Charging pile state prediction method, computer readable medium and equipment

Abstract

The invention provides a charging pile state prediction method, a computer readable medium and equipment, and the method is a health index and double attention optimization GCN charging pile state prediction method, and comprises the steps of load data acquisition, dynamic wavelet noise reduction, health index construction and health state prediction. According to the principle, firstly, load data of a charging pile are collected and processed, data integration is conducted on server platform data and voltage, current and power data in a charging station, and data processing is conducted; secondly, performing noise reduction processing on the original data by utilizing dynamic threshold wavelet transform to obtain a charging pile state data matrix; thirdly, based on the voltage, current and power matrix of the charging pile load data, health factors of a single state index of the charging pile are constructed through a distance sparseness algorithm, and the health factors are fused into a health index according to an improved objective weighting method; and finally, predicting the health state of the charging pile by adopting a double-attention optimization graph convolutional neural network.

Description

Technical Field

[0001] The present application relates to the technical field of charging piles, and in particular to a charging pile state prediction method, computer-readable medium and device. Background Art

[0002] In the face of the rapidly developing and large-scale electric vehicle charging pile market, it is necessary to implement standardized and normalized management of electric vehicle charging equipment. The scientific management and comprehensive evaluation of existing charging piles have become increasingly prominent. Charging pile status prediction based on status monitoring and on-site inspection enables charging station managers and charging pile maintenance and repair personnel to timely understand and grasp the operating status of charging piles, and use this as a basis for decision-making to guide the daily maintenance of charging piles, conduct rapid diagnosis and precise maintenance, determine the degree of charging pile failure, and formulate economical, reliable, reasonable and efficient maintenance and periodic maintenance plans for them, which will help achieve the ideal goal of extending the service life of charging piles, take positive measures in advance to do a good job in fault warning and hidden danger investigation, and help reduce the hidden failure rate of existing charging piles.

[0003] However, due to noise interference and the accuracy of the algorithm itself, the prediction accuracy of the charging pile status needs to be improved. Summary of the invention

[0004] The present application provides a charging pile state prediction method, computer-readable medium and device, which have the advantage of achieving more accurate state prediction of the charging pile.

[0005] The technical solution of this application is as follows:

[0006] On the one hand, the present application provides a charging pile state prediction method, comprising the following steps:

[0007] Load data acquisition: Collect the power load of individual charging piles to obtain charging pile load data;

[0008] Dynamic wavelet noise reduction: wavelet transform is used to reduce the noise of the obtained charging pile load data;

[0009] Health index construction: After obtaining the charging pile load index data after noise reduction, the distance sparsity algorithm DSA is used to construct the health factor of a single load index according to the distance distribution of the data. Based on the improved objective weighting method CRITIC, multiple index health factors are integrated into the health index of the charging pile operation status.

[0010] Health status prediction: After obtaining the health index of the charging pile, the original data and the health index are integrated to construct a multi-dimensional input data matrix D n×mThe topological structure of the graph constructed by the graph convolutional neural network GCN is used to identify the correlation between different features, and the local attention mechanism and the global attention mechanism are combined to extract the long-term and short-term temporal dependencies of the time series, thereby realizing accurate status prediction of the charging pile.

[0011] Furthermore, in the load data acquisition step, the charging pile load data includes voltage, current and power data at the input / output end of the charging pile.

[0012] Furthermore, the method of dynamic wavelet denoising is as follows: Assume that the noisy signal can be expressed as:

[0013] x(t)=s(t)+n(t)

[0014] Where x(t) is the original signal, s(t) is the characteristic signal, and n(t) is the noise signal composed of various interferences;

[0015] Wavelet transform decomposes the signal into a superposition of a series of wavelet functions. Wavelet transform provides decomposition in scale and position, or frequency and time through a series of convolution operations:

[0016]

[0017] Where W represents the wavelet coefficient, a is the scaling parameter inversely proportional to the center frequency; b is the translation parameter of the positioning signal; ψ * (t) is the complex conjugate of the wavelet basis ψ(t);

[0018] Determine the dynamic hard threshold and filter the decomposed wavelet coefficients through the dynamic hard threshold:

[0019]

[0020] where H∈R L is the judgment coefficient, L is the length of the wavelet coefficient, and φ is the feature judgment output;

[0021] Calculate the feature judgment output φ: For the scale dependency of the wavelet coefficients, extract the scale feature information through one-dimensional convolution, and then obtain the final output φ of the feature judgment through Softmax(·) i :

[0022] φ i =Softmax(CNN(W i ))

[0023] Then, the feature judgment output φ i Convert it into an optimizable threshold judgment coefficient H, and directly convert the feature decision into a hard operation through logical judgment, using Gumbel distribution and Softmax function to sample on the probability distribution:

[0024] h i =Softmax(((log(φ i )+g i ) / ε),i=0,1

[0025]

[0026] where g0 and g1 are samples drawn from the Gumbel(0,1) distribution, and ε is the control distribution h i The smoothness function, Returns the sampling result of Gumbel-Softmax; forward represents the feedforward stage; backward represents the back-propagation stage.

[0027] Furthermore, the distance sparsity algorithm DSA determines the probability distribution of the indicator under normal conditions based on the logarithmic distance between the data, and thus determines the operating status of the indicator through the probability range;

[0028] Probability density calculation of distance sparsity:

[0029]

[0030]

[0031]

[0032]

[0033]

[0034] Where v is the base of the logarithmic function, satisfying v∈R,v>0,v≠1,μ is the average value of the transformed indicator data, σ is the standard deviation, and f(z) is the Gaussian probability density function of z;

[0035] Perform numerical conversion calculations:

[0036]

[0037] H i ∈(0,1) and is the health factor of a single indicator. The closer it is to 1, the more normal the operating status of the load indicator is; the closer it is to 0, the more abnormal the operating status of the load indicator is.

[0038] Furthermore, the health status prediction step includes: original load data Fusion with health index HI to construct multidimensional input matrix D n×m , the obtained multidimensional data matrix D n×mInput into the graph convolutional neural network GCN for feature extraction, multidimensional data matrix D n×m for:

[0039]

[0040] Among them, the length of the original load data is n, and the dimension is m-1. Including the collected current, voltage and power data; the charging pile health index HI is expressed as [d 1m ,d 2m ...d nm ] T ;

[0041] The data attribute graph G = (V, E), which consists of a node set V and an edge set E. i ∈V represents a node of V, e ij =(v i ,v j )∈E, indicating the edge from node i to node j, N(v i )={u∈V|(v i ,u)∈E} is the neighborhood of node i, and the adjacency matrix A∈R N×N As shown in the following formula, the feature matrix X∈R n×d ,x i ∈R d , represents the feature vector of node i;

[0042]

[0043] The space-time graph is represented as a discrete time dynamic graph, that is, G = {G1, G2..., G T}, the calculation method of the graph convolutional neural network is:

[0044]

[0045] I represents the input feature of each layer. For the initial input layer, feature I is D, so I l represents the input feature of the lth layer of the network, σ is the nonlinear activation function, W l represents the weight matrix of the layer, D represents the degree matrix of the graph, and A represents the adjacency matrix;

[0046] Secondly, the output of the GCN layer is transmitted to the LSTM network with a dual attention mechanism to further explore the temporal correlation of features; the dual attention mechanism includes an LSTM network with a local attention mechanism and an LSTM network with a global attention mechanism. The formula for calculating the local attention value can be written as:

[0047]

[0048]

[0049]

[0050] Among them, s refers to the window size of the local attention mechanism; Is the final output of the LSTM network during state prediction and the hidden state at time j The matching score between is the trainable weight matrix, The context vector of a single-layer neural network representing the input linear activation function;

[0051] The formula for calculating the global attention value can be written as:

[0052]

[0053]

[0054]

[0055] in, is the trainable weight matrix, Is the final output of the LSTM network during state prediction and the hidden state at time j The matching score between The context vector of a single-layer neural network representing the input linear activation function;

[0056] Finally, the local attention value and the global attention value are input into a feedforward neural network with a linear activation function to predict the health status. The optimization function is as follows:

[0057]

[0058] g(·) represents the ReLu() activation function, b is the bias, W is the trainable weight matrix in the neural network, L is the loss function in the neural network, and Y is the health index of the charging pile.

[0059] On the other hand, the present application provides a computer-readable medium, which stores a computer program. When the computer program is called by a computer, it executes the charging pile state prediction method described in the above solution.

[0060] On the other hand, the present application provides a device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is called by the processor, the charging pile state prediction method described in the above scheme is executed.

[0061] In summary, the beneficial effects of this application are:

[0062] 1. The present invention performs dynamic threshold wavelet transform on the voltage, current and power data obtained from the charging pile. The threshold function reflects different processing strategies for wavelet coefficients exceeding and below the threshold. The noise-related components are dynamically eliminated through point-by-point hard thresholds to reduce most of the noise coefficients and retain key information.

[0063] 2. Based on the charging pile load index data after noise reduction, the present invention constructs the health factor of a single load index according to the distance distribution of the data through the distance sparsity algorithm (DSA), and uses the improved objective weighting method (CRITIC) to integrate multiple index health factors into the health index of the charging pile operation status. The health index can reflect the real-time status of the charging pile;

[0064] 3. The present invention adopts the topological structure of the graph constructed by the graph convolutional neural network (GCN) to identify the correlation between different features, introduces the LSTM network with a dual attention mechanism to predict the time series, and combines the local attention mechanism and the global attention mechanism to extract the long-term and short-term dependencies of the time series, thereby realizing accurate state prediction of the charging pile. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flow chart of the charging pile state prediction method of the present application;

[0066] Figure 2 It is a schematic diagram of the dynamic threshold wavelet denoising model of the present application;

[0067] Figure 3 It is a schematic diagram of the property graph structure;

[0068] Figure 4 This is a schematic diagram of the dual attention mechanism optimized GCN health status prediction model in this application. DETAILED DESCRIPTION

[0069] The specific implementation of the present application is described in detail below with reference to the accompanying drawings.

[0070] Embodiment: The embodiment of the present application provides a charging pile state prediction method, which is a GCN charging pile state prediction method that integrates health index and dual attention optimization. The process is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0071] Step (1): Load data acquisition

[0072] The data collection of the present invention mainly collects the power load conditions of the individual charging piles through the collector inside the charging piles, and the load data includes the voltage, current and power data of the input / output ends of the charging piles.

[0073] Step (2): Dynamic wavelet denoising

[0074] like Figure 2 As shown in the figure, after obtaining the charging pile load data, wavelet transform is used to reduce the noise of the data. Wavelet transform is the process of filtering and denoising the signal according to the frequency distribution characteristics of the signal and noise. The decomposed wavelet coefficients are filtered through the wavelet basis, the number of decomposition layers, the appropriate threshold and the threshold function. The effective signal is retained to the maximum extent, the noise is eliminated to the maximum extent, and finally a more effective signal is obtained.

[0075] (2.1) Wavelet function

[0076] Assume that the noisy signal can be expressed as:

[0077] x(t)=s(t)+n(t)

[0078] Where x(t) is the original signal, s(t) is the characteristic signal, and n(t) is the noise signal composed of various interferences.

[0079] Wavelet transform decomposes the signal into a superposition of a series of wavelet functions. Wavelet transform provides decomposition in terms of scale and position, or frequency and time through a series of convolution operations.

[0080]

[0081] Where W represents the wavelet coefficient, a is a scaling parameter inversely proportional to the center frequency, b is the translation parameter of the positioning signal, ψ * (t) is the complex conjugate of the wavelet basis ψ(t).

[0082] (2.2) Dynamic Hard Threshold

[0083] After wavelet transform, the coefficients of n(t) are small and evenly distributed, while the coefficients of s(t) are concentrated. Therefore, setting the coefficients of noise to zero is the core idea of ​​thresholding. The key point of thresholding is to determine which coefficients should be retained (features) and which coefficients should be removed (noise). The threshold can be expressed as:

[0084]

[0085] where H∈R L is the judgment coefficient, L is the length of the wavelet coefficient, and φ is the feature judgment output.

[0086] First, the feature judgment output φ needs to be calculated. For the intra-scale dependence of the wavelet coefficients, the present invention extracts the intra-scale feature information through one-dimensional convolution, and then obtains the final output φ of the feature judgment through Softmax(·) i :

[0087] φ i =Softmax(CNN(Wi ))

[0088] Then, the feature judgment output φ i Convert it into an optimizable threshold judgment coefficient H, and directly convert the feature decision into a hard operation through logical judgment, using Gumbel distribution and Softmax function to sample on the probability distribution:

[0089] h i =Softmax(((log(φ i )+g i ) / ε),i=0,1

[0090]

[0091] where g0 and g1 are samples drawn from the Gumbel(0,1) distribution, and ε is the control distribution h i The smoothness function, Returns the sampling result of Gumbel-Softmax; forward represents the feedforward stage; backward represents the back-propagation stage.

[0092] Step (3): Health index construction

[0093] (3.1) Distance sparsity algorithm

[0094] After obtaining the charging pile load index data after noise reduction, the distance sparsity algorithm (DSA) is used to construct the health factor of a single load index according to the distance distribution of the data. Based on the improved objective weighting method (CRITIC), multiple index health factors are integrated into the health index of the charging pile operation status.

[0095] The distance sparsity algorithm (DSA) determines the probability distribution of the indicator under normal conditions based on the logarithmic distance between the data, and thus determines the operating status of the indicator under the probability range.

[0096] Probability density calculation of distance sparsity:

[0097]

[0098]

[0099]

[0100]

[0101]

[0102] Where v is the base of the logarithmic function, satisfying v∈R,v>0,v≠1,μ is the average value of the transformed indicator data, σ is the standard deviation, and f(z) is the Gaussian probability density function of z.

[0103] Since the present invention focuses on the abnormal conditions of overcurrent, overvoltage, overload, undercurrent and undervoltage of the charging pile, the data are all intermediate indicator data, and the health index needs to be constructed as extremely large indicator data, so numerical conversion calculation is required:

[0104]

[0105] H i ∈(0,1) and is the health factor of a single indicator. The closer it is to 1, the more normal the operating status of the load indicator is; the closer it is to 0, the more abnormal the operating status of the load indicator is.

[0106] (3.2) Improved objective weighting method

[0107] After obtaining the health factors of several indicators, construct the health factor matrix H = [H1, H2...H n ] T The present invention adopts an improved objective weighting method to weight and fuse H into a health index HI that reflects the overall status of the charging pile.

[0108] The CRITIC method uses contrast intensity and index conflict to determine the weight of the evaluation index. The present invention improves the traditional CRITIC weight method and selects the coefficient of variation as the contrast intensity for calculation. The larger the coefficient of variation, the more obvious the fluctuation of the index; the conflict between the indicators is measured by the distance correlation coefficient between the indicators. If the correlation coefficient between the two indicators is larger, it means that the positive correlation between the indicators is higher, and the conflict between the two indicators is lower.

[0109] First, calculate the coefficient of variation:

[0110]

[0111] Where V j is the coefficient of variation of the jth indicator, σ j is the standard deviation of the jth term, is the average of the jth item.

[0112] Secondly, calculate the distance correlation coefficient r and the conflict quantitative index value:

[0113] dcov(h i ,h j )=E(|h i -h' i ||h j -h' j |)+E(|hi -h' i |)E(|h j -h' j |)-2E(|h i -h" i ||h j -h" j |)

[0114]

[0115] Among them, h i ,h j represents a vector of indicator data of length n, (h i ,h j ),(h' i ,h' j ),(h” i ,h” j ) are random variables that follow the same distribution and are independent of each other.

[0116] The quantitative index value of the conflict between the jth index and other indexes is:

[0117]

[0118] According to the weight determination principle of the CRITIC method, the final objective weight of the indicator contains the comparison strength and conflict information between indicators. j represents the amount of information contained in the jth evaluation index, then C j It can be expressed as:

[0119]

[0120] Finally, calculate the indicator weight, the indicator's C j The larger the value, the greater the amount of information, and the corresponding indicator weight value is also larger. The objective weight calculation formula of the jth indicator is as follows:

[0121]

[0122] After obtaining the health factor matrix H = [H1,H2...H n ] T And its corresponding objective weight parameter W=[w1,w2...w n ], the health index is

[0123] Step (4) Health status prediction

[0124] After obtaining the health index HI of the charging pile, the original data is fused with HI to construct a multi-dimensional input data matrix D n×mThe topological structure of the graph constructed by the graph convolutional neural network (GCN) is used to identify the correlation between different features, and the local attention mechanism and the global attention mechanism are combined to extract the long-term and short-term temporal dependencies of the time series, so as to achieve accurate status prediction of the charging pile.

[0125] Raw load data Fusion with health index HI to construct multidimensional input matrix D n×m , the obtained multidimensional data matrix D n×m Input into the graph convolutional neural network GCN for feature extraction, multidimensional data matrix D n×m for

[0126]

[0127] Among them, the length of the original load data is n, and the dimension is m-1. Including current, voltage and power data collected by the system; the charging pile health index HI is expressed as [d 1m ,d 2m ...d nm ] T .

[0128] First, the obtained multidimensional data matrix D n×m The input is fed into a graph convolutional neural network (GCN) for feature extraction. GCN is a neural network suitable for analyzing graph data. It takes graph structure and node information as the input of the model. The output type of GCN can be defined according to different downstream application tasks. GCN relies on the relationship between nodes to obtain information in the graph and optimizes the message passing mechanism between graph nodes to obtain the dependency relationship between different nodes. It has become a widely used graph analysis method.

[0129] like Figure 3 As shown, the attribute graph G = (V, E) consists of a node set V and an edge set E. i ∈V represents a node of V, e ij =(v i ,v j )∈E, indicating the edge from node i to node j, N(v i )={u∈V|(v i ,u)∈E} is the neighborhood of node i, and the adjacency matrix A∈R N×N As shown in the following formula, the feature matrix X∈R n×d ,x i ∈R d , represents the feature vector of node i.

[0130]

[0131] The space-time graph is represented as a discrete time dynamic graph, that is, G = {G1, G2..., G T}, the calculation method of the graph convolutional neural network is:

[0132]

[0133] I represents the input feature of each layer. For the initial input layer, feature I is D, so I l represents the input feature of the lth layer of the network, σ is the nonlinear activation function, W l represents the weight matrix of the layer, D represents the degree matrix of the graph, and A represents the adjacency matrix.

[0134] Secondly, if Figure 4 As shown in Figure 1, the output of the GCN layer is transferred to the LSTM network with a dual attention mechanism to further explore the temporal correlation of features. The dual attention mechanism includes an LSTM network with a local attention mechanism and an LSTM network with a global attention mechanism. The formula for calculating the local attention value can be written as:

[0135]

[0136]

[0137]

[0138] Among them, s refers to the window size of the local attention mechanism; Is the final output of the LSTM network during state prediction and the hidden state at time j The matching score between is the trainable weight matrix, Context vector for a single-layer neural network that represents the input to a linear activation function.

[0139] The formula for calculating the global attention value can be written as:

[0140]

[0141]

[0142]

[0143] in, is the trainable weight matrix, Is the final output of the LSTM network during state prediction and the hidden state at time j The matching score between Context vector for a single-layer neural network that represents the input to a linear activation function.

[0144] Finally, the local attention value and the global attention value are input into a feedforward neural network with a linear activation function to predict the health status. The optimization function is as follows:

[0145]

[0146] g(·) represents the ReLu() activation function, b is the bias, W is the trainable weight matrix in the neural network, L is the loss function in the neural network, and Y is the health index of the charging pile.

[0147] An embodiment of the present application further provides a computer-readable medium, wherein the computer-readable medium stores a computer program, and when the computer program is called by a computer, the charging pile state prediction method described in the above scheme is executed.

[0148] An embodiment of the present application further provides a device, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is called by the processor, the charging pile state prediction method described in the above scheme is executed.

[0149] The above is only a preferred implementation of the present application. It should be pointed out that a person skilled in the art can make several modifications and improvements without departing from the inventive concept of the present application, and these all fall within the scope of protection of the present application.

Claims

1. A charging pile state prediction method, characterized in that: The steps include: Load data acquisition: Collect the power load of individual charging piles to obtain charging pile load data; Dynamic wavelet noise reduction: wavelet transform is used to reduce the noise of the obtained charging pile load data; Health index construction: After obtaining the charging pile load index data after noise reduction, the distance sparsity algorithm DSA is used to construct the health factor of a single load index according to the distance distribution of the data. Based on the improved objective weighting method CRITIC, multiple index health factors are integrated into the health index of the charging pile operation status. Health status prediction: After obtaining the health index of the charging pile, the original data is fused with the health index to construct a multi-dimensional input data matrix. The topological structure of the graph constructed by the graph convolutional neural network GCN is used to identify the correlation between different features. The local attention mechanism and the global attention mechanism are combined to extract the long-term and short-term time dependencies of the time series, thereby realizing accurate status prediction of the charging pile.

2. The charging pile state prediction method according to claim 1, characterized in that: In the load data acquisition step, the charging pile load data includes the voltage, current and power data of the charging pile input / output ends.

3. The charging pile state prediction method according to claim 1, characterized in that: The method of dynamic wavelet denoising is: Assume that the noisy signal can be expressed as: x(t)=s(t)+n(t) Where x(t) is the original signal, s(t) is the characteristic signal, and n(t) is the noise signal composed of various interferences; Wavelet transform decomposes the signal into a superposition of a series of wavelet functions. Wavelet transform provides decomposition in scale and position, or frequency and time through a series of convolution operations: Where W represents the wavelet coefficient, a is the scaling parameter inversely proportional to the center frequency; b is the translation parameter of the positioning signal; ψ * (t) is the complex conjugate of the wavelet basis ψ(t); Determine the dynamic hard threshold and filter the decomposed wavelet coefficients through the dynamic hard threshold: where H∈R L is the judgment coefficient, L is the length of the wavelet coefficient, and φ is the feature judgment output; Calculate the feature judgment output φ: For the scale dependency of the wavelet coefficients, extract the scale feature information through one-dimensional convolution, and then obtain the final output φ of the feature judgment through Softmax(·) i : f i =Softmax(CNN(W i )) Then, the feature judgment output φ i Convert it into an optimizable threshold judgment coefficient H, and directly convert the feature decision into a hard operation through logical judgment, using Gumbel distribution and Softmax function to sample on the probability distribution: h i =Softmax(((log(φ i )+g i ) / ε),i=0.1 where g0 and g1 are samples drawn from the Gumbel(0,1) distribution, and ε is the control distribution h i The smoothness function, Returns the sampling result of Gumbel-Softmax; forward represents the feedforward stage; backward represents the back-propagation stage.

4. The charging pile state prediction method according to claim 3, characterized in that: The distance sparsity algorithm DSA determines the probability distribution of the indicator under normal conditions based on the logarithmic distance between the data, and thus determines the operating status of the indicator through the probability range; Probability density calculation of distance sparsity: Where v is the base of the logarithmic function, satisfying v∈R,v>0,v≠1,μ is the average value of the transformed indicator data, σ is the standard deviation, and f(z) is the Gaussian probability density function of z; Perform numerical conversion calculations: H i ∈(0,1) and is the health factor of a single indicator. The closer it is to 1, the more normal the health of the load indicator operation status is. The closer the distance is to 0, the more abnormal the operating status of the load indicator is.

5. The charging pile state prediction method according to claim 4, characterized in that: After obtaining the health factors of several indicators, construct the health factor matrix H = [H1, H2...H n ] T ; The improved objective weighting method is used to weight H into a health index HI that reflects the overall status of the charging pile: First, calculate the coefficient of variation: Where V j is the coefficient of variation of the j-th indicator, σ j is the standard deviation of the jth term, is the average of the jth item; Secondly, calculate the distance correlation coefficient r and the conflict quantitative index value: dcov(h i ,h j )=E(|h i -h′ i ||h j -h′ j |)+E(|h i -h′ i |)E(|h j -h′ j |)-2E(|h i -h″ i ||h j -h′ j |) dCov 2 (h i )=(1 / T 2 )∑ j,k ((h ij -h ik )-∑ k (h ij -h ik )-∑ j (h ij -h ik )+∑ j,k (h ij -h ik )) 2 Among them, h i ,h j represents a vector of indicator data of length n, (h i ,h j ),(h′ i ,h′ j ),(h″ i ,h″ j ) are random variables that follow the same distribution and are independent of each other; The quantitative index value of the conflict between the jth index and other indexes is: Let C j represents the amount of information contained in the jth evaluation index, then C j It can be expressed as: Finally, calculate the indicator weight, the indicator's C j The larger the value, the greater the amount of information, and the corresponding indicator weight value is also larger. The objective weight calculation formula of the jth indicator is as follows: After obtaining the health factor matrix H = [H1,H2...H n ] T And its corresponding objective weight parameter W=[w1,w2...w n ], the health index is 6. The charging pile state prediction method according to claim 1, characterized in that: The health status prediction step includes: original load data Fusion with health index HI to construct multidimensional input matrix D n×m , the obtained multidimensional data matrix D n×m Input into the graph convolutional neural network GCN for feature extraction, multidimensional data matrix D n×m for: Among them, the length of the original load data is n, and the dimension is m-1. Including the collected current, voltage and power data; the charging pile health index HI is expressed as [d 1m ,d 2m ...d nm ] T ; The data attribute graph G = (V, E), which consists of a node set V and an edge set E. i ∈V represents a node of V, e ij =(v i ,v j )∈E, indicating the edge from node i to node j, N(v i )={u∈V|(v i ,u)∈E} is the neighborhood of node i, and the adjacency matrix A∈R N×N As shown in the following formula, the feature matrix X∈R n×d ,x i ∈R d , represents the feature vector of node i; The space-time graph is represented as a discrete time dynamic graph, that is, G = {G1, G2..., G T }, the calculation method of the graph convolutional neural network is: I represents the input feature of each layer. For the initial input layer, feature I is D, so I l represents the input feature of the lth layer of the network, σ is the nonlinear activation function, W l represents the weight matrix of the layer, D represents the degree matrix of the graph, and A represents the adjacency matrix; Secondly, the output of the GCN layer is transmitted to the LSTM network with a dual attention mechanism to further explore the temporal correlation of features; the dual attention mechanism includes an LSTM network with a local attention mechanism and an LSTM network with a global attention mechanism. The formula for calculating the local attention value can be written as: Among them, s refers to the window size of the local attention mechanism; Is the final output of the LSTM network during state prediction and the hidden state at time j The matching score between is the trainable weight matrix, The context vector of a single-layer neural network representing the input linear activation function; The formula for calculating the global attention value can be written as: in, is the trainable weight matrix, Is the final output of the LSTM network during state prediction and the hidden state at time j The matching score between The context vector of a single-layer neural network representing the input linear activation function; Finally, the local attention value and the global attention value are input into a feedforward neural network with a linear activation function to predict the health status. The optimization function is as follows: g(·) represents the ReLu() activation function, b is the bias, W is the trainable weight matrix in the neural network, L is the loss function in the neural network, and Y is the health index of the charging pile.

7. A computer-readable medium, characterized in that The computer-readable medium stores a computer program, and when the computer program is called by a computer, the charging pile state prediction method according to any one of claims 1 to 6 is executed.

8. A device, characterized in that It includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is called by the processor, the charging pile state prediction method according to any one of claims 1 to 6 is executed.

Images