Search

CN-122009213-A - Ebike user behavior prediction and personalized power adjustment method

CN122009213ACN 122009213 ACN122009213 ACN 122009213ACN-122009213-A

Abstract

The application provides Ebike user behavior prediction and personalized power regulation method which comprises the steps of collecting Ebike multi-dimensional operation data, wherein the multi-dimensional operation data comprise user static physiological data, dynamic riding behavior data and scene perception data, generating riding behavior fingerprints of a user through characteristic engineering processing on the multi-dimensional operation data, generating scene-intention prediction results through multi-scale time sequence modeling and intention decoding processing on the riding behavior fingerprints, calling a personalized power regulation rule base, integrating a scene power base line, user behavior adaptation rules and battery protection constraint rules, executing dynamic calibration on the scene-intention prediction results to generate an adaptive power regulation command, executing power parameter calibration on the adaptive power regulation command, synchronously feeding back battery energy states to a regulation module, dynamically regulating boosting proportion and torque change rate, and carrying out cooperative adaptation of Ebike power output and user behavior, driving scenes and battery states.

Inventors

  • XIAO XIAOTAO
  • GUO XINFENG
  • Kuai qiang
  • SUN ZHEN
  • CHEN ZHIPEI

Assignees

  • 深圳唯乐高科技有限公司

Dates

Publication Date
20260512
Application Date
20260212

Claims (10)

  1. 1.A Ebike user behavior prediction and personalized power adjustment method, comprising the steps of: Step 1, acquiring Ebike multi-dimensional operation data, wherein the multi-dimensional operation data comprises user static physiological data, dynamic riding behavior data and scene perception data, and generating riding behavior fingerprints of a user through characteristic engineering processing on the multi-dimensional operation data; step 2, generating a scene-intention prejudging result for the riding behavior fingerprint through multi-scale time sequence modeling and intention decoding processing; Step 3, a personalized power adjustment rule base is called, the rule base integrates a scene power baseline, a user behavior adaptation rule and a battery protection constraint rule, and dynamic calibration of power parameters is carried out on the scene-intention prejudging result to generate an adaptation type power adjustment instruction; and 4, executing power parameter calibration on the adaptive power adjustment instruction, synchronously feeding back the battery energy state to the adjustment module, and dynamically adjusting the power assisting proportion and the torque change rate to carry out Ebike cooperative adaptation of power output, user behavior, driving scene and battery state.
  2. 2. The method of claim 1, wherein step 1 comprises: step 11, data purification integration is carried out on the static physiological data, the dynamic riding behavior data and the scene perception data of the user, abnormal interference data are removed, and user identifications are associated to generate a riding fit rule feature set; Step 12, constructing riding behavior trend features and riding parameter association features for the riding fit rule feature set, wherein the riding behavior trend features reflect the change rule of riding parameters, and the riding parameter association features reflect the cooperative relationship among different riding parameters so as to generate a riding time sequence association feature set; And 13, performing weight self-adaptive fusion on the riding time sequence association characteristic group to dynamically adjust the weight duty ratio of the static physiological characteristic, the dynamic behavior characteristic and the scene perception characteristic according to the scene type so as to generate the riding behavior fingerprint.
  3. 3. The method according to claim 2, wherein step 11 comprises: step 111, carrying out format standardization processing on the user static physiological data, and unifying data measurement units and storage formats to generate riding adaptation standardized physiological data; Step 112, performing outlier rejection on the dynamic riding behavior data, and judging and filtering parameter values exceeding the normal riding range based on the behavior rationality rule base to generate riding regularity behavior data; And 113, stripping redundant information from the scene perception data, reserving scene parameters related to riding behaviors, and performing related binding on the riding adaptation standardized physiological data and the riding regular behavior data to generate a riding adaptation rule feature set.
  4. 4. The method of claim 2, wherein step 12 comprises: Step 121, calculating the change rate of time sequence parameters in the riding adaptive gauge feature set, and capturing the increasing and decreasing trend of the parameters along with time so as to generate riding behavior trend features; Step 122, performing correlation analysis on different types of riding parameters in the riding fit rule feature set, and constructing an association mapping relation between the parameters to generate riding parameter association features; And 123, integrating the riding behavior trend features and the riding parameter association features, and sorting according to the data dimension to generate a riding time sequence association feature set.
  5. 5. The method according to claim 2, wherein step 13 comprises: Step 131, constructing a scene type recognition model, and classifying scenes of scene perception data to determine the current riding scene type; step 132, setting a feature weight distribution rule for the scene type, wherein the weight duty ratio of the related features is correspondingly adjusted under different scenes; and 133, performing weighted fusion operation on each feature in the riding time sequence association feature group according to the weight distribution rule to generate a riding behavior fingerprint matched with the current riding scene type.
  6. 6. The method of claim 1, wherein step 2 comprises: Step 21, multi-scale convolution feature extraction is carried out on the riding behavior fingerprint, local and global behavior features are captured in parallel through a first-size convolution kernel and a second-size convolution kernel, and the size of the second-size convolution kernel is larger than that of the first-size convolution kernel so as to generate a riding multi-scale feature map; Step 22, inputting the riding multi-scale feature map into a bidirectional time sequence network, and excavating the front-back dependency relationship of the behavior feature to generate a riding time sequence associated feature vector; And step 23, performing intention decoding on the riding time sequence associated feature vector, and generating a scene-intention prejudgement result by combining scene feature matching.
  7. 7. The method of claim 6, wherein step 21 comprises: Step 211, carrying out local feature extraction on the riding behavior fingerprint by adopting a first-size convolution check to obtain riding local original features, and carrying out association strengthening treatment on the riding local original features based on association relation of instantaneous riding parameters so as to generate a riding local feature map; Step 212, performing global feature extraction on the riding behavior fingerprint by adopting a second-size convolution kernel to obtain a riding global original feature, wherein the size of the second-size convolution kernel is larger than that of the first-size convolution kernel, and determining a riding global trend feature based on a behavior change trend in a continuous period by performing time sequence trend aggregation processing on the riding global original feature so as to generate a riding global feature map; and 213, carrying out channel attention weighted fusion on the riding local feature map and the riding global feature map to generate a riding multi-scale feature map.
  8. 8. The method of claim 6, wherein step 22 comprises: step 221, splitting the riding multi-scale feature map into riding time sequence feature segments according to time dimension; Step 222, inputting the riding time sequence feature segments into a forward path and a reverse path of a bidirectional time sequence network, and respectively extracting historical behavior influence and future behavior trend to obtain riding bidirectional path output features; And 223, fusing the output characteristics of the riding bidirectional paths, and constructing a characteristic expression containing a complete time sequence dependency relationship to generate a riding time sequence associated characteristic vector.
  9. 9. The method of claim 1, wherein step 3 comprises: step 31, matching scene types in scene-intention prejudgement results with scene power baselines in a personalized power regulation rule base, and determining a riding basic power parameter range; step 32, calling a user behavior adaptation rule for the user power demand in the scene-intention prejudging result based on the riding basic power parameter range so as to adjust and adjust the power assisting proportion and the torque change rate parameter to obtain riding adjusted power parameters; And 33, limiting the boundary of the riding adjusted power parameters according to the battery protection constraint rule to generate an adaptive power adjustment instruction.
  10. 10. The method of claim 1, wherein step 4 comprises: step 41, carrying out layering analysis on charge and discharge parameters, power assisting proportion parameters and torque change rate parameters in the adaptive power adjustment instruction, and extracting reference values and adjustment thresholds of the parameters to generate a power adjustment parameter analysis set; Step 42, collecting the residual capacity of the battery, the consistency of the voltage of the battery core and the working state of the energy battery in real time, integrating to form a battery energy state data set, feeding back the battery energy state data set to an adjusting module, and triggering parameter dynamic calibration logic; Step 43, constructing a 'demand-supply' adaptive model based on a battery energy state data set and a power regulation parameter analysis set, when the residual electric quantity of the battery is sufficient, adjusting the power assisting proportion according to the user behavior characteristics to match the power generating habit, and when the residual electric quantity of the battery is low, optimizing the torque change rate on the premise of keeping basic power; And step 44, performing scene adaptation verification on the adjusted boosting proportion and the torque change rate by combining road condition characteristics of the current driving scene, and matching the engagement fluency measurement of the power output at different stages of start-stop, acceleration and uniform speed so as to perform dynamic collaborative adaptation of Ebike power output and user behaviors, driving scenes and battery states.

Description

Ebike user behavior prediction and personalized power adjustment method Technical Field The application relates to the technical field of electric power-assisted bicycle control, in particular to a Ebike user behavior prediction and personalized power adjustment method. Background Along with the popularization of green trip concepts, ebike has become an important tool for urban commute, outdoor body building and leisure and tour, and the requirements of users on the suitability and the comfortableness of power output in the riding process are increasingly improved. The physiological characteristics and riding habits of different users are obviously different, and the requirements for power are different in the same driving scene, so that the power adjustment of Ebike can be accurately matched with the individual characteristics and scene change of the users. In the prior art, a common Ebike power adjustment scheme is to preset a fixed power mode, and determine the power assisting proportion by switching gear positions, and the key thought is that limited gear positions are divided based on rated power of a motor, each gear position corresponds to a fixed power assisting coefficient, a user manually switches according to subjective feeling, and the system only outputs auxiliary power with fixed proportion according to the current gear position. According to the scheme, a basic power assisting function is realized by simplifying control logic, and complex data processing and model operation are not needed. However, the fixed mode power adjustment scheme has obvious technical defects that the power output cannot be dynamically adjusted by combining individual characteristics of a user with real-time riding behaviors, so that the power adaptability is poor. For example, the fixed power assisting proportion is difficult to meet the power demands of users with different heights and weights at the same time, the pedal frequency and pedal force habit of the users can not be matched, the situation that the power assisting and the manpower power are asynchronous often occurs, the riding smoothness and the comfortableness are affected, and the problem is the core pain point to be solved by the technical scheme of the application. Disclosure of Invention In order to solve the technical problems, the application provides a Ebike user behavior prediction and personalized power adjustment method to solve the problem of insufficient power adaptability in the prior art. The technical scheme provided by the embodiment of the application is as follows: a Ebike user behavior prediction and personalized power adjustment method, comprising the steps of: Step 1, acquiring Ebike multi-dimensional operation data, wherein the multi-dimensional operation data comprises user static physiological data, dynamic riding behavior data and scene perception data, and generating riding behavior fingerprints of a user through characteristic engineering processing on the multi-dimensional operation data; step 2, generating a scene-intention prejudging result for the riding behavior fingerprint through multi-scale time sequence modeling and intention decoding processing; Step 3, a personalized power adjustment rule base is called, the rule base integrates a scene power baseline, a user behavior adaptation rule and a battery protection constraint rule, and dynamic calibration of power parameters is carried out on the scene-intention prejudging result to generate an adaptation type power adjustment instruction; and 4, executing power parameter calibration on the adaptive power adjustment instruction, synchronously feeding back the battery energy state to the adjustment module, and dynamically adjusting the power assisting proportion and the torque change rate to carry out Ebike cooperative adaptation of power output, user behavior, driving scene and battery state. Optionally, step 1 includes: step 11, data purification integration is carried out on the static physiological data, the dynamic riding behavior data and the scene perception data of the user, abnormal interference data are removed, and user identifications are associated to generate a riding fit rule feature set; Step 12, constructing riding behavior trend features and riding parameter association features for the riding fit rule feature set, wherein the riding behavior trend features reflect the change rule of riding parameters, and the riding parameter association features reflect the cooperative relationship among different riding parameters so as to generate a riding time sequence association feature set; And 13, performing weight self-adaptive fusion on the riding time sequence association characteristic group to dynamically adjust the weight duty ratio of the static physiological characteristic, the dynamic behavior characteristic and the scene perception characteristic according to the scene type so as to generate the riding behavior fingerprint. Optionally, step 11 includes: ste