Electric vehicle charging safety risk identification method

Abstract

The invention relates to the technical field of charging piles, in particular to an electric vehicle charging safety risk identification method, which comprises the following steps that electric vehicle user data and a charging pile local multi-dimensional data set are constructed, and the multi-dimensional data set comprises a training set and a test set; deploying a deep training model in the charging pile, and training the deep training model; after a parameter uploading request is received, the charging pile uploads parameters of an internally trained auto-encoder model to the center node; the center node aggregates the received parameters; the aggregated parameters are issued to each charging pile, and the charging piles update the parameters of the auto-encoder model after receiving the aggregated parameters; each charging pile performs electric vehicle charging safety risk identification on newly received charging pile data by using the parameter-updated auto-encoder model, and calculates the number of times of electric vehicle charging safety risks; parameters of the method can be updated in real time, and the method can be popularized to charging safety risk identification of a larger-scale charging pile.

Description

Technical Field

[0001] The present invention relates to the technical field of charging piles, and in particular to a method for identifying safety risks in charging of electric vehicles. Background Art

[0002] With the popularity of new energy vehicles, charging piles, as nodes of the charging network, support the interaction between users, vehicles, operators and power grids, and have begun to be deployed on a large scale. However, with the increase in the frequency of use of charging piles, factors such as hardware wear, equipment instability, user usage habits, and network data attacks will cause charging pile electric vehicle charging safety risks, which brings difficulties to electric vehicle users and also brings great burdens and obstacles to the equipment operation and maintenance of power grid companies and operators. Therefore, how to accurately identify the charging safety risks of electric vehicles has become a problem that needs to be solved urgently.

[0003] In the related art, the identification of electric vehicle charging safety risks mostly relies on the hardware or dynamic data of a single charging pile, and the single data or multiple data are fused and evaluated as the identification result; however, the above method has low efficiency and accuracy, and lacks perception and real-time update of changes in equipment and users; in addition, the above method is generally called through displayed data, but the problem of malicious attacks and leakage during the transmission process is difficult to avoid, and relying solely on limited charging pile data is likely to cause local optimality, which is difficult to promote to larger-scale charging pile electric vehicle charging safety risk identification. Summary of the invention

[0004] In view of at least one of the above technical problems, the present invention provides a method for identifying electric vehicle charging safety risks, which uses a training model to collect data and processes the data through an encoder model to monitor electric vehicle charging safety risks.

[0005] According to a first aspect of the present invention, a method for identifying electric vehicle charging safety risks is provided, comprising the following steps:

[0006] Constructing a local multidimensional dataset of electric vehicle user data and charging piles, wherein the multidimensional dataset includes a training set and a test set;

[0007] Deploy the deep training model in the charging pile and train the deep training model;

[0008] After receiving the parameter upload request, the charging pile uploads the parameters of the internally trained autoencoder model to the central node;

[0009] The central node aggregates the received parameters;

[0010] The aggregated parameters are sent to each charging pile, and the charging pile updates the parameters of the autoencoder model after receiving the aggregated parameters;

[0011] Each charging pile uses the autoencoder model with updated parameters to identify the electric vehicle charging safety risks of the newly received charging pile data and calculate the number of electric vehicle charging safety risks.

[0012] In some embodiments of the present invention, the contents of the training set and the test set include charging pile hardware data and user data;

[0013] The charging pile hardware data includes charging efficiency, power factor, single-gun output power, mean time between failures, mean time to repair, actual charging pile availability, voltage limiting capability, current limiting capability, output voltage error, output current error, ripple factor, real-time remaining power, battery temperature, battery health status, and charge and discharge times;

[0014] The user data includes the time of connecting to the charging pile, the time of completing charging, the time of disconnecting from the charging pile, the measured transmission energy, the charging transaction session ID, the charging pile ID, the time zone, the pilot signal of the time series, the charging current of the time series, the user ID, the estimated time of leaving the charging pile and the estimated user's demand for charging energy.

[0015] In some embodiments of the present invention, training the deep training model includes reconstruction loss function construction, encoder data dimensionality reduction, decoder data recovery, reconstruction loss function calculation and electric vehicle charging safety risk threshold evaluation.

[0016] In some embodiments of the present invention, in the construction of the reconstruction loss function, the calculation formula of the reconstruction loss function is:

[0017]

[0018] Among them, L(θ,φ) is the loss function, N is the number of samples, and D θ represents the output of the decoder (parameter is θ), E φ is the output of the encoder (parameter is φ), x j is a data sample with d dimensions; φ and θ are the parameters of the encoder and decoder, and the corresponding encoder and decoder are represented as E and D.

[0019] In some embodiments of the present invention, the calculation formula for the encoder data dimensionality reduction is:

[0020] R j =E φ (x j )

[0021] The calculation formula for the decoder data recovery is:

[0022] (x j )′=D θ (E φ (x j ))

[0023] Among them, E φ is the output of the encoder (the input is x j , parameter is φ), let the encoder reduce the data to the output R j , D θ represents the output of the decoder (the input is E φ , parameter is θ), that is, the recovered data, that is, (x j )′.

[0024] In some embodiments of the present invention, the reconstruction loss function calculation and electric vehicle charging safety risk threshold evaluation include the following steps:

[0025] The parameters φ and θ are optimized iteratively to minimize the reconstruction loss function. The electric vehicle charging safety risk threshold is δ, which is adjusted according to the actual training situation.

[0026] In some embodiments of the present invention, the charging pile uploads the parameters of the internally trained autoencoder model to the central node, including: training the autoencoder N rounds, calculating the variance of the autoencoder parameters, sorting the autoencoder parameters by variance and setting the variance threshold, and uploading the effective parameters of the autoencoder.

[0027] In some embodiments of the present invention, the autoencoder is trained for N rounds using a deep training model, and the autoencoder parameters after each round of training are counted;

[0028] The formula for calculating the variance of the autoencoder parameters is:

[0029]

[0030] Among them, θ j ,φ j are the parameters of the i-th and j-th round decoders and encoders, and is the mean of the decoder and encoder parameters of the i-th and j-th rounds, var(θ j ) and var(φ j ) is the variance of the decoder and encoder parameters of the i-th and j-th rounds.

[0031] In some embodiments of the present invention, the autoencoder parameters are sorted by variance and the variance threshold is set, including: sorting all model parameters from large to small according to the variance calculation formula, and the variance threshold ξ; uploading the effective parameters of the autoencoder includes removing the parameters ranked in front ξ and behind ξ according to the variance threshold ξ, and then uploading the remaining parameters to the cloud center node.

[0032] In some embodiments of the present invention, the parameters are aggregated including the following steps:

[0033] The parameters sent by each charging pile are numerically sorted according to parameter type, the variance of each parameter is calculated, and then the model parameters are sorted from large to small according to the variance, the top 20% and bottom 20% parameters are removed, and the geometric mean of the remaining parameters is calculated;

[0034] After the parameters are aggregated, a charging pile reputation record is also included, and the charging pile reputation record includes the following steps: the central node simultaneously records the charging pile IDs corresponding to the parameters ranked in the top 20% and the bottom 20%; after any charging pile is recorded three times, if the model parameters uploaded three times are ranked in the top 20% or the bottom 20%, it will not participate in the next round of parameter upload.

[0035] The beneficial effects of the present invention are as follows: the present invention uploads and re-aggregates the parameters of the encoder model, transmits the aggregated parameters to the charging pile, the charging pile updates the encoder parameters, and uses the updated parameters to detect the charging pile; compared with the prior art, the present invention has higher efficiency and accuracy, and the parameters can be updated in real time, which can be extended to the larger-scale charging pile electric vehicle charging safety risk identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 This is a flowchart of the steps of the method for identifying the charging safety risks of electric vehicles in an embodiment of the present invention;

[0038] Figure 2 This is a flow chart of the steps of aggregating received parameters by a central node in an embodiment of the present invention;

[0039] Figure 3 is a transformation flow chart of data conversion in an embodiment of the present invention;

[0040] Figure 4This is a relationship diagram of the charging pile and the central node during operation in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0042] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only and do not represent the only implementation method.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0044] like Figures 1 to 4 The electric vehicle charging safety risk identification method shown includes the following steps:

[0045] S100: Build a local multidimensional dataset of electric vehicle user data and charging piles, which includes a training set and a test set;

[0046] S200: deploying a deep training model in a charging pile, and training the deep training model;

[0047] S300: After receiving the parameter upload request, the charging pile uploads the parameters of the internally trained autoencoder model to the central node;

[0048] S400: The central node aggregates the received parameters;

[0049] S500: Sending the aggregated parameters to each charging pile, and the charging pile updates the parameters of the autoencoder model after receiving the aggregated parameters;

[0050] S600: Each charging pile uses the autoencoder model with updated parameters to identify the electric vehicle charging safety risks for the newly received charging pile data, and calculates the number of electric vehicle charging safety risks.

[0051] like Figure 1 and Figure 4As shown in , the multidimensional data set collects various data of the charging pile and the user data of the charging pile 02. The training set is used to train the deep training model, and the test set is used to verify the accuracy and generalization ability of the model; the autoencoder model is deployed and trained inside the charging pile so that it can learn the data characteristics of the charging pile; the charging pile uploads the parameters of the trained autoencoder model to the central node for global parameter aggregation and optimization; the time for uploading the model parameters to the central node can be after the charging pile completes the training of the autoencoder model, or after receiving the upload parameter request; the algorithm for aggregating the parameters by the central node can use the distributed parameter aggregation method, such as the aggregation algorithm in federated learning, to aggregate the parameters to obtain the globally optimal model parameters, thereby achieving model optimization; the central node sends the aggregated optimal model parameters to each charging pile, updates its internal autoencoder model, and enables it to have a stronger ability to identify electric vehicle charging safety risks; each charging pile uses the autoencoder model with updated parameters to process the newly received charging pile data, and identifies whether there is an electric vehicle charging safety risk by comparing the difference between the actual data and the model predicted data.

[0052] Through deep learning and distributed parameter updating, the parameters of the autoencoder model are continuously optimized to enable it to better adapt to the actual situation of the charging pile and improve the accuracy of electric vehicle charging safety risk identification; real-time identification of electric vehicle charging safety risks can effectively prevent charging piles from being maliciously attacked or abused, and ensure the safety and stability of the electric vehicle charging process; distributed parameter aggregation is adopted, and there is no need to upload the original data of the charging pile to the central node, which effectively protects the data privacy of the charging pile and the user; through an automated electric vehicle charging safety risk identification mechanism, manual intervention and operation and maintenance costs are reduced, and the operation and maintenance efficiency of the charging pile system is improved; compared with the prior art, the present invention has higher efficiency and accuracy, and the parameters will be updated in real time, which can be extended to larger-scale charging pile electric vehicle charging safety risk identification.

[0053] In some embodiments of the present invention, Figure 3 As shown in , the contents of the training set and the test set include charging pile hardware data 01 and user data 02;

[0054] Charging pile hardware data 01 includes charging efficiency, power factor, single-gun output power, mean time between failures, mean time to repair failures, actual charging pile utilization efficiency, voltage limiting capability, current limiting capability, output voltage error, output current error, ripple factor, real-time remaining power, battery temperature, battery health status, and charge and discharge times;

[0055] Charging efficiency reflects the ability of charging piles to convert electrical energy into battery storage energy. Changes in electric vehicle charging safety risks may cause equipment aging or failure. Power factor measures the efficiency of charging piles in the charging process. Low power factor may indicate harmonic or reactive power problems. Single-gun output power is the power output of each charging gun, which is used to monitor the usage of each charging port. Mean time between failures and mean time to repair faults are key indicators for evaluating the reliability and maintenance efficiency of charging piles. The actual available efficiency of charging piles is an efficiency indicator that reflects the actual usage of charging piles, which is affected by many factors, such as user demand and equipment failure. Voltage limiting energy The force and current limiting capabilities reflect the charging pile's ability to control the output voltage and current, and are important parameters for protecting the safety of batteries and charging facilities; the output voltage error and output current error are important indicators for measuring the output accuracy of the charging pile. Excessive errors may damage the battery or affect the charging efficiency; the ripple factor is an indicator for measuring the smoothness of the output current or voltage waveform. A high ripple factor may have a negative impact on battery life; the real-time remaining power refers to the power of the charging pile's internal energy storage device, not the battery power, and is used to monitor the charging pile's own energy reserves; the battery temperature, battery health status, and number of charge and discharge times can be used to evaluate the charging pile's ability to manage the battery.

[0056] User data 02 includes the time of connecting to the charging pile, the time of completing charging, the time of disconnecting from the charging pile, the measured transmission energy, the charging transaction session ID, the charging pile ID, the time zone, the pilot signal of the time series, the charging current of the time series, the user ID, the estimated time of leaving the charging pile and the estimated user's demand for charging energy.

[0057] The time of connecting to the charging pile, the time of completing charging, and the time of disconnecting from the charging pile record the user's charging behavior, which is used to analyze charging habits and demand patterns; the measured transmission energy is the actual electric energy transmitted to the electric vehicle, which is used for billing and efficiency evaluation; the charging transaction session ID, charging pile ID, and user ID are data that uniquely identify each charging transaction, charging pile, and user, which facilitates data tracking and analysis; due to the time difference in different regions, the time zone needs to be determined to ensure the consistency and comparability of the data; the time series pilot signal and the time series charging current provide real-time data during the charging process, which are used to analyze the dynamic changes in the charging process; the estimated time of leaving the charging pile and the estimated user's demand for charging energy are estimates based on historical data or user behavior predictions, which can be used to optimize the scheduling and energy management of charging piles.

[0058] In some embodiments of the present invention, training the deep training model includes reconstruction loss function construction, encoder data dimensionality reduction, decoder data recovery, reconstruction loss function calculation and electric vehicle charging safety risk threshold evaluation.

[0059] The reconstruction loss function is a key indicator to measure the difference between the decoder output and the original input. This loss function is usually used to evaluate the decoder's ability to restore the original data. The reconstruction loss function includes mean square error, cross entropy loss or other custom loss functions for specific data characteristics. The meaning of encoder data dimensionality reduction is that the encoder is responsible for mapping the input data, that is, high-dimensional data 03, to a lower-dimensional space, such as Figure 3 As shown in , this process is implemented through a multi-layer neural network, in which each layer learns higher-level features of the input data. The purpose of dimensionality reduction is to remove redundancy and noise in the data while retaining enough information so that the decoder can reconstruct the original data; the decoder is the inverse process of the encoder, recovering the original data from the output of the encoder, that is, from the low-dimensional data 04. Continue to refer to Figure 3 , the decoder is also implemented through a multi-layer neural network. Each layer attempts to map the low-dimensional representation back to the dimension of the original data. The performance of the decoder directly affects the quality of the reconstructed data, and then affects the ability of the entire autoencoder model to identify electric vehicle charging safety risks; during the training process, the reconstruction loss function is used to calculate the error between the decoder output and the original input. This error value is used as a gradient signal in the back-propagation algorithm to optimize the parameters of the encoder and decoder. By minimizing the reconstruction loss, the model can learn how to compress and restore data more effectively; in order to use the trained autoencoder model for electric vehicle charging safety risk identification, it is necessary to set a threshold to distinguish between normal data and electric vehicle charging safety risk data. This threshold is usually determined based on the value of the reconstruction loss. During the training process, the reconstruction loss distribution of the training data can be calculated, and a reasonable threshold can be set based on the distribution. In the test phase, if the reconstruction loss of the new data exceeds this threshold, the data is considered to be an electric vehicle charging safety risk.

[0060] In some embodiments of the present invention, in constructing the reconstruction loss function, the calculation formula of the reconstruction loss function is:

[0061]

[0062] Among them, L(θ,φ) is the loss function, N is the number of samples, and D φ represents the output of the decoder (parameter is θ), E φ is the output of the encoder (parameter is φ), x j is a data sample with d dimensions; φ and θ are the parameters of the encoder and decoder, and the corresponding encoder and decoder are represented as E and D.

[0063] The reconstruction loss function type is the application of mean square error in the autoencoder model to measure D θ (E φ (x j )) and the original input data sample xj The difference between E φ Denotes the encoder function with parameter φ, which maps the input data to a low-dimensional space, D θ represents the decoder function with parameters θ, which recovers the original data from the low-dimensional representation. L(θ, φ) is the loss function to be minimized, which depends on the parameters φ and θ of the encoder and decoder; The average loss of all N samples can be calculated, which helps the model to be more stable during training and reduce the excessive impact of a single sample on the overall loss function; (x j -D θ (E φ (x j ))) 2 Calculate a single sample x j The squared error of the reconstruction loss measures the original input x j With decoder output D θ (E φ (x j )); It should be pointed out here that φ and θ will be updated during the training process through optimization algorithms such as gradient descent.

[0064] In some embodiments of the present invention, the calculation formula for encoder data dimensionality reduction is:

[0065] R j =E φ (x j )

[0066] The calculation formula for decoder data recovery is:

[0067] (x j )′=D θ (E φ (x j ))

[0068] Among them, E φ is the output of the encoder (the input is x j , parameter is φ), let the encoder reduce the data to the output R j , D θ represents the output of the decoder (the input is E φ , parameter is θ), that is, the recovered data, that is, (x j )′.

[0069] x j is the jth input data sample, E φ is the encoder function with parameter φ, R j is the encoded low-dimensional representation. The encoder is a neural network that maps the input data to a low-dimensional space through multiple layers of nonlinear transformations.j )′=D θ (E φ (x j )), D θ is a decoder function with parameter θ, which tries to transform the encoded low-dimensional representation R j An approximation x of the original data is recovered from j ,Similarly, the decoder is also a neural network that maps the low-dimensional representation back to the dimensions of the original data through multiple layers of nonlinear transformations.

[0070] In some embodiments of the present invention, the reconstruction loss function calculation and electric vehicle charging safety risk threshold evaluation include the following steps:

[0071] The parameters φ and θ are optimized iteratively to minimize the reconstruction loss function. The electric vehicle charging safety risk threshold is δ, which is adjusted according to the actual training situation.

[0072] The electric vehicle charging safety risk threshold δ is used for application on charging piles. This threshold can be a percentage of the distribution, or determined based on other statistical criteria. In the test set or actual application, for new data samples, its reconstruction loss is calculated. If it exceeds the electric vehicle charging safety risk threshold δ, the sample is considered to be an electric vehicle charging safety risk.

[0073] In some embodiments of the present invention, the charging pile uploads the parameters of the internally trained autoencoder model to the central node, including: training the autoencoder N rounds, calculating the variance of the autoencoder parameters, sorting the autoencoder parameters by variance and setting the variance threshold, and uploading the effective parameters of the autoencoder.

[0074] The deep training process usually involves multiple iterations. In each iteration, the model calculates the loss based on the current parameters and input data, and updates the parameters through the back-propagation algorithm to minimize the loss. After the training is completed, the charging pile needs to calculate the variance of the model parameters. Parameter variance is an indicator of the discreteness of the distribution of parameter values. For each parameter in the neural network, such as weights and biases, the variance of the distribution of the change or final value during the entire training process can be calculated to identify which parameters have a greater impact on the performance of the model and which parameters are relatively unimportant or redundant. The charging pile will sort the parameters according to the calculated parameter variance. After sorting, a threshold is set to distinguish between "important" parameters and "unimportant" parameters. Parameters with larger variances are generally considered to be more important because their changes have a greater impact on model performance. Finally, the charging pile will only upload "valid" parameters whose variance exceeds the set threshold to the central node, which can significantly reduce the amount of data that needs to be transmitted while retaining the parameters that have the greatest impact on model performance. After receiving these parameters, the central node can rebuild or update the global autoencoder model for further analysis or deployment.

[0075] In some embodiments of the present invention, the autoencoder is trained for N rounds using a deep training model, and the autoencoder parameters after each round of training are counted;

[0076] The formula for calculating the variance of the autoencoder parameters is:

[0077]

[0078] Among them, θ j ,φ j are the parameters of the i-th and j-th round decoders and encoders, and is the mean of the decoder and encoder parameters of the i-th and j-th rounds, var(θ j ) and var(φ j ) is the variance of the decoder and encoder parameters of the i-th and j-th rounds.

[0079] The importance of parameters can be determined through variance calculation. The calculated variance is used to determine the importance of parameters by threshold. The higher the importance, the greater the impact on model performance. Finally, the charging pile will only upload the "valid" parameters whose variance exceeds the set threshold to the central node.

[0080] In some embodiments of the present invention, the autoencoder parameters are sorted by variance and the variance threshold is set, including: sorting all model parameters from large to small according to the variance calculation formula, and the variance threshold ξ; uploading the effective parameters of the autoencoder includes removing the parameters ranked in front ξ and behind ξ according to the variance threshold ξ, and then uploading the remaining parameters to the cloud center node.

[0081] The conversion format involved in parameter upload can be JSON, Protobuf or binary format. By filtering out some parameters, it can ensure that valid data is transmitted to the central cloud node, ensuring data validity and model training effect.

[0082] In some embodiments of the present invention, Figure 2 As shown in , parameter aggregation includes the following steps:

[0083] S410: sorting the parameters sent by each charging pile by parameter type, calculating the variance of each parameter, and then sorting the model parameters from large to small according to the variance, removing the top 20% and bottom 20% parameters, and calculating the geometric mean of the remaining parameters;

[0084] S420: The central node simultaneously records the charging pile IDs corresponding to the parameters ranked in the top 20% and the bottom 20%. After any charging pile is recorded three times, if the model parameters uploaded three times are ranked in the top 20% or the bottom 20%, it will not participate in the next round of parameter upload.

[0085] After the parameters are aggregated, the charging pile reputation record is also included. The charging pile reputation record includes the following steps:

[0086] Collect model parameters from each charging pile. These parameters include voltage, current, temperature, charging efficiency and other types of values. The collected data may exist in different formats, such as JSON, XML or binary format, depending on the communication protocol between the charging pile and the central node; variance calculation is to quantify the degree of discreteness of a set of values, which is independent of the order of the values. Variance is a statistic that measures the degree of discreteness of data. High variance means that the difference between data points is large, which may indicate that the parameter has a large variability between different charging piles; according to a predetermined ratio, remove the parameters with the largest and smallest variances. The purpose of this step is to remove parameters that may be less important due to the safety risk value of electric vehicle charging or low correlation.

[0087] During the variance sorting and screening process, the IDs of charging piles whose variances are in the top 20% and bottom 20% are recorded at the same time. These IDs will be used for subsequent credit records and analysis; for each charging pile, check the number of times it is recorded in the top 20% or bottom 20%. If a charging pile appears in these extreme positions multiple times, such as 3 times or more, this may indicate that there are problems with the data quality of the charging pile, such as sensor failure, data tampering, or electric vehicle charging safety risk operations. Based on the results of the credit check, decide whether to prohibit certain charging piles from participating in the next round of parameter uploads; prohibited charging piles need to be further inspected, repaired or replaced to ensure the accuracy and reliability of their data; the credit record of the charging pile is updated during each parameter aggregation process to ensure its accuracy and effectiveness.

[0088] Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for identifying electric vehicle charging safety risks, characterized in that: The following steps are involved: Constructing a local multidimensional dataset of electric vehicle user data and charging piles, wherein the multidimensional dataset includes a training set and a test set; Deploy the deep training model in the charging pile and train the deep training model; After receiving the parameter upload request, the charging pile uploads the parameters of the internally trained autoencoder model to the central node; The central node aggregates the received parameters; The aggregated parameters are sent to each charging pile, and the charging pile updates the parameters of the autoencoder model after receiving the aggregated parameters; Each charging pile uses the autoencoder model with updated parameters to identify the electric vehicle charging safety risks of the newly received charging pile data and calculate the number of electric vehicle charging safety risks.

2. The electric vehicle charging safety risk identification method according to claim 1 is characterized in that: The content of the training set and the test set includes charging pile hardware data and user data; The charging pile hardware data includes charging efficiency, power factor, single-gun output power, mean time between failures, mean time to repair, actual charging pile availability, voltage limiting capability, current limiting capability, output voltage error, output current error, ripple factor, real-time remaining power, battery temperature, battery health status, and charge and discharge times; The user data includes the time of connecting to the charging pile, the time of completing charging, the time of disconnecting from the charging pile, the measured transmission energy, the charging transaction session ID, the charging pile ID, the time zone, the pilot signal of the time series, the charging current of the time series, the user ID, the estimated time of leaving the charging pile and the estimated user's demand for charging energy.

3. The electric vehicle charging safety risk identification method according to claim 1 is characterized in that: Training the deep training model includes reconstruction loss function construction, encoder data dimensionality reduction, decoder data recovery, reconstruction loss function calculation and electric vehicle charging safety risk threshold evaluation.

4. The electric vehicle charging safety risk identification method according to claim 3 is characterized in that: In the reconstruction loss function construction, the calculation formula of the reconstruction loss function is: Among them, L(θ,φ) is the loss function, N is the number of samples, and D θ represents the output of the decoder (parameter is θ), E φ is the output of the encoder (parameter is φ), x j is a data sample with d dimensions; φ and θ are the parameters of the encoder and decoder, and the corresponding encoder and decoder are represented as E and D.

5. The electric vehicle charging safety risk identification method according to claim 4 is characterized in that: The calculation formula for the encoder data dimensionality reduction is: R j =E φ (x j ) The calculation formula for the decoder data recovery is: (x j )′=D θ (E φ (x j )) Among them, E φ is the output of the encoder (the input is x j , parameter is φ), let the encoder reduce the data to the output R j , D θ represents the output of the decoder (the input is E φ , parameter is θ), that is, the recovered data, that is, (x j )′.

6. The electric vehicle charging safety risk identification method according to claim 5 is characterized in that: The reconstruction loss function calculation and electric vehicle charging safety risk threshold evaluation include the following steps: The parameters φ and θ are optimized iteratively to minimize the reconstruction loss function. The electric vehicle charging safety risk threshold is δ, which is adjusted according to the actual training situation.

7. The electric vehicle charging safety risk identification method according to claim 1 is characterized in that: The charging pile uploads the parameters of the internally trained autoencoder model to the central node, including: training the autoencoder for N rounds, calculating the variance of the autoencoder parameters, sorting the autoencoder parameters according to the variance and setting the variance threshold, and uploading the effective parameters of the autoencoder.

8. The electric vehicle charging safety risk identification method according to claim 7 is characterized in that: The autoencoder is trained N rounds using a deep training model, and the autoencoder parameters after each round of training are counted; The autoencoder parameter variance calculation formula is: Among them, θ j ,φ j are the parameters of the i-th and j-th round decoders and encoders, and is the mean of the decoder and encoder parameters of the i-th and j-th rounds, var(θ j ) and var(φ j ) is the variance of the decoder and encoder parameters of the i-th and j-th rounds.

9. The electric vehicle charging safety risk identification method according to claim 7, characterized in that: The autoencoder parameters are sorted by variance and the variance threshold is set, including: sorting all model parameters from large to small according to the variance calculation formula, and the variance threshold ξ; the autoencoder effective parameter uploading includes removing the parameters ranked in front ζ and behind ζ according to the variance threshold ζ, and then uploading the remaining parameters to the cloud center node.

10. The electric vehicle charging safety risk identification method according to claim 1, characterized in that: The parameters are aggregated including the following steps: The parameters sent by each charging pile are numerically sorted according to parameter type, the variance of each parameter is calculated, and then the model parameters are sorted from large to small according to the variance, the top 20% and bottom 20% parameters are removed, and the geometric mean of the remaining parameters is calculated; After the parameters are aggregated, a charging pile reputation record is also included, and the charging pile reputation record includes the following steps: the central node simultaneously records the charging pile IDs corresponding to the parameters ranked in the top 20% and the bottom 20%; after any charging pile is recorded three times, if the model parameters uploaded three times are ranked in the top 20% or the bottom 20%, it will not participate in the next round of parameter upload.

Images