A method and system for detecting automobile charging safety based on target environment data

Abstract

The present invention provides a method and system for detecting automobile charging safety based on target environmental data. The method collects real-time and historical environmental data, cleans, standardizes and normalizes the collected data, establishes a machine learning model based on the preprocessed data and predicts charging safety events, sends the detection data and prediction results to an interactive interface and displays them in a graphical form simultaneously, and sends signals for decision-making and alarming. The nonlinear regression algorithm is used to fully consider the influence of multivariate independent variables, and the multivariate independent variables are combined to be used as one variable to improve the accuracy of the prediction. K-means clustering is used to obtain the thresholds or proportions of different variable data that affect automobile safety, and random forest technology is used to fully consider the impact of multiple variables on automobile safety for probabilistic prediction. Compared with simple decision tree prediction and manual empirical threshold judgment analysis, the prediction accuracy is higher and the automation level is higher.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric vehicle charging methods, and specifically refers to a method and system for detecting vehicle charging safety based on target environment data. Background Art

[0002] In recent years, with the intensification of environmental pollution, the harm of traditional fuel vehicles to the environment has received more and more attention. With the introduction of relevant national policies and strong support for new energy vehicle companies, new energy vehicles will occupy an important position in the future automobile market. The charging safety of electric vehicles is a prerequisite for the normal operation of electric vehicles. At present, there have been reports at home and abroad that electric vehicles have experienced accidents such as reduced charging efficiency and battery aging due to environmental factors during charging. The efficiency of electric vehicle charging, battery aging, charging safety such as short circuit, open circuit, leakage, etc. are closely related to the external environment of charging. Therefore, real-time detection and early warning of charging equipment under different external environments are very important for the charging safety of electric vehicles. Summary of the invention

[0003] A method for detecting automobile charging safety based on target environment data comprises the following steps:

[0004] S1. Collect real-time and historical environmental data, charging pile data, battery data and power grid data. Environmental data include air temperature, air humidity, air pressure, rainfall, wind speed, electromagnetic interference intensity and dust. Charging pile data include the rated current of the charging pile, the rated voltage of the charging pile, the charging interface type and communication protocol of the charging pile, the aging degree of the charging interface of the charging pile and the battery temperature of the charging pile; battery data include battery temperature, battery aging degree, battery chemical composition and rated power of the battery, battery charging interface type and communication protocol, battery charging interface aging degree and battery shell damage data, where rainfall refers to the rainwater flow rate per unit time through a unit area of ​​a horizontal plane; power grid data include three-phase voltage, three-phase current and power grid frequency;

[0005] S2, data preprocessing: cleaning, standardization and normalization of the data collected in S1;

[0006] S3, establishing a machine learning model based on the data preprocessed in S2 and predicting charging safety events, where charging safety events include decreased battery charging efficiency and increased battery aging;

[0007] S4. Establish a machine learning model based on the data preprocessed in S2 and predict charging safety events, where charging safety events include charging short circuit, leakage, and charging interruption;

[0008] S5, the detection data of S2, the prediction results of S3 and S4 are sent to the interactive interface and displayed synchronously in the form of charts, and signals are sent for decision-making and alarm.

[0009] Furthermore, S3 includes the following sub-steps:

[0010] S31, collect historical data in S2 for feature engineering processing;

[0011] S32, the data processed by feature engineering is used as variables to perform nonlinear regression analysis model training, the regression coefficients of various variables are predicted by gradient descent method, the battery charging efficiency and battery aging are fitted to obtain the fitting curve, the real-time data in S2 is collected to evaluate and tune the model, and the time series analysis algorithm is used to perform trend analysis on the obtained fitting curve;

[0012] S33. Draw a trend line based on the trend analysis results, calculate the confidence interval of the trend line and set the threshold range of the battery charging efficiency and the battery aging degree, and determine whether the confidence interval is completely within the threshold range. If so, it indicates that the charging efficiency is normal or the aging degree is within the expected range. If not, send a signal for decision-making and alarm, and at the same time send the above data analysis information to the interactive interface and display it in the form of a chart.

[0013] Furthermore, the S4 includes the following sub-steps:

[0014] S41, extracting features from S2 data;

[0015] S42, collect continuous variables corresponding to charging short circuit, leakage, and charging interruption in history, use K-means clustering algorithm to obtain the center point K of different types of continuous variables corresponding to each safety event, collect this type of continuous variables in real time, and when new data is input, if the change in K value is less than the preset threshold A, determine that the real-time collected continuous variable falls within the risk threshold B, and the risk threshold B refers to a data set of a certain type of continuous variables corresponding to the above safety events. The continuous variables in this set are predicted to be the main factors causing the above safety events, where the continuous variable categories include environmental data, battery aging data, battery charging interface aging degree, and charging pile charging interface aging degree;

[0016] S43. Collect the categorical variables corresponding to the charging short circuit, leakage, and charging interruption in history, and count the proportion of various variables in a certain categorical variable corresponding to each safety event. The proportion is used as the weight. The larger the weight, the more likely the variable is to cause the corresponding safety event; the categorical variables include the rated current of the charging pile, the rated voltage of the charging pile, the charging interface type and communication protocol of the charging pile, the battery chemical composition, the rated power of the battery, and the charging interface type and communication protocol of the battery;

[0017] S44. Collect binary variables corresponding to charging short circuit, leakage, and charging interruption in history, and count the proportion of yes and no in the binary variables corresponding to each safety event respectively. The proportion is used as the weight. The larger the weight, the more likely the variable is to cause the corresponding safety event. The binary variable refers to the battery shell damage data;

[0018] S45. Collect historical environmental data, charging pile data, battery data, and power grid data, generate different decision trees by randomly combining variables, use the random deep forest algorithm to generate an integrated model for the above different decision trees, and use the real-time collected data as input into the integrated model to predict the probability of occurrence of different types of safety incidents, and issue warnings and decisions for probabilities that exceed the threshold.

[0019] Furthermore, the generation of different decision trees in S45 includes the following sub-steps:

[0020] S451, first randomly select one of the variables as a root node, and randomly generate the remaining variables as leaf nodes, to generate multiple decision trees with multiple branches;

[0021] S452. Use the recursive splitting method to make logical judgments and probability predictions on the decision tree in S451. When a node is a continuous variable, a binary choice judgment is made based on whether the value of the continuous variable detected in real time is within the threshold range; when a node is a categorical variable, identify the variable type in the categorical variable, and make multivariate choice judgments based on the different types of the categorical variable; when a node is a binary variable, identify the binary variable and make a binary choice judgment based on yes or no.

[0022] Furthermore, the specific implementation process of calculating the confidence interval of the trend line and setting the threshold range of the battery charging efficiency and the battery aging degree in step S33 is as follows:

[0023] S331: Constructing a deep learning model for threshold confidence intervals;

[0024] S332: generating feature label data by using feature engineering to calculate the confidence interval function of the trend line, to set the battery charging efficiency range function, and to set the threshold range function of the battery aging degree, wherein a part of the data is used as training data and a part of the data is used as test data;

[0025] S333: input the training data into the deep learning model of the threshold confidence interval for training, and stop the training when the model training meets the 90% prediction accuracy condition;

[0026] S334: deploying the trained deep learning model of the threshold confidence interval in an actual environment for real-time adjustment of the confidence interval and setting the threshold range of the battery charging efficiency and the battery aging degree;

[0027] Furthermore, the specific implementation process of step S43 of adjusting the proportion in real time through deep learning model prediction analysis is as follows:

[0028] S431: Constructing a deep learning model of classification weights;

[0029] S432: Labeling the classification variables corresponding to charging short circuit, leakage, and charging interruption through feature engineering to generate feature label data, part of which is used for training and part of which is used for testing data;

[0030] S433: inputting the training data into the deep learning model of classification weights for training, and stopping the training when the model training meets the condition of 90% prediction accuracy;

[0031] S434: Deploy the trained deep learning model of classification weights in an actual environment for real-time adjustment of the weights of charging short circuit, leakage, and charging interruption classifications.

[0032] Furthermore, the specific implementation process of step S44 is to adjust the weight of each binary variable in real time through deep learning model prediction analysis:

[0033] S441: Constructing deep learning models for binary variables;

[0034] S442: Labeling binary variables corresponding to charging short circuit, leakage, and charging interruption through feature engineering to generate feature label data, part of which is used for training and part of which is used for testing data;

[0035] S443: inputting the training data into a deep learning model of binary variables for training, and stopping the training when the model training meets a 90% prediction accuracy condition;

[0036] S444: Deploy the trained deep learning model of binary variables in an actual environment to adjust the weights of the binary variables of charging short circuit, leakage, and charging interruption in real time.

[0037] The present invention also includes a system for detecting automobile charging safety based on target environment data, the system comprising a data acquisition module, a preprocessing module, a machine learning module, a decision module, an alarm module and a human-computer interaction module;

[0038] The data acquisition module is used to collect real-time and historical environmental data, charging pile data, battery data and power grid data. Environmental data includes air temperature, air humidity, air pressure, rainfall, wind speed, electromagnetic interference intensity and dust. Charging pile data includes the rated current of the charging pile, the rated voltage of the charging pile, the charging interface type and communication protocol of the charging pile, the aging degree of the charging interface of the charging pile and the battery temperature of the charging pile; battery data includes battery temperature, battery aging degree, battery chemical composition and rated power of the battery, battery charging interface type and communication protocol, battery charging interface aging degree and battery shell damage data. Rainfall refers to the flow of rainwater per unit time through a unit area of ​​a horizontal plane; power grid data includes three-phase voltage, three-phase current and power grid frequency;

[0039] The preprocessing module is used to clean, standardize and normalize the collected data;

[0040] The machine learning module is used to establish a machine learning model and predict charging safety events, where charging safety events include decreased battery charging efficiency, increased battery aging, charging short circuit, leakage, and charging interruption;

[0041] The decision module is used to make corresponding decisions based on the prediction results of the machine learning module;

[0042] The alarm module is used to issue an alarm based on the prediction results of the machine learning module;

[0043] The human-computer interaction module is used to display data collection information and data analysis information in the form of charts and is used for touch screen control system operation.

[0044] Further, the machine learning module includes a feature engineering submodule, a nonlinear regression analysis submodule, a time series analysis submodule, a confidence calculation submodule, and a threshold judgment submodule;

[0045] The feature engineering submodule is used to extract features from the historical data collected by the data collection module;

[0046] The nonlinear regression analysis submodule is used to train the nonlinear regression analysis model for the data after feature extraction, predict the regression coefficient of each variable by the gradient descent method, perform model fitting on the battery charging efficiency and battery aging, and obtain the fitting curve;

[0047] The time series analysis submodule is used to perform trend analysis on the obtained fitting curve;

[0048] The confidence calculation submodule is used to generate a trend line according to the trend analysis results and calculate the confidence interval of the trend line;

[0049] The threshold judgment submodule is used to judge whether the confidence interval is completely within the threshold range of the battery charging efficiency and the battery aging degree, and send the judgment result.

[0050] Furthermore, the machine learning module also includes a K-means clustering module, a risk threshold judgment module, a categorical variable proportion statistics module, a binary variable proportion statistics module, and a random forest algorithm module;

[0051] The K-means clustering module collects the continuous variables corresponding to the charging short circuit, leakage, and charging interruption in history. After feature processing, the K-means clustering algorithm is used to obtain the center point K of different types of continuous variables corresponding to each safety event.

[0052] The risk threshold judgment module is set to receive the real-time collection of this type of continuous variables, and set the threshold A. According to the center point K and threshold A, the range of the risk threshold B is calculated using the mean square error formula. The mean square error value of the historical data and the center point K is calculated using the mean square error formula. After the new data is input, the mean square error value after the new input number is calculated in real time, and the difference is made with the previous value. If the difference is within the range of threshold A, the new input value falls within the risk threshold B;

[0053] The classification variable proportion statistics module is used to identify the variable type in the classification variable corresponding to each security event in the historical data, and count the proportion of the variable type;

[0054] The binary variable proportion statistics module is used to identify the variable type in the binary variable corresponding to each safety event in the historical data, and count the proportion of the variable type, i.e. "yes" or "no". Whether the battery shell is damaged or not is determined by the convolutional upscaling network.

[0055] The random deep forest algorithm module collects historical environmental data, charging pile data, battery data, and power grid data, generates different decision trees by randomly combining variables, and uses the random deep forest algorithm to generate an integrated model for the above different decision trees. The real-time collected data is input into the integrated model to predict the probability of occurrence of different types of security incidents, and for probabilities that exceed the threshold, early warnings and decisions are made.

[0056] Beneficial Effects

[0057] By collecting real-time and historical environmental data, charging pile data, and battery data, the machine learning model can be used to predict the charging efficiency trend and battery aging trend caused by changes in the external environment in real time, providing early warning and decision-making basis for charging efficiency decline and battery aging;

[0058] Specifically, the nonlinear regression algorithm is used to predict in real time the charging efficiency trend and battery aging trend caused by changes in the external environment. The nonlinear regression algorithm fully considers the impact of multiple independent variables, combines the multiple independent variables into one variable, and improves the accuracy of the prediction. K-means clustering is used to obtain the threshold or proportion of different variable data that affect vehicle safety, and the random forest technology is used to fully consider the impact of multiple variables on vehicle safety for probabilistic prediction. Compared with simple decision tree prediction and manual empirical threshold judgment analysis, the prediction is more accurate and has a higher level of automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 : A step diagram of a method for detecting automobile charging safety based on target environment data according to a first embodiment of the present invention;

[0060] Figure 2 : A working step diagram of S3 according to the first embodiment of the present invention;

[0061] Figure 3 : A working step diagram of S4 according to the first embodiment of the present invention;

[0062] Figure 4 : A block diagram of a system for detecting automobile charging safety based on target environment data according to a second embodiment of the present invention;

[0063] Figure 5 : Machine learning module diagram according to the second embodiment of the present invention. DETAILED DESCRIPTION

[0064] Embodiment 1

[0065] refer to Figures 1 to 3 , a method for detecting automobile charging safety based on target environment data, comprising the following steps:

[0066] S1. Collect real-time and historical environmental data, charging pile data, battery data and power grid data. Environmental data include air temperature, air humidity, air pressure, rainfall, wind speed, electromagnetic interference intensity and dust. Charging pile data include the rated current of the charging pile, the rated voltage of the charging pile, the charging interface type and communication protocol of the charging pile, the aging degree of the charging interface of the charging pile and the battery temperature of the charging pile; battery data include battery temperature, battery aging degree, battery chemical composition and rated power of the battery, battery charging interface type and communication protocol, battery charging interface aging degree and battery shell damage data, where rainfall refers to the rainwater flow rate per unit time through a unit area of ​​a horizontal plane; power grid data include three-phase voltage, three-phase current and power grid frequency;

[0067] S2, data preprocessing: cleaning, standardization and normalization of the data collected in S1;

[0068] S3, establishing a machine learning model based on the data preprocessed by S2 and predicting charging safety events, where charging safety events include decreased battery charging efficiency and increased battery aging;

[0069] S4. Establish a machine learning model based on the data preprocessed in S2 and predict charging safety events, where charging safety events include charging short circuit, leakage, and charging interruption;

[0070] S5, the detection data of S2, the prediction results of S3 and S4 are sent to the interactive interface and displayed in the form of charts simultaneously, and signals are sent for decision-making and alarm.

[0071] Furthermore, S3 includes the following sub-steps:

[0072] S31, collect historical data in S2 for feature engineering processing;

[0073] S32, the data processed by feature engineering is used as variables to perform nonlinear regression analysis model training, the regression coefficients of various variables are predicted by gradient descent method, the battery charging efficiency and battery aging are fitted to obtain the fitting curve, the real-time data in S2 is collected to evaluate and tune the model, and the time series analysis algorithm is used to perform trend analysis on the obtained fitting curve;

[0074] S33. Draw a trend line based on the trend analysis results, calculate the confidence interval of the trend line and set the threshold range of the battery charging efficiency and the battery aging degree, and determine whether the confidence interval is completely within the threshold range. If so, it indicates that the charging efficiency is normal or the aging degree is within the expected range. If not, send a signal for decision-making and alarm, and at the same time send the above data analysis information to the interactive interface and display it in the form of a chart.

[0075] Furthermore, the S4 includes the following sub-steps:

[0076] S41, extracting features from S2 data;

[0077] S42, collect continuous variables corresponding to charging short circuit, leakage, and charging interruption in history, use K-means clustering algorithm to obtain the center point K of different types of continuous variables corresponding to each safety event, collect this type of continuous variables in real time, and when new data is input, if the change in K value is less than the preset threshold A, determine that the real-time collected continuous variable falls within the risk threshold B, and the risk threshold B refers to a data set of a certain type of continuous variables corresponding to the above safety events. The continuous variables in this set are predicted to be the main factors causing the above safety events, where the continuous variable categories include environmental data, battery aging data, battery charging interface aging degree, and charging pile charging interface aging degree;

[0078] S43. Collect the categorical variables corresponding to the charging short circuit, leakage, and charging interruption in history, and count the proportion of various variables in a certain categorical variable corresponding to each safety event. The proportion is used as the weight. The larger the weight, the more likely the variable is to cause the corresponding safety event; the categorical variables include the rated current of the charging pile, the rated voltage of the charging pile, the charging interface type and communication protocol of the charging pile, the battery chemical composition, the rated power of the battery, and the charging interface type and communication protocol of the battery;

[0079] S44. Collect binary variables corresponding to charging short circuit, leakage, and charging interruption in history, and count the proportion of yes and no in the binary variables corresponding to each safety event respectively. The proportion is used as the weight. The larger the weight, the more likely the variable is to cause the corresponding safety event. The binary variable refers to the battery shell damage data;

[0080] S45. Collect historical environmental data, charging pile data, battery data, and power grid data, generate different decision trees by randomly combining variables, use the random deep forest algorithm to generate an integrated model for the above different decision trees, and use the real-time collected data as input into the integrated model to predict the probability of occurrence of different types of safety incidents, and issue warnings and decisions for probabilities that exceed the threshold.

[0081] Furthermore, the generation of different decision trees in S45 includes the following sub-steps:

[0082] S451, first randomly select one of the variables as a root node, and randomly generate the remaining variables as leaf nodes, to generate multiple decision trees with multiple branches;

[0083] S452. Use the recursive splitting method to make logical judgments and probability predictions on the decision tree in S451. When a node is a continuous variable, a binary choice judgment is made based on whether the value of the continuous variable detected in real time is within the threshold range; when a node is a categorical variable, identify the variable type in the categorical variable, and make multivariate choice judgments based on the different types of the categorical variable; when a node is a binary variable, identify the binary variable and make a binary choice judgment based on yes or no.

[0084] Furthermore, the specific implementation process of calculating the confidence interval of the trend line and setting the threshold range of the battery charging efficiency and the battery aging degree in step S33 is as follows:

[0085] S331: Constructing a deep learning model for threshold confidence intervals;

[0086] S332: generating feature label data by using feature engineering to calculate the confidence interval function of the trend line, to set the battery charging efficiency range function, and to set the threshold range function of the battery aging degree, wherein a part of the data is used as training data and a part of the data is used as test data;

[0087] S333: input the training data into the deep learning model of the threshold confidence interval for training, and stop the training when the model training meets the 90% prediction accuracy condition;

[0088] S334: deploying the trained deep learning model of the threshold confidence interval in an actual environment for real-time adjustment of the confidence interval and setting the threshold range of the battery charging efficiency and the battery aging degree;

[0089] Through the above steps, the threshold value and the confidence interval can be adjusted in real time according to the actual environment, so as to improve the accuracy of the prediction in step S33.

[0090] Furthermore, the specific implementation process of step S43 of adjusting the proportion in real time through deep learning model prediction analysis is as follows:

[0091] S431: Constructing a deep learning model of classification weights;

[0092] S432: Labeling the classification variables corresponding to charging short circuit, leakage, and charging interruption through feature engineering to generate feature label data, part of which is used for training and part of which is used for testing data;

[0093] S433: inputting the training data into the deep learning model of classification weights for training, and stopping the training when the model training meets the condition of 90% prediction accuracy;

[0094] S434: Deploy the trained deep learning model of classification weights in an actual environment for real-time adjustment of the weights of charging short circuit, leakage, and charging interruption classifications.

[0095] Furthermore, the specific implementation process of step S44 is to adjust the weight of each binary variable in real time through deep learning model prediction analysis:

[0096] S441: Constructing deep learning models for binary variables;

[0097] S442: Labeling binary variables corresponding to charging short circuit, leakage, and charging interruption through feature engineering to generate feature label data, part of which is used for training and part of which is used for testing data;

[0098] S443: inputting the training data into a deep learning model of binary variables for training, and stopping the training when the model training meets a 90% prediction accuracy condition;

[0099] S444: Deploy the trained deep learning model of binary variables in an actual environment to adjust the weights of the binary variables of charging short circuit, leakage, and charging interruption in real time.

[0100] Embodiment 2

[0101] refer to Figure 4 and Figure 5 , the present invention also includes a system for detecting automobile charging safety based on target environment data, the system comprising a data acquisition module 1, a preprocessing module 2, a machine learning module 3, a decision module 4, an alarm module 5 and a human-computer interaction module 6;

[0102] The data acquisition module 1 is used to collect real-time and historical environmental data, charging pile data, battery data and power grid data. Environmental data includes air temperature, air humidity, air pressure, rainfall, wind speed, electromagnetic interference intensity and dust. Charging pile data includes the rated current of the charging pile, the rated voltage of the charging pile, the charging interface type and communication protocol of the charging pile, the aging degree of the charging interface of the charging pile and the battery temperature of the charging pile; battery data includes battery temperature, battery aging degree, battery chemical composition and rated power of the battery, battery charging interface type and communication protocol, battery charging interface aging degree and battery shell damage data, wherein rainfall refers to the rainwater flow per unit time through a unit area of ​​a horizontal plane; power grid data includes three-phase voltage, three-phase current and power grid frequency;

[0103] The preprocessing module 2 is used to clean, standardize and normalize the collected data;

[0104] The machine learning module 3 is used to establish a machine learning model and predict charging safety events, where charging safety events include decreased battery charging efficiency, increased battery aging, charging short circuit, leakage, and charging interruption;

[0105] The decision module 4 is used to make corresponding decisions according to the prediction results of the machine learning module;

[0106] The alarm module 5 is used to issue an alarm according to the prediction result of the machine learning module;

[0107] The human-computer interaction module 6 is used to display the data collection information and data analysis information in the form of charts and graphs, and is used for the operation of the touch screen control system.

[0108] Further, the machine learning module 3 includes a feature engineering submodule 31, a nonlinear regression analysis submodule 32, a time series analysis submodule 33, a confidence calculation submodule 34, and a threshold judgment submodule 35;

[0109] The feature engineering submodule 31 is used to extract features from the historical data collected by the data collection module;

[0110] The nonlinear regression analysis submodule 32 is used to perform nonlinear regression analysis model training on the data after feature extraction, predict the regression coefficient of each variable by gradient descent method, perform model fitting on battery charging efficiency and battery aging, and obtain a fitting curve;

[0111] The time series analysis submodule 33 is used to perform trend analysis on the obtained fitting curve;

[0112] The confidence calculation submodule 33 is used to generate a trend line according to the trend analysis result and calculate the confidence interval of the trend line;

[0113] The threshold judgment submodule 34 is used to judge whether the confidence interval is completely within the threshold range of the battery charging efficiency and the battery aging degree, and send the judgment result.

[0114] Furthermore, the machine learning module 3 also includes a K-means clustering module 36, a risk threshold judgment module 37, a classification variable proportion statistics module 38, a binary variable proportion statistics module 39, and a random forest algorithm module 40;

[0115] The K-means clustering module 36 collects continuous variables corresponding to charging short circuit, leakage, and charging interruption in history, and after feature processing, uses the K-means clustering algorithm to obtain the center point K of different types of continuous variables corresponding to each safety event;

[0116] The risk threshold judgment module 37 is configured to receive and collect this type of continuous variable in real time, and set the threshold A. According to the center point K and the threshold A, the range of the risk threshold B is calculated using the mean square error formula, and the mean square error value of the historical data and the center point K is calculated using the mean square error formula. After the new data is input, the mean square error value after the new input number is calculated in real time, and the difference is made with the previous value. If the difference is within the range of the threshold A, the new input value falls within the risk threshold B;

[0117] The classification variable proportion statistics module 38 is used to identify the variable type in the classification variable corresponding to each security event in the historical data, and to count the proportion of the variable type;

[0118] The binary variable proportion statistics module 39 is used to identify the variable type in the binary variable corresponding to each safety event in the historical data, and to count the proportion of the variable type, i.e., "yes or no". Whether the battery shell is damaged or not is determined by a convolutional upgrade network.

[0119] The random deep forest algorithm module 40 collects historical environmental data, charging pile data, battery data, and power grid data, generates different decision trees by randomly combining variables, and uses the random deep forest algorithm to generate an integrated model for the above different decision trees. The real-time collected data is input into the integrated model to predict the probability of occurrence of different types of security incidents, and for probabilities that exceed the threshold, early warnings and decisions are made.

[0120] The present invention and its embodiments are described above, which is not restrictive. What is shown in the accompanying drawings is only one of the embodiments of the present invention, and the actual content is not limited to this. In short, if ordinary technicians in this field are inspired by it and do not deviate from the purpose of the invention, they can creatively design structural methods and embodiments similar to the technical scheme, which should all fall within the scope of protection of the present invention.

Claims

1. A method for detecting automobile charging safety based on target environment data, characterized in that: The steps include: S1, collect real-time and historical environmental data, charging pile data, battery data and power grid data; S2, data preprocessing: cleaning, standardization and normalization of the data collected in S1; S3, establishing a machine learning model based on the data preprocessed by S2 and predicting charging safety events, where charging safety events include decreased battery charging efficiency and increased battery aging; S4. Establish a machine learning model based on the data preprocessed in S2 and predict charging safety events, where charging safety events include charging short circuit, leakage, and charging interruption; The S4 comprises the following sub-steps: S41, extracting features from S2 data; S42, collect continuous variables corresponding to charging short circuit, leakage, and charging interruption in history, use K-means clustering algorithm to obtain the center point K of different types of continuous variables corresponding to each safety event, collect this type of continuous variables in real time, and when new data is input, if the change in K value is less than the preset threshold A, determine that the real-time collected continuous variable falls within the risk threshold B, and the risk threshold B refers to a data set of a certain type of continuous variables corresponding to the above safety events. The continuous variables in this set are predicted to be the main factors causing the above safety events, where the continuous variable categories include environmental data, battery aging data, battery charging interface aging degree, and charging pile charging interface aging degree; S43. Collect the categorical variables corresponding to the charging short circuit, leakage, and charging interruption in history, and count the proportion of various variables in a certain categorical variable corresponding to each safety event. The proportion is used as the weight, and the proportion is adjusted in real time through the deep learning model prediction analysis. The larger the weight, the more likely the variable is to cause the corresponding safety event; The categorical variables include the rated current of the charging pile, the rated voltage of the charging pile, the charging interface type and communication protocol of the charging pile, the battery chemical composition, the rated power of the battery, the charging interface type and communication protocol of the battery; S44. Collect binary variables corresponding to charging short circuit, leakage, and charging interruption in history, and count the proportion of yes and no in the binary variables corresponding to each safety event respectively. The proportion is used as the weight. The proportion is adjusted in real time through the deep learning model prediction analysis. The larger the weight, the more likely the variable is to cause the corresponding safety event; the binary variable refers to the battery shell damage data; S45. Collect historical environmental data, charging pile data, battery data, and power grid data, generate different decision trees by randomly combining variables, use the random deep forest algorithm to generate an integrated model for the above different decision trees, use the real-time collected data to input the integrated model to predict the probability of occurrence of different types of security events, and make early warnings and decisions for probabilities that exceed the threshold; The generation of different decision trees in S45 includes the following sub-steps: S451, first randomly select one of the variables as a root node, and randomly generate the remaining variables as leaf nodes, to generate multiple decision trees with multiple branches; S452, using the recursive splitting method to make logical judgments and probability predictions for the decision tree in S451, when a certain node is a continuous variable, a binary choice judgment is made based on whether the value of the continuous variable detected in real time is within the threshold range; when a certain node is a categorical variable, the variable type in the categorical variable is identified, and a multivariate choice judgment is made based on the different types of the categorical variable; when a certain node is a binary variable, the binary variable is identified, and a binary choice judgment is made based on yes or no; S5, the detection data of S2, the prediction results of S3 and S4 are sent to the interactive interface and displayed in the form of charts simultaneously, and signals are sent for decision-making and alarm.

2. The method for detecting automobile charging safety based on target environment data according to claim 1, characterized in that: The S3 comprises the following sub-steps: S31, collect historical data in S2 for feature engineering processing; S32, the data processed by feature engineering is used as variables to perform nonlinear regression analysis model training, the regression coefficients of various variables are predicted by gradient descent method, the battery charging efficiency and battery aging are fitted to obtain the fitting curve, the real-time data in S2 is collected to evaluate and tune the model, and the time series analysis algorithm is used to perform trend analysis on the obtained fitting curve; S33. Draw a trend line based on the trend analysis results, calculate the confidence interval of the trend line and set the threshold range of the battery charging efficiency and the battery aging degree, and determine whether the confidence interval is completely within the threshold range. If so, it indicates that the charging efficiency is normal or the aging degree is within the expected range. If not, send a signal for decision-making and alarm, and at the same time send the above data analysis information to the interactive interface and display it in the form of a chart.

3. The method for detecting automobile charging safety based on target environment data as claimed in claim 2, characterized in that: The specific implementation process of step S33 for calculating the confidence interval of the trend line and setting the threshold range of the battery charging efficiency and the battery aging degree is as follows: S331: Constructing a deep learning model for threshold confidence intervals; S332: generating feature label data by using feature engineering to calculate the confidence interval function of the trend line, to set the battery charging efficiency range function, and to set the threshold range function of the battery aging degree, wherein a part of the data is used as training data and a part of the data is used as test data; S333: inputting the training data into a deep learning model with a threshold confidence interval for training, and stopping the training when the model training meets a 90% prediction accuracy condition; S334: Deploy the trained deep learning model of the threshold confidence interval in the actual environment for real-time adjustment of the confidence interval and setting of the threshold range of the battery charging efficiency and the battery aging degree.

4. The method for detecting automobile charging safety based on target environment data according to claim 1, characterized in that: The specific implementation process of step S43 is to adjust the proportion in real time through deep learning model prediction analysis as follows: S431: Constructing a deep learning model of classification weights; S432: Labeling the classification variables corresponding to charging short circuit, leakage, and charging interruption through feature engineering to generate feature label data, part of which is used for training and part of which is used for testing data; S433: inputting the training data into the deep learning model of classification weights for training, and stopping the training when the model training meets the condition of 90% prediction accuracy; S434: Deploy the trained deep learning model of classification weights in an actual environment for real-time adjustment of the weights of charging short circuit, leakage, and charging interruption classifications.

5. The method for detecting automobile charging safety based on target environment data according to claim 1, characterized in that: The specific implementation process of step S44 is to adjust the weight of each binary variable in real time through deep learning model prediction analysis: S441: Constructing deep learning models for binary variables; S442: Labeling binary variables corresponding to charging short circuit, leakage, and charging interruption through feature engineering to generate feature label data, part of which is used for training and part of which is used for testing data; S443: inputting the training data into a deep learning model of binary variables for training, and stopping the training when the model training meets a 90% prediction accuracy condition; S444: Deploy the trained deep learning model of binary variables in an actual environment to adjust the weights of the binary variables of charging short circuit, leakage, and charging interruption in real time.

6. A system for detecting automobile charging safety based on target environment data, characterized in that: It includes a data acquisition module (1), a preprocessing module (2), a machine learning module (3), a decision module (4), an alarm module (5) and a human-computer interaction module (6); The data acquisition module (1) is used to collect real-time and historical environmental data, charging pile data, battery data and power grid data; The preprocessing module (2) is used to clean, standardize and normalize the collected data; The machine learning module (3) is used to establish a machine learning model and predict charging safety events, wherein the charging safety events include decreased battery charging efficiency, increased battery aging, charging short circuit, leakage, and charging interruption; The decision module (4) is used to make corresponding decisions according to the prediction results of the machine learning module; The alarm module (5) is used to issue an alarm according to the prediction result of the machine learning module; The human-computer interaction module (6) is used to display the data collection information and data analysis information in the form of charts and is used for the operation of the touch screen control system; The machine learning module (3) comprises a feature engineering submodule (31), a nonlinear regression analysis submodule (32), a time series analysis submodule (33), a confidence calculation submodule (34), and a threshold judgment submodule (35); The feature engineering submodule (31) is used to extract features from the historical data collected by the data collection module; The nonlinear regression analysis submodule (32) is used to perform nonlinear regression analysis model training on the data after feature extraction, predict the regression coefficient of each variable by gradient descent method, perform model fitting on battery charging efficiency and battery aging, and obtain a fitting curve; The time series analysis submodule (33) is used to perform trend analysis on the obtained fitting curve; The confidence calculation submodule (34) is used to generate a trend line according to the trend analysis result and calculate the confidence interval of the trend line; The threshold judgment submodule (35) is used to judge whether the confidence interval is completely within the threshold range of the battery charging efficiency and the battery aging degree, and send the judgment result; The machine learning module (3) further includes a K-means clustering module (36), a risk threshold judgment module (37), a categorical variable proportion statistics module (38), a binary variable proportion statistics module (39), and a random forest algorithm module (40); The K-means clustering module (36) collects continuous variables corresponding to charging short circuit, leakage, and charging interruption in history, and after feature processing, uses the K-means clustering algorithm to obtain the center point K of different types of continuous variables corresponding to each safety event; The risk threshold judgment module (37) is configured to receive and collect the continuous variable of this type in real time, and set the threshold A. According to the center point K and the threshold A, the range of the risk threshold B is calculated using the mean square error formula. The mean square error value of the historical data and the center point K is calculated using the mean square error formula. After the new data is input, the mean square error value after the new input number is calculated in real time, and the difference is made with the previous value. If the difference is within the range of the threshold A, the new input value falls within the risk threshold B. The classification variable proportion statistics module (38) is used to identify the variable type in the classification variable corresponding to each security event in the historical data, and to count the proportion of the variable type; The binary variable proportion statistics module (39) is used to identify the variable type in the binary variable corresponding to each safety event in the historical data, and to count the proportion of the variable type, i.e., "yes" or "no". Whether the battery shell is damaged or not is determined by a convolutional upscaling network. The random deep forest algorithm module (40) collects historical environmental data, charging pile data, battery data, and power grid data, generates different decision trees by randomly combining variables, uses the random deep forest algorithm to generate an integrated model for the above different decision trees, uses real-time collected data as input into the integrated model to predict the probability of occurrence of different types of security events, and issues warnings and decisions for probabilities that exceed thresholds.

Images