Logistics vehicle braking energy recovery method

Abstract

The invention provides a logistics vehicle braking energy recovery method. The logistics vehicle braking energy recovery method comprises the following steps that A, the whole vehicle driving process is divided into a plurality of data segments; b, extracting a plurality of vehicle speed related data in each data segment to form a data matrix set; c, carrying out dimension reduction operation on the data matrix set, then carrying out clustering analysis, and outputting a clustering result as a neural network; d, calculating fluctuation characteristic parameters and segmentation characteristic parameters by taking the data matrix set obtained in the step B as standard characteristic parameters; e, processing the standard characteristic parameters, the fluctuation characteristic parameters and the segmentation characteristic parameters, and selecting an optimal parameter subset to train a neural network; and H, vehicle speed data are obtained in real time, corresponding feature data are obtained from the optimal feature subset, a corresponding working condition category is obtained through a neural network, the proportion of motor braking in the total braking force is controlled by a vehicle control unit, and energy recovery is conducted. According to the invention, the braking energy can be recovered to the maximum extent on the premise of ensuring the braking safety.

Description

Technical Field

[0001] The present invention relates to the fields of speed planning for autonomous vehicles and energy optimization control management for hybrid vehicles, and particularly to a method for recovering braking energy of a logistics vehicle. Background Art

[0002] A Braking Energy Recovery System refers to a system applied to automobiles or rail transit that can convert the heat energy generated during braking into mechanical energy and store it in a capacitor, and can quickly release the energy when in use.

[0003] In the current braking energy recovery system of commercial vehicles, the proportion of electric braking in the overall braking process is closely related to the mechanical gear of the vehicle. However, existing braking energy recovery systems often cannot adaptively adjust according to the actual driving conditions and dynamic changes in the mechanical gear, resulting in limited energy recovery efficiency and difficulty in fully exploiting the potential of electric braking. Even if there is adaptive adjustment in the current technology, different driving conditions have different requirements. Therefore, formulating different electric braking intensities according to different driving conditions can better improve the energy recovery efficiency. Summary of the Invention

[0004] The present invention aims to provide a method for recovering braking energy of a logistics vehicle. The present invention can achieve rapid identification and response to changes in working conditions, has real-time performance, can quickly respond to changes in vehicle working conditions, ensure timely adjustment of the energy recovery strategy, maximize the recovery of braking energy on the premise of ensuring braking safety, and improve the endurance.

[0005] The technical solution of the present invention is as follows: The method for recovering braking energy of the logistics vehicle includes the following steps: A. Call the vehicle driving process record database, divide each vehicle driving process recorded therein into multiple data segments, and each data segment consists of a period of idle time and a subsequent driving period; B. Based on the vehicle speed data of each data segment, perform the following data acquisition: average driving speed, maximum speed, average acceleration, maximum acceleration, average deceleration, maximum deceleration, idle time ratio, deceleration time ratio, acceleration time ratio, uniform speed time ratio, speed standard deviation, acceleration standard deviation; Construct a data matrix composed of the above data for each data segment respectively to form a data matrix set; C. Perform dimensionality reduction operation on the data matrix set, and then input it into a clustering model with the initial clustering center optimized by the particle swarm algorithm for clustering analysis to obtain a clustering result; D. Use the data matrix set obtained in step B as the standard characteristic parameters; calculate the fluctuation characteristic parameters and segmented characteristic parameters based on the vehicle speed data for each data segment; E. Use the recursive feature elimination method based on random forest classification as the basic model to continuously iterate and eliminate the worst parameters for the standard characteristic parameters, fluctuation characteristic parameters, and segmented characteristic parameters. Use the 5-fold cross-validation method to verify the accuracy when different numbers of features are used, and determine the optimal number of selected features; F. Process using the recursive feature elimination method with the optimal number of features, and select to obtain the optimal parameter subset; G. Use the optimal parameter subset as the input and the clustering result of step C as the output to train the learning vector quantization neural network to obtain the trained learning vector quantization neural network; H. Obtain the vehicle speed data in real time, obtain the corresponding feature data from the optimal feature subset according to the vehicle speed data, then input it into the trained learning vector quantization neural network, output to obtain the corresponding operating condition category, determine the proportion of the electric motor braking in the total braking force according to the operating condition category, input it into the vehicle's vehicle controller, and the vehicle controller controls the proportion of the electric motor braking in the total braking force and performs energy recovery.

[0006] In the said step B, when obtaining data for each data segment, data preprocessing is required. First, fill in the missing vehicle speed values, then process the abnormal acceleration data, delete the pulse data and idle data, fill in the missing vehicle speed values, and remove the noise data.

[0007] The method for filling in the missing vehicle speed values includes the following steps: For data with a deceleration less than -16.5 km / h, check whether there are zero values in the data of the next four unit times. If there are zero values, use linear interpolation to fill in the data; for data with an acceleration greater than 6 km / h, check whether there are zero values in the data of the previous four unit times. If there are zero values, use linear interpolation to fill in the data; The process of processing the abnormal acceleration data includes the following steps: For data with a maximum acceleration exceeding 6 km / h of the vehicle, adjust it to 6 km / h; for data with a maximum deceleration exceeding -16.5 km / h, adjust it to -16.5 km / h.

[0008] The said pulse data refers to: Check that the current vehicle speed value is not zero and the acceleration is greater than 6 km / h, and at the same time, both the vehicle speed values of the previous three unit times and the vehicle speed values of the next three unit times are 0. Then, determine that the current vehicle speed value is a pulse value and set it to 0; The said idle data refers to: Data with a vehicle speed of 10 km / h continuously for 20 seconds, or data with a speed of zero for more than 180 seconds.

[0009] The described noise data removal includes the following steps: These random noises are filtered using moving average filtering, and the calculation formula is as follows: (1) Where y represents the filtered vehicle speed sequence, x represents the original vehicle speed data sequence, n represents the current position, and k represents the window index.

[0010] In the described step D, the fluctuation characteristic parameters are specifically: the average value of the square of the acceleration, the frequency of speed change greater than 2 km / h within every 100 m, the frequency of speed change greater than 10 km / h within every 100 m, the frequency of speed change greater than 2 km / h within every 100 s, and the frequency of speed change greater than 10 km / h within every 100 s, which are used as the fluctuation characteristic parameters; The sectional characteristic parameters are specifically: the vehicle speed proportion of 0 - 10 km / h, the vehicle speed proportion of 10 - 20 km / h, the vehicle speed proportion of 20 - 30 km / h, the vehicle speed proportion of 30 - 40 km / h, the vehicle speed proportion of 40 - 50 km / h, the vehicle speed proportion of 50 - 60 km / h, the vehicle speed proportion of 60 - 70 km / h, the vehicle speed proportion of 70 - 80 km / h, the vehicle speed proportion of 80 - 90 km / h, and the vehicle speed proportion of 90 - 100 km / h.

[0011] In the described step H, the specific corresponding working condition categories output by the neural network are: plain highway, hilly highway, and national road.

[0012] In the described step H, initially, at least vehicle speed data input for more than 180 seconds of vehicle travel is required for the learning vector quantization neural network to output the corresponding working condition; during subsequent operation, the time interval for the neural network to output the corresponding working condition category is updated every 3 seconds.

[0013] In the described step H, the specific proportion of the electric motor brake in the total braking force determined according to the working condition category is: 30% for plain highway, 50% for hilly highway, and 40% for national road.

[0014] Through high - speed data processing and real - time decision - making algorithms, the present invention realizes the rapid identification and response to working condition changes, quickly identifies the corresponding working conditions, timely allocates the proportion of the electric motor brake in braking, thereby maximizing the recovery of braking energy, and at the same time has high adaptability in the face of different situations and has good application prospects. Brief Description of the Drawings

[0015] Figure 1 is the flow chart of the present invention; Figure 2 is the schematic diagram of the working condition effect of the present invention; Figure 3 Schematic diagram of the predicted working condition effect of the test set according to the present invention; Figure 4 Schematic diagram of the predicted working condition effect of the real-time input data in the embodiment of the present invention; Figure 5 Schematic diagram of the improvement effect of the actual driving braking energy recovery efficiency of the present invention; Figure 6 Schematic diagram of the principle of the learning vector quantization neural network in the present invention. Detailed implementation manners Embodiment 1

[0016] The braking energy recovery method for a logistics vehicle includes the following steps: A. Call the vehicle driving process record database, and divide each vehicle driving process recorded therein into a plurality of data segments, each data segment consisting of a period of idle time and a subsequent driving period; B. Based on the vehicle speed data of each data segment, perform the following data acquisition: average driving speed, maximum speed, average acceleration, maximum acceleration, average deceleration, maximum deceleration, idle time ratio, deceleration time ratio, acceleration time ratio, constant speed time ratio, speed standard deviation, acceleration standard deviation; When performing data acquisition for each data segment, data preprocessing is required. First, fill in the missing values of the vehicle speed, then process the abnormal acceleration data, delete the pulse data and idle data, fill in the missing values of the vehicle speed, and remove the noise data.

[0017] The method for filling in the missing values of the vehicle speed includes the following steps: For data with a deceleration less than -16.5 km / h, check whether there are zero values in the data of the next four unit times. If there are zero values, use linear interpolation to fill in the data; for data with an acceleration greater than 6 km / h, check whether there are zero values in the data of the previous four unit times. If there are zero values, use linear interpolation to fill in the data; where each unit time is 1 second; The process for processing abnormal acceleration data includes the following steps: for data with a maximum acceleration exceeding 6 km / h of the vehicle, adjust it to 6 km / h; for data with a maximum deceleration exceeding -16.5 km / h, adjust it to -16.5 km / h.

[0018] The pulse data refers to: checking that the current vehicle speed value is not zero and the acceleration is greater than 6 km / h, and at the same time satisfying that the vehicle speed values of the previous three unit times and the vehicle speed values of the next three unit times are all 0, then determining that the current vehicle speed value is a pulse value and setting it to 0; where each unit time is 1 second; The idle speed data mentioned above refers to: data where the vehicle speed is 10 km / h for 20 consecutive seconds, or data where the speed is zero for more than 180 seconds.

[0019] The steps for removing noise data are as follows: These random noises are filtered using moving average filtering, and the calculation formula is as follows: (1) Where y represents the filtered vehicle speed sequence, x represents the original vehicle speed data sequence, n represents the current position, and k represents the window index.

[0020] For each data segment, a data matrix composed of these data is constructed respectively to form a data matrix set; C. Perform dimensionality reduction operations on the data matrix set, and then input it into a clustering model with the initial clustering center optimized by the particle swarm algorithm for clustering analysis to obtain the clustering result; D. Use the data matrix set obtained in step B as the standard feature parameters; calculate the fluctuation feature parameters and segment feature parameters based on the vehicle speed data of each data segment; The fluctuation feature parameters are specifically: the average value of the square of the acceleration, the frequency of speed change greater than 2 km / h within every 100 m, the frequency of speed change greater than 10 km / h within every 100 m, the frequency of speed change greater than 2 km / h within every 100 s, and the frequency of speed change greater than 10 km / h within every 100 s, which are used as the fluctuation feature parameters; The segment feature parameters are specifically: the vehicle speed proportion of 0 - 10 km / h, the vehicle speed proportion of 10 - 20 km / h, the vehicle speed proportion of 20 - 30 km / h, the vehicle speed proportion of 30 - 40 km / h, the vehicle speed proportion of 40 - 50 km / h, the vehicle speed proportion of 50 - 60 km / h, the vehicle speed proportion of 60 - 70 km / h, the vehicle speed proportion of 70 - 80 km / h, the vehicle speed proportion of 80 - 90 km / h, and the vehicle speed proportion of 90 - 100 km / h.

[0021] E. Use the recursive feature elimination method based on the random forest classification as the basic model to continuously iterate and eliminate the worst parameters for the standard feature parameters, fluctuation feature parameters, and segment feature parameters, and use the 5 - fold cross - validation method to verify the accuracy when different numbers of features are used to determine the optimal number of selected features; F. Process using the recursive feature elimination method with the optimal number of features to select the optimal parameter subset; G. Use the optimal parameter subset as the input and the clustering result in step C as the output to train the learning vector quantization neural network to obtain the trained learning vector quantization neural network; H. Obtain vehicle speed data in real time, obtain corresponding feature data from the optimal feature subset according to the speed data, then input it into the trained learning vector quantization neural network, output the corresponding operating condition category, determine the proportion of electric motor braking in the total braking force according to the operating condition category, input it into the vehicle's vehicle controller, and the vehicle controller controls the proportion of electric motor braking in the total braking force and performs energy recovery.

[0022] The corresponding operating condition categories output by the neural network are specifically: plain highway, hilly highway, and national road.

[0023] Initially, at least the vehicle speed data input for more than 180 seconds of vehicle travel needs to be obtained before the learning vector quantization neural network will output the corresponding operating condition; during subsequent operation, the time interval for the neural network to output the corresponding operating condition category is updated every 3 seconds.

[0024] The proportion of electric motor braking in the total braking force determined according to the operating condition category is specifically: 30% for plain highway, 50% for hilly highway, and 40% for national road. Embodiment 2

[0025] Use the Trunksim dynamics model as the vehicle simulation model, and use the method of Embodiment 1 to perform energy control simulation in the model. The specific operations are as follows: The vehicle speed curves under three operating conditions are collected as Figure 2 shown. The average vehicle speed on the plain highway is 80 km / h, the driving is relatively stable, and there are few bumps; the average vehicle speed on the hilly highway is relatively high at 70 km / h, the driving fluctuation is obvious, and there are many bumps; the average vehicle speed on the national road is relatively low at 40 - 60 km / h, the driving fluctuation is obvious, and there are many bumps.

[0026] In the simulation experiment, input the vehicle speed data set that combines the above three operating conditions. The driver steps on the accelerator pedal and brake pedal accordingly according to the vehicle speed in the data set. When the driving time exceeds 180 seconds, the vehicle controller will use the current vehicle speed and the vehicle speed in the previous 180 seconds as a data segment, calculate the optimal feature parameters of this data segment, input them into the neural network as input, predict the current operating condition category, and the vehicle controller internally selects the proportion of electric motor braking in the total braking force according to the predicted category, updates the data segment every 3 seconds, and also updates the predicted category. When braking occurs, that is, when the brake pedal opening is greater than 0, judge whether the current battery SOC is less than 90% to prevent overcharging and damaging the battery, and judge whether the brake pedal opening is less than 60% to prevent failure to perform emergency braking in time. When all the above conditions are met, the motor torque reverses and becomes the generator mode, and the reversed torque of the motor is used as a part of the braking torque.

[0027] Figure 3 The total number of segments in the test set is 194, and the recognition accuracy rate is: 94.33%; Figure 4 The accuracy rate of predicting working conditions for real-time input data is 80.3%; Figure 5 The driving time periods are divided in the order of plain highway, hilly highway, and national road. At the beginning, it is on the plain highway. Since the vehicle speed on the plain highway is relatively high and braking is less, the proportion of electric motor braking in the total braking force is set relatively low at 30%. Therefore, the difference in the remaining battery charge SOC with and without using the present invention is not obvious. When it comes to the hilly highway, without using the present invention, in order to ensure braking performance, only a more conservative proportion of electric motor braking force / total braking force can be set, and more braking energy recovery cannot be carried out. However, when the present invention predicts the working condition on the hilly highway, due to the scarcity of vehicles on the high-speed road and the increase in braking times under this working condition, when the braking energy recovery condition is met, the proportion of electric motor braking in the total braking force is allowed to be increased to 50% for more energy recovery. Finally, when it comes to the national road working condition, the present invention predicts driving on the national road. Since there are more vehicles on the national road, the proportion of electric motor braking in the total braking force needs to be reduced to 40%. Without using the present invention and being unable to predict the working condition, if a relatively high proportion of electric motor braking in the total braking force is set under this working condition, a larger braking pedal opening is required to ensure braking performance, thus affecting driving comfort. If the proportion of electric motor braking in the total braking force is relatively low, the braking energy cannot be fully recovered. The driving distance of this journey is 82.6 km. With the present invention, the SOC drops from 80% to 41.6%, and without using the present invention, the SOC drops from 80% to 40.8%, saving 0.8% of the power.

Claims

1. A method for recovering braking energy of a logistics vehicle, comprising the following steps: A. Call the vehicle driving process record database, divide each vehicle driving process recorded therein into multiple data segments, and each data segment consists of an idle time period followed immediately by a driving time period; B. Based on the vehicle speed data of each data segment, perform the following data acquisition: average driving speed, maximum speed, average acceleration, maximum acceleration, average deceleration, maximum deceleration, idle time ratio, deceleration time ratio, acceleration time ratio, uniform speed time ratio, speed standard deviation, acceleration standard deviation; Construct a data matrix composed of the above data for each data segment respectively, and form a data matrix set; C. Perform dimensionality reduction operation on the data matrix set, and then input it into a clustering model with the initial clustering center optimized by the particle swarm algorithm for clustering analysis to obtain a clustering result; D. Use the data matrix set obtained in step B as standard characteristic parameters; Calculate and obtain fluctuation characteristic parameters and segment characteristic parameters based on the vehicle speed data of each data segment; E. Use the recursive feature elimination method based on the random forest classification as the basic model to continuously iterate and eliminate the worst parameters for the standard characteristic parameters, fluctuation characteristic parameters and segment characteristic parameters, and use the 5-fold cross-validation method to verify the accuracy when different numbers of features are used, and determine the selected optimal number of features; F. Process with the recursive feature elimination method using the optimal number of features to select and obtain an optimal parameter subset; G. Use the optimal parameter subset as the input and the clustering result of step C as the output to train a learning vector quantization neural network to obtain a trained learning vector quantization neural network; H. Real-time obtain the vehicle speed data of the vehicle, obtain the corresponding feature data from the optimal feature subset according to the vehicle speed data, then input it into the trained learning vector quantization neural network, output to obtain the corresponding working condition category, determine the proportion of the electric brake in the total braking force according to the working condition category, input it into the vehicle's vehicle controller, and the vehicle controller controls the proportion of the electric brake in the total braking force and performs energy recovery.

2. The method for recovering braking energy of a logistics vehicle according to claim 1, wherein: In the step B, when performing data acquisition on each data segment, data preprocessing needs to be performed. First, fill in the missing values of the vehicle speed, then process the abnormal acceleration data, delete the pulse data and idle data, fill in the missing values of the vehicle speed, and remove the noise data.

3. The method for recovering braking energy of a logistics vehicle according to claim 2, wherein: The method for filling in the missing values of the vehicle speed includes the following steps: For data with a deceleration less than -16.5 km / h, check whether there are zero values in the data of the next four unit times. If there are zero values, use linear interpolation to fill in the data; for data with an acceleration greater than 6 km / h, check whether there are zero values in the data of the previous four unit times. If there are zero values, use linear interpolation to fill in the data; The abnormal acceleration data processing process described above includes the following steps: For data where the maximum acceleration of the vehicle exceeds 6 km / h, adjust it to 6 km / h; for data where the maximum deceleration exceeds -16.5 km / h, adjust it to -16.5 km / h.

4. The method for recovering braking energy of a logistics vehicle according to claim 2, wherein: The pulse data mentioned above refers to: checking that the current vehicle speed value is not zero and the acceleration is greater than 6 km / h, and at the same time satisfying that the vehicle speed values in the previous three unit time intervals and the vehicle speed values in the next three unit time intervals are all 0. Then, it is determined that the current vehicle speed value is a pulse value and set to 0. The idle data mentioned above refers to: data where the vehicle speed is 10 km / h continuously for 20 seconds, or data where the speed is zero for more than 180 seconds.

5. The method for recovering braking energy of a logistics vehicle according to claim 2, wherein: The removal of noise data includes the following steps: Use moving average filtering to filter and process these random noises. The calculation formula is as follows: （1） Where y represents the vehicle speed sequence after filtering, x represents the original vehicle speed data sequence, n represents the current position, and k represents the index of the window.

6. The method for recovering braking energy of a logistics vehicle according to claim 1, wherein: In step D mentioned above, the fluctuation characteristic parameters are specifically: the average value of the square of the acceleration, the frequency of speed change greater than 2 km / h within every 100 m, the frequency of speed change greater than 10 km / h within every 100 m, the frequency of speed change greater than 2 km / h within every 100 s, and the frequency of speed change greater than 10 km / h within every 100 s, which are used as the fluctuation characteristic parameters; The segmented characteristic parameters are specifically: the vehicle speed ratio of 0 - 10 km / h, the vehicle speed ratio of 10 - 20 km / h, the vehicle speed ratio of 20 - 30 km / h, the vehicle speed ratio of 30 - 40 km / h, the vehicle speed ratio of 40 - 50 km / h, the vehicle speed ratio of 50 - 60 km / h, the vehicle speed ratio of 60 - 70 km / h, the vehicle speed ratio of 70 - 80 km / h, the vehicle speed ratio of 80 - 90 km / h, and the vehicle speed ratio of 90 - 100 km / h.

7. The method for recovering braking energy of a logistics vehicle according to claim 1, wherein: In step H mentioned above, the specific corresponding working condition categories obtained by the neural network output are: plain highway, hilly highway, and national road.

8. The method for recovering braking energy of a logistics vehicle according to claim 7, wherein: In step H mentioned above, initially, at least vehicle speed data input for more than 180 seconds of vehicle travel is required for the learning vector quantization neural network to output the corresponding working condition; during subsequent operation, the time interval for the neural network to output the corresponding working condition category is updated every 3 seconds.

9. The method for recovering braking energy of a logistics vehicle according to claim 8, wherein: In step H mentioned above, the specific proportion of the electric brake in the total braking force determined according to the working condition category is: 30% for plain highway, 50% for hilly highway, and 40% for national road.

Images