Cable joint fault prediction method based on temperature difference long short-term memory network learning

Abstract

The application provides a cable joint fault prediction method based on temperature difference long short-term memory network learning, first, the temperature of multiple position points of a cable intermediate joint accessory is collected, and the temperature difference values of the position points and the environment temperature are calculated; second, the time sequence temperature difference values of the multiple position points are taken as model input, and a fault prediction model based on a long short-term memory network is established; then, a data set is constructed to train the model, and a multiverse optimization algorithm is used to optimize the key parameters of the model; finally, the time sequence temperature difference values of multiple position points of intermediate joints of multiple power cables are collected, and the established model is used to predict the intermediate joint fault.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of cable joint fault prediction identification and machine learning, and particularly relates to a cable joint fault prediction method based on temperature difference long short-term memory network learning. BACKGROUND

[0002] With the rapid development of power systems and the implementation of old city reconstruction projects, power cables are increasingly widely used in power networks. Power cables play a crucial role in the entire distribution network, especially in cities where land is scarce. In view of the environmental aesthetics, power cables are used to connect the entire distribution network. In the process of using power cables, the cable joint is the weakest link in the distribution network and also the device with the highest failure rate. The contact resistance of the cable joint is large, and it is in a closed space with poor heat dissipation conditions. When the cable is overloaded, the temperature of the cable joint rises rapidly. The high temperature will cause the insulation layer of the cable joint to age rapidly, the contact resistance will increase, the temperature will rise, and a vicious cycle will be formed, which will cause the cable to ground or short circuit, and even cause the electrical equipment to burn, fire, and large-scale power outage, causing serious accidents and threatening the safe operation of the power grid. Therefore, detecting the temperature change of the cable joint can determine whether a fault has occurred in time, which is of great significance to improving the stability of the power system.

[0003] At present, in order to ensure the safe operation of the cable joint, the following methods are usually used for monitoring:

[0004] The first method is to paint the cable joint. Color paint is applied to the cable joint, and then the color change of the color paint is observed to determine the heating condition. When the color of the color paint becomes darker and the paint peels off, it indicates that the joint has overheated. It should be noted that sometimes the color of the color paint changes little, which can easily lead to misjudgment, so this method should be used with caution.

[0005] The second method is to paste a temperature test sheet on the cable joint. The temperature test sheet is used to observe whether the electrical joint is overheating. The temperature test sheet has three colors: red 80℃, green 70℃, and yellow 60℃. When the temperature test sheet melts, it indicates that the joint has overheated. The temperature test sheet is more intuitive and is widely used, but the disadvantage is that the temperature test sheet is often shaken off when the switch is opened and closed.

[0006] The third is to install temperature sensors and data collectors in the cable connector, upload the collected temperature data to the background server through wireless transmission, display it in combination with the background, and the background server judges the collected temperature data and alarms when the temperature is higher than the set threshold. However, this method is greatly affected by changes in ambient temperature, and may be misjudged when the surrounding environment is abnormal or in hot summer weather. In addition, temperature online monitoring is used in the cable intermediate joint, and the fault cable is often accompanied by a sudden change in temperature before breakdown, which can be reflected to the remote monitoring platform through the temperature sensor, but this method has the problem that the fault point is known, but the fault has often developed very seriously at this time, and the intelligent monitoring intervention and manual intervention are too late, and it does not have the function of early warning.

[0007] At present, the relatively optimal cable intermediate joint fault detection method is to install temperature sensors and data collectors in the cable connector, upload the collected temperature data to the background server through wireless transmission, display it in combination with the background, and the background server judges the collected temperature data and alarms when the temperature is higher than the set threshold. However, this method is greatly affected by changes in ambient temperature, and may be misjudged when the surrounding environment is abnormal or in hot summer weather. In addition, temperature online monitoring is used in the cable intermediate joint, and the fault cable is often accompanied by a sudden change in temperature before breakdown, which can be reflected to the remote monitoring platform through the temperature sensor, but this method has the problem that the fault point is known, but the fault has often developed very seriously at this time, and the intelligent monitoring intervention and manual intervention are too late, and it does not have the function of early warning. SUMMARY

[0008] Therefore, in view of the defects and deficiencies of the prior art, the present application provides a cable intermediate joint fault prediction method based on temperature difference long short-term memory network learning: first, the temperature of multiple position points of the cable intermediate joint accessories is collected, and the temperature difference value of these position points and the ambient temperature is calculated; second, the time series of multiple position point temperature difference values is used as model input, and a fault prediction model based on long short-term memory network is established; then, the data set is constructed to train the model, and the multiverse optimization algorithm is used to optimize the key parameters of the model; finally, the time series temperature difference values of multiple position points of the intermediate joint of multiple power cables are collected, and the established model is used to predict the intermediate joint fault.

[0009] The application adopts time series temperature difference values of multiple position points of cable intermediate joints as data driving of fault prediction, can realize long-term dependence of driving information, automatically mines potential correlation between temperature change and fault, and has good model generalization performance and high prediction accuracy by using a model optimized by a multiverse algorithm.

[0010] The application specifically adopts the following technical solutions:

[0011] A cable joint fault prediction method based on temperature difference long short-term memory network learning has the characteristics that: firstly, the temperatures of multiple position points of a cable intermediate joint accessory are collected, and temperature difference values of the position points and an environment temperature are calculated; secondly, time series multiple position point temperature difference values are taken as model input, and a long short-term memory network-based fault prediction model is established; thirdly, a data set is constructed to train the model, and a multiverse optimization algorithm is used to optimize key parameters of the model; and finally, time series temperature difference values of multiple position points of multiple power cable intermediate joints are collected, and the established model is used to predict intermediate joint faults.

[0012] Further, collection points A1 and B1 are arranged at two ends of the intermediate joint, and collection points A2, A3, A4 and B2, B3, B4 are arranged at symmetric two ends, respectively, the collection points are spaced apart by 50 CM, the temperature difference values of the position points and the environment temperature are taken as the basis of joint temperature change, the collected environment temperature is t0, the temperature values collected at the eight positions are t1-t8, and the temperature difference values of the eight positions are:

[0013] T i =|t i -t0|,i∈[1,8]。

[0014] Further, the long short-term memory network-based fault prediction model includes four parts of data input, deep network, output prediction, and parameter optimization:

[0015] Data input: the time series temperature difference values of the eight positions are taken as the input of the model, x t ={T1,T2,…,T8}, the time series data is standardized by using z-score, then the time series data X is segmented, X={X1,X2,…,X L}, and the segmented L window data is taken as the input of the model;

[0016] Deep network: the network layer adopts LSTM to build a deep learning network, a single-layer LSTM network is used to build the model, the input data is first input into the embedding layer as the input of LSTM, the output of each LSTM layer is used as the input of the next LSTM layer, and the outputs P={P1, P2, …, P L} of all LSTM layers are mapped into a one-dimensional data through a fully connected layer;

[0017] Output prediction: the one-dimensional data mapped by the fully connected layer is activated and output by a Softmax function, and whether a failure occurs within 12 hours in the future is predicted;

[0018] Parameter optimization: in the construction of the joint failure prediction model, the segmentation window length L and the number of LSTM hidden layers S are optimized by using a multiverse optimization algorithm, and the root mean square error RMSE between the model output and the theoretical output is used as the loss function of the model training.

[0019] Further, the method specifically comprises the following steps:

[0020] Step S1: collecting temperature difference time series data of 8 position points around the cable middle joint to construct a data set, using z-score to standardize the time series data, 70% as training data and 30% as test data;

[0021] Step S2: building a failure prediction model based on LSTM and initializing the model parameters, wherein the segmentation window length L and the number of LSTM network hidden layers S are used as the parameters to be optimized:

[0022] Step S3: initializing the multiverse optimization algorithm, initializing L and S, and the maximum number of iterations and the universe position;

[0023] Step S4: model training and prediction based on L and S parameters, the root mean square error RMSE between the model output and the theoretical output is used as the fitness value of the model training, that is, the universe expansion rate;

[0024] Step S5: universe expansion rate calculation, arranging the universe according to the universe expansion rate and selecting a white hole through a roulette mechanism;

[0025] Step S6: updating the wormhole existence probability W EP and the travel distance rate T DR , so as to perform more accurate local search within the obtained global optimal range;

[0026] Step S7: calculating the current universe expansion rate: if the universe expansion rate is better than the current universe expansion rate, updating the current universe expansion rate, otherwise keeping the current universe;

[0027] Step S8: performing individual position update of the universe to find the optimal individual;

[0028] Step S9: termination condition judgment: if the maximum number of iterations is met, output the corresponding L and S parameter values, otherwise increase the iteration number by 1 and return to execute step S4;

[0029] Step S10: establish a fault prediction model based on the optimal L and S parameters and verify it using a test data set;

[0030] Step S11: collect the temperature difference time series data of 8 position points in real time on site and use the established fault prediction model to predict the cable intermediate joint fault.

[0031] Compared with the prior art, the present application and the preferred schemes thereof use the time series temperature difference values of multiple position points of the cable intermediate joint as the data driving for fault prediction, can realize long-term dependence of driving information, automatically mine the potential correlation between temperature change and fault, and have good model generalization performance and high prediction accuracy by using the model optimized by the multiverse algorithm, the method has small influence of environmental temperature, high prediction accuracy, and strong early warning function. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 is an LSTM network structure diagram.

[0033] Figure 2 is an intermediate joint multi-position temperature acquisition schematic diagram in the embodiment of the present application.

[0034] Figure 3 is a cable intermediate joint fault prediction model framework diagram in the embodiment of the present application.

[0035] Figure 4 is a cable intermediate joint fault prediction implementation flowchart in the embodiment of the present application.

[0036] Figure 5 is a model training loss value change curve diagram in the embodiment of the present application.

[0037] Figure 6 is a model prediction accuracy change curve diagram in the embodiment of the present application.

[0038] Figure 7 is a fault early warning effect diagram of different methods in the embodiment of the present application. DETAILED DESCRIPTION

[0039] In the following, specific embodiments of the present application will be described in detail with reference to the accompanying drawings, and those skilled in the art can clearly understand the present application and implement the present application according to these detailed descriptions. The features in each different embodiment can be combined to obtain new implementation manners, or replace some features in some embodiments to obtain other preferred implementation manners without departing from the principles of the present application.

[0040] As Figures 1-7 shown, in order to make the features and advantages of the present patent more obvious and easy to understand, the following specific examples are described in detail as follows:

[0041] 1. Long Short-Term Memory Learning Network (LSTM)

[0042] As a kind of back propagation recurrent neural network, LSTM is used to learn the correlation between long time span data, which improves the simple nodes of traditional neural network into the form of storage unit, and the standard LSTM cell structure is as shown in Figure 1 The input gate, forget gate and output gate in the LSTM model at time t are i t ,f t ,o t respectively; c t and c t-1 are the cell state vectors at time t and t-1 respectively; h t is the prediction value at time t; x t is the input value at time t; σ = 1 / [1+exp(-x)] (sigmoid function) is the activation function of i t ,f t ,o t ; the hyperbolic tangent function tanh(x) = [exp(x)-exp(-x)] / [exp(x)+exp(-x)] is the activation function of c t .

[0043] The control mode in the single unit block in the LSTM model is composed of input gate, forget gate and output gate, and the sigmoid activation unit function is used in each of the three types of gates in the unit block to determine whether they are started, so that the change of model state and the addition of information can selectively flow through the unit block. The control mode of LSTM model structure is beneficial to selectively remember the information on the time series of dam deformation, and meets the needs of long-term and short-term correlation analysis in time series, and the specific implementation formula is as follows:

[0044] i t = σ(W xi x t +W hi h t-1 +W ci c t-1 +b i ), (1)

[0045] f t = σ(W xf x t +W hf h t-1 +Wcf c t-1 +b f (2)

[0046] c t =f t *c t-1 +i t *tanh(W xc x t +W hc h t-1 +b c (3)

[0047] o t＝ σ(W xo x t +W ho h t-1 +W co c t +b o (4)

[0048] h t =o t *tanh(c t ), (5)

[0049] In the formula: W xi W hi W ci W xf W hf W cf W xc W hc W xo W ho W co b is the weighting coefficient; i ,b f ,b c ,b o h is the bias vector. t-1 This is the predicted value at time t-1.

[0050] 2. Multiverse Optimization Algorithm

[0051] The main idea of the Multiverse Optimization Algorithm (MVO) is to establish a model based on the three main concepts of the multiverse theory, namely white holes, black holes, and wormholes. A white hole is a celestial body that only emits and does not absorb, a black hole absorbs everything in the universe, a wormhole is like a time-space tunnel connecting white holes and black holes, transporting individuals to any corner of the universe, even from one universe to another. The multiverse reaches a stable state through the three. Define the candidate solution as a universe, and the fitness of the candidate solution as the expansion rate of the universe. MVO algorithm can effectively balance the relationship between global optimization and local search, has the characteristics of few adjustment parameters, strong global optimization ability, fast convergence speed, etc., and is gradually applied to solve various optimization problems. The implementation of MVO algorithm mainly includes the following steps:

[0052] (1) Set a group of N universes to search in a D-dimensional target space, and initialize them

[0053]

[0054] (2) To establish a mathematical model between white and black holes and exchange objects between universes, use the roulette mechanism. In each iteration, sort the universes according to their expansion rate (fitness), and select a white hole through roulette betting.

[0055]

[0056] where, is the jth parameter of the ith universe, NI(X i is the normalized expansion rate of the ith universe, r1 is a random number between 0 and 1, is the jth parameter of the kth universe selected by the roulette mechanism.

[0057] (3) The probability of wormhole existence W EP increases linearly during iteration, and the travel distance rate T DR continuously decreases during iteration to perform more accurate local search within the obtained global best range. The adaptive formulas of the two coefficients are formula (8) and formula (9)

[0058]

[0059]

[0060] where, W EPmin is the minimum value of W EP , W EPmax is the maximum value of W EP , l is the current iteration number, L is the maximum iteration number, p defines the development accuracy during iteration, and the higher the value of p, the faster the local search.

[0061] (4) Update the universe position and find the optimal individual.

[0062] When r2 < W EP ,

[0063]

[0064] When r2 >= W EP ,

[0065]

[0066] Where x j j represents the jth parameter of the best universe formed at present, ub j and lb j represent the upper and lower limits of the j variable, r2, r3 and r4 are random numbers between [0, 1].

[0067] 3. Cable intermediate joint fault prediction based on temperature difference long short-term memory network learning

[0068] In order to reduce the influence of environmental temperature, realize the long-term dependence of temperature change information, automatically mine the potential correlation between temperature change and fault, collect the temperatures of multiple position points of the cable intermediate joint accessories, calculate the temperature difference values of these position points and the environmental temperature, use the time series of multiple position point temperature difference values as driving data, establish a fault prediction model based on long short-term memory network, make the model have strong early warning function, and use a multiverse optimization algorithm to optimize the key parameters of the model, and improve the prediction accuracy of the model.

[0069] 3.1 Cable intermediate joint multi-position temperature difference collection

[0070] At present, the commonly used cable intermediate joint fault detection method is to install temperature sensors on the cable connector to collect temperature values, and then to analyze the collected temperature data in real time. When the temperature is higher than the set threshold, the fault alarm is given. However, this method is greatly affected by environmental temperature changes, and may produce false judgments when the surrounding environment is abnormal or in high-temperature weather in summer.

[0071] The present application collects the temperatures of multiple position points of the cable intermediate joint accessories, as shown in the schematic diagram Figure 2 , the collection points A1 and B1 are arranged at both ends of the intermediate joint, and then the collection points A2, A3, A4 and B2, B3, B4 are arranged at the symmetric two ends, respectively. The interval between these collection points is 50CM. In order to reduce the influence of environmental temperature change, the temperature difference values of these position points and the environmental temperature are used as the basis for joint temperature change. Assuming that the collected environmental temperature is t0, the temperature values collected at 8 positions are t1-t8, and the temperature difference values of 8 positions are:

[0072] T i = |t i -t0, i∈[1,8] (12)

[0073] The cable intermediate joint contact resistance is large, and it is in a closed space with poor heat dissipation conditions. When the cable is normally operated, the temperature values of the eight positions are equivalent to the ambient temperature, and the temperature difference values of the eight positions are small. When the cable is abnormally operated, the temperature of the cable joint rapidly rises, the high temperature will cause the insulation layer of the cable joint to rapidly age, the contact resistance will increase, the temperature rise will accelerate, forming a vicious cycle, the temperature difference values of the eight positions will continue to rise, the closer to the intermediate joint, the more serious the heating, the larger the temperature difference value near the intermediate position, and the acceleration of temperature rise is also large.

[0074] 3.2 Fault prediction model establishment based on multi-universe optimized LSTM

[0075] The cable intermediate joint fault evolution trend is reflected in the temperature changes of multiple position points near the joint. These data have natural continuity in time and strong correlation and causality in the time series before and after. In this embodiment, the potential internal correlation between the joint fault evolution trend and the temperature changes of multiple position points is automatically mined by establishing a fault prediction model based on LSTM to realize early prediction of the joint fault.

[0076] The overall framework of the cable intermediate joint fault prediction model constructed in this embodiment is shown in Figure 3 , which includes data input, deep network, output prediction, parameter optimization, and the like.

[0077] Data input: the temperature difference value time series of the eight positions are taken as the input of the model, x t ={T1, T2, …, T8}, the time series data is standardized by z-score, and then the time series data X is segmented, X={X1, X2, …, X L}, which is segmented into L window data as the model input.

[0078] Deep network: the LSTM of Figure 1 is used to build a deep learning network, a single-layer LSTM network is used to build the model, the input data is first input into the embedding layer as the input of the LSTM, the output of each LSTM layer is taken as the input of the next LSTM layer, and the outputs P={P1, P2, …, P L} of all LSTM layers are mapped into a one-dimensional data through the fully connected layer (FC).

[0079] Output prediction: the one-dimensional data mapped by the fully connected layer (FC) is activated and output through the Softmax function to predict whether a fault will occur within 12 hours in the future.

[0080] Parameter optimization: In the construction of the joint fault prediction model, a large number of parameters are involved, among which the segmentation window length L and the number of LSTM hidden layers S are the most critical. In order to achieve better fault prediction effect, the multi-universe optimization algorithm is used to optimize these two parameters, and the root mean square error (RMSE) of the model output and the theoretical output is taken as the loss function of the model training.

[0081] 3.3 Fault prediction implementation process and steps

[0082] Based on the above, the implementation process of the cable intermediate joint fault prediction method based on multi-universe optimization LSTM is as shown in Figure 3 and Figure 4 The specific implementation steps are as follows:

[0083] Step 1, collect the temperature difference time series data of 8 position points near the cable intermediate joint to construct the data set, use z-score to standardize the time series data, 70% as training data, 30% as test data.

[0084] Step 2, build a fault prediction model based on LSTM, and initialize the model parameters, among which the segmentation window length L and the number of LSTM network hidden layers S are the parameters to be optimized.

[0085] Step 3, initialize the multi-universe optimization algorithm, initialize L and S, and the maximum number of iterations and the position of the universe.

[0086] Step 4, model training and prediction based on L and S parameters, the root mean square error (RMSE) of the model output and the theoretical output is taken as the fitness value of the model training, that is, the universe expansion rate.

[0087] Step 5, universe expansion rate calculation, arrange the universe according to the universe expansion rate and select a white hole through the roulette mechanism (7).

[0088] Step 6, update the wormhole existence probability W EP and the travel distance rate T DR in order to carry out more accurate local search within the obtained global optimal range.

[0089] Step 7, calculate the current universe expansion rate. If the universe expansion rate is better than the current universe expansion rate, update the current universe expansion rate, otherwise keep the current universe.

[0090] Step 8, perform individual position update of the universe according to formula (10) and formula (11) to find the optimal individual.

[0091] Step 9, termination condition judgment. If the maximum number of iterations is met, the corresponding L and S parameter values are output, otherwise the number of iterations is increased by 1 and step 4 is returned to be executed.

[0092] Step 10, establishing a fault prediction model based on the optimal L and S parameters and verifying it with a test data set.

[0093] Step 11, collecting 8 position point temperature difference time series data in real time on site to predict the cable intermediate joint fault by the established fault prediction model.

[0094] 4. Specific examples and explanations

[0095] This embodiment takes the monitoring data of a substation of Fuzhou Power Supply Company as an example to establish a power cable intermediate joint fault prediction model for fault prediction, analyze and compare the prediction results, and verify the applicability and superiority of the method of this embodiment.

[0096] Taking the 10K power cable of a substation of Fuzhou Power Supply Company as the test object, the temperature difference time series data of 8 position points of each cable intermediate joint accessory were collected from February 1, 2022 to August 15, 2022, and the sampling interval of the time series data was 1 time / minute. A total of 26000 data were collected, and the current 8 position point temperature difference time series data were used as input and whether a fault occurred within 12 hours in the future was used as output to construct a data set. The first 18200 data were selected as training data, and the last 7800 data were selected as test data.

[0097] First, the fault prediction model is trained. The initial value of the learning rate is set to 0.001, the learning rate is adjusted once every 100 training times (multiplied by 0.1 based on the original learning rate), dropout is set to 0.5 to suppress overfitting, batchsize is 20, and the multiverse optimization algorithm is used to optimize L and S parameters during training. The number of universes is set to 30, the range of L is 1-100, the range of S is 10-100, and the maximum number of iterations is 300. The training set is input into the model for training, and the loss value of the model training is as shown in Figure 5 After 50 iterations, the model begins to converge, the loss value of the test set is small and stable after convergence, overfitting is effectively avoided, and after training, the optimal value of L is 36 and the optimal value of S is 320.

[0098] To verify the advantages of the prediction model of this embodiment, a classic CNN network of deep learning is used to construct a prediction model, an LSTM is used to construct a prediction model, and the multiverse optimization LSTM prediction model of this embodiment is used. The same training set and test set are used for model training and testing, and the prediction accuracy of the three models is as shown in Figure 5As shown: the prediction model based on the CNN network takes 8 position point temperature differences as input and does not have time series characteristics, and the test set accuracy is low; the prediction model based on the LSTM network takes 8 position point temperature difference time series data as input, and the test set accuracy is improved, but the training process L, S parameters cannot reach the optimal state, resulting in a certain gap between the test set accuracy and the training set accuracy after the model converges, and overfitting phenomenon is prone to occur; the MVO+LSTM prediction model of the embodiment takes 8 position point temperature difference time series data as input, can automatically mine the potential internal correlation between the joint fault evolution trend and the temperature change of multiple position points, has high test set accuracy and training set accuracy, and has good generalization performance of the model after optimizing LSTM with multiverse, effectively avoiding overfitting phenomenon.

[0099] To verify the effectiveness and advantages of the method of the embodiment, different models are trained using the same training set, 500 time series data of normal state and fault in the future 12 hours are selected from the test set for testing, and the test results are shown in Table 1: Currently, a single point temperature value at the middle joint of the cable is collected, a higher temperature threshold is set for judgment, this method is greatly affected by environmental temperature changes, and may produce misjudgment when the surrounding environment is abnormal or in summer high temperature weather, the correct rate of prediction is only 81.8%; the prediction model based on the CNN network is used, 8 position point temperature differences near the joint are collected as input, the correct rate of prediction is improved, the correct rate of prediction is 88.8%, because the input data does not have time series characteristics, the correct rate of prediction needs to be improved; the prediction model based on the LSTM network is used, 8 position point temperature difference time series data near the middle joint are collected as input, which can automatically mine the potential internal correlation between the joint fault evolution trend and the temperature change of multiple position points, the correct rate of prediction reaches 93.7%, but the training process L, S parameters cannot reach the optimal state, and the correct rate needs to be further improved; the MVO+LSTM prediction model of the embodiment takes 8 position point temperature difference time series data as input, and can effectively predict normal state and fault, the generalization performance of the model after optimizing LSTM with multiverse is good, and the correct rate of prediction is as high as 98.5%, the method of the embodiment can automatically mine the potential internal correlation between the fault evolution trend and the temperature change, is less affected by the environmental temperature, has good model generalization performance and high prediction accuracy.

[0100] Table 1 prediction accuracy of different methods

[0101]

[0102] To verify the early warning effect of the fault prediction method of the embodiment, a 10K power cable of a substation of Fuzhou Power Supply Company of State Grid was taken as a test object, a certain intermediate joint of the 10K power cable failed on June 12, 2022, the temperature difference time series data of 8 position points near the joint 32 hours before the failure of the intermediate joint were collected for testing, and the early warning effects of the four different methods were as shown in Figure 6 and Figure 7 The threshold judgment method is greatly affected by the change of the environment temperature, and a failure misjudgment occurs at the 5-hour time point, and the warning is given only when the temperature rises sharply, and the warning is given only 2 hours in advance; the prediction model method based on the CNN network uses the temperature differences of 8 position points as input, and the warning is given 4 hours in advance, but the fluctuation is large; the warning effect of the prediction model based on the LSTM network is good, and the warning can be given 5 hours in advance, but the model parameters L and S are not in the optimal state, and the warning effect needs to be further improved; the MVO+LSTM prediction model method of the embodiment can give a warning 5 hours in advance, has good early warning effect, and is stable.

[0103] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application and not to limit them, although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced, without departing from the spirit and scope of the present application, any modification or equivalent replacement, which should be covered within the protection scope of the claims of the present application.

[0104] The present patent is not limited to the above best embodiments, and anyone can derive other various forms of cable joint fault prediction methods based on temperature difference long short-term memory network learning under the inspiration of the present patent, and any equivalent changes and modifications made within the scope of the patent application of the present application should be covered within the scope of the present patent.

Claims

1. A cable joint fault prediction method based on temperature difference long short-term memory network learning, characterized by: Firstly, the temperature of multiple position points of the cable middle joint accessory is collected, and the temperature difference value of the position points and the ambient temperature is calculated; secondly, the time series of the multiple position point temperature difference values is taken as the model input, and a fault prediction model based on a long short-term memory network is established; Then, the data set is constructed to train the model, and the multiverse optimization algorithm is used to optimize the key parameters of the model; finally, the time series temperature difference values of multiple position points of the middle joint of multiple power cables are collected, and the established model is used to predict the middle joint fault; The collection points A1 and B1 are arranged at two ends of the intermediate joint, and the collection points A2, A3, A4 and B2, B3, B4 are arranged at symmetric two ends respectively, the interval between the collection points is 50 CM, the temperature difference value between the temperature of the position points and the ambient temperature is taken as the basis of the joint temperature change, the collected ambient temperature is , the temperature values collected at 8 positions are , and the temperature difference values of the 8 positions are: The long short-term memory network fault prediction model includes four parts: data input, deep network, output prediction, and parameter optimization: Data input: the time series of temperature difference values of 8 positions as the input of the model, , the time series data X is segmented into L window data as the input of the model after being standardized by z-score, , Deep network: the network layer adopts LSTM to build a deep learning network, a single-layer LSTM network is used to build the model, the input data is first input into the embedding layer as the input of the LSTM, the output of each LSTM layer is used as the input of the next LSTM layer, and the outputs of all LSTM layers are mapped into a one-dimensional data by the full connection layer . Output prediction: the fully connected layer is mapped into one-dimensional data, and the Softmax function is activated to output, predicting whether a fault will occur within 12 hours in the future; Parameter optimization: in the construction of the joint fault prediction model, the segmentation window length L and the number of LSTM hidden layers S are optimized using the multiverse optimization algorithm, and the root mean square error RMSE of the model output and the theoretical output is taken as the loss function of the model training. 2.The cable joint fault prediction method based on the temperature difference length short-time memory network learning according to claim 1, characterized in that, Specifically, the following steps are included: Step S1: collect the temperature difference time series data of 8 position points around the cable middle joint to construct a data set, and use z-score to standardize the time series data, 70% as training data and 30% as test data; Step S2: establish a fault prediction model based on LSTM, and initialize the model parameters, wherein the segmentation window length L and the number of LSTM network hidden layers S are taken as the parameters to be optimized: Step S3: initialize the multiverse optimization algorithm, initialize L and S, and the maximum number of iterations and the universe position; Step S4: train and predict the model based on the L and S parameters, and take the root mean square error RMSE of the model output and the theoretical output as the fitness value of the model training, that is, the universe expansion rate; Step S5: universe expansion rate calculation, arrange the universe according to the universe expansion rate and select a white hole through the roulette mechanism; Step S6: updating wormhole existence probability and travel distance rate in order to make more accurate local search within the obtained global best range; Step S7: calculate the current universe expansion rate: if the universe expansion rate is better than the current universe expansion rate, update the current universe expansion rate, otherwise keep the current universe; Step S8: execute individual position update of the universe to find the optimal individual; Step S9: termination condition judgment: if the maximum number of iterations is met, output the corresponding L and S parameter values, otherwise increase the iteration number by 1 and return to step S4; Step S10: establish a fault prediction model based on the optimal L and S parameters, and verify it using the test data set; Step S11: collect the temperature difference time series data of 8 position points in real time through the established fault prediction model to predict the cable middle joint fault.

Images