Disclosure of Invention
(one) solving the technical problems
Aiming at the defects of the prior art, the application provides a new energy automobile power supply safety detection method, which solves the problems that the prior art lacks of acquisition of a plurality of key parameters, and the importance of the characteristics cannot be accurately estimated by manually selecting the characteristics, so that the understanding of a model on the real situation is not accurate enough, and the prediction accuracy, generalization and stability are limited.
(II) technical scheme
In order to achieve the above object, the present application provides a new energy automobile power safety detection method, which is characterized in that the method comprises:
the method comprises the steps of data collection and data flow establishment, wherein the data collection and data flow establishment are used for collecting multidimensional information inside and outside an automobile, including speed, acceleration, battery temperature, current and voltage, and the collection device comprises an on-board sensor and a CAN bus and establishes real-time data flow;
feature screening, namely performing feature screening by adopting a random forest model, sorting by an importance index of the features, and selecting the features with the top ranking;
simulating a battery application scene, simulating the battery application scene through a model according to actual working conditions and driving habits, acquiring real-time working condition data and driving habit parameters including speed, acceleration and environmental temperature from a vehicle sensor, a navigation system and a driving recorder, preprocessing the collected working condition data and driving habit parameters, removing abnormal values, filling missing values, and establishing a simulation model to simulate the battery application scene based on the collected working condition data and driving habit parameters;
traditional machine learning prediction, performing traditional machine learning prediction by using an XGBoost model, and classifying and regression predicting the safety risk of the battery according to real-time data;
time series prediction, which is to predict the state and performance of the battery, including capacity fade and temperature change, using an RF model;
the integration improvement is to integrate the traditional machine learning prediction result and the time sequence prediction result;
result evaluation, namely evaluating according to actual prediction results;
model interpretation, the extent of contribution of the model to the predicted outcome is interpreted using SHAP values.
The application is further arranged to: the feature screening step specifically comprises the following steps:
step 1, collecting data required by battery safety detection, including battery parameters, environment information and battery states;
step 2, preprocessing the data, including missing value processing, outlier processing and data normalization;
step 3, dividing the data set into a training set and a testing set by adopting cross verification;
step 4, selecting a random forest class and a random forest class to define parameters of the model;
step 5, calculating importance indexes of the features through a trained random forest model;
step 6, sorting the features according to the feature importance indexes, selecting the features with the top ranking, manually determining a threshold according to the actual monitoring requirements of corresponding data, and reserving the features with the importance indexes higher than the threshold;
step 7, selecting the feature ranked at the front as a final feature set according to the ranking result for subsequent model training and security risk prediction;
the application is further arranged to: based on the collected working condition data and driving habit parameters, a simulation model is established in the simulated battery application scene to simulate the battery application scene, and the method specifically comprises the following steps:
a. collecting charge and discharge characteristic data of a battery, including charge and discharge efficiency and capacity fading;
establishing a physical model of the battery, including an electrochemical model and a heat conduction model;
the working condition data and the driving habit parameters are combined and input into a physical model for simulation;
b. collecting battery application data, including charge and discharge data and environmental data;
analyzing and modeling the collected data, and selecting statistical models such as Markov models and time sequence models;
according to the working condition data and the driving habit parameters, inputting the working condition data and the driving habit parameters into a statistical model for simulation;
c. selecting battery management system BMS simulation software MATLAB/Simulink, AMESim;
according to the actual working condition and driving habit parameters, configuring a battery model and an environment model in simulation software;
running simulation software to simulate application scenes of the battery under different conditions, including a charging and discharging process, temperature change and current change;
analyzing simulation results, and evaluating possible safety risks under different conditions;
the application is further arranged to: the specific implementation steps of classifying and regression predicting the safety risk of the battery by adopting the XGBoost model in the traditional machine learning prediction include:
preparing a dataset comprising features and labels, wherein the features are input variables for prediction, and the labels are output variables of corresponding battery safety risks;
preprocessing a data set, including missing value processing, outlier processing, data normalization and standardization;
dividing a data set into a training set and a testing set by adopting cross verification, wherein the training set is used for model training, and the testing set is used for evaluating model performance;
determining an XGBoost model and defining relevant parameters of the model;
training the XGBoost model by using a training set;
predicting the test set and the real-time data by using a trained XGBoost model;
the application is further arranged to: in the time sequence prediction step, the specific implementation steps of predicting the state and the performance of the battery by using the RF model are as follows:
preparing a dataset comprising time series features and corresponding status and performance tags;
preprocessing a data set, including missing value processing, outlier processing, data normalization and standardization;
dividing the time sequence data set into a training set and a testing set by adopting cross validation;
determining a random forest model and defining model parameters;
training the random forest model by using a training set;
performing time sequence prediction on the test set and the real-time data by using a trained random forest model;
the application is further arranged to: in the integrated improvement step, weights are distributed to the traditional machine learning prediction result and the time sequence prediction result and are used for weighting and combining the prediction results;
linearly combining the traditional machine learning prediction result and the time sequence prediction result by using weights to obtain a final integrated prediction result;
performing model evaluation by using the integrated prediction result;
the application is further arranged to: in the result evaluation step, an evaluation index is determined according to the properties and requirements of the task, wherein the evaluation index comprises the area under an AUC value curve, the accuracy, the recall rate and the F1 fraction;
calculating the selected evaluation index by using the real label and the predicted label;
the application is further arranged to: in the result evaluation step, an evaluation index is determined according to the properties and requirements of the task, wherein the evaluation index comprises area under an AUC (automatic curve) value curve, accuracy, recall rate and F1 score, and the specific accuracy, recall rate and F1 score are calculated:
the accuracy rate calculation formula:
where TP is True examples (True posives) and FP is False positive examples (False posives).
The recall rate calculation formula:
where TP is True examples (True posives) and FN is False negative examples (False negative).
F1 Score (F1 Score) calculation formula: f1 Score=2 x (Precision x Recall)/(precision+recall).
(III) beneficial effects
The application provides a new energy automobile power supply safety detection method. The beneficial effects are as follows:
the method has the advantages that the inside and outside multidimensional information of the automobile is collected, the inside and outside multidimensional information comprises the speed, the acceleration, the battery temperature, the current and the voltage, and the real-time data flow is established through the vehicle-mounted sensor and the CAN bus, so that richer data are obtained, the detection process is more real-time and dynamic, in the aspect of feature screening, a random forest model is adopted for feature screening, the features ranked forward are ordered through importance indexes of the features, and compared with a simple feature selection method in the traditional method, the random forest model CAN evaluate the importance of the features more accurately, and the feature selection effect is improved.
In addition, in the aspect of simulating a battery application scene, the battery application scene is simulated through a model, real-time working condition data and driving habit parameters are obtained by utilizing a vehicle sensor, a navigation system and a driving recorder data source, compared with the traditional method which is only based on analysis of static data, the method can better simulate the real battery service condition, adopts an XGBoost model to conduct traditional machine learning prediction, uses an RF model to conduct time sequence prediction, can more accurately predict the safety risk, state and performance of a battery, improves the accuracy and precision of prediction, integrates the traditional machine learning prediction result and the time sequence prediction result, further improves the prediction precision and stability, combines the prediction results of different models, and provides more reliable prediction results.
In the aspect of result evaluation, evaluation is carried out according to an actual prediction result, the contribution degree of a model to the prediction result is interpreted by using a SHAP value method, the prediction accuracy rating is concerned, meanwhile, the interpretation and interpretation of the model prediction result are provided, and the interpretability and the trust degree of the model are enhanced.
The method solves the problems that in the prior art, the acquisition of a plurality of key parameters is lacking, the importance of the characteristics cannot be accurately evaluated by manually selecting the characteristics, so that the understanding of the model on the real situation is not accurate enough, and the prediction accuracy, generalization and stability are limited.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Example 1
Referring to fig. 1, the application provides a new energy automobile power supply safety detection method, which comprises the following steps:
s1, data collection and data flow establishment, wherein the data collection and data flow establishment are used for collecting multi-dimensional information inside and outside an automobile, including speed, acceleration, battery temperature, current and voltage, and the collection device comprises an on-board sensor and a CAN bus and establishes real-time data flow; ensuring timely data transmission and processing
S2, feature screening, namely performing feature screening by adopting a random forest model, sorting by importance indexes of the features, and selecting the features with the top ranking
The method specifically comprises the following steps:
collecting data required for battery safety detection, including battery parameters, environmental information and battery status;
preprocessing data, including missing value processing, outlier processing and data normalization;
dividing the data set into a training set and a testing set by adopting cross validation; the training set is used for feature screening and model training, and the testing set is used for evaluating model performance;
selecting a RandomastClassifier and a RandomfortrRaegressor, and defining parameters of a model;
calculating importance indexes of the features through a trained random forest model; the implementation is based on average reduction of the non-purity (Mean Decrease Impurity), in particular as follows:
python
import numpy as np
import matplotlib.pyplot as plt
training random forest model
rf＝RandomForestClassifier(n_estimators＝100,max_depth＝10)
rf.fit(X_train,y_train)
Importance of # acquisition feature
importances＝rf.feature_importances_
Importance of # visual features
indices＝np.argsort(importances)[::-1]
plt.figure()
plt.title("Feature Importances")
plt.bar(range(X.shape[1]),importances[indices],color＝"r",align＝"center")
plt.xticks(range(X.shape[1]),feature_names[indices],rotation＝90)
plt.xlim([-1,X.shape[1]])
plt.tight_layout()
plt.show()
Sorting the features according to the feature importance indexes, selecting the features with the top ranking, manually determining a threshold according to the actual monitoring requirements of corresponding data, and reserving the features with the importance indexes higher than the threshold;
the method comprises the following steps:
python
setting feature importance threshold #
threshold＝0.01
# Filter characteristics
selected_features＝[feature for feature,importance in zip(feature_names,importances)if importance>threshold]
Selecting the feature with the top ranking as a final feature set according to the ranking result for subsequent model training and security risk prediction;
the specific formulas and parameters in the feature screening step are as follows:
feature importance, indicating the extent to which each feature contributes to model prediction;
-x_train: training set feature data;
-y_train: training set tag data;
-number of features of shape [1 ];
feature_names;
s3, simulating a battery application scene, simulating the battery application scene through a model according to actual working conditions and driving habits, acquiring real-time working condition data and driving habit parameters including speed, acceleration and environmental temperature from a vehicle sensor, a navigation system and a driving recorder, preprocessing the collected working condition data and driving habit parameters, removing abnormal values, filling missing values, and establishing a simulation model to simulate the battery application scene based on the collected working condition data and driving habit parameters;
based on the collected working condition data and driving habit parameters, a simulation model is established to simulate the application scene of the battery, and the method specifically comprises the following steps:
a. collecting charge and discharge characteristic data of a battery, including charge and discharge efficiency and capacity fading;
establishing a physical model of the battery, including an electrochemical model and a heat conduction model;
the working condition data and the driving habit parameters are combined and input into a physical model for simulation;
application scene of simulated battery under different conditions
b. Collecting battery application data, including charge and discharge data and environmental data;
analyzing and modeling the collected data, and selecting statistical models such as Markov models and time sequence models;
according to the working condition data and the driving habit parameters, inputting the working condition data and the driving habit parameters into a statistical model for simulation;
simulating application scenes of the battery under different conditions;
c. selecting battery management system BMS simulation software MATLAB/Simulink, AMESim;
according to the actual working condition and driving habit parameters, configuring a battery model and an environment model in simulation software;
running simulation software to simulate application scenes of the battery under different conditions, including a charging and discharging process, temperature change and current change;
analyzing simulation results to evaluate security risks possibly existing under different conditions
S4, traditional machine learning prediction is carried out, XGBoost model is used for carrying out traditional machine learning prediction, and safety risks of the battery are classified and regression prediction is carried out according to real-time data
The specific implementation steps of classifying and regression predicting the safety risk of the battery by adopting the XGBoost model are as follows:
preparing a dataset comprising features and labels, wherein the features are input variables for prediction, and the labels are output variables of corresponding battery safety risks;
preprocessing a data set, including missing value processing, outlier processing, data normalization and standardization;
dividing a data set into a training set and a testing set by adopting cross verification, wherein the training set is used for model training, and the testing set is used for evaluating model performance;
determining an XGBoost model and defining relevant parameters of the model;
the specific model definition steps are as follows:
python
import xgboost as xgb
# definition XGBoost classification model
xgb_model＝xgb.XGBClassifier(learning_rate＝0.1,n_estimators＝100,max_depth＝3)
# definition XGBoost regression model
xgb_model＝xgb.XGBRegressor(learning_rate＝0.1,n_estimators＝100,max_depth＝3)
Wherein the method comprises the steps of
-learning_rate: the learning rate is used for controlling the weight updating speed of each tree;
-n_evators: the number of trees determines the number of trees in the XGBoost model;
-max_depth: the depth of the tree, the complexity and fitting capacity of the tree are controlled;
training the XGBoost model by using a training set;
python
xgb_model.fit(X_train,y_train)
wherein:
-x_train: feature data of the training set;
y_train: tag data of the training set;
predicting the test set and the real-time data by using a trained XGBoost model;
python
prediction＝xgb_model.predict(X_test)
wherein:
-x_test: characteristic data of the test set and the real-time data;
-prediction: a prediction result representing a prediction of a safety risk of the battery;
s5, time sequence prediction, namely, predicting the state and the performance of the battery by using an RF model to perform time sequence prediction, wherein the time sequence prediction comprises capacity fading and temperature change
In the time series prediction step, the specific implementation steps of predicting the state and performance of the battery using the RF model are as follows:
preparing a dataset comprising time series features and corresponding status and performance tags;
preprocessing a data set, including missing value processing, outlier processing, data normalization and standardization;
dividing the time sequence data set into a training set and a testing set by adopting cross validation;
determining a random forest model and defining model parameters;
specific:
python
from sklearn.ensemble import RandomForestRegressor
# definition random forest model
rf_model＝RandomForestRegressor(n_estimators＝100,max_depth＝3)
Wherein:
-n_evators: the number of decision trees determines the number of trees in the random forest model;
-max_depth: the depth of the tree, the complexity and fitting capacity of the tree are controlled;
training the random forest model by using a training set;
python
rf_model.fit(X_train,y_train)
wherein:
-x_train: feature data of the training set;
y_train: the label data of the training set, namely the state and the performance of the battery;
performing time sequence prediction on the test set and the real-time data by using a trained random forest model;
python
prediction＝rf_model.predict(X_test)
wherein:
-x_test: characteristic data of the test set and the real-time data;
-prediction: a prediction result, which is a predicted value indicating the state and performance of the battery;
s6, integration improvement: integrating traditional machine learning prediction results and time sequence prediction results
Weight is distributed to the traditional machine learning prediction result and the time sequence prediction result and used for weighting and combining the prediction results; the weight is determined through experimental verification;
linearly combining the traditional machine learning prediction result and the time sequence prediction result by using weights to obtain a final integrated prediction result;
specific:
python
alpha=0.7# integration weight value, determined from experimentation and validation
ensemble_prediction＝alpha*prediction_XGBoost+(1-alpha)*prediction_RF
Wherein:
-alpha: integrating weights to represent the relative importance of the traditional machine learning prediction result and the time sequence prediction result;
performing model evaluation by using the integrated prediction result; using evaluation indexes including accuracy and Root Mean Square Error (RMSE), and adjusting and optimizing according to evaluation result to further improve performance of integrated model
S7, evaluating results according to actual prediction results
Determining evaluation indexes including area under an AUC value curve, accuracy, recall rate and F1 fraction according to the properties and requirements of the task;
calculating the selected evaluation index by using the real label and the predicted label;
calculation of specific AUC values:
python
from sklearn.metrics import roc_auc_score
auc＝roc_auc_score(true_labels,predicted_labels)
specific accuracy, recall, F1 score calculation:
the accuracy rate calculation formula:
where TP is True examples (True posives) and FP is False positive examples (False posives).
The recall rate calculation formula:
where TP is True examples (True posives) and FN is False negative examples (False negative).
F1 Score (F1 Score) calculation formula: f1 Score=2 x (Precision x Recall)/(precision+recall);
specific:
python
from sklearn.metrics import precision_score,recall_score,f1_score
precision＝precision_score(true_labels,predicted_labels)
recall＝recall_score(true_labels,predicted_labels)
f1＝f1_score(true_labels,predicted_labels)
the evaluation result reflects the prediction accuracy, recall rate and prediction capability of the model;
s8, model interpretation, namely using SHAP values to interpret the contribution degree of the model to the prediction result; improving the interpretability and interpretability of a model
Calculating SHAP values of the features, and evaluating the contribution degree of each feature to the prediction result;
specific:
python
import shap
calculation of SHAP value #
explainer＝shap.Explainer(model)
shap_values＝explainer.shap_values(data)
Wherein:
-model: a traditional machine learning model, a time series model;
-data: test set and actual application data;
using the contribution degree of SHAP value visual characteristics to explain the influence of the model on the prediction result;
specific:
python
shap.summary_plot(shap_values,data)
generating a feature importance graph to display the influence of the SHAP value of each feature on the prediction result;
interpreting the prediction result of the single sample using a SHAP value single sample interpretation method;
specific:
python
shap.force_plot(explainer.expected_value,shap_values[index],data[index])
a force directed graph is generated to show the degree of contribution of a feature to a single sample.
Example 2
S1, data collection and data flow establishment, wherein the data collection and data flow establishment are used for collecting multi-dimensional information inside and outside an automobile, including speed, acceleration, battery temperature, current and voltage, and the collection device comprises an on-board sensor and a CAN bus and establishes real-time data flow; ensuring timely data transmission and processing
S2, performing State of Charge (SOC) evaluation, and evaluating the energy storage condition of the battery by monitoring the Charge and discharge states of the battery; monitoring the temperature change of the battery, and ensuring that the battery works in a proper temperature range; battery short-circuit protection: detecting whether a short circuit condition exists in a battery circuit, and timely taking measures to prevent overheating, burning or explosion of the battery; and (3) battery balance control: each single battery in the battery pack is charged and discharged in an equalizing manner, so that the charge states of the batteries are relatively consistent, and potential safety hazards caused by unbalance among the batteries in the battery pack are reduced; and (3) diagnosing battery faults: by monitoring the working parameters and the state of the battery, faults in the battery system are timely identified and eliminated, and the normal operation of the vehicle is ensured.
The power supplies of 10 groups of new energy automobiles were tested as in example 1 and example 2, respectively, and were set as experiment 1 and experiment 2, respectively.
Table 1: power supply safety detection parameter
| Group of | Data accuracy/% | Potential risk early warning rate/% | Risk early warning accuracy/% |
| Experiment 1 group | 99.5 | 98.8 | 98.7 |
| Experiment 2 group | 95.5 | 46.5 | 15.3 |
In the present application, the above is combined with the above matters:
in the aspect of data collection and data stream establishment, multi-dimensional information inside and outside an automobile is collected, real-time data streams including speed, acceleration, battery temperature, current and voltage are established through a vehicle-mounted sensor and a CAN bus, richer data are obtained, the detection process is more real-time and dynamic, in the aspect of feature screening, a random forest model is adopted for feature screening, ranking is carried out through importance indexes of features, features with top ranking are selected, and compared with a simple feature selection method in a traditional method, the random forest model CAN evaluate the importance of features more accurately, and the effect of feature selection is improved.
In addition, in the aspect of simulating a battery application scene, the battery application scene is simulated through a model, real-time working condition data and driving habit parameters are obtained by utilizing a vehicle sensor, a navigation system and a driving recorder data source, compared with the traditional method which is only based on analysis of static data, the method can better simulate the real battery service condition, adopts an XGBoost model to conduct traditional machine learning prediction, uses an RF model to conduct time sequence prediction, can more accurately predict the safety risk, state and performance of a battery, improves the accuracy and precision of prediction, integrates the traditional machine learning prediction result and the time sequence prediction result, further improves the prediction precision and stability, combines the prediction results of different models, and provides more reliable prediction results.
In the aspect of result evaluation, evaluation is carried out according to an actual prediction result, the contribution degree of a model to the prediction result is interpreted by using a SHAP value method, the prediction accuracy rating is concerned, meanwhile, the interpretation and interpretation of the model prediction result are provided, and the interpretability and the trust degree of the model are enhanced.
The power supply safety detection method provided by the application is used for carrying out safety detection by extracting multi-dimensional information inside and outside an automobile, including vehicle sensor data, battery system operation data and environment data, so that the detection is more comprehensive and accurate, meanwhile, a random forest and SHAP combined model is used for screening the characteristics, a machine learning model is used for screening the characteristics, the characteristics related to safety risks are more accurately selected, the prediction precision is improved by combining a traditional machine learning model and a time sequence prediction model in an integrated and improved mode, the advantages and diversity among different models are fully utilized, the method is more suitable for scenes of real-time monitoring and prediction, real-time data streams are received, and the characteristics are extracted, predicted and decision support is carried out in real time.
The above embodiments may be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented in software, the above-described embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the processes or functions described in accordance with embodiments of the present application are produced in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains one or more sets of available media. The usable medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium may be a solid state disk.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, and are not repeated herein.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other forms.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Finally: the foregoing description of the preferred embodiments of the application is not intended to limit the application, but to enable any modification, equivalent or improvement to be made without departing from the spirit and principles of the application.