Electric vehicle remaining mileage prediction under quantitative driving behavior

Abstract

The invention provides a novel intelligent method for predicting the remaining mileage of a new energy automobile, the method is applied to data collection and model training of the automobile under complex working conditions, and the result predicted by a neural model is closer to a real value by establishing a driving behavior quantitative factor and is more accurate than the previous prediction method; according to the method, establishment of a neural network model under a complex working condition is realized, soc, a driving behavior quantification factor, an environment temperature, an electrical appliance load and a speed working condition are taken as input, so that various influence factors can be considered in a prediction influence range, and a prediction error range is further reduced.

Description

Technical Field

[0001] The invention relates to the technical field of new energy vehicles, and in particular to a method for predicting the remaining mileage of an electric vehicle based on a driving behavior quantification factor. Background Art

[0002] As an intuitive expression of the endurance of new energy vehicle power batteries, the remaining mileage is directly related to the driver's judgment on driving new energy vehicles. As an in-depth and intuitive study of the RUL (remaining life) prediction of new energy power batteries, the remaining mileage prediction of new energy vehicles is of great significance to the development of automotive batteries. The remaining mileage prediction has a similar method to the remaining life prediction. By collecting car driving data such as SOC data, battery temperature, ambient temperature data, etc., a suitable model is selected to be established, and the collected data is used as input values. By training the model, an accurate prediction value is obtained. Since the actual remaining mileage is different from the experimental remaining mileage under normal and unchanged working conditions, changes in various conditions during driving can affect the remaining mileage value. Therefore, it is necessary to establish an excellent model to optimize the collected important related data so that the predicted value obtained by the final model can be as close to the true value as possible. Summary of the invention

[0003] The purpose of the present invention is to provide a method for predicting the remaining mileage of an electric vehicle based on a driving behavior quantification factor, which can comprehensively consider the influence of vehicle state parameters and driving behavior on the remaining mileage and improve the accuracy of prediction.

[0004] To achieve the above objectives, the present invention adopts the following technical solutions: S1 Data collection and preprocessing:

[0005] S1 Data collection and preprocessing: Collect historical driving data of new energy vehicles, classify and clean the data, remove outliers, missing values ​​and duplicate values, and divide the data.

[0006] S2 Driving behavior quantification: Based on the cleaned data, driving behavior evaluation indicators are selected, each indicator is weighted and the weight coefficient is evaluated; the driving behavior quantification factor is obtained by normalizing the weighted sum.

[0007] S3 Neural Network Model Construction and Training: Build an LSTM neural network model, take vehicle state parameters (SOC, ambient temperature, speed, etc.) and driving behavior quantification factors as input, train the model; use the model car to verify the training results.

[0008] S4 Remaining mileage prediction: The vehicle status parameters and driving behavior quantitative factors collected in real time are input into the trained LSTM neural network model to obtain the predicted value of the remaining mileage.

[0009] ] Preferably, S1 includes the following steps: S1-1, processing missing values, abnormal values ​​and repeated values ​​of the data to eliminate the influence of abnormal data. S1-2, dividing the cleaned data, first dividing the driving events, that is, the process from vehicle startup to shutdown, including acceleration, deceleration, constant speed and idling and other driving processes. Each driving event is divided into multiple driving segments of more than 20 seconds, and the end segment of the event less than 20 seconds is ignored.

[0010] Preferably, S2 comprises the following steps:

[0011] S2-1. Seven evaluation indicators were selected from the aspects of vehicle speed and acceleration to digitize driving behavior. At the same time, three driving behavior evaluation indicators, namely, the proportion of rapid acceleration time, the proportion of high-speed time and the proportion of energy recovery, were analyzed.

[0012] S2-2. Analysis of sudden acceleration duration: The vehicle speed and acceleration in short driving segments are extracted, and a method for judging the threshold line of sudden acceleration events based on the 90% percentile line of the segmented interval is proposed. Taking 10km / h as a segmented interval, different speed intervals are analyzed independently, and the 90% percentile line of each speed interval is used as the threshold line for judging sudden acceleration events, so as to avoid the difficulty of identifying high-speed events with the threshold line.

[0013] S2-3. Analysis of the proportion of high-speed time: The speed range is divided into low speed, medium speed and high speed through the K-means clustering method and the intra-cluster error sum of squares method to obtain the proportion of high-speed time.

[0014] ] S2-4. The importance of the evaluation indicators is evaluated through the random forest algorithm, and the importance of each evaluation indicator to the energy consumption of the drive motor is used as the weight coefficient.

[0015] S2-4. Calculation of driving behavior quantitative factors.

[0016] Preferably, the S3 includes the following steps: S3-1. Taking the driving behavior quantization factor and time as input, the neural network is trained using the original data.

[0017] Preferably, the S4 includes the following steps: S4-1, predicting the remaining mileage through an LSTM neural network. BRIEF DESCRIPTION OF THE DRAWINGS FIG1 is a basic flow chart of a method for predicting the remaining range of an electric vehicle based on a driving behavior quantification factor according to the present invention; FIG2 is a digital processing of driving behavior, in which seven evaluation indicators are selected from the aspects of vehicle speed and acceleration; FIG3 is a specific implementation method for the distribution of vehicle speed and acceleration; FIG4 is a division of the center points of the clustering results for low speed, medium speed and high speed; FIG5 is a weight coefficient of the importance of each indicator;

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] 1 , the present invention provides a method for predicting the remaining range of an electric vehicle based on a driving behavior quantification factor, comprising the following steps:

[0022] S1. Data collection and preprocessing: Collect historical driving data of new energy vehicles, classify and clean the data, remove outliers, missing values ​​and duplicate values, and divide the data.

[0023] S2. Driving behavior quantification: Based on the cleaned data, driving behavior evaluation indicators are selected, each indicator is weighted and the weight coefficient is evaluated; the driving behavior quantification factor S2-1 is obtained by normalizing the weighted sum. Considering the availability and operability of the evaluation indicators, 7 evaluation indicators are selected from the vehicle speed and acceleration to digitize the driving behavior, as shown in Figure 2; at the same time, the three driving behavior evaluation indicators of rapid acceleration time proportion, high-speed time proportion and energy recovery proportion are analyzed.

[0024] ] Analysis of sudden acceleration duration: The speed and acceleration in the short driving segment are extracted and summarized to obtain the distribution of speed and acceleration in Figure 3. This paper proposes a threshold line judgment method for sudden acceleration events based on the 90% percentile line of the segmented interval. Taking 10km / h as a segmented interval, different speed intervals are analyzed independently, and the 90% percentile line of each speed interval is used as the threshold line for sudden acceleration event judgment, so as to avoid the difficulty of identifying high-speed events by the threshold line.

[0025] Analysis of the proportion of high-speed time: The mean vehicle speed and drive motor energy consumption of each short-trip segment are summarized, and the K-means clustering algorithm is used to cluster the mean vehicle speed and drive motor energy consumption. The number of clusters is specified to be 3, and the average value of the clustering silhouette coefficient is the largest. Combined with the elbow method, the sum of squared errors within the cluster decreases most significantly when the number of clusters is 3. Combining the two methods, the clustering effect is best when K=3. The speed range is divided into low speed, medium speed and high speed. The center point of the clustering result is shown in Figure 4.

[0026] ] S2-2. This method uses the random forest algorithm to evaluate the importance of the evaluation indicators, and uses the importance of each evaluation indicator to the drive motor energy consumption as a weight coefficient. The basic method of random forest is: divide the original data set into n training sets and n test sets for random sampling, obtain test set samples and training set samples, and input the 7 defined driving behavior evaluation indicators; establish a decision tree model, establish n decision trees, bring the test set data into the trained decision tree model, and obtain n results through model calculation. The average of the results can obtain the predicted value of the drive motor energy consumption.

[0027] ] Before using the random forest algorithm to evaluate the importance of feature indicators, it is necessary to process the valid data in the data set. 70% of the data set is used as the training set to train the model; 30% of the data set is used as the test set to verify the effectiveness of the model. After model training, the importance weight coefficients of each indicator are obtained, as shown in Figure 5. The order of the evaluation index weight coefficients is: high-speed time proportion, vehicle speed average, rapid acceleration time, energy recovery proportion, acceleration average, vehicle speed standard deviation, acceleration standard deviation.

[0028] S2-3, driving behavior quantitative factors are determined by various driving behavior evaluation indicators. Because these six evaluation indicators are negatively correlated with energy consumption, while this evaluation indicator is positively correlated with energy consumption, In order to eliminate the impact of different evaluation indicators, the evaluation indicators are first normalized. a , P h , P r This is normalized data. m 、V std 、a m 、a std These four evaluation indicators are normalized. The weight coefficients of each evaluation indicator for motor energy consumption are obtained in the above analysis. The weight coefficients are multiplied by the normalized values ​​of the evaluation indicators to obtain the weighted sum, and the driving behavior quantification factor Q is obtained. The calculation formula is: , where n represents the number of evaluation indicators; represents the normalized evaluation index; Represents the weight coefficient of the evaluation index.

[0029] S3. From the operating data of four randomly selected electric vehicles, the influencing factors are extracted as the input of the LSTM neural network at time t, and all input factors are normalized to improve the training effect of the model, and the remaining driving range is used as the output of the model at time t. In the present invention, in order to further verify the training effect of the prediction model, the total data volume is divided into a training set and a test set. The ratios of the training set and the test set are 7:3, 8:2 and 9:1 respectively, and the data are not randomly sampled; when training the model, the sequence length is set to 20, the batch size is 256, the number of network nodes is 128, the number of iterations is 100 times, and the optimizer uses the Adam algorithm.

[0030] The input and output of the LSTM neural network model at time t are defined as , = (remaining mileage), Where is the input at time t, is the output at time t,; They are driving behavior quantification factor, ambient temperature, device power load rate, and speed condition. The remaining mileage is output through LSTM prediction.

[0031] The implementation described above is only a description of the preferred mode of the present invention, and does not limit the scope of the present invention. Without departing from the design spirit of the present invention, ordinary technicians in this field may make various modifications and improvements to the technical solution of the present invention, which should all fall within the determined protection scope of the invention right.

Claims

1. A method for predicting the remaining mileage of an electric vehicle based on a driving behavior quantification factor, characterized in that: The method includes the following steps: S1 Data collection and preprocessing: Collect historical driving data of electric vehicles, and clean, classify and preprocess the data to remove outliers, missing values ​​and duplicate values; S2 Driving behavior quantification: Based on the preprocessed data, select evaluation indicators related to driving behavior, quantify each indicator and assign weights to obtain driving behavior quantification factors; S3 Model construction and training: Construct a neural network model, take vehicle state parameters and driving behavior quantification factors as input, and remaining mileage as output, and train and optimize the model; S4 Remaining mileage prediction: Collect vehicle state parameters and driving behavior data of electric vehicles in real time, input them into the trained neural network model, and obtain the predicted value of the remaining mileage.

2. The method for predicting the remaining range of an electric vehicle according to claim 1, wherein: The driving behavior quantification step further includes: S2-1 selecting driving behavior evaluation indicators, including but not limited to accelerator pedal opening, brake pedal opening, steering angle, driving speed and driving time; S2-2 normalizing each evaluation indicator to eliminate dimensional differences; S2-3 assigning corresponding weights to each evaluation indicator according to the degree of influence of each evaluation indicator on the remaining mileage; S2-4 multiplying the normalized evaluation indicator value with its corresponding weight, and summing them to obtain the driving behavior quantification factor.

3. The method for predicting the remaining range of an electric vehicle according to claim 1 or 2, wherein: The neural network model adopts a long short-term memory (LSTM) network structure, which can process time series data and capture the dynamic relationship between driving behavior and remaining mileage.

4. A system for predicting remaining mileage of an electric vehicle, the system comprising: S4-1 data acquisition module, used to collect vehicle status parameters and driving behavior data of electric vehicles in real time; S4-2 data processing module, used for preprocessing the collected data and quantifying driving behavior; S4-3 prediction model module, including a neural network model trained according to the method described in claim 1, 2 or 3, used for receiving the output of the data processing module and outputting the predicted value of the remaining mileage; S4-5 display module, used for displaying the predicted value of the remaining mileage to the driver.

5. The electric vehicle remaining range prediction system according to claim 4, further comprising: S5-1 Model update module, used to update and optimize the neural network model regularly or according to specific conditions to improve prediction accuracy.

6. An electric vehicle, characterized in that: The electric vehicle is integrated with the electric vehicle remaining range prediction system according to claim 4 or 5.

7. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for predicting the remaining range of an electric vehicle according to claim 1, 2 or 3.

8. A computing device for realizing the prediction of the remaining range of an electric vehicle, comprising: S8-1 processor; S8-2: a memory storing a computer program executable on a processor, wherein the program, when executed by the processor, implements the steps of the method for predicting the remaining range of an electric vehicle according to claim 1, 2 or 3.

Images