Underground cable fault risk early warning method and device based on CEEMDAN-NGO-BiLSTM

Abstract

The invention discloses an underground cable fault risk early warning method and device based on CEEMDAN-NGO-BiLSTM. The method comprises the following steps: acquiring a multi-working-condition multi-source time sequence data set; dividing the multi-source time series data set into a training set, a verification set and a test set; respectively decomposing data from different sources in the time sequence data set into k modal components; importing k modal components obtained by decomposition of the training set into BiLSTM for training to obtain a trained model; importing k modal components obtained by decomposing the verification set into the trained model to evaluate and calculate the risk diagnosis precision; according to the risk diagnosis precision index, optimizing the hyper-parameters of the BiLSTM to obtain an optimized underground cable risk early warning model; and importing the k modal components of the test set into the underground cable risk early warning model to calculate output values, and superposing the output k values to calculate an evaluation index so as to realize operation risk early warning. According to the invention, the risk of the underground cable can be accurately assessed and even pre-warned in advance.

Description

Technical Field

[0001] The present invention relates to the technical field of cable risk early warning, and in particular to a method and device for early warning of underground cable fault risks based on CEEMDAN-NGO-BiLSTM. Background Art

[0002] With the widespread application of power cables in urban power grid construction, the breakdown failures caused by insulation aging, electrical treeing and external force damage of the cross-linked polyethylene cables put into operation in the early stage have occurred repeatedly, and are on the rise. This situation has seriously threatened the reliable operation of the cable network. In addition, the cable overload phenomenon causes the cable temperature to be too high, which will also accelerate aging. At present, some cables have gradually reached their expected lifespan. Once a fault occurs, it is extremely difficult to find it. Fault finding not only consumes a lot of manpower and material resources, but also brings immeasurable power outage losses, affects the daily life of residents and the normal operation of other industries, and causes adverse social impacts and significant economic losses. Therefore, it is of great practical significance to grasp the operation status of the cable in real time, evaluate its fault risk level and even warn it.

[0003] At present, the main means of detecting underground cables include temperature detection and partial discharge detection. However, the general temperature detection method can only monitor overheating faults and cannot issue risk warnings for cable insulation breakdown. The partial discharge detection method is easily affected by noise and has low detection accuracy.

[0004] When an underground cable has an overheating fault, it is usually accompanied by changes in temperature and humidity and the generation of special gases. In addition, before an underground cable has an insulation breakdown fault, it is often accompanied by partial discharge, and correspondingly some special gases will be generated. The grounding loop current of the cable can also reflect the current carrying capacity of the cable and the buffer layer fault information to a certain extent. Therefore, the comprehensive monitoring and prediction of multi-source time series data can realize the assessment and early warning of various fault risks.

[0005] However, multi-source time series data is complex and has different characteristics between multiple data. It is difficult to predict multi-source time series data with a unified method. Therefore, it is necessary to use a decomposition algorithm to decompose the data modally. Data with different characteristics or sources can show main features at different modal layers. In addition, the bidirectional long short-term memory neural network is a good neural network model for solving time series prediction problems, but its model is relatively complex and the parameter adjustment is slow. Therefore, an optimization algorithm is also needed to assist in parameter adjustment and achieve rapid training of the model.

[0006] Therefore, how to reasonably perform data mining on multi-source data, realize the fault risk level assessment of underground cables, and the early warning method and device before the fault occurs are problems that people in this field urgently need to solve. Summary of the invention

[0007] In view of this, it is necessary to provide an underground cable fault risk warning method and device based on CEEMDAN-NGO-BiLSTM to solve the problem of how to accurately evaluate the underground cable risk and even provide early warning.

[0008] The present invention provides an underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM, comprising:

[0009] Real-time acquisition of multi-source data under multiple working conditions to form a multi-condition multi-source time series data set D;

[0010] The multi-condition and multi-source time series data set D is randomly divided into the training set D according to the ratio of 7:1.5:1.5. train , validation set D valid and the test set D test ;

[0011] The time series data set D in the past period of time is transformed into a fully adaptive noise ensemble empirical mode decomposition algorithm (CEEMDAN). train , D valid and D test The data from different sources are decomposed into k modal components respectively;

[0012] The training set D train The k modal components obtained by decomposition are introduced into a bidirectional long short-term memory neural network (BiLSTM) for training to obtain a trained model;

[0013] The validation set D valid The k modal components obtained by decomposition are imported into the trained model to evaluate the accuracy of risk diagnosis;

[0014] During the training process, the Northern Goshawk Optimization Algorithm (NGO) was used to optimize the hyperparameters of the bidirectional long short-term memory neural network according to the calculated risk diagnosis accuracy index;

[0015] Obtain the optimized underground cable risk warning model;

[0016] Finally, the test set D test The k modal components are imported into the underground cable risk warning model to calculate the output value, and the output k values ​​are superimposed to calculate the evaluation index to achieve early warning of operation risks.

[0017] Furthermore, the multiple operating conditions include normal operating conditions, overheating fault operating conditions and insulation breakdown fault operating conditions; the multi-source data include underground cable time series temperature data, ground loop current data, gas particle concentration data and humidity data.

[0018] Furthermore, the fully adaptive noise set empirical mode decomposition algorithm CEEMDAN is used to decompose the time series data set D train , D valid and D test The data from different sources are decomposed into k modal components, including:

[0019] Add Gaussian white noise to the original time series signal;

[0020] Perform a layer of decomposition on the signal after adding noise to obtain the first layer of modal components and residuals;

[0021] Perform a second-layer decomposition on the first-layer residual and the remaining signal of the decomposition to obtain the second-layer modal components and residuals;

[0022] Repeat the above decomposition process until the residual signal becomes a monotonic function and the modal decomposition is completed.

[0023] Furthermore, the training set D train The k modal components obtained by decomposition are introduced into the bidirectional long short-term memory neural network BiLSTM for training, and the trained model is obtained, including:

[0024] The time series data are imported into the network according to the normal time series (forward) and the reverse time series (backward), respectively, where the output true value corresponding to the time series data under normal operation is 0, the output true value corresponding to the time series data under overheating fault operation is 1, and the output true value corresponding to the time series data under insulation breakdown fault operation is 2;

[0025] Calculate the forward forget gate and backward forget gate according to the initialized network, and select the information to be forgotten;

[0026] Calculate the forward memory gate and the backward memory gate according to the initialized network, and select the information to be memorized;

[0027] Calculate the cell state at the current moment;

[0028] The forward information and the backward information are superimposed to calculate the output gate and the hidden layer state at the current moment;

[0029] Update the next iteration and finally output the trained network.

[0030] Furthermore, in the training process, the Northern Goshawk Optimization Algorithm NGO is used to optimize the hyperparameters of the bidirectional long short-term memory neural network according to the calculated risk diagnosis accuracy index, including:

[0031] The Northern Goshawk Optimization Algorithm is used to optimize the batch size and learning rate of the bidirectional long short-term memory neural network, which includes two stages.

[0032] In the first stage, the northern goshawk randomly selects a prey and quickly attacks it. This stage is called the prey recognition stage, and the mathematical expression is as follows:

[0033] P i =X k ,i=1,2,...,N;k=1,2,...,i-1,...,N

[0034]

[0035] Among them, P i is the location of the prey selected by the i-th northern goshawk; F pi is the objective function value, i.e., the fitness value, k is a random natural number in [1, N], x i new,l1 is the new state of the i-th northern goshawk, x i new,p1 is its new state in the jth dimension, F i new,p1 is the corresponding fitness value, r is a random number in [0,1], and the value of I is 1 or 2;

[0036] The second stage is the pursuit and escape stage. At this time, the northern goshawk is extremely fast and can catch prey at any time and place. Assuming that the northern goshawk is in an attack position with a radius of R during hunting, the mathematical expression is as follows:

[0037]

[0038] Among them, t is the current iteration number; T is the maximum iteration number; X i new,p2 is the new position of the i-th northern goshawk; x i,j new,p2 is the new position of the i-th northern goshawk in the j-dimension, x i,j p2 is the new position of the jth dimension of the ith prey after the second stage update; F i new,p2 is the objective function value of the i-th northern goshawk after updating based on the second stage;

[0039] After chasing the prey, the optimization of hyperparameters is completed.

[0040] Furthermore, the test set D test The k modal components of the underground cable risk warning model are imported to calculate the output value, and the output k values ​​are superimposed to calculate the evaluation index, including:

[0041] The k modal components are respectively imported into the corresponding trained bidirectional long short-term memory neural network model, and k output values ​​are calculated;

[0042] Superimpose k output values ​​to generate the final value. If the value is in the interval (-∞, 0.2], the cable is in an excellent operating state; if the value is in the interval (0.2, 0.5], the cable is in a good operating state; if the value is in the interval (0.5, 0.8], the cable is in an overheating risk operating state; if the value is in the interval (0.8, 1.2], the cable is in a normal overheating fault state; if the value is in the interval (1.2, 1.8], the cable is in a severe overheating fault state; if the value is in the interval (1.8, 2.2], the cable is in an insulation breakdown fault state; if the value is in the interval (2.2, +∞], the cable is in a severe insulation breakdown fault state;

[0043] Based on the above status indicators, the risk level of underground cables can be assessed and early warning of risks can be provided.

[0044] A CEEMDAN-NGO-BiLSTM-based underground cable fault risk warning device comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the CEEMDAN-NGO-BiLSTM-based underground cable fault risk warning method is implemented.

[0045] Compared with the prior art, the beneficial effects of the present invention include:

[0046] 1. The underground cable fault risk warning method provided integrates multiple data sources and is more reliable. It can not only make a level assessment of the fault risk of underground cables, but also make a fault warning before the fault occurs, and has multiple fault risk assessment capabilities.

[0047] 2. The underground cable fault risk warning method based on CEEMDAN-NGO-BiLSTM provided first uses the CEEMDAN algorithm to perform modal decomposition on time series data to address the problem of inconsistent features of multi-source data, and displays the main features of different data sources on different modal layers to facilitate subsequent data training; secondly, BiLSTM is used to better capture the contextual information of time series data; on this basis, the NGO optimization algorithm is also used to optimize and shorten the parameter adjustment process of BiLSTM, making its trained network model more optimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flow chart of the underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM provided by the present invention;

[0049] Figure 2A schematic diagram of an underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM provided by the present invention;

[0050] Figure 3 A flowchart of step S3 provided by the present invention;

[0051] Figure 4 A flow chart of step S4 provided by the present invention;

[0052] Figure 5 A flow chart of step S6 provided by the present invention; DETAILED DESCRIPTION

[0053] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0054] Example 1

[0055] The embodiment of the present invention provides an underground cable fault risk warning method based on CEEMDAN-NGO-BiLSTM, combined with Figure 1 and Figure 2 Come and see, Figure 1 A flowchart of the underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM provided by the present invention, Figure 2 The schematic diagram of the underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM provided by the present invention is the underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM, including steps S1 to S8, wherein:

[0056] In step S1, multi-source data such as underground cable time series temperature data, grounding circulation data, gas particle concentration data and humidity data under normal operation, overheating fault operation and insulation breakdown fault operation are collected in real time to form a multi-condition multi-source time series data set D;

[0057] In step S2, the multi-operating-state multi-source time series data set D is randomly divided into the training set D and the training set D according to the ratio of 7:1.5:1.5. train , validation set D valid and the test set D test ;

[0058] In step S3, the time series data set D in the past period of time is transformed into train , D valid and D test The data from different sources are decomposed into k modal components respectively;

[0059] In step S4, the training set D train The k modal components obtained by decomposition are introduced into a bidirectional long short-term memory neural network (BiLSTM) for training to obtain a trained model;

[0060] In step S5, the validation set D valid The k modal components obtained by decomposition are imported into the trained model to evaluate the accuracy of risk diagnosis;

[0061] In step S6, the Northern Goshawk Optimization Algorithm (NGO) is used during the training process to optimize the hyperparameters of the bidirectional long short-term memory neural network according to the calculated risk diagnosis accuracy index;

[0062] In step S7, an optimized underground cable risk warning model is obtained;

[0063] In step S8, the test set D is finally test The k modal components are imported into the underground cable risk warning model to calculate the output value, and the output k values ​​are superimposed to calculate the evaluation index to achieve early warning of operation risks.

[0064] Preferably, the multi-condition and multi-source data set includes: collecting data of various conditions, such as normal operating conditions, overheating fault operating conditions and insulation breakdown fault operating conditions; collecting multi-source data, such as temperature time series data, ground loop current time series data, gas particle concentration time series data and humidity time series data.

[0065] Preferably, combined Figure 3 Come and see, Figure 3 The step S3 flow chart provided by the present invention includes steps S31 to S34, wherein:

[0066] In step S31, Gaussian white noise is added to the original time series signal;

[0067] In step S32, the signal after adding noise is decomposed in one layer to obtain the first layer modal components and residuals;

[0068] In step S33, the first layer residual and the remaining signal of the decomposition are subjected to a two-layer decomposition to obtain the second layer modal components and residual;

[0069] In step S34, the above decomposition process is repeated until the residual signal becomes a monotonic function, and the modal decomposition is completed.

[0070] As a specific embodiment, the embodiment of the present invention uses the CEEMDAN algorithm to perform modal decomposition on temperature time series data, ground loop current time series data, gas particle concentration time series data and humidity time series data.

[0071] Preferably, the process of adding Gaussian white noise to the original time series signal is expressed by the following formula:

[0072]

[0073] in, is the signal after adding noise, η0 is the amplitude adjustment coefficient, W i (n) is Gaussian white noise obeying N(0,1), and X(n) is the original time series signal.

[0074] Preferably, the signal after adding noise is decomposed into one layer to obtain the first layer modal component and residual as shown in the following formula:

[0075]

[0076] in, is the first layer modal component, I is the number of decomposition layers, and r1(n) is the residual.

[0077] Preferably, the first layer residual and the remaining signal of the decomposition are subjected to a two-layer decomposition to obtain the second layer modal components and residuals, which is expressed by the following formula:

[0078]

[0079] in, is the second-layer modal component, r2(n) is the second-layer modal component residual, EMD1 is a layer of empirical mode decomposition, and the Hilbert spectrum is obtained.

[0080] Preferably, after completing the modal decomposition, the original time series data can be expressed by the following formula:

[0081]

[0082] in, is the k-th layer modal component, R(n) is the final residual, and K is uniformly set to 5 in this embodiment.

[0083] Preferably, combined Figure 4 Come and see, Figure 4 The present invention provides a flow chart of step S4, including steps S41 to S46, wherein:

[0084] In step S41, the timing data is used as input and imported into the network according to normal timing (forward) and reverse timing (backward), and the output is the cable status corresponding to the timing data;

[0085] In step S42, a forward forget gate and a backward forget gate are calculated according to the initialized network, and information to be forgotten is selected;

[0086] In step S43, the forward memory gate and the backward memory gate are calculated according to the initialized network, and the information to be memorized is selected;

[0087] In step S44, the cell state at the current moment is calculated;

[0088] In step S45, the forward information and the backward information are superimposed to calculate the output gate and the hidden layer state at the current moment;

[0089] In step S46, the next iteration is updated and the trained network is finally output.

[0090] As a specific embodiment, the embodiment of the present invention imports the 5-layer modal components of temperature time series data, ground loop current time series data, gas particle concentration time series data and humidity time series data after CEEMDAN modal decomposition as input into the BiLSTM neural network for training, and outputs the cable status corresponding to the time series data, wherein the output true value corresponding to the time series data under normal operation is 0, the output true value corresponding to the time series data under overheating fault operation is 1, and the output true value corresponding to the time series data under local fault operation is 2.

[0091] Preferably, the forward propagation process of the BiLSTM network is expressed by the following formula:

[0092] f t =σ(W f ×[h t-1 ,x t ]+b f )

[0093] i t =σ(W i ×[h t-1 ,x t ]+b i )

[0094] o t =σ(W o ×[h t-1 ,x t ]+b o )

[0095] c t =f t ⊙c t-1 +i t ⊙tanh(W c ×[h t-1 ,x t ]+b c )

[0096] h t =o t ⊙tanh(c t )

[0097] Among them, f t represents the forget gate, i t represents the input gate, o t represents the output gate, σ represents the sigmoid function, ⊙ represents element-by-element multiplication, b represents the learnable bias parameter, W represents the weight matrix, [h t-1 ,x t ] means to change the hidden state h of the previous time step t-1 and the current input x t For splicing, c t Indicates the cell state.

[0098] Preferably, the back propagation process of the BiLSTM network is expressed by the following formula:

[0099]

[0100] The process of backward propagation is similar to that of forward propagation, and the corresponding function names are distinguished by wavy lines.

[0101] Preferably, combined Figure 5 , Figure 5 The present invention provides a flow chart of step S6, including step S61 to step S64, wherein:

[0102] In step S61, two hyperparameters of the batch size and learning rate of the bidirectional long short-term memory neural network are extracted as the target "prey" of the northern goshawk optimization algorithm;

[0103] In step S62, in the first stage, the northern goshawk randomly selects a prey and quickly attacks it, which is called the prey identification stage;

[0104] In step S63, the second stage is the pursuit and escape stage, at which the northern goshawk is extremely fast and can catch prey at any time and place.

[0105] In step S64, the optimization of the hyperparameters is completed after the prey is chased.

[0106] As a specific embodiment, the embodiment of the present invention extracts 10 hyperparameters, namely, the learning rate and batch size of the five network models after BiLSTM training of the five layers of modal components after modal decomposition, and optimizes the hyperparameters respectively using the Northern Goshawk algorithm.

[0107] Preferably, the process of the prey identification stage is expressed by the following formula:

[0108] P i =X k ,i=1,2,...,N;k=1,2,...,i-1,...,N

[0109]

[0110] Among them, P i is the location of the prey selected by the i-th northern goshawk; F pi is the objective function value, i.e., the fitness value, k is a random natural number in [1, N], x i new,l1 is the new state of the i-th northern goshawk, x i new,p1 is its new state in the jth dimension, F i new,p1 is the corresponding fitness value. r is a random number in [0,1]. In addition, the value of I is 1 or 2;

[0111] Preferably, the process of the pursuit and escape stage is expressed by the following formula:

[0112]

[0113]

[0114] Among them, t is the current iteration number; T is the maximum iteration number; X i new,p2 is the new position of the i-th northern goshawk; x i,j new,p2 is the new position of the i-th northern goshawk in the j-dimension, x i,j p2 is the new position of the jth dimension of the ith prey after the second stage update; F i new,p2 is the objective function value of the i-th northern goshawk after updating based on the second stage;

[0115] Preferably, the optimization of the hyperparameters is completed after the prey is chased.

[0116] Preferably, the test set D test The five modal components of the underground cable risk warning model are imported to calculate the output value, and the five output values ​​are superimposed to calculate the evaluation indicators including:

[0117] The five modal components are respectively imported into the corresponding trained bidirectional long short-term memory neural network models, and five output values ​​are calculated;

[0118] The 5 output values ​​are superimposed to generate the final value. If the value is in the interval (-∞, 0.2], the cable is in an excellent operating state; if the value is in the interval (0.2, 0.5], the cable is in a good operating state; if the value is in the interval (0.5, 0.8], the cable is in an overheating risk operating state; if the value is in the interval (0.8, 1.2], the cable is in a normal overheating fault state; if the value is in the interval (1.2, 1.8], the cable is in a severe overheating fault state; if the value is in the interval (1.8, 2.2], the cable is in an insulation breakdown fault state; if the value is in the interval (2.2, +∞], the cable is in a severe insulation breakdown fault state;

[0119] As a specific embodiment, the output results calculated in the test set import model of the embodiment of the present invention are shown in the following table:

[0120] Table 1

[0121]

[0122]

[0123] Example 2

[0124] An embodiment of the present invention provides an underground cable fault risk warning device based on CEEMDAN-NGO-BiLSTM, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the underground cable fault risk warning method as described above is implemented.

[0125] The present invention discloses an underground cable fault risk early warning method and device based on CEEMDAN-NGO-BiLSTM. The method comprises real-time acquisition of underground cable time series temperature data, grounding loop current data, gas particle concentration data, humidity data and other multi-source data under normal operation, overheating fault operation and insulation breakdown fault operation to form a multi-condition multi-source time series data set D; the multi-source time series data set D is randomly divided into a training set D according to a ratio of 7:1.5:1.5 train , validation set D valid and the test set D test ; Use the fully adaptive noise ensemble empirical mode decomposition algorithm (CEEMDAN) to transform the time series data set D over a period of time train , D valid and D test The data from different sources are decomposed into k modal components; the training set D train The k modal components obtained by decomposition are introduced into a bidirectional long short-term memory neural network (BiLSTM) for training to obtain a trained model; the validation set D validThe k modal components obtained by decomposition are imported into the trained model to evaluate the accuracy of risk diagnosis. During the training process, the Northern Goshawk Optimization Algorithm (NGO) is used to optimize the hyperparameters of the bidirectional long short-term memory neural network according to the calculated risk diagnosis accuracy index. The optimized underground cable risk warning model is obtained. Finally, the test set D test The k modal components are imported into the underground cable risk warning model to calculate the output value, and the output k values ​​are superimposed to calculate the evaluation index to achieve early warning of operation risks.

[0126] The technical solution of the present invention uses the CEEMDAN-NGO-BiLSTM method to evaluate and warn the fault risk level of underground cables. On the basis of better capturing the contextual information of time series data, the algorithm also solves the problem of inconsistent features of multi-source data, making the trained network model more optimized, and ultimately realizing the fault risk assessment and warning of underground cables.

[0127] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for warning underground cable fault risk based on CEEMDAN-NGO-BiLSTM, characterized in that: include: Real-time acquisition of multi-source data under multiple working conditions to form a multi-condition multi-source time series data set D; The multi-condition and multi-source time series data set D is randomly divided into the training set D according to the ratio of 7:1.5:1.

5. train , validation set D valid and the test set D test ; The time series data set D in the past period of time is transformed into a fully adaptive noise ensemble empirical mode decomposition algorithm CEEMDAN. train , D valid and D test The data from different sources are decomposed into k modal components respectively; The training set D train The k modal components obtained by decomposition are introduced into the bidirectional long short-term memory neural network BiLSTM for training to obtain the trained model; The validation set D valid The k modal components obtained by decomposition are imported into the trained model to evaluate the accuracy of risk diagnosis; During the training process, the Northern Goshawk Optimization Algorithm NGO was used to optimize the hyperparameters of the bidirectional long short-term memory neural network according to the calculated risk diagnosis accuracy index; Obtain the optimized underground cable risk warning model; Finally, the test set D test The k modal components are imported into the underground cable risk warning model to calculate the output value, and the output k values ​​are superimposed to calculate the evaluation index to achieve early warning of operation risks.

2. The underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM according to claim 1 is characterized in that: The multiple operating conditions include normal operating conditions, overheating fault operating conditions and insulation breakdown fault operating conditions; the multi-source data include underground cable time series temperature data, ground loop current data, gas particle concentration data and humidity data.

3. The underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM according to claim 1 is characterized in that: The fully adaptive noise set empirical mode decomposition algorithm CEEMDAN is used to decompose the time series data set D in the past period of time. train , D valid and D test The data from different sources are decomposed into k modal components, including: Add Gaussian white noise to the original time series signal; Perform a layer of decomposition on the signal after adding noise to obtain the first layer of modal components and residuals; Perform a second-layer decomposition on the first-layer residual and the remaining signal of the decomposition to obtain the second-layer modal components and residuals; Repeat the above decomposition process until the residual signal becomes a monotonic function and the modal decomposition is completed.

4. The underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM according to claim 1 is characterized in that: The training set D train The k modal components obtained by decomposition are introduced into the bidirectional long short-term memory neural network BiLSTM for training, and the trained model is obtained, including: The time series data are imported into the network according to the normal time series and the reverse time series respectively, wherein the output true value corresponding to the time series data under normal operation is 0, the output true value corresponding to the time series data under overheating fault operation is 1, and the output true value corresponding to the time series data under local fault operation is 2; Calculate the forward forget gate and backward forget gate according to the initialized network, and select the information to be forgotten; Calculate the forward memory gate and the backward memory gate according to the initialized network, and select the information to be memorized; Calculate the cell state at the current moment; The forward information and the backward information are superimposed to calculate the output gate and the hidden layer state at the current moment; Update the next iteration and finally output the trained network.

5. The underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM according to claim 1 is characterized in that: The training process uses the Northern Goshawk Optimization Algorithm NGO to optimize the hyperparameters of the bidirectional long short-term memory neural network according to the calculated risk diagnosis accuracy index, including: The Northern Goshawk Optimization Algorithm is used to optimize the batch size and learning rate of the bidirectional long short-term memory neural network, which includes two stages. In the first stage, the northern goshawk randomly selects a prey and quickly attacks it. This stage is called the prey recognition stage, and the mathematical expression is as follows: P i =X k ,i=1,2,...,N;k=1,2,...,i-1,...,N Among them, P i is the location of the prey selected by the i-th northern goshawk; F pi is the objective function value, i.e., the fitness value, k is a random natural number in [1, N], x i new,l1 is the new state of the i-th northern goshawk, x i new,p1 is its new state in the jth dimension, F i new,p1 is the corresponding fitness value, r is a random number in [0,1], and the value of I is 1 or 2; The second stage is the pursuit and escape stage. At this time, the northern goshawk is extremely fast and can catch prey at any time and place. Assuming that the northern goshawk is in an attack position with a radius of R during hunting, the mathematical expression is as follows: Among them, t is the current iteration number; T is the maximum iteration number; X i new,p2 is the new position of the i-th northern goshawk; x i,j new,p2 is the new position of the i-th northern goshawk in the j-dimension, x i,j p2 is the new position of the jth dimension of the ith prey after the second stage update; F i new,p2 is the objective function value of the i-th northern goshawk after updating based on the second stage; After chasing the prey, the optimization of hyperparameters is completed.

6. The underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM according to claim 1 is characterized in that: The test set D test The k modal components of the underground cable risk warning model are imported to calculate the output value, and the output k values ​​are superimposed to calculate the evaluation index, including: The k modal components are respectively imported into the corresponding trained bidirectional long short-term memory neural network model, and k output values ​​are calculated; Superimpose k output values ​​to generate the final value. If the value is in the interval (-∞, 0.2], the cable is in an excellent operating state; if the value is in the interval (0.2, 0.5], the cable is in a good operating state; if the value is in the interval (0.5, 0.8], the cable is in an overheating risk operating state; if the value is in the interval (0.8, 1.2], the cable is in a normal overheating fault state; if the value is in the interval (1.2, 1.8], the cable is in a severe overheating fault state; if the value is in the interval (1.8, 2.2], the cable is in an insulation breakdown fault state; if the value is in the interval (2.2, +∞], the cable is in a severe insulation breakdown fault state; Based on the above status indicators, the risk level of underground cables can be assessed and early warning of risks can be provided.

7. An underground cable fault risk warning device based on CEEMDAN-NGO-BiLSTM, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the underground cable fault risk early warning method based on CEEMDAN-NGO-BiLSTM according to any one of claims 1 to 6 is implemented.

Images