Intelligent switching method and system for power communication network

Abstract

The invention discloses an intelligent switching method and system for a power communication network, and belongs to the technical field of communication and informationization. A historical database of the power communication network is constructed, and an initial value of the system is defined; performing standardization processing on historical data by adopting Z-score, calculating a Pearson's correlation coefficient between each alarm data and a fault type, and screening an initial attribute; weighting the alarm data by using a wavelet neural network model; fault diagnosis is carried out by adopting an improved convolutional neural network; the improved particle swarm algorithm is adopted to carry out intelligent switching on the path where the fault point is located, the fault recovery speed is increased, and the recovered power system has high robustness.

Description

Technical Field

[0001] The present invention relates to the field of communication technology and information technology, and more specifically to an intelligent switching method and system for a power communication network. Background Art

[0002] With the continuous development of my country's economy, the stable, safe and reliable operation of the power system has become increasingly important. As a key component of the intelligent dispatching and modern management of the power system, the power communication network carries various dispatching services and plays a vital role. However, the failure of the power communication network will not only reduce the transmission quality of the communication service, but in serious cases, it will even affect the safe production of the power system.

[0003] Therefore, how to improve the accuracy of fault determination and rapid recovery in the power communication network has become a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0004] In view of this, the present invention provides a method and system for intelligent switching of a power communication network, which can greatly improve the fault recovery speed of the power communication network and maintain the safe and stable operation of the power grid.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] On the one hand, the present invention provides a method for intelligent switching of a power communication network, comprising the following steps:

[0007] (1) Build a historical database of the power communication network and define the initial value of the system. In order to reflect the intelligence of the intelligent switching of the power communication network, the initialization value can be given based on the historical data experience value, or it can be adjusted by the staff.

[0008] Specifically, define X as the historical database of the power communication network, including the power communication network alarm data, fault types, and each node of the power communication network and its related topological diagrams. Define the initial values ​​of the system: alarm attribute screening threshold R; dynamic variable scale convolution layer size L; improved convolutional neural network convergence threshold H; improved convolutional neural network maximum iteration number F; improved particle swarm algorithm global search inertia factor maximum value ω' max and the minimum value ω' min ; Improve the maximum number of iterations of the convolutional neural network T'; Improve the particle swarm algorithm adjustment coefficient α, α∈[0,1]; Improve the first acceleration constant β of the particle swarm algorithm 1 and the second acceleration constant β 2 ; Improve the speed coefficient λ of particle swarm algorithm.

[0009] (2) Filter initial attributes

[0010] The present invention adopts the Pearson correlation coefficient to screen out attributes valuable for fault diagnosis.

[0011] 1) Data Standardization

[0012] The Z-score is used to standardize the data of the historical database X.

[0013] 2) Calculate the Pearson correlation coefficient

[0014] Calculate the Pearson correlation coefficient between each alarm attribute and the fault type.

[0015] 3) Filter attributes

[0016] According to the value of the Pearson correlation coefficient, the data set X1 with a correlation coefficient greater than the alarm attribute screening threshold R is finally selected as the data set X1 used next.

[0017] (3) Using the wavelet neural network model to weight the alarm data

[0018] The present invention adopts a wavelet neural network model as a method for weighting alarm information. The weighting process is: first, design the structure of the wavelet neural network, including the number and connection mode of neurons in the input layer, hidden layer and output layer; then, use the prepared data set to train and verify the wavelet neural network model; finally, use the trained wavelet neural network model to perform weighted operations on new alarm information and participate in subsequent model fault diagnosis. The specific steps are as follows:

[0019] 1) Screening training samples

[0020] From the historical data set X1 after the initial attributes are determined, group P is selected as the training sample set, and group Q is selected as the verification sample set.

[0021] 2) Wavelet neural network structure design

[0022] Design the structure of the wavelet neural network, including the number and connection of neurons in the input layer, hidden layer, and output layer. The hidden layer uses the Sigmoid activation function, and the output layer uses the linear activation function.

[0023] 3) Wavelet neural network model training

[0024] The back-propagation algorithm is used for model training, and the gradient descent method is used to continuously train the model to find the appropriate displacement factor, scale factor, and connection weight, so that the input layer (alarm information) and the output layer (alarm weight) have a better correspondence and obtain the optimal model.

[0025] 4) Model Validation

[0026] The validation set Q is used to verify the performance of the model. The present invention uses the mean square error (MSE) to calculate the error index between the predicted result and the true label. The lower the value of this evaluation index, the better the model performance.

[0027] 5) Weighted operation of alarm information

[0028] The new alarm information data is input into the trained wavelet neural network model to obtain the output result, so that each alarm information is assigned a reasonable weight value, which reflects its importance and urgency in the current operation and maintenance scenario.

[0029] (4) Using improved convolutional neural network for fault diagnosis

[0030] In order to improve the efficiency of feature learning and reduce the impact of degradation, dynamic variable scale convolution is used to optimize the traditional convolutional neural network for fault diagnosis. The specific steps are as follows:

[0031] 1) Construct a convolutional neural network with dynamic variable scale convolution optimization

[0032] ①Construction of dynamic variable scale convolution layer

[0033] The preprocessed alarm data is input into multiple parallel one-dimensional convolutional layers of the dynamic variable scale convolutional layer, and the number of channels of these parallel convolutional layers is the same but the convolution kernel size is different. The dynamic variable scale convolutional layer of the present invention is composed of one-dimensional convolutional layers of size L, 2L and 4L, each convolutional layer contains 8 convolution kernels, the step size is 1, and the value of L is dynamically adjusted during training. Padding is used to ensure that the size of the output of each convolutional layer is consistent. The output features of each convolutional layer are cascaded to obtain a feature vector for the input of the next pooling layer.

[0034] ②Pooling layer construction

[0035] The pooling layer is set after the dynamic variable scale convolution layer, and its function is feature mapping. In the pooling layer, the weighted sum and bias increase of each local feature output by the dynamic variable scale convolution layer are performed to obtain new features based on the nonlinear function.

[0036] ③Fully connected layer construction

[0037] The task of the fully connected layer is to represent the feature map obtained by dynamic variable scale convolution and pooling operations with a one-dimensional feature vector, and then recognize the pattern through the classifier.

[0038] 2) Compare the output of the fully connected layer with the actual value, and calculate the error between the two through the loss function.

[0039] 3) Determine whether the network converges (less than the improved convolutional neural network convergence threshold H for convergence). If so, proceed to step 5); otherwise, proceed to step 4).

[0040] 4) Network back propagation and weight update

[0041] The error value calculated in step 2) is fed back to each layer of the network, and the weights of each layer are corrected. The back propagation algorithm is used to calculate the gradient, which represents the contribution of each weight to the network output error. Then, the weight of the network is updated according to the learning rate and gradient, and the error value is calculated again until the convergence condition is met.

[0042] 5) The loss function value is used to determine whether the network meets the requirements or reaches the maximum number of iterations F of the improved convolutional neural network. If the network meets or reaches the maximum number of iterations F of the improved convolutional neural network, the process proceeds to step 6), otherwise it proceeds to step 2) to adjust and modify the model parameters.

[0043] 6) Fault diagnosis result output

[0044] The trained model is used for actual fault diagnosis tasks. For each sample to be diagnosed, it is input into the model and the prediction result output by the model is obtained. Then, the prediction result is matched with the possible fault type and the final fault diagnosis result is output, including the fault point location and fault type.

[0045] (5) Using an improved particle swarm algorithm to intelligently switch the path where the fault point is located

[0046] The network structure topology diagram involved in the fault point location in step (4) is found in the database, and the fault path where the fault point location is located is intelligently switched according to the fault type. The specific process is as follows:

[0047] 1) In the particle swarm algorithm, each faulty communication node load and disconnecting switch passed by a fault path are single particles. The initial velocity and position of the particle are obtained by random method. The fitness function takes the number of segmented switches of the particle, the path length and its line load value as key indicators.

[0048] 2) It is proposed to use nonlinear variable inertia factor to calculate the global search inertia factor.

[0049] 3) Update the particle velocity and position using the improved global search inertia factor.

[0050] 4) Determine whether the number of repetitions has exceeded the maximum number of iterations of the improved particle swarm algorithm. If so, output the next best position of the switching path; if not, repeat steps 2) and 3) to continue iterating.

[0051] On the other hand, the present invention further proposes a power communication network intelligent switching system, which is used to implement the above-mentioned power communication network intelligent switching method, comprising:

[0052] Data storage module, used to build the power communication network history database and define the system initial value;

[0053] Data preprocessing module, used to standardize historical data, calculate the Pearson correlation coefficient between each alarm data and fault type, and screen initial attributes;

[0054] A data weighting module is used to weight the alarm data using a wavelet neural network model;

[0055] A fault diagnosis module, used for performing fault diagnosis on weighted alarm data;

[0056] Intelligent switching module, used to use improved particle swarm algorithm to intelligently switch the path where the fault point is located;

[0057] The output module is used to output the fault point location, fault type and intelligent switching suggestions.

[0058] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses a method and system for intelligent switching of an electric power communication network, which constructs a historical database of the electric power communication network and defines the initial value of the system; uses Z-score to standardize the historical data, calculates the Pearson correlation coefficient between each alarm data and the fault type, and screens the initial attributes; uses a wavelet neural network model to weight the alarm data; uses dynamic variable scale convolution to optimize the traditional convolutional neural network for fault diagnosis; and uses an improved particle swarm algorithm to intelligently switch the path where the fault point is located, which not only improves the fault recovery speed, but also makes the restored power system have stronger robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0060] Figure 1 It is a flow chart of the intelligent switching method of the power communication network of the present invention;

[0061] Figure 2 This is a flowchart of the weighted processing of alarm data of the present invention;

[0062] Figure 3 This is a comparison chart of the fault diagnosis accuracy of the present invention and other three methods in the power communication network. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] On the one hand, an embodiment of the present invention discloses a method for intelligent switching of a power communication network, comprising the following steps:

[0065] (1) Constructing a historical database of the power communication network and defining system initial values. In order to reflect the intelligence of the intelligent switching of the power communication network in this embodiment, the initialization value can be given based on the historical data experience value, and can also be adjusted by the staff.

[0066] Specifically, define X as the historical database of the power communication network, including the power communication network alarm data, fault types, and each node of the power communication network and its related topological diagrams. Define the initial values ​​of the system: alarm attribute screening threshold R; dynamic variable scale convolution layer size L; improved convolutional neural network convergence threshold H; improved convolutional neural network maximum iteration number F; improved particle swarm algorithm global search inertia factor maximum value ω' max and the minimum value ω' min ; Improve the maximum number of iterations of the convolutional neural network T'; Improve the particle swarm algorithm adjustment coefficient α, α∈[0,1]; Improve the first acceleration constant β of the particle swarm algorithm 1 and the second acceleration constant β 2 ; Improve the speed coefficient λ of particle swarm algorithm.

[0067] (2) Filter initial attributes

[0068] In the large-scale and complex power communication network, a huge amount of alarm data is generated every day. However, not all alarm information is of practical value for identifying and solving the root fault. On the contrary, the mixed noise data will interfere with the maintenance personnel's accurate diagnosis of the root cause of the fault. The present invention uses the Pearson correlation coefficient to screen the attributes that are valuable for diagnosing the fault.

[0069] 1) Data Standardization

[0070] Since the dimensions or units of the variables are different, or their numerical ranges vary greatly, it is not conducive to more accurate calculation of the Pearson correlation coefficient. Therefore, each attribute variable needs to be standardized before calculating the Pearson correlation coefficient of each attribute. The present invention uses Z-score to standardize the data of the historical database X.

[0071] 2) Calculate the Pearson correlation coefficient

[0072] After data standardization is completed, the Pearson correlation coefficient r between each alarm attribute and the fault type is calculated. The specific formula is as follows:

[0073]

[0074] Among them, x i and i are the observed values ​​of alarm attributes and fault types, respectively. and are their means respectively, and n is the number of observations.

[0075] 3) Filter attributes

[0076] According to the value of the Pearson correlation coefficient r, the data set X1 with a correlation coefficient greater than the alarm attribute screening threshold R is finally selected as the data set X1 used next.

[0077] (3) Using the wavelet neural network model to weight the alarm data

[0078] The many attributes of the alarm data of the power communication network have different importance, and the contributions of attributes of different importance are different. Therefore, it is necessary to comprehensively consider the differences in the attributes of each alarm in combination with weights. Simply relying on subjective experience to determine the weights has no scientific basis. The present invention uses a wavelet neural network model to weight the alarm information. The weighting process is to first perform a wavelet transform on the alarm information data after selecting the initial attributes, and decompose the signal into wavelet coefficients at different frequencies and scales. Then, the structure of the wavelet neural network is designed, including the number and connection method of the neurons in the input layer, hidden layer, and output layer; then, the prepared data set is used to train and verify the wavelet neural network model; finally, the trained wavelet neural network model is used to perform weighted operations on the new alarm information to participate in subsequent model fault diagnosis. For specific steps, please refer to Figure 2 , the detailed steps are as follows:

[0079] 1) Screening training samples

[0080] From the data set X1 after the initial attributes are determined, group P is selected as the training sample set, and group Q is selected as the verification sample set.

[0081] 2) Wavelet neural network structure design

[0082] Design the structure of the wavelet neural network, including the number of neurons and the connection method of the input layer, hidden layer, and output layer. The hidden layer uses the Sigmoid activation function, and the output layer uses the linear activation function. The number of units in the input layer and the output layer are N and M respectively, and the output formula is as follows:

[0083]

[0084] In the formula, is the input and output of the pth sample; h represents the activation function; a j represents the size factor of the wavelet; d j represents the displacement factor; W jn1 represents the connection weight between the input layer and the jth hidden layer; V mj Represents the connection weight between the output layer and the jth hidden layer.

[0085] 3) Wavelet neural network model training

[0086] The back propagation algorithm is used for model training, where the standard error expression of the output value is as follows:

[0087]

[0088] Where E is the standard error of the output value, For the expected output of the pth sample, the gradient descent method is used to continuously train the model to find the appropriate displacement factor, scale factor and connection weight, so that the input layer (alarm information) and the output layer (alarm weight) have a better correspondence and obtain the optimal model.

[0089] 4) Model Validation

[0090] The validation set Q is used to verify the performance of the model. The present invention uses the mean square error (MSE) to calculate the error index between the predicted result and the true label. The specific formula is as follows:

[0091]

[0092] Where i = 1, 2, .. n2, n2 is the prediction length, is the actual value, y2 i The lower the value of this evaluation index, the better the model performance.

[0093] 5) Weighted operation of alarm information

[0094] The new alarm information data is input into the trained wavelet neural network model to obtain the output result, so that each alarm information is assigned a reasonable weight value, which reflects its importance and urgency in the current operation and maintenance scenario.

[0095] (4) Using improved convolutional neural network for fault diagnosis

[0096] In order to improve the efficiency of feature learning and reduce the impact of degradation, dynamic variable scale convolution is used to optimize the traditional convolutional neural network for fault diagnosis. The detailed steps are as follows:

[0097] 1) Construct a convolutional neural network with dynamic variable scale convolution optimization

[0098] ①Construction of dynamic variable scale convolution layer

[0099] The preprocessed alarm data is input into multiple parallel one-dimensional convolutional layers of the dynamic variable scale convolutional layer, and the number of channels of these parallel convolutional layers is the same but the convolution kernel size is different. The dynamic variable scale convolutional layer of the present invention is composed of one-dimensional convolutional layers of size L, 2L and 4L, each convolutional layer contains 8 convolution kernels, the step size is 1, and the value of L is dynamically adjusted during training. Padding is used to ensure that the size of the output of each convolutional layer is consistent. The output features of each convolutional layer are cascaded to obtain a feature vector for the input of the next pooling layer.

[0100] ②Pooling layer construction

[0101] The pooling layer is set after the dynamic variable scale convolution layer, and its function is feature mapping. In the pooling layer, the local features output by the dynamic variable scale convolution layer are weighted and added with bias, and new features are obtained based on nonlinear functions. After the local features output by the dynamic variable scale convolution layer are pooled, on the one hand, the feature map size is reduced, and the network calculation complexity is reduced; on the other hand, the features are further compressed to improve the accuracy of feature extraction.

[0102] ③Fully connected layer construction

[0103] The task of the fully connected layer is to represent the feature map obtained by the dynamic variable scale convolution and pooling operation with a one-dimensional feature vector, and then identify the pattern through the classifier. The classifier selected by the present invention is the Softmax logistic regression function. Let B = {b 1 ,b 2 ,.....,b T} represents the input sample set of the fully connected layer, let C = {c 1 ,c 2 ,....,c K} represents a supervised pattern recognition category set. In the Softmax logistic regression process, the input sample b i Classification: k The probability can be expressed as:

[0104]

[0105] In the formula, Represents the input sample b i With category c k The correlation between them, i = 1, 2, .. T, k = 1, 2, .. K, K represents the number of categories, and T represents the number of samples.

[0106] 2) Fault features are extracted through dynamic variable-scale convolution and pooling, and the output of the fully connected layer is compared with the actual value, and the error between the two is calculated through the loss function.

[0107] 3) Determine whether the network converges (less than the improved convolutional neural network convergence threshold H for convergence). If so, proceed to step 5); otherwise, proceed to step 4).

[0108] 4) Network back propagation and weight update

[0109] The error value calculated in step 2) is fed back to each layer of the network, and the weights of each layer are corrected. The back propagation algorithm is used to calculate the gradient, which represents the contribution of each weight to the network output error. Then, the weight of the network is updated according to the learning rate and gradient, and the error value is calculated again until the convergence condition is met.

[0110] 5) The loss function value is used to determine whether the network meets the requirements or reaches the maximum number of iterations F of the improved convolutional neural network. If it meets or reaches the maximum number of iterations of the improved convolutional neural network, proceed to step 6), otherwise proceed to step 2) to adjust and modify the model parameters.

[0111] 6) Fault diagnosis result output

[0112] The trained model is used for actual fault diagnosis tasks. For each sample to be diagnosed, it is input into the model and the prediction result output by the model is obtained. Then, the prediction result is matched with the possible fault type and the final fault diagnosis result is output, including the fault point location and fault type.

[0113] (5) Using an improved particle swarm algorithm to intelligently switch the path where the fault point is located

[0114] The network structure topology diagram involved in the fault point location in step (4) is found in the database, and the fault path where the fault point location is located is intelligently switched according to the fault type. The specific process is as follows:

[0115] 1) In the particle swarm algorithm, each faulty communication node load and disconnecting switch passed by a fault path are single particles. The initial velocity and position of the particle are obtained by random method. The fitness function takes the number of segmented switches of the particle, the path length and its line load value as key indicators.

[0116] 2) Calculate the global search inertia factor. Since the traditional particle swarm algorithm has limited global optimization ability and convergence speed, a fixed global search inertia factor will weaken the global optimization ability and reduce the convergence speed of the algorithm. This patent proposes to use a nonlinear variable inertia factor to improve the performance of the particle swarm algorithm. The formula is as follows:

[0117] ω'=ω' min -(ω' min -ω' max )α t-1 (6)

[0118] ω' represents the global search inertia factor, ω' max and ω' min are the maximum and minimum values ​​of the global search inertia factor, α is the adjustment coefficient, α∈[0,1], and t is the number of iterations of the improved particle swarm algorithm.

[0119] 3) Update particle velocity using improved global search inertia factor and location The formula is:

[0120]

[0121] β 1 , β 2 To improve the first and second acceleration constants of the particle swarm algorithm, They represent the individual extreme value of i'-dimensional particles and the optimal solution of global variables, and are the current moving speed and location respectively, λ is the speed coefficient, and r represents a random number between 0 and 1.

[0122] 4) Determine whether the number of repetitions has exceeded the maximum number of iterations T' of the improved particle swarm algorithm. If so, output the next best position of the switching path; if not, repeat steps 2) and 3) to continue iterating.

[0123] On the other hand, the present invention further proposes a power communication network intelligent switching system, which is used to implement the above-mentioned power communication network intelligent switching method, comprising:

[0124] Data storage module, used to build the power communication network history database and define the system initial value;

[0125] Data preprocessing module, used to standardize historical data, calculate the Pearson correlation coefficient between each alarm data and fault type, and screen initial attributes;

[0126] A data weighting module is used to weight the alarm data using a wavelet neural network model;

[0127] A fault diagnosis module, used for performing fault diagnosis on weighted alarm data;

[0128] Intelligent switching module, used to use improved particle swarm algorithm to intelligently switch the path where the fault point is located;

[0129] The output module is used to output the fault point location, fault type and intelligent switching suggestions.

[0130] In order to illustrate the implementation effect of the present invention, the power communication network fault diagnosis method (WNN-SCNN) proposed in the present invention is experimentally compared with traditional convolutional neural networks (CNN), recurrent neural networks (RNN) and Apriori algorithm. Figure 3 This is a rendering of the fault diagnosis accuracy of the present invention and the traditional convolutional neural network, recurrent neural network and Apriori algorithm under the same conditions. It can be seen that the power communication network fault diagnosis method proposed by the present invention fully considers the impact of alarm attributes and traditional convolutional neural networks on fault diagnosis. By screening and weighting the alarm information and using dynamic variable scale convolution to optimize the traditional convolutional neural network, the pertinence of feature extraction is significantly improved, and the efficiency and accuracy of power communication network fault diagnosis are effectively improved. The improved particle swarm algorithm uses a nonlinear variable inertia factor to improve the performance of the particle swarm algorithm. The fitness function uses the number of segmented switches, path length and line load value of the particle as key indicators. These innovative measures not only improve the fault recovery speed, but also make the power system more robust after recovery.

[0131] The beneficial effects of the present invention are as follows: the present invention provides a scientific and reasonable method for intelligent switching of power communication networks, which can greatly improve the fault recovery speed of the power communication network and reduce power grid losses.

[0132] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0133] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent switching of a power communication network, characterized in that: The following steps are involved: Build a historical database for the power communication network and define the initial values ​​of the system; Pearson correlation coefficient was used to screen initial attributes; Using wavelet neural network model to weight the alarm data; Use improved convolutional neural network for fault diagnosis; An improved particle swarm algorithm is used to intelligently switch the path where the fault point is located.

2. A method for intelligent switching of a power communication network according to claim 1, characterized in that: The process of using the Pearson correlation coefficient to screen the initial attributes is specifically as follows: (1) Use Z-score to standardize historical database data; (2) Calculate the Pearson correlation coefficient r between each alarm attribute and the fault type. The specific formula is as follows: Among them, x i and i are the observed values ​​of alarm attributes and fault types, respectively. and are their means respectively, n is the number of observations; (3) According to the value of the Pearson correlation coefficient r, the data with a correlation coefficient greater than the alarm attribute screening threshold R is finally selected as the data to be used next.

3. The method for intelligent switching of a power communication network according to claim 1, characterized in that: The specific process of weighting the alarm data using the wavelet neural network model is as follows: (1) Select group P as the training sample set and group Q as the validation sample set from the historical data set after the initial attributes are determined; (2) Design the structure of the wavelet neural network, including the number and connection method of neurons in the input layer, hidden layer and output layer. The hidden layer uses the Sigmoid activation function and the output layer uses the linear activation function. The number of units in the input layer and output layer are N and M respectively. The output formula is as follows: In the formula, is the input and output of the pth sample; h represents the activation function; a j represents the size factor of the wavelet; d j represents the displacement factor; W jn1 represents the connection weight between the input layer and the jth hidden layer; V mj represents the connection weight between the output layer and the jth hidden layer; (3) The back propagation algorithm is used for model training, where the standard error expression of the output value is as follows: Where E is the standard error of the output value, For the expected output of the pth sample, the gradient descent method is used to continuously train the model to find the appropriate displacement factor, scale factor and connection weight, so that the input layer (alarm information) and the output layer (alarm weight) have a better correspondence and obtain the optimal model; (4) Use the validation set Q to verify the performance of the model, and use the mean square error (MSE) to calculate the error index between the predicted result and the true label. The specific formula is as follows: Where i = 1, 2, .. n2, n2 is the prediction length, is the actual value, y2 i is the predicted value. The lower the value of this evaluation index, the better the model performance. (5) The new alarm information data is input into the trained wavelet neural network model to obtain the output result, so that each alarm information is assigned a reasonable weight value, which reflects its importance and urgency in the current operation and maintenance scenario.

4. The method for intelligent switching of a power communication network according to claim 1, characterized in that: The fault diagnosis process using the improved convolutional neural network is specifically as follows: (1) Construct a convolutional neural network with dynamic variable scale convolution optimization; (2) Extract fault features through dynamic variable-scale convolution and pooling, and compare the output of the fully connected layer with the actual value to obtain the error value between the two; (3) Determine whether the network converges (less than the improved convolutional neural network convergence threshold H for convergence). If so, proceed to step (5); otherwise, proceed to step (4); (4) Network back propagation and weight update: use the back propagation algorithm to calculate the gradient, update the network weights according to the learning rate and gradient, and calculate the error value again until the convergence condition is met or the maximum number of iterations F of the improved convolutional neural network is reached; (5) judging whether the network meets the requirements or reaches the maximum number of iterations F of the improved convolutional neural network through the loss function value; if the network meets or reaches the maximum number of iterations F of the improved convolutional neural network, the process goes to step (6); otherwise, the process goes to step (2) to adjust and modify the model parameters; (6) Output fault diagnosis results.

5. A method for intelligent switching of a power communication network according to claim 4, characterized in that: The process of constructing the dynamic variable scale convolution optimized convolutional neural network is specifically as follows: (1) Dynamically scaled convolutional layer construction. The dynamically scaled convolutional layer of the present invention is composed of one-dimensional convolutional layers of sizes L, 2L and 4L. Each convolutional layer contains 8 convolution kernels with a step size of 1. The value of L is dynamically adjusted during training. Padding is used to ensure that the output size of each convolutional layer is consistent. The output features of each convolutional layer are cascaded to obtain a feature vector for input to the next pooling layer. (2) Pooling layer construction: The pooling layer is set after the dynamic variable scale convolution layer. Its function is feature mapping. In the pooling layer, the local features output by the dynamic variable scale convolution layer are weighted summed and biased, and new features are obtained based on the nonlinear function. (3) Construction of the fully connected layer. The task of the fully connected layer is to represent the feature map obtained by the dynamic variable scale convolution and pooling operation with a one-dimensional feature vector, and then recognize the pattern through the classifier. The classifier selected by the present invention is the Softmax logistic regression function. Let B = {b1, b2, ..., b T } represents the input sample set of the fully connected layer, let C = {c1,c2,....,c K } represents a supervised pattern recognition category set. In the Softmax logistic regression process, the input sample b i Classification: k The probability can be expressed as: In the formula, Represents the input sample b i With category c k The correlation between them, i = 1, 2, .. T, k = 1, 2, .. K, K represents the number of categories, and T represents the number of samples.

6. The method for intelligent switching of a power communication network according to claim 1, characterized in that: The improved particle swarm algorithm is used to intelligently switch the path where the fault point is located. The specific process is as follows: (1) In the particle swarm algorithm, each faulty communication node load and disconnecting switch that a fault path passes through is a single particle. The initial velocity and position of the particle are obtained by random method. The fitness function uses the number of segmented switches of the particle, the path length and its line load value as key indicators. (2) Calculate the global search inertia factor. Since the traditional particle swarm algorithm has limited global optimization ability and convergence speed, a fixed global search inertia factor will weaken the global optimization ability and reduce the convergence speed of the algorithm. This patent proposes to use a nonlinear variable inertia factor to improve the performance of the particle swarm algorithm. The formula is as follows: oh'=oh' min -(oh' min -oh' max )a t-1 ω' represents the global search inertia factor, ω' max and ω' min are the maximum and minimum values ​​of the global search inertia factor, α is the adjustment coefficient, α∈[0,1], and t is the number of iterations of the improved particle swarm algorithm; (3) Update particle velocity using the improved global search inertia factor and location The formula is: β1 and β2 are the first and second acceleration constants of the improved particle swarm algorithm, They represent the individual extreme value of i'-dimensional particles and the optimal solution of global variables, and are the current moving speed and location respectively, λ is the speed coefficient, and r represents a random number between 0 and 1; (4) Determine whether the number of repetitions has exceeded the maximum number of iterations T' of the improved particle swarm algorithm. If so, output the next best position of the switching path; if not, repeat steps (2) and (3) to continue iterating.

7. An intelligent switching system for a power communication network, used to implement the above-mentioned intelligent switching method for a power communication network, characterized in that: include: Data storage module, used to build the power communication network history database and define the system initial value; Data preprocessing module, used to standardize historical data, calculate the Pearson correlation coefficient between each alarm data and fault type, and screen initial attributes; A data weighting module is used to weight the alarm data using a wavelet neural network model; A fault diagnosis module, used for performing fault diagnosis on weighted alarm data; Intelligent switching module, used to use improved particle swarm algorithm to intelligently switch the path where the fault point is located; The output module is used to output the fault point location, fault type and intelligent switching suggestions.

Images