Disclosure of Invention
Based on the above problems, the invention aims to provide a distributed energy storage network resource regulation decision method which can comprehensively utilize multi-level data, improve the accuracy of energy demand prediction, enhance the flexibility and the intelligence level of a regulation strategy, regulate resources by adopting a global optimization method, and improve the robustness and the reliability of a system.
The distributed energy storage network resource regulation and control decision method comprises the following steps:
S1, multi-level data acquisition and processing, namely acquiring multi-level data of a distributed energy storage system, wherein the multi-level data comprises microscopic equipment operation data, mesoscopic regional power grid data and macroscopic whole network data, and synchronously processing the multi-level data;
S2, a bidirectional interactive user demand response is carried out, wherein a bidirectional interactive mechanism with a user side is established, the user power demand and energy storage willingness are collected, a demand response signal is sent to the user when the power is tension or the power price fluctuates, the user is encouraged to adjust the power consumption behavior, and corresponding economic incentives are given;
S3, dynamic energy route optimization, namely constructing a dynamic energy route model of the distributed energy storage network based on multi-level data and a user demand response result, predicting future energy demand and supply conditions, wherein the construction of the dynamic energy route model comprises the following steps:
S31, energy demand prediction, namely, based on historical energy demand and supply data, constructing an energy demand prediction model based on a machine learning algorithm by considering data of each level and feedback of user demand response, wherein input characteristics comprise historical load, electricity price fluctuation and user response data, and the preprocessed multi-level data and the preprocessed user demand response data are used as model input to perform joint training;
s32, analyzing user response data, predicting behavior changes of the user under different electricity prices and demand response signals, establishing a user response model, and quantifying energy storage charging and discharging behaviors of the user under different situations;
S33, a dynamic optimization algorithm is used for constructing a dynamic optimization model aiming at minimizing energy transmission loss and charge-discharge loss, solving an optimization problem by using a distributed algorithm, and ensuring calculation efficiency and instantaneity;
S34, a dynamic energy routing model is established based on the energy demand prediction and dynamic optimization model, the energy transmission path is dynamically adjusted based on data collected in real time and a prediction result, and the charging and discharging strategies of the energy storage system are optimized, so that efficient energy transmission and use are ensured.
Preferably, the multi-level data collection and processing of S1 specifically includes:
S11, acquiring microscopic level data, namely acquiring voltage, current, power and temperature data through a sensor arranged on the energy storage equipment;
S12, mesoscopic hierarchical data acquisition, namely installing monitoring equipment at each node of the regional power grid, and acquiring regional load and regional voltage level data;
s13, macro-level data acquisition, namely acquiring whole-network load, frequency and voltage data through a monitoring unit of a power grid dispatching center;
And S14, multi-level data synchronous processing, namely summarizing the micro-level equipment operation data, the mesoscopic level regional power grid data and the macro-level whole-network data to a central data processing platform, performing time stamp alignment on the data from different sources by using a time sequence synchronization technology, and integrating the multi-level data into a unified data set by using a data fusion algorithm.
Preferably, the bi-directional interactive user demand response in S2 specifically includes:
S21, user side data acquisition and monitoring, namely installing an intelligent ammeter at the user side, and monitoring and acquiring electricity utilization data of a user in real time, wherein the electricity utilization data comprise electricity consumption amount, electricity utilization time and electricity utilization equipment type, and acquiring data of self-owned energy storage equipment of the user, including energy storage energy, charge and discharge states and equipment health conditions, through an intelligent sensor;
s22, establishing a two-way interaction mechanism, namely establishing a user interaction platform based on the mobile application and the webpage platform, checking real-time electricity price, electricity consumption condition and energy storage state information through the user interaction platform, and setting personal preference and response willingness;
S23, user demand response and regulation, namely when the power is tension or the electricity price fluctuates, sending a demand response signal to a user through a user interaction platform, and suggesting the user to adjust the electricity consumption behavior, including deferring electricity consumption or starting the self energy storage equipment;
S24, an economic incentive mechanism, namely implementing a dynamic electricity price mechanism according to the load of the power grid and the electricity price condition of the market, wherein a user charges at the low electricity price and discharges at the high electricity price so as to obtain the benefit of electricity price difference, and giving economic incentive to the user participating in demand response;
and S25, cleaning and standardizing the data responded by the user to generate a demand response data set reflecting the electricity consumption behavior and the energy storage willingness of the user.
Preferably, the construction of the energy demand prediction model is based on an ELM model of an extreme learning machine, and the construction of the energy demand prediction model based on the ELM comprises the following steps:
Model input construction, which comprises historical load, regional load, electricity price fluctuation and user response data;
model construction and joint training:
Designing an ELM model architecture, determining the number of neurons of an input layer, a hidden layer and an output layer, randomly generating and fixing weights and biases of the hidden layer, only training the weights of the output layer, integrating a multi-level unified data set and a demand response data set into a comprehensive data set, taking the comprehensive data set as model input, and carrying out combined training, wherein the training is as follows:
dividing the comprehensive data set into a training set, a verification set and a test set, and ensuring that the training and evaluation data of the model are independent;
randomly initializing, namely randomly generating the weight and the bias of the ELM hidden layer, and fixing the weight and the bias;
Training the model, namely training the model by utilizing a training set, calculating the weight of an output layer by a least square method, minimizing a prediction error, evaluating the performance of the model on a verification set, adjusting the super parameters of the model according to a verification result, testing the model on a test set, and evaluating the generalization capability and the prediction precision of the model.
Preferably, the training of the ELM model includes randomly generating hidden layer parameters and solving for output weights by using a least squares method, and specifically includes:
Mapping of input layer to hidden layer:
let the input data set beWhereinIs the input feature vector for sample a,Is the corresponding output of the device,Represents the real space of the dimension n,Representing an m-dimensional real space;
The output of the hidden layer is expressed as: where X is the input matrix, W is the randomly generated weight matrix, b is the randomly generated bias vector,Is an activation function;
Mapping from the hidden layer to the output layer, wherein the relation between the hidden layer output and the output weight is as follows:
where H is the hidden layer output matrix,Is an output weight matrix, Y is a target output matrix;
Solving the output weight by a least square method:
Wherein, the method comprises the steps of, wherein,Is the Moore-Penrose pseudo-inverse of H.
Preferably, the user response model is based on a gradient lift tree GBT model, and the construction of the user response model specifically comprises the following steps:
Extracting key features in the demand response data set, including response time, response quantity, response frequency and electricity price sensitivity;
the user response model is established by utilizing the user response data, and the user response model is established specifically as follows:
classifying and clustering the users, namely identifying different types of user groups, wherein the user groups are divided into high-response users and low-response users;
the feature engineering is to construct a feature vector of user response, which comprises historical response data and electricity price fluctuation data;
Constructing a series of weak learners by the GBT model to improve the performance of the model, and setting parameters of the GBT model, wherein the parameters comprise the number of trees, the maximum depth of the trees, the learning rate and the minimum sample division number;
Training and verifying, namely performing model training by using a user response data set, evaluating model performance through cross verification, and adjusting model parameters;
User behavior change prediction, namely simulating user situations under different electricity prices and demand response signals, generating prediction data under various situations, predicting the behavior change of the user under the different electricity prices and demand response signals by using a GBT model, and calculating response probability and response quantity of the user under the different situations:
The input feature is to input simulated electricity prices and demand response signals and user history response data into the GBT model,
Prediction output GBT model outputs the response probability of the user under the current situationAnd response volume;
The energy storage charging and discharging behavior quantification is realized by simulating the charging and discharging behavior of the energy storage equipment of a user under different conditions, and specifically comprises the following steps:
defining response strategies of users under different electricity prices and demand response signals;
the behavior quantification is to quantify energy storage charging and discharging behaviors of a user under different conditions based on a user response model, and calculate specific charging and discharging quantity and time;
And the benefit evaluation is used for evaluating the economic benefits of the user under different response strategies, including electricity fee saving, incentive rewards and the like, and quantifying the comprehensive benefits of response behaviors.
Preferably, the energy storage charging and discharging behavior quantification specifically includes:
defining response strategies of users in different situations:
A high electricity price period, in which the user reduces electricity consumption or discharge;
a low electricity price period in which a user increases charge or decreases discharge;
Behavior quantification, namely calculating energy storage charge and discharge amounts of users under different conditionsAnd:
;
;
Wherein MaxCharge and MaxDischarge are the maximum charge and discharge capabilities of the user equipment;
the benefit evaluation is to calculate the economic benefit of the user under different response strategies:
;
Wherein,Representing electricity price;
the incentive rewards of the users participating in the demand response are evaluated, and the incentive rewards comprise electric charge discounts and point rewards.
Preferably, the constructing a dynamic optimization model for minimizing energy transmission loss and charge-discharge loss specifically includes:
the optimization objective function is as follows:
;
Wherein,Is the loss power of the ith energy transmission path,Is the transmission time of the ith energy transmission path,Is the charging loss power of the jth energy storage device,Is the charging time of the jth energy storage device,Is the discharge loss power of the kth energy storage device,N, M, L is the number of the energy transmission path, the charged energy storage device and the discharged energy storage device respectively;
Constraint conditions:
User response behavior constraints-user behavior data predicted using a user response model, including response probabilities at different electricity prices and demand response signalsAnd response volumeIncorporating the user response data into an optimization model, defining user response behavior constraints:
;
;
Wherein,AndThe response probabilities of the energy storage devices j and k in the current context,、Respectively the maximum charge and discharge of the energy storage device,Representing the expected energy of the energy storage device j in a charging operation,Representing the expected energy of the energy storage device k in a discharging operation;
Energy balance constraint: Wherein, the method comprises the steps of,、、Energy transmission, charge and discharge, respectively;
Energy storage device capacity constraints:
;
;
Wherein,AndThe upper limits of the charge and discharge capacities of the energy storage device, respectively;
The solving the optimization problem by using the distributed algorithm specifically comprises:
Decomposing the global optimization problem into a plurality of sub-problems, wherein each sub-problem corresponds to one energy storage device or transmission line, and decomposing by using an ADMM algorithm;
solving the sub-problems, namely independently solving each sub-problem locally and calculating corresponding charging power, discharging power and transmission power;
global coordination, namely summarizing the solving result of each sub-problem through a global coordinator, updating global variables, and ensuring the satisfaction of global energy balance and constraint conditions;
And (3) iterative updating, namely exchanging information between the sub-problems and the global coordinator in each round of iteration, updating respective optimization variables, and gradually converging to a global optimal solution.
Preferably, the dynamic energy routing model construction in S34 specifically includes:
S341, outputting results based on the energy demand prediction and dynamic optimization model, wherein the results comprise predicted energy demand, user response behavior prediction data and an optimization strategy for minimizing energy transmission loss and charge-discharge loss;
Constructing an energy routing model, defining a selection and scheduling strategy of a transmission path, and comprising the following steps:
Defining nodes and edges in a distributed energy storage network, and establishing a network topological structure, wherein the nodes comprise energy storage equipment, power generation equipment and a load center, and the edges comprise energy transmission lines;
S342, dynamically adjusting an energy transmission path:
real-time data acquisition, namely acquiring data of the running states of the energy storage system and the power grid in real time through a distributed sensor network, wherein the data comprise voltage, current, power, energy requirements and user responses;
a path adjustment algorithm, which dynamically adjusts the energy transmission path based on the data collected in real time and the energy demand prediction result;
S343, optimizing a charging and discharging strategy of the energy storage system:
Based on the energy demand prediction and the user response behavior model, optimizing the charge-discharge strategy of the energy storage system specifically comprises the following steps:
the method comprises the steps of carrying out calibration on an energy demand prediction result by utilizing real-time data, improving the accuracy of prediction, dynamically adjusting the charging and discharging time and power of an energy storage system according to a charging and discharging strategy output by an optimization model so as to minimize energy loss and economic cost;
s344, outputting the optimized energy transmission path and the energy storage regulation strategy:
Outputting the optimized energy transmission path, ensuring energy to be transmitted to a demand node to meet load demands, and issuing the optimized energy storage regulation strategy to each energy storage device to monitor the execution condition of the energy storage device in real time.
Preferably, the network topology is expressed as: wherein V represents a set of nodes and E represents a set of edges;
the path selection algorithm uses a shortest path algorithm to calculate an optimal energy transmission path:
Wherein, the method comprises the steps of, wherein,Is a sideIs used for the weight of the (c),Is a path selection variable;
Path weight calculation: Wherein, the method comprises the steps of, wherein,AndAs the weight coefficient of the light-emitting diode,In order to transfer the lost power of the energy,Is the energy transmission time;
The path adjustment algorithm is updated in real time based on the path, expressed as the formula:
Wherein, the method comprises the steps of, wherein,Is the path weight at the time t,AndTransmission loss power and transmission time at time t;
And (3) self-adaptive adjustment: Wherein, the method comprises the steps of, wherein,Is to adjust the step size of the step,Is the transmission loss power at time t +1,Is the actual demand at the time t,Is the energy forecast demand at time t.
The invention has the beneficial effects that:
According to the invention, the energy demand prediction model based on the Extreme Learning Machine (ELM) is established by collecting and fusing multi-level data of microscopic level, mesoscopic level and macroscopic level, the accuracy of energy demand prediction is improved, and the energy storage charging and discharging behaviors of a user under different conditions are quantized by combining the analysis results of the user response model, so that the prediction results are more accurate. The optimized energy transmission path and energy storage regulation strategy are dynamically adjusted based on data and prediction results acquired in real time, so that efficient energy transmission and use are ensured, and the utilization rate and economic benefit of the energy storage system are maximized.
According to the invention, a bidirectional interaction mechanism with the user side is established, the electricity demand and the energy storage wish of the user are collected, a demand response signal is sent to the user when the electricity tension or the electricity price fluctuates, the user is encouraged to adjust the electricity consumption behavior, and corresponding economic incentive is given. The user response model predicts response probability and response quantity by analyzing the behavior change of the user under different electricity prices and demand response signals, and improves the adaptability of the system to the user demand change.
According to the invention, based on the energy demand prediction and dynamic optimization model, an energy routing model is constructed, the charge and discharge strategies of an energy transmission path and an energy storage system are adjusted in real time by using a distributed optimization algorithm, and the high efficiency and reliability of energy transmission and energy storage regulation and control are ensured by self-adaptive adjustment and iterative optimization, so that the flexibility and response capability of the system are improved.
Detailed Description
The present invention will be described in further detail with reference to the following examples.
As shown in fig. 1-2, the distributed energy storage network resource regulation decision method comprises the following steps:
s1, multi-level data acquisition and processing, namely acquiring multi-level data of a distributed energy storage system, wherein the multi-level data comprises microscopic equipment operation data (voltage, current, power and temperature), mesoscopic regional power grid data (regional load and regional voltage level) and macroscopic whole-network data (whole-network load, frequency, voltage and the like), and synchronously processing the multi-level data;
S2, a bidirectional interactive user demand response is carried out, wherein a bidirectional interactive mechanism with a user side is established, the user power demand and energy storage willingness are collected, a demand response signal is sent to the user when the power is tension or the power price fluctuates, the user is encouraged to adjust the power consumption behavior (such as deferring the power consumption or starting the self energy storage equipment), and corresponding economic incentive is given;
S3, dynamic energy route optimization, namely constructing a dynamic energy route model of the distributed energy storage network based on multi-level data and a user demand response result, predicting future energy demand and supply conditions, wherein the construction of the dynamic energy route model comprises the following steps:
S31, energy demand prediction, namely, based on historical energy demand and supply data, taking data of each level and feedback of user demand response into consideration, constructing an energy demand prediction model based on a machine learning algorithm, wherein input characteristics comprise historical load, electricity price fluctuation and user response data, and taking preprocessed multi-level data and user demand response data (i.e. a unified data set and a comprehensive data set) as model input for joint training;
s32, analyzing user response data, predicting behavior changes of the user under different electricity prices and demand response signals, establishing a user response model, and quantifying energy storage charging and discharging behaviors of the user under different situations;
S33, a dynamic optimization algorithm is used for constructing a dynamic optimization model aiming at minimizing energy transmission loss and charge-discharge loss, solving an optimization problem by using a distributed algorithm, and ensuring calculation efficiency and instantaneity;
s34, a dynamic energy routing model is established based on an energy demand prediction and dynamic optimization model, an energy transmission path is dynamically adjusted based on data and a prediction result acquired in real time, a charge and discharge strategy of an energy storage system is optimized, efficient energy transmission and use are ensured, the model input comprises multi-level data acquired in real time and a user demand response result, and the model input is output as the optimized energy transmission path and the energy storage regulation strategy.
The main step framework of the dynamic energy routing model is as follows:
node and edge definition, namely establishing a topological structure of the energy storage network, and defining nodes in the network and edges for connecting the nodes.
Path selection algorithm the optimal path from the starting node to the target node is calculated using a shortest path algorithm, such as Dijkstra.
And calculating the transmission loss and time of the path, and taking the transmission loss and time as the weight of path selection.
And updating the real-time path, namely dynamically adjusting the path weight according to the data acquired in real time, and recalculating the optimal path to ensure the energy efficient transmission.
And (3) charge and discharge optimization, namely optimizing the charge and discharge strategy of the energy storage system according to the energy demand prediction and the user response behavior model, and minimizing the energy loss and the economic cost.
And the response behavior regulation and control is that the charge and discharge behaviors of the energy storage equipment at the user end are optimized according to the prediction result of the user response model, and the flexibility and the response capability of the system are improved.
The multi-level data acquisition and processing of S1 provides the basic data that play a key role in the energy demand prediction and dynamic optimization in S3.
And S2, the bidirectional interactive user demand response directly influences the user demand response analysis and the execution of a dynamic optimization algorithm in S3 by collecting the user power demand and response willingness, so that the model can reflect the user behavior change.
And S3, generating a dynamic energy routing scheme by using a prediction and optimization algorithm based on the data of the S1 and the S2 by using the dynamic energy routing optimization model.
S1, multi-level data acquisition and processing specifically comprises:
S11, acquiring microscopic level data, namely acquiring voltage, current, power and temperature data through a sensor arranged on energy storage equipment, transmitting the acquired data to a local edge computing node by utilizing a wireless sensing network for preliminary processing, denoising and filtering the data on the edge computing node, and generating a preprocessed equipment operation data set;
S12, mesoscopic hierarchical data acquisition, namely installing monitoring equipment at each node of the regional power grid, acquiring regional load and regional voltage level data, transmitting the data acquired by the monitoring equipment to a regional control center by using a regional communication network, and integrating the data at the regional control center to generate a regional power grid data set;
S13, macro-level data acquisition, namely acquiring whole network load, frequency and voltage data through a monitoring unit of a power grid dispatching center, transmitting the data to a database of the power grid dispatching center by utilizing a power special communication network, and summarizing the data in the power grid dispatching center to generate a whole network data set;
And S14, synchronously processing multi-level data, namely summarizing the operation data of the equipment at a micro level, the regional power grid data at a mesoscopic level and the whole-network data at a macro level to a central data processing platform, aligning time stamps of the data from different sources by utilizing a time sequence synchronization technology, ensuring timeliness and consistency of the data, integrating the multi-level data into a unified data set through a data fusion algorithm, and covering multi-dimensional information of the equipment, the region and the whole network.
Converting the data of different data sources into a unified unit, normalizing the data of different ranges to the same range (such as 0 to 1) so that the data have the same scale, facilitating subsequent processing and fusion, applying different weights to the data of different layers according to the importance and the credibility of the data sources by using a weighted average method, calculating a weighted average value, and fusing the data of a micro-level, a mesoscopic level and a macroscopic level to generate a unified data set. The relevance and complementarity between the data are considered in the fusion process, so that the integrity and consistency of the fused data are ensured.
The two-way interactive user demand response in S2 specifically includes:
S21, user side data acquisition and monitoring, namely installing an intelligent ammeter at the user side, monitoring and acquiring electricity consumption data of the user in real time, including electricity consumption, electricity consumption time and electricity consumption equipment types, acquiring data of self-contained energy storage equipment of the user through an intelligent sensor, including energy storage energy, charge and discharge states and equipment health conditions, and transmitting the data of the user side to a central data processing platform through a wireless network by utilizing the Internet of things technology for unified storage and management;
S22, establishing a two-way interaction mechanism, namely establishing a user interaction platform based on a mobile Application (APP) and a webpage platform, checking real-time electricity price, electricity consumption condition and energy storage state information through the user interaction platform, and setting personal preference and response willingness;
s23, user demand response and regulation, namely, when the electric power is tension or the electricity price fluctuates, a demand response signal is sent to a user through a user interaction platform, the user is recommended to adjust electricity consumption behavior, the electricity consumption is delayed or the self-contained energy storage equipment is started, a plurality of response options are provided for the user to select, such as immediate response, partial response, temporary non-response and the like, and the user is allowed to set an automatic response rule;
s24, an economic incentive mechanism, namely implementing a dynamic electricity price mechanism according to the power grid load and the market electricity price condition, wherein a user charges at the low electricity price and discharges at the high electricity price to obtain the electricity price difference income, and giving economic incentive to the user participating in the demand response, wherein the economic incentive comprises electricity fee discount, point rewards and cash return;
And the user is encouraged to continuously participate in the demand response by feeding back the effect of the demand response and the obtained economic incentive to the user through the user interaction platform, so that the response capability of the whole system is improved.
And S25, cleaning and standardizing the data responded by the user to generate a demand response data set reflecting the electricity consumption behavior and the energy storage willingness of the user.
The construction of the energy demand prediction model is based on an Extreme Learning Machine (ELM) model, and the construction of the energy demand prediction model based on the ELM comprises the following steps:
Model input construction, which comprises historical load, regional load, electricity price fluctuation and user response data;
model construction and joint training:
Designing an ELM model architecture, determining the number of neurons of an input layer, a hidden layer and an output layer, randomly generating and fixing weights and biases of the hidden layer, only training the weights of the output layer, integrating a multi-level unified data set and a demand response data set into a comprehensive data set, taking the comprehensive data set as model input, and carrying out combined training, wherein the training is as follows:
dividing the comprehensive data set into a training set, a verification set and a test set, and ensuring that the training and evaluation data of the model are independent;
randomly initializing, namely randomly generating the weight and the bias of the ELM hidden layer, and fixing the weight and the bias;
Training a model, namely training the model by utilizing a training set, calculating the weight of an output layer by a least square method, minimizing a prediction error, evaluating the performance of the model on a verification set, adjusting the super parameters (the number of hidden layer neurons) of the model according to a verification result, testing the model on a test set, and evaluating the generalization capability and the prediction precision of the model;
After the model is deployed, continuously collecting real-time data, periodically updating model parameters by using an online learning method, improving the adaptability and prediction accuracy of the model, monitoring the prediction performance of the model in real time, and continuously optimizing the model structure and parameter setting by an error analysis and feedback mechanism.
Training of the ELM model comprises the steps of randomly generating hidden layer parameters and solving output weights by using a least square method, and specifically comprises the following steps:
Mapping of input layer to hidden layer:
let the input data set beWhereinIs the input feature vector for sample a,Is the corresponding output of the device,Represents the real space of the dimension n,Representing an m-dimensional real space;
The output of the hidden layer is expressed as: where X is the input matrix, W is the randomly generated weight matrix, b is the randomly generated bias vector,Is an activation function (tan is selected);
Mapping from the hidden layer to the output layer, wherein the relation between the hidden layer output and the output weight is as follows:
where H is the hidden layer output matrix,Is an output weight matrix, Y is a target output matrix;
Solving the output weight by a least square method:
Wherein, the method comprises the steps of, wherein,Is the Moore-Penrose pseudo-inverse of H.
The ELM model is applied to the invention, and the energy demand prediction model based on the ELM is constructed by the following steps:
Input features including historical load, electricity price fluctuations, user response data, meteorological data, regional load and voltage levels are selected from the integrated dataset and normalized.
ELM model training:
Input layer, set the input feature vector as;
The hidden layer is used for randomly generating a weight matrix W and a bias vector b and calculating hidden layer output H;
an output layer for solving the output weight by least square method using training data;
Model performance is assessed using the validation set, and model hyper-parameters (hidden layer neuron number) are adjusted according to the validation results.
And continuously collecting real-time data, and periodically updating model parameters by using incremental training.
And (3) deploying the trained ELM model to predict the real-time energy demand, monitoring the prediction performance of the model in real time, and continuously optimizing the model structure and parameter setting through an error analysis and feedback mechanism.
Through the calculation steps, the energy demand prediction model based on the Extreme Learning Machine (ELM) can be constructed based on historical energy demand and supply data and combined with data of each level and feedback of user demand response, the preprocessed multi-level data and user demand response data are used as model input, combined training is carried out, prediction accuracy and adaptability of the model are improved, and reliable data support is provided for dynamic energy route optimization.
The user response model is based on a gradient lift tree GBT model, and the construction of the user response model specifically comprises the following steps:
Extracting key features in the demand response data set, including response time, response quantity, response frequency and electricity price sensitivity;
the user response model is established by utilizing the user response data, and the user response model is established specifically as follows:
classifying and clustering the users, namely identifying different types of user groups, wherein the user groups are divided into high-response users and low-response users, and the high-response users and the low-response users are realized by adopting K-means clustering;
the feature engineering is to construct a feature vector of user response, which comprises historical response data and electricity price fluctuation data;
Constructing a series of weak learners (decision trees) by the GBT model to improve the performance of the model, and setting parameters of the GBT model, wherein the parameters comprise the number of the trees, the maximum depth of the trees, the learning rate and the minimum sample division number;
Training and verifying, namely performing model training by using a user response data set, evaluating model performance through cross verification, and adjusting model parameters;
training is carried out based on GBT algorithm, and the specific steps comprise:
Initial model the first weak learner, typically a simple regression tree, is trained.
And calculating a prediction residual of the current model, namely, calculating the difference between the true value and the predicted value.
Iterative training, in each iteration, training a new regression tree to fit the residual of the previous iteration and updating the overall model.
And model combination, namely carrying out weighted average on the prediction results of all the weak learners to obtain a final prediction model.
GBT model training detailed steps:
initial model: typically, the initial modelSet as the mean of the target values on the training set,A is the number of samples;
Residual calculation for each round of iterationCalculating residual error:
Wherein, the method comprises the steps of, wherein,Is a loss function, e.g. square error;
Fitting residual errors by training a new regression treeFitting residual;
The update model is as follows: Wherein, the method comprises the steps of, wherein,The learning rate is usually set to a small value of 0.1;
the final model is the weighted sum of all weak learners:
。
User behavior change prediction, namely simulating user situations under different electricity prices and demand response signals, generating prediction data under various situations, predicting the behavior change of the user under the different electricity prices and demand response signals by using a GBT model, and calculating response probability and response quantity of the user under the different situations:
The input feature is to input simulated electricity prices and demand response signals and user history response data into the GBT model,
Prediction output GBT model outputs the response probability of the user under the current situationAnd response volume;
The energy storage charging and discharging behavior quantification is realized by simulating the charging and discharging behavior of the energy storage equipment of a user under different conditions, and specifically comprises the following steps:
Defining response strategies of users under different electricity prices and demand response signals, such as deferred electricity utilization, advanced charge, delayed discharge and the like;
the behavior quantification is to quantify energy storage charging and discharging behaviors of a user under different conditions based on a user response model, and calculate specific charging and discharging quantity and time;
And the benefit evaluation is used for evaluating the economic benefits of the user under different response strategies, including electricity fee saving, incentive rewards and the like, and quantifying the comprehensive benefits of response behaviors.
The energy storage charging and discharging behavior quantification specifically comprises the following steps:
defining response strategies of users in different situations:
A high electricity price period, in which the user reduces electricity consumption or discharge;
a low electricity price period in which a user increases charge or decreases discharge;
Behavior quantification, namely calculating energy storage charge and discharge amounts of users under different conditionsAnd:
;
;
Wherein MaxCharge and MaxDischarge are the maximum charge and discharge capabilities of the user equipment;
the benefit evaluation is to calculate the economic benefit of the user under different response strategies:
;
Wherein,Representing electricity price;
the incentive rewards of the users participating in the demand response are evaluated, and the incentive rewards comprise electric charge discounts and point rewards.
Through the specific steps, a user response model can be established based on the user response data, the behavior change of the user under different electricity prices and demand response signals is predicted by using a Gradient Boost Tree (GBT) algorithm, the energy storage charging and discharging behaviors of the user under different situations are quantized, and a scientific basis and an optimization strategy are provided for the resource regulation of the distributed energy storage network.
The construction of the dynamic optimization model aiming at minimizing the energy transmission loss and the charge-discharge loss specifically comprises the following steps:
the optimization objective function is as follows:
the optimization objective function is as follows:
;
Wherein,Is the loss power of the ith energy transmission path,Is the transmission time of the ith energy transmission path,Is the charging loss power of the jth energy storage device,Is the charging time of the jth energy storage device,Is the discharge loss power of the kth energy storage device,N, M, L is the number of the energy transmission path, the charged energy storage device and the discharged energy storage device respectively;
Constraint conditions:
User response behavior constraints-user behavior data predicted using a user response model, including response probabilities at different electricity prices and demand response signalsAnd response volumeIncorporating the user response data into an optimization model, defining user response behavior constraints:
;
;
Wherein,AndThe response probabilities of the energy storage devices j and k in the current context,、Respectively the maximum charge and discharge of the energy storage device,Representing the expected energy of the energy storage device j in a charging operation,Representing the expected energy of the energy storage device k in a discharging operation;
Energy balance constraint: Wherein, the method comprises the steps of,、、Energy transmission, charge and discharge, respectively;
Energy storage device capacity constraints:
;
;
Wherein,AndThe upper limits of the charge and discharge capacities of the energy storage device, respectively;
the solving of the optimization problem using a distributed algorithm specifically includes:
The problem decomposition is to decompose the global optimization problem into a plurality of sub-problems, wherein each sub-problem corresponds to one energy storage device or transmission line, and the decomposition is carried out by utilizing ADMM (Alternating Direction Method of Multipliers) algorithm;
solving the sub-problems, namely independently solving each sub-problem locally, and calculating corresponding charging power, discharging power and transmission power, wherein the method comprises the following steps of:
;
;
;
Wherein,Is the maximum transmission power of the energy transmission;
global coordination, namely summarizing the solving result of each sub-problem through a global coordinator, updating global variables, and ensuring the satisfaction of global energy balance and constraint conditions;
and (3) in each round of iteration, exchanging information between the sub-problems and the global coordinator, updating respective optimization variables, gradually converging to a global optimal solution, and adopting an iterative updating formula as follows:
;
;
。
Wherein,Is a lagrange multiplier, x, z represent variables, c is a vector representing the right constant term of the constraint, E and B are matrices for mapping variables x and z into the constraint,Is a regularization parameter which is a function of the data,、Is an objective function with respect to variables x and z.
Through the steps, a dynamic optimization model aiming at minimizing energy transmission loss and charge and discharge loss can be constructed, and the optimization problem is solved by using a distributed algorithm, so that the efficient regulation and optimization of the distributed energy storage network are realized.
The dynamic energy routing model construction in S34 specifically includes:
S341, model input, wherein the model input data comprises an energy demand prediction result based on the output result of the energy demand prediction and dynamic optimization model, and the model input data comprises predicted energy demand, user response behavior prediction data and an optimization strategy for minimizing energy transmission loss and charge-discharge lossUser response behavior prediction dataAndOutput strategy of dynamic optimization modelAnd;
Constructing an energy routing model, defining a selection and scheduling strategy of a transmission path, and comprising the following steps:
Defining nodes and edges in a distributed energy storage network, establishing a network topology structure, wherein the nodes comprise energy storage equipment, power generation equipment and a load center, and the edges comprise energy transmission lines;
the path selection algorithm is used for calculating an optimal energy transmission path based on the path selection algorithm;
calculating the weight of a transmission path according to the energy demand prediction and dynamic optimization results, wherein the weight comprises energy transmission loss and transmission time;
S342, dynamically adjusting an energy transmission path:
real-time data acquisition, namely acquiring data of the running states of the energy storage system and the power grid in real time through a distributed sensor network, wherein the data comprise voltage, current, power, energy requirements and user responses;
a path adjustment algorithm, which dynamically adjusts the energy transmission path based on the data collected in real time and the energy demand prediction result;
S343, optimizing a charging and discharging strategy of the energy storage system:
Based on the energy demand prediction and the user response behavior model, optimizing the charge-discharge strategy of the energy storage system specifically comprises the following steps:
the method comprises the steps of carrying out calibration on an energy demand prediction result by utilizing real-time data, improving the accuracy of prediction, dynamically adjusting the charging and discharging time and power of an energy storage system according to a charging and discharging strategy output by an optimization model so as to minimize energy loss and economic cost;
s344, outputting the optimized energy transmission path and the energy storage regulation strategy:
Outputting the optimized energy transmission path, ensuring energy to be transmitted to a demand node to meet load demands, and issuing the optimized energy storage regulation strategy to each energy storage device to monitor the execution condition of the energy storage device in real time.
The network topology is expressed as: wherein V represents a set of nodes and E represents a set of edges;
the path selection algorithm uses a shortest path algorithm to calculate the optimal energy transmission path:
Wherein, the method comprises the steps of, wherein,Is a sideIs (e.g. energy loss),Is a path selection variable (if edgesIn the path then 1, otherwise 0;
Path weight calculation: Wherein, the method comprises the steps of, wherein,AndAs the weight coefficient of the light-emitting diode,In order to transfer the lost power of the energy,Is the energy transmission time;
The path adjustment algorithm is updated in real time based on the path, expressed as the formula:
Wherein, the method comprises the steps of, wherein,Is the path weight at the time t,AndTransmission loss power and transmission time at time t;
And (3) self-adaptive adjustment: Wherein, the method comprises the steps of, wherein,Is to adjust the step size of the step,Is the transmission loss power at time t +1,Is the actual demand at the time t,Is the energy forecast demand at time t.
Optimizing a charge-discharge strategy of the energy storage system, and optimizing a formula of charge-discharge:
;
Constraint conditions:
;
;
Predictive calibration: real-time data calibration formula:
Wherein, the method comprises the steps of, wherein,Is the calibration coefficient of the device,Is the actual energy demand at time t;
outputting an optimized energy transmission path and an energy storage regulation strategy:
Path output:
The optimal path formula:;
And (3) strategy release:
And (3) outputting an energy storage regulation strategy:
;
。
Wherein,、Charging and discharging strategies are respectively adopted.