A battery management method and system for new energy vehicles

Abstract

The present application relates to the field of battery management technology for new energy vehicles, and particularly relates to a battery management method and system for new energy vehicles. The method comprises the following steps: obtaining the historical discharge power of a new energy vehicle control terminal and the historical driving environment state in a high-precision map, filtering and smoothing the historical discharge power to obtain a power fluctuation smooth curve; secondly, according to the historical driving environment state, performing semi-slope start-stop power mapping on the power fluctuation smooth curve to generate semi-slope start-stop power mapping data, and obtaining energy consumption progressive law through energy consumption increment behavior learning; finally, designing a battery remaining endurance prediction architecture according to the energy consumption progressive law, and sending it to the new energy vehicle terminal to perform battery management. The present application optimizes the battery management technology for new energy vehicles, making the battery management technology for new energy vehicles more perfect.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of battery management technology for new energy vehicles, and in particular to a battery management method and system for new energy vehicles. BACKGROUND

[0002] In the battery management of new energy vehicles, how to efficiently and scientifically evaluate the remaining endurance capability of the battery is an important field of research and application. The remaining endurance of the battery is not only related to the health status of the battery itself, but also closely related to driving environment, driving habits, road conditions and other factors. For example, under complex working conditions such as uphill and sudden acceleration, the energy consumption of the battery fluctuates greatly, which requires the battery management system to accurately predict the remaining battery power and endurance time under different driving conditions. In addition, with the continuous development of high-precision maps and intelligent transportation technologies, new energy vehicles have an increasing demand for environmental information. Through high-precision maps, real-time data of vehicle driving state and surrounding environment are obtained, providing more accurate basis for battery management, which has become a trend in future development. However, the traditional battery management method for new energy vehicles has the problem of inaccurate energy consumption analysis under the start-stop state of the road half-slope state, resulting in large errors in the judgment of the remaining endurance of the battery. SUMMARY

[0003] Therefore, it is necessary to provide a battery management method and system for new energy vehicles to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a battery management method for new energy vehicles, the method comprising the following steps:

[0005] Step S1: obtaining historical discharge power through a new energy vehicle control terminal; and obtaining historical driving environment state through a high-precision map carried by the new energy vehicle control; filtering and smoothing the historical discharge power to obtain a power fluctuation smooth curve;

[0006] Step S2: mapping start-stop power under road half-slope state according to the power fluctuation smooth curve based on the historical driving environment state to obtain half-slope start-stop power mapping data; performing energy consumption increment behavior learning on the half-slope start-stop power mapping data to generate an energy consumption progression rule;

[0007] Step S3: designing a battery remaining endurance prediction architecture according to the energy consumption progression rule to obtain a remaining endurance prediction architecture; sending the remaining endurance prediction architecture to the new energy vehicle terminal to perform battery management for new energy vehicles.

[0008] Preferably, the present application also provides a battery management system for new energy vehicles for performing the battery management method for new energy vehicles as described above, the battery management system for new energy vehicles comprising:

[0009] a data acquisition module configured to acquire historical discharge power through a new energy vehicle control terminal, and acquire a historical driving environment state through a high-precision map carried by the new energy vehicle control, and perform filtering and smoothing processing on the historical discharge power to obtain a power fluctuation smooth curve;

[0010] a behavior learning module configured to perform start-stop power mapping of the power fluctuation smooth curve under a road half-slope state according to the historical driving environment state to obtain half-slope start-stop power mapping data, and perform energy consumption increment behavior learning on the half-slope start-stop power mapping data to generate an energy consumption progression law;

[0011] an architecture design module configured to perform battery residual endurance prediction architecture design according to the energy consumption progression law to obtain a residual endurance prediction architecture, and send the residual endurance prediction architecture to a new energy vehicle terminal to perform new energy vehicle battery management.

[0012] The beneficial effects of this invention lie in its ability to analyze in detail the impact of different driving conditions on battery power consumption by acquiring historical discharge power data from the control terminal of new energy vehicles and historical driving environment status provided by high-precision maps. By filtering and smoothing the historical discharge power to generate a power fluctuation smoothing curve, noise caused by instantaneous fluctuations can be effectively removed, extracting a more stable and representative power change trend. This not only improves data accuracy but also provides more accurate foundational data for subsequent power mapping and energy consumption learning, enhancing the battery management system's adaptability and predictive capabilities to dynamic driving environments. Based on historical driving environment status, start-stop power mapping is performed for special road conditions such as hill starts. By analyzing the battery power demand under different slopes, hill start-stop power mapping data is generated. This process accurately captures battery power fluctuations under complex road conditions, providing key data for further energy consumption behavior learning. Through incremental energy consumption behavior learning, the system can gradually summarize the progressive energy consumption law of the battery under specific conditions, revealing the gradual pattern of battery energy consumption under different driving scenarios and road conditions. This progressive energy consumption law not only improves the predictive accuracy of the battery management system but also provides a theoretical basis for optimizing battery efficiency and extending battery life. Based on the energy consumption progression patterns learned in previous studies, a battery remaining range prediction architecture was designed. This architecture can accurately predict the remaining driving range based on real-time driving status, road conditions, and battery health. The introduction of this innovative architecture not only provides more accurate range predictions in changing driving environments but also adjusts battery management strategies according to real-time changes. Sending this prediction architecture to the new energy vehicle terminal allows the battery management system to dynamically adjust the battery charging and discharging strategy based on real-time data and prediction results, maximizing battery energy utilization and range performance, thereby improving the user's driving experience and battery efficiency. This invention is an improvement on a traditional battery management method for new energy vehicles, addressing the problem of inaccurate energy consumption analysis in start-stop states on inclines, which leads to large errors in judging remaining battery range. It improves the accuracy of energy consumption analysis in start-stop states on inclines and reduces errors in judging remaining battery range. Attached Figure Description

[0013] Figure 1 A flowchart illustrating the steps of a battery management method for new energy vehicles;

[0014] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0015] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S3. DETAILED DESCRIPTION

[0016] Referring to Figures 1 to 3 A battery management method for a new energy vehicle, the method comprising the following steps:

[0017] Step S1: obtaining historical discharge power through a new energy vehicle control terminal; and obtaining historical driving environment state through a high-precision map carried by the new energy vehicle control; performing filtering and smoothing processing on the historical discharge power to obtain a power fluctuation smoothing curve;

[0018] Step S2: performing start-stop power mapping of the power fluctuation smoothing curve under a road half-slope state according to the historical driving environment state to obtain half-slope start-stop power mapping data; performing energy consumption increment behavior learning on the half-slope start-stop power mapping data to generate an energy consumption progression rule;

[0019] Step S3: designing a residual endurance prediction architecture according to the energy consumption progression rule to obtain a residual endurance prediction architecture; and sending the residual endurance prediction architecture to a new energy vehicle terminal to perform battery management for the new energy vehicle.

[0020] In the embodiment of the present application, reference is made to Figure 1 The above is a step flowchart of the battery management method for the new energy vehicle, and in the present example, the battery management method for the new energy vehicle comprises the following steps:

[0021] Step S1: obtaining historical discharge power through a new energy vehicle control terminal; and obtaining historical driving environment state through a high-precision map carried by the new energy vehicle control; performing filtering and smoothing processing on the historical discharge power to obtain a power fluctuation smoothing curve;

[0022] In the embodiment of the application, the battery management module in the new energy vehicle control terminal collects historical discharge power data of the power battery under different driving conditions. The data includes the product of current and voltage obtained at a sampling frequency of 1 Hz within the interval of 0s to 3600s, the discharge power range is between 0kW and 150kW, the original power curve obtained by sampling has noise and missing points, in order to ensure the stability of subsequent processing, the missing points are filled by linear interpolation, in the case of missing three consecutive points, the missing points are filled by cubic spline interpolation, a complete power fluctuation curve is obtained, then the power fluctuation curve is smoothed by moving average filtering with a sliding window width of 15 points, the boundary points are extended by mirror image to ensure that the edge of the smoothed curve is not distorted, the high frequency fluctuation amplitude of the filtered power fluctuation is reduced from an average of 4kW to 1kW, the smoothed curve output by filtering is used as the input data for subsequent energy consumption feature analysis, and the historical driving environment state is obtained through the vehicle controller and the high-precision map data interface, the high-precision map contains road slope information with a sampling accuracy of 1m and a range of-12° to +12°, and traffic flow sampling results in each historical time period, the traffic flow data is obtained by counting the number of vehicles passing per minute, the range is 0 vehicles per minute to 80 vehicles per minute, after aligning with the power fluctuation smoothing curve by time stamp, complete driving environment and power coupling data are formed, which prepares for subsequent half-slope start-stop power mapping.

[0023] Step S2: mapping the power fluctuation smoothing curve to the start-stop power under the road half-slope state according to the historical driving environment state to obtain half-slope start-stop power mapping data; performing energy consumption increment behavior learning on the half-slope start-stop power mapping data to generate an energy consumption progressive rule;

[0024] In the embodiment of the present application, according to the road geometry and traffic flow data extracted from the historical driving environment state, the power fluctuation smooth curve is cut off at the power output segment when the slope value is in the interval of 2° to 6° and the traffic flow is greater than 40 vehicles per minute, and a power distribution set under the semi-slope start-stop state is constructed. In the power distribution set, the threshold segmentation method is used to extract the rapid transition process of the power from 0kW to more than 20kW when the vehicle switches from start to stop, and the time interval and peak value of adjacent transitions are recorded to obtain semi-slope start-stop power mapping data. The data is expressed in the form of a two-dimensional mapping table, where one dimension is the slope angle sampling accuracy of 0.1°, and the other dimension is the traffic flow sampling interval of 10 vehicles per minute. In the mapping table, any combination of slope and traffic flow can correspond to a start-stop power distribution sequence. The distribution sequence is input into the energy consumption increment behavior learning process. The behavior learning does not depend on the neural network model, but calculates the energy consumption growth rate of different power segments during semi-slope start-stop through the gradient incremental statistical method. In this process, the energy consumption progression law is obtained by hourly statistics of energy consumption growth rate. This law can accurately describe the energy accumulation mode of the vehicle in the semi-slope start-stop scene and is represented in the form of a numerical curve. The cumulative energy growth rate generated by the energy consumption progression law curve within the time span of 0s to 1800s gradually evolves from an initial value of 0.2% per second to a later value of 0.05% per second.

[0025] Step S3: according to the energy consumption progression law, a battery residual endurance prediction architecture is designed to obtain a residual endurance prediction architecture; and the residual endurance prediction architecture is sent to a new energy vehicle terminal to perform battery management for the new energy vehicle.

[0026] In the embodiment of the present application, the energy consumption progression law obtained in the previous stage is input into the battery residual endurance prediction architecture design process. First, the gradient segment in the energy consumption progression law curve is feature extracted, and the feature parameters formed by the energy consumption in different slope intervals, including time dimension cumulative energy growth rate, gradient decline trend value, and power peak occurrence frequency, are taken as input feature sets. Then, a quantitative relationship between battery energy consumption and residual capacity is established through a prediction structure based on stepwise regression algorithm. In this process, the mapping relationship between the residual energy of the actual power battery with a capacity of 50kWh and the integral value of the energy consumption progression law curve is established. After regression fitting, it is predicted that in the simulation scenario, when the initial battery capacity is 30kWh, the residual endurance prediction value is about 95km under the semi-slope start-stop environment with an average slope of 4° and an average traffic flow of 60 vehicles per minute. Subsequently, the obtained residual endurance prediction architecture is written into the new energy vehicle terminal control module in the form of a data structure, and the control module performs residual endurance management operation according to the prediction architecture, thereby completing the complete implementation of the new energy vehicle battery management system.

[0027] Step S1 includes the following steps:

[0028] Step S11: obtaining historical discharge power through the new energy vehicle control terminal; and obtaining historical driving environment state through the high-precision map carried by the new energy vehicle control;

[0029] Step S12: drawing a fluctuation curve of the historical discharge power, and marking a time stamp to construct a power fluctuation curve;

[0030] Step S13: filling missing values of the power fluctuation curve to obtain a power fluctuation filling curve;

[0031] Step S14: performing filtering and smoothing processing on the power fluctuation filling curve to obtain a power fluctuation smoothing curve.

[0032] In the embodiment of the application, the battery management module built in the new energy vehicle control terminal collects historical discharge power data of the electric vehicle during operation, and the collection mode is to obtain instantaneous power value by multiplying voltage value and current value at a sampling frequency of 1 Hz, wherein the current sampling range is 0A to 500A, the voltage sampling range is 200V to 800V, the power value range is finally between 0kW to 150kW, the cumulative collection time is 0s to 3600s, a total of 3600 data points, at the same time of collection, the data is stored with time stamp information for subsequent processing, at the same time, the driving environment state is obtained through the high-precision map associated with the control terminal, the road data provided by the high-precision map includes road slope information with an accuracy of 0.1°, the road transverse curvature radius range is between 30m to 5000m, the longitudinal slope range is between -12° to 12°, and the road traffic flow information is also obtained, the flow is obtained by counting the number of vehicles passing through the detection area per unit time, the statistical frequency is 60s once, the traffic flow effective range is 0 vehicle per minute to 80 vehicle per minute, the historical driving environment state and the historical discharge data are corresponded through time stamp to form a record set, providing basic data for the next data processing link.

[0033] After the historical discharge power data collection and the time stamp synchronization of the driving environment state are completed, the power value obtained according to the 1Hz sampling period is subjected to fluctuation curve drawing. First, the power value is arranged in a two-dimensional coordinate system with the horizontal axis as time and the vertical axis as power according to the sampling sequence. The change curve of the historical power output is drawn with a time interval of 1s. The curve nodes are marked with accurate time stamps, for example, the 0s to 3600s have corresponding power point data. The original power fluctuation curve is formed in a stable time sequence. The fluctuation curve can completely reflect the dynamic change position of the battery discharge process in the experimental scene. The power fluctuation curve formed through the process contains the power change of the vehicle in the start-stop process under different slope conditions when the longitudinal fluctuation range is large. The time stamp marking ensures that the driving environment data at the same time point is synchronized to the power curve. Therefore, the entire curve not only reflects the discharge change but also one-to-one corresponds to the external environment state, thereby providing a prerequisite for missing value processing.

[0034] The power fluctuation curve constructed by the time stamp marking is subjected to missing value filling in the case of data loss in the sampling. The missing value determination rule is that if the difference between the continuous sampling points and the front and rear power exceeds the limited threshold of 50kW and the corresponding position is a null value, it is defined as a missing point. In the experimental data, the missing rate is about 0.8% of the total data amount. For the single-point missing condition, the linear interpolation method is used to determine the interpolation value through the power difference between the front and rear two sampling points. For the continuous 2-point missing condition, the quadratic interpolation method is used to obtain the interpolation value through the power trend of the adjacent 3 points. For the continuous 3-point to 5-point missing interval, the cubic spline interpolation is used to ensure the continuity and smoothness of the curve. Under the sampling conditions of the experiment, the new power data sequence is obtained after the interpolation is completed, and the power fluctuation filling curve is formed again according to the time stamp. The filled curve does not have blank points in the entire sampling period, and the overall trend of the observed data is consistent with the original curve, with balanced transition and naturalness. The fluctuation characteristics can be completely retained.

[0035] After the power fluctuation filling curve is constructed, in order to further reduce high-frequency noise interference and highlight long-term trends, the filling curve is filtered and smoothed, and the filtering method adopted is a processing method based on the combination of moving average filtering and low-pass filtering. Firstly, in the moving average process, the sliding window width is set to 15 points, that is, 15s time range, the current sampling point power and the power of the previous and subsequent 7 points are averaged to replace the current value. This method can smooth the power curve in the instantaneous sharp fluctuation, ensure the maximum trend signal, and then further filter the high-frequency part of the curve by a first-order low-pass digital filter with a cutoff frequency of 0.2Hz in the output sequence. In this process, the filter adopts Butterworth function coefficients to ensure the smoothness of the passband, and finally the power fluctuation smoothing curve is obtained. The curve is overall gentle in the 0s to 3600s sampling interval and the short-time jitter is obviously reduced. The peak point is reduced from the average adjacent point difference of 3kW before filtering to 1kW. The output result is used as the input data for subsequent half-slope start-stop mapping analysis.

[0036] Step S2 comprises the following steps:

[0037] Step S21: extracting road geometry and traffic flow in the historical driving environment state;

[0038] Step S22: mapping the power fluctuation smoothing curve according to the traffic flow and the road geometry in the road half-slope state to obtain half-slope start-stop power mapping data;

[0039] Step S23: performing energy increment behavior learning on the half-slope start-stop power mapping data to output energy increment learning data in the half-slope start-stop state;

[0040] Step S24: quantifying the half-slope start-stop energy progressive law under the time dimension promotion of the energy increment learning data to generate an energy progressive law.

[0041] As an example of the present application, reference is made to Figure 2 In this example, the step S2 comprises:

[0042] Step S21: extracting road geometry and traffic flow in the historical driving environment state;

[0043] In the embodiment of the application, after the historical driving environment state is acquired, the road geometric shape information therein is first extracted, and the extraction process includes reading slope data, radius of curvature data, road longitudinal length data and road segment identification information from the high-precision map. In the test example, the total length of the road involved is 12km, the road segment accuracy is 1m, each unit segment has a slope value and a radius of curvature value, the slope range is-6° to +8°, the radius of curvature range is 45m to 3500m, the traffic flow extraction adopts a time synchronization method to align the data with a sampling frequency of 1 minute to the time stamp corresponding to the power data, the traffic flow data of each road segment is counted by the number of vehicles passing through, and the traffic flow value ranges from 0 vehicles per minute to 75 vehicles per minute. When the extracted road geometric shape and traffic flow are matched, a two-dimensional array structure is used for storage, the first dimension is the road segment number, and the second dimension includes the slope value and the traffic flow value of the corresponding segment. This process ensures that each road segment carries the corresponding driving environment state, forming a data set that can be directly coupled with the power fluctuation smooth curve, providing a basic input for subsequent semi-slope start-stop power mapping.

[0044] Step S22: mapping the power fluctuation smooth curve to the start-stop power in the road semi-slope state according to the traffic flow and the road geometric shape, to obtain semi-slope start-stop power mapping data;

[0045] In the embodiment of the application, after the road geometric shape and traffic flow are extracted, the power fluctuation smooth curve is mapped to the corresponding road slope and traffic flow. The mapping logic is that when the slope angle is in the interval of 2° to 6° and the traffic flow is greater than or equal to 40 vehicles per minute, the power value at the corresponding time point is divided into the semi-slope start-stop power segment set. In the experimental sampling, a total of 162 start-stop processes are obtained, each start-stop process is defined as a starting action from the power 0kW rising to more than 20kW, and a stopping action from the power 10kW falling to 0kW. The time difference and amplitude of these power changes are recorded as start-stop mapping points to generate semi-slope start-stop power mapping data. The data is presented in a three-dimensional mapping table, the first dimension is the slope sampling value range 2° to 6° with an accuracy of 0.1°, the second dimension is the traffic flow statistical value interval with every 10 vehicles per minute as a gear, and the third dimension is the start-stop power change data sequence, which includes the starting power peak value, the stopping power valley value, the start-stop process duration and other parameters. The mapping data can accurately depict the power output characteristics of the road segment semi-slope start-stop, and ensure the integrity of the input information in the subsequent energy consumption learning stage.

[0046] Step S23: energy consumption increment behavior learning on the semi-slope start-stop power mapping data, to output energy consumption increment learning data in the semi-slope start-stop state;

[0047] In the embodiment of the present application, the obtained semi-slope start-stop power mapping data is subjected to energy increment behavior learning, and the behavior learning is carried out in a way based on step-by-step energy increment statistics and differential analysis, and the specific process is that the power in each start-stop mapping segment is integrated with respect to time to obtain an energy value, and the energy value is divided by the time duration to calculate the energy consumption change rate per unit time, and the energy consumption change rate per unit time is recorded in the energy increment sequence. In the actual test data, the average energy consumption interval of each start-stop process is between 0.25kWh and 0.9kWh. By comparing the energy increment sequences under different slope and traffic conditions, the energy increment amplitude between adjacent data is calculated by the differential method, and the growth direction and the absolute value of the growth amplitude are recorded. After arranging these differential results, the energy increment learning data under the semi-slope start-stop scene can be obtained. In the whole process, the segmented statistics method is used to classify and statistics according to every 0.5° of slope and every 10 vehicles per minute of traffic flow, so that the energy increment results present a multi-dimensional data set structure. This structure ensures that the learning data can truly reflect the energy increment characteristics under the environmental conditions.

[0048] In another embodiment, after obtaining the semi-slope start-stop power mapping data, first, differential operation is performed on the power mapping curve to calculate the power change between adjacent time points to obtain a power change sequence. Then, the power change sequence is synchronized with the vehicle speed sequence and the slope change sequence, and the energy increment per unit time under the semi-slope start-stop condition is calculated by using the weighted cumulative method. The non-linear relationship between the energy accumulation value and the power change is gradually calculated within a 600s time period by using the window sliding method. In this process, the overlapping window method is introduced to ensure data smoothness. For example, in the interval of 10% slope and 0.12 vehicles per second flow, the cumulative energy of the vehicle within 600s is 4.5kWh, and the total instantaneous power fluctuation obtained by differential is 5.2kW. Then, the correlation analysis is performed on the two, and the energy increment learning data is generated. This data contains the influence degree of power fluctuation on energy accumulation.

[0049] Step S24: The energy increment learning data is subjected to semi-slope start-stop energy progressive law quantification under the time dimension push, and the energy progressive law is generated.

[0050] In the embodiment of the present application, after obtaining the energy consumption increment learning data in the half-slope start-stop state, the energy consumption increment learning data is placed in the time dimension for progressive rule quantization processing. First, the energy consumption learning data is arranged in time sequence, and the energy consumption increment in each 30s window in the interval of 0s to 1800s is segmented and counted. The average, peak value and variance of the energy consumption increment in the window are calculated. The parameters of each window are used to generate a new time sequence. Then, the time sequence is subjected to mutation point detection. The mutation point detection adopts a judgment rule based on the ratio of sliding standard deviation to mean value. When the difference between the mean value of the window and the adjacent window exceeds 5%, it is counted as a mutation. The detection result forms a stage marker of energy consumption growth. Then, the stage marker sequence is subjected to curve convergence processing. The exponential smoothing method is used to reduce the continuity of mutation and extract the progressive trend. Finally, the energy consumption progressive rule is generated. The rule is expressed in the form of a numerical sequence. Specifically, the energy consumption increment gradually progresses from the initial average value of 0.015kWh every 30s to 0.005kWh every 30s in the total time of 1800s. The sequence provides a quantitative expression of the energy consumption change of the half-slope start-stop under the time dimension promotion, which can be used as an input condition for the subsequent endurance prediction architecture design.

[0051] In another embodiment, after obtaining the energy consumption increment learning data, the energy consumption progressive rule quantization processing under the time dimension promotion is performed. First, the energy consumption increment learning data is divided into multiple periods according to time sequence, and each period has a fixed length of 900s. The average, variance and maximum fluctuation amplitude of the energy consumption increment change are calculated in each period. Then, the time series trend analysis method is used to smooth the energy consumption increment change through the methods of moving average and exponential weighted moving average. Then, the change trend of energy consumption with time is extracted by using the piecewise linear fitting method. For example, in the section with a slope of 12% and a traffic flow of 0.15 vehicles per second, the average value of the energy consumption increment in the 900s period is 5.3kWh, the variance is 0.9, and the maximum fluctuation amplitude is 2.1kWh. Then, these quantitative characteristics are aggregated to generate the energy consumption progressive rule. The rule data clearly indicates the progressive change characteristics of energy consumption with time under different slope and flow conditions, providing basic data for the subsequent battery remaining endurance prediction architecture.

[0052] Step S23 includes the following steps:

[0053] The half-slope start-stop frequency and half-slope inclination change in the half-slope start-stop power mapping data are extracted; and vehicle load information is obtained;

[0054] The gravity potential energy change variance is calculated according to the vehicle load information and the half-slope inclination change in the half-slope start-stop state, so as to obtain the gravity potential energy change variance;

[0055] Based on the gravity potential energy change variance, the instantaneous output energy consumption increment index is obtained by performing instantaneous output energy consumption increment index analysis on the gravity change correlation in the half-slope start-stop power mapping data.

[0056] According to the half-slope start-stop frequency and the instantaneous output energy increment index, the additional loss of the motor driving current is derived, and additional loss data is obtained.

[0057] Based on the additional loss data, the energy increment behavior learning is carried out, and the energy increment learning data in the half-slope start-stop state is output.

[0058] In the embodiment of the application, the specific embodiment of extracting the half-slope start-stop frequency and the half-slope inclination change in the half-slope start-stop power mapping data is that all the segmented data with the slope angle between 2° and 6° in the constructed half-slope start-stop power mapping data is searched, the process of rapidly increasing the power from 0kW to more than 20kW is defined as one starting action, the process of decreasing the power from 10kW to 0kW is defined as one stopping action, the power fluctuation smoothing data in the 3600s sampling period is traversed and counted point by point, and the half-slope start-stop times 162 are accumulated, the start-stop frequency range is between 2 times and 5 times per minute, the real-time slope value corresponding to each start-stop action occurrence point is extracted, the minimum value of the slope fluctuation range is 2.1°, the maximum value is 5.8°, the slope change sequence is used to represent the slope fluctuation characteristics in the whole time dimension, and the half-slope inclination change is obtained by the slope difference between adjacent start-stop actions, the average 0.6° is the main fluctuation amplitude, and the maximum slope difference is 1.2°.

[0059] The specific embodiment of obtaining the vehicle load information is that the full load weight of the test vehicle and the actual weight of the current test working condition are recorded in the same statistical period, the vehicle empty weight is 1650kg, the additional load in the test scene is 350kg, and the total test load is 2000kg, the load parameter is introduced as a fixed input into the subsequent potential energy calculation link, and it is ensured that the load value remains unchanged in each time segment.

[0060] The embodiment of calculating the gravity potential energy change variance in the half-slope start-stop state according to the vehicle load information and the half-slope inclination change is that the corresponding slope change sequence and the corresponding load parameter are taken in each minute window, the road longitudinal height difference corresponding to the segmented slope is calculated, and the gravity potential energy change amount of the vehicle in the slope segment is obtained, the average slope conversion height difference is 0.7m to 2.1m in the test, and the gravity potential energy change amount ranges from about 13730J to 41190J, the slope change amount in each minute is taken as a group of samples, the variance value of the samples is calculated as the gravity potential energy change variance, the variance measures the energy fluctuation stability generated by the slope fluctuation in each start-stop process, and the average value of the variance in the example data is , the maximum value reaches , wherein The square of the energy represents the degree of dispersion or fluctuation of the energy change. Based on the variance of the gravitational potential energy change, the specific embodiment of the instantaneous output energy consumption increment index analysis between the correlation of the gravity change and the half-slope start-stop power mapping data is to correlate the calculated gravitational potential energy change variance with the corresponding power mapping sequence point by point. Each start-stop process is based on a time segment. The instantaneous energy consumption sequence calculated by power integration is standardized with the potential energy change variance. Then, the gradient change rate of the standardized result is calculated along the time axis. The rate is defined as the instantaneous output energy consumption increment index, with a numerical range of 0.02 to 0.15 and an average of 0.08. The instantaneous output energy consumption increment index curve is formed, which shows the coupling relationship between the instantaneous power demand and the gravitational potential energy change under the condition of high-frequency half-slope start-stop. This provides quantitative input for subsequent energy consumption increment behavior learning.

[0061] Under the premise that the half-slope start-stop power mapping data and the instantaneous output energy consumption increment index have been obtained, the start-stop frequency in the entire sampling period is first extracted. In the experimental data, the total sampling time is 3600s, and the start-stop action is counted 162 times. The average start-stop frequency per minute is 2.7 times, and the highest frequency interval reaches 5 times per minute. This start-stop frequency data is input into the motor current state mapping to evaluate the occurrence of high-load current. The instantaneous output energy consumption increment index is used to measure the additional impact of power growth on motor current during each start-stop process. When the energy consumption increment index value reaches 0.12, the motor current peak value appears between 420A and 450A, while the normal stable running interval motor current is about 180A to 220A. Therefore, the increase of instantaneous current is triggered by start-stop action. The start-stop frequency and the increment index are used to jointly construct the additional loss of motor drive current. The specific method is to square the part of each start-stop peak current exceeding the steady-state value and multiply it by the duration to obtain the additional power loss caused by current fluctuation. Then, the loss is summed for all start-stop processes. For example, in the experimental data, the cumulative additional loss of 1.9kWh is obtained in 162 start-stop processes, which is equivalent to 11.7Wh additional loss per start-stop on average. This loss data is output in the form of an additional loss sequence, providing input for the next step of energy consumption increment learning.

[0062] The additional current loss sequence is statistically learned, first, the additional loss data is arranged in time sequence to obtain the time cumulative increment curve from 0s to 3600s, the curve shows a larger slope peak in the early 500s, the additional loss growth rate of each start-stop process is between 0.04Wh and 0.06Wh per second, and in the later period, the curve tends to gradually flatten, and the growth rate is reduced to 0.01Wh per second to 0.02Wh per second, in order to more clearly capture the additional energy consumption accumulation process, the difference between the additional loss of each sampling point and the previous sampling point is recorded, and the loss gradient sequence is constructed, which reflects the phased change of the energy consumption behavior under the start-stop cluster, and a sliding window width of 60s is used to locally average the gradient sequence, and a more stable energy consumption increment trend is obtained, and on this basis, the energy consumption increment learning data is formed, which will show different structures under different slope intervals and traffic flow intervals, for example, under the condition that the slope angle is 4° and the traffic flow is 60 vehicles per minute, the energy consumption increment learning data reflects that the average additional energy consumption of each start-stop accounts for 12% of the baseline energy consumption, and under the condition that the slope angle is 2.5° and the traffic flow is 40 vehicles per minute, the proportion decreases to 7%, finally, the energy consumption increment learning data completely records the coupling relationship between the additional motor driving current loss and the external environmental conditions, and serves as the basis input of the subsequent energy consumption progressive law quantization and endurance prediction architecture.

[0063] The instantaneous output energy consumption increment index analysis includes:

[0064] The slope gravity resistance component and the slope inertia force component in different stages are calculated based on the variance of the change of gravitational potential energy.

[0065] The instantaneous driving force required by the vehicle is analyzed according to the slope gravity resistance component and the slope inertia force component.

[0066] The driving force instantaneous power correlation data is obtained by coupling the driving force instantaneous power of the half-slope start-stop power mapping data;

[0067] The instantaneous time sequence integral power is obtained by time sequence integration of the driving force instantaneous power correlation data.

[0068] The instantaneous output energy consumption increment index is obtained by analyzing the instantaneous output energy consumption increment index between the gravitational change correlation according to the instantaneous time sequence integral power.

[0069] In the half-slope start-stop scene in the embodiment of the application, the slope gravity resistance component and the slope inertia force component need to be derived based on the variance of the change of gravitational potential energy, the total mass of the experimental vehicle is set to 2000kg, the slope angle of each slope section is provided by a high-precision map, and the resolution is 0.1°, the slope value is extracted in a sampling period of 1s , first calculate the gravity resistance component of the ramp, the method is to carry out point by point operation of the vehicle gravity and the sine function of the slope angle, the specific process is to take the vehicle gravity G = m x g, wherein m = 2000 kg, g = 9.8 m / s², G = 19600 N is obtained, then calculate =G×sin , when =2°, sin2°≈0.0349 is taken, and ≈19600N×0.0349≈684N, when =6°, sin6°≈0.1045 is taken, and ≈19600N×0.1045≈2046N, so the gravity resistance component of the whole section is obtained by point by point operation, for the calculation of the inertia force component of the ramp, first obtain the acceleration value a = (v(t+1)-v(t)) / Δt by dividing the difference between adjacent velocity data points by the time interval Δt = 1 s, take the vehicle mass 2000 kg multiplied by the acceleration to obtain the instantaneous inertia force =m×a, when v(t) = 0 m / s and v(t+1) = 0.8 m / s, the acceleration a = 0.8 m / s², =1600 N, when the speed rises faster, for example from 0 m / s to 1.1 m / s, the acceleration a = 1.1 m / s², =2200 N, store all the sampling points of , that is, the inertia force component of the vehicle on the ramp is obtained.

[0070] According to the obtained gravity resistance component Fg and inertia force component Fi of the ramp, the instantaneous driving force of the vehicle is calculated point by point, the method is to add them up at each time to obtain Fd = Fg + Fi, the calculation process uses 1 s as the time window, and the instantaneous driving force curve is formed by accumulating point by point in the whole process of each start-stop segment, in the example, when the slope =4°, Fg = 19600 x sin4° ≈ 1366 N, when the acceleration a = 0.6 m / s², Fi = 1200 N, then the instantaneous driving force Fd = 1366 N + 1200 N = 2566 N, store the Fd values of all time segments according to the 162 start-stop segments respectively to obtain the complete instantaneous driving force time sequence, in the experimental data, the minimum driving force is observed to be 950 N, and the maximum driving force reaches 3500 N, which ensures that the mechanical requirements under different start-stop conditions are covered.

[0071] After getting the instantaneous driving force curve, it needs to be associated with the half-slope start-stop power mapping data to calculate the instantaneous power corresponding to the driving force. The method is to multiply the instantaneous driving force Fd and the instantaneous speed v collected synchronously point by point, that is, P = Fd x v. When v = 3 m / s and Fd = 1800 N, then the instantaneous power P = 1800 N x 3 m / s = 5400 W = 5.4 kW. When the speed rises to 5 m / s and the driving force is 2650 N, then P = 2650 N x 5 m / s = 13250 W = 13.25 kW. All time points are calculated one by one to get the driving force instantaneous power sequence, and then aligned with the power fluctuation smoothing curve by time stamp. Each sampling point has the actual speed and driving force to calculate the power value. The coupling result is the driving force instantaneous power association data. This data is arranged as a time segmented matrix. The number of columns is the time point, and the number of rows is the different start-stop segments to ensure that the power data can completely correspond to the mechanical input.

[0072] After the construction of the driving force instantaneous power association data is completed, the time sequence integration is performed on each start-stop segment to obtain the instantaneous time sequence integration power. The integration process is to accumulate the power value P(t) on the time sequence at each start-stop. The integration step is Δt = 1 s. The process is In the example data, the driving force instantaneous power sequence of a certain start-stop is 5 kW, 7 kW, 10 kW, 12 kW, and 9 kW, which lasts for 5 s. The cumulative power consumption is 5 kW x 1 s + 7 kW x 1 s + 10 kW x 1 s + 12 kW x 1 s + 9 kW x 1 s = 43 kWs = 43000 J. This value represents the total energy consumed by the start-stop action. The energy consumption range of all 162 start-stop segments is 20 kJ to 65 kJ, with an average of 41 kJ. The instantaneous time sequence integration power is stored in the form of sequence data according to the time index to form a complete instantaneous time sequence integration power.

[0073] After the time sequence integration power set is formed, it is associated with the calculated gravity potential energy variance to obtain the instantaneous output energy consumption increment index. The specific method is to normalize the ratio of the integral power E of each start-stop to the corresponding gravity potential energy variance , and I = E / , so that the unit is converted to a dimensionless quantitative index. Then, the first-order difference of the time dimension sequence I is calculated, that is, ΔI = I(t+1) - I(t), to observe the sensitivity of power demand to slope potential fluctuation. Finally, the difference sequence is averaged through a sliding window with a window length of 5 start-stops to obtain the smoothed instantaneous energy consumption increment index. In the example observation, when the slope is 5.8°, the integral energy is about 63 kJ, and the gravity potential energy variance is about , the corresponding increment index is 63 / ≈0.012, the continuous start-stop difference obtains the maximum value 0.15, the minimum value 0.02, and the average 0.08, the index completely describes the intensity relationship between instantaneous energy consumption and gravity fluctuation under the start-stop action, and finally forms the instantaneous output energy consumption increment index sequence, which provides direct input for motor additional loss derivation.

[0074] Wherein the motor driving current additional loss derivation includes:

[0075] The motor high load current state is analyzed based on the half-slope start-stop frequency;

[0076] The high-rate discharge state of the battery is analyzed based on the motor high load current state, and the high-rate discharge state is obtained.

[0077] According to the heat square level growth trend of the battery internal resistance;

[0078] Based on the heat square level growth trend and the instantaneous output energy consumption increment index, the heat loss regression analysis is carried out, and then the energy utilization loss gradient is quantified.

[0079] According to the energy utilization loss gradient and the instantaneous output energy consumption increment index, the motor driving current additional loss derivation is carried out, and the additional loss data is obtained.

[0080] In the embodiment of the application, the half-slope start-stop frequency is counted hour by hour, and the motor current data corresponding to the start-stop segment is synchronously collected, so as to analyze the high load current state of the motor. The vehicle is set to run in a scene with a slope interval of 2° to 6° for 3600s, and a total of 162 start-stop events occur, with an average start-stop frequency of 2.7 times per minute, and the highest frequency reaches 5 times per minute. When analyzing the motor current of each start-stop segment, the steady-state driving current ranges from 180A to 220A, and if the current peak value in a single start-stop process is higher than 300A, it is defined as the motor high load current state. Further analysis found that among the 162 start-stop events, the current peak value of 118 events exceeded 350A, of which 27 events reached more than 400A, and the highest peak value appeared in the case of a slope of 5.8°, with a current instantaneous value of 450A, which lasted for 2.5s to 4.5s. In order to ensure the accuracy of the statistics, the current sequence obtained by the sensor is first sampled at 1Hz, and then the start-stop interval is intercepted successively, and the current peak value and duration in the interval are calculated. The analysis results show that the total cumulative time of the high load current state is about 390s, accounting for 10.8% of the total running time of 3600s. The high load current sequence is extracted separately and indexed by time to obtain the motor high load current state data set, which provides physical input conditions for subsequent analysis of high-rate discharge behavior, and ensures that the coupling form of start-stop frequency and current fluctuation under different slope scenarios is quantitatively recorded.

[0081] After obtaining the high load current state of the motor, the high rate discharge state of the battery is quantitatively analyzed. The nominal capacity of the experimental vehicle's power battery pack is 50 Ah. When the motor appears high load current, the instantaneous discharge rate of the battery needs to be calculated. The formula is Discharge rate = current / rated capacity. For example, when the current is 350 A, the rate is 7C, and when the current is 450 A, the rate is 9C. The discharge rate values of 118 high load segments are calculated respectively. In the entire experimental data, 7C state appears 59 times, 8C state appears 28 times, and 9C state appears 31 times. The maximum duration of 6s small section remains at 9C discharge level. In order to ensure the accuracy of the results, all current values are directly converted using the original sampling signal of the sensor, and then each start-stop segment is associated with a timestamp to form a discharge rate sequence. The sequence length is consistent with the number of high load segments, and each entry contains four indicators: start time, duration, current peak, and discharge rate. Further statistics on energy output show that under 7C rate conditions, the unit start-stop energy release is about 12Wh to 18Wh, and under 9C rate conditions, the energy release range rises to 20Wh to 28Wh. The analysis results effectively correspond the high load current to the high rate state inside the battery, realize the mapping of the motor working condition to the battery discharge characteristics, and lay a data foundation for the subsequent heat growth trend research.

[0082] After completing the determination of the high rate discharge state of the battery, the internal resistance thermal effect of the battery is analyzed in detail, and the heat square level growth trend caused by the internal resistance under high rate state is determined. The experimental condition is that the average value of the internal resistance of the power battery pack is 2.5mΩ. When calculating the thermal power, the formula P = I² × R is used, where I is the sampling current and R is the internal resistance. The current in all high rate discharge segments is substituted point by point into the calculation, and the results show that when the current is 350A, the instantaneous thermal power is 350 × 350 × 0.0025 = 306W, when the current is 400A, the instantaneous thermal power is 400 × 400 × 0.0025 = 400W, and when the current is 450A, the thermal power reaches 450 × 450 × 0.0025 = 506W, showing a strict square relationship trend. In the 162 start-stop segments, the thermal power is statistically analyzed, and the thermal power range is concentrated between 255W and 506W, and most of the intervals are between 300W and 450W. Further combined with the time window, the cumulative heat energy value of each start-stop is obtained. For example, in a high load segment of 4s, the current is about 350A, the average thermal power is 306W, and the cumulative heat energy is about 1224J. In a 450A segment of 6s, the cumulative heat energy is more than 3000J. Then the heat energy sequence of all segments is fitted, and the fitting result verifies that the I²R rule conforms to the current square level growth characteristics. Through this result, a quantitative data set of battery heat growth under high rate conditions is established, which is used to support the subsequent energy utilization rate loss gradient and additional loss derivation, and ensures that the influence of thermal loss is correctly quantified.

[0083] After the calculation of the battery high-rate discharge state and the square level growth trend of the battery internal resistance heat power, in order to further quantify its influence on energy utilization rate, it is necessary to establish the regression relationship between heat power and instantaneous output energy consumption increment index. First, the heat power sequence obtained by experiment is standardized. The minimum value of heat power in the experimental sampling point is 225W, and the maximum value is 506W. The maximum and minimum normalization method is used to map it to the interval of 0 to 1, and the normalized heat power sequence is obtained. Similarly, the instantaneous output energy consumption increment index is normalized. The original value is in the interval of 0.02 to 0.15, and it is converted to the range of 0 to 1 accordingly. Then the normalized heat power and the normalized index are input as variables, and the output target is defined as the relative proportion of battery heat loss in total output power, that is, the energy utilization rate loss rate. In the experimental segment, when the motor power is 10kW, the heat power loss rate is 0.03 when the heat power is 300W, and when the motor power is 20kW, the heat power loss rate is 0.0225 when the heat power is 450W. After all the sample distribution points are plotted on the two-dimensional coordinate plane, the least square regression is used to fit the relationship straight line of energy loss rate and input double variables, and the fitting residual mean square error converges within 0.002. The loss rate gradient result obtained by experiment shows a trend of increasing with the increase of instantaneous output energy consumption increment index, for example, the loss rate near the index 0.02 is 0.021 on average, the loss rate near the index 0.10 is 0.028 on average, and the loss rate in the interval of index 0.12 to 0.15 is further increased to 0.031. The trend shows that in the high-rate state, the heat square growth and the energy consumption increment effect superimpose and amplify the overall system energy utilization rate decline, and finally form the energy utilization rate loss gradient data sequence, which provides quantitative input basis for subsequent loss derivation.

[0084] After the energy utilization loss gradient data is obtained, the gradient needs to be coupled with the time integral power to output the motor drive current additional loss data. First, the time integral power of each start-stop segment is counted, and the energy range of the 162 start-stop segments in the experimental results is distributed between 20 kJ and 65 kJ, with an average of 41 kJ. The energy utilization loss gradient corresponding to the segment is multiplied point by point with the time integral power of the segment to obtain the additional energy consumption value, the formula is ΔE=gradient x integral power. For example, in a certain segment, the time integral power is 42 kJ, and the energy utilization loss gradient is 0.028, then the additional loss ΔE of the segment is 1176 J. In another segment, the time integral power is 60 kJ, and the gradient is 0.031, then the additional loss is 1860 J. The total value of the additional loss of the entire sampling period is about 1.95 MJ by accumulating the calculation results of all segments, and the average value of the additional loss of a single start-stop is about 12 kJ by equally dividing the cumulative results. On this basis, the energy additional loss sequence is formed with the timestamp as the index, each data in the sequence contains three fields of time integral power, energy utilization loss gradient and additional loss value, which completely retains the dissipation dynamics of the actual process. The additional loss sequence is sent as a new input into the energy consumption increment behavior learning unit, so that the learning result not only contains the cumulative amount of original power consumption, but also includes the additional loss effect induced by the current square trend in the high load state, so as to ensure that the new energy vehicle battery management method has more accurate energy evaluation basis in the subsequent endurance prediction link.

[0085] Step S24 includes the following steps:

[0086] Analyzing the non-linear energy consumption increment evolution data in the energy consumption increment learning data under the time dimension driving;

[0087] Performing unordered peak mutation analysis on the non-linear energy consumption increment evolution data to obtain increment unordered peak mutation data;

[0088] Performing mutation peak jump convergence on the non-linear energy consumption increment evolution data according to the increment unordered peak mutation data to obtain peak jump convergence data;

[0089] Quantifying the non-linear energy consumption increment evolution data under the time dimension driving according to the Logistic algorithm and the peak jump convergence data to generate energy consumption progressive law.

[0090] In the embodiment of the present application, after the energy consumption increment learning data is formed, the evolution trend thereof in the time dimension is analyzed non-linearly, the time interval 0s to 3600s is divided into a window of 60s, a total of 60 time period data points are obtained, and the average energy consumption increment of each time period is used to construct an energy consumption evolution sequence. In the sequence, there is an obvious nonlinear feature, which is that the growth rate in the early stage is greater than that in the later stage. The average increment rate in the interval 0s to 600s is 0.045 Wh / s, and the average increment rate in the interval 3000s to 3600s is only 0.012 Wh / s. In order to further depict the nonlinear evolution feature, a segmented gradient calculation method is used. In each 60s window, the local gradient change rate is obtained by the difference method, and finally the nonlinear energy consumption increment evolution data of the half-slope start-stop is formed. The data takes time as the horizontal coordinate and the energy consumption change rate per unit time as the vertical coordinate, and clearly reflects the non-stationary distribution rule of vehicle energy consumption change with time. After obtaining the nonlinear energy consumption increment evolution data, the peak value is analyzed for disorder mutation. First, the full sequence is searched for peak values. When a local data point is greater than 5% of the adjacent points, it is defined as a peak point. The statistical result shows that there are 47 peak values in the whole sequence. Then the time interval between the peak points is calculated, and the standard deviation of the interval is used to measure the orderliness of the peak value distribution. The result shows that the standard deviation reaches 84s, indicating that the peak value appears with disorder jitter. The distribution of each peak value is analyzed. Some peak values are more than twice the average value, for example, the peak value at 1320s is 0.085 Wh / s, and the average value is only 0.038 Wh / s. Therefore, it is classified as a mutation peak value. The whole mutation peak value sequence is extracted separately and recorded according to the time stamp to form the increment disorder peak mutation data, which is used as the input of the subsequent convergence processing. After obtaining the increment disorder peak mutation data, the mutation peak jump convergence processing is performed. First, all the mutation peak values are arranged in time sequence. When the difference between two consecutive peak values exceeds 1.5 times the average peak value standard deviation, it is defined as a mutation jump point. The statistical result shows that there are 11 mutation jump points. The jump sections are applied to the exponential weighted average smoothing method for convergence calculation. The smoothing coefficient is set to 0.3. The convergence value is calculated point by point and replaces the original value, so as to obtain the peak jump convergence data. The data represents the stabilization trend of the energy consumption increment peak value with time, eliminates the interference of strong disorder fluctuations, and at the same time maintains the global feature of energy consumption evolution.After obtaining the peak jump convergence data, the semi-slope start-stop energy consumption progressive law is quantified based on the Logistic algorithm. First, the converged data is standardized to the interval of 0 to 1 as the input variable of the Logistic curve, the time is taken as the independent variable, and the converged data is taken as the dependent variable. The sequence is fitted by using the characteristics of the Logistic function, and the piecewise iterative search method is used in the fitting process to determine the inflection point position of the curve. The experimental results show that the inflection point is located at about 1100s, and then the energy consumption increment evolution gradually slows down. The lower limit of the fitted curve is 0.01, and the upper limit is stable at 0.09. It is shown that the energy consumption follows a nonlinear progressive law from fast to slow in the entire journey. Through this quantification process, the energy consumption progressive law is generated and output in the form of a complete sequence, including time stamp, progressive value and iteration parameter, which ensures that the law of energy consumption evolution with time can be expressed in the form of mathematical curve and actual value.

[0091] Step S3 comprises the following steps:

[0092] Step S31: selecting features of the energy consumption progressive law to obtain an energy consumption progressive feature law;

[0093] Step S32: memory regression learning of the energy consumption progressive feature law based on a BP neural network to obtain energy consumption law memory learning data;

[0094] Step S33: battery remaining endurance prediction architecture design according to the energy consumption law memory learning data to obtain a remaining endurance prediction architecture;

[0095] Step S34: sending the remaining endurance prediction architecture to a new energy vehicle terminal to perform battery management for the new energy vehicle.

[0096] As an example of the present application, reference is made to Figure 3 In this example, the step S3 comprises:

[0097] Step S31: selecting features of the energy consumption progressive law to obtain an energy consumption progressive feature law;

[0098] In the embodiment of the present application, after the energy consumption progressive law is generated, feature selection is first performed to extract key law parameters affecting the prediction of the remaining energy consumption of the battery, and the energy consumption progressive law is divided into multiple feature dimensions, including the progressive gradient in the time dimension, the nonlinear peak density distribution, the energy consumption increment mean value, the deviation rate, and the convergence speed of the progressive law. These features are calculated and extracted according to the statistical order. In the experimental data, the gradient mean value of the energy consumption progressive curve in the 0s-1200s interval is 0.035 Wh / s, the mean value in the 1200s-2400s interval is 0.021 Wh / s, and the mean value in the 2400s-3600s interval decreases to 0.011 Wh / s. The nonlinear peak density is the number of fluctuation peak values in each 600s interval, and the statistical results are 17 times, 11 times and 6 times respectively. The above parameters are constructed into a feature vector to form a sample sequence in time window units, ensuring that the complexity of the progressive law is preserved in the feature structure, thereby forming the energy consumption progressive feature law.

[0099] Step S32: memory regression learning of the energy consumption progressive feature law based on the BP neural network is performed to obtain energy consumption law memory learning data;

[0100] In the embodiment of the present application, after the energy consumption progressive feature law is established, memory regression learning is performed thereon. The feature law sequence is input into a neural network learning framework with an error back propagation mechanism. The connection weight and node state are propagated and corrected layer by layer through segmented input of the feature vector. The input features include the average gradient, peak density and convergence speed in the time window and other parameters. A 3-layer topological structure is set during network initialization. The input layer has 5 nodes corresponding to the 5 parameters of the features. The hidden layer has 8 nodes. The output layer has 1 node corresponding to the energy consumption prediction value. The number of training samples is set to 60 time window data points. The forward propagation is used to calculate the weighted sum and output the prediction value one by one. Then, the error value is calculated by comparing the prediction value with the real energy consumption data. The weights of each layer are corrected in the reverse direction according to the error. The cycle iteration is performed 1000 times until the mean square error converges below 0.003 as the termination condition. Finally, the energy consumption law memory learning data is obtained. The data represents a predictive output that retains the historical feature memory.

[0101] Step S33: a battery remaining endurance prediction architecture design is performed according to the energy consumption law memory learning data to obtain a remaining endurance prediction architecture;

[0102] In the embodiment of the present application, after completing the memory regression learning, the battery remaining endurance prediction architecture is designed based on the energy consumption law learning data, the energy consumption law memory learning data is directly mapped with the initial capacity and the remaining capacity of the battery, the total capacity of the power battery of the experimental vehicle is 50kWh, the initial electric quantity is set to 30kWh, the average unit time loss rate is 0.42kWh per hour under the condition of the slope angle of 4° and the traffic flow of 60 vehicles per minute, the rate is projected into the remaining 30kWh electric quantity to obtain the theoretical endurance time of about 71 hours, and the endurance mileage of about 2485km is converted from the average speed of 35km / h of the vehicle, under the actual condition, the progressive law correction parameter is introduced, the early energy consumption rate of 0.62kWh per hour and the late energy consumption rate of 0.35kWh per hour are weighted and superimposed to obtain the final endurance prediction value of 95km, the prediction value is structured and represented in the prediction architecture, including an input parameter module, an energy consumption memory curve module and a prediction output module, and finally a remaining endurance prediction architecture is formed.

[0103] Step S34: sending the remaining endurance prediction architecture to the new energy vehicle terminal to perform the battery management of the new energy vehicle.

[0104] In the embodiment of the present application, after obtaining the remaining endurance prediction architecture, it is deployed into the new energy vehicle terminal to perform the battery management operation, in the terminal transmission process, the data frame structure is used for transmission, the prediction architecture number is marked in the header of the data frame, the feature parameters and the memory learning weight values are contained in the middle part, and the prediction output rules are in the tail part, the prediction architecture is written into the running cache by the battery management unit in the receiving end, and the remaining endurance mileage is calculated in real time according to the sampling values such as current, voltage and vehicle speed in subsequent driving by calling the mapping relationship in the prediction architecture, the latest prediction value is constantly output through the refresh frequency of 1s, the energy consumption law and the remaining endurance estimation of the battery are kept consistent, and the implementation operation of the battery management of the new energy vehicle is completed.

[0105] Step S32 includes the following steps:

[0106] The energy consumption progressive feature law is processed by time sequence segmentation, the gradient sequence of each segment is extracted, and the energy consumption progressive gradient sequence is obtained;

[0107] The number of layers, the number of nodes and the initial weight distribution of the BP neural network are processed according to the energy consumption progressive gradient sequence, and the initial topology structure of the BP neural network is obtained;

[0108] The energy consumption progressive gradient sequence is processed by multi-layer forward propagation through the initial topology structure of the BP neural network, the weighted sum and the activation function output are calculated layer by layer, and the energy consumption progressive multi-layer forward output data is obtained;

[0109] The energy consumption progressive multi-layer forward output data is used for memory regression learning to obtain energy consumption law memory learning data.

[0110] In the embodiment of the present application, a segmented window is established with time as the horizontal axis in the data set of the energy consumption progressive characteristic law, the sampling time is 0s to 3600s in the experiment, the window length is set to 60s, 60 time periods are divided, each period includes 60 energy consumption data points, the energy consumption gradient value of adjacent sampling points is calculated on each segment by using the first-order difference method, that is, the energy consumption value of the current point is subtracted from the energy consumption value of the previous time point, and then divided by the interval 1s to obtain the energy consumption change rate per unit time, for example, if the energy consumption at the 10s is 0.45Wh and the energy consumption at the 9s is 0.41Wh in the first segment, the gradient is 0.04Wh / s, the same operation is performed on all sampling points to obtain the complete gradient sequence in the segment, the gradient trajectory is recorded in the 60 time periods by repeating the above process, the energy consumption progressive gradient is formed, and finally a data set including 60 sample trajectories is obtained, each trajectory represents an energy consumption change rate sequence in a time window. According to the specific embodiment of the layer number, node number and initial weight distribution processing of the BP neural network based on the energy consumption progressive gradient, the length of each gradient sequence is first counted as 60, the number of input layer nodes of the BP neural network is determined as 60, and the number of output layer nodes is set as 1 for outputting the predicted value, 2 hidden layers are set in the middle, the number of nodes of the first hidden layer is 30, and the number of nodes of the second hidden layer is 15, then the connection weights are initialized, the method is to generate an initial weight matrix according to the interval [-0.5, 0.5] random distribution, the bias node is uniformly initialized as 0.1, and the network is ensured to be in the non-saturated interval in the initial stage, and the initial topological structure of the BP neural network is obtained after the above processing.

[0111] The specific embodiment of the multi-layer forward propagation processing of the energy consumption progressive gradient sequence by the initial topological structure of the BP neural network is that after the input layer receives each energy consumption progressive gradient sequence, the input vector is multiplied and added with the first hidden layer weight matrix point by point to obtain the weighted sum, then the weighted sum is input to the Sigmoid function for activation, the output range is between 0 and 1, the output vector is transmitted to the second hidden layer, the same weighting and activation operation is repeated, the output of the second hidden layer is formed, and then transmitted to the output layer for weighting and activation once to obtain the predicted result value, the forward propagation process is run on the 60 sequences one by one to generate energy consumption progressive multi-layer forward output data, which is represented as an energy consumption prediction value corresponding to each time period.

[0112] According to the specific embodiment of the memory regression learning of the energy consumption progressive multi-layer forward output data, the error is calculated by comparing the predicted value obtained by the forward propagation with the original energy consumption true value, the square error is used as the loss function, the error is recorded in time sequence, for example, in the interval of 300s to 360s, the true energy consumption is 0.52Wh and the predicted output is 0.49Wh, then the error is 0.03Wh, the error is transmitted back to the nodes of each layer of the network, the weight value and the bias parameter are corrected by using the gradient descent algorithm, the updated weight value is used for the next round of forward propagation, the whole process is repeated on all 60 sample sequences, the iteration number is set to 1000 times, until the mean square error converges within the threshold value 0.003, and finally the energy consumption law memory learning data is formed, which retains the evolution process of the energy consumption law in the time dimension sequence and can predict the energy consumption trend in the future time period.

[0113] The application also provides a battery management system for new energy vehicles for executing the battery management method for new energy vehicles as described above, which comprises:

[0114] A data acquisition module is configured to acquire historical discharge power through a new energy vehicle control terminal, acquire historical driving environment states through a high-precision map carried by the new energy vehicle control terminal, and perform filtering and smoothing processing on the historical discharge power to obtain a power fluctuation smoothing curve.

[0115] A behavior learning module is configured to perform start-stop power mapping of the power fluctuation smoothing curve under a road half-slope state according to the historical driving environment states to obtain half-slope start-stop power mapping data, and perform energy consumption incremental behavior learning on the half-slope start-stop power mapping data to generate an energy consumption progressive law.

[0116] An architecture design module is configured to perform battery residual endurance prediction architecture design according to the energy consumption progressive law to obtain a residual endurance prediction architecture, and send the residual endurance prediction architecture to a new energy vehicle terminal to execute battery management for new energy vehicles.

[0117] The above description is only a specific embodiment of the application, so that those skilled in the art can understand or implement the application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A battery management method for new energy vehicles, characterized in that, The method comprises the following steps: Step S1: obtaining historical discharge power through a new energy vehicle control terminal; and obtaining historical driving environment state through a high-precision map carried by the new energy vehicle control; filtering and smoothing the historical discharge power to obtain a power fluctuation smooth curve; Step S2: performing start-stop power mapping of the power fluctuation smooth curve under a road half-slope state according to the historical driving environment state to obtain half-slope start-stop power mapping data; performing energy consumption increment behavior learning on the half-slope start-stop power mapping data to generate an energy consumption progression rule; wherein step S2 comprises: Step S21: extracting road geometry and traffic flow in the historical driving environment state; Step S22: performing start-stop power mapping of the power fluctuation smooth curve under a road half-slope state according to the traffic flow and road geometry to obtain half-slope start-stop power mapping data; Step S23: performing energy consumption increment behavior learning on the half-slope start-stop power mapping data to output energy consumption increment learning data under a half-slope start-stop state; Step S24: quantifying the energy consumption progression rule of the half-slope start-stop under time dimension driving according to the energy consumption increment learning data to generate the energy consumption progression rule; wherein step S24 comprises: analyzing the energy consumption increment learning data to obtain half-slope start-stop non-linear energy consumption increment evolution data under time dimension driving; performing disordered peak mutation analysis on the non-linear energy consumption increment evolution data to obtain increment disordered peak mutation data; performing mutation peak jump convergence on the non-linear energy consumption increment evolution data according to the increment disordered peak mutation data to obtain peak jump convergence data; quantifying the energy consumption progression rule of the half-slope start-stop under time dimension driving based on the Logistic algorithm and the peak jump convergence data to generate the energy consumption progression rule; Step S3: designing a battery remaining endurance prediction architecture according to the energy consumption progression rule to obtain a remaining endurance prediction architecture; and sending the remaining endurance prediction architecture to a new energy vehicle terminal to perform new energy vehicle battery management.

2. The battery management method for new energy vehicles according to claim 1, characterized in that, Step S1 comprises the following steps: Step S11: obtaining historical discharge power through a new energy vehicle control terminal; and obtaining historical driving environment state through a high-precision map carried by the new energy vehicle control; Step S12: performing fluctuation curve drawing on the historical discharge power and performing time stamp marking to construct a power fluctuation curve; Step S13: performing missing value filling on the power fluctuation curve to obtain a power fluctuation filling curve; Step S14: filtering and smoothing the power fluctuation filling curve to obtain a power fluctuation smooth curve.

3. The battery management method for new energy vehicles according to claim 1, characterized in that, Step S23 comprises the following steps: extracting half-slope start-stop frequency and half-slope inclination change in the half-slope start-stop power mapping data; and obtaining vehicle load information; performing gravity potential energy change variance calculation under a half-slope start-stop state according to the vehicle load information and the half-slope inclination change to obtain gravity potential energy change variance; performing instantaneous output energy consumption increment index analysis on the half-slope start-stop power mapping data based on the gravity potential energy change variance to obtain an instantaneous output energy consumption increment index; According to the half-slope start-stop frequency and the instantaneous output energy consumption increment index, the motor driving current additional loss is derived, and additional loss data is obtained; Based on the additional loss data, the energy consumption increment behavior learning is performed, and the energy consumption increment learning data in the half-slope start-stop state is output.

4. The battery management method for new energy vehicles according to claim 3, characterized in that, The instantaneous output energy consumption increment index analysis includes: Based on the variance of the gravitational potential energy, the slope gravity resistance component and the slope inertia force component in different stages are calculated; According to the slope gravity resistance component and the slope inertia force component, the instantaneous driving force required by the vehicle is analyzed; According to the instantaneous driving force required by the vehicle, the driving force instantaneous power coupling is performed on the half-slope start-stop power mapping data, and the driving force instantaneous power coupling data is obtained; The driving force instantaneous power coupling data is time-integrated to obtain the instantaneous time-integrated power; According to the instantaneous time-integrated power, the instantaneous output energy consumption increment index is analyzed to obtain the instantaneous output energy consumption increment index.

5. The battery management method for new energy vehicles according to claim 3, characterized in that, The motor driving current additional loss derivation includes: Based on the half-slope start-stop frequency, the motor high-load current state is analyzed; Based on the motor high-load current state, the battery high-rate discharge state is analyzed, and the high-rate discharge state is obtained; According to the high-rate discharge state, the heat square level growth trend of the battery internal resistance is analyzed; Based on the heat square level growth trend and the instantaneous output energy consumption increment index, the heat loss regression analysis is performed, and then the energy utilization loss gradient is quantified; According to the energy utilization loss gradient, the motor driving current additional loss is derived, and additional loss data is obtained.

6. The battery management method for new energy vehicles according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: selecting features of energy consumption progression rules to obtain energy consumption progression feature rules; Step S32: based on BP neural network, memory regression learning is performed on the energy consumption progression feature rules to obtain energy consumption rule memory learning data; Step S33: according to the energy consumption rule memory learning data, the battery remaining endurance prediction architecture is designed to obtain the remaining endurance prediction architecture; Step S34: the remaining endurance prediction architecture is sent to the new energy vehicle terminal to perform new energy vehicle battery management.

7. The battery management method for new energy vehicles according to claim 6, characterized in that, Step S32 includes the following steps: The energy consumption progression feature rules are time-integrated and segmented to extract the gradient sequence of each segment to obtain the energy consumption progression gradient sequence; According to the energy consumption progression gradient sequence, the number of layers, the number of nodes and the initial weight distribution of the BP neural network are allocated to obtain the initial topology structure of the BP neural network; Through the initial topology structure of the BP neural network, the energy consumption progression gradient sequence is processed by multi-layer forward propagation, and the weighted sum and activation function output are calculated layer by layer to obtain the energy consumption progression multi-layer forward output data; According to the energy consumption progression multi-layer forward output data, memory regression learning is performed to obtain energy consumption rule memory learning data.

8. A battery management system for a new energy vehicle, characterized in that, The new energy vehicle battery management system is used to perform the new energy vehicle battery management method as claimed in claim 1, which comprises: A data acquisition module is used to acquire historical discharge power through a new energy vehicle control terminal, and to acquire historical driving environment state through a high-precision map carried by the new energy vehicle control terminal; the historical discharge power is filtered and smoothed to obtain a power fluctuation smoothing curve; The behavior learning module is configured to perform start-stop power mapping of the power fluctuation smoothing curve in a road half-slope state according to a historical driving environment state, to obtain half-slope start-stop power mapping data; and to perform energy consumption increment behavior learning on the half-slope start-stop power mapping data, to generate an energy consumption progression law. The architecture design module is configured to perform battery residual endurance prediction architecture design according to the energy consumption progression law, to obtain a residual endurance prediction architecture; and to send the residual endurance prediction architecture to a new energy vehicle terminal, to perform battery management for the new energy vehicle.

Images