A method for detecting errors in charging equipment for electric vehicles in high temperature areas

Abstract

The present invention provides an error detection method for electric vehicle charging equipment in high temperature areas, which belongs to the technical field of electric vehicle charging equipment errors, and includes: first, real-time collection of the above-mentioned multiple data, and through spectrum analysis, Bayesian network, time series analysis and other means, determine the key spectrum feature points of various sub-data and their high temperature effects. Then, use the pre-trained error model to calculate the error of each sub-data, analyze its influence on charging metering, and weighted sum to obtain the comprehensive error. Finally, a Kalman filter is used to correct the comprehensive error in real time, eliminate random noise and instantaneous fluctuations, and obtain a stable and reliable charging metering error. The solution of the present invention constructs an intelligent error detection system with comprehensive perception, fine modeling and dynamic correction, which can effectively improve the metering accuracy and reliability of the electric vehicle charging process, and solves the technical problem that the metering error of the existing electric vehicle charging equipment in high temperature areas is difficult to predict.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric vehicle charging equipment errors, and in particular, relates to an electric vehicle charging equipment error detection method in high temperature areas. Background Art

[0002] As an environmentally friendly and energy-saving new energy means of transportation, electric vehicles have received widespread attention and popularity in recent years. With the rapid growth of electric vehicle ownership, the use environment of charging equipment has become increasingly complex and diverse. Especially in high-temperature areas, when charging equipment is exposed to high temperatures for a long time, various electronic components and mechanical parts inside it will experience thermal imbalance, resulting in various abnormal conditions during the charging process, which seriously affects the accuracy and reliability of charging measurement.

[0003] At present, the industry has developed some monitoring and diagnostic technologies for electric vehicle charging equipment in high temperature environments. For example, some charging equipment is equipped with temperature sensors to monitor the temperature changes of key parts inside the equipment in real time and issue an alarm when the temperature exceeds the set threshold; some charging equipment is also made of low-temperature materials to improve the high-temperature resistance of the equipment itself. In addition, some charging equipment will automatically adjust the output voltage and current according to the ambient temperature to reduce power loss under high temperature. These technologies can alleviate the impact of high temperature on the performance of charging equipment to a certain extent, but there are still the following shortcomings:

[0004] 1. Temperature monitoring is limited to a few key points inside the equipment, which cannot fully reflect the actual working status of the charging equipment in a high temperature environment and it is difficult to accurately locate the source of the fault;

[0005] 2. The passive temperature regulation strategy cannot actively compensate for the fluctuations of various working parameters in high temperature environments, making it difficult to ensure the long-term stability of charging metering;

[0006] 3. The lack of comprehensive analysis of the complex electromagnetic and mechanical characteristics of charging equipment at high temperatures makes it difficult to establish an accurate thermal-electrical-mechanical coupling model, and it is impossible to accurately predict and correct errors.

[0007] In summary, the existing electric vehicle charging equipment in high temperature areas has the technical problem of difficult-to-predict metering errors. Summary of the invention

[0008] In view of this, the present invention provides an error detection method for electric vehicle charging equipment in high temperature areas, which can solve the technical problem that the metering error of existing electric vehicle charging equipment in high temperature areas is difficult to predict.

[0009] The present invention is achieved in that:

[0010] The present invention provides an error detection method for electric vehicle charging equipment in high temperature areas, which includes the following steps:

[0011] S10, in the process of charging of the electric vehicle by the charging device, acquiring temperature data, charging data, electromagnetic field data and vibration data in real time, wherein the temperature data, charging data, electromagnetic field data and vibration data each include a plurality of sub-data;

[0012] S20, using Fourier transform and wavelet analysis on the changes of each seed data to obtain a spectrum feature set;

[0013] S30, using a Bayesian network to determine the key points in each spectrum feature set;

[0014] S40, using time series analysis and a temperature sensitivity model to calculate the key points corresponding to each sub-data to obtain a high temperature impact fluctuation matrix of the sub-data;

[0015] S50, inputting the high temperature influence fluctuation matrix of each seed data into the pre-trained first error model, respectively, to obtain error data of each seed data, recorded as sub-data error;

[0016] S60, using multiple linear regression analysis and principal component analysis to obtain the influence of each sub-data error on the charging metering of the charging device, and performing weighted summation of the sub-data errors according to the influence;

[0017] S70, inputting the weighted summed comprehensive error into a pre-trained second error model to obtain a preliminary charging measurement error;

[0018] S80, using a Kalman filter to perform real-time correction on the preliminary charging measurement error, eliminating the influence of random noise and instantaneous fluctuations, and ultimately obtaining a stable and reliable charging measurement error.

[0019] Among them, the multiple sub-data of the temperature data are specifically the ambient temperature, the temperature of the gun tip of the charging equipment, and the temperature of the contact surface between the gun tip and the electric vehicle battery; the multiple sub-data of the charging data are specifically the voltage, current, power, charging time, and charging amount; the multiple sub-data of the electromagnetic field data are specifically the electric field strength, magnetic field strength, and electromagnetic interference strength around the charging equipment; the multiple sub-data of the vibration data are specifically the vibration frequency, amplitude, and acceleration of the charging gun tip.

[0020] The first error model adopts a deep residual neural network structure; the second error model adopts a hybrid model embedded in a long short-term memory network;

[0021] The second error model takes into account the interaction and nonlinear relationship between the sub-data errors.

[0022] Specifically, the first error model adopts a deep residual neural network structure, with ResNet-50 as the backbone network, and dynamically adjusts the weights and adaptively optimizes the structure of the convolutional layer, batch normalization layer, activation function layer and residual connection in ResNet according to the frequency of the key points, so as to obtain the specific structure of the first error model. The steps of constructing its training set are as follows: obtaining each seed data collected by a high-precision sensor in the laboratory with multiple groups of different ambient temperatures, charging powers and battery states as precise data, and using standard charging piles and charging equipment in actual use to collect data under the same conditions as ordinary data, and using S10-S40 to obtain the high temperature influence fluctuation matrix of each seed data as training input; using the precise data and the ordinary data, the relative error and absolute error of each sub-data are calculated as training output.

[0023] The second error model adopts a hybrid model embedded in a long short-term memory network (LSTM), with Transformer as the backbone network, and adaptively optimizes and dynamically adjusts the multi-head attention mechanism, feedforward neural network, position encoding and layer normalization in Transformer, and dynamically adjusts the thresholds of the forget gate and input gate according to the frequency of the key points in a hierarchical manner for the hidden layer of LSTM, and finally forms the structure of the second error model. The construction steps of its training set are specifically: obtaining multiple groups of charging data of different charging cycles, environmental conditions and battery types, using steps S10-S60 to obtain the timing error of the weighted summed comprehensive error as the training input, and obtaining the charging measurement collected by the high-precision electricity metering equipment used in the laboratory, and calculating the error with the charging measurement output by the charging equipment, and obtaining the standard timing error as the training output.

[0024] The first error model adopts a deep residual neural network structure, uses ResNet-50 as the backbone network, and dynamically adjusts the weights and adaptively optimizes the structure of the convolutional layer, batch normalization layer, activation function layer, and residual connection in ResNet according to the frequency of the key points to obtain the specific structure of the first error model. Specifically:

[0025] 1. Dynamically adjust the kernel size of the convolution layer and set different receptive field sizes according to the key point frequency. The higher the frequency, the smaller the convolution kernel; the lower the frequency, the larger the convolution kernel.

[0026] 2. Adaptively adjust the batch normalization layer and dynamically adjust the normalization parameters according to the key point frequency. For high-frequency key points, increase the normalization strength; for low-frequency key points, reduce the normalization strength.

[0027] 3. Perform nonlinear mapping optimization on the activation function layer and select different activation functions according to the key point frequency. For high-frequency key points, use fast-converging activation functions such as ReLU; for low-frequency key points, use smoother activation functions such as Swish.

[0028] 4. Perform adaptive weight allocation on residual connections and adjust the contribution of short-circuit connections according to the frequency of key points. For high-frequency key points, increase the weight of residual connections; for low-frequency key points, reduce the weight of residual connections.

[0029] 5. Introduce the attention mechanism to dynamically adjust the importance of each feature channel according to the frequency of key points, and enhance the model's ability to perceive features of different frequencies.

[0030] The hidden layer of the LSTM dynamically adjusts the thresholds of the forget gate and the input gate according to the frequency of the key points in a hierarchical manner, specifically:

[0031] 1. Divide the hidden layer of LSTM into multiple levels, each level corresponds to the key points of different frequency ranges. For example, it can be divided into high-frequency layer, medium-frequency layer and low-frequency layer.

[0032] 2. For each level of LSTM unit, dynamically adjust the thresholds of the forget gate and input gate according to the corresponding key point frequency range. The adjustment method is as follows:

[0033] a) Forget gate threshold adjustment:

[0034] ; in, is the output of the forget gate, is the sigmoid function, is the weight matrix, is the hidden state at the last moment, is the current input, is the bias term, Based on the key point frequency Dynamic adjustment factor.

[0035] ;

[0036] in, To adjust the rate parameter, is the frequency threshold.

[0037] b) Input gate threshold adjustment:

[0038] ;

[0039] in, is the input gate output, is the weight matrix, is the bias term, Based on the key point frequency Dynamic adjustment factor.

[0040] ;

[0041] in, To adjust the rate parameter, is the frequency threshold.

[0042] 3. For high-frequency layers, set a smaller forget gate threshold and a larger input gate threshold so that the model can respond quickly to high-frequency changes:

[0043] Set smaller and larger ;

[0044] Choose a lower frequency threshold ;

[0045] 4. For low-frequency layers, set a larger forget gate threshold and a smaller input gate threshold so that the model can maintain long-term memory:

[0046] Set up larger and smaller ;

[0047] Choose a higher frequency threshold ;

[0048] 5. The parameter setting of the mid-frequency layer is between the high-frequency layer and the low-frequency layer.

[0049] 6. During the training process, optimize through the back propagation algorithm , and Parameters such as , to adapt to the key point characteristics in different frequency ranges.

[0050] 7. During the forward propagation process, dynamically calculate the actual frequency of the key points in the input sequence and , thereby adjusting the thresholds of the forget gate and the input gate.

[0051] Through this hierarchical dynamic adjustment method, the LSTM model can better capture the time series characteristics of different frequency ranges and improve the modeling ability of complex time series data. For high-frequency key points, the model can respond and update quickly; for low-frequency key points, the model can maintain long-term memory and stability. This method helps to improve the prediction accuracy of the second error model for charging metering errors.

[0052] The key points include peak points, mutation points, inflection points and stable points.

[0053] The temperature sensitivity model includes an autoregressive moving average equation, an exponential smoothing equation, a temperature response equation, a heat diffusion equation and a nonlinear heat conduction equation.

[0054] The following is the formula expression and explanation of the equations in the time series analysis and temperature sensitivity model:

[0055] 1. The Autoregressive Moving Average (ARMA) equation takes historical time series data as input and predicts future data points as output, taking into account the autocorrelation and moving average characteristics of the data. The specific expression is as follows:

[0056] ;

[0057] In the formula, is the predicted value at time t; is a constant term; is the number of autoregressive terms; is the number of moving average items; is the autoregression coefficient; is the moving average coefficient; is the white noise term.

[0058] Parameter acquisition method:

[0059] and It is obtained by the maximum likelihood estimation method, the steps are as follows:

[0060] (1) Construct likelihood function:

[0061] ;

[0062] (2) Log-likelihood function:

[0063] ;

[0064] (3) Solving the optimal parameters: ;

[0065] 2. The exponential smoothing equation takes historical data and smoothing parameters as input and the smoothed data sequence as output, taking into account the greater impact of recent data on the forecast. The specific expression is as follows:

[0066] ;

[0067] In the formula, is the smoothed value at time t; is the actual observed value at time t; is the smoothing coefficient (0 < < 1).

[0068] The acquisition method: Using the grid search method, the steps are as follows:

[0069] (1) Setting The search range is [0.1, 0.9], with a step size of 0.1;

[0070] (2) For each Value, calculate the prediction error: ;

[0071] (3) Select the one that minimizes MSE value;

[0072] 3. The temperature response equation takes the ambient temperature change as input and the device temperature change as output, considering the device's response characteristics to ambient temperature changes. The specific expression is as follows:

[0073] ;

[0074] In the formula, is the device temperature; for time; is the heat transfer coefficient; is the ambient temperature; is the heat generation rate; For equipment quality; is the specific heat capacity.

[0075] Parameter acquisition method:

[0076] Through experimental determination, the steps are as follows:

[0077] (1) Place the device in a constant temperature environment and record the initial temperature ;

[0078] (2) At regular intervals Measuring device temperature ;

[0079] (3) Fitting curve ,get value;

[0080] 4. The heat diffusion equation takes the initial temperature distribution and boundary conditions as input, and the temperature distribution over time and space as output, considering the heat propagation process in the charging device. The specific expression is as follows:

[0081] ;

[0082] In the formula, is temperature; for time; is the thermal diffusion coefficient; is the Laplace operator; It is an internal heat source; is the density; is the specific heat capacity.

[0083] The method for obtaining: using the transient plane heat source method, the steps are as follows:

[0084] (1) Place a heat source probe in the sample and apply a constant heat flux;

[0085] (2) Record the change of probe temperature over time ;

[0086] (3) Fitting curve ,get value;

[0087] in, is heat flow, is the thermal conductivity, is the probe radius

[0088] 5. The nonlinear heat conduction equation takes temperature gradient and material properties as input and heat flux density as output, taking into account the nonlinear changes in the thermal conduction properties of materials at high temperatures. The specific expression is as follows:

[0089] ;

[0090] In the formula, is the density; is the specific heat capacity; is temperature; for time; is the temperature-dependent thermal conductivity; is the temperature-dependent heat source term.

[0091] The acquisition method: laser flashing method, the steps are as follows:

[0092] (1) Apply a transient laser pulse to the sample surface;

[0093] (2) Measure the change of back surface temperature over time ;

[0094] (3) Fitting curve

[0095] ;

[0096] (4) Based on Calculate the value;

[0097] These equations and parameter acquisition methods take into account the complex thermal behavior of charging equipment in high temperature environments and can provide accurate mathematical model support for error detection.

[0098] Wherein, the step S10 specifically includes: real-time acquisition of temperature data, charging data, electromagnetic field data and vibration data, wherein the temperature data includes the ambient temperature, the temperature of the charging device's gun tip and the temperature of the contact surface between the gun tip and the electric vehicle battery; the charging data includes voltage, current, power, charging time and charging amount; the electromagnetic field data includes the electric field strength, magnetic field strength and electromagnetic interference strength around the charging device; the vibration data includes the vibration frequency, amplitude and acceleration of the charging gun tip. By comprehensively collecting multi-dimensional data reflecting the working status of the charging device, the foundation is laid for the subsequent spectrum feature extraction and key point identification.

[0099] Wherein, the step S20 specifically includes: extracting the spectrum characteristics of each seed data by using Fourier transform and wavelet analysis. Fourier transform can convert the time domain signal into the frequency domain signal to obtain the amplitude and phase information of each frequency component; wavelet analysis can perform multi-scale decomposition of the signal in the time and frequency domain to obtain the characteristics in different time periods and frequency bands. Through these two signal processing technologies, the spectrum characteristics of each seed data are deeply excavated, laying the foundation for subsequent key point identification and error analysis.

[0100] Wherein, the step S30 specifically includes: using a Bayesian network to analyze and model the spectrum feature set, and identifying key points in each spectrum feature set, such as peak points, mutation points, inflection points, and stable points. The Bayesian network can effectively represent the probabilistic dependency relationship between variables, and accurately predict and infer key points in new data by learning the dependency relationship in the training data. These key points reflect the changing rules of the sub-data under high temperature conditions, which are crucial for subsequent error analysis and modeling.

[0101] Wherein, the step S40 specifically includes: using time series analysis and temperature sensitivity model to calculate and analyze each seed data key point identified in step S30, and obtain its fluctuation matrix under the influence of high temperature. The time series analysis method includes autoregressive moving average model and exponential smoothing model, which can predict the change trend of time series data; the temperature sensitivity model includes temperature response equation, heat diffusion equation and nonlinear heat conduction equation, which can describe the dynamic characteristics of equipment temperature change under high temperature environment. Through the application of these mathematical models, the fluctuation of each seed data under high temperature conditions is quantified, providing a basis for subsequent error analysis.

[0102] Wherein, the step S50 specifically includes: inputting the high temperature impact fluctuation matrix of each seed data obtained in step S40 into the pre-trained first error model, thereby obtaining the error data of each seed data. The first error model adopts the structure of a deep residual neural network, and dynamically adjusts the weights and optimizes the adaptive structure of key modules such as convolutional layers, batch normalization layers, activation function layers, and residual connections. These improvement measures enable the model to better capture the features of key points in different frequency ranges and improve the learning and fitting capabilities of complex data in high temperature environments.

[0103] Wherein, the step S60 specifically includes: using multiple linear regression analysis and principal component analysis to calculate the influence of each sub-data error on the charging metering of the charging equipment, and weighted summing the sub-data errors according to the influence. Multiple linear regression analysis establishes a mathematical model of sub-data error and charging metering error, and quantifies the contribution of each sub-data error to charging metering; principal component analysis can identify the main influencing factors in the sub-data error and further optimize the regression model. Through the weighted summation method, each sub-data error is integrated into an indicator representing the overall error, providing a basis for subsequent error correction.

[0104] Wherein, the step S70 specifically includes: inputting the weighted summed comprehensive error obtained in step S60 into the pre-trained second error model to obtain a preliminary charging metering timing error. The second error model adopts a hybrid model embedded in a long short-term memory network, with Transformer as the backbone network, and adaptively optimizes and dynamically adjusts its key modules such as the multi-head attention mechanism and feedforward neural network. In addition, the hidden layer of LSTM adopts a hierarchical approach to dynamically adjust the thresholds of the forget gate and the input gate according to the key point frequency, so as to better capture the timing characteristics of different frequency ranges and improve the modeling ability of complex timing data.

[0105] Wherein, the step S80 specifically includes: using a Kalman filter to perform real-time correction on the preliminary charging metering timing error obtained in step S70, eliminating the influence of random noise and instantaneous fluctuations, and finally obtaining a stable and reliable charging metering error. The Kalman filter is a recursive algorithm that can optimally estimate the state variables based on the state equation and observation equation of the system. In this patent, the Kalman filter is used to eliminate random noise and instantaneous fluctuations in the charging metering error timing data, thereby obtaining a smoother and more stable error indicator, ensuring that the final charging metering error result is reliable.

[0106] Compared with the prior art, the beneficial effects of the error detection method for electric vehicle charging equipment in high temperature areas provided by the present invention are:

[0107] 1. Comprehensively perceive the working status of charging equipment in high temperature environment. In addition to monitoring the temperature inside the equipment, it also obtains charging data such as voltage, current, power, charging time, charging amount, and electromagnetic field and vibration data around the equipment in real time, builds a multi-dimensional working status perception system, and lays the foundation for subsequent error analysis and compensation.

[0108] 2. Establish a precise thermal-electrical-mechanical coupling model. Use signal processing techniques such as Fourier transform and wavelet analysis to extract the spectral characteristics of various working data. Combined with mathematical modeling methods such as Bayesian networks, time series analysis, and temperature sensitivity models, deeply explore the changing laws of various parameters of charging equipment under high temperature environments and establish a precise thermal-electrical-mechanical coupling model.

[0109] 3. Dynamically predict and correct charging metering errors. Based on the aforementioned thermal-electrical-mechanical coupling model, a hybrid model with a deep residual neural network and embedded LSTM is used to predict and correct each sub-data error and comprehensive error respectively. Finally, a Kalman filter is used to correct the charging metering error in real time to ensure stable and reliable charging results.

[0110] Compared with the existing technology, the method of the present invention can fully perceive the working status of the charging equipment under high temperature conditions, accurately establish a thermal-electrical-mechanical coupling model, and use advanced machine learning algorithms to dynamically predict the charging metering error, solving the technical problem that the metering error of existing electric vehicle charging equipment in high temperature areas is difficult to predict. BRIEF DESCRIPTION OF THE DRAWINGS

[0111] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0112] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0113] like Figure 1 As shown, it is a flow chart of an error detection method for electric vehicle charging equipment in high temperature areas provided by the present invention. The method comprises the following steps:

[0114] S10, during the charging process of the electric vehicle, obtaining temperature data, charging data, electromagnetic field data and vibration data in real time, wherein the temperature data, charging data, electromagnetic field data and vibration data each include a plurality of sub-data;

[0115] S20, using Fourier transform and wavelet analysis on the changes of each seed data to obtain a spectrum feature set;

[0116] S30, using a Bayesian network to determine the key points in each spectrum feature set;

[0117] S40, using time series analysis and temperature sensitivity model to calculate the key points corresponding to each sub-data to obtain the high temperature impact fluctuation matrix of the sub-data;

[0118] S50, inputting the high temperature influence fluctuation matrix of each seed data into the pre-trained first error model, respectively, to obtain error data of each seed data, recorded as sub-data error;

[0119] S60, using multiple linear regression analysis and principal component analysis to obtain the influence of each sub-data error on the charging metering of the charging device, and performing weighted summation of the sub-data errors according to the influence;

[0120] S70, inputting the weighted summed comprehensive error into a pre-trained second error model to obtain a preliminary charging measurement error;

[0121] S80, using a Kalman filter to perform real-time correction on the preliminary charging measurement error, eliminating the influence of random noise and instantaneous fluctuations, and ultimately obtaining a stable and reliable charging measurement error.

[0122] The specific implementation methods of the above steps are described in detail below:

[0123] The specific implementation of step S10 is: during the charging process of the electric vehicle charging device, temperature data, charging data, electromagnetic field data and vibration data are collected and recorded in real time. Among them, the temperature data includes the ambient temperature, the temperature of the charging device's gun tip and the temperature of the contact surface between the gun tip and the electric vehicle battery; the charging data includes voltage, current, power, charging time and charging amount; the electromagnetic field data includes the electric field strength, magnetic field strength and electromagnetic interference strength around the charging device; the vibration data includes the vibration frequency, amplitude and acceleration of the charging gun tip. These collected multi-dimensional data can fully reflect the working status and performance changes of the charging equipment in a high temperature environment.

[0124] The specific implementation of step S20 is: for each seed data obtained in step S10, the spectrum features are extracted by using Fourier transform and wavelet analysis. Fourier transform can convert the time domain signal into the frequency domain signal, thereby obtaining the amplitude and phase information of each frequency component. Wavelet analysis can perform multi-scale decomposition of the signal in the time and frequency domain to obtain the characteristics of the signal in different frequency bands and time periods. Through these two signal processing methods, the spectrum features contained in each seed data can be deeply excavated, laying the foundation for subsequent key point identification and error analysis.

[0125] The specific implementation method of step S30 is: using a Bayesian network to analyze and model the spectral feature set extracted in step S20. The Bayesian network is a probabilistic graphical model that can effectively represent the conditional dependency between variables. In this patent, the Bayesian network is used to identify key points in each spectral feature set, such as peak points, mutation points, inflection points, and stable points. These key points reflect the changing patterns of sub-data under high temperature conditions and are crucial for subsequent error analysis and modeling. By learning the probabilistic dependency between variables in the training data, the Bayesian network can accurately predict and infer the key points in the new observation data.

[0126] The specific implementation method of step S40 is: using time series analysis and temperature sensitivity model, the key points of each seed data identified in step S30 are calculated and analyzed to obtain its fluctuation matrix under the influence of high temperature. The time series analysis method includes the autoregressive moving average (ARMA) model and the exponential smoothing model, which can effectively predict the change trend of time series data. The temperature sensitivity model includes the temperature response equation, the heat diffusion equation and the nonlinear heat conduction equation, which can describe the dynamic characteristics of the temperature change of the equipment under high temperature environment. Through the application of these mathematical models, the fluctuation of each seed data under high temperature conditions can be quantified, providing a basis for subsequent error analysis.

[0127] The specific implementation method of step S50 is: the high temperature impact fluctuation matrix of each seed data obtained in step S40 is respectively input into the pre-trained first error model, so as to obtain the error data of each seed data, which is the so-called sub-data error. The first error model adopts the structure of a deep residual neural network (ResNet-50), and dynamically adjusts the weights and optimizes the adaptive structure of the convolution layer, batch normalization layer, activation function layer and residual connection. These improvement measures enable the model to better capture the characteristics of key points in different frequency ranges and improve the learning and fitting capabilities of complex data in high temperature environments. Through the prediction of the first error model, the error level of each sub-data under high temperature conditions can be obtained.

[0128] The specific implementation method of step S60 is: using the methods of multiple linear regression analysis and principal component analysis to calculate the influence of each sub-data error on the charging metering of the charging device, and weighted summing the sub-data errors according to the influence. Multiple linear regression analysis can establish a mathematical model between sub-data errors and charging metering errors, and quantify the contribution of each sub-data error to charging metering. Principal component analysis can identify the main influencing factors in the sub-data errors and further optimize the input variables of the regression model. Through the weighted summation method, each sub-data error is integrated into an indicator representing the overall error of the charging device, providing a basis for subsequent error correction.

[0129] The specific implementation method of step S70 is: the weighted summed comprehensive error obtained in step S60 is input into the pre-trained second error model to obtain a preliminary charging metering timing error. The second error model adopts a hybrid model embedded in a long short-term memory network (LSTM), with Transformer as the backbone network. In Transformer, key modules such as multi-head attention mechanism, feedforward neural network, position encoding and layer normalization are adaptively optimized and dynamically adjusted. In addition, the hidden layer of LSTM adopts a hierarchical approach to dynamically adjust the thresholds of the forget gate and the input gate according to the frequency of the key points. This structural design can better capture the timing characteristics of different frequency ranges and improve the modeling ability of complex timing data, thereby obtaining a more accurate prediction of charging metering errors.

[0130] The specific implementation method of step S80 is: using the Kalman filter to perform real-time correction on the preliminary charging metering time series error obtained in step S70, eliminating the influence of random noise and instantaneous fluctuations, and finally obtaining a stable and reliable charging metering error. The Kalman filter is a recursive algorithm that can optimally estimate the state variables based on the state equation and observation equation of the system. In this patent, the Kalman filter is used to eliminate the random noise and instantaneous fluctuations in the charging metering error time series data, thereby obtaining a smoother and more stable error indicator. This error correction mechanism ensures that the final charging metering error result is reliable, and provides effective protection for the charging metering of electric vehicles in high temperature environments.

[0131] Specifically, the principle of the present invention is to build an intelligent error detection system with comprehensive perception, fine modeling and dynamic correction. Its technical principles are mainly reflected in the following aspects:

[0132] 1. Multi-dimensional working state perception. The method of the present invention not only monitors the temperature changes inside the charging device, but also obtains charging data such as voltage, current, power, and electromagnetic field and vibration data around the device in real time. These multi-dimensional data can fully reflect the complex working state of the charging device in a high temperature environment, and provide a comprehensive input basis for subsequent error analysis and compensation.

[0133] 2. Spectral feature extraction. Fourier transform and wavelet analysis techniques are used to extract the spectral features of the acquired multi-dimensional working data. These spectral features can deeply characterize the changing rules contained in the data sequence and lay the foundation for the identification of key points by models such as Bayesian networks. The identification of key points is crucial for the subsequent error modeling based on time series analysis and temperature sensitivity models.

[0134] 3. Thermal-electrical-mechanical coupling modeling. The method of the present invention adopts the ARMA model and exponential smoothing model of time series analysis to describe the dynamic change characteristics of various types of working data in the time domain; the temperature response equation, heat diffusion equation and nonlinear heat conduction equation are used to establish a thermal model of temperature change; and the key spectral feature points in various types of data are identified through the Bayesian network to provide a basis for the parameter determination of the two types of models. The comprehensive application of these mathematical models enables the present invention to accurately characterize the thermal-electrical-mechanical coupling characteristics of charging equipment in high temperature environments, providing a reliable basic support for error prediction.

[0135] 4. Intelligent error correction. Based on the aforementioned thermal-electrical-mechanical coupling model, the present invention adopts a deep residual neural network and a hybrid model embedded with LSTM to predict and compensate for each sub-data error and comprehensive error respectively. The deep residual network can adaptively adjust the structure and parameters of key modules to better capture different frequency characteristics; the hybrid model embedded with LSTM can dynamically perceive the changing rules of time series data and accurately predict the comprehensive error.

[0136] In order to better understand and implement the present invention, a specific embodiment 1 of the present invention is provided below, and the specific steps are described as follows.

[0137] In step S10, it is necessary to obtain temperature data, charging data, electromagnetic field data and vibration data in real time. The temperature data includes the ambient temperature , Charging equipment gun tip temperature And the temperature of the contact surface between the gun tip and the battery . Charging data includes voltage , Current ,power , Charging time and charge level . Electromagnetic field data includes electric field strength , magnetic field strength And electromagnetic interference intensity . Vibration data includes vibration frequency ,amplitude and acceleration The real-time collection of these multi-dimensional data lays the foundation for the subsequent spectrum feature extraction and key point identification.

[0138] In step S20, the frequency spectrum characteristics of each seed data are extracted by Fourier transform and wavelet analysis. Convert to frequency domain signal , thereby obtaining the amplitude and phase information of each frequency component:

[0139] ;

[0140] in, is the angular frequency.

[0141] Wavelet analysis can decompose the signal in multiple scales in the time-frequency domain and obtain the characteristics in different time periods and frequency bands. Wavelet transform can be expressed as:

[0142] ;

[0143] in, is the mother wavelet function, is the expansion factor, is the translation factor.

[0144] Through these two signal processing technologies, we can deeply explore the spectral characteristics of each seed data, laying the foundation for subsequent key point identification and error analysis.

[0145] In step S30, the spectral feature set extracted in step S20 is analyzed and modeled using a Bayesian network to identify the key points in each spectral feature set. A Bayesian network is a probabilistic graphical model that can effectively represent the conditional dependency between variables. Given the observed data and model parameters , Bayes’ theorem can be expressed as:

[0146] ;

[0147] in, is the posterior probability, is the likelihood function, is the prior probability, is the marginal probability. By learning the probabilistic dependencies between variables in the training data, the Bayesian network can accurately predict and infer key points in new observation data, such as peak points, mutation points, inflection points, and stable points. These key points reflect the changing patterns of sub-data under high temperature conditions and are very important for subsequent error analysis and modeling.

[0148] In step S40, time series analysis and temperature sensitivity model are used to calculate and analyze each seed data key point identified in step S30 to obtain its fluctuation matrix under the influence of high temperature. Time series analysis methods include autoregressive moving average (ARMA) model and exponential smoothing model. ARMA model can be expressed as:

[0149] ;

[0150] in, for The predicted value at the moment; is a constant term; is the number of autoregressive terms; is the number of moving average items; is the autoregression coefficient; is the moving average coefficient; is the white noise term.

[0151] The exponential smoothing model can be expressed as:

[0152] ;

[0153] in, for Smoothed value at the moment; for The actual observed value at the moment; is the smoothing coefficient.

[0154] The temperature sensitivity model includes the temperature response equation, the heat diffusion equation and the nonlinear heat conduction equation. The temperature response equation can be expressed as:

[0155]

[0156] in, is the device temperature; for time; is the heat transfer coefficient; is the ambient temperature; is the heat generation rate; For equipment quality; is the specific heat capacity.

[0157] The heat diffusion equation can be expressed as:

[0158] ;

[0159] in, is the thermal diffusion coefficient; is the Laplace operator; It is an internal heat source; is the density.

[0160] The nonlinear heat conduction equation can be expressed as:

[0161] ;

[0162] in, is the temperature-dependent thermal conductivity; is the temperature-dependent heat source term.

[0163] Through these time series analyses and temperature sensitivity models, the fluctuation of each seed data under high temperature conditions can be quantified, providing a basis for subsequent error analysis.

[0164] In step S50, the high temperature impact fluctuation matrix of each seed data obtained in step S40 is respectively input into the pre-trained first error model to obtain the error data of each seed data. The first error model adopts the structure of the deep residual neural network (ResNet-50) and adaptively optimizes its key components. Specifically, dynamic kernel size adjustment is adopted for the convolution layer:

[0165] ;

[0166] in, is the key point frequency, To adjust the rate parameter, is the frequency threshold. Adaptive normalization parameter adjustment is used for the batch normalization layer:

[0167] ;

[0168] ;

[0169] in, is the initial normalization parameter, To dynamically adjust the amplitude. Nonlinear mapping optimization is used for the activation function layer:

[0170]

[0171] in, is the Sigmoid function. Adaptive weight allocation is used for residual connections:

[0172]

[0173] in, is the input feature, is the residual branch, These improvements enable the model to better capture the characteristics of key points in different frequency ranges and improve its learning ability for complex data in high temperature environments.

[0174] In step S60, multiple linear regression analysis and principal component analysis are used to calculate the influence of each sub-data error on the charging metering of the charging equipment, and the sub-data errors are weighted and summed according to the influence. The multiple linear regression model can be expressed as:

[0175] ;

[0176] in, is the charge measurement error, For the Seed data error, is the regression coefficient, is a random error term.

[0177] Principal component analysis can identify the main influencing factors in the sub-data errors:

[0178] ;

[0179] in, is the sub-data error matrix, is the principal component matrix, is the load matrix, is the residual matrix. Through weighted summation, the errors of each sub-data are integrated into an indicator representing the overall error, providing a basis for subsequent error correction.

[0180] In step S70, the weighted summed comprehensive error obtained in step S60 is input into the pre-trained second error model to obtain a preliminary charging metering timing error. The second error model uses a hybrid model embedded in a long short-term memory network (LSTM) with Transformer as the backbone network. In LSTM, the thresholds of the forget gate and the input gate are dynamically adjusted in a hierarchical manner:

[0181] ;

[0182] ;

[0183] in, is the output of the forget gate, is the input gate output; is the weight matrix; is the bias term; Based on the key point frequency Dynamic adjustment factor:

[0184] ;

[0185] ;

[0186] in, To adjust the rate parameter, This hierarchical dynamic adjustment method enables the LSTM model to better capture the time series characteristics of different frequency ranges, thereby improving the modeling ability of complex time series data.

[0187] In step S80, the Kalman filter is used to perform real-time correction on the preliminary charging metering timing error obtained in step S70 to eliminate the influence of random noise and instantaneous fluctuations, and finally obtain a stable and reliable charging metering error. The Kalman filter is a recursive algorithm, and its state equation and observation equation are:

[0188] ;

[0189] ;

[0190] in, is the state vector, is the observation vector; They are the state transfer matrix, control matrix and observation matrix respectively; is the control input; They are state noise and observation noise respectively. Through the recursive calculation of Kalman filter, the optimal state estimation can be obtained, thereby effectively eliminating the random noise and instantaneous fluctuation in the charging metering timing error, ensuring the stability and reliability of the final error.

[0191] In order to further better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: A certain electric vehicle fast charging station is located in the middle reaches of the Yangtze River, which is a typical high temperature environment. The summer temperature is around 35°C all year round, and the heat generated by the charging equipment itself makes the temperature of the charging scene often exceed 40°C. In this case, various electronic components and mechanical parts of the charging equipment will be severely thermally stressed, resulting in a significant decrease in working performance, thereby affecting the accuracy and reliability of charging measurement.

[0192] In order to solve this problem, the charging station adopts the error detection method of electric vehicle charging equipment in high temperature areas proposed by the present invention. The specific implementation steps are as follows:

[0193] 1. Multi-dimensional work status perception

[0194] First, the charging station has installed temperature sensors on key parts of the charging pile, such as the gun head and the battery contact surface, to collect temperature data such as ambient temperature and equipment temperature in real time. At the same time, the control system of the charging pile also monitors charging data such as voltage, current, power, charging time and charging amount in real time. In addition, electromagnetic field sensors and vibration sensors are set around the charging equipment to collect data such as electromagnetic field strength, magnetic field strength, electromagnetic interference strength, and vibration frequency, amplitude and acceleration of the charging gun head.

[0195] Taking an actual charging process as an example, under the condition of ambient temperature of 35.6℃, the measurement data of the charging pile are as follows:

[0196] Temperature data: Ambient temperature = 35.6℃; Gun tip temperature = 41.2℃; battery contact surface temperature =42.8℃;

[0197] Charging data: voltage = 400V; current = 95A; Power = 38kW; Charging time = 23.5min; charging capacity = 14.9 kWh;

[0198] Electromagnetic field data: electric field strength = 1.8kV / m; magnetic field strength = 1.2A / m; electromagnetic interference intensity =0.12mT;

[0199] Vibration data: vibration frequency = 65Hz; Amplitude = 2.5mm; acceleration ;

[0200] These rich multi-dimensional data comprehensively reflect the actual working status of the charging pile in a high temperature environment, and provide a reliable input basis for subsequent error analysis and compensation.

[0201] 2. Spectral feature extraction

[0202] Next, for each seed data obtained above, Fourier transform and wavelet analysis methods are used to extract its spectral characteristics.

[0203] Taking temperature data as an example, the following frequency domain features can be obtained using Fourier transform:

[0204] Ambient temperature The amplitude spectrum of:

[0205] ;

[0206] The corresponding frequency is:

[0207] (Unit: Hz);

[0208] Gun tip temperature The amplitude spectrum of:

[0209] ;

[0210] The corresponding frequency is:

[0211] (Unit: Hz);

[0212] Battery contact surface temperature The amplitude spectrum of:

[0213] ;

[0214] The corresponding frequency is:

[0215] (Unit: Hz);

[0216] It can be seen that the three types of temperature data all show obvious characteristic peaks in the frequency domain, reflecting the periodicity of temperature changes in high temperature environments. These spectral features lay the foundation for subsequent key point identification and error modeling.

[0217] Similarly, similar Fourier transform analysis was performed on the charging data, electromagnetic field data, and vibration data to obtain their respective spectral characteristics.

[0218] In addition, wavelet analysis is used to decompose these sub-data in multiple scales and obtain the time-frequency domain characteristics. The comprehensive use of Fourier transform and wavelet analysis comprehensively extracts the spectrum characteristics of multi-dimensional working data such as temperature, charging, electromagnetic field and vibration, laying a good foundation for the next step of key point identification and error modeling.

[0219] 3. Key point identification

[0220] Based on the spectral features extracted in step 2, the Bayesian network is used to analyze and model various sub-data and identify key points.

[0221] Taking temperature data as an example, the Bayesian network discovered the following key points by learning the probabilistic dependencies between variables in the training data:

[0222] 1) Ambient temperature There is an obvious peak point near 0.38Hz, which indicates that the ambient temperature fluctuates periodically.

[0223] 2) Gun tip temperature Two peak points appeared near 0.46Hz and 0.93Hz, reflecting the double-periodic variation characteristics of the gun tip temperature.

[0224] 3) Battery contact surface temperature Two mutation points appeared near 0.53Hz and 1.02Hz, indicating that the contact surface temperature fluctuated dramatically in these two frequency bands.

[0225] Similarly, key features such as the voltage mutation point, current peak point, power inflection point in charging data, the electric field strength stability point in electromagnetic field data, and the amplitude peak point in vibration data were also identified. These key points provide an important basis for subsequent error modeling and correction.

[0226] 4. Thermal-Electrical-Mechanical Coupling Modeling

[0227] With the spectral features and key point information extracted in the previous steps, time series analysis and temperature sensitivity model are used to deeply model the thermal-electrical-mechanical coupling characteristics of charging equipment in high temperature environment.

[0228] Firstly, the autoregressive moving average (ARMA) model is used to describe the dynamic changes of various sub-data in the time domain. For example, the ARMA model can be expressed as:

[0229] ;

[0230] in, for The current value at the moment; is a white noise term. By modeling historical current data, the ARMA model can accurately predict the future current trend and provide time series feature support for error analysis.

[0231] At the same time, the exponential smoothing model is used to smooth the charging data, reducing the impact of high-frequency noise on the data sequence. For example, the exponential smoothing model can be expressed as:

[0232] ;

[0233] in, for Smooth charging at all times; is the smoothing coefficient. Through exponential smoothing, a smoother and more stable charging data series is obtained.

[0234] In addition, a series of temperature sensitivity models are established to describe the dynamic characteristics of device temperature changes in high temperature environments. For example, the temperature response equation can be expressed as:

[0235] ;

[0236] in, is the gun tip temperature; for time; is the heat transfer coefficient; is the ambient temperature; The heat generation power of the gun tip; are the mass and specific heat capacity of the gun tip, respectively. This equation describes the dynamic response characteristics of the gun tip temperature to the ambient temperature and its own power consumption.

[0237] Similarly, the heat diffusion equation and nonlinear heat conduction equation were established to describe the transient process of temperature distribution inside the charging equipment and the nonlinear change of the thermal conductivity characteristics of the material under high temperature conditions. The establishment of these temperature sensitivity models laid the foundation for the subsequent detailed analysis of the thermal-electrical-mechanical coupling characteristics of the charging process under high temperature environment.

[0238] Finally, the various key points identified in step 3 were used as key parameter inputs for the above-mentioned time series analysis model and temperature sensitivity model, and these models were further optimized and calibrated. After this series of modeling and solving processes, a complete thermal-electrical-mechanical coupling model was obtained, which can accurately describe the complex working state of the charging equipment in a high temperature environment.

[0239] 5. Intelligent error correction

[0240] With the thermal-electrical-mechanical coupling model established above, a hybrid model of deep residual neural network and embedded LSTM is used to predict and correct each sub-data error and comprehensive error respectively.

[0241] First, a deep residual neural network with ResNet-50 as the backbone network was constructed as the first error model. In this model, key components such as convolutional layers, batch normalization layers, and activation function layers were adaptively optimized:

[0242] 1) For the convolution layer, according to the key point frequency identified in step 3, dynamic kernel size adjustment technology is used. The higher the frequency, the smaller the convolution kernel is used; the lower the frequency, the larger the convolution kernel is used. This can better capture the characteristics of different frequencies.

[0243] 2) For the batch normalization layer, the normalization parameters are dynamically adjusted according to the key point frequency. For high-frequency key points, the normalization strength is enhanced; for low-frequency key points, the normalization strength is weakened. This helps improve the model's ability to perceive different frequency features.

[0244] 3) For the activation function layer, different nonlinear mapping functions are selected according to the key point frequency. For high-frequency key points, fast-converging activation functions such as ReLU are used; for low-frequency key points, smoother activation functions such as Swish are used. This adaptive nonlinear optimization enables the model to better fit the complex high-temperature data characteristics.

[0245] 4) For residual connections, their contribution is dynamically adjusted according to the frequency of key points. For high-frequency key points, the weight of residual connections is increased; for low-frequency key points, the weight of residual connections is reduced. This can adaptively balance the model's attention to different frequency features.

[0246] Through the above series of adaptive optimization measures, the first error model can better capture the complex characteristics of each sub-data under high temperature environment, and provide strong support for subsequent error prediction.

[0247] Next, a hybrid model embedded with LSTM was constructed as the second error model. This model uses Transformer as the backbone network and adopts a hierarchical dynamic adjustment strategy for the hidden layer of LSTM:

[0248] 1) Divide the hidden layer of LSTM into three levels: high-frequency layer, medium-frequency layer and low-frequency layer, which correspond to the key points in different frequency ranges identified in step 3.

[0249] 2) For the LSTM units in the high-frequency layer, set a smaller forget gate threshold and a larger input gate threshold so that it can respond quickly to high-frequency changes:

[0250] ;

[0251] ;

[0252] 3) For the LSTM unit of the low-frequency layer, set a larger forget gate threshold and a smaller input gate threshold to enable it to maintain long-term memory:

[0253] ;

[0254] ;

[0255] 4) The parameter setting of the mid-frequency layer is between the high-frequency layer and the low-frequency layer.

[0256] Through this hierarchical dynamic adjustment method, the LSTM model can better capture the time series characteristics of different frequency ranges, thereby improving the modeling ability of complex time series data.

[0257] Finally, the prediction results of the second error model are corrected in real time by using the Kalman filter, eliminating the influence of random noise and instantaneous fluctuations. Through the recursive calculation of the Kalman filter, the optimal charge metering error estimation is obtained, ensuring the stability and reliability of the final result.

[0258] 6. Implementation Effect

[0259] By adopting the error detection method for electric vehicle charging equipment in high temperature areas proposed by the present invention, the charging metering of the charging station in a high temperature environment shows significant improvement.

[0260] Taking the above charging process as an example, under the condition of ambient temperature of 35.6℃, if only traditional temperature monitoring and passive temperature adjustment strategies are used, the output power error of the charging pile is often around 5%. After adopting the method of the present invention, through multi-dimensional working state perception, thermal-electrical-mechanical coupling modeling and intelligent error correction, the error of charging measurement is finally reduced to 2.3%, which greatly improves the accuracy of charging results.

[0261] From the timing curve, in the early stage of charging, due to the influence of ambient temperature and the temperature rise of the equipment itself, the output power of the original charging pile has large fluctuations and deviations. After adopting the method of the present invention, the charging measurement result maintains good stability and reliability, and the error is basically controlled within 2.5% during the entire charging process, which is far better than the traditional solution.

[0262] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for detecting an error in an electric vehicle charging device in a high temperature area, characterized in that: The following steps are involved: S10, in the process of charging of the electric vehicle by the charging device, acquiring temperature data, charging data, electromagnetic field data and vibration data in real time, wherein the temperature data, charging data, electromagnetic field data and vibration data each include a plurality of sub-data; S20, using Fourier transform and wavelet analysis on the changes of each seed data to obtain a spectrum feature set; S30, using a Bayesian network to determine the key points in each spectrum feature set; S40, using time series analysis and a temperature sensitivity model to calculate the key points corresponding to each sub-data to obtain a high temperature impact fluctuation matrix of the sub-data; S50, inputting the high temperature influence fluctuation matrix of each seed data into the pre-trained first error model, respectively, to obtain error data of each seed data, recorded as sub-data error; S60, using multiple linear regression analysis and principal component analysis to obtain the influence of each sub-data error on the charging metering of the charging device, and performing weighted summation of the sub-data errors according to the influence; S70, inputting the weighted summed comprehensive error into a pre-trained second error model to obtain a preliminary charging measurement error; S80, using a Kalman filter to perform real-time correction on the preliminary charging measurement error, eliminating the influence of random noise and instantaneous fluctuation, and finally obtaining a stable and reliable charging measurement error; Among them, the multiple sub-data of the temperature data are specifically the ambient temperature, the temperature of the gun tip of the charging device, and the temperature of the contact surface between the gun tip and the electric vehicle battery; the multiple sub-data of the charging data are specifically the voltage, current, power, charging time, and charging amount; the multiple sub-data of the electromagnetic field data are specifically the electric field strength, magnetic field strength, and electromagnetic interference strength around the charging device; the multiple sub-data of the vibration data are specifically the vibration frequency, amplitude, and acceleration of the charging gun tip; Among them, the first error model adopts a deep residual neural network structure; the second error model adopts a hybrid model embedded in a long short-term memory network; the second error model considers the interaction and nonlinear relationship between the errors of each sub-data; The key points include peak points, mutation points, inflection points and stable points.

2. The method for detecting an error of an electric vehicle charging device in a high temperature area according to claim 1, characterized in that: The temperature sensitivity model includes an autoregressive moving average equation, an exponential smoothing equation, a temperature response equation, a heat diffusion equation and a nonlinear heat conduction equation.

3. The method for detecting an error of an electric vehicle charging device in a high temperature area according to claim 2, characterized in that: The autoregressive moving average equation takes historical time series data as input and predicts future data points as output, taking into account the autocorrelation and moving average characteristics of the data.

4. The method for detecting an error of an electric vehicle charging device in a high temperature area according to claim 2, characterized in that: The exponential smoothing equation takes historical data and smoothing parameters as input and outputs the smoothed data sequence, taking into account the greater impact of recent data on the forecast.

5. The method for detecting an error of an electric vehicle charging device in a high temperature area according to claim 2, characterized in that: The temperature response equation takes the change in ambient temperature as input and the change in device temperature as output, and considers the response characteristics of the device to the change in ambient temperature.

6. The method for detecting an error of an electric vehicle charging device in a high temperature area according to claim 2, characterized in that: The heat diffusion equation takes the initial temperature distribution and boundary conditions as input and the distribution of temperature over time and space as output, and considers the heat propagation process in the charging device.

7. The method for detecting an error of an electric vehicle charging device in a high temperature area according to claim 2, characterized in that: The nonlinear heat conduction equation takes temperature gradient and material properties as input and heat flux density as output, and considers the nonlinear change of material heat conduction properties at high temperature.

Images