Abnormal data diagnosis method and system for charging piles based on AP clustering and LSTM network

Abstract

The present invention discloses a charging pile abnormal data diagnosis method and system based on AP clustering and LSTM network, which uses the charging pile metering unit to measure the pile end data in real time, and at the same time receives the vehicle end BMS information transmitted to the charging pile. Use standard metering instruments to collect real-time electric energy data of the complete charging cycle of the car and the pile on site, and establish a difference sample library. Perform AP cluster analysis on the difference sample library, introduce the Pearson correlation analysis model, perform correlation calculation, determine the influence degree of the main cause of the data difference, and use different error threshold judgment criteria. Use LSTM neural network to establish a time series abnormal data diagnosis model, calculate the error index of vehicle and pile charging metering data in real time, input the data exceeding the error threshold range into the diagnosis model, and realize abnormal data identification and abnormal cause diagnosis. The present invention not only makes the diagnosis result more authentic and reliable, but also objective and comprehensive.

Description

Technical Field

[0001] The present invention relates to a method and system for diagnosing abnormal data of vehicle charging piles based on AP clustering and LSTM neural network, and belongs to the technical field of charging pile fault models. Background Art

[0002] With the popularization of electric vehicles and charging facilities, faults in the charging process are becoming more frequent, affecting the further promotion of electric vehicles, and more importantly, endangering the personal safety of users. A survey conducted by the China Electric Vehicle Charging Infrastructure Promotion Alliance on the use of charging facilities in ten cities in China showed that the average failure rate of public charging facilities reached 20%, and frequent failures seriously affected their charging life. Electric vehicle safety accidents are increasing year by year, especially the safety problems generated during the charging process seriously restrict the vigorous development of the new energy vehicle industry. The fault diagnosis capability of charging facilities has become one of the main issues affecting their safe and stable operation, especially when the key components of the charging module and energy conversion module fail, it will affect the safe working performance of the entire equipment. At present, the fault diagnosis system of charging facilities is not perfect enough, the fault diagnosis effect is not good enough, and the fault diagnosis strategy is not smart enough. It is difficult to determine the source of faults with complex correlations, which has also become a problem for the further promotion of electric vehicles. Summary of the invention

[0003] Purpose of the invention: In order to realize abnormal data identification and abnormal cause diagnosis during vehicle-pile interaction, the present invention provides a charging pile abnormal data diagnosis method based on AP clustering and LSTM network.

[0004] Technical solution: To achieve the above purpose, the technical solution adopted by the present invention is:

[0005] A charging pile abnormal data diagnosis method based on AP clustering and LSTM network includes the following steps:

[0006] Step 1: Use the charging pile metering unit to measure the pile-side data in real time, and receive the vehicle-side BMS information transmitted to the charging pile at the same time. Use standard metering instruments to collect real-time power data of the complete charging cycle of the vehicle and the pile on site. Difference the pile-side data recorded and calculated by the charging pile metering unit and the power data and battery temperature data measured by the standard metering instrument to establish a difference sample library.

[0007] Step 2: Perform AP self-clustering analysis on the data anomaly sample library, and the data is automatically divided into several categories according to its own characteristics.

[0008] Step 3: Introduce the Pearson correlation analysis model to calculate the correlation between the electric energy data difference and battery temperature difference errors and the parameters recorded by the metering unit category by category, determine the influence of the main cause of the data difference according to the size of the correlation coefficient, and use different error threshold judgment standards based on the degree of influence.

[0009] Step 4: Use the results of cluster analysis as input parameters and use the LSTM neural network to establish a time series abnormal data diagnosis model.

[0010] Step 5: Calculate the error index of vehicle-pile charging metering data in real time, and input the data exceeding the error threshold range into the diagnostic model to identify abnormal data and causes.

[0011] Preferred: In step 2, AP self-clustering analysis is performed on the data abnormal sample library, and the data is automatically divided into several categories according to its own characteristics:

[0012] Step 2.1: Establish the objective function of AP clustering based on the established difference sample library.

[0013] The objective function of AP clustering is as follows:

[0014]

[0015] in, represents the clustering objective function value, is called a variable node. is the number of nodes, , , All are function nodes.

[0016] Constraint function 1:

[0017]

[0018]

[0019] in, To solve the first constraint of the objective function, ensure When it is selected as the center by other points, it is also its own center. Solve the maximum value of the first constraint of the objective function.

[0020] Constraint function 2:

[0021]

[0022] in, To solve the second constraint of the objective function, each data point can only choose one point as its center, that is, the center is unique. =1, point For point center.

[0023]

[0024] in, Represents a similarity measure of the input Combine the functions together, where is the opposite of the Euclidean distance, By two constraint functions and Keep making corrections until it stops.

[0025] Step 2.2 Set the intermediate transmission information, update the information according to the maximum sum rule, and finally use the information value to calculate .

[0026]

[0027]

[0028]

[0029] α ij = { ∑ k = j max [ ρ kj , 0 ] , i = j , min [ 0 , ρ ij ] + max k ≠ j [ ρ kj , 0 ] , i = j ,

[0030] in: and Responsible information and available information, respectively. express As The appropriate size of the center, passing from point arrive Available information The role is to transfer from point To point information, indicating that under the condition that information has been obtained from other data points, it is used as The degree of possibility of the center. and All of them are intermediate variables of the AP clustering algorithm. After multiple iterations of the AP clustering algorithm, the relationship matrix between all the data points can be obtained. , so as to achieve the purpose of automatically clustering sample sets into several categories according to data characteristics.

[0031] Step 2.3 To avoid oscillation, the AP algorithm introduces an attenuation coefficient when updating information Each message is set to its previous update value. times the value of this information update times. Among them, the attenuation coefficient is a real number between 0 and 1, and here it is 0.5. and Iteration value:

[0032] ;

[0033] in, represents the attenuation coefficient, Indicates assignment, Indicates responsibility information, Represents prior responsibility information, Indicates the information value obtained by prior responsibility information, Indicates available information. represents the prior available information, Indicates the information value obtained by the prior available information, Indicates the number of iterations.

[0034] Preferably: Step 3 specifically includes the following steps:

[0035] Step 3.1, calculate the correlation between the two variables of electric energy data difference and battery temperature difference error and the parameters recorded by the metering unit respectively to solve the Pearson correlation coefficient.

[0036] Step 3.2, determine the influencing factors that cause the data difference based on the size of the correlation coefficient, and use different error threshold judgment criteria based on the main cause.

[0037] Preferred: Based on the main cause, different error threshold judgment criteria are used. Different error threshold judgments are used based on the relationship between the prior experience charging pile charging failure triggering factors and the error threshold. The relationship between the prior experience charging pile charging failure triggering factors and the error threshold is as follows:

[0038] Temperature Effects: Temperature changes can cause an error threshold in charging efficiency between 5% and 20%, depending on the magnitude of the temperature change and the type of charger.

[0039] Charging pile module failure: Module protection or failure may cause charging interruption, and the error threshold is 100%.

[0040] Communication problem between the charging station and the vehicle: Communication interruption may cause charging failure, with an error threshold of 100%.

[0041] Frequent power outages during charging: May result in insufficient charge. The error threshold depends on the frequency and duration of power outages.

[0042] Charger problem: may cause a significant drop in charging efficiency, and the error threshold depends on the specific condition of the charger.

[0043] Preferably, the LSTM neural network includes an input unit, a forget gate, an input gate, and an output gate, wherein the input unit is a difference data sample after AP clustering. The output unit is a difference data sample that considers the influence of historical data time series and retains some data features through the forget gate, the input gate, and the output gate.

[0044] Forget gate expression:

[0045] f t = σ ( W f [ h t − 1 , x t ] + b f )

[0046] in, represents the forgetting value, , They represent the weight and bias of the forget gate respectively. Represents the current input value. The value of the previous state of the cell. Represents the activation function.

[0047] The input gate is composed of the activation layer and the tanh layer, which determines the information that needs to be updated. The output expression of the activation layer is:

[0048] i t = σ ( W i [ h t − 1 , x i ] + b i )

[0049] in, represents the output of the activation layer, , denote the weight and bias of the activation layer respectively.

[0050] The hyperbolic tangent activation function forms the tanh layer, creating a candidate value that may be added to the cell state. .

[0051] C t ~ = tanh ( W c [ h t − 1 , x t ] + b c )

[0052] in, is the candidate value, is the tanh layer bias.

[0053] Candidate value Combine the old state of the cell Update to new status :

[0054]

[0055] in, Represents the multiplication of corresponding elements.

[0056] The new state of the cell is calculated and combined with the output gate to form the final output containing the forgetting and memory parameters:

[0057] O t = σ ( W o [ h t − 1 , x i ] + b o )

[0058] in, Indicates the final output, , Represent the weight and bias of the output gate respectively.

[0059] Preferably, the method for establishing a time series abnormal data diagnosis model using an LSTM neural network in step 4 is as follows:

[0060] 4.2.1 Data preprocessing: Clean, remove noise and label the difference data after AP clustering, assign a category label of the corresponding abnormal data type to each sample. At the same time, divide the data set into training set, validation set and test set.

[0061] 4.2.2 Model settings: Set learning rate, batch size and number of iterations, initialize weights and biases.

[0062] 4.2.3 Model training: The preprocessed data training set is input into the LSTM neural network model through the neural network input gate to calculate the prediction results. The loss function value is calculated based on the prediction results and the true labels, and the loss function is optimized using the gradient descent method. At the end of each training cycle, the model performance is evaluated using the validation set, and the learning rate and batch size hyperparameters are adjusted based on the validation results to improve the model performance.

[0063] 4.2.4 Model testing: Use the test set to evaluate the model performance and obtain the final performance indicators. Deploy the trained LSTM neural network model to the actual application scenario to provide users with real-time diagnostic suggestions or warnings.

[0064] Optimum: The formula for the error index of the vehicle charging pile metering data is:

[0065]

[0066] Among them, RMSE is the error index of vehicle charging metering data. is the ideal value for measuring electric energy. is the actual value of the metered electric energy, and n is the number of samples.

[0067] Preferably: in step 1, the pile-end data includes voltage, current, usage time, temperature and humidity, and line loss; the vehicle-end BMS information includes battery cell voltage, battery state of charge, and battery temperature.

[0068] Another object of the present invention is to provide a charging pile abnormal data diagnosis system based on AP clustering and LSTM network, characterized in that: it is used to implement the charging pile abnormal data diagnosis method based on AP clustering and LSTM network, including a sample library establishment unit, a clustering unit, an error threshold judgment unit, an LSTM neural network unit, and an identification unit, wherein:

[0069] The sample library establishment unit is used to measure the pile-end data in real time using the charging pile metering unit, and at the same time receive the vehicle-end BMS information transmitted to the charging pile. The real-time power data of the complete charging cycle of the vehicle and the pile is collected on-site using standard metering instruments. The pile-end data recorded and calculated by the charging pile metering unit and the power data and battery temperature data measured by the standard metering instrument are subtracted to establish a difference sample library.

[0070] The clustering unit is used to perform AP self-clustering analysis on the data abnormality sample library, and the data is automatically divided into several categories according to its own characteristics.

[0071] The error threshold judgment unit is used to introduce the Pearson correlation analysis model, calculate the correlation between the electric energy data difference and the battery temperature difference errors and the parameters recorded by the metering unit category by category, determine the influence degree of the main cause of the data difference according to the size of the correlation coefficient, and adopt different error threshold judgment standards based on the influence degree.

[0072] The LSTM neural network unit is used to use the results obtained after cluster analysis as input parameters and use the LSTM neural network to establish a time series abnormal data diagnosis model.

[0073] The identification unit is used to calculate the error index of the vehicle-pile charging metering data in real time, input the data exceeding the error threshold range into the diagnosis model, and realize the identification of abnormal data and causes.

[0074] Another object of the present invention is to provide a computer system, comprising a memory and a processor, wherein the memory is used to store a computer program / instruction. The processor is used to execute the computer program / instruction to implement the charging pile abnormal data diagnosis method based on AP clustering and LSTM network.

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] 1. This method combines vehicle BMS information and charging pile metering information to construct a difference error index. At the same time, it takes into account the possible abnormal influence of vehicle and pile charging data, making the diagnosis result more authentic and reliable.

[0077] 2. This method uses the AP self-clustering algorithm and does not require the number of clusters and cluster centers to be set in advance. It can quickly identify data centers, and on this basis, correlation analysis is performed to determine the main causes of influencing factors, which is more objective and comprehensive.

[0078] 3. Charging pile detection data is affected by the user's living habits, and the amount of data varies greatly in different periods. Using LSTM neural network to build a diagnostic model can fully consider the temporal characteristics of charging data distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 This is a flow chart for abnormal data diagnosis of vehicle charging piles based on AP clustering and LSTM neural network;

[0080] Figure 2 It is a schematic diagram of the principle of similarity propagation clustering algorithm;

[0081] Figure 3 Schematic diagram of the long short-term memory neural network algorithm principle. DETAILED DESCRIPTION

[0082] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0083] A charging pile abnormal data diagnosis method based on AP clustering and LSTM network uses the charging pile metering unit to measure the pile end data (voltage, current, usage time, temperature and humidity, line loss) in real time, and at the same time receives the vehicle-end BMS information (battery cell voltage, battery state of charge, battery temperature) and other parameters transmitted to the charging pile. Use standard metering instruments to collect real-time power data of the complete charging cycle of the car and the pile on site. The pile end data recorded and calculated by the charging pile metering unit and the power data and battery temperature data measured by the standard metering instrument are subtracted to establish a difference sample library. AP self-clustering analysis is performed on the sample data and divided into several categories according to the data characteristics. Correlation analysis is performed on the classified data class by class, and the main influencing factors of the fault and the main cause of the difference in vehicle and pile charging data are determined according to the calculated correlation coefficient, and different error threshold standards are used based on the main cause. The results obtained after AP clustering analysis are used as input to establish a time-series diagnostic network model using an LSTM neural network. The error index (root mean square error) of the vehicle and pile charging metering data is calculated in real time, and the data exceeding the metering error threshold is input into the diagnostic model to achieve the expected diagnostic function. Figure 1-3 As shown, the following steps are included:

[0084] Step 1: For the vehicle-pile interaction data during the charging process, two measurement methods are used to collect data. The specific methods are as follows:

[0085] The charging pile metering unit is used to measure the pile-end data (voltage, current, usage time, temperature and humidity, line loss) in real time, and at the same time receive the vehicle-end BMS information (battery cell voltage, battery state of charge, battery temperature) and other parameters transmitted to the charging pile. Standard metering instruments are used to collect real-time power data of the complete charging cycle of the vehicle and the pile on site. The pile-end data recorded and calculated by the charging pile metering unit is subtracted from the power data and battery temperature data measured by the standard metering instrument to establish a difference sample library.

[0086] Step 2: Perform AP self-clustering analysis on the established difference sample library and randomly divide the data into several categories according to their own characteristics. Figure 2 As shown, the specific steps are as follows:

[0087] The objective function of AP clustering is as follows:

[0088]

[0089] in is called a variable node, , , All are function nodes.

[0090] Constraint function 1:

[0091]

[0092]

[0093] To solve the first constraint of the objective function, ensure When selected as the center by other points, it is also its own center;

[0094] Constraint function 2:

[0095]

[0096] To solve the second constraint of the objective function, each data point can only choose one point as its center, that is, the center is unique. =1, point For point center.

[0097]

[0098] The similarity measure of the input Combined together, is the opposite of the Euclidean distance. Objective function By two constraint functions and Keep making corrections until it stops.

[0099] Step 2.2: Set the intermediate transmission information, update the information according to the maximum sum rule, and finally use the information value to calculate .

[0100]

[0101]

[0102]

[0103] α ij = { ∑ k = j max [ ρ kj , 0 ] , i = j , min [ 0 , ρ ij ] + max k ≠ j [ ρ kj , 0 ] , i = j ,

[0104] in: and They are called responsible information and available information respectively. express As The appropriate size of the center, passing from point arrive Information between; Available information The role is to transfer from point To point information, indicating that under the condition that information has been obtained from other data points, it is used as The degree of possibility of the center. and All of them are intermediate variables of the AP clustering algorithm. After multiple iterations of the AP clustering algorithm, the relationship matrix between all the data points can be obtained. , so as to achieve the purpose of automatically clustering sample sets into several categories according to data characteristics.

[0105] Step 2.3 To avoid oscillation, the AP algorithm introduces an attenuation coefficient when updating information Each message is set to its previous update value. times the value of this information update times. Among them, the attenuation coefficient is a real number between 0 and 1, and here it is 0.5. and Iteration value:

[0106] ;

[0107] in, represents the attenuation coefficient, Indicates assignment, Indicates responsibility information, Represents prior responsibility information, Indicates the information value obtained by prior responsibility information, Indicates available information. represents the prior available information, Indicates the information value obtained by the prior available information, Indicates the number of iterations.

[0108] Step 3: Calculate the correlation between the two variables of power data difference and battery temperature difference error and the parameters recorded by the metering unit (voltage, current, usage time, ambient temperature and humidity, line loss, battery charge state) to solve the Pearson correlation coefficient. The expression of the Pearson correlation coefficient is:

[0109]

[0110] in, represents the Pearson correlation coefficient, is the covariance of the two samples, and are the standard deviations of the two samples. Parameters such as voltage, current, usage time, temperature and humidity, line loss, battery charge status, etc.; variables For parameters such as power data, battery temperature data, etc. is the number of variables.

[0111] The influencing factors that lead to the data difference are determined based on the size of the correlation coefficient, and different error threshold judgment standards are adopted based on the main cause. According to prior experience, the relationship between the common charging pile charging failure factors and the error threshold is as follows:

[0112] Temperature Effects: Temperature changes can cause an error threshold in charging efficiency between 5% and 20%, depending on the magnitude of the temperature change and the type of charger.

[0113] Charging pile module failure: Module protection or failure may cause charging interruption, and the error threshold is 100%.

[0114] Communication problem between the charging station and the vehicle: Communication interruption may cause charging failure, with an error threshold of 100%.

[0115] Frequent power outages during charging: May result in insufficient charge. The error threshold depends on the frequency and duration of power outages.

[0116] Charger problem: may cause a significant drop in charging efficiency, and the error threshold depends on the specific condition of the charger.

[0117] Exception Type Error threshold Temperature Effect 5%-20% Charging pile module failure 100% Car-pile communication problem 100% Frequent power outages during charging —— Charger Problems ——

[0118] Step 4.1 uses the results of cluster analysis as input parameters and uses the LSTM neural network to establish a time series abnormal data diagnosis model.

[0119] like Figure 3 As shown, some parameters of the model are set as follows:

[0120] The cells in the model are cells that retain the memory of historical data features; the input units are difference data samples after AP clustering; the output units are difference data samples that consider the impact of historical data timing and retain some data features through forget gate, input gate, and output gate calculations.

[0121] Forget gate expression:

[0122] f t = σ ( W f [ h t − 1 , x t ] + b f )

[0123] in, , Represent the weight and bias of the forget gate respectively; Represents the current input value; The value of the previous state of the cell; Represents the activation function. The smaller the value of , the more past values ​​are forgotten.

[0124] The input gate is composed of the activation layer and the tanh layer, which determines the information that needs to be updated. The output expression of the activation layer is:

[0125] i t = σ ( W i [ h t − 1 , x i ] + b i )

[0126] in, , denote the weight and bias of the activation layer respectively.

[0127] The hyperbolic tangent activation function forms the tanh layer, creating a candidate value that may be added to the cell state. .

[0128] C t ~ = tanh ( W c [ h t − 1 , x t ] + b c )

[0129] Candidate value Combine the old state of the cell Update to new status :

[0130]

[0131] in, Represents the multiplication of corresponding elements.

[0132] The new state of the cell is calculated and combined with the output gate to form the final output containing the forgetting and memory parameters:

[0133] O t = σ ( W o [ h t − 1 , x i ] + b o )

[0134] in, , Represent the weight and bias of the output gate respectively.

[0135] Step 4.2 The steps of the anomaly diagnosis algorithm of the LSTM recurrent neural network are as follows:

[0136] 4.2.1 Data preprocessing: Clean, remove noise and label the difference data after AP clustering, assign a category label of the corresponding abnormal data type to each sample. At the same time, divide the data set into training set, validation set and test set.

[0137] 4.2.2 Model settings: set learning rate, batch size and number of iterations, initialize weights and biases;

[0138] 4.2.3 Model training: The preprocessed data training set is input into the LSTM model through the neural network input gate to calculate the prediction results; the loss function value is calculated based on the prediction results and the true labels, and the loss function is optimized using the gradient descent method; at the end of each training cycle, the validation set is used to evaluate the model performance (accuracy, recall rate), and the learning rate, batch size and other hyperparameters are adjusted according to the validation results to improve the model performance.

[0139] 4.2.4 Model testing: Use the test set to evaluate the model performance and obtain the final performance indicators. Deploy the trained LSTM model to the actual application scenario to provide users with real-time diagnostic suggestions or warnings.

[0140] Step 5. Calculate the error index (root mean square error) of the vehicle charging data in real time, and input the data exceeding the error threshold range into the LSTM diagnostic model to identify abnormal data and causes.

[0141]

[0142] Among them, RMSE is the error index of vehicle charging metering data. is the ideal value for measuring electric energy. is the actual value of the metered electric energy, and n is the number of samples.

[0143] In another embodiment, a charging pile abnormal data diagnosis system based on AP clustering and LSTM network is provided, characterized in that: it is used to implement the charging pile abnormal data diagnosis method based on AP clustering and LSTM network, including a sample library establishment unit, a clustering unit, an error threshold judgment unit, an LSTM neural network unit, and an identification unit, wherein:

[0144] The sample library establishment unit is used to measure the pile-end data in real time using the charging pile metering unit, and at the same time receive the vehicle-end BMS information transmitted to the charging pile. The real-time power data of the complete charging cycle of the vehicle and the pile is collected on-site using standard metering instruments. The pile-end data recorded and calculated by the charging pile metering unit and the power data and battery temperature data measured by the standard metering instrument are subtracted to establish a difference sample library.

[0145] The clustering unit is used to perform AP self-clustering analysis on the data abnormality sample library, and the data is automatically divided into several categories according to its own characteristics.

[0146] The error threshold judgment unit is used to introduce the Pearson correlation analysis model, calculate the correlation between the electric energy data difference and the battery temperature difference errors and the parameters recorded by the metering unit category by category, determine the influence degree of the main cause of the data difference according to the size of the correlation coefficient, and adopt different error threshold judgment standards based on the influence degree.

[0147] The LSTM neural network unit is used to use the results obtained after cluster analysis as input parameters and use the LSTM neural network to establish a time series abnormal data diagnosis model.

[0148] The identification unit is used to calculate the error index of the vehicle-pile charging metering data in real time, input the data exceeding the error threshold range into the diagnosis model, and realize the identification of abnormal data and causes.

[0149] In another embodiment, a computer system is provided, including a memory and a processor, wherein the memory is used to store a computer program / instruction. The processor is used to execute the computer program / instruction to implement the charging pile abnormal data diagnosis method based on AP clustering and LSTM network.

[0150] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A charging pile abnormal data diagnosis method based on AP clustering and LSTM network, characterized in that: The following steps are involved: Step 1: Use the charging pile metering unit to measure the pile-side data in real time, and receive the vehicle-side BMS information transmitted to the charging pile; Use standard metering instruments to collect real-time power data of the complete charging cycle of the vehicle and the charging pile on site; subtract the pile-end data recorded and calculated by the charging pile metering unit from the power data and battery temperature data measured by the standard metering instruments to establish a difference sample library; Step 2: Perform AP self-clustering analysis on the abnormal sample library of the difference sample library data, and the data is automatically divided into several categories according to its own characteristics; establish the objective function of AP clustering based on the established difference sample library; the objective function of AP clustering is as follows: in, represents the clustering objective function value, is called a variable node. is the number of nodes, , , All are function nodes; constraint function 1: in, To solve the first constraint of the objective function, ensure When it is selected as the center by other points, it is also its own center. Solve the maximum value of the first constraint of the objective function; constraint function 2: in, To solve the second constraint of the objective function, each data point can only choose one point as its center, that is, the center is unique; when =1, point For point the center of; in, Represents a similarity measure of the input The functions combined together, where is the opposite of the Euclidean distance, By two constraint functions and Keep making corrections until it stops; Step 3: Introduce the Pearson correlation analysis model to calculate the correlation between the power data difference and the battery temperature difference error and the parameters recorded by the metering unit. Determine the influence of the main cause of the data difference according to the correlation coefficient, and use different error threshold judgment standards based on the influence degree. Step 4: Use the results of cluster analysis as input parameters and use the LSTM neural network to establish a time series abnormal data diagnosis model; Step 5: Calculate the error index of vehicle-pile charging metering data in real time, and input the data exceeding the error threshold range into the diagnostic model to identify abnormal data and causes.

2. The charging pile abnormal data diagnosis method based on AP clustering and LSTM network according to claim 1 is characterized in that: In step 2, set the intermediate transmission information, update the information according to the maximum sum rule, and finally use the information value to calculate ; in: and Responsible information and available information, respectively. express As The appropriate size of the center, passing from point arrive Information between; Available information The role is to transfer from point To point information, indicating that under the condition that information has been obtained from other data points, it is used as The degree of possibility of the center; and All of them are intermediate variables of the AP clustering algorithm. After multiple iterations of the AP clustering algorithm, the relationship matrix between all the data points can be obtained. , to achieve the purpose of automatically clustering sample sets into several categories based on data features; Each message is set to its previous update value. Multiply the value of this information update times, that is, the t+1th time and The iteration value of is: ;in, represents the attenuation coefficient, Indicates assignment, Indicates responsibility information, Represents prior responsibility information, Indicates the information value obtained by prior responsibility information, Indicates available information. represents the prior available information, Indicates the information value obtained by the prior available information, Indicates the number of iterations.

3. The charging pile abnormal data diagnosis method based on AP clustering and LSTM network according to claim 2 is characterized in that: Step 3 specifically includes the following steps: Step 3.1, calculating the correlation between the two variables of electric energy data difference and battery temperature difference error and the parameters recorded by the metering unit for each category to solve the Pearson correlation coefficient; Step 3.2, determining the influencing components of the main cause of the data difference according to the size of the correlation coefficient, and using different error threshold judgment standards based on the main cause.

4. The charging pile abnormal data diagnosis method based on AP clustering and LSTM network according to claim 3 is characterized by: Based on the method of using different error threshold judgment standards for the main cause, different error thresholds are used for judgment through the relationship between the prior experience charging pile charging failure triggering factors and the error threshold. The relationship between the prior experience charging pile charging failure triggering factors and the error threshold is as follows: Temperature influence: Temperature changes may cause the error threshold of charging efficiency to be between 5% and 20%, depending on the amplitude of temperature change and the type of charging pile; Charging pile module failure: Module protection or failure may cause charging interruption, and the error threshold is 100%; Communication problems between the charging pile and the vehicle: Communication interruption may cause charging failure, and the error threshold is 100%; Frequent power outages during charging: May result in insufficient charging, and the error threshold depends on the frequency and duration of power outages; Charger problems: May cause a significant drop in charging efficiency, and the error threshold depends on the specific condition of the charger.

5. The charging pile abnormal data diagnosis method based on AP clustering and LSTM network according to claim 4 is characterized in that: The LSTM neural network includes an input unit, a forget gate, an input gate, and an output gate. The input unit is a difference data sample after AP clustering; the output unit is a difference data sample that considers the influence of historical data time series and retains some data features after calculation by the forget gate, input gate, and output gate; the forget gate expression is: in, represents the forgetting value, , Represent the weight and bias of the forget gate respectively; Represents the current input value; The value of the previous state of the cell; Represents the activation function; the input gate is composed of the activation layer and the tanh layer, which determines the information that needs to be updated; the output expression of the activation layer is: in, represents the output of the activation layer, , Represent the weight and bias of the activation layer respectively; the hyperbolic tangent activation function constitutes the tanh layer, creating a candidate value that may be added to the cell state ; in, is the candidate value, is the tanh layer bias; candidate value Combine the old state of the cell Update to new status : in, Represents the multiplication of corresponding elements; the new state of the cell is calculated and combined with the output gate to form the final output containing forgetting and memory parameters: in, Indicates the final output, , Represent the weight and bias of the output gate respectively.

6. The charging pile abnormal data diagnosis method based on AP clustering and LSTM network according to claim 5 is characterized by: The method of using LSTM neural network to establish a time series abnormal data diagnosis model in step 4 is as follows: Step 4.2.1 Data preprocessing: Clean, remove noise and label the difference data after AP clustering, assign a category label of the corresponding abnormal data type to each sample; at the same time, divide the data set into training set, validation set and test set; Step 4.2.2 Model setup: set learning rate, batch size and number of iterations, initialize weights and biases; Step 4.2.3 Model training: Input the preprocessed data training set into the LSTM neural network model through the neural network input gate and calculate the prediction results; calculate the loss function value based on the prediction results and the true labels, and optimize the loss function using the gradient descent method; at the end of each training cycle, use the validation set to evaluate the model performance, and adjust the learning rate and batch size hyperparameters based on the validation results to improve the model performance; Step 4.2.4 Model testing: Use the test set to evaluate the model performance and obtain the final performance indicators; deploy the trained LSTM neural network model to the actual application scenario to provide users with real-time diagnostic suggestions or warnings.

7. The charging pile abnormal data diagnosis method based on AP clustering and LSTM network according to claim 6 is characterized by: The formula for the error index of vehicle charging metering data is: Among them, RMSE is the error index of vehicle charging metering data. is the ideal value for measuring electric energy. is the actual value of the metered electric energy, and n is the number of samples.

8. The charging pile abnormal data diagnosis method based on AP clustering and LSTM network according to claim 7 is characterized by: In step 1, the data on the pile end include voltage, current, usage time, temperature and humidity, and line loss. The BMS information on the vehicle end includes battery cell voltage, battery state of charge, and battery temperature.

9. A charging pile abnormal data diagnosis system based on AP clustering and LSTM network, characterized by: The method for diagnosing abnormal data of a charging pile based on AP clustering and LSTM network is used to implement any one of claims 1-8, comprising a sample library establishment unit, a clustering unit, an error threshold judgment unit, an LSTM neural network unit, and an identification unit, wherein: the sample library establishment unit is used to use the charging pile metering unit to measure the pile end data in real time, and at the same time receive the vehicle end BMS information transmitted to the charging pile; use standard metering instruments to collect real-time electric energy data of the complete charging cycle of the vehicle and the pile on site; the pile end data recorded and calculated by the charging pile metering unit and the electric energy data and battery temperature data measured by the standard metering instrument are subtracted to establish a difference sample library; the clustering unit is used to perform AP self-clustering on the data abnormality sample library Analysis, the data is automatically divided into several categories according to its own characteristics; the error threshold judgment unit is used to introduce the Pearson correlation analysis model, and calculate the correlation between the electric energy data difference and the battery temperature difference error and the parameters recorded by the metering unit, and determine the influence of the main cause of the data difference according to the size of the correlation coefficient, and adopt different error threshold judgment standards based on the influence degree; the LSTM neural network unit is used to use the results obtained after cluster analysis as input parameters, and use the LSTM neural network to establish a time series abnormal data diagnosis model; the identification unit is used to calculate the error index of the vehicle charging metering data in real time, and input the data exceeding the error threshold range into the diagnosis model to realize the identification of abnormal data and causes.

10. A computer system, characterized in that: It includes a memory and a processor, the memory is used to store computer programs / instructions; the processor is used to execute the computer programs / instructions to implement the charging pile abnormal data diagnosis method based on AP clustering and LSTM network as described in any one of claims 1-8.

Images