based on each power curve to determine a first parallel module, controlling the first parallel module to generate a power adjustment instruction and sending the power adjustment instruction to a second parallel module adjacent to the first parallel module, so that each second parallel module performs primary power adjustment according to the power adjustment instruction and a power adjustment target value corresponding to each second parallel module to obtain a power adjustment quantity;
acquiring historical power data of each parallel module, and constructing a power trend prediction model of each parallel module according to the historical power data so as to obtain a power prediction sequence according to the power trend prediction model;
And correcting the power adjustment quantity of each parallel module through the power prediction sequence of each parallel module to obtain a power adjustment correction quantity, and controlling each parallel module to carry out secondary power adjustment based on the power adjustment correction quantity of the parallel module so as to realize the power adjustment of each parallel module in the flexible equipment.
As one preferable solution, the analyzing the real-time power data to obtain a power adjustment target value and a power curve of each parallel module includes:
Performing noise reduction and filtering processing on the real-time power data through wavelet transformation to obtain a first power data set, and performing data standardization processing on the first power data set according to a Kalman filter to obtain a second power data set;
and inputting the second power data set into a feedforward neural network to perform state calculation to obtain the running state of each parallel module, comparing the running state with a preset state interval, and performing power mutual feedback numerical calculation on the corresponding parallel module according to a comparison result to obtain the power regulation target value of each parallel module.
As one preferable solution, the calculating the power mutual feedback value of the parallel modules according to the comparison result to obtain the power adjustment target value of each parallel module includes:
Taking the parallel modules with the running states exceeding the preset state interval as overrun running parallel modules, and taking the rest parallel modules as normal running parallel modules;
Determining the power value to be regulated of each overrun parallel module, and sequencing the priority of each overrun parallel module according to the relation between the power value to be regulated and a preset regulation power interval;
and selecting an adjustment target corresponding to each overrun parallel module from each normal operation parallel module based on a priority order according to the distance between each normal operation parallel module and each overrun parallel module so as to perform power mutual feedback numerical calculation, and obtaining a power adjustment target value of each parallel module.
As one preferable solution, the analyzing the real-time power data to obtain a power adjustment target value and a power curve of each parallel module further includes:
Noise filtering is carried out on the real-time power data through a band elimination filter to obtain a third power data set, and weighting processing is carried out on the third power data set through a hanning window function to obtain a fourth power data set;
Correcting fluctuation points in the fourth power data set according to the self-adaptive Kalman filter, and fitting the corrected fourth power data set by adopting a feedforward neural network to obtain a primary power curve of each parallel module;
Extracting key features in the primary power curves, calculating fundamental wave components and harmonic components of the key features through Fourier transformation, comparing the fundamental wave components and the harmonic components with a preset harmonic distortion rate threshold, marking a curve section exceeding the preset harmonic distortion rate threshold as a switching distortion section, and removing the switching distortion section to obtain power curves of the parallel modules, wherein the key features are used for representing high-frequency power fluctuation components of the corresponding parallel modules under non-ideal switching conditions.
As one preferable mode, the evaluating the power state of each parallel module based on each power curve to determine a first parallel module, controlling the first parallel module to generate a power adjustment command and sending the power adjustment command to a second parallel module adjacent to the first parallel module, and the method includes:
Performing four-layer decomposition on each power curve by adopting discrete wavelet transformation, and reconstructing a decomposition result by wavelet coefficients to obtain a characteristic data matrix so as to extract a power characteristic vector sequence from the characteristic data matrix;
Inputting the power characteristic vector sequence into a mapping relation model established by a radial basis function support vector regressor for processing to obtain a current power state of a corresponding parallel module, and calculating the Euclidean distance between the current power state and a preset average division threshold value to obtain a power state evaluation value of the corresponding parallel module;
And taking the parallel module with the power state evaluation value exceeding a preset performance index as a first parallel module, controlling the first parallel module to generate a power adjustment instruction, and sending the power adjustment instruction to a second parallel module adjacent to the first parallel module.
As one preferable solution, the foregoing method for enabling each second parallel module to perform primary power adjustment according to the power adjustment command and the corresponding power adjustment target value, to obtain a power adjustment amount includes:
according to the power adjustment instruction, each second parallel module carries out quantization constraint on a power adjustment target value corresponding to the second parallel module by using a gradient algorithm and minimizing a power climbing rate variance as an objective function, and generates a power adjustment sequence comprising an adjustment target value, adjustment time and adjustment direction;
Calculating a climbing speed value of each second parallel module by a Gaussian elimination method based on a power adjustment sequence of each second parallel module, performing amplitude limiting treatment on the climbing speed value exceeding a preset climbing speed, and distributing power adjustment increment of the corresponding second parallel module according to a preset attenuation proportion;
and taking the power adjustment increment and the corresponding climbing rate value as a state quantity and a control quantity respectively, and calculating by a Lagrangian multiplier method to obtain the power adjustment quantity of each second parallel module.
As one preferable solution, the calculating the power adjustment amount of each second parallel module by the lagrangian multiplier method includes:
Taking the power adjustment rate constraint and the load power constraint of each second parallel module as constraint conditions, and taking the power climbing rate of each parallel module in the flexible equipment as a target model;
And solving the target model through a Lagrangian multiplier method, updating the climbing speed value according to a steepest descent method when iteration stop criteria are not met, and iteratively solving the target model through the Lagrangian multiplier method based on the updated climbing speed value until the iteration stop criteria are met, and outputting power adjustment quantity of each parallel module in the flexible equipment when iteration turns are stopped, wherein the iteration stop criteria are that the change rate of power adjustment increment with adjacent turns is smaller than a preset power adjustment increment change threshold.
As one preferable solution, the obtaining historical power data of each parallel module, and constructing a power trend prediction model of each parallel module according to the historical power data, so as to obtain a power prediction sequence according to the power trend prediction model, includes:
Collecting historical power data of each parallel module through a rolling window, and carrying out normalization processing on the historical power data in the rolling window by taking a rated power value as a reference, so as to train a long-short-time memory neural network as sample data, and obtain a power trend prediction model corresponding to the parallel module to generate a power trend prediction sequence;
Extracting oscillation characteristics of the historical power data based on fast Fourier transformation, and weighting a power trend prediction sequence of a parallel module to which the historical power data belong according to the inverse of an oscillation period to obtain a power prediction compensation sequence;
and correcting the power prediction compensation sequence according to a Kalman filter to obtain the power prediction sequence of each parallel module.
As one preferable solution, the correcting the power adjustment amount of each parallel module through the power prediction sequence of each parallel module to obtain a power adjustment correction amount and controlling each parallel module to perform secondary power adjustment based on the power adjustment correction amount thereof includes:
Collecting real-time power data, real-time temperature data, real-time voltage data and real-time current data of each parallel module in the flexible equipment after secondary power adjustment, taking the real-time power data, the real-time voltage data and the real-time current data as training samples to train the deep neural network, and outputting the change rate of load power;
triggering parameter correction of a controller in the flexible equipment to obtain a parameter correction value when the load power change rate exceeds a preset power change rate threshold;
based on the parameter correction value, optimizing the power adjustment correction quantity of each parallel module by adopting a genetic algorithm to obtain a power adjustment correction optimization sequence, and taking the ratio of the load power change rate of each parallel module to the rated power of the parallel module as a compensation coefficient;
And carrying out weighting treatment on the power adjustment correction optimization sequence through the compensation coefficient to obtain a power adjustment compensation quantity, and controlling each corresponding parallel module to carry out three times of power adjustment based on the self power adjustment compensation quantity so as to realize power adjustment of each parallel module in the flexible equipment.
A second aspect of the present invention provides a power conditioning system for a parallel module of flexible devices, comprising:
the data processing module is used for acquiring real-time power data of each parallel module in the flexible equipment, and analyzing and processing the real-time power data to obtain power regulation target values and power curves of the parallel modules;
The primary adjusting module is used for evaluating the power states of the parallel modules based on the power curves to determine a first parallel module, controlling the first parallel module to generate a power adjusting instruction and sending the power adjusting instruction to a second parallel module adjacent to the first parallel module, so that the second parallel modules perform primary power adjustment according to the power adjusting instruction and the corresponding power adjusting target values to obtain power adjustment amounts;
The power prediction module is used for acquiring historical power data of each parallel module, and constructing a power trend prediction model of each parallel module according to the historical power data so as to obtain a power prediction sequence according to the power trend prediction model;
And the secondary adjusting module is used for correcting the power adjusting quantity of each parallel module through the power prediction sequence of each parallel module to obtain a power adjusting correction quantity and controlling each parallel module to carry out secondary power adjustment based on the power adjusting correction quantity of the parallel module so as to realize the power adjustment of each parallel module in the flexible equipment.
Compared with the prior art, the embodiment of the invention has the beneficial effects that at least one of the following points is adopted:
(1) The working state of each parallel module can be quickly known by acquiring the power data of each parallel module in the flexible equipment in real time; by sending power adjustment instructions to adjacent parallel modules so as to enable the adjacent parallel modules to combine the corresponding power adjustment target values to carry out power adjustment, the accuracy and timeliness of adjustment can be ensured, and information sharing and collaborative adjustment among the parallel modules are realized;
(2) According to historical power data of each parallel module, a power regulation strategy based on predictive control is introduced to predict future power variation trend of each parallel module, power regulation quantity is corrected in advance, oscillation in the power climbing process is restrained, and accurate power dynamic regulation of each parallel module in flexible equipment is realized;
(3) The invention integrates a plurality of intelligent links such as data acquisition, analysis processing, prediction optimization and the like, and realizes the accurate control and optimization adjustment of the parallel modules of the flexible equipment through an intelligent control strategy so as to reduce manual intervention and improve the production efficiency and the automation level.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings and examples, in which it is evident that the embodiments described are only some, but not all embodiments of the invention, and these examples are provided for a more thorough and complete disclosure of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the description of the present application, the terms "first," "second," "third," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first", "a second", "a third", etc. may explicitly or implicitly include one or more such feature. In the description of the present application, unless otherwise indicated, the meaning of "a plurality" is two or more.
In the description of the present application, unless explicitly stated or limited otherwise, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected, mechanically connected, electrically connected, directly connected, indirectly connected via an intervening medium, or in communication between two elements. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and the like are used herein for descriptive purposes only and not to indicate or imply that the system or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore should not be construed as limiting the application. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items. The specific meaning of the above terms in the present application will be understood in specific cases by those of ordinary skill in the art.
In the description of the present application, it should be noted that all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs unless defined otherwise. The terminology used in the description of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application, as the particular meaning of the terms described above in the present application will be understood to those of ordinary skill in the art in the detailed description of the application.
In an embodiment, as shown in fig. 1, a first aspect of the present invention provides a power adjustment method for a parallel module of a flexible device, including:
s1, acquiring real-time power data of each parallel module in flexible equipment, and analyzing and processing the real-time power data to obtain power regulation target values and power curves of the parallel modules;
the invention particularly adopts a voltage transformer and a current transformer to acquire real-time data from an embedded parallel module in flexible equipment at a fixed data acquisition frequency, wherein the sampling frequency is set to acquire 100 data points per second, the acquired real-time power data comprises a voltage fundamental wave value, a current fundamental wave value and a power phase value, and the acquisition is carried out aiming at the numerical values of the voltage data range between plus and minus 380V, the current data range between plus and minus 200 amperes and the phase angle range between plus and minus 180 degrees, so that the sampling distortion caused by data fluctuation is avoided.
In an embodiment, the analyzing the real-time power data to obtain a power adjustment target value and a power curve of each parallel module includes:
Performing noise reduction and filtering processing on the real-time power data through wavelet transformation to obtain a first power data set, and performing data standardization processing on the first power data set according to a Kalman filter to obtain a second power data set;
and inputting the second power data set into a feedforward neural network to perform state calculation to obtain the running state of each parallel module, comparing the running state with a preset state interval, and performing power mutual feedback numerical calculation on the corresponding parallel module according to a comparison result to obtain the power regulation target value of each parallel module.
In the wavelet transformation filtering noise reduction processing process, db4 wavelet basis function is selected to carry out four-layer decomposition on the acquired voltage fundamental wave value, current fundamental wave value and power phase value, high-frequency noise and random disturbance are removed, main frequency components are reserved, a first power data set obtained after filtering has good data continuity, data standardization is carried out through a Kalman filter, abnormal mutation points are corrected, and a stable and reliable second power data set is obtained;
Then, according to the voltage fundamental wave value, the current fundamental wave value and the power phase value contained in the second power data set, carrying out numerical calculation on the current running state of each parallel module by combining a feedforward neural network, acquiring a preset power threshold interval, a voltage threshold interval and a current threshold interval which are obtained based on historical data statistics from a controller of flexible equipment as reference, comparing each data acquisition period with the calculated running state, and carrying out power mutual feedback numerical calculation on the corresponding parallel module according to a comparison result to obtain a power regulation target value of each parallel module;
When the feedforward neural network calculates the running state, the input layer comprises three neurons of a power value, a voltage value and a current value, the hidden layer adopts ten neurons, the output layer corresponds to three running states, namely a normal running state, a critical running state and an overrun running state, and threshold intervals corresponding to the states are obtained through statistics of historical running data, for example, the normal running power threshold interval is within a range of plus or minus ten percent of rated power, the critical running power threshold interval is within a range of plus or minus twenty percent of rated power, and the overrun running power threshold interval exceeds the rated power by plus or minus twenty percent.
The invention uses wavelet transformation to carry out noise reduction filtering treatment on real-time power data, can effectively remove noise and interference in the data, improves the accuracy and reliability of the data, can further reduce errors and uncertainties in the data and improve the stability and consistency of the data by carrying out data standardization treatment on a Kalman filter, realizes accurate prediction and evaluation on the running state of a parallel module by learning and simulating complex system behaviors through strong nonlinear mapping capability of a feedforward neural network, and can ensure that the parallel module always runs in an optimal state by adopting an intelligent regulation mechanism and improve the overall performance and stability of the system.
In an embodiment, the calculating the power mutual feedback value of the parallel modules according to the comparison result to obtain the power adjustment target value of each parallel module includes:
Taking the parallel modules with the running states exceeding the preset state interval as overrun running parallel modules, and taking the rest parallel modules as normal running parallel modules;
Determining the power value to be regulated of each overrun parallel module, and sequencing the priority of each overrun parallel module according to the relation between the power value to be regulated and a preset regulation power interval;
and selecting an adjustment target corresponding to each overrun parallel module from each normal operation parallel module based on a priority order according to the distance between each normal operation parallel module and each overrun parallel module so as to perform power mutual feedback numerical calculation, and obtaining a power adjustment target value of each parallel module.
Specifically, when the power mutual feedback value (which can be considered as the power adjustment quantity that a single parallel module needs to feed back to another parallel module can be positive, negative or zero) is calculated, based on the relation between the running state of each parallel module and a preset state interval, each parallel module is divided into an overrun running parallel module and a normal running parallel module, the power value to be adjusted of the overrun running parallel module, namely the value exceeding the rated power corresponding to the preset state interval, is determined, and the power adjustment is carried out on the parallel modules in the overrun running state.
For the scene of larger power adjustment quantity, the smaller the priority value is, the higher the adjustment priority is, the adjustment is carried out from the low power adjustment quantity parallel module, the transition is gradually carried out to the high power adjustment quantity parallel module, and the high power mutation is avoided, for example, when a plurality of parallel modules simultaneously need power adjustment, the priority of the parallel module with the power adjustment quantity being less than ten percent of rated power is 1, the priority of the parallel module with the power adjustment quantity being between ten percent and twenty percent of rated power is 2, the priority of the parallel module with the power adjustment quantity being greater than twenty percent of rated power is 3, and the like.
The method comprises the steps of selecting an adjusting object based on a priority order and carrying out power mutual feedback numerical calculation, wherein the adjusting object is selected from normal operation parallel modules according to the distance between the adjusting object and the overrun operation parallel module, and is generally provided by a normal operation state parallel module close to (i.e. within a preset distance interval), for example, when the power of one parallel module exceeds thirty percent of a rated value, the feedback quantity required by the target parallel module is thirty percent of the rated value (the feedback quantity is the power adjustment target value of the target parallel module), the feedback quantity is negative because the feedback quantity is a receiver, and the adjacent two parallel modules respectively provide fifteen percent of the power adjustment quantity, namely fifteen percent of the rated value of the target parallel module is required to be fed back to the target parallel module by the parallel module because the feedback quantity is positive. In the power mutual feedback process, stability is ensured through a preset attenuation proportion, and the preset attenuation proportion is set to be five percent, so that a power regulation target value of each parallel module can be obtained.
The method and the device accurately identify the over-limit operation parallel modules and the normal operation parallel modules by comparing the operation states of the parallel modules with the preset state interval, are favorable for rapidly positioning the problem parallel modules, and ensure that the parallel modules with high priority can be preferentially regulated for the priority sorting of the over-limit operation parallel modules, so that the operation states of the whole system are more effectively controlled, the negative influence caused by the over-limit operation of the parallel modules is reduced, the power regulation target values of all the parallel modules can be obtained through the power mutual feedback numerical calculation, the scientificity and rationality of regulation measures are ensured, and the scheme realizes the global optimization of the whole system by comprehensively considering the operation states, the priorities, the distances and other factors of the parallel modules.
In an embodiment, the analyzing the real-time power data to obtain a power adjustment target value and a power curve of each parallel module further includes:
Noise filtering is carried out on the real-time power data through a band elimination filter to obtain a third power data set, and weighting processing is carried out on the third power data set through a hanning window function to obtain a fourth power data set;
Correcting fluctuation points in the fourth power data set according to the self-adaptive Kalman filter, and fitting the corrected fourth power data set by adopting a feedforward neural network to obtain a primary power curve of each parallel module;
Extracting key features in the primary power curves, calculating fundamental wave components and harmonic components of the key features through Fourier transformation, comparing the fundamental wave components and the harmonic components with a preset harmonic distortion rate threshold, marking a curve section exceeding the preset harmonic distortion rate threshold as a switching distortion section, and removing the switching distortion section to obtain power curves of the parallel modules, wherein the key features are used for representing high-frequency power fluctuation components of the corresponding parallel modules under non-ideal switching conditions.
According to the fluctuation characteristics of the third power data set, the power data at different moments are weighted by using a Hanning window function, the window length is set to be five times of the switching period, attenuation weights with a center weight value of one and an edge weight value of two zero are given to data points in the window, and a fourth power data set is obtained through weighted average calculation;
Aiming at the abrupt change point in the fourth power data set, setting an observed quantity as a power data sequence, setting a state quantity as a power mean value and a power variance by using a self-adaptive Kalman filter, correcting fluctuation points which deviate from the range of three times of the power mean value and the power variance, setting twelve time sequence neurons through an input layer of a feedforward neural network, eight neurons through an implicit layer, and setting four neurons through an output layer to respectively correspond to the amplitude, the phase, the frequency and the waveform characteristics of a power curve so as to fit the corrected power data, and obtaining a primary power curve of each parallel module;
The method comprises the steps of extracting key characteristics such as zero-point crossover frequency, fluctuation amplitude ratio, waveform distortion rate, power factor and the like from a primary power curve, calculating fundamental wave components and harmonic components of the characteristics through Fourier transformation, comparing the fundamental wave components and harmonic components with a preset harmonic distortion rate threshold, marking a section exceeding the threshold as a switching distortion section, and removing the switching distortion section to obtain the power curve of each parallel module, wherein the zero-point crossover frequency reflects the basic period of power fluctuation when the power curve characteristic is extracted, the fluctuation amplitude ratio is usually around 50 Hz, the relative size between wave crests and troughs is smaller than 0.1 when the power curve is in normal operation, the waveform distortion rate describes the deviation degree of the curve and an ideal sine wave, the power factor represents the ratio of active power to apparent power, calculating the fundamental wave components and the first 40 times of harmonic components through 1024-point fast Fourier transformation, and marking a time section exceeding the threshold as the switching distortion section when the power curve is in normal operation.
The DC converter with the switching frequency of 2 kilohertz can generate oscillation due to the influence of parasitic inductance and parasitic capacitance in the switching process, so that a power curve has high-frequency fluctuation, the high-frequency components are effectively suppressed after the band-stop filtering and the weighted average processing, the residual low-frequency components can reflect the load change characteristic, when the load suddenly changes, step response can be formed on the power curve, and a smooth transition curve can be obtained through Kalman filtering and neural network fitting, so that the subsequent characteristic extraction and analysis are facilitated. The invention accurately corrects fluctuation points in the power data set through the self-adaptive Kalman filter, thereby improving the stability and consistency of the data, adopting the strong nonlinear mapping capability of the feedforward neural network, fitting the corrected data to generate primary power curves of all parallel modules by adopting the characteristic of being capable of fitting a complex power change relation well, and the key characteristics in the primary power curves can represent high-frequency power fluctuation components of the parallel modules under the non-ideal switching condition, provide important basis for subsequent harmonic analysis, accurately identify and reject curve sections exceeding a threshold value, and be beneficial to eliminating power distortion caused by improper switching operation or equipment fault and improving the accuracy and reliability of the power curves.
S2, evaluating the power states of the parallel modules based on the power curves to determine a first parallel module, controlling the first parallel module to generate a power adjustment instruction and sending the power adjustment instruction to second parallel modules adjacent to the first parallel module, so that the second parallel modules perform primary power adjustment according to the power adjustment instruction and the corresponding power adjustment target values to obtain a power adjustment quantity;
In an embodiment, the evaluating the power state of each parallel module based on each power curve to determine a first parallel module, controlling the first parallel module to generate a power adjustment command and sending the power adjustment command to a second parallel module adjacent to the first parallel module, and the evaluating includes:
Performing four-layer decomposition on each power curve by adopting discrete wavelet transformation, and reconstructing a decomposition result by wavelet coefficients to obtain a characteristic data matrix so as to extract a power characteristic vector sequence from the characteristic data matrix;
Inputting the power characteristic vector sequence into a mapping relation model established by a radial basis function support vector regressor for processing to obtain a current power state of a corresponding parallel module, and calculating the Euclidean distance between the current power state and a preset average division threshold value to obtain a power state evaluation value of the corresponding parallel module;
And taking the parallel module with the power state evaluation value exceeding a preset performance index as a first parallel module, controlling the first parallel module to generate a power adjustment instruction, and sending the power adjustment instruction to a second parallel module adjacent to the first parallel module.
Specifically, the method comprises the steps of performing four-layer decomposition on a power curve by adopting discrete wavelet transformation, selecting a second-layer high-frequency coefficient to calculate a power fluctuation amplitude, a third-layer high-frequency coefficient to calculate a power fluctuation period, a fourth-layer low-frequency coefficient to calculate a power change rate and an adjacent layer coefficient to calculate a power phase difference according to an energy distribution rule of wavelet coefficients, and obtaining a standardized characteristic data matrix through wavelet coefficient reconstruction to extract a power characteristic vector sequence from the data matrix, wherein the second-layer high-frequency coefficient reflects short-term change characteristics of the power fluctuation, the corresponding time scale is a millisecond level, the power fluctuation amplitude is obtained through square sum of calculated coefficients, the third-layer high-frequency coefficient reflects mid-term change characteristics of the power fluctuation, the corresponding time scale is a second level, the power fluctuation period is obtained through zero crossing point interval of the calculated coefficients, the fourth-layer low-frequency coefficient reflects long-term trend of the power change, the corresponding time scale is a minute level, and the power change rate is obtained through difference of the calculated coefficients;
the method comprises the steps of training by using historical data to establish a power state mapping relation model through a radial basis function support vector regressor, taking a standardized power characteristic vector sequence as input to obtain a current power state of a corresponding parallel module, taking state data with power average division rate exceeding ninety-five percent in the historical data as a preset average division threshold value, and calculating the Euclidean distance between the current power state and the preset average division threshold value to obtain a power state evaluation value, wherein when the radial basis function support vector regressor establishes the power state mapping relation, the width parameter of a kernel function is set to be 0.5, a penalty factor is set to be 100, an optimal parameter combination is selected through cross verification, the selection of the preset average division threshold value is based on the state with the best power average effect in the historical data, the power deviation of each parallel module in the states is less than five percent, the power fluctuation amplitude is less than ten percent of the rated value, and the standard deviation of the power fluctuation period is less than twenty percent of the average value;
And finally, setting an equipartition performance judging rule, controlling the parallel modules with the power state evaluation value exceeding twenty percent of the preset performance index to serve as first parallel modules and generating a power adjusting instruction to be sent to the second parallel modules adjacent to the first parallel modules through the optical fiber ring network, wherein each parallel module in the flexible equipment can be the first parallel module and can also be the second parallel modules of other first parallel modules at the same time, and the adjacent modules can be understood as being within a preset distance range, but the specific distance numerical value is not limited. The power regulation command adopts hexadecimal format, including parallel module address, power regulation target value and regulation priority, for example, the power regulation command format of a certain parallel module is that the starting bit 0xAA, parallel module address 0x01, power value 0x0F, priority 0x02 and ending bit 0xFF, the data interaction is implemented between all parallel modules by using baud rate 115200 through serial communication interface, data frame interval time is 10 ms, feedback checker can ensure data transmission correctness through cyclic redundancy check, in addition, when the equipartition performance is determined, the calculation of Euclidean distance considers three dimensionalities of power amplitude difference, phase difference and fluctuation period difference, when the difference of a certain dimensionality exceeds the corresponding threshold value, the power equipartition request is triggered, and the optical fiber ring network adopts bidirectional redundancy architecture.
In the power sharing control process of the parallel modules, when the power of a certain parallel module exceeds the rated value by thirty percent due to abrupt change of load, the power characteristics extracted through wavelet transformation show that the power fluctuation amplitude is increased, the fluctuation period is shortened, the state evaluation value output by the support vector regressor is obviously deviated from a preset sharing threshold value, after a power sharing request is triggered, adjacent parallel modules respond one by one according to the token transmission sequence, and the power deviation of each parallel module after power adjustment is reduced to within five percent of the rated value. The method and the device enhance the expression capability and robustness of the feature data through multi-level decomposition and reconstruction, extract the power feature vector sequence from the feature data matrix to convert complex power signals into simple and efficient feature vectors, provide convenience for subsequent processing and analysis, perform Euclidean distance calculation and power state assessment through a high-precision mapping relation model to quantify the deviation degree of the power states of the parallel modules, provide important basis for subsequent control and adjustment, and further reduce power fluctuation and loss through cooperative adjustment among adjacent parallel modules, thereby being beneficial to improving the stability and reliability of the system.
In an embodiment, the step of enabling each second parallel module to perform power adjustment once according to the power adjustment command and the corresponding power adjustment target value to obtain a power adjustment amount includes:
according to the power adjustment instruction, each second parallel module carries out quantization constraint on a power adjustment target value corresponding to the second parallel module by using a gradient algorithm and minimizing a power climbing rate variance as an objective function, and generates a power adjustment sequence comprising an adjustment target value, adjustment time and adjustment direction;
Calculating a climbing speed value of each second parallel module by a Gaussian elimination method based on a power adjustment sequence of each second parallel module, performing amplitude limiting treatment on the climbing speed value exceeding a preset climbing speed, and distributing power adjustment increment of the corresponding second parallel module according to a preset attenuation proportion;
and taking the power adjustment increment and the corresponding climbing rate value as a state quantity and a control quantity respectively, and calculating by a Lagrangian multiplier method to obtain the power adjustment quantity of each second parallel module.
Specifically, the invention supports that the maximum power adjustment target value of the parallel module is plus or minus thirty percent of rated power, utilizes a gradient algorithm to minimize the power climbing rate variance as an optimization target, carries out quantization constraint on the power adjustment quantity of each parallel module to generate a power adjustment sequence comprising an adjustment target value, adjustment time and adjustment direction, calculates the deviation between the current climbing rate and the average rate in each iteration of the gradient algorithm, carries out power adjustment quantity correction on the parallel module with larger deviation, considers the physical constraint of the parallel module, has the power adjustment upper limit not exceeding twenty percent of the rated power, has the adjustment time not less than 10 milliseconds, has the adjustment direction consistent with the load change trend, and comprises the target power value of each time point;
Then calculating the maximum climbing rate meeting the adjustment capacity of all the parallel modules through a consistency constraint operator based on the power adjustment sequence of each second parallel module, solving the climbing rate distribution problem by using a Gaussian elimination method to obtain climbing rate values of each parallel module, carrying out amplitude limiting treatment on the rate values exceeding the adjustment range, and distributing power adjustment increment according to a preset attenuation proportion; when calculating the maximum climbing rate, the consistency constraint operator comprehensively considers the adjustment capability difference of each parallel module, for example, the maximum adjustment rate of the parallel module with the power of 1000 watts is 100 watts per millisecond, the maximum adjustment rate of the parallel module with the power of 500 watts is 50 watts per millisecond, the adjustment capability is consistent as a consistency constraint condition, a mathematical model is established by combining the working characteristics of each parallel module, the model is used for describing the interaction among the modules and the output characteristics of the whole parallel module, an optimization algorithm (such as a genetic algorithm, a particle swarm algorithm and the like) or a numerical solution method (such as a Newton iteration method, a gradient descent method and the like) is adopted, and the maximum speed of the output power of the parallel module changing along with time, namely the maximum climbing rate, is solved on the premise of meeting the consistency constraint condition;
And finally, selecting the power adjustment increment as a state quantity and the corresponding climbing rate as a control quantity, and solving the power adjustment quantity of each corresponding parallel module by using a Lagrange multiplier method, wherein in the state space expression, the power value is used as the state quantity to describe the running state of the parallel module, the climbing rate is used as the control quantity to reflect the dynamic characteristic of power change, and the Lagrange multiplier method also introduces power balance constraint and rate consistency constraint when solving the optimal control problem.
The invention adopts a gradient algorithm, takes the minimum power climbing rate variance as an objective function, carries out quantization constraint on the power adjustment target values of all adjacent parallel modules, can ensure the stability and consistency in the power adjustment process, avoids the impact on a system caused by overlarge power fluctuation, adopts a Gaussian elimination method to calculate and obtain the climbing rate value of each corresponding parallel module, and ensures the accuracy of a calculation result.
In an embodiment, the calculating the power adjustment amount of each of the second parallel modules by the lagrangian multiplier method includes:
Taking the power adjustment rate constraint and the load power constraint of each second parallel module as constraint conditions, and taking the power climbing rate of each parallel module in the flexible equipment as a target model;
And solving the target model through a Lagrangian multiplier method, updating the climbing speed value according to a steepest descent method when iteration stop criteria are not met, and iteratively solving the target model through the Lagrangian multiplier method based on the updated climbing speed value until the iteration stop criteria are met, and outputting power adjustment quantity of each parallel module in the flexible equipment when iteration turns are stopped, wherein the iteration stop criteria are that the change rate of power adjustment increment with adjacent turns is smaller than a preset power adjustment increment change threshold.
Specifically, when solving a target model through a Lagrangian multiplier method, the invention introduces a power balance constraint and a rate consistency constraint, and stops iteration when the power adjustment increment change rate of two adjacent iterations is less than five percent, wherein the step factor of the steepest descent method is set to be 0.5, and the iteration solving process is continued according to the optimized climbing rate value until the iteration stopping criterion is met, and the power adjustment quantity of each parallel module when stopping iteration rounds is output. The optimal climbing rate distribution scheme can be obtained through multiple rounds of iterative computation, all parallel modules synchronously adjust power according to the uniform rate, power conflict and oscillation among the parallel modules are avoided, the response time of the whole equipartition process is less than 500 milliseconds, and the power equipartition precision can be superior to five percent. The invention takes the uniform power climbing rate of each parallel module in the flexible equipment as a target model, is beneficial to ensuring the coordination and stability of the parallel modules in the operation process, can reduce the power difference between the parallel modules by optimizing the power climbing rate, and avoids the condition of overload or insufficient power, thereby improving the operation efficiency and reliability of the whole flexible equipment.
S3, acquiring historical power data of each parallel module, and constructing a power trend prediction model of each parallel module according to the historical power data so as to obtain a power prediction sequence according to the power trend prediction model;
in one embodiment, step S3 includes:
Collecting historical power data of each parallel module through a rolling window, and carrying out normalization processing on the historical power data in the rolling window by taking a rated power value as a reference, so as to train a long-short-time memory neural network as sample data, and obtain a power trend prediction model corresponding to the parallel module to generate a power trend prediction sequence;
Extracting oscillation characteristics of the historical power data based on fast Fourier transformation, and weighting a power trend prediction sequence of a parallel module to which the historical power data belong according to the inverse of an oscillation period to obtain a power prediction compensation sequence;
and correcting the power prediction compensation sequence according to a Kalman filter to obtain the power prediction sequence of each parallel module.
Specifically, a time window with a fixed length of 1 second is selected through a rolling window calculator, power data in the window is normalized by taking a rated power value as a reference, twelve time sequence neurons are arranged on an input layer of a long-short-time memory neural network to capture time sequence characteristics of power change, eight neurons are arranged on an hidden layer to extract deep layer characteristics of power change, four neurons are arranged on an output layer to respectively correspond to four prediction points in 100 milliseconds in the future, model training is carried out on the basis of a pre-acquired standard power curve to obtain a power trend prediction model so as to generate a power trend prediction sequence, wherein the training data is selected from a climbing section in the standard power curve, and the three typical scenes of stable climbing, oscillation climbing and step climbing are included;
Setting a prediction time domain to be 100 milliseconds by utilizing a variable predictor, acquiring a historical power state value sampled in a 1 millisecond period from a real-time database, calculating a frequency spectrum component in a range from 0 to 500 hertz by utilizing a fast Fourier transform, extracting a main vibration frequency and an amplitude in a power oscillation characteristic, and carrying out weighting processing on a power trend prediction sequence according to the inverse of the oscillation period as a weighting coefficient to generate a power prediction compensation sequence;
And finally, establishing a prediction error minimization target through a sliding time domain controller, setting state quantity as a power mean value, a variance and an observed quantity as real-time power values by using a Kalman filter, carrying out self-adaptive correction on power prediction compensation values to obtain power prediction sequences of all parallel modules, correcting the prediction compensation values through self-adaptive gain, and reducing the response speed of a control loop and inhibiting power overshoot by introducing a damping compensation term when the oscillation amplitude is large.
The invention combines a plurality of advanced algorithms and technologies, can efficiently process and predict the power data of the parallel modules, provides powerful support for real-time control and optimization of a parallel module system, can generate a highly accurate power prediction sequence through the synergistic effect of a plurality of steps such as normalization processing, FFT (fast Fourier transform) extraction oscillation characteristics, weighting processing, kalman filter correction and the like, provides powerful guarantee for stable operation of the system, predicts future power variation trend of each parallel module by introducing a power adjustment strategy based on prediction control, and corrects the power adjustment quantity in advance so as to inhibit oscillation in the power climbing process, thereby realizing accurate power dynamic adjustment of each parallel module in flexible equipment.
S4, correcting the power adjustment quantity of each parallel module through the power prediction sequence of each parallel module to obtain power adjustment correction quantity, and controlling each parallel module to carry out secondary power adjustment based on the power adjustment correction quantity of the parallel module so as to realize power adjustment of each parallel module in the flexible equipment;
Specifically, the power adjustment quantity is decomposed by a piecewise linear interpolation calculator according to a fixed slope based on a power prediction sequence, the power adjustment quantity is divided into a plurality of time periods, the fixed adjustment slope is kept in each period, the typical slope is set to be ten percent of rated power per millisecond, then the power feedback value of each parallel module is read from a serial bus based on a token ring network, the power deviation exceeding a prediction interval is subjected to proportional correction by a feedforward compensator, the power adjustment correction quantity is obtained to carry out power adjustment on the corresponding parallel module, and the adjustment priority is distributed according to the size of a power adjustment increment, so that real-time suppression of power oscillation is realized.
In addition, when one of the parallel modules has an over-high temperature fault, the state monitor detects the over-limit temperature within 3 milliseconds, immediately reduces the priority of the parallel module to the lowest, limits the power below 300 watts, and after the other parallel modules are regrouped, the first group of parallel modules complete power sharing within 5 milliseconds, the sharing precision is better than two percent, the second group of parallel modules are adjusted in place within 10 milliseconds, and the last group of parallel modules are used as standby and are gradually adjusted to the sharing state within 15 milliseconds. When the load power is suddenly changed, the hierarchical optimization calculator preferentially adjusts the output power of the first group of parallel modules from fifty percent to eighty percent of rated power, so that the core parallel modules can respond to load change rapidly, other groups of parallel modules gradually follow according to a preset response time sequence, the whole equipartition process is completed within 20 milliseconds, and even if power overshoot occurs to the individual parallel modules during the equipartition process, fluctuation can be controlled within ten percent of rated value through feedforward compensation.
In an embodiment, the correcting the power adjustment amount of each parallel module through the power prediction sequence of each parallel module to obtain a power adjustment correction amount and controlling each parallel module to perform secondary power adjustment based on the power adjustment correction amount of the parallel module includes:
Collecting real-time power data, real-time temperature data, real-time voltage data and real-time current data of each parallel module in the flexible equipment after secondary power adjustment, taking the real-time power data, the real-time voltage data and the real-time current data as training samples to train the deep neural network, and outputting the change rate of load power;
triggering parameter correction of a controller in the flexible equipment to obtain a parameter correction value when the load power change rate exceeds a preset power change rate threshold;
based on the parameter correction value, optimizing the power adjustment correction quantity of each parallel module by adopting a genetic algorithm to obtain a power adjustment correction optimization sequence, and taking the ratio of the load power change rate of each parallel module to the rated power of the parallel module as a compensation coefficient;
And carrying out weighting treatment on the power adjustment correction optimization sequence through the compensation coefficient to obtain a power adjustment compensation quantity, and controlling each corresponding parallel module to carry out three times of power adjustment based on the self power adjustment compensation quantity so as to realize power adjustment of each parallel module in the flexible equipment.
Specifically, the invention collects real-time power data, real-time temperature data, real-time voltage data and real-time current data of each parallel module in the flexible equipment after secondary power adjustment, trains a deep neural network as training samples, sets an input layer containing twelve feature vectors, an hidden layer containing eight neurons of each layer of three layers and an output layer containing four optimization parameters through the deep neural network, and respectively establishes mapping relations for temperature change, voltage fluctuation, current fluctuation, environmental temperature and environmental humidity data under different load rates so as to output the load power change rate through the trained model;
triggering correction when the change rate of load power exceeds twenty percent of rated value or the environmental parameter exceeds a preset range, carrying out online automatic correction on the parameter of a controller of flexible equipment, carrying out optimization calculation on the power average parameter by utilizing a genetic algorithm with the chromosome length of 32 bits according to the corrected parameter correction value, mapping the parameter value of the controller through binary coding, setting the crossover probability to eighty percent and the variation probability to twenty percent, reading the parameter value of the proportional gain between one to ten zero percent, the integral time constant between one to one hundred milliseconds and the response dead zone value between one to five percent of rated power from a controller parameter library, and calculating the fitness function value by utilizing the power average deviation and the adjustment time weighting to generate a power adjustment correction optimization sequence;
The ratio of the load power change rate of each parallel module to the rated power of the parallel module is used as a compensation coefficient, for example, when the load power is changed from 500 watts to 800 watts within 10 milliseconds, the power change rate is 30 watts per millisecond, the compensation coefficient takes a value of 0.03, a sliding average calculator carries out smoothing treatment on the compensated control parameters within a window of 50 milliseconds, the impact of parameter jump on the control performance is avoided, then the power adjustment correction optimization sequence is weighted through the compensation coefficient to obtain the power adjustment compensation quantity, and each corresponding parallel module is controlled to carry out power adjustment again based on the power adjustment compensation quantity so as to realize the power adjustment of each parallel module in the flexible equipment.
In the cooperative control process of the parallel module under different working conditions, the change of the ambient temperature directly influences the heat radiation capability of the parallel module, so that the power regulation characteristic is influenced, the response speed of the controller is automatically reduced when the temperature rises through the parameter mapping relation established by the deep neural network, overshoot is avoided, the response speed is properly improved when the temperature falls, and the dynamic performance is improved. The change of the load power reflects the dynamic characteristics of the electric equipment, the control parameters obtained by genetic algorithm optimization can adapt to different load change rates, the response dead zone value is increased when the high power is changed, the control sensitivity is reduced, the response dead zone value is reduced when the low power is changed, the control precision is improved, and the dynamic self-adaption of power average control is realized.
The invention can search the optimal solution in the complex parameter space by triggering the parameter correction and the optimization of the genetic algorithm, thereby obtaining the power adjustment correction optimization sequence, takes the ratio of the load power change rate of each parallel module to the rated power of the parallel module as a compensation coefficient to carry out the weighting treatment on the power adjustment correction optimization sequence, and considers the difference and the actual working state between the parallel modules, so that the power adjustment is more flexible and accurate. The established flow optimization mechanism of the autonomous cooperative control algorithm adjusts parameters of the cooperative control algorithm in a self-adaptive manner, and autonomously optimizes a power balance strategy according to the actual working condition of the flexible equipment and the change of the external environment, so that the power adjustment among parallel modules is dynamically self-adaptive, the problem of unbalanced power adjustment among the parallel modules in the flexible equipment is effectively solved, and the stability and the reliability of the system are improved.
Aiming at the problems of inconsistent power regulation states, inconsistent power climbing rates and the like of the parallel modules in the flexible equipment, the invention acquires the power data of the parallel modules in real time, and determines the power regulation target value of the parallel modules through an autonomous cooperative control algorithm. And when the judging result is not satisfied, starting power average regulation among the parallel modules, and cooperatively determining the power regulation quantity of each parallel module by adopting a distributed optimization algorithm so that the power climbing rates of the parallel modules tend to be consistent. Meanwhile, a power regulation strategy based on predictive control is introduced, the future power variation trend of the parallel module is predicted, the power regulation quantity is corrected in advance, and the oscillation in the power climbing process is restrained.
The embodiment of the application designs a power adjustment method aiming at parallel modules of flexible equipment, which is difficult to accurately adjust the power of each parallel module in the flexible equipment by the existing control algorithm, and realizes the acquisition of real-time power data of each parallel module in the flexible equipment, and the analysis and the processing of the real-time power data to obtain power adjustment target values and power curves of each parallel module; the power state of each parallel module is evaluated based on each power curve to determine a first parallel module, the first parallel module is controlled to generate a power adjustment instruction and send the power adjustment instruction to a second parallel module adjacent to the first parallel module, so that each second parallel module carries out primary power adjustment according to the power adjustment instruction and a corresponding power adjustment target value to obtain a power adjustment quantity, historical power data of each parallel module is obtained, a power trend prediction model of each parallel module is constructed according to the historical power data to obtain a power prediction sequence according to the power trend prediction model, the power adjustment quantity of each parallel module is corrected through the power prediction sequence of each parallel module to obtain a power adjustment correction quantity, and each parallel module is controlled to carry out secondary power adjustment based on the power adjustment correction quantity of the parallel module, so that accurate power dynamic adjustment of each parallel module in flexible equipment is realized.
Although the steps in the flowcharts described above are shown in order as indicated by arrows, these steps are not necessarily executed in order as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders.
In another embodiment, as shown in fig. 2, a second aspect of the present invention provides a power conditioning system for a parallel module of a flexible device, comprising:
the data processing module 10 is used for acquiring real-time power data of each parallel module in the flexible equipment, and analyzing and processing the real-time power data to obtain power regulation target values and power curves of each parallel module;
a primary adjustment module 20, configured to evaluate a power state of each parallel module based on each power curve, so as to determine a first parallel module, control the first parallel module to generate a power adjustment instruction, and send the power adjustment instruction to a second parallel module adjacent to the first parallel module, so that each second parallel module performs primary power adjustment according to the power adjustment instruction and a corresponding power adjustment target value, and obtain a power adjustment amount;
the power prediction module 30 is configured to obtain historical power data of each parallel module, and construct a power trend prediction model of each parallel module according to the historical power data, so as to obtain a power prediction sequence according to the power trend prediction model;
And the secondary adjustment module 40 is configured to correct the power adjustment amount of each parallel module through the power prediction sequence of each parallel module, obtain a power adjustment correction amount, and control each parallel module to perform secondary power adjustment based on the power adjustment correction amount of the parallel module, so as to implement power adjustment of each parallel module in the flexible device.
It should be noted that, each module in the power adjustment system for the parallel module of the flexible device may be implemented in whole or in part by software, hardware, or a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules. For a specific limitation of a power adjustment system for a parallel module of a flexible device, see the limitation of a power adjustment method for a parallel module of a flexible device hereinabove, the two have the same functions and roles, and are not described herein.
In summary, the invention relates to the technical field of information processing, and discloses a power regulation method and a power regulation system for parallel modules of flexible equipment, wherein real-time power data of each parallel module in the flexible equipment are analyzed and processed to obtain power regulation target values and power curves of each parallel module; the power state of the parallel module to which the power curve belongs is evaluated based on the power curve to determine a first parallel module, a generated power adjusting instruction is controlled to be sent to the second parallel module adjacent to the first parallel module, so that each second parallel module can carry out power adjustment according to the power adjusting instruction and the corresponding power adjusting target value of the second parallel module, a power adjusting quantity is obtained, a power adjusting strategy based on predictive control is introduced according to historical power data of each parallel module, future power change trend of each parallel module is predicted, the power adjusting quantity is corrected in advance, oscillation in the power climbing process is restrained, and accurate power dynamic adjustment of each parallel module in flexible equipment is achieved.
In this specification, each embodiment is described in a progressive manner, and all the embodiments are directly the same or similar parts referring to each other, and each embodiment mainly describes differences from other embodiments. In particular, for system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, as relevant to see a section of the description of method embodiments. It should be noted that, any combination of the technical features of the foregoing embodiments may be used, and for brevity, all of the possible combinations of the technical features of the foregoing embodiments are not described, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The foregoing examples represent only a few preferred embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the application. It should be noted that modifications and substitutions can be made by those skilled in the art without departing from the technical principles of the present application, and such modifications and substitutions should also be considered to be within the scope of the present application. Therefore, the protection scope of the patent of the application is subject to the protection scope of the claims.