LLM-based vehicle charging pile operation state evaluation and prediction method

Abstract

The invention provides an LLM-based vehicle charging pile operation state evaluation and prediction method, and belongs to the field of vehicle charging piles. Data such as alarm records and reasons of the automobile charging pile, charging records, direct current charging pile BMS data, heartbeat data, communication protocol response duration, charging current, charging voltage, gun line temperature, the maximum temperature of a battery pack, loss metering charging degrees, fault conditions, maintenance records, models and specifications are collected, and data standardization storage is carried out. And cleaning, denoising and normalizing the collected data of the automobile charging pile. A large amount of historical data is used to enable the model to learn the complex relationship between the operation state of the charging pile and each piece of collected data. And combining the trained large model with a traditional evaluation and prediction algorithm. And analyzing and processing the real-time operation data of the charging pile by using the constructed evaluation and prediction model.

Description

Technical Field

[0001] The present invention relates to the field of vehicle charging piles, and particularly to a method for evaluating and predicting the operating state of a vehicle charging pile based on LLM. Background Art

[0002] With the rapid development of the electric vehicle industry, charging piles, as key facilities for electric vehicle charging, their operating state directly affects the charging experience of electric vehicle users and the safety of charging facilities. Traditional methods for evaluating the operating state of charging piles mainly rely on manual inspections and simple sensor monitoring, making it difficult to achieve accurate prediction and timely warning of the operating state of charging piles. Therefore, developing a method that can efficiently and accurately evaluate and predict the operating state of charging piles based on large model technology has important practical significance and application value. Summary of the Invention

[0003] To solve the above technical problems, the present invention provides a method for evaluating and predicting the operating state of a vehicle charging pile based on LLM, which can effectively avoid problems such as single data dimension and insufficient prediction accuracy in the traditional evaluation method for evaluating the operating state of charging piles, and effectively reduce the operating cost of charging piles.

[0004] The technical solution of the present invention is as follows:

[0005] A method for evaluating and predicting the operating state of a vehicle charging pile based on LLM, which collects data such as alarm records and reasons, charging records, DC charging pile BMS data, heartbeat data, communication protocol response duration, charging current, charging voltage, gun line temperature, maximum battery pack temperature, calculated loss of charging degrees, fault conditions, maintenance records, models, specifications, etc. of the vehicle charging pile, and performs standardized storage of the data. The collected data of the vehicle charging pile is cleaned, denoised, and normalized. A large amount of historical data is used to let the model learn the complex relationship between the operating state of the charging pile and each collected data. The trained large model is combined with traditional evaluation and prediction algorithms. The constructed evaluation and prediction model is used to analyze and process the real-time operating data of the charging pile.

[0006] Furthermore,

[0007] The collected data of the vehicle charging pile is cleaned, denoised, and normalized. Data cleaning and denoising mainly remove data that significantly deviates from the actual data in the collected data, and normalization mainly reduces the impact of individual indicators with large deviations on the overall result.

[0008] Data cleaning:

[0009] Through the data cleaning method of introducing a cleaning coefficient (k), unified data cleaning work is carried out on the collected standard data;

[0010] Outlier detection is performed by an improved 3-sigma rule, and the formula is:

[0011] Qutliers = {x||x–u|>3kσ|}

[0012] where μ is the mean of the data, k is the cleaning coefficient of the data category, and σ is the standard deviation of the data;

[0013] Data denoising:

[0014] Introduce an outlier coefficient a, fit the data through a regression model, and calculate the data outliers;

[0015] e i =|y i –y| / y*a

[0016] y i is the actual value, y is the fitted value, and a is the outlier system of the data category;

[0017] If e i exceeds the set threshold H e , then this data is considered an outlier;

[0018] Data normalization processing:

[0019] Introduce a normalization coefficient (g) to normalize the collected data.

[0020] X norm =(X-X min ) / (X max -X min )*g

[0021] X is the original data, X min and X max are the minimum and maximum values of this feature data respectively, and X norm is the normalized data; g is the normalization coefficient of the data category; in this way, data in different formats and magnitudes are converted into a unified standard format, with a value range between ([0,1]) for subsequent analysis and modeling.

[0022] Furthermore,

[0023] Use a large amount of historical data to let the model learn the complex relationship between the operating status of the charging pile and each collected data, and continuously adjust the model parameters through the backpropagation algorithm to minimize the loss function. Continuously adjust the model parameters to make the loss function value the smallest, thereby improving the accuracy and generalization ability of the model.

[0024] Select the Transformer architecture as the basis of the large model. In the Transformer architecture, the following calculation formula is adopted: (Attention(Q, K, V) = softmax(QK T \sqrt(d k ))V

[0025] where Q, K, and V are input vectors, and d k is the dimension of the key vector K;

[0026] Use the multi-head attention mechanism to calculate the attention in parallel, capture the dependencies in different aspects of the data, and the calculation formula is:

[0027] MultiHead(Q, K, V) = Concat(head1,…,head h )W O

[0028] head i = Attention(QW i Q , KW i K , VW i V )

[0029] where h is the number of heads, and QW i Q , KW i K , VW i V and W O are trainable weight matrices.

[0030] Furthermore,

[0031] Combine the trained large model with traditional evaluation and prediction algorithms to give full play to the powerful data analysis ability of the large model and the targeted advantages of traditional algorithms, and build a more accurate and reliable evaluation and prediction model for the operating status of electric vehicle chargers.

[0032] Adopt the ARIMA model in time series analysis, and the prediction formula is:

[0033]

[0034] where, is the autoregressive coefficient, θ i is the moving average coefficient, e t+n-i is the past prediction error, and h is the prediction step.

[0035] Furthermore,

[0036] Using the constructed evaluation and prediction model, analyze and process the real-time operation data of the charging pile. Evaluate the current operation status of the charging pile by setting thresholds. According to the learning and analysis results of the model, predict the possible fault types and fault times of the charging pile in the future period of time.

[0037] Evaluate the current operation status of the charging pile by setting thresholds. For the obtained status P of the electric vehicle charging pile, if Pmin < P < Pmax, it is determined that the charging pile is operating normally; otherwise, it is considered that there is a potential fault risk. Here, Pmin and Pmax are thresholds set under the normal operation conditions of the electric vehicle charging pile.

[0038] According to the learning and analysis results of the model, predict the possible fault types and fault times of the charging pile in the future period of time. By establishing a fault probability logistic regression model:

[0039]

[0040] where x is the input feature vector, w is the weight vector, b is the bias term, and P(y = 1|x) represents the probability of the charging pile having a fault. When the predicted probability exceeds the set fault threshold, an early warning is issued to remind the operation and maintenance personnel to perform maintenance and repair in a timely manner. Description of the Drawings

[0041] Figure 1 is a schematic diagram of the working process of the present invention. Detailed Embodiments

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0043] The present invention provides a method for evaluating and predicting the operation status of an electric vehicle charging pile based on an LLM. The method includes steps such as data collection, data preprocessing, model construction, evaluation and prediction, and warning feedback. By introducing large model technologies such as deep learning and machine learning, accurate evaluation and prediction of the operation status of the charging pile are realized.

[0044] The technical solution includes the following content:

[0045] Data collection of electric vehicle charging piles:

[0046] 1. Collect data such as BMS data, heartbeat data, communication protocol response duration, charging current, charging voltage, gun line temperature, maximum battery pack temperature, and loss-counted charging degrees of DC chargers for electric vehicles through the cloud fast charging protocol, and store the collected data in a standardized manner according to a unified coding rule.

[0047] 2. Collect data such as alarm records and reasons, charging records, fault conditions, maintenance records, models, and specifications of chargers through the charger operation platform, and store the collected data in a standardized manner according to a unified coding rule.

[0048] Data preprocessing for electric vehicle chargers:

[0049] Data cleaning method.

[0050] Perform unified data cleaning on the collected standard data through a data cleaning method that introduces a cleaning coefficient (k), and distinguish the influence degree of important data and unimportant data on the model.

[0051] Perform outlier detection through the improved 3-sigma rule. The formula is:

[0052] Qutliers = {x||x – u|>3kσ|}

[0053] Among them, μ is the mean of the data, k is the cleaning coefficient of the data category (the specific data content is calculated based on historical data), and σ is the standard deviation of the data

[0054] Data denoising method.

[0055] Introduce an outlier coefficient (a), fit the data through a regression model, and calculate the data outliers.

[0056] e i =|y i –y| / y*a

[0057] y i is the actual value, y is the fitted value, and a is the outlier system of the data category (calculated based on historical data)

[0058] If e i exceeds the set threshold H e , then this data is considered an outlier.

[0059] Data normalization method.

[0060] Introduce a normalization coefficient (g) to normalize the collected data.

[0061] X norm =(X - X min ) / (X max -Xmin )*g

[0062] Let \(X\) be the original data, \(X\) min and \(X\) max are respectively the minimum and maximum values in the feature data. \(X\) norm is the data after normalization. \(g\) is the normalization coefficient for the data category. In this way, data in different formats and magnitudes are converted into a unified standard format, with a value range between \((0, 1)\) for subsequent analysis and modeling.

[0063] Construction and training of a large model for evaluating the operating status of electric vehicle chargers:

[0064] Select the Transformer architecture as the basis of the large model. In the Transformer architecture, the following calculation formula is used: \(\text{Attention}(Q, K, V)=\text{softmax}(\frac{QK}{\sqrt{d}})V\) T \sqrt{d} k )V

[0065] where \(Q\), \(K\), and \(V\) are input vectors, and \(d\) k is the dimension of the key vector \(K\).

[0066] The multi - head attention mechanism can capture dependencies in different aspects of the data by calculating attention in parallel through multiple heads. The calculation formula is:

[0067] \(\text{MultiHead}(Q, K, V)=\text{Concat}(head1,\cdots,head_h)W\) h )W O

[0068] head i =\text{Attention}(QW^{h_i}, KW^{h_i}, VW^{h_i})\) i Q ,KW i K ,VW i V )

[0069] where \(h\) is the number of heads, and \(QW^{h_i}\), \(KW^{h_i}\), \(VW^{h_i}\), and \(W\) i Q 、KW i K 、VW i V and \(W\) O are trainable weight matrices.

[0070] During the training process, a large amount of historical data is used to let the model learn the complex relationship between the operating status of the charger and various influencing factors. The model parameters are continuously adjusted through the backpropagation algorithm to minimize the loss function. Common loss functions such as the mean squared error loss function (MSE):

[0071]

[0072] where n is the number of samples, y i is the true value, is the predicted value of the model. Continuously adjust the model parameters to minimize the loss function value, thereby improving the accuracy and generalization ability of the model.

[0073] Method for evaluating and predicting the fusion of operating state evaluation prediction models.

[0074] Combine the trained large model with traditional evaluation and prediction algorithms, and adopt the ARIMA model in time series analysis. The prediction formula is:

[0075]

[0076] where, is the autoregressive coefficient, θ i is the moving average coefficient, e t+n-i is the past prediction error, and h is the prediction step.

[0077] Evaluation of operating state evaluation prediction:

[0078] Evaluate the operating state of the current charging pile by setting thresholds. For the obtained state P of the vehicle charging pile, if Pmin < P < Pmax, it is determined that the charging pile is operating normally; otherwise, it is considered that there is a potential fault risk. Among them, Pmin and Pmax are thresholds set according to the normal operation of the vehicle charging pile.

[0079] According to the learning and analysis results of the model, predict the possible fault types and fault times of the charging pile in the future. By establishing a fault probability logistic regression model:

[0080]

[0081] where x is the input feature vector, w is the weight vector, b is the bias term, and P(y = 1|x) represents the probability of the charging pile having a fault. When the predicted probability exceeds the set fault threshold, an early warning is issued in advance to remind the operation and maintenance personnel to perform maintenance and repair in a timely manner.

[0082] The above are only the preferred embodiments of the present invention, which are only used to illustrate the technical solutions of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for evaluating and predicting the operating status of a vehicle charging pile based on LLM, characterized in that: Collect the alarm records and reasons of the car charging pile, charging records, DC charging pile BMS data, heartbeat data, communication protocol response time, charging current, charging voltage, gun line temperature, battery pack maximum temperature, damage charging degree, fault conditions, maintenance records, models, and specification data. The collected data are stored in a standardized manner according to unified coding rules, and the operating status of the car charging pile is evaluated and predicted in combination with the improved car charging pile operating status evaluation and prediction large model.

2. The method according to claim 1, characterized in that The above data are collected and preprocessed, and finally the LLM model is constructed and trained to obtain a large model for evaluating and predicting the operating status of automobile charging piles.

3. The method according to claim 1 or 2, characterized in that: Clean, denoise and normalize the collected data of car charging piles; Data cleaning and denoising are performed to remove data that deviates from the actual data, and normalization is performed to reduce the impact of indicators whose deviations exceed the set values ​​on the overall results.

4. The method according to claim 3, characterized in that Data cleaning: By introducing the data cleaning method of cleaning coefficient (k), the collected standard data are uniformly cleaned; Outlier detection is performed using the improved 3-sigma rule, and the formula is: Qutliers={x||x–u|>3kσ|} Among them, μ is the mean of the data, k is the cleaning coefficient of the category to which the data belongs, and σ is the standard deviation of the data; Data denoising: Introduce the outlier coefficient a, fit the data through the regression model, and calculate the data outliers; and i =|and i –y| / y*a y i is the actual value, y is the fitted value, and a is the classification outlier system to which the data belongs; If e i Exceeds the set threshold H e , then the data is considered to be an outlier; Data normalization processing: The normalization coefficient (g) is introduced to normalize the collected data; X norm =(X-X min ) / (X max -X min )*g X is the original data, X min and X max are the minimum and maximum values ​​of the feature data, respectively, norm is the normalized data; g is the normalization coefficient of the category to which the data belongs; in this way, data of different formats and magnitudes are converted into a unified standard format with a value range between ([0,1]) for subsequent analysis and modeling.

5. The method according to claim 3, characterized in that: Use historical data to let the model learn the complex relationship between the operating status of the charging pile and the various collected data. Use the back propagation algorithm to continuously adjust the model parameters to minimize the loss function. Continuously adjust the model parameters to minimize the loss function value, thereby improving the accuracy and generalization ability of the model.

6. The method according to claim 5, characterized in that The Transformer architecture is used as the basis of the large model. In the Transformer architecture, the following calculation formula is used: (Attention(Q,K,V)=softmax(QK T \sqrt(d k ))V Among them, Q, K, V are input vectors, d k is the dimension of the key vector K; The multi-head attention mechanism is used to calculate attention in parallel to capture the dependency management of different aspects of the data. The calculation formula is: MultiHead(Q,K,V)=Concat(head1,…,head h )W O head i =Attention(QW i Q ,KW i K ,VW i V ) Where h is the number of heads, QW i Q , KW i K 、VW i V and W O is a trainable weight matrix.

7. The method according to claim 5, characterized in that Combine the trained big model with the traditional evaluation and prediction algorithm, and use the big model data analysis capabilities and the targetedness of the traditional algorithm to build a more accurate and reliable vehicle charging pile operation status evaluation and prediction model.

8. The method according to claim 7, characterized in that Using the ARIMA model in time series analysis, the prediction formula is: in, is the autoregressive coefficient, θ i is the moving average coefficient, e t+n-i is the past prediction error, and h is the prediction step size.

9. The method according to claim 7, characterized in that: The constructed evaluation and prediction model is used to analyze and process the real-time operation data of the charging pile. The current operation status of the charging pile is evaluated by setting thresholds. Based on the learning and analysis results of the model, the type and time of possible faults that may occur in the charging pile in the future are predicted.

10. The method according to claim 9, wherein the operating state of the current charging pile is evaluated by setting thresholds. For the obtained state P of the vehicle charging pile, if Pmin < P < Pmax, it is determined that the charging pile is operating normally; otherwise, there is considered to be a potential fault risk. Here, Pmin and Pmax are thresholds set according to the normal operation of the vehicle charging pile; based on the learning and analysis results of the model, the possible fault types and fault times of the charging pile in the next period of time are predicted; by establishing a fault probability logistic regression model: where x is the input feature vector, w is the weight vector, b is the bias term, and P(y = 1|x) represents the probability of the charging pile having a fault; when the predicted probability exceeds the set fault threshold, an early warning is issued to remind the operation and maintenance personnel to perform maintenance and repair in a timely manner.

Images