A method and system for evaluating the state of a distribution network cable based on a long short-term memory network

Abstract

The application discloses a kind of long short-term memory network-based distribution network cable state evaluation method and system, it is related to distribution network cable evaluation technical field, by collecting and distribution network cable operation insulation state related historical characteristic parameters and the historical insulation state of corresponding distribution network cable;Weighted Bagging algorithm is used to fuse historical characteristic parameters, and obtain weighted fusion characteristic parameters;Long short-term memory network-based distribution network cable insulation state evaluation model is constructed;Based on the weighted fusion characteristic parameters and the historical insulation state of corresponding distribution network cable, distribution network cable insulation state evaluation model is trained, the characteristic parameters of the distribution network cable to be evaluated are input into the distribution network cable insulation state evaluation model, and evaluation result is obtained.The application real-time acquisition of the characteristic parameters of distribution network cable is input into the distribution network cable insulation state evaluation model trained, and the state of distribution network cable can be automatically evaluated, to realize the all-weather, online monitoring of distribution network cable.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of distribution network cable evaluation, and more particularly to a distribution network cable state evaluation method and system based on a long short-term memory network. BACKGROUND

[0002] At present, cables play an extremely important role in the transmission of electric energy in the power system, and their operating state directly affects the safety and stability of the power system. The design life of a cable is generally 20 to 30 years. As the operating life of a cable increases, its insulation will gradually age, and the overall insulation state will gradually deteriorate. For a cable that has been in operation for many years, the overall insulation is seriously aged. At the same time, if the operating environment of the cable is poor, such as being wet, containing corrosive substances, and being severely radiated, the overall insulation will age more rapidly. At this time, the operating life of the cable is likely to be shorter than the designed service life. Therefore, the evaluation of the insulation state of the cable is particularly important. If the insulation state of the cable cannot be well evaluated to accurately estimate the replacement time, cable failure will easily occur, which will lead to the shutdown or even loss of control of a large electrical system, causing serious economic losses and social impacts.

[0003] With the development of smart grid technology, the insulation state of a distribution network cable can be evaluated to understand the insulation condition of the cable. In the traditional way, the insulation state of the distribution network cable is evaluated by online monitoring. Many cities have lagged behind in municipal planning, and the cable laying environment is poor. In this environment, the online monitoring method for evaluating the insulation state of the distribution network cable will lead to inaccurate evaluation results.

[0004] Therefore, how to improve the evaluation accuracy of the insulation state of the distribution network cable is a problem that needs to be solved by those skilled in the art. SUMMARY

[0005] Therefore, the present application provides a distribution network cable state evaluation method and system based on a long short-term memory network to solve the problems in the background art.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0007] On the one hand, a distribution network cable state evaluation method based on a long short-term memory network is provided, comprising:

[0008] Collecting historical feature parameters related to the insulation state of the distribution network cable and the historical insulation state of the corresponding distribution network cable;

[0009] Fusing the historical feature parameters by using a weighted Bagging algorithm to obtain fused feature parameters after weighting;

[0010] The long short-term memory network-based distribution network cable insulation state evaluation model is constructed;

[0011] The distribution network cable insulation state evaluation model is trained based on the weighted fused feature parameters and the historical insulation states of the corresponding distribution network cables, and a trained distribution network cable insulation state evaluation model is obtained.

[0012] The feature parameters of the distribution network cable to be evaluated are input into the distribution network cable insulation state evaluation model, and an evaluation result is obtained.

[0013] Optionally, the historical feature parameters are relative temperature rises of surface temperatures of the distribution network cables and ambient temperatures and partial discharge signals of the distribution network cables.

[0014] Optionally, the specific process of training the distribution network cable insulation state evaluation model comprises:

[0015] The LSTM network structure is initialized: the number of LSTM hidden units is set to numHiddemUnits=40, the number of categories is set to numClass=4, the feature dimension is set to D=14, the maximum number of iterations is set to maxEpoch=40, the batch size is set to miniBatchSize=512, the random gradient descent is selected as the optimization algorithm, and the cross-entropy loss is selected as the loss function.

[0016] The weighted fused feature parameters and the historical insulation states of the corresponding distribution network cables are input into the LSTM network.

[0017] The LSTM model is started to be trained, and the training process is an iteration process. In each iteration, the cross-entropy loss is calculated through forward propagation, the objective function is minimized, and the model parameters are updated through back propagation.

[0018] Optionally, before the historical feature parameters are fused by using the weighted Bagging algorithm, the historical feature parameters related to the insulation states of the distribution network cables and the historical insulation states of the corresponding distribution network cables are cleaned by using a stacked deep auto-encoder network, and the feature parameter information of the distribution network cables is processed by using the deep auto-encoder.

[0019] Optionally, the method further comprises:

[0020] The feature parameters of the distribution network cable to be evaluated are collected in real time by using a sensor;

[0021] The collected feature parameters are input into the trained stacked deep auto-encoder network to be denoised and processed by dimension reduction, and clean and low-dimensional feature parameters are obtained.

[0022] The processed feature parameter data is input into the trained distribution network cable insulation state evaluation model to complete evaluation of the distribution network cable insulation state.

[0023] Optionally, the method further comprises evaluating the performance of the distribution network cable insulation state evaluation model by:

[0024] Q(F) = Quality(F) TestData

[0025] wherein Testdata is a test data set, and Q(F) represents an AUC value of the distribution network cable insulation state evaluation model F on the test data set.

[0026] In another aspect, a distribution network cable state evaluation system based on a long short-term memory network is provided, comprising:

[0027] a collection module configured to collect historical feature parameters related to the insulation state of the distribution network cable and historical insulation states of the corresponding distribution network cable;

[0028] a fusion module configured to fuse the historical feature parameters by using a weighted Bagging algorithm to obtain weighted fused feature parameters;

[0029] a model construction module configured to construct a distribution network cable insulation state evaluation model based on a long short-term memory network;

[0030] a model training module configured to train the distribution network cable insulation state evaluation model based on the weighted fused feature parameters and the historical insulation states of the corresponding distribution network cable to obtain a trained distribution network cable insulation state evaluation model;

[0031] an evaluation module configured to input feature parameters of a distribution network cable to be evaluated into the distribution network cable insulation state evaluation model to obtain an evaluation result.

[0032] Optionally, the system further comprises an information cleaning module configured to clean the collected historical feature parameters related to the insulation state of the distribution network cable and the historical insulation states of the corresponding distribution network cable by using a stack-type deep auto-encoder network, and to perform dimension reduction processing on the feature parameter information of the distribution network cable by using the deep auto-encoder.

[0033] Compared with the prior art, the method and system for evaluating the state of distribution network cables based on a long short-term memory network are provided, historical characteristic parameters related to the insulation state of the distribution network cables and historical insulation states of the corresponding distribution network cables are collected, the historical characteristic parameters are fused by using a weighted Bagging algorithm to obtain weighted fused characteristic parameters, a distribution network cable insulation state evaluation model based on a long short-term memory network is constructed, the distribution network cable insulation state evaluation model is trained based on the weighted fused characteristic parameters and the historical insulation states of the corresponding distribution network cables to obtain a trained distribution network cable insulation state evaluation model, and the characteristic parameters of the distribution network cable to be evaluated are input into the distribution network cable insulation state evaluation model to obtain an evaluation result. The characteristic parameters of the distribution network cable are collected in real time and input into the trained distribution network cable insulation state evaluation model, so that the state of the distribution network cable can be automatically evaluated, and all-weather and online monitoring of the distribution network cable is realized. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.

[0035] Figure 1 The drawings are the method flowchart provided by the present application. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0037] The embodiment of the present application discloses a method for evaluating the state of distribution network cables based on a long short-term memory network, as shown in Figure 1 The method comprises the following steps:

[0038] Collecting historical characteristic parameters related to the insulation state of the distribution network cables and historical insulation states of the corresponding distribution network cables;

[0039] Fusing the historical characteristic parameters by using a weighted Bagging algorithm to obtain weighted fused characteristic parameters;

[0040] Constructing a distribution network cable insulation state evaluation model based on a long short-term memory network;

[0041] training the distribution network cable insulation state evaluation model based on the weighted fused feature parameters and the historical insulation state of the corresponding distribution network cable, to obtain the trained distribution network cable insulation state evaluation model;

[0042] inputting the feature parameters of the distribution network cable to be evaluated into the distribution network cable insulation state evaluation model, to obtain an evaluation result.

[0043] wherein the output of the long short-term memory network is weighted and combined by the Bagging algorithm, specifically:

[0044]

[0045] wherein y j is the jth historical feature parameter, and is the jth weak learner in the Bagging algorithm, λ j is the weight of each weak learner, y is the final equalizer output, and T is the number of weak learners. The weight λ j may be set according to the accuracy of each weak learner, as follows:

[0046]

[0047] wherein e j represents the error rate of the jth weak learner, e max and e min represent the maximum and minimum values of the error rates of all weak learners. The weight λ j is updated each time the historical feature parameters related to the insulation state of the distribution network cable and the historical insulation state of the corresponding distribution network cable are collected.

[0048] In one specific embodiment, the historical feature parameters are the relative temperature rise of the surface temperature of the distribution network cable and the ambient temperature, and the partial discharge signal of the distribution network cable.

[0049] In one specific embodiment, the specific process of training the distribution network cable insulation state evaluation model includes:

[0050] The LSTM network can process data with input samples as sequences, and the length of the sample sequence is variable. Compared with general recurrent network models, LSTM solves the problems of gradient disappearance and gradient explosion in the training process. For non-sequence data input, LSTM can also be used as a classifier, and its effect is usually better than traditional classifiers such as KNN, SVM, etc. First, the fused feature parameters are input into the LSTM as the input of the LSTM, and the final result is output through the LSTM cell. Then, a fully connected layer and a Soft max layer are used, and finally a vector with a length of 4 (the number of categories) is output.

[0051] The basic LSTM network structure used in the present application is defined as follows: in an LSTM unit, the LSTM accepts the current fusion feature parameter f i as input, and outputs the hidden state and the cell state as the input of the next stage, where d is the dimension of the output vector. The LSTM network contains three gates: the input gate g i , the forget gate g f and the output gate g o , as well as a control gate g g , W i , W f , W o , W g are the corresponding weight matrices respectively. Denote H i as the concatenation of the current stage input feature vector f i and the previous stage output hidden state h i , then the output hidden state h i+1 and the cell state c i+1 can be updated by the following formulas:

[0052] g i = σ(W u *H i );

[0053] g f = σ(W f *H i );

[0054] g o = σ(W o *H i );

[0055] g g = tanh(W c *H i );

[0056] c i+1 = g f ⊙c i + g i ⊙g g ;

[0057] h i+1 = g o ⊙tanh(c i );

[0058] In the formula, σ represents a sigmoid function, and represents multiplication of corresponding elements. In order to express the above formula in a simpler form, W is recorded as a combination of the four weight matrices, and the above formula is simplified in the form of LSTM(·) into the form of the following formula:

[0059] (h i+1 ,c i+1 )=LSTM(H i ,c i ,W);

[0060] In the present embodiment, when the LSTM network structure is initialized, the number of LSTM hidden units numHiddemUnits is set to 40, the number of classes numClass is set to 4, the feature dimension D is set to 14, the maximum number of iterations maxEpoch is set to 40, the batch size miniBatchSize is set to 512, and the stochastic gradient descent (SGD) is selected as the optimization algorithm, and the cross-entropy loss is selected as the loss function. The training process is an iterative process; each iteration calculates the cross-entropy loss by forward propagation, minimizes the objective function, and updates the model parameters by back propagation, and finally obtains the LSTM classification model.

[0061] In a specific embodiment, before the historical feature parameters are fused by using the weighted Bagging algorithm, the historical feature parameters related to the insulation state of the distribution network cable and the historical insulation state of the corresponding distribution network cable collected are cleaned by constructing a stacked deep autoencoder network, and the feature parameter information of the distribution network cable is processed by dimension reduction by using the deep autoencoder.

[0062] Specifically, the stacked deep autoencoder network is constructed to clean the historical feature parameters related to the insulation state of the distribution network cable and the historical insulation state of the corresponding distribution network cable collected, the historical data of the distribution network cable is constructed into a matrix X, X = {x(1), x(2), x(3), … x(N)}, x(i) ∈ R M , and a certain amount of "damage noise" is added to the matrix X to obtain noise-containing data χ, satisfying x ~ qD(χ|X), qD being a noise distribution form, i.e., the "damage noise" is added according to qD distribution, and y is obtained by encoding χ based on the autoencoder network, so that y can approach or reconstruct the original input X.

[0063] The specific process is as follows:

[0064] (1) Construct a deep autoencoder model, set the network learning rate ε, the sparse parameter ρ parameter, and randomly initialize the model connection weight W and the offset b;

[0065] (2) Set the batch training number and the number of iterations in the forward propagation algorithm, execute the forward propagation algorithm, and calculate the average activation amount ρj ;

[0066] (3) constructing a loss function based on the input and output of the deep auto-encoder;

[0067] (4) performing a back propagation method, updating network weight parameters W and b based on a stochastic gradient descent algorithm, completing the training of the deep auto-encoder, realizing the noise reduction and dimensionality reduction of the distribution network equipment operation and maintenance information.

[0068] In one specific embodiment, further comprising:

[0069] Real-time acquisition of characteristic parameters of the distribution network cable to be evaluated by using a sensor;

[0070] The collected characteristic parameters are input into the trained stack deep auto-encoder network for noise reduction and dimensionality reduction processing to obtain clean and low-dimensional characteristic parameters;

[0071] The processed characteristic parameter data is input into the trained distribution network cable insulation state evaluation model to complete the evaluation of the insulation state of the distribution network cable.

[0072] In one specific embodiment, comprising: further comprising evaluating the performance of the distribution network cable insulation state evaluation model by:

[0073] Q(F) = Quality(F) TestData

[0074] Wherein, TestData is a test data set, and Q(F) represents the AUC value of the distribution network cable insulation state evaluation model F on the test data set.

[0075] On the other hand, a distribution network cable state evaluation system based on a long short-term memory network is provided, comprising:

[0076] The acquisition module acquires historical characteristic parameters related to the insulation state of the distribution network cable and the historical insulation state of the corresponding distribution network cable;

[0077] The fusion module fuses the historical characteristic parameters by using a weighted Bagging algorithm to obtain weighted fused characteristic parameters;

[0078] The model construction module constructs a distribution network cable insulation state evaluation model based on a long short-term memory network;

[0079] The model training module trains the distribution network cable insulation state evaluation model based on the weighted fused characteristic parameters and the historical insulation state of the corresponding distribution network cable to obtain a trained distribution network cable insulation state evaluation model;

[0080] The evaluation module inputs the characteristic parameters of the power distribution cable to be evaluated into a power distribution cable insulation state evaluation model to obtain an evaluation result.

[0081] In one specific embodiment, an information cleaning module is further included, and the stack type deep auto-encoder network is configured to clean the collected historical characteristic parameters related to the insulation state of the power distribution cable and the historical insulation state of the corresponding power distribution cable, and the deep auto-encoder is used to perform dimension reduction processing on the characteristic parameter information of the power distribution cable.

[0082] The various embodiments are described in the present specification in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts of the various embodiments can be mutually referred to. For the apparatus disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0083] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A distribution network cable status assessment method based on long short-term memory network, characterized in that: include: Collect historical characteristic parameters related to the insulation status of distribution network cables and the historical insulation status of the corresponding distribution network cables; The historical feature parameters are fused using a weighted Bagging algorithm to obtain weighted fused feature parameters; Construct a distribution network cable insulation status assessment model based on long short-term memory network; Based on the weighted fusion feature parameters and the historical insulation status of the corresponding distribution network cables, a distribution network cable insulation status assessment model is trained to obtain a trained distribution network cable insulation status assessment model; Inputting characteristic parameters of the distribution network cable to be evaluated into the distribution network cable insulation status evaluation model to obtain an evaluation result; The specific process of training the distribution network cable insulation condition assessment model includes: Initialize the LSTM network structure: set the number of LSTM hidden units numHiddemUnits = 40, the number of classes numClass = 4, the feature dimension D = 14, the maximum number of iterations maxEpoch = 40, the batch size miniBatchSize = 512, and select stochastic gradient descent as the optimization algorithm and cross entropy loss as the loss function; use the weighted fusion feature parameters and the historical insulation status of the corresponding distribution network cable as inputs to the LSTM network and start training the LSTM model. The training process is an iterative process. In each iteration, the cross entropy loss is calculated through forward propagation to minimize the objective function, and the model parameters are updated through backpropagation. In an LSTM unit, LSTM accepts the current fusion feature parameters f i As input, and output hidden state and cell status As input to the next stage, d is the output vector dimension; the LSTM network contains three gates: input gate , the Forget Gate and output gate , and a control door , are the corresponding weight matrices respectively; H i Input feature vector from the current stage f i And the hidden state output of the previous stage h i Spliced ​​together, then the hidden state of the output h i+1 and cell status c i+1 Updated by the following formula: c i+1 ＝ g f ⊙ c i + g i ⊙ g g ; h i+1 = g o ⊙tanh( c i ); Where σ represents the sigmoid function, ⊙ represents the multiplication of corresponding elements; Before using the weighted Bagging algorithm to fuse the historical characteristic parameters, the method also includes constructing a stacked deep autoencoder network to clean the collected historical characteristic parameters related to the insulation status of the distribution network cable and the historical insulation status of the corresponding distribution network cable, and using the stacked deep autoencoder network to perform dimensionality reduction processing on the distribution network cable characteristic parameter information; Also includes: Use sensors to collect the characteristic parameters of the distribution network cables to be evaluated in real time; Inputting the collected feature parameters into the trained stacked deep autoencoder network for noise reduction and dimensionality reduction to obtain clean, low-dimensional feature parameters; The processed feature parameter data is input into the trained distribution network cable insulation status assessment model to complete the assessment of the distribution network cable insulation status; The performance of the distribution network cable insulation condition assessment model is evaluated by: Among them, TestData is the test data set, and Q(F) represents the AUC value of the distribution network cable insulation status assessment model F on the test data set.

2. A distribution network cable status assessment method based on long short-term memory network according to claim 1, characterized in that: The historical characteristic parameters are the relative temperature rise between the surface temperature of the distribution network cable and the ambient temperature and the partial discharge signal of the distribution network cable.

3. A distribution network cable status assessment system based on long short-term memory network, characterized in that: A distribution network cable status assessment method based on a long short-term memory network according to any one of claims 1 to 2 is applied, comprising: The acquisition module collects historical characteristic parameters related to the insulation status of the distribution network cables and the historical insulation status of the corresponding distribution network cables; A fusion module uses a weighted Bagging algorithm to fuse the historical feature parameters to obtain weighted fusion feature parameters; Model building module, which builds a distribution network cable insulation status assessment model based on long short-term memory network; The model training module trains the distribution network cable insulation status assessment model based on the weighted fusion feature parameters and the historical insulation status of the corresponding distribution network cables to obtain a trained distribution network cable insulation status assessment model; The evaluation module inputs the characteristic parameters of the distribution network cable to be evaluated into the distribution network cable insulation status evaluation model to obtain an evaluation result.

4. A distribution network cable status assessment system based on long short-term memory network according to claim 3, characterized in that: It also includes an information cleaning module, which constructs a stacked deep autoencoder network to clean the collected historical characteristic parameters related to the insulation status of the distribution network cable and the historical insulation status of the corresponding distribution network cable, and uses the stacked deep autoencoder network to perform dimensionality reduction processing on the characteristic parameter information of the distribution network cable.

Images