CN105137265A - Insulator leakage current prediction method - Google Patents

Abstract

The invention relates to an insulator leakage current prediction method, comprising a leakage current measurement system, a humidity measurement system and an operation voltage measurement system, and establishing a nerve network model using equivalent impedance and relative humidity as input, and the equivalent impedance under saturated humidity as output; the initial data collected by the leakage current measurement system, the humidity measurement system and the operation voltage measurement system is used as the training sample of a nerve network; the leakage current value under non-saturated humidity is calculated to the leakage current value under saturated humidity. According to the invention, the nerve network model using equivalent impedance and relative humidity as input, and the equivalent impedance under saturated humidity as output is built to research the relation between moisture and leakage current, and to predict insulator leakage current.

Description

Insulator Leakage Current Forecasting Methodology

Technical field

The invention belongs to technical field of electric system protection, relate to the curent change of insulator under the state of making moist, be specifically related to a kind of insulator Leakage Current Forecasting Methodology.

Background technology

Criterion of Polluted Insulator external insulating strength under the condition of making moist can sharply decline, under working voltage, just likely pollution flashover can occur.Pollution flashover accident often causes large area, has a power failure for a long time, causes huge economic loss.By the research to Mechanism of Contamination Flashover, find that the evolution that leakage current and pollution flashover discharge is closely related, and leakage current is convenient to continuous monitoring, contain abundant information, concentrated expression pollution level, damp degree, insulator can bear the impact of the factor such as voltage and insulator shape.The leakage current change of insulator simultaneously reflects the cumulative change process of insulator contamination, so the filthy leakage current of insulator is for judging that external insulation state has great importance.

Summary of the invention

In order to overcome the deficiencies in the prior art, the invention provides a kind of insulator Leakage Current Forecasting Methodology, by set up with equiva lent impedance and relative humidity be input, with the neural network model of the equiva lent impedance under saturated humidity for exporting, the relation of research wetness and Leakage Current, predicts insulator Leakage Current.

Technical scheme of the present invention is: a kind of insulator Leakage Current Forecasting Methodology, comprise Leakage Current measuring system, humidity measurement system and working voltage measuring system, setting up with equiva lent impedance and relative humidity is input, with the neural network model of the equiva lent impedance under saturated humidity for exporting, the primary data that Leakage Current measuring system, humidity measurement system and working voltage measuring system gather as the training sample of neural network, by the leakage current values of leakage current values reduction under saturated humidity under unsaturation humidity.Described equiva lent impedancefor:wherein I_hfor Leakage Current maximal value, U_rfor insulator working voltage, L is the total leakage distance of insulator, and the implication of equiva lent impedance is under working voltage, when maximum Leakage Current flows through insulator surface, and the average resistance on its unit Leakage Current.Described neural network comprises input layer, hidden layer and output layer, and the transport function of described hidden layer adopts S type function---hyperbolic tangent function,described neural network adopts LM fast algorithm to train, and weighed value adjusting rate elects Δ w=(J as^tj+ μ I)^-1j^te, in formula, J is the Jacobian matrix of error to weights differential, and e is error vector, and μ is self-adaptative adjustment scalar.Described neural network comprises two hidden layers, and hidden layer neuron number is all 8.The input equiva lent impedance of described neural network is less than 3.3M Ω/m.Described training sample adopts 800 points, and wherein random selecting 770 data are trained neural network, and other 30 data points are as the verification msg of neural network.

The present invention has following good effect: be input by setting up with equiva lent impedance and relative humidity, with the neural network model of the equiva lent impedance under saturated humidity for exporting, study the relation of wetness and Leakage Current, predicting insulator Leakage Current.

Accompanying drawing explanation

Fig. 1 is specific embodiment of the invention neural network structure model.

Embodiment

Contrast accompanying drawing below, by the description to embodiment, the specific embodiment of the present invention is as the effect of the mutual alignment between the shape of involved each component, structure, each several part and annexation, each several part and principle of work, manufacturing process and operation using method etc., be described in further detail, have more complete, accurate and deep understanding to help those skilled in the art to inventive concept of the present invention, technical scheme.

The present invention is monitored by damping device, air extractor and the measurement instrument humidity to test box house and is adjusted, by changing the leakage current that moisture measurement different surfaces makes moist under state, high pressure accesses from below, for preventing the flashover of high-voltage connection under high humility, the bonding organic insulating sheath in high-voltage connection rod surface.Ground wire connects leakage current measurement system, in order to measure leakage current from the top of organic glass casing.By the data that the record training sample as neural network of the present invention.Training sample adopts 800 points, and wherein random selecting 770 data are trained neural network, and other 30 data points are as the verification msg of neural network.

Due to the insulator in actual motion, its working voltage, the parameters such as leakage distance are all different, in order to make result of study can be applied to insulator under different running status, are recording I_hbasis on, consider the impact of working voltage and insulator leakage distance, a new parameter can be introduced, be defined as equiva lent impedancewherein I_hfor Leakage Current maximal value, U_rfor insulator working voltage, L is the total leakage distance of insulator, and the implication of equiva lent impedance is under working voltage, when maximum Leakage Current flows through insulator surface, and the average resistance on its unit Leakage Current.When maximum leakage electric current flows through insulator surface, average resistance on its unit leakage distance, its concentrated expression U_r, I_hwith the acting in conjunction of L.

The object of this neural network is by the leakage current values of leakage current values reduction under saturated humidity under unsaturation humidity, therefore, that chooses is input as equiva lent impedance (r) and relative humidity (HR), using the equiva lent impedance under saturated humidity as output.Because small area analysis is easily disturbed, External Insulation state estimation has little significance, and therefore, the input equiva lent impedance of neural network is less than 3.3M Ω/m.

Choosing of artificial neural network parameter has important impact to the performance of network and training speed.First the transport function of hidden layer neuron will be determined.Sigmoid function has had the non-linear behavior required for classification, have again realize needed for LMS (LeastMeanSquare) learning algorithm can micro-characteristic, simultaneously the input-output characteristic of sigmoid function also relatively human brain, has better bionical effect.Therefore this project neural network hidden layer have employed a kind of conventional S shape transport function---hyperbolic tangent function, and its expression formula is as follows:

The improvement of BP algorithm mainly contains 2 approach, and one is adopt didactic learning algorithm, and another kind adopts more effective optimized algorithm.The neural network of this project adopts Levenberg-Marquardt fast algorithm to train.This algorithm have employed the algorithm of self-adaptative adjustment learning rate, and faster than the speed of other gradient algorithm is many, but needs more internal memory.The weighed value adjusting rate of this algorithm elects Δ w=(J as^tj+ μ I)^-1j^te, in formula, J is the Jacobian matrix of error to weights differential, and e is error vector, and μ is self-adaptative adjustment scalar.

Hidden layer and neuronic number have larger impact for the performance of network, when a hidden layer, at neuron number less (5,8,10), time, the misdiagnosis rate of network is comparatively large, and when neuron number is 15, neural network performance is more excellent, neuron number rises to 20h, and network performance is deteriorated again; When two hidden layers, the training error of network and misdiagnosis rate are all than less when only having a hidden layer, and when two hidden layer neuron numbers are all 8, network performance reaches optimum, so the structure of the artificial neural network of book foundation is two hidden layers, hidden layer neuron number is all 8.

Above by reference to the accompanying drawings to invention has been exemplary description; obvious specific implementation of the present invention is not subject to the restrictions described above; as long as have employed the improvement of the various unsubstantialities that method of the present invention is conceived and technical scheme is carried out; or design of the present invention and technical scheme directly applied to other occasion, all within protection scope of the present invention without to improve.

Claims (7)

1. an insulator Leakage Current Forecasting Methodology, it is characterized in that, comprise Leakage Current measuring system, humidity measurement system and working voltage measuring system, setting up with equiva lent impedance and relative humidity is input, with the neural network model of the equiva lent impedance under saturated humidity for exporting, the primary data that Leakage Current measuring system, humidity measurement system and working voltage measuring system gather as the training sample of neural network, by the leakage current values of leakage current values reduction under saturated humidity under unsaturation humidity.

2. insulator Leakage Current Forecasting Methodology according to claim 1, is characterized in that, described equiva lent impedancefor:wherein I_hfor Leakage Current maximal value, U_rfor insulator working voltage, L is the total leakage distance of insulator, and the implication of equiva lent impedance is under working voltage, when maximum Leakage Current flows through insulator surface, and the average resistance on its unit Leakage Current.

3. insulator Leakage Current Forecasting Methodology according to claim 2, it is characterized in that, described neural network comprises input layer, hidden layer and output layer, and the transport function of described hidden layer adopts S type function---hyperbolic tangent function,$f\left(x\right)=\tanh \left(x\right)=\frac{1-e^{-x}}{1+e^{-x}}.$

4. insulator Leakage Current Forecasting Methodology according to claim 2, is characterized in that, described neural network adopts LM fast algorithm to train, and weighed value adjusting rate elects Δ w=(J as^tj+ μ I)^-1j^te, in formula, J is the Jacobian matrix of error to weights differential, and e is error vector, and μ is self-adaptative adjustment scalar.

5. insulator Leakage Current Forecasting Methodology according to claim 3, is characterized in that, described neural network comprises two hidden layers, and hidden layer neuron number is all 8.

6. insulator Leakage Current Forecasting Methodology according to claim 2, is characterized in that, the input equiva lent impedance of described neural network is less than 3.3M Ω/m.

7. insulator Leakage Current Forecasting Methodology according to claim 6, is characterized in that, described training sample adopts 800 points, and wherein random selecting 770 data are trained neural network, and other 30 data points are as the verification msg of neural network.