A charging pile fault prediction method based on time series data enhancement

Through the timing data enhancement technology, Kalman filtering and timing generation adversarial network processing charging pile data is solved, the sample imbalance in charging pile fault prediction is improved, the accuracy of fault prediction and the generalization ability of the model are improved, and the intelligence is ensured. The stable operation of the power grid and the popularization of electric vehicles.

CN118643449BActive Publication Date: 2024-11-22国网(山东)电动汽车服务有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411105062.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-11-22
Estimated Expiration
2044-08-13

AI Technical Summary

Technical Problem

The existing charging pile fault prediction technology is scarce due to the scarcity of fault data compared to normal data, which leads to the imbalance of positive and negative samples, which reduces the model's ability to identify fault patterns and affects the quality and efficiency of charging services.

Method used

Using the method based on time-series data enhancement, the noise is filtered through Kalman filtering, the timing generation adversarial network is established for data enhancement, and the multi-layer perceptron network is used to predict the failure data of the charging pile, which solves the sample imbalance problem and improves the model Generalization ability.

Benefits of technology

It improves the accuracy of charging pile fault prediction, enhances the identification ability of the model, ensures the stable operation of the smart grid and the popularization of electric vehicles, and provides technical support for the innovation of smart charging technology and the efficient utilization of clean energy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118643449B_ABST
    Figure CN118643449B_ABST
Patent Text Reader

Abstract

The present invention proposes a charging pile fault prediction method based on time series data enhancement, which belongs to the technical field of charging piles and includes the following steps: S1: collecting the source data set of the charging pile operation, filtering the noise of the source data set through Kalman filtering, and obtaining the intermediate data set; S2: establishing a time series generative adversarial network to achieve data enhancement of the intermediate data set; S3: based on the enhanced intermediate data set, through a multi-layer perceptron network, predicting the charging pile fault data, and obtaining the target data set; S4: based on the intermediate data set and the target data set, obtaining the final charging pile fault prediction method. The charging pile fault prediction method proposed by the present invention can effectively improve the operating stability of the charging pile and the accuracy of fault detection, and is of great significance to the reliability and safety of the intelligent charging network.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of charging piles, and in particular relates to a charging pile fault prediction method based on time series data enhancement. Background Art

[0002] With the widespread use of electric vehicles, charging piles, as key facilities of smart grids, are of great significance for ensuring power supply and promoting green travel. However, charging piles may encounter various faults in actual operation. If these faults are not discovered and handled in time, they will seriously affect the quality and efficiency of charging services and may even have an adverse impact on the power grid.

[0003] In the existing charging pile fault prediction technology, since fault data is relatively scarce compared to normal operation data, there is an imbalance between positive and negative samples. This imbalance problem makes the fault prediction model tend to be biased towards the majority class (i.e. normal operation data) during training, thereby reducing the model's ability to identify fault patterns. Therefore, a charging pile data processing method is urgently needed to solve the sample imbalance problem in order to improve the accuracy of charging pile fault prediction. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, a charging pile fault prediction method based on time series data enhancement is provided.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] This technical solution proposes a charging pile fault prediction method based on time series data enhancement, which includes the following steps:

[0007] S1: Collect the source data set of the charging pile operation, filter the noise of the source data set through Kalman filtering, and obtain the intermediate data set;

[0008] S2: Establish a temporal generative adversarial network to achieve data enhancement for the intermediate data set;

[0009] S3: Based on the enhanced intermediate data set, the charging pile fault data is predicted through a multi-layer perceptron network to obtain the target data set;

[0010] S4: Based on the intermediate data set and the target data set, the final charging pile fault prediction method is obtained.

[0011] Preferably, in S1, the method for filtering noise comprises the following steps:

[0012] S11: Predict the state vector at the next time based on the current state vector. The formula is as follows:

[0013] (1);

[0014] In the formula, Represents the time step k The time step is based on the information k +1 status prediction, Indicates the current time step k The state vector of represents the state transfer matrix, represents process noise;

[0015] S12: According to the current estimation error covariance matrix, predict the estimation error covariance matrix of the next time step. The formula is as follows:

[0016] (2);

[0017] In the formula, Represents the time step k The time step is based on the information k +1 error covariance matrix prediction, Indicates the current time step k The estimated error covariance matrix of represents the state transfer matrix, represents the process noise covariance;

[0018] S13: Calculate the Kalman gain through the observation matrix and the observation noise covariance. The formula is as follows:

[0019] (3);

[0020] In the formula, Represents the time step k The Kalman gain, H represents the observation matrix, represents the observation noise covariance;

[0021] S14: Based on the current state vector and the estimated error covariance, the Kalman gain is used to update to obtain an accurate state estimate and remove the noise component in the signal. The formula is as follows:

[0022] (4);

[0023] In the formula, represents the updated state vector, represents the state vector before update, represents the Kalman gain, Represents the time step k The observation data, H represents the observation matrix, represents the updated estimation error covariance matrix, represents the estimated error covariance matrix before updating, I Represents the identity matrix.

[0024] Preferably, in S2, the data enhancement method comprises the following steps:

[0025] S21: A temporal generative adversarial network is established based on a gated recurrent unit. The gated recurrent unit uses an update gate and a reset gate to extract and learn data features. The update gate controls the retention of information from the previous time step, and the reset gate selectively forgets past information and retains important information. The formula is as follows:

[0026] (5);

[0027] In the formula, Indicates the current time step t The input feature vector is Represents the previous time step t A hidden state of -1, Represents the time step t Update the output of the gate, Represents the time step t Reset the output of the gate, Represents the time step t Candidate hidden states of Represents the time step t The final hidden state of

[0028] represents the weights and biases of the update gate, forget gate, candidate hidden state, , Represents the sigmoid activation function and the tanh activation function;

[0029] S22: Based on the automatic encoding component, the automatic encoding and decoding of data features is realized. The automatic encoding component includes an embedding function and a recovery function. The formula is as follows:

[0030] (6);

[0031] In the formula, Indicates the current time step t The input feature vector is Represents input The embedding vector after the embedding function, Represents the embedding vector The eigenvector after the restoration function is Represents the embedding vector The intermediate state vector mapped to the supervision space is used for subsequent supervised training. represents the corresponding gated recurrent unit;

[0032] S23: The adversarial generation component consists of a generator and a discriminator. The generator formula is as follows:

[0033] (7);

[0034] In the formula, Indicates the current time step t The input is random Gaussian noise, represents the hidden state generated by the generator, Indicates hidden state Representation mapped to the supervision space, Indicates hidden state The vector after the restoration function, represents the corresponding gated recurrent unit;

[0035] The discriminator formula is as follows:

[0036] (8);

[0037] In the formula, Represents the generator after noise The hidden state generated after input, Indicates hidden state Representation mapped to the supervision space, Represents input data The embedding vector after the embedding function, Represents the discrimination result of the discriminator after the corresponding input, represents the corresponding gated recurrent unit;

[0038] S24: The model is trained by minimizing the loss through an optimization algorithm. In the temporal generative adversarial network, there are three loss functions. The loss function formula is as follows:

[0039] (9);

[0040] In the formula, Represents the cross entropy loss of the discriminator’s discrimination result, N represents the number of samples, represents the discrimination result of the discriminator after the corresponding input, γ is the weight parameter, Represents the loss of the generator, including and ;

[0041] in, Calculate the embedding vector of the input The intermediate vector mapped to the supervision space The difference between Calculation Input The feature vector after the generator output is restored The difference between and Respectively represent variance calculation and mean calculation, v is the parameter;

[0042] Represents the loss of supervised learning, calculating the input The eigenvector after the restoration function The difference between The intermediate vector between the supervision space The difference between.

[0043] Preferably, in S3, the data prediction method comprises the following steps:

[0044] S31: The multi-layer perceptron structure includes input layer, hidden layer, and output layer. The formula is as follows:

[0045] (10);

[0046] In the formula, It's a neuron j Activation, Yes Connection Input i and hidden neurons j The weight of is the input signal, is the bias term of the hidden layer, n is the number of neurons in the input layer;

[0047] is the activation of the output layer, is the connection of hidden neurons j and output neurons k The weight of is the output layer bias term, m is the number of neurons in the hidden layer, Represents the sigmoid activation function.

[0048] Preferably, in S4, the charging pile fault prediction method includes the following steps:

[0049] S41: Data set merging, the formula is as follows:

[0050] (11);

[0051] In the formula, x 1 to x nIt represents the intermediate data set obtained after the n source data sets are filtered by Kalman. g 1 to g n represents the intermediate dataset after data augmentation, t 1 to t n Represents the input data after the data set is merged;

[0052] S42: Based on the data from S41, it is input into the multi-layer perceptron network for model training to predict charging pile failures. The formula is as follows:

[0053] (12);

[0054] Where L represents the output of the multi-layer perceptron With the true value The difference between them is used to predict the charging pile failure by minimizing the loss.

[0055] Compared with the prior art, the present invention has the following advantages:

[0056] This application applies data enhancement technology to time series data analysis, studies data mining methods and the application of data enhancement technology in the problem of charging pile fault prediction, and gives a charging pile fault prediction method based on time series data enhancement, which solves the problem of scarce fault samples, enhances the generalization ability of the model, and makes fault prediction more accurate, providing strong technical support for the stable operation of smart grids and the popularization of electric vehicles, promoting the innovation of smart charging technology and the efficient use of clean energy, and can effectively improve the operating stability of charging piles and the accuracy of fault detection, which is of great significance to the reliability and safety of smart charging networks.

[0057] This application can effectively alleviate the problem of insufficient charging pile fault samples, improve the accuracy of charging pile fault prediction, and can predict possible charging pile faults in advance for the smart grid system, which will help grid operators take timely measures to ensure the continuity and reliability of charging services. It can help charging pile operators optimize maintenance plans, reduce unexpected downtime, and improve the utilization rate and service quality of charging equipment. It will help to achieve coordinated optimization of charging piles and other energy management systems and improve overall energy utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0059] Figure 1 It is the overall flow chart of the present invention;

[0060] Figure 2This is a diagram of a charging pile fault prediction model in the present invention. DETAILED DESCRIPTION

[0061] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0062] like Figure 1-Figure 2 As shown, this embodiment proposes a charging pile fault prediction method based on time series data enhancement, including the following steps:

[0063] S1: Collect the source data set of the charging pile operation, filter the noise of the source data set through Kalman filtering, and obtain the intermediate data set;

[0064] S2: Establish a temporal generative adversarial network to achieve data enhancement for the intermediate data set;

[0065] S3: Based on the enhanced intermediate data set, the charging pile fault data is predicted through a multi-layer perceptron network to obtain the target data set;

[0066] S4: Based on the intermediate data set and the target data set, the final charging pile fault prediction method is obtained.

[0067] In S1, the method for filtering noise includes the following steps:

[0068] S11: Predict state vector: Predict the state vector at the next time based on the current state vector. The formula is as follows:

[0069] (1);

[0070] In the formula, Represents the time step k The time step is based on the information k +1 status prediction, Indicates the current time step k The state vector of represents the state transfer matrix, represents process noise;

[0071] S12: Predict the estimation error covariance matrix: According to the current estimation error covariance matrix, predict the estimation error covariance matrix of the next time step. The formula is as follows:

[0072] (2);

[0073] In the formula, Represents the time stepk The time step is based on the information k +1 error covariance matrix prediction, Indicates the current time step k The estimated error covariance matrix of represents the state transfer matrix, represents the process noise covariance;

[0074] S13: Calculate Kalman gain: Calculate Kalman gain through the observation matrix and observation noise covariance. The formula is as follows:

[0075] (3);

[0076] In the formula, Represents the time step k The Kalman gain, H represents the observation matrix, represents the observation noise covariance;

[0077] S14: Update state and estimated error covariance: Based on the current state vector and estimated error covariance, the Kalman gain is used to update to obtain accurate state estimation, remove noise components in the signal, and smooth the data. The formula is as follows:

[0078] (4);

[0079] In the formula, represents the updated state vector, represents the state vector before update, represents the Kalman gain, Represents the time step k The observation data, H represents the observation matrix, represents the updated estimation error covariance matrix, represents the estimated error covariance matrix before updating, I Represents the identity matrix.

[0080] In S2, the data enhancement method includes the following steps:

[0081] S21: A temporal generative adversarial network is established based on a gated recurrent unit. The gated recurrent unit uses an update gate and a reset gate to extract and learn data features. The update gate controls the retention of information from the previous time step, and the reset gate selectively forgets past information and retains important information. The formula is as follows:

[0082] (5);

[0083] In the formula, Indicates the current time stept The input feature vector is Represents the previous time step t A hidden state of -1, Represents the time step t Update the output of the gate, Represents the time step t Reset the output of the gate, Represents the time step t Candidate hidden states of Represents the time step t The final hidden state of

[0084] represents the weights and biases of the update gate, forget gate, candidate hidden state, , Represents the sigmoid activation function and the tanh activation function;

[0085] S22: In addition to the generative adversarial component, the temporal generative adversarial network also includes an automatic encoding component. Based on the automatic encoding component, the automatic encoding and decoding of data features is realized to enhance the quality of generated data. The automatic encoding component includes an embedding function and a recovery function. The formula is as follows:

[0086] (6);

[0087] In the formula, Indicates the current time step t The input feature vector is Represents input The embedding vector after the embedding function, Represents the embedding vector The eigenvector after the restoration function is Represents the embedding vector The intermediate state vector mapped to the supervision space is used for subsequent supervised training to help the generative adversarial component better capture the potential patterns in the data. represents the corresponding gated recurrent unit;

[0088] S23: The adversarial generation component consists of a generator and a discriminator. The generator formula is as follows:

[0089] (7);

[0090] In the formula, Indicates the current time step t The input is random Gaussian noise, represents the hidden state generated by the generator, Indicates hidden state Representation mapped to the supervised space, Indicates hidden state The vector after the restoration function, represents the corresponding gated recurrent unit;

[0091] The discriminator formula is as follows:

[0092] (8);

[0093] In the formula, Represents the generator after noise The hidden state generated after input, Indicates hidden state Representation mapped to the supervision space, Represents input data The embedding vector after the embedding function, Represents the discrimination result of the discriminator after the corresponding input, represents the corresponding gated recurrent unit;

[0094] S24: Loss function: The model is trained by minimizing the loss through an optimization algorithm. In the temporal generative adversarial network, there are three loss functions. The loss function formula is as follows:

[0095] (9);

[0096] In the formula, Represents the cross entropy loss of the discriminator’s discrimination result, N represents the number of samples, represents the discrimination result of the discriminator after the corresponding input, γ is the weight parameter, Represents the loss of the generator, including and ;

[0097] in, Calculate the embedding vector of the input The intermediate vector mapped to the supervision space The difference between Calculation Input The feature vector after the generator output is restored The difference between and Respectively represent variance calculation and mean calculation, v is a parameter, set to 100;

[0098] Represents the loss of supervised learning, calculating the input The eigenvector after the restoration function The difference between The intermediate vector between the supervision space The difference between.

[0099] Through the above three loss functions, the temporal generative adversarial network can ensure the accuracy of generated data and achieve high-quality data enhancement.

[0100] In S3, the data prediction method includes the following steps:

[0101] S31: Multilayer Perceptron: The structure of a multilayer perceptron is similar to the information transmission between neurons, including input layer, hidden layer, and output layer. The formula is as follows:

[0102] (10);

[0103] In the formula, It's a neuron j Activation, Yes Connection Input i and hidden neurons j The weight of is the input signal, is the bias term of the hidden layer, n is the number of neurons in the input layer;

[0104] is the activation of the output layer, is the connection of hidden neurons j and output neurons k The weight of is the output layer bias term, m is the number of neurons in the hidden layer, Represents the sigmoid activation function.

[0105] In S4, the charging pile fault prediction method includes the following steps:

[0106] S41: Data set merging, the formula is as follows:

[0107] (11);

[0108] In the formula, x 1 to x n It represents the intermediate data set obtained after the n source data sets are filtered by Kalman. g 1 to g n represents the intermediate dataset after data augmentation, t 1 to t n Represents the input data after the data set is merged;

[0109] S42: Based on the data from S41, it is input into the multi-layer perceptron network for model training to predict charging pile failures. The formula is as follows:

[0110] (12);

[0111] Where L represents the output of the multi-layer perceptron With the true value The difference between them is used to predict the failure of charging piles by minimizing the loss, and the situations under different proportions of generated data are compared to ensure the effectiveness of the time series adversarial generation network.

[0112] By applying data enhancement technology to time series data analysis, the problem of scarce fault samples is solved, the generalization ability of the model is enhanced, and fault prediction is more accurate. This provides strong technical support for the stable operation of smart grids and the popularization of electric vehicles, promotes the innovation of smart charging technology and the efficient use of clean energy, and can effectively improve the operating stability of charging piles and the accuracy of fault detection, which is of great significance to the reliability and safety of smart charging networks.

[0113] The above method can effectively alleviate the problem of insufficient charging pile fault samples, improve the accuracy of charging pile fault prediction, and can predict possible faults of charging piles in advance for the smart grid system, which helps grid operators take timely measures to ensure the continuity and reliability of charging services. It can help charging pile operators optimize maintenance plans, reduce unexpected downtime, and improve the utilization rate and service quality of charging equipment. It helps to achieve coordinated optimization of charging piles and other energy management systems and improve overall energy utilization efficiency.

[0114] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A charging pile fault prediction method based on time series data enhancement, characterized in that: The following steps are involved: S1: Collect the source data set of the charging pile operation, filter the noise of the source data set through Kalman filtering, and obtain the intermediate data set; S2: Establish a temporal generative adversarial network to achieve data enhancement for the intermediate data set; S3: Based on the enhanced intermediate data set, the charging pile fault data is predicted through a multi-layer perceptron network to obtain the target data set; S4: Based on the intermediate data set and the target data set, the final charging pile fault prediction method is obtained; In S1, the method for filtering noise comprises the following steps: S11: Predict the state vector at the next time based on the current state vector. The formula is as follows: (1); In the formula, Represents the time step k The time step is based on the information k +1 status prediction, Indicates the current time step k The state vector of represents the state transfer matrix, represents process noise; S12: According to the current estimation error covariance matrix, predict the estimation error covariance matrix of the next time step. The formula is as follows: (2); In the formula, Represents the time step k The time step is based on the information k +1 error covariance matrix prediction, Indicates the current time step k The estimated error covariance matrix of represents the state transfer matrix, represents the process noise covariance; S13: Calculate the Kalman gain through the observation matrix and the observation noise covariance. The formula is as follows: (3); In the formula, Represents the time step k The Kalman gain, H represents the observation matrix, represents the observation noise covariance; S14: Based on the current state vector and the estimated error covariance, the Kalman gain is used to update to obtain an accurate state estimate and remove the noise component in the signal. The formula is as follows: (4); In the formula, represents the updated state vector, represents the state vector before update, represents the Kalman gain, Represents the time step k The observation data, H represents the observation matrix, represents the updated estimation error covariance matrix, represents the estimated error covariance matrix before updating, I represents the identity matrix; In S2, the data enhancement method comprises the following steps: S21: A temporal generative adversarial network is established based on a gated recurrent unit. The gated recurrent unit uses an update gate and a reset gate to extract and learn data features. The update gate controls the retention of information from the previous time step, and the reset gate selectively forgets past information and retains important information. The formula is as follows: (5); In the formula, Indicates the current time step t The input feature vector is Represents the previous time step t A hidden state of -1, Represents the time step t Update the output of the gate, Represents the time step t Reset the output of the gate, Represents the time step t The candidate hidden states of Represents the time step t The final hidden state of represents the weights and biases of the update gate, forget gate, candidate hidden state, , Represents the sigmoid activation function and the tanh activation function; S22: Based on the automatic encoding component, the automatic encoding and decoding of data features is realized. The automatic encoding component includes an embedding function and a recovery function. The formula is as follows: (6); In the formula, Indicates the current time step t The input feature vector is Represents input The embedding vector after the embedding function, Represents the embedding vector The eigenvector after the restoration function is Represents the embedding vector The intermediate state vector mapped to the supervision space is used for subsequent supervised training. represents the corresponding gated recurrent unit; S23: The adversarial generation component consists of a generator and a discriminator. The generator formula is as follows: (7); In the formula, Indicates the current time step t The input is random Gaussian noise, represents the hidden state generated by the generator, Indicates hidden state Representation mapped to the supervision space, Indicates hidden state The vector after the restoration function, represents the corresponding gated recurrent unit; The discriminator formula is as follows: (8); In the formula, Represents the generator after noise The hidden state generated after input, Indicates hidden state Representation mapped to the supervision space, Represents input data The embedding vector after the embedding function, Represents the discrimination result of the discriminator after the corresponding input, represents the corresponding gated recurrent unit; S24: The model is trained by minimizing the loss through an optimization algorithm. In the temporal generative adversarial network, there are three loss functions. The loss function formula is as follows: (9); In the formula, Represents the cross entropy loss of the discriminator’s discrimination result, N represents the number of samples, represents the discrimination result of the discriminator after the corresponding input, γ is the weight parameter, Represents the loss of the generator, including and ; in, Calculate the embedding vector of the input The intermediate vector mapped to the supervision space The difference between Calculation Input The feature vector after the generator output is restored The difference between and Respectively represent variance calculation and mean calculation, v is the parameter; Represents the loss of supervised learning, calculating the input The eigenvector after the restoration function The difference between The intermediate vector between the supervision space The difference between.

2. A charging pile fault prediction method based on time series data enhancement according to claim 1, characterized in that: In S3, the data prediction method The following steps are involved: S31: The multi-layer perceptron structure includes input layer, hidden layer, and output layer. The formula is as follows: (10); In the formula, It's a neuron j Activation, Yes Connection Input i and hidden neurons j The weight of is the input signal, is the bias term of the hidden layer, n is the number of neurons in the input layer; is the activation of the output layer, is the connection of hidden neurons j and output neurons k The weight of is the output layer bias term, m is the number of neurons in the hidden layer, Represents the sigmoid activation function.

3. The charging pile fault prediction method based on time series data enhancement according to claim 1 is characterized in that: In S4, the charging pile fault prediction method includes the following steps: S41: Data set merging, the formula is as follows: (11); In the formula, x 1 to x n It represents the intermediate data set obtained after the n source data sets are filtered by Kalman. g 1 to g n represents the intermediate dataset after data augmentation, t 1 to t n Represents the input data after the data set is merged; S42: Based on the data from S41, it is input into the multi-layer perceptron network for model training to predict charging pile failures. The formula is as follows: (12); Where L represents the output of the multi-layer perceptron With the true value The difference between them is used to predict the charging pile failure by minimizing the loss.