A method for predicting the parking time of a vehicle with a battery sensor

Abstract

The present invention discloses a method for predicting the parking time of a car with a battery sensor, and the main steps are: 1. Extracting a data set of corresponding quantity and dimension from a car networking big data platform, and performing data preprocessing and feature engineering; 2. According to the battery-related data and constructed feature variables as training sets and test sets, respectively selecting appropriate machine learning algorithms for model training, and selecting the model with the best model evaluation index to output the prediction result of the parking time; 3. Complete the data set establishment and model training of the user's next car use time prediction model, and judge the risk level of power shortage. The method of the present invention can more accurately predict the timing of starting the vehicle to charge the battery, and provide more accurate guidance to the car owner, so that the car owner can avoid the problem of being unable to start the engine and inconvenience in use due to battery reasons, thereby improving the user experience.

Description

Technical Field

[0001] The invention relates to the field of after-sales vehicle quality monitoring, and in particular to a method for predicting the parking time of a car with a battery sensor. Background Art

[0002] After the vehicle is parked and turned off, the car battery provides the power consumption requirements of all controllers and loads in the vehicle. Even when the vehicle enters a dormant state, that is, each controller enters a low-power mode, there will still be power consumption. Generally, the car battery is designed to provide low-power parking time of about 30 days. However, if the user performs some improper operations, or installs other power-consuming equipment, or parks the vehicle for a long time without using it, or the vehicle is disturbed by the environment during parking, resulting in abnormal discharge, such as interference from radio frequency signals, or other situations, the car battery will be over-consumed until it is lower than the vehicle's starting power, making it impossible for the user to start the vehicle, resulting in extreme complaints from the user and the inability to use the vehicle. Therefore, it is necessary to timely predict the current parking time and the risk of the next power shortage after the vehicle is parked and turned off, so as to ensure that the vehicle can be started normally after parking.

[0003] In the prior art, for example, Chinese patent document CN 112763920 discloses a battery detection method, device and computer storage medium, which predicts the current battery remaining charge state through the battery remaining charge state, battery remaining life state, battery voltage, battery current, battery temperature and the vehicle's parking time, and then uses a machine learning regression model to predict the current battery remaining charge state, and then compares the remaining charge with an empirical threshold to determine whether there is a risk of low power. This method predicts the risk of low battery to a certain extent, but there are still the following problems: First, it does not take into account the abnormal discharge of the vehicle during parking, which will accelerate power consumption and affect the predicted value; second, it uses the parking time as input to predict the current remaining charge, which lacks timeliness. It is impossible to predict how long the parking can last after the engine is turned off, and whether there is a risk of low battery. If the user is far away from the vehicle when it is predicted that there is a risk of low battery, it may still cause low battery; third, the judgment of low battery risk is not based solely on the current remaining power status, and a comprehensive judgment should be made in combination with the prediction of the user's next use time. For example, it is predicted that the battery will be low in 24 hours, but if it is predicted that the user will use the car within 12 hours, then the current risk should be low, otherwise it will be high risk, which is more accurate.

[0004] In another Chinese patent publication, such as Chinese patent document CN 111896883, a method for predicting the parking time supported by a vehicle battery and an early warning method are disclosed. Compared with the above patent method, it goes a step further by using the historical data of the vehicle battery and parking time as a training set, and with the help of a machine learning algorithm, the current parking time can be predicted after this parking. However, this method still has the following problems: First, it still does not consider the abnormal discharge of the vehicle during parking; second, the determination of the risk of power shortage is not combined with the prediction of the user's next car use time for comprehensive determination. Combining the methods of the above two patents, in the model training link, only the training model of non-time series data is used, and the characteristics of the Internet of Vehicles data as time series data are not considered. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a method for predicting the parking time of a car with a battery sensor. The method is a method for predicting the parking time of a vehicle and the risk of power shortage based on Internet of Vehicles data through a machine learning algorithm, so as to timely predict the current parking time and the risk of power shortage next time after the vehicle is parked and turned off, so as to ensure that the vehicle can be started normally after parking.

[0006] The technical solution of the present invention is achieved in this way:

[0007] A method for predicting the parking time of a vehicle with a battery sensor, characterized by comprising the following steps:

[0008] 1) Extracting data sets of corresponding quantity and dimension from the Internet of Vehicles big data platform and performing data preprocessing and feature engineering, including the following steps:

[0009] A. Data extraction: Extract a certain number of N vehicles from the Internet of Vehicles Big Data Platform, and the battery-related data for a continuous period of time. Automatically calculate the low-battery tags based on the uploaded data. If the calculated number of low-battery vehicles is less than the set number M, re-extract the data of N vehicles until the number of low-battery tags is greater than the set number M.

[0010] B. Data preprocessing: Process the data extracted in the data extraction phase for outliers;

[0011] C. Constructing the characteristic variable of battery life: The battery life state of the vehicle is obtained by using the extracted voltage value of the vehicle at the time of starting through the empirical threshold value;

[0012] D. Construct a characteristic variable of abnormal power consumption of a sample: obtain the sleep-to-rest time point of the vehicle's internal network through the network heartbeat flag data extracted from the data in step A, and determine whether the wake-up is actively woken up by the user based on the user operation data at that moment. If it is not actively woken up by the user, it is an abnormal wake-up. At the same time, calculate the number of abnormal wake-ups and the cumulative duration of abnormal wake-ups between each shutdown of the vehicle and the next start-up or the time when the low-power label appears;

[0013] E. Construct the residual charge characteristic variable at the beginning of a sample: determine the battery SOC value when the vehicle is turned off by using the ignition off data;

[0014] F. Construct the residual charge characteristic variable at the end of a sample: the battery SOC value is divided into two cases: the battery low charge label appears or the battery low charge label does not appear;

[0015] G. Construct a parking time variable for a sample: Calculate the time from the time the engine is turned off to the time the user powers on or the time when the power-off tag appears through relevant data;

[0016] 2) Based on the battery-related data and the constructed characteristic variables as the training set and the test set, respectively select appropriate machine learning algorithms for model training, and select the model with the best model evaluation index to output the prediction result of the parking time, including the following steps:

[0017] A. Prepare and split the model training data set: Use the preprocessed battery sensor data to construct the feature variables as the feature variables for training the parking time prediction model, use the constructed parking time variable of a single sample as the prediction label of the prediction model, and divide the entire data set into a training set and a test set according to a certain ratio;

[0018] B. Data normalization: Map the feature variables and predicted label data of the data set to the interval [0,1];

[0019] C. Model training: Select a variety of regression algorithms for model training, and select the optimal parameters of each model through cross-validation;

[0020] D. Model selection: Input the test set into each model, obtain the mean square error of the prediction results of each model, and select the model with the smallest mean square error as the final prediction model;

[0021] E. Obtaining the parking time prediction result: normalizing and restoring the prediction value output by the selected prediction model;

[0022] 3) Complete the establishment of the data set and model training of the user's next car usage time prediction model, and determine the level of battery shortage risk. Compare the parking time prediction result with the user's next car usage time prediction result to obtain the level of battery shortage risk.

[0023] In this way, the prediction method of the present invention can more accurately predict the timing of starting the vehicle to charge the battery, and provide more accurate guidance to the car owner, so that the car owner can avoid the inability to start the engine due to battery problems, thereby improving the user experience. In addition, the present invention uses the historical data of the battery to obtain the longest parking time that the battery can support. This data can be used as one of the indicators of the health status of the battery, providing a reference for battery design for engineers in the battery field.

[0024] Further: the battery-related data includes battery sensor data, user vehicle-related data and vehicle network heartbeat data, mainly engine operation status EngineStatus, power status feedback PowerStatusFeedback, lock signal LockSignal, driver's door lock status DriverDoorLockStatus, anti-theft status KeyAlarmStatus, unlock signal UnlockSignal, battery sensor SOC value, battery sensor voltage signal U_BATT, vehicle outside temperature signal EnvironmentalTemp, vehicle network sleep flag Sleepflag, vehicle network wake-up flag Awakeflag, each data has a timestamp. The present invention obtains these data, which are the basic data of vehicle operation, and obtaining these data lays the foundation for subsequent calculation, judgment and prediction.

[0025] Further: when the vehicle network sleep flag Sleepflag changes from 0 to 1, and the vehicle network wake-up flag Awakeflag changes from 0 to 1, it is determined that a wake-up occurs at this moment, and whether there is user active wake-up data within 2 minutes before and after the wake-up moment. Active wake-up data includes changes in the unlock signal UnlockSignal, the lock signal LockSignal, the driver's side door lock status DriverDoorLockStatus, and the anti-theft status KeyAlarmStatus from the armed state to the disarmed state. If no active wake-up data is found, the current wake-up is determined to be an abnormal wake-up. In this way, the active wake-up or abnormal wake-up state of the vehicle can be accurately distinguished, which is convenient for collecting and calculating the abnormal wake-up state.

[0026] Furthermore: the characteristic variables of the training parking time prediction model include the battery sensor SOC value, the battery sensor voltage signal U_BATT, the outdoor temperature signal EnvironmentalTemp, the number of abnormal wake-up times of a sample, the continuous abnormal wake-up time of a sample, the initial SOC value of a sample, the SOC value at the end of a sample, and the battery life status. The constructed parking time variable of a sample is used as the prediction label of the prediction model, and the data set is divided into a training set and a test set according to the ratio of 8:2.

[0027] Further: In the data normalization step, the mapping formula is Among them, max is the maximum value of the sample data, and min is the minimum value of the sample data.

[0028] Further: In the model training step, XGBOOST, GBDT, and SVR algorithms are selected for model training, and the optimal parameters of each model are selected through five-fold cross validation; the training set is further divided into five parts, four of which are used as training sets, and the remaining one is used as a test set for model training and evaluation, and the model parameters that minimize the mean square error of the model are selected. After the four algorithms of XGBOOST, GBDT, and SVR, and after five-fold cross validation, the model parameters that minimize the mean square error of the model are selected more accurately, and the better model parameters are obtained.

[0029] Further: The normalized reduction formula is x * =x*(max-min)+min. The normalized reduction formula selected by the present invention is determined to facilitate system execution.

[0030] Further: the parking time of a bicycle every day is the time when the user powers on minus the previous shutdown time or the time when the power-down tag appears minus the previous shutdown time. The average parking time of a day is the sum of the parking times whose start and end times are both within the same day divided by the number of parking times that meet this condition. The characteristic variable of a sample is an array composed of the average parking time of 7 consecutive days, and the predicted tag is the average parking time of the 8th day. In this way, the calculation method of parameters such as the parking time of a bicycle every day and the average parking time of a day is determined, which is convenient for system execution.

[0031] Further: the determination of the power shortage risk level comprises the following steps:

[0032] A. Dataset establishment: Calculate the average parking time of a bicycle every day. The feature variable of a sample is an array consisting of the average parking time of S consecutive days, and the predicted label is the average parking time of the S+1th day;

[0033] B. Model training and prediction: Use time series algorithm to train the model and adjust the parameters so that the error of the prediction value is less than 1 hour, and output the prediction model;

[0034] C. Determination of the risk level of battery failure: The risk level of battery failure is determined by comparing the parking time prediction result with the next time the user uses the car.

[0035] Further, the magnitude of the power-loss risk is as follows: high risk means that the vehicle's parking time is less than the user's maximum possible remaining vehicle time, or the vehicle's parking time is greater than the user's maximum possible remaining vehicle time, but less than 1 day; medium risk means that the vehicle's parking time is greater than the user's maximum possible remaining vehicle time, but less than 3 days; low risk means that the vehicle's parking time is greater than the user's maximum possible remaining vehicle time, but less than 1 week; no risk means that the vehicle's parking time is greater than 1 week. In this way, four levels of power-loss risk are determined, which is convenient for system execution.

[0036] Further: the battery-related data for continuous time is for one consecutive month.

[0037] Further: M is an integer number ranging from 3% to 8% of N.

[0038] In summary, the present invention has the following beneficial effects:

[0039] 1. The method of the present invention can more accurately predict the timing of starting the vehicle to charge the battery, and provide more accurate guidance to the car owner, so that the car owner can avoid the inability to start the engine due to battery problems, thereby improving the user experience.

[0040] 2. The present invention uses the historical data of the battery to obtain the longest parking time that the battery can support. This data can be used as one of the indicators of the health status of the battery and provide a battery design reference for engineers in the battery field.

[0041] 3. The present invention utilizes the historical data of batteries to obtain a portrait of the longest parking time that the battery of each vehicle can support, thereby providing favorable conditions and technical support for the OEM's operating customers; utilizes the historical data of batteries to obtain a portrait of the longest parking time that the battery of each vehicle can support, thereby providing a parking time dimension for the OEM to create battery features, enriching the data dimension of the OEM's vehicle parts, and providing data support for the subsequent digitization of the vehicle.

[0042] 4. The present invention uses the critical state characteristics of the battery for starting the engine to make the prediction more accurate and more reliable; it uses the historical data of the battery to promptly remind the user to charge the battery (start the vehicle), which to a large extent avoids the occurrence of battery depletion and the inconvenience of being unable to start the engine due to battery depletion, cultivates users' good car-using habits, and extends the service life of the battery and the entire vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Parking duration prediction model dataset preparation flow chart;

[0044] Figure 2 Parking duration prediction model training and prediction flow chart;

[0045] Figure 3 Power shortage risk level output flow chart. DETAILED DESCRIPTION

[0046] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0047] See also Figure 1 — Figure 3 The present invention provides a method for predicting the parking time of a car with a battery sensor, and the steps are as follows:

[0048] The method for predicting the parking time of a vehicle with a battery sensor according to the present invention comprises the following steps:

[0049] 1. Extract the corresponding number and dimension of data sets from the Internet of Vehicles big data platform, and perform data preprocessing and feature engineering, see Figure 1 , including the following steps:

[0050] A. Data extraction: Extract N vehicles (N is 1000 or more) from the Internet of Vehicles big data platform, battery sensor data for a continuous period of time (such as one month), user vehicle-related data, and vehicle network heartbeat data, including engine running status EngineStatus, power status feedback PowerStatusFeedback, lock signal LockSignal, driver's side door lock status DriverDoorLockStatus, anti-theft status KeyAlarmStatus, unlock signal UnlockSignal, battery sensor SOC value, battery sensor voltage signal U_BATT, outdoor temperature signal EnvironmentalTemp, vehicle network sleep flag Sleepflag, vehicle network wake-up flag Awakeflag, and each data is timestamped. The power-loss tag is automatically calculated based on the uploaded data. The calculation rules of the power-loss tag are obtained through statistical analysis. The voltage values ​​of 1,000 normal vehicles are extracted and a box diagram of the voltage value is drawn. The power-loss threshold is set to be lower than the lower limit of the voltage box diagram. It is currently set to 6.1V. This value needs to be set based on the data extracted from different vehicle models. It is not a fixed value. If the number of vehicles with power-loss tags is less than the set number M (such as 50 or 60 vehicles), the data of 1,000 vehicles will be re-extracted until the number of power-loss tags is greater than the set number M (such as 50 or 60 vehicles). M can be set to an integer number of 3%-8% of N, and N and M are both integers.

[0051] B. Data preprocessing: The data extracted in the data extraction stage are processed for outliers. The first principle of outlier processing is to delete samples with invalid values. The second principle is to fill missing values ​​with the average value.

[0052] C. Construct battery life characteristic variables: The battery life status of the vehicle is obtained by extracting the voltage value of the vehicle at the start-up time through the empirical threshold value. The judgment condition at the start-up time is that the engine running state EngineStatus changes from the stop state to the running state. According to engineering experience, the normal value of the battery voltage at the moment of car start-up should be greater than 9V. It is now set that if the battery voltage is greater than 9V at the moment of start-up, it is healthy, 8-9V is sub-healthy, and below 8V is recommended for replacement.

[0053] D. Construct a characteristic variable of abnormal power consumption for a sample: The abnormal power consumption characteristic consists of the number of abnormal wake-ups and the duration of abnormal wake-ups. When the vehicle network sleep flag Sleepflag changes from 0 to 1, and the vehicle network wake-up flag Awakeflag changes from 0 to 1, it is determined that a wake-up occurs at this moment. Within 2 minutes before and after the wake-up moment, it is determined whether there is user active wake-up data. The active wake-up data includes changes in the unlocking signal UnlockSignal, the locking signal LockSignal, the driver's side door lock status DriverDoorLockStatus, and the anti-theft status KeyAlarmStatus from the armed state to the disarmed state. If no active wake-up data is found, the current wake-up is determined to be an abnormal wake-up. At the same time, the number of abnormal wake-ups and the cumulative abnormal wake-up time length are calculated between each vehicle shutdown and the next startup or the time when the low-power label appears. The shutdown judgment condition is that the engine running state EngineStatus changes from the running state to the stopped state, and the power status feedback PowerStatusFeedback changes from the ON gear to the OFF gear.

[0054] E. Construct the charge characteristic variable at the initial stage of a sample: judge the battery SOC value when the vehicle is turned off by the ignition-off data. The battery SOC value is the state of charge, which reflects the remaining capacity of the battery and is expressed as the percentage of the remaining capacity to the battery capacity.

[0055] F. Construct the residual charge characteristic variable at the end of a sample: it is divided into two situations: the low-power label appears or the low-power label does not appear. If the low-power label does not appear, it is the battery SOC value when the user powers on. The power status feedback PowerStatusFeedback changes from OFF to ON as the judgment condition for powering on. If the low-power label appears, it is the SOC value when the label appears.

[0056] G. Construct a parking time variable for a sample: If there is no low-battery tag, the parking time is the time when the user powers on minus the previous shutdown time. If there is a low-battery tag, the parking time is the time when the low-battery tag appears minus the previous shutdown time.

[0057] Secondly, based on the battery-related data and the constructed characteristic variables (here refers to the characteristic variables constructed in steps C, D, E, F, and G of the above steps) as the training set and test set, select appropriate machine learning algorithms for model training, and select the model with the best model evaluation index to output the prediction result of the parking time, see Figure 2 , including the following steps:

[0058] A. Prepare and split the model training data set: Use the preprocessed battery sensor related data to construct feature variables (here refers to the feature variables constructed in steps C, D, E, F, and G of the above steps) as feature variables for training the parking time prediction model, including the battery sensor SOC value, the battery sensor voltage signal U_BATT, the outdoor temperature signal EnvironmentalTemp, the number of abnormal wake-up times of a sample, the continuous abnormal wake-up time of a sample, the initial SOC value of a sample, the SOC value at the end of a sample, and the battery life status. The constructed parking time variable of a sample is used as the prediction label of the prediction model, and the data set is divided into a training set and a test set in a ratio of 8:2.

[0059] B. Data normalization: Map the feature variables and predicted label data of the data set to the interval [0,1]. The mapping formula is: Among them, max is the maximum value of the sample data, and min is the minimum value of the sample data.

[0060] C. Model training: XGBOOST, GBDT, and SVR algorithms were selected for model training respectively, and the optimal parameters of each model were selected through 5-fold cross validation. That is, the training set was divided into 5 parts of data, 4 of which were used as training sets, and the remaining 1 part of the data was used as the test set for model training and evaluation, and the model parameters that minimized the mean square error of the model were selected.

[0061] D. Model selection: Input the test set into each model with the optimal parameters, obtain the mean square error of the prediction results of each model, and select the model with the smallest mean square error as the final prediction model.

[0062] E. Obtain the parking time prediction result: Normalize the prediction value output by the selected prediction model, and the formula is x * =x*(max-min)+min.

[0063] Finally, the data set and model training of the prediction model for the next time the user uses the car are completed, and the risk level of power shortage is determined. The level of power shortage risk is obtained by comparing the parking time prediction result with the prediction result of the next time the user uses the car. Figure 3 , specifically including the following steps:

[0064] A. Dataset establishment: Calculate the average parking time of a bicycle every day. The parking time is the time when the user powers on minus the previous shutdown time or the time when the low-power label appears minus the previous shutdown time. The average parking time of a day is the sum of the parking time when the start and end times of the parking time calculation are both within the same day divided by the number of parking times that meet this condition. The characteristic variable S of a sample is 7 days, which is an array composed of the average parking time of 7 consecutive days. The predicted label is the average parking time on the 8th day.

[0065] B. Model training and prediction: The time series algorithm LSTM is used for model training and parameter adjustment so that the error of the prediction value is less than 1 hour, and the prediction model is output.

[0066] C. Determination of the risk level of battery failure: The risk level of battery failure is determined by comparing the parking time prediction result with the prediction result of the next time the user uses the car. High risk means that the parking time of the vehicle is less than the maximum possible remaining time of the user, or the parking time of the vehicle is greater than the maximum possible remaining time of the user, but less than 1 day; medium risk means that the parking time of the vehicle is greater than the maximum possible remaining time of the user, but less than 3 days; low risk means that the parking time of the vehicle is greater than the maximum possible remaining time of the user, but less than 1 week; no risk means that the parking time of the vehicle is greater than 1 week.

[0067] Finally, it should be noted that the above examples of the present invention are merely examples for illustrating the present invention, and are not intended to limit the embodiments of the present invention. Although the applicant has described the present invention in detail with reference to the preferred embodiments, for those of ordinary skill in the art, other different forms of changes and modifications can be made based on the above description. It is impossible to list all the embodiments here. Any obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for predicting the parking time of a car with a battery sensor, characterized in that The steps include: 1) Extracting data sets of corresponding quantity and dimension from the Internet of Vehicles big data platform and performing data preprocessing and feature engineering, including the following steps: A. Data extraction: Extract a certain number of N vehicles from the Internet of Vehicles Big Data Platform, and the battery-related data for a continuous period of time. Automatically calculate the low-battery tags based on the uploaded battery-related data. If the calculated number of low-battery vehicles is less than the set number M, re-extract the data of N vehicles until the number of low-battery tags is greater than the set number M. B. Data preprocessing: Perform outlier processing on the relevant data extracted in the data extraction phase; C. Constructing the characteristic variable of battery life: The battery life state of the vehicle is obtained by using the extracted voltage value of the vehicle at the time of starting through the empirical threshold value; D. Construct a sample of abnormal power consumption characteristic variables: obtain the sleep-to-rest time point of the vehicle's internal network through the network heartbeat flag data extracted from step A, and determine whether the wake-up is actively woken up by the user based on the user operation data at that moment. If it is not actively woken up by the user, it is an abnormal wake-up. At the same time, calculate the number of abnormal wake-ups and the cumulative duration of abnormal wake-ups between each shutdown of the vehicle and the next start-up or the time when the low-power label appears; E. Construct the residual charge characteristic variable at the initial stage of a sample: determine the battery SOC value when the vehicle is turned off through the engine running status EngineStatus point shutdown data in the battery related data; F. Construct the residual charge characteristic variable at the end of a sample: the battery SOC value is divided into two cases: the battery low charge label appears or the battery low charge label does not appear; G. Construct a parking time variable for a sample: Calculate the time from the time the engine is turned off to the time the user powers on or the time when the battery-deficient tag appears through battery-related data; 2) Based on the battery-related data and the constructed characteristic variables as the training set and the test set, respectively select appropriate machine learning algorithms for model training, and select the model with the best model evaluation index to output the prediction result of the parking time, including the following steps: A. Prepare and split the model training data set: Use the preprocessed battery sensor data of the battery-related data to construct the feature variables as the feature variables for training the parking time prediction model, use the constructed parking time variable of a single sample as the prediction label of the prediction model, and divide the entire data set into a training set and a test set according to a certain ratio; B. Data normalization: Map the feature variables and predicted label data of the data set to the interval [0,1]; C. Model training: Select a variety of regression algorithms for model training, and select the optimal parameters of each model through cross-validation; D. Model selection: Input the test set into each model, obtain the mean square error of the prediction results of each model, and select the model with the smallest mean square error as the final prediction model; E. Obtaining the parking time prediction result: normalizing and restoring the prediction value output by the selected prediction model; 3) Complete the establishment of the data set and model training of the prediction model for the user's next car use time, and determine the level of battery shortage risk; compare the parking time prediction result with the user's next car use time prediction result to obtain the level of battery shortage risk; The dataset is established by calculating the average parking time of a bicycle every day. The characteristic variable of a sample is an array consisting of the average parking time of S consecutive days, and the predicted label is the average parking time of the S+1th day.

2. The method for predicting the parking time of a vehicle with a battery sensor according to claim 1, characterized in that: The battery-related data includes battery sensor data, user vehicle-related data and vehicle network heartbeat data, which are engine operating status EngineStatus, power status feedback PowerStatusFeedback, lock signal LockSignal, driver's side door lock status DriverDoorLockStatus, anti-theft status KeyAlarmStatus, unlock signal UnlockSignal, battery sensor SOC value, battery sensor voltage signal U_BATT, outdoor temperature signal EnvironmentalTemp, vehicle network sleep flag Sleepflag, vehicle network wake-up flag Awakeflag, and each data is timestamped.

3. The method for predicting the parking time of a vehicle with a battery sensor according to claim 2, characterized in that: When the vehicle network sleep flag Sleepflag changes from 0 to 1, and the vehicle network wake-up flag Awakeflag changes from 0 to 1, it is determined that a wake-up occurs at this moment. Within 2 minutes before and after the wake-up time, it is determined whether there is user active wake-up data. The active wake-up data includes the unlocking signal UnlockSignal, the locking signal LockSignal, the driver's side door lock status DriverDoorLockStatus changes, and the anti-theft status KeyAlarmStatus changes from the armed state to the disarmed state. If no active wake-up data is found, the current wake-up is determined to be an abnormal wake-up.

4. A method for predicting the parking time of a vehicle with a battery sensor according to any one of claims 1 to 3, characterized in that: The characteristic variables of the training parking time prediction model include the battery sensor SOC value, the battery sensor voltage signal U_BATT, the vehicle outside temperature signal EnvironmentalTemp, the number of abnormal wake-up times of a sample, the continuous abnormal wake-up time of a sample, the initial SOC value of a sample, the SOC value at the end of a sample, and the battery life status. The constructed parking time variable of a sample is used as the prediction label of the prediction model, and the data set is divided into a training set and a test set in a ratio of 8:

2.

5. A method for predicting the parking time of a vehicle with a battery sensor according to any one of claims 1 to 3, characterized in that: In the data normalization step, the mapping formula is Among them, max is the maximum value of the sample data, and min is the minimum value of the sample data.

6. A method for predicting the parking time of a vehicle with a battery sensor according to any one of claims 1 to 3, characterized in that: In the model training step, XGBOOST, GBDT, and SVR algorithms were selected for model training, and the optimal parameters of each model were selected through 5-fold cross validation; the training set was further divided into 5 parts of data, 4 of which were used as training sets, and the remaining 1 was used as a test set for model training and evaluation, and the model parameters that minimized the model mean square error were selected.

7. A method for predicting the parking time of a vehicle with a battery sensor according to any one of claims 1 to 3, characterized in that: The normalized reduction formula is x * =x*(max-min)+min.

8. A method for predicting the parking time of a vehicle with a battery sensor according to any one of claims 1 to 3, characterized in that: The parking time of a bicycle each day is the time when the user powers on the bicycle minus the previous shutdown time, or the time when the low-power label appears minus the previous shutdown time. The average parking time of the day is the sum of the parking times whose start and end times are both within the same day divided by the number of parking times that meet this condition. The characteristic variable S of a sample is an array of the average parking times for 7 consecutive days, and the predicted label is the average parking time on the 8th day.

9. A method for predicting the parking time of a vehicle with a battery sensor according to any one of claims 1 to 3, characterized in that: The method of determining the power shortage risk level includes the following steps: A. Dataset establishment: Calculate the average parking time of a bicycle every day. The feature variable of a sample is an array consisting of the average parking time of S consecutive days, and the predicted label is the average parking time of the S+1th day; B. Model training and prediction: Use time series algorithm to train the model and adjust the parameters so that the error of the prediction value is less than 1 hour, and output the prediction model; C. Determination of the risk level of battery failure: The risk level of battery failure is determined by comparing the predicted parking time with the predicted time of the next use of the car by the user.

10. A method for predicting the parking time of a vehicle with a battery sensor according to claim 9, characterized in that: The levels of the battery failure risk are as follows: high risk means that the vehicle's parking time is less than the user's maximum possible remaining vehicle usage time, or the vehicle's parking time is greater than the user's maximum possible remaining vehicle usage time, but less than 1 day; medium risk means that the vehicle's parking time is greater than the user's maximum possible remaining vehicle usage time, but less than 3 days; low risk means that the vehicle's parking time is greater than the user's maximum possible remaining vehicle usage time, but less than 1 week; no risk means that the vehicle's parking time is greater than 1 week.

11. A method for predicting the parking time of a vehicle with a battery sensor according to any one of claims 1 to 3, characterized in that: The battery related data of the continuous time is a continuous month.

12. The method for predicting the parking time of a vehicle with a battery sensor according to claim 4, characterized in that: The M is an integer number ranging from 3% to 8% of N.

Images