Charging pile fault intelligent diagnosis method and system

Abstract

The invention discloses a charging pile fault intelligent diagnosis method and system. The method comprises the following steps: acquiring historical operation work order data of a fault charging pile by using a historical charging operation work order; feature factors of the historical operation work order data of the fault charging pile are extracted, and the feature factors are preprocessed; constructing a training feature data set according to the preprocessed feature factors, and training a bidirectional long-short term memory recurrent neural network model and a residual network model based on the training feature data set; a trained bidirectional long-short-term memory recurrent neural network model and a residual network model are fused by setting a threshold value, and fault prediction is performed on the state of the currently operated charging pile according to the fused model, so that early warning notification can be performed on the charging pile with potential safety hazards in advance and the fault type can be fed back. The overall operation and maintenance efficiency of the charging station is improved, effective service is provided for safe and stable operation of the electric automobile charging pile, and sustainable and healthy development of the new energy automobile industry is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of charging facility fault diagnosis, and in particular to a charging pile fault intelligent diagnosis method and system. Background Art

[0002] With the rapid development of the electric vehicle industry around the world, charging facilities, as an important infrastructure, have also ushered in unprecedented development opportunities and challenges. Electric vehicles are favored for their environmentally friendly characteristics such as zero emissions and low noise, and are gradually becoming an important part of future travel. However, as one of the key links supporting the electric vehicle industry chain, the reliability and safety of charging facilities are directly related to the daily travel experience of the majority of car owners and the development prospects of the entire new energy vehicle industry.

[0003] Although the construction of charging facilities has been rapidly promoted in recent years, there are still many challenges in the actual operation process. Among them, the frequent occurrence of charging equipment failures not only affects the user experience, but also increases operating costs and reduces the service level of the charging network. When faced with complex and changeable fault conditions, the existing charging station operation and maintenance system often shows problems such as insufficient fault identification ability and slow fault response, which to a certain extent restricts the improvement of the quality of charging services. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a charging pile fault intelligent diagnosis method and system to solve the problems of inaccurate identification of charging equipment fault categories and untimely fault feedback in current charging stations as a whole.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an intelligent diagnosis method for charging pile faults, comprising: obtaining historical operation work order data of faulty charging piles using charging historical operation work orders; extracting characteristic factors of the historical operation work order data of the faulty charging piles, and preprocessing the characteristic factors; constructing a training feature data set based on the preprocessed characteristic factors, and training a bidirectional long short-term memory recurrent neural network model and a residual network model based on the training feature data set; fusing the trained bidirectional long short-term memory recurrent neural network model and the residual network model by setting a threshold, and predicting the fault status of the currently operating charging pile according to the fused model.

[0008] As a preferred solution of the intelligent diagnosis method for charging pile faults of the present invention, the method of obtaining the historical operation work order data of the faulty charging pile by using the historical charging operation work order includes:

[0009] Obtain the user's charging history operation work order data through the background database, including: charging pile number, charging pile name, charging pile type, fault condition, fault start time, fault end time, fault acceptance time, fault clearing time, fault clearing method, fault code, fault description and work order ID; according to the fault clearing time in the charging history operation work order data, obtain the data of the faulty charging pile history operation work order before and after the fault clearing time, including the charging pile number, charging date, charging start time, charging end time, charging model, whether charging is completed and related information of the charging end reason;

[0010] Filter the acquired data of the historical operation work orders of the faulty charging pile according to business needs, leaving the required data including: charging pile number, charging start time, charging end time, whether charging is completed and order end reason, and remove invalid orders and unnecessary data;

[0011] According to the charging time and the reason for order termination, the acquired data of the historical operation work orders of the faulty charging piles are classified into three categories: valid charging equipment orders, faulty charging equipment orders and failed charging equipment orders.

[0012] As a preferred solution of the intelligent diagnosis method for charging pile faults of the present invention, the characteristic factors for extracting the historical operation work order data of the faulty charging pile include:

[0013] Establishing a target value vector and a characteristic value vector according to the historical operation work order data of the faulty charging pile, wherein the target value vector is a single target value;

[0014] The number of eigenvalue vectors is the number of work orders. The total number of orders for each charging pile per day is counted, and the total number of orders within k days from the prediction time point is used as the first characteristic factor. The total number of faulty orders and the total number of invalid orders for each charging pile on the day of the prediction time point are counted as the second characteristic factor. The number of different order termination reasons for the same charging pile on the day of the prediction time point is counted as the third characteristic factor.

[0015] As a preferred solution of the intelligent diagnosis method for charging pile faults of the present invention, the preprocessing of the characteristic factors includes:

[0016] The characteristic factor data of different dimensions are linearly normalized, and the formula is expressed as follows:

[0017]

[0018] Among them, i represents the index of the data sample, j represents the index of the feature factor, and x ij represents the original value of the i-th sample on the j-th feature, x′ij Represents data x ij The result of normalization, parameter k is used to define the range of the feature value subset, w represents the total number of features, x j.min represents the smallest value among all eigenvectors of the jth eigenfactor, x j.max Represents the maximum value of the jth eigenfactor of all eigenvectors;

[0019] The Newton interpolation method is used to interpolate the normalized data. The formula is:

[0020] p n (x)=f[x0]+f[x0,x1](x-x0)+f[x0,x1,x2](x-x0)(x-x1)+…+f[x0,x1,…,x n ](x-x0)(x-x1)…(xx n-1 )

[0021] Among them, p n (x) represents the position of the independent variable, x represents the value of the independent variable, x0,x 1 ,…,x n-1 represents the horizontal coordinate of the interpolation node, n represents the total number of interpolation nodes, f[x0,x1,…,x n ] represents x0,x1,…,x n-1 The n-1th order difference quotient of the interpolation node is used to represent the rate of change of the function value between different interpolation points;

[0022] The n-1th order difference quotient of the interpolation node is calculated by the recursive formula:

[0023] f[x i ]=f(x i )

[0024] Among them, f[x i ] is the zero-order difference quotient, indicating the interpolation node x i The function value f(x i );

[0025]

[0026] Among them, f[x i ,x i+1 ,…,x i+k ] is the k-order difference quotient, indicating the function value at the interpolation node x i ,x i+1 ,…,x i+k The rate of change between, the value of i ranges from 0 to nk-1.

[0027] As a preferred solution of the charging pile fault intelligent diagnosis method of the present invention, wherein: training a bidirectional long short-term memory recurrent neural network model and a residual network model based on the training feature data set includes:

[0028] After label encoding of the preprocessed characteristic factors, training samples and label data are generated according to the data of the first time range as input and the data of the second time range as labels;

[0029] Dividing the processed data into a training set and a test set according to a certain ratio, wherein the training set is used to train a bidirectional long short-term memory recurrent neural network model;

[0030] The input gate in the unit of the bidirectional long short-term memory recurrent neural network model controls the retention of the memory of the previous moment, the forget gate helps the network selectively discard the memory that is no longer important, the candidate memory cell accepts new input and updates the memory according to the control condition of the input gate in the unit, and the output gate selectively outputs the memory at the current time step. The output h of the bidirectional long short-term memory recurrent neural network model t It is expressed as:

[0031] h t =[h t f ;h t b ]

[0032] Among them, h t f and h t b represent the forward hidden state and the reverse hidden state respectively.

[0033] As a preferred solution of the charging pile fault intelligent diagnosis method of the present invention, wherein: training the bidirectional long short-term memory recurrent neural network model and the residual network model based on the training feature data set also includes:

[0034] The Adam algorithm is selected to optimize the bidirectional long short-term memory recurrent neural network model and the residual network model. The first-order momentum term formula of the Adam algorithm optimizer parameter is:

[0035]

[0036] Where m represents the first-order momentum term, β1 represents the attenuation coefficient, represents the gradient at the current time step, L represents the objective function, and W represents the parameter;

[0037] The second-order momentum term formula for the Adam algorithm optimizer parameters is:

[0038]

[0039] Where g represents the second-order momentum term, and β2 represents the attenuation coefficient;

[0040] The deviation formula of the modified first-order momentum term and the deviation formula of the second-order momentum term are:

[0041]

[0042] in, represents the modified first-order momentum term, represents the modified second-order momentum term;

[0043] Adam's learning parameter update error formula is:

[0044]

[0045] Among them, Δθ is the learning parameter, η represents the learning rate, and ε represents a very small constant;

[0046] The test set is used to test the bidirectional long short-term memory recurrent neural network model and the residual network model to obtain a fault classification diagnosis and output a fault probability value.

[0047] As a preferred solution of the intelligent diagnosis method for charging pile faults of the present invention, the fault prediction of the current operating charging pile status includes:

[0048] By setting the threshold, the trained bidirectional long short-term memory recurrent neural network model and the residual network model are fused, and the feature data is sent to the residual network weight layer, the nonlinear activation function layer and the batch normalization weight layer in turn. The processed results are then sent to multiple residual units, and then the fused output results are obtained through the batch activation function layer. The formula is:

[0049]

[0050] r c =LSTM(f(x c ,W c )+x c )

[0051] Among them, y c represents the fusion output, represents the feature vector of the original input, Represents the feature vector output by the residual network, d represents the dimension of the output vector, Represents the concatenation operation of vectors, r c Represents the features of each dimension output by the Bi-LSTM residual network, x crepresents each dimension of the original input, f(·,·) represents the residual correlation function of the residual network, and W c Represents the parameter matrix of the residual layer network;

[0052] For each test sample, record the four classification evaluation indicators of the model output, namely, accuracy, precision, recall, and F1 value, and determine the balance point between accuracy and recall based on actual needs and scenarios;

[0053] A threshold is set according to the balance point. When the fault probability value output by the model is greater than the threshold, the charging pile currently in operation is determined to be in a fault state. When the fault probability value output by the model is less than the threshold, the charging pile currently in operation is determined to be in a normal state.

[0054] In a second aspect, the present invention provides a charging pile fault intelligent diagnosis system, comprising:

[0055] A data acquisition unit, used to acquire historical operation work order data of a faulty charging pile using a charging historical operation work order;

[0056] A data processing unit, used to extract characteristic factors of the historical operation work order data of the faulty charging pile, and pre-process the characteristic factors;

[0057] A model training unit, used for constructing a training feature data set according to the preprocessed feature factors, and training a bidirectional long short-term memory recurrent neural network model and a residual network model based on the training feature data set;

[0058] The fault prediction unit is used to fuse the trained bidirectional long short-term memory recurrent neural network model and the residual network model by setting a threshold, and predict the fault status of the currently operating charging pile according to the fused model.

[0059] In a third aspect, the present invention provides a computing device, comprising:

[0060] Memory and processor;

[0061] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method are implemented.

[0062] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which implement the steps of the method when executed by a processor.

[0063] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention utilizes the charging history operation work order to obtain the historical operation work order data of the faulty charging pile; extracts the characteristic factors of the historical operation work order data of the faulty charging pile, and preprocesses the characteristic factors; constructs a training feature data set according to the preprocessed characteristic factors, and trains a bidirectional long short-term memory recurrent neural network model and a residual network model based on the training feature data set; fuses the trained bidirectional long short-term memory recurrent neural network model and the residual network model by setting a threshold, and predicts the fault status of the currently operating charging pile according to the fused model, so as to give early warning notifications to charging piles with safety hazards and feedback the fault type, thereby improving the overall operation and maintenance efficiency of charging stations, providing effective services for the safe and stable operation of electric vehicle charging piles, and promoting the sustainable and healthy development of the new energy vehicle industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0065] Figure 1 This is an overall flow chart of a charging pile fault intelligent diagnosis method according to an embodiment of the present invention;

[0066] Figure 2 This is a flow chart of a feature extraction and data processing method of a charging pile fault intelligent diagnosis method according to an embodiment of the present invention;

[0067] Figure 3 A flow chart of a bidirectional recurrent neural network for a charging pile fault intelligent diagnosis method according to an embodiment of the present invention is established;

[0068] Figure 4 A schematic diagram of a charging pile operation and maintenance work order of a charging pile fault intelligent diagnosis method according to an embodiment of the present invention;

[0069] Figure 5 A schematic diagram of historical operation work order data of a faulty charging pile in a method for intelligent diagnosis of charging pile faults according to an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0071] Example 1

[0072] Reference Figure 1 to Figure 5 As an embodiment of the present invention, a charging pile fault intelligent diagnosis method is provided, which specifically includes the following steps:

[0073] S100: Using the charging history operation work order to obtain the historical operation work order data of the faulty charging pile;

[0074] In an embodiment of the present application, the user's charging history operation work order data is obtained through the background database, including: charging pile number, charging pile name, charging pile type, fault condition, fault start time, fault end time, fault acceptance time, fault clearing time, fault clearing method, fault code, fault description and work order ID. According to the fault clearing time in the charging history operation work order data, the data of the faulty charging pile history operation work order before and after the fault clearing time is obtained, including the charging pile number, charging date, charging start time, charging end time, charging model, whether charging is completed and related information of the charging end reason. The specific data sample is Figure 4 and Figure 5 As shown;

[0075] In the embodiment of the present application, the acquired data of the historical operation work order of the faulty charging pile is filtered and processed according to the business needs, and the required data is retained, including: the charging pile number, the charging start time, the charging end time, whether the charging is completed and the order end reason, and invalid orders and unnecessary data are eliminated;

[0076] In the embodiment of the present application, the acquired data of the historical operation work orders of the faulty charging piles are classified into three categories: valid charging equipment orders, charging equipment fault orders, and charging equipment failure orders according to the charging time and the order termination reason;

[0077] It should be noted that: (1) valid charging equipment orders: orders that start and end normally, the user uses the charging pile to complete the charging order, the charging duration is within the set reasonable range, and the power consumption is in line with expectations; (2) charging equipment failure orders: orders that start and end abnormally, the order may be terminated due to external damage or internal failure of the charging pile; (3) charging equipment failure orders: orders with too short charging time and no practical reference significance, which may be due to the operation and maintenance personnel frequently testing the charging pile or the user quickly canceling the charging order. When the order duration is less than θ t Or when the power consumption is less than W, it is defined as an invalid order, where θ t and W is the set threshold.

[0078] S200: extracting characteristic factors of historical operation work order data of faulty charging piles, and preprocessing the characteristic factors;

[0079] In the embodiments of the present application, Figure 2 As shown in Figure 1, the characteristic factors for extracting historical operation work order data of faulty charging piles include:

[0080] Establish the target value vector Y based on the historical operation work order data of the faulty charging pile 1×m and the eigenvalue vector X m×n , the formula is:

[0081]

[0082] Among them, Y 1×m Indicates that the target value vector is a single target value, y i ∈{1,0}, yi=1 means the charging pile fails, y i =0 means that the charging pile has no fault, m means the number of orders, that is, the number of feature vectors is m, X m×n Each row vector of the set representing the eigenvectors is a eigenvector, each eigenvector has n eigenvector factors, there are a total of m eigenvectors, and they correspond one-to-one to the target value;

[0083] It should be noted that the target value vector is a single target value, i.e., whether the charging pile is faulty. A faulty charging pile is represented by 1, and no fault is represented by 0. The number of eigenvalue vectors is the number of work orders.

[0084] Furthermore, the total number of orders for each charging pile per day is counted, and the total number of orders within k days from the prediction time point is taken as the first characteristic factor, recorded as Q i (i=1,2,...,k), where Q i Indicates the total number of orders on the i-th day from the prediction time point, with x i The i-th element of one of the eigenvectors in the table, that is:

[0085] x i =Q i (i=1,2,...,k)

[0086] Count the total number of fault orders Q for each charging pile at the predicted time point on the day gi And the number of invalid orders Q wi As the second characteristic factor, Q gi represents the total number of faulty orders on the i-th day at the prediction time point, Q wi Indicates the total number of invalid orders on the i-th day at the forecast time point. That is:

[0087] x i+k =Q gi (i=1,2,...,k)

[0088] x i+2k =Q wi (i=1,2,...,k)

[0089] Count the number of different order termination reasons for the same charging pile at the predicted time point on the same day Q yi (i=1,2,...,w) as the third characteristic factor, Q yi Indicates the number of reasons for the end of the i-th order. There are at most i reasons for the end of an order, where 3k+w=q, that is:

[0090] x 3k+i =Q yi (i=1,2,...,w)

[0091] In the embodiment of the present application, preprocessing the characteristic factors includes:

[0092] The characteristic factor data of different dimensions are processed uniformly so that the data of different units can be used as the input parameters of the neural network at the same time, providing a basis for subsequent analysis. The linear normalization is expressed as:

[0093]

[0094] Among them, X is the original data, X max is the maximum value in the data, X min is the minimum value in the data;

[0095] Then the linear normalization formula of the eigenvector is expressed as:

[0096]

[0097] Among them, i represents the index of the data sample, j represents the index of the feature factor, and x ij represents the original value of the i-th sample on the j-th feature, x′ ijRepresents data x ij The result of normalization, parameter k is used to define the range of the eigenvalue subset, and the specific feature range is processed separately to ensure that the sum of the ratios of these eigenvalues ​​is 1. w represents the total number of features, x j.min represents the smallest value among all eigenvectors of the jth eigenfactor, x j.max Represents the maximum value of the jth eigenfactor of all eigenvectors;

[0098] Furthermore, the Newton interpolation method is used to interpolate the normalized data to solve the problem of incomplete data and partial data missing due to equipment, storage and other reasons. The advantage is that new data points can be dynamically added to the interpolation node without recalculating all coefficients. The formula is expressed as:

[0099] p n (x)=f[x0]+f[x0,x1](x-x0)+f[x0,x1,x2](x-x0)(x-x1)+

[0100] …+f[x0,x1,…,x n ](x-x0)(x-x1)…(xx n-1 )

[0101] Among them, p n (x) represents the position of the independent variable, x represents the value of the independent variable, x0, x1, …, x n-1 represents the horizontal coordinate of the interpolation node, n represents the total number of interpolation nodes, f[x0,x1,…,x n ] represents x0,x1,…,x n-1 The n-1th order difference quotient of the interpolation node is used to represent the rate of change of the function value between different interpolation points;

[0102] Furthermore, the n-1th order difference quotient of the interpolation node is calculated by the recursive formula:

[0103] f[x i ]=f(x i )

[0104] Among them, f[x i ] is the zero-order difference quotient, indicating the interpolation node x i The function value f(x i );

[0105]

[0106] Among them, f[x i ,x i+1 ,…,x i+k ] is the k-order difference quotient, indicating the function value at the interpolation node xi ,x i+1 ,…,x i+k The rate of change between, the value of i ranges from 0 to nk-1.

[0107] S300: constructing a training feature data set according to the preprocessed feature factors, and training a bidirectional long short-term memory recurrent neural network model and a residual network model based on the training feature data set;

[0108] In the embodiment of the present application, the pre-processed characteristic factors are label-encoded, the fault type label is mapped to 0 and 1, and the fault type is encoded. For example, "none" fault is coded as "000", "other" fault is coded as "001", "screen fault" fault is coded as "002", "module fault" fault is coded as "003", "offline fault" fault is coded as "004", and the data of the first time range is used as input and the data of the second time range is used as labels to generate training samples and label data;

[0109] It should be noted that the first time range is the past 7 days and the second time range is the 8th day;

[0110] Furthermore, the processed data is divided into a training set and a test set according to a certain ratio. The training set is used to train the bidirectional long short-term memory recurrent neural network model. In this experiment, the data set is divided according to a ratio of 7:3, with 70% of the data used as the training set and 30% of the data used as the test set.

[0111] Further, such as Figure 3 As shown in the figure, the input gate control in the unit of the bidirectional long short-term memory recurrent neural network model retains the memory of the previous moment, and the positive input gate i t f and the reverse input gate i t b The formulas are:

[0112] i t f =σ(W i x t +b i +U i h t-1 )

[0113] i t b =σ(W i x t +b i +U i h t+1 )

[0114] Among them, x tis the input of the time step at time t, σ is the activation function, h t-1 is the hidden state at the previous moment t-1, W i and U i is the weight from the input layer to the hidden layer, b i is the bias input to the hidden layer; the calculation of the reverse long short-term memory recurrent neural network is performed from back to front, h t+1 t+1 represents the next time step;

[0115] The forget gate helps the network selectively discard memories that are no longer important. t f and the reverse forget gate f t b The formulas are:

[0116] f t f =σ(W f x t +b f +U f h t-1 )

[0117] f t b =σ(W f x t +b f +U f h t+1 )

[0118] Among them, W f and U f is the corresponding weight, b f is the corresponding bias;

[0119] The candidate memory cell receives new input and updates the memory according to the control conditions of the input gate in the cell. t f and reverse candidate memory cells g t b The formulas are:

[0120] g t f =tanh(W g x t +b g +U g h t-1 )

[0121] g t b =tanh(W g x t +b g +Ug h t+1 )

[0122] Among them, W g and U g is the corresponding weight, b g is the corresponding bias;

[0123] Updated forward and reverse memory cell formula c t f and c t b They are:

[0124] c t f =f t ⊙c t-1 +i t ⊙g t

[0125] c t b =f t ⊙c t-1 +i t g t

[0126] Among them, ⊙ represents element-by-element multiplication, c t is the updated memory cell, c t-1 is the memory cell of the previous time step, i t is the input value, g t Represents the current candidate memory cell;

[0127] The output gate selectively outputs the memory at the current time step, and the forward output gate o t f and the reverse output gate o t b The formulas are:

[0128] o t f =σ(W o x t +b o +U o h t-1 )

[0129] o t b =σ(W o x t +b o +U o h t+1 )

[0130] Among them, W o and U ois the corresponding weight, b o is the corresponding bias;

[0131] The forward hidden state h at time t t f and the reverse hidden state h t b for:

[0132] h t f =o t ⊙tanh(c t )

[0133] h t b =o t ⊙tanh(c t )

[0134] Among them, the initial state at the beginning is zero, o t represents the current output gate, c t represents updated memory cells;

[0135] Furthermore, the output h of the bidirectional long short-term memory recurrent neural network model t It is expressed as:

[0136] h t =[h t f ;h t b ]

[0137] Among them, h t f and h t b represent the forward hidden state and the reverse hidden state respectively.

[0138] In the embodiment of the present application, the Adam algorithm is selected to optimize the bidirectional long short-term memory recurrent neural network model and the residual network model. The first-order momentum term of the Adam algorithm optimizer parameter is used to measure the gradient of the parameter, and the learning rate of the model is adjusted to stabilize the training process. The variance of the parameter update is reduced by using the statistical information of the past gradient. The formula is:

[0139]

[0140] Where m represents the first-order momentum term, β1 represents the attenuation coefficient, represents the gradient at the current time step, L represents the objective function, and W represents the parameter;

[0141] The second-order momentum term of the Adam algorithm optimizer parameter is used to measure the gradient variance of the parameter. The first-order momentum term formula is constrained to avoid too fast parameter updates and improve stability. The formula is:

[0142]

[0143] Where g represents the second-order momentum term, and β2 represents the attenuation coefficient;

[0144] Since the parameter estimate may be biased towards 0 at the beginning, it is necessary to compensate for the first-order moment estimate to reduce the deviation of the optimization process. The deviation formula of the modified first-order momentum term and the deviation formula of the second-order momentum term are:

[0145]

[0146] in, represents the modified first-order momentum term, represents the modified second-order momentum term;

[0147] Combining the above characteristics of adaptive learning rate, ability to handle sparse gradients, and use of historical gradient information, the first-order and second-order momentum terms after deviation correction are used to adjust the step size, and the learning rate is combined to control the speed of parameter update. The learning parameter update error formula of Adam is:

[0148]

[0149] Among them, Δθ is the learning parameter, η represents the learning rate, and ε represents a very small constant;

[0150] Furthermore, the test set is used to test the bidirectional long short-term memory recurrent neural network model and the residual network model to obtain the fault classification diagnosis and output the fault probability value.

[0151] S400: The trained bidirectional long short-term memory recurrent neural network model and residual network model are integrated by setting a threshold, and the fault status of the currently operating charging pile is predicted according to the integrated model;

[0152] In the embodiment of the present application, the trained bidirectional long short-term memory recurrent neural network model and the residual network model are fused by setting a threshold, and the feature data is sequentially sent to the residual network weight layer, the nonlinear activation function layer and the batch normalization weight layer, and then the processed result is sent to multiple residual units, and then the fusion output result is obtained through the batch activation function layer. The formula is:

[0153]

[0154] r c =LSTM(f(x c ,W c)+x c )

[0155] Among them, y c represents the fusion output, represents the feature vector of the original input, Represents the feature vector output by the residual network, d represents the dimension of the output vector, Represents the concatenation operation of vectors, r c Represents the features of each dimension output by the Bi-LSTM residual network, x c represents each dimension of the original input, f(·,·) represents the residual correlation function of the residual network, and W c Represents the parameter matrix of the residual layer network;

[0156] For each test sample, the four classification evaluation indicators of the model output, namely, accuracy, precision, recall and F1 value, are recorded. The formula is expressed as follows:

[0157]

[0158] Among them, TP represents the number of faulty data sets identified as charging fault conditions, TN represents the number of data sets correctly identified as charging faults, FP represents the number of samples incorrectly predicted as normal charging data, and FN represents the number of samples incorrectly predicted as faulty data.

[0159] Determine the balance point between accuracy and recall based on actual needs and scenarios. A higher threshold can improve accuracy but may lead to a lower recall, while a lower threshold may have the opposite effect. Usually, a threshold can be selected that makes accuracy and recall close to a balance.

[0160] The threshold is set to 0.5 according to the balance point. When the fault probability value output by the model is greater than the threshold, the charging pile currently in operation is judged to be in a fault state. When the fault probability value output by the model is less than the threshold, the charging pile currently in operation is judged to be in a normal state.

[0161] From the above, it can be seen that the present invention uses the charging history operation work order to obtain the historical operation work order data of the faulty charging pile; extracts the characteristic factors of the historical operation work order data of the faulty charging pile, and preprocesses the characteristic factors; constructs a training feature data set according to the preprocessed characteristic factors, and trains a bidirectional long short-term memory recurrent neural network model and a residual network model based on the training feature data set; fuses the trained bidirectional long short-term memory recurrent neural network model and the residual network model by setting a threshold, and predicts the fault status of the currently operating charging pile according to the fused model, so as to give early warning notifications to charging piles with safety hazards and feedback the fault type, thereby improving the overall operation and maintenance efficiency of the charging station, providing effective services for the safe and stable operation of electric vehicle charging piles, and promoting the sustainable and healthy development of the new energy vehicle industry.

[0162] The above is a schematic scheme of a charging pile fault intelligent diagnosis method of this embodiment. It should be noted that the technical scheme of the charging pile fault intelligent diagnosis system and the technical scheme of the above-mentioned charging pile fault intelligent diagnosis method belong to the same concept. For details not described in detail in the technical scheme of the charging pile fault intelligent diagnosis system in this embodiment, please refer to the description of the technical scheme of the above-mentioned charging pile fault intelligent diagnosis method.

[0163] This embodiment also provides a charging pile fault intelligent diagnosis system, including:

[0164] A data acquisition unit, used to acquire historical operation work order data of a faulty charging pile using a charging historical operation work order;

[0165] A data processing unit, used to extract characteristic factors of historical operation work order data of faulty charging piles and pre-process the characteristic factors;

[0166] A model training unit, used for constructing a training feature data set according to the preprocessed feature factors, and training a bidirectional long short-term memory recurrent neural network model and a residual network model based on the training feature data set;

[0167] The fault prediction unit is used to fuse the trained bidirectional long short-term memory recurrent neural network model and the residual network model by setting a threshold, and predict the fault status of the currently operating charging pile according to the fused model.

[0168] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.

[0169] This embodiment also provides a computing device, which includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for intelligent diagnosis of charging pile faults is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0170] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method proposed in the above embodiment is implemented.

[0171] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0172] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0173] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0174] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0175] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0176] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0177] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0178] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0179] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A charging pile fault intelligent diagnosis method, characterized in that: include: Use the charging history operation work order to obtain the historical operation work order data of the faulty charging pile; Extracting characteristic factors of the historical operation work order data of the faulty charging pile, and preprocessing the characteristic factors; Constructing a training feature data set according to the preprocessed feature factors, and training a bidirectional long short-term memory recurrent neural network model and a residual network model based on the training feature data set; The trained bidirectional long short-term memory recurrent neural network model and the residual network model are fused by setting a threshold, and the fault status of the currently operating charging pile is predicted based on the fused model.

2. The intelligent diagnosis method for charging pile faults according to claim 1, characterized in that: The method of obtaining the historical operation work order data of the faulty charging pile by using the historical charging operation work order includes: Obtain the user's charging history operation work order data through the background database, including: charging pile number, charging pile name, charging pile type, fault condition, fault start time, fault end time, fault acceptance time, fault clearing time, fault clearing method, fault code, fault description and work order ID; according to the fault clearing time in the charging history operation work order data, obtain the data of the faulty charging pile history operation work order before and after the fault clearing time, including the charging pile number, charging date, charging start time, charging end time, charging model, whether charging is completed and related information of the charging end reason; Filter the acquired data of the historical operation work orders of the faulty charging pile according to business needs, leaving the required data including: charging pile number, charging start time, charging end time, whether charging is completed and order end reason, and remove invalid orders and unnecessary data; According to the charging time and the reason for order termination, the acquired data of the historical operation work orders of the faulty charging piles are classified into three categories: valid charging equipment orders, faulty charging equipment orders and failed charging equipment orders.

3. The intelligent diagnosis method for charging pile faults according to claim 2, characterized in that: The characteristic factors for extracting the historical operation work order data of the faulty charging pile include: Establishing a target value vector and a characteristic value vector according to the historical operation work order data of the faulty charging pile, wherein the target value vector is a single target value; The number of eigenvalue vectors is the number of work orders. The total number of orders for each charging pile per day is counted, and the total number of orders within k days from the prediction time point is used as the first characteristic factor. The total number of faulty orders and the total number of invalid orders for each charging pile on the day of the prediction time point are counted as the second characteristic factor. The number of different order termination reasons for the same charging pile on the day of the prediction time point is counted as the third characteristic factor.

4. The intelligent diagnosis method for charging pile faults according to claim 3, characterized in that: Preprocessing the characteristic factors includes: The characteristic factor data of different dimensions are linearly normalized, and the formula is expressed as follows: Among them, i represents the index of the data sample, j represents the index of the feature factor, and x ij represents the original value of the i-th sample on the j-th feature, x′ ij Represents data x ij The result of normalization, parameter k is used to define the range of the feature value subset, w represents the total number of features, x j.min represents the smallest value among all eigenvectors of the jth eigenfactor, x j.max Represents the maximum value of the jth eigenfactor of all eigenvectors; The Newton interpolation method is used to interpolate the normalized data. The formula is: p n (x)=f[x0]+f[x0,x1](x-x0)+f[x0,x1,x2](x-x0)(x-x1)+…+f[x0,x1,…,x n ](x-x0)(x-x1)…(xx n-1 ) Among them, p n (x) represents the position of the independent variable, x represents the value of the independent variable, x0, x1, …, x n-1 represents the horizontal coordinate of the interpolation node, n represents the total number of interpolation nodes, f[x0,x1,…,x n ] represents x0,x1,…,x n-1 The n-1th order difference quotient of the interpolation node is used to represent the rate of change of the function value between different interpolation points; The n-1th order difference quotient of the interpolation node is calculated by the recursive formula: f[x i ]=f(x i ) Among them, f[x i ] is the zero-order difference quotient, indicating the interpolation node x i The function value f(x i ); Among them, f[x i ,x i+1 ,…,x i+k ] is the k-order difference quotient, indicating the function value at the interpolation node x i ,x i+1 ,…,x i+k The rate of change between, the value of i ranges from 0 to nk-1.

5. The intelligent diagnosis method for charging pile faults according to claim 4, characterized in that: Training a bidirectional long short-term memory recurrent neural network model and a residual network model based on the training feature data set includes: After label encoding of the preprocessed characteristic factors, training samples and label data are generated according to the data of the first time range as input and the data of the second time range as labels; Dividing the processed data into a training set and a test set according to a certain ratio, wherein the training set is used to train a bidirectional long short-term memory recurrent neural network model; The input gate in the unit of the bidirectional long short-term memory recurrent neural network model controls the retention of the memory of the previous moment, the forget gate helps the network selectively discard the memory that is no longer important, the candidate memory cell accepts new input and updates the memory according to the control condition of the input gate in the unit, and the output gate selectively outputs the memory at the current time step. The output h of the bidirectional long short-term memory recurrent neural network model t It is expressed as: h t =[h t f ;h t b ] Among them, h t f and h t b represent the forward hidden state and the reverse hidden state respectively.

6. The intelligent diagnosis method for charging pile faults according to claim 5, characterized in that: Training the bidirectional long short-term memory recurrent neural network model and the residual network model based on the training feature data set also includes: The adaptive moment estimation Adam algorithm is selected to optimize the bidirectional long short-term memory recurrent neural network model and the residual network model. The first-order momentum term formula of the Adam algorithm optimizer parameter is: Where m represents the first-order momentum term, β1 represents the attenuation coefficient, represents the gradient at the current time step, L represents the objective function, and W represents the parameter; The second-order momentum term formula for the Adam algorithm optimizer parameters is: Where g represents the second-order momentum term, and β2 represents the attenuation coefficient; The deviation formula of the modified first-order momentum term and the deviation formula of the second-order momentum term are: in, represents the modified first-order momentum term, represents the modified second-order momentum term; Adam's learning parameter update error formula is: Among them, Δθ is the learning parameter, η represents the learning rate, and ε represents a very small constant; The test set is used to test the bidirectional long short-term memory recurrent neural network model and the residual network model to obtain a fault classification diagnosis and output a fault probability value.

7. The intelligent diagnosis method for charging pile faults according to claim 6, characterized in that: The fault prediction of the current operating charging pile status includes: By setting the threshold, the trained bidirectional long short-term memory recurrent neural network model and the residual network model are fused, and the feature data is sent to the residual network weight layer, the nonlinear activation function layer and the batch normalization weight layer in turn. The processed results are then sent to multiple residual units, and then the fused output results are obtained through the batch activation function layer. The formula is: r c =LSTM(f(x c ,W c )+x c ) Among them, y c represents the fusion output, represents the feature vector of the original input, Represents the feature vector output by the residual network, d represents the dimension of the output vector, Represents the concatenation operation of vectors, r c Represents the features of each dimension output by the Bi-LSTM residual network, x c represents each dimension of the original input, f(·,·) represents the residual correlation function of the residual network, and W c Represents the parameter matrix of the residual layer network; For each test sample, record the four classification evaluation indicators of the model output, namely, accuracy, precision, recall, and F1 value, and determine the balance point between accuracy and recall based on actual needs and scenarios; A threshold is set according to the balance point. When the fault probability value output by the model is greater than the threshold, the charging pile currently in operation is determined to be in a fault state. When the fault probability value output by the model is less than the threshold, the charging pile currently in operation is determined to be in a normal state.

8. A system using the charging pile fault intelligent diagnosis method according to any one of claims 1 to 7, characterized in that: include: A data acquisition unit, used to acquire historical operation work order data of a faulty charging pile using a charging historical operation work order; A data processing unit, used to extract characteristic factors of the historical operation work order data of the faulty charging pile, and pre-process the characteristic factors; A model training unit, used for constructing a training feature data set according to the preprocessed feature factors, and training a bidirectional long short-term memory recurrent neural network model and a residual network model based on the training feature data set; The fault prediction unit is used to fuse the trained bidirectional long short-term memory recurrent neural network model and the residual network model by setting a threshold, and predict the fault status of the currently operating charging pile according to the fused model.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having computer program instructions stored thereon, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images