Bicycle gear self-adaptive decision-making method, system and equipment based on multi-modal sensor cooperative processing and medium

Abstract

The invention relates to the technical field of bicycles. The invention provides a bicycle gear self-adaptive decision-making method, system and device based on multi-modal sensor cooperative processing and a medium. The method comprises the steps that motion parameters and environment parameters of a bicycle are collected in real time through a multi-modal sensor set; performing space-time alignment and noise filtering processing on the motion parameters and the environment parameters to generate standardized feature vectors; performing riding mode recognition processing on the standardized feature vector through a lightweight neural network model, and outputting a current riding mode category; based on a preset gear configuration template library and the current riding mode category, self-adaptive strategy matching processing is carried out, and a self-adaptive gear decision instruction is generated; and the gear switching operation corresponding to the adaptive gear decision instruction is executed through the preset closed-loop control unit, so that the technical effects of improving the decision precision, enhancing the adaptive capability of the riding scene and reducing mechanical loss are achieved.

Description

Technical Field

[0001] This invention relates to the field of bicycle technology, and in particular to a bicycle gear adaptive decision-making method, system, device, and medium based on multimodal sensor collaborative processing. Background Technology

[0002] With the popularization of intelligent transportation and healthy travel concepts, bicycles are playing an increasingly important role in urban commuting, fitness, and environmentally friendly travel. Adaptive gear selection is a core element in improving riding efficiency and optimizing energy distribution, and it is also a key focus and challenge in intelligent cycling system research. Effective adaptive gear selection aims to dynamically adjust the gear ratio based on real-time riding conditions, achieving smooth gear shifting and an optimal energy-efficient riding experience.

[0003] However, in traditional technologies, gear selection primarily relies on single parameter thresholds such as speed or cadence to trigger shifts. At the data monitoring level, traditional cycling computers only focus on basic information like speed and mileage, neglecting the fusion of multimodal data such as pressure, posture, and environment, leading to biased gear selection decisions. At the strategy setting level, the use of preset fixed gear templates makes it difficult to adapt to diverse riding modes such as climbing and high-speed cruising, limiting the improvement in riding efficiency. At the mechanical execution level, PWM servo shifting uses an open-loop control method, failing to consider vibration feedback from the transmission system, which not only causes shifting shocks but also exacerbates gear wear. Summary of the Invention

[0004] Therefore, it is necessary to provide a bicycle gear adaptive decision-making method, system, device, and medium based on multimodal sensor collaborative processing to address the above-mentioned technical problems, so as to improve decision-making accuracy, enhance the adaptability of riding scenarios, and reduce mechanical wear.

[0005] In a first aspect, this application provides a bicycle gear selection adaptive decision-making method based on multimodal sensor collaborative processing, the method comprising:

[0006] The bicycle's motion parameters and environmental parameters are collected in real time using a multimodal sensor array.

[0007] Spatiotemporal alignment and noise filtering are performed on motion parameters and environmental parameters to generate standardized feature vectors;

[0008] A lightweight neural network model is used to process the standardized feature vector for cycling pattern recognition, and the current cycling pattern category is output.

[0009] Based on the preset gear configuration template library and the current riding mode category, an adaptive strategy matching process is performed to generate an adaptive gear decision instruction.

[0010] The gear shifting operation corresponding to the adaptive gear shifting decision command is executed by a preset closed-loop control unit.

[0011] In one embodiment, an adaptive strategy matching process is performed based on a preset gear configuration template library and the current riding mode category to generate an adaptive gear decision instruction, including:

[0012] Call the gear configuration template in the preset gear configuration template library. The gear configuration template includes gear switching threshold and gear interval ratio parameters.

[0013] Based on the current riding mode category, a dynamic strategy is matched to generate an initial strategy matching result;

[0014] The gear shifting threshold is dynamically optimized based on the user's historical riding data to generate an optimized gear shifting threshold.

[0015] The gear interval ratio parameter is dynamically optimized based on the mechanical characteristics of the bicycle to generate the optimized gear interval ratio parameter.

[0016] The initial strategy matching result is updated based on the optimized gear shifting threshold and the optimized gear interval ratio parameter, and the strategy matching result is generated.

[0017] Adaptive gear selection instructions are generated based on the strategy matching results.

[0018] In one embodiment, the gear shifting threshold is dynamically optimized based on the user's historical riding data to generate an optimized gear shifting threshold, including:

[0019] Extract cycling mode distribution characteristics and gear efficiency indicators from users' historical cycling data to generate cycling habit analysis data;

[0020] Based on cycling habit analysis data, an adaptive optimization strategy is matched to generate threshold adjustment parameters;

[0021] Using the following formula, the gear shifting threshold is dynamically and iteratively adjusted based on the threshold adjustment parameters through an incremental learning mechanism to generate an optimized gear shifting threshold:

[0022]

[0023] in, This indicates the optimized gear shift threshold. Indicates the initial gear shift threshold. Indicates the incremental learning rate. This represents the number of historical cycling data samples. Indicates the first Time decay weights for each sample This indicates the efficiency index of gear positions. This indicates a pattern matching indicator function.

[0024] In one embodiment, the gear interval ratio parameter is dynamically optimized based on the mechanical characteristics of the bicycle to generate an optimized gear interval ratio parameter, including:

[0025] Extract real-time mechanical state parameters of bicycle transmission to generate a mechanical characteristic dataset;

[0026] Based on the preset transmission efficiency optimization model, the mechanical characteristic dataset is dynamically analyzed and processed to generate gear ratio adjustment coefficients.

[0027] Using the following formula, based on the gear ratio adjustment coefficient, an adaptive smoothing algorithm is used to incrementally optimize the gear interval ratio parameter, generating the optimized gear interval ratio parameter:

[0028]

[0029] in, This indicates the optimized gear interval ratio parameter. This represents the initial gear interval ratio parameter vector. This represents element-wise multiplication. Indicates the adaptive smoothing coefficient. This represents the transmission efficiency loss function. Indicates the mechanical wear factor. Indicates the start time for optimization. Indicates the optimization termination time. Represents a time variable. This indicates the gear interval ratio parameter.

[0030] In one embodiment, the gear shifting operation corresponding to the adaptive gear shifting decision command is executed through a preset closed-loop control unit, including:

[0031] Based on the closed-loop control unit, the adaptive gear decision command is parsed into target gear parameters and switching trajectory constraints;

[0032] A drive signal sequence matching the target gear parameters is generated based on a dynamic sliding diaphragm control algorithm;

[0033] A progressive execution strategy is used to perform time-series scheduling on the driving signal sequence to generate segmented execution instructions;

[0034] Based on the switching trajectory constraints, the gear shifting action corresponding to the segmented execution command is executed, and the vibration feedback signal of the mechanical transmission system is collected in real time.

[0035] The drive signal sequence is corrected online based on the vibration feedback signal until the gear shifting operation is completed.

[0036] In one embodiment, a lightweight neural network model is used to process the standardized feature vector for cycling pattern recognition, and the current cycling pattern category is output, including:

[0037] Multimodal feature decoupling is performed on the standardized feature vectors to generate velocity feature subspace, pressure feature subspace, attitude feature subspace and environment feature subspace;

[0038] By using a bidirectional long short-term memory network, the temporal dependencies in each feature subspace of velocity, pressure, attitude and environment are captured respectively, and temporal enhanced feature vectors are generated.

[0039] Spatial correlation modeling of temporal augmentation feature vectors across modal dimensions is performed using 3D convolutional kernels to generate fused vector feature tensors;

[0040] Dynamically weighted and aggregated fused feature tensors based on an attention weight allocation mechanism to generate pattern recognition feature vectors;

[0041] The current cycling mode category is output by classifying the pattern recognition feature vectors using a pre-defined fully connected classifier.

[0042] In one embodiment, the motion parameters and environmental parameters of the bicycle are collected in real time using a multimodal sensor array, including:

[0043] Wheel speed pulse signals are acquired using a speed sensor;

[0044] The change in resistance of the crank strain gauge is collected using a pressure sensor;

[0045] The Euler angle and angular velocity of the vehicle body are measured using an attitude sensor;

[0046] Detects obstacle distance information using environmental perception sensors;

[0047] Motion parameters include wheel speed pulse signals, crank strain gauge resistance changes, vehicle body Euler angles, and angular velocities; environmental parameters include obstacle distance information.

[0048] Secondly, this application also provides a bicycle gear adaptive decision-making system based on multimodal sensor cooperative processing, the system comprising:

[0049] The multi-source sensing module is used to collect the bicycle's motion parameters and environmental parameters in real time through a multi-modal sensor array;

[0050] The data normalization module is used to perform spatiotemporal alignment and noise filtering on motion parameters and environmental parameters, and generate standardized feature vectors.

[0051] The pattern recognition module is used to process the standardized feature vectors for cycling pattern recognition using a lightweight neural network model and output the current cycling pattern category.

[0052] The strategy decision module is used to perform adaptive strategy matching processing based on the preset gear configuration template library and the current riding mode category, and generate adaptive gear decision instructions.

[0053] The execution control module is used to execute the gear switching operation corresponding to the adaptive gear decision command through a preset closed-loop control unit.

[0054] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0055] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0056] This application provides a bicycle gear-adaptive decision-making method, system, device, and medium based on multimodal sensor collaborative processing. It utilizes a multimodal sensor array to collect real-time bicycle motion and environmental parameters, breaking the limitations of traditional single-parameter monitoring. By integrating multi-dimensional data such as speed, pressure, posture, and environment, it provides more comprehensive information support for gear-shifting decisions. The collected motion and environmental parameters undergo spatiotemporal alignment and noise filtering. Clock synchronization unifies the data's time dimension, and filtering algorithms eliminate interference, generating standardized feature vectors to ensure data accuracy and consistency, laying a solid foundation for subsequent precise analysis. A lightweight neural network model is used to process the standardized feature vectors for riding mode recognition. This model can deeply mine multimodal data correlations, capture temporal changes in riding states, and output the current riding mode category, making gear-shifting decisions more aligned with actual riding scenarios.

[0057] When performing adaptive strategy matching based on a pre-set gear configuration template library and the current riding mode category, it not only relies on the template library to obtain the basic gear parameters of the corresponding mode, but also dynamically optimizes the decision logic in combination with the actual scenario, generating more adaptable adaptive gear decision commands, effectively enhancing the adaptability to different riding scenarios. Gear switching operations are executed through a pre-set closed-loop control unit, which can provide real-time feedback and correction during execution, avoiding the shift shock problem that is prone to occur in traditional open-loop control, reducing wear on mechanical parts, and lowering mechanical losses. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart of a bicycle gear adaptive decision-making method based on multimodal sensor collaborative processing, according to one embodiment of the present invention.

[0060] Figure 2 This is a flowchart illustrating the gear shifting operation corresponding to the adaptive gear decision command executed by a preset closed-loop control unit in one embodiment of the present invention.

[0061] Figure 3 This is a structural diagram of a bicycle gear adaptive decision-making system based on multimodal sensor collaborative processing, according to one embodiment of the present invention. Detailed Implementation

[0062] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0063] First, the application scenarios of the embodiments of this application are described. In the embodiments of this application, a bicycle gear adaptive decision-making method, system, device, and medium based on multimodal sensor collaborative processing are provided, applicable to scenarios such as urban commuting cycling, outdoor fitness cycling, and professional cycling training. For example, in urban commuting scenarios, cyclists frequently need to deal with complex situations such as traffic lights, pedestrians crossing, and changes in road slope. This application can dynamically adjust the gear based on real-time collected data such as vehicle speed, vehicle posture, and distance to surrounding obstacles to adapt to different states such as starting and stopping, smooth driving, or short-distance acceleration. In outdoor fitness cycling scenarios, facing diverse terrains such as mountains and plains, the gear switching logic can be optimized by sensing parameters such as pedaling pressure and vehicle tilt angle, helping cyclists to rationally allocate physical energy and improve fitness results. In professional cycling training scenarios, historical training data and real-time cycling modes can be combined to generate gear strategies adapted to training goals (such as endurance improvement and sprint training), assisting in the efficient execution of training plans.

[0064] In illustrative purposes, the bicycle gear adaptive decision-making method, system, device and medium based on multimodal sensor collaborative processing provided in this application embodiment can also be applied to other application scenarios, such as the intelligent operation and maintenance scenario of shared bicycles (reducing mechanical wear and tear of shared vehicles and extending service life through gear adaptive adjustment) and the elderly leisure riding scenario (improving riding safety and comfort through smooth gear switching and safety warning linkage). This is only an example and does not limit the specific application scenario.

[0065] like Figure 1 As shown, this application provides a bicycle gear adaptive decision-making method based on multimodal sensor cooperative processing, the method including:

[0066] S101: Real-time acquisition of bicycle motion and environmental parameters through a multimodal sensor array.

[0067] For example, the synchronous acquisition function of a multimodal sensor array is used to ensure that various sensors work collaboratively at a preset frequency and time reference. The speed sensor in the multimodal sensor array acquires the bicycle wheel rotation speed parameters, the pressure sensor acquires the change in the strain resistance of the bicycle crank, the attitude sensor acquires the Euler angles and angular velocity parameters of the bicycle frame, and the environmental perception sensor acquires the distance parameters of obstacles around the bicycle. After acquisition, the multimodal sensor array summarizes the above parameters to obtain the bicycle's motion parameters and environmental parameters.

[0068] S102: Perform spatiotemporal alignment and noise filtering on motion parameters and environmental parameters to generate standardized feature vectors.

[0069] For example, spatiotemporal alignment processing is performed on the collected motion parameters and environmental parameters of the bicycle. During the spatiotemporal alignment process, the bicycle's motion parameters and environmental parameters are unified to the same time axis reference according to the clock synchronization mechanism of the central main controller, eliminating time deviations when data is collected from different sensors and ensuring that the bicycle's motion parameters and environmental parameters are consistent in the time dimension. After the spatiotemporal alignment processing is completed, noise filtering processing is performed on the time-consistent bicycle motion parameters and environmental parameters.

[0070] During the noise filtering process, a preset filtering algorithm is used to remove invalid data from the bicycle's motion and environmental parameters caused by mechanical vibration, environmental interference, and other factors, retaining only valid data. After the noise filtering process is completed, the bicycle's motion and environmental parameters, which have undergone spatiotemporal alignment and noise filtering, are subjected to feature standardization processing, converting various parameters into feature data with a unified dimension and value range. Through the above series of processes, a standardized feature vector is generated.

[0071] S103: Uses a lightweight neural network model to process the standardized feature vector for cycling pattern recognition and outputs the current cycling pattern category.

[0072] For example, a standardized feature vector is input into a pre-built lightweight neural network model. The model decouples the standardized feature vector into multimodal features, separating it into velocity, pressure, posture, and environmental feature sub-vectors. Then, it captures the temporal dependencies of each feature sub-vector, generating a temporally enhanced feature vector. This temporally enhanced feature vector is then fused across modalities to generate a fused feature tensor. A built-in classification decision module processes the fused feature tensor for cycling mode recognition, distinguishing the mode type according to preset mode determination rules. Finally, the lightweight neural network model outputs the current cycling mode category.

[0073] S104: Based on the preset gear configuration template library and the current riding mode category, perform adaptive strategy matching processing to generate adaptive gear decision instructions.

[0074] For example, a pre-set gear configuration template library is invoked to extract basic gear parameters adapted to the current scenario. These basic parameters include gear switching thresholds and gear interval ratios corresponding to different riding modes. The current riding mode category is matched with the mode types in the pre-set gear configuration template library to select the corresponding initial gear strategy. The gear switching thresholds in the initial gear strategy are optimized by combining historical user riding data, and the gear interval ratios in the initial gear strategy are adjusted according to the mechanical characteristics of the bicycle. The optimized parameters are integrated to generate a complete strategy, which is then converted into a control signal format recognizable by the execution unit to generate adaptive gear decision instructions.

[0075] S105: Executes the gear shifting operation corresponding to the adaptive gear shift decision command through a preset closed-loop control unit.

[0076] For example, the adaptive gear selection decision command is input to a preset closed-loop control unit. The preset closed-loop control unit parses the command to extract the target gear parameters and gear shifting trajectory constraints. Based on the extracted parameters and conditions, an adapted drive signal sequence is generated, and the drive signal sequence is sent to the gear shifting component in stages according to the segmented execution logic. During gear shifting, the preset closed-loop control unit collects vibration feedback signals from the gear shifting component and the mechanical transmission system in real time. If deviations exist, the drive signal sequence is corrected online. Afterward, the preset closed-loop control unit completes the gear shifting operation corresponding to the adaptive gear selection decision command.

[0077] This application provides an embodiment of a bicycle gear selection adaptive decision-making method based on multimodal sensor collaborative processing. This method uses a multimodal sensor array to collect bicycle motion and environmental parameters in real time, breaking the limitations of traditional single-parameter monitoring. It integrates multi-dimensional data such as speed, pressure, posture, and environment, providing more comprehensive information support for gear selection decisions. The collected motion and environmental parameters undergo spatiotemporal alignment and noise filtering. Clock synchronization unifies the data's time dimension, and filtering algorithms eliminate interference, generating standardized feature vectors to ensure data accuracy and consistency, laying a solid foundation for subsequent precise analysis. A lightweight neural network model is used to process the standardized feature vectors for riding mode recognition. This model can deeply mine multimodal data correlations, capture temporal changes in riding states, and output the current riding mode category, making gear selection decisions more aligned with actual riding scenarios.

[0078] When performing adaptive strategy matching based on a pre-set gear configuration template library and the current riding mode category, it not only relies on the template library to obtain the basic gear parameters of the corresponding mode, but also dynamically optimizes the decision logic in combination with the actual scenario, generating more adaptable adaptive gear decision commands, effectively enhancing the adaptability to different riding scenarios. Gear switching operations are executed through a pre-set closed-loop control unit, which can provide real-time feedback and correction during execution, avoiding the shift shock problem that is prone to occur in traditional open-loop control, reducing wear on mechanical parts, and lowering mechanical losses.

[0079] In one embodiment, an adaptive strategy matching process is performed based on a preset gear configuration template library and the current riding mode category to generate an adaptive gear decision instruction, including:

[0080] (1) Call the gear configuration template in the preset gear configuration template library. The gear configuration template includes the gear switching threshold and gear interval ratio parameters.

[0081] For example, the corresponding gear configuration template is extracted from the preset gear configuration template library according to the preset retrieval rules to ensure that the subsequent strategy construction has complete basic parameter basis.

[0082] Among them, the gear shift threshold is a critical parameter used to determine whether a gear shift is triggered, and the gear interval ratio parameter is a parameter used to determine the size of the interval between different gears.

[0083] (2) Based on the current cycling mode category, match the corresponding dynamic strategy to generate the initial strategy matching result.

[0084] For example, feature parsing is performed on the current riding mode category to extract key features, including riding speed range and road gradient features. These key features are then compared one by one with the mode features corresponding to each dynamic strategy in the preset gear configuration template library. The dynamic strategy with the highest feature matching degree is selected, and an initial strategy matching result is generated based on this dynamic strategy.

[0085] Among them, dynamic strategies refer to the pre-designed strategy schemes for different riding mode categories in the pre-set gear configuration template library. Each dynamic strategy includes gear adjustment logic and basic parameters, and corresponds to specific riding scenario requirements, providing a basic framework for adapting the initial strategy matching results to the scenario.

[0086] (3) Based on the user's historical riding data, the gear shift threshold is dynamically optimized to generate the optimized gear shift threshold.

[0087] For example, historical user cycling data is retrieved from storage, categorized, and extracted to identify data related to gear shifting, such as the speed and cadence at which the user triggers gear shifting in different cycling modes. A data analysis model is then used to uncover user gear usage preferences and physical adaptation patterns in different scenarios. Based on these patterns, the gear shifting threshold in the initial strategy matching results is gradually adjusted, and an optimized gear shifting threshold is generated after multiple rounds of adaptive verification.

[0088] The user's historical cycling data includes records of gear shifting in various cycling scenarios, such as urban commuting and mountain biking. It also includes data on changes in cycling parameters, such as speed, cadence, and pedaling pressure, as well as user feedback on gear shifting. This data directly reflects the user's cycling habits and physical characteristics, providing data support for optimizing gear shifting thresholds.

[0089] (4) Based on the mechanical characteristics of the bicycle, the gear interval ratio parameter is dynamically optimized to generate the optimized gear interval ratio parameter.

[0090] For example, real-time operational data of a bicycle's mechanical system is collected using sensor components. This data, combined with inherent parameters from the bicycle's mechanical profile (such as transmission system structural parameters and gear specifications), is used to analyze the bicycle's mechanical characteristics. These characteristics include transmission efficiency, gear meshing accuracy, and servo motor response speed. The impact of different gear interval ratios on the mechanical stability and lifespan under the current mechanical state is determined. Based on the analysis results, the gear interval ratio parameters in the initial strategy matching results are finely adjusted. After verification through mechanical load simulation, optimized gear interval ratio parameters are generated.

[0091] Among them, bicycle mechanical characteristics refer to the performance parameters and operating characteristics of the bicycle's mechanical system. The bicycle mechanical system includes the transmission system, gear shifting system, etc., and its performance parameters and operating characteristics include, for example, the power transmission efficiency of the transmission system, the degree of gear wear, and the response time and accuracy of the servo motor.

[0092] (5) Update the initial strategy matching result based on the optimized gear shift threshold and the optimized gear interval ratio parameter, and generate the strategy matching result.

[0093] For example, the parameter structure of the initial strategy matching result is parsed to determine the position and function of the gear shift threshold and gear interval ratio parameters in the initial strategy matching result. The optimized gear shift threshold and optimized gear interval ratio parameters are substituted into their corresponding positions to replace the initial parameters. At the same time, the gear adjustment logic in the strategy is adaptively modified to ensure that the optimized parameters and logic can work together. After strategy integrity verification, the strategy matching result is generated.

[0094] Among them, strategy integrity verification refers to a multi-dimensional check of the updated strategy matching results. The check includes logical coherence, parameter rationality, and scenario adaptability, ensuring that the parameters in the strategy do not conflict with the adjustment logic and can meet the comprehensive needs of the current riding mode, user habits, and bicycle mechanical status, thus ensuring that the strategy matching results are feasible for actual execution.

[0095] (6) Generate adaptive gear decision instructions based on strategy matching results.

[0096] For example, the strategy matching result is parsed to convert the gear adjustment requirements in the strategy into standardized parameter commands, such as target gear and shift speed. Based on the bicycle actuator's communication protocol and control standard, the standardized parameter commands are encoded to generate control signals conforming to the actuator's data reception format. After command correctness verification, an adaptive gear decision command is generated. Command correctness verification includes checking whether the command conforms to the actuator's operating range and whether there are any logical errors.

[0097] In one embodiment, the gear shifting threshold is dynamically optimized based on the user's historical riding data to generate an optimized gear shifting threshold, including:

[0098] (1) Extract the distribution characteristics of riding modes and gear efficiency indicators from the user's historical riding data to generate riding habit analysis data.

[0099] For example, historical cycling data of users in different cycling scenarios and time periods are retrieved from the user's cycling record storage medium. The data is then analyzed in multiple dimensions using data analysis tools to extract cycling pattern distribution characteristics that reflect the user's cycling preferences, as well as gear efficiency indicators that can measure the effectiveness of gear usage. After that, the extracted features and indicators are systematically integrated to generate cycling habit analysis data.

[0100] The cycling mode distribution characteristics include information such as the proportion of different cycling modes used by users in various cycling scenarios, including flat roads, uphill, and downhill, as well as the frequency of mode switching. The gear efficiency index involves quantitative data on the increase in cycling speed and energy consumption during gear shifts.

[0101] (2) Based on cycling habit analysis data, an adaptive optimization strategy is matched to generate threshold adjustment parameters.

[0102] For example, using the generated cycling habit analysis data as the core basis, the adaptive optimization strategy that best matches the cycling habit analysis data is selected from a pre-built adaptive optimization strategy library through feature comparison, logical matching, and other methods. Based on the selected strategy, the specific adjustment rules and parameter ranges for the gear shifting threshold are defined, thereby generating threshold adjustment parameters.

[0103] The adaptive optimization strategy library contains a variety of strategies that are adapted to different cycling habits. Each strategy includes unique parameter adjustment logic and weight allocation methods to meet the personalized needs of different users.

[0104] (3) Using the following formula, based on the threshold adjustment parameters, the gear shifting threshold is dynamically and iteratively adjusted through an incremental learning mechanism to generate the optimized gear shifting threshold:

[0105]

[0106] in, This indicates the optimized gear shift threshold. Indicates the initial gear shift threshold. Indicates the incremental learning rate. This represents the number of historical cycling data samples. Indicates the first Time decay weights for each sample This indicates the efficiency index of gear positions. This indicates a pattern matching indicator function.

[0107] For example, based on the generated threshold adjustment parameters, the initial gear shifting threshold is dynamically and iteratively adjusted through an incremental learning mechanism. In each iteration, multiple factors are comprehensively considered, including the time decay weight of historical riding data samples (samples from different times have different impacts on the current adjustment, with newer samples usually having a higher impact weight), gear efficiency indicators (the actual efficiency brought by gear shifting), and pattern matching indicator functions (to determine the matching degree between the current riding mode and the strategy to determine the adjustment direction). After multiple rounds of repeated optimization iterations, an optimized gear shifting threshold is generated.

[0108] The incremental learning mechanism continuously updates and adjusts parameters based on newly acquired cycling data, ensuring that threshold adjustments always align with the user's latest cycling habits. The combined effects of time decay weighting, gear efficiency metrics, and pattern matching indicators make threshold adjustments more targeted and precise.

[0109] In one embodiment, the gear interval ratio parameter is dynamically optimized based on the mechanical characteristics of the bicycle to generate an optimized gear interval ratio parameter, including:

[0110] (1) Extract the real-time mechanical state parameters of the bicycle transmission and generate a mechanical characteristic dataset.

[0111] For example, various types of sensors, such as temperature sensors, displacement sensors, and pressure sensors, installed in key parts of the bicycle drivetrain, are used to continuously monitor parameters such as the operating temperature of the transmission components, gear meshing clearance, drive belt tension, and bearing vibration amplitude. The sensors convert the collected physical signals into electrical signals in real time and transmit them to the data acquisition unit. After preprocessing the electrical signals, such as filtering and amplification, the data acquisition unit summarizes all parameter information to generate a mechanical characteristic dataset that includes various mechanical state parameters.

[0112] Among them, temperature sensors are used to detect the temperature of transmission components such as gears and bearings, displacement sensors are used to measure displacement such as gear meshing clearance, and pressure sensors are used to sense pressure values ​​such as transmission belt tension.

[0113] A high-precision inductive displacement sensor can be selected and installed inside the bicycle gearbox near the critical meshing position of the driving and driven gears to ensure that the sensor's detection end is aligned with the meshing clearance area of ​​the two gears. This displacement sensor is based on the inductive measurement principle. It emits an electromagnetic signal of a specific frequency to the gear meshing area. When the meshing clearance changes slightly due to wear, vibration, or other factors during transmission, the phase and amplitude of the electromagnetic signal received by the sensor will change accordingly. The sensor converts this signal change into an electrical signal corresponding to the displacement.

[0114] In real-world cycling scenarios, the gear meshing clearance may dynamically fluctuate under different load conditions. The displacement sensor continuously monitors this fluctuation, outputting an electrical signal to the data acquisition unit in real time. The data acquisition unit filters and amplifies this signal, converting it into quantifiable gear meshing clearance data. This data, along with parameters such as the temperature of the transmission components collected by the temperature sensor and the tension of the transmission belt collected by the pressure sensor, is then incorporated into the mechanical characteristic dataset.

[0115] (2) Based on the preset transmission efficiency optimization model, the mechanical characteristic dataset is dynamically analyzed and processed to generate the gear ratio adjustment coefficient.

[0116] For example, a mechanical characteristic dataset is input into a pre-defined transmission efficiency optimization model. This model is constructed based on the dynamic equations of the transmission system and the theory of mechanical losses, and includes a transmission efficiency calculation unit and a mechanical loss analysis unit. The transmission efficiency calculation unit calculates the actual transmission efficiency of the current transmission system based on the parameters in the mechanical characteristic dataset, combined with information such as the transmission ratio and power input / output. The mechanical loss analysis unit analyzes the magnitude of various mechanical losses, such as gear friction loss, bearing friction loss, and transmission belt elastic sliding loss. The model integrates the transmission efficiency calculation results and the mechanical loss analysis results to determine the direction and magnitude of the gear interval ratio adjustment required to improve transmission efficiency, thereby generating a gear ratio adjustment coefficient.

[0117] Among them, the transmission efficiency optimization model is trained with a large amount of experimental and simulation data of transmission systems, and can accurately reflect the relationship between transmission efficiency and various mechanical parameters and gear interval ratios.

[0118] (3) Using the following formula, based on the gear ratio adjustment coefficient, the gear interval ratio parameter is incrementally optimized using an adaptive smoothing algorithm to generate the optimized gear interval ratio parameter:

[0119]

[0120] in, This indicates the optimized gear interval ratio parameter. This represents the initial gear interval ratio parameter vector. This represents element-wise multiplication. Indicates the adaptive smoothing coefficient. This represents the transmission efficiency loss function. Indicates the mechanical wear factor. Indicates the start time for optimization. Indicates the optimization termination time. Represents a time variable. This indicates the gear interval ratio parameter.

[0121] For example, based on the gear ratio adjustment coefficient, an adaptive smoothing algorithm is used to smooth the gear ratio adjustment coefficient, avoiding overly drastic adjustments to the gear interval ratio parameter due to sudden changes in the coefficient. Combining the transmission efficiency loss function, the mechanical wear factor, and the time range from the optimization start time to the optimization end time, element-wise multiplication is performed on the initial gear interval ratio parameter vector. At each time point, based on the current transmission efficiency loss and mechanical wear condition, each element in the initial gear interval ratio parameter vector is adjusted incrementally. Through continuous iterative optimization, the optimized gear interval ratio parameter is generated.

[0122] The adaptive smoothing algorithm employs methods such as moving averages to process the adjustment coefficients, ensuring a smooth parameter adjustment process. The transmission efficiency loss function is constructed based on the difference between the theoretical optimal value and the actual value of transmission efficiency, used to quantify the degree of transmission efficiency loss and guide the adjustment direction of the gear interval ratio parameter. The mechanical wear factor is obtained through monitoring and evaluating the wear degree of mechanical components, reflecting the impact of mechanical wear on transmission efficiency and making the adjustment more consistent with the actual use of the machinery.

[0123] like Figure 2 As shown, the gear shifting operation corresponding to the adaptive gear shift decision command is executed through a preset closed-loop control unit, including:

[0124] S201: Based on the closed-loop control unit, the adaptive gear decision command is parsed into target gear parameters and switching trajectory constraints.

[0125] For example, based on preset instruction parsing rules, the encoding format and parameter identifiers of the adaptive gear decision instruction are interpreted to extract target gear parameters, such as the specific value of the target gear and the corresponding transmission ratio. Simultaneously, switching trajectory constraints are extracted, including the allowed speed variation range and acceleration peak limit during gear switching, thus clarifying the target and constraint boundaries for subsequent gear switching operations.

[0126] S202: Generate a drive signal sequence that matches the target gear parameters based on the dynamic sliding diaphragm control algorithm.

[0127] For example, the dynamic sliding diaphragm control algorithm is based on the theory of sliding diaphragm variable structure control and combines it with the dynamic model of the gear actuator. By calculating the sliding diaphragm surface and control law of the system, it generates a drive signal sequence that matches the target gear parameters. The drive signal sequence includes the values ​​and variation patterns of control signals such as voltage and current acting on the gear actuator at different times, so as to drive the actuator to move towards the target gear.

[0128] The dynamic model includes the inertia and damping characteristics of the actuator. Gear shifting components include servo motors and transmission mechanisms.

[0129] S203: The driving signal sequence is processed by a progressive execution strategy to generate segmented execution instructions.

[0130] For example, based on the phased characteristics of gear shifting, the drive signal sequence is divided into multiple execution phases, and the start time, duration, and signal strength gradient of the drive signal in each phase are determined. Through the above timing scheduling, the continuous drive signal sequence is transformed into phased control commands, generating segmented execution commands, so that the gear shifting process proceeds step by step and smoothly.

[0131] S204: Based on the switching trajectory constraints, execute the gear shifting action corresponding to the segmented execution command, and collect the vibration feedback signal of the mechanical transmission system in real time.

[0132] For example, based on switching trajectory constraints, the gear shifting action corresponding to the segmented execution command is executed. During execution, the gear shifting mechanism gradually completes the mechanical actions of gear shifting according to the segmented execution command, such as gear meshing and drive belt adjustment. Vibration sensors installed at key parts of the mechanical transmission system are used to collect vibration feedback signals of the mechanical transmission system in real time during the gear shifting process. These signals include information such as vibration frequency and amplitude, which are used to reflect the smoothness of the gear shifting action and the operating status of the mechanical system.

[0133] Key components of the mechanical transmission system include gearboxes and drive shafts.

[0134] S205: Perform online correction processing on the drive signal sequence based on the vibration feedback signal until the gear shifting operation is completed.

[0135] For example, the vibration feedback signal is filtered and its features are extracted. The processed signal is then compared with the switching trajectory constraints and the vibration characteristics during normal gear shifting. If the vibration signal exceeds the allowable range, it indicates a deviation in the current gear shifting action. The control unit corrects the control parameters in the drive signal sequence online according to the type and degree of the deviation. The corrected drive signal sequence then drives the gear actuator again, and this process is repeated iteratively until the gear shifting operation is completed, ensuring that gear shifting is both precise and maintains the smooth operation of the mechanical system.

[0136] Control parameters include signal amplitude and frequency.

[0137] In one embodiment, a lightweight neural network model is used to process the standardized feature vector for cycling pattern recognition, and the current cycling pattern category is output, including:

[0138] (1) Perform multimodal feature decoupling on the standardized feature vectors to generate velocity feature subspace, pressure feature subspace, attitude feature subspace and environment feature subspace.

[0139] For example, multimodal feature decoupling processing is performed on the standardized feature vectors, including reading the attribute labels of each dimension of the features in the standardized feature vectors and dividing the standardized feature vectors into dimensions based on the feature attributes. Feature dimensions directly related to cycling speed are grouped together to construct a speed feature subspace; feature dimensions related to pedaling pressure and transmission pressure are grouped together to construct a pressure feature subspace; feature dimensions related to vehicle tilt angle and center of gravity position are grouped together to construct a posture feature subspace; and feature dimensions related to external temperature and road slope are grouped together to construct an environmental feature subspace. Through the above division process, it is ensured that the features within each feature subspace are highly correlated, and that the features in different subspaces are independent of each other.

[0140] Among them, feature attribute labels are pre-labeled category identifiers for each dimension of the standardized feature vector, used to clarify the type of cycling information reflected by the feature and to provide a basis for dimensional division. Feature attributes include categories such as motion state, mechanics, posture, and environment.

[0141] (2) Temporal dependencies in each feature subspace of velocity feature subspace, pressure feature subspace, posture feature subspace and environment feature subspace are captured by bidirectional long short-term memory network to generate temporal enhanced feature vectors.

[0142] For example, a bidirectional long short-term memory (LSTM) network is used, where the velocity feature subspace, pressure feature subspace, attitude feature subspace, and environmental feature subspace are input into four independent processing branches of the network. Each branch's LSM network includes a forward temporal processing unit and a backward temporal processing unit. The forward temporal processing unit extracts temporal correlation information from the start to the end of the feature sequence, such as the velocity variation from low to high. The backward temporal processing unit extracts temporal correlation information from the end to the start of the feature sequence, such as the pressure decay from high to low. The outputs of the two units are concatenated and fused to generate temporally enhanced feature vectors corresponding to each feature subspace.

[0143] The four independent processing branches maintain consistent structures and parameters to ensure uniform standards for extracting temporal information from different feature subspaces and avoid feature deviations caused by branch differences.

[0144] (3) Spatial correlation modeling of temporal enhancement feature vectors in cross-modal dimensions is performed by three-dimensional convolution kernel to generate fusion vector feature tensors.

[0145] For example, a three-dimensional convolutional kernel is constructed, with dimensions corresponding to the feature channel dimension, time series dimension, and modal interaction dimension, respectively. The temporal enhancement feature vectors generated from each feature subspace are concatenated in modal order to generate a multimodal temporal feature matrix, which is then input into the three-dimensional convolutional layer. The three-dimensional convolutional kernel captures the correlation between different features within the same modality in the feature channel dimension, enhances temporal continuity in the time series dimension, and uncovers the collaborative relationships between features of different modalities at the same time node in the modal interaction dimension. Through multiple rounds of convolutional operations and feature mapping, the multimodal temporal feature matrix is ​​transformed into a fused vector feature tensor.

[0146] (4) Dynamically weighted aggregate the fusion feature tensor based on the attention weight allocation mechanism to generate pattern recognition feature vector.

[0147] For example, an attention weight allocation mechanism is introduced to perform global average pooling on the fused feature tensor, obtaining a global feature description vector for the fused feature tensor. This global feature description vector is input into a fully connected layer, and an activation function is used to calculate the attention weight values ​​corresponding to each spatial location of the fused feature tensor. The magnitude of the weight value represents the contribution of the feature at that location to cycling pattern recognition. The attention weight values ​​are then multiplied element-wise with the fused feature tensor, strengthening feature regions with high contribution and suppressing feature regions with low contribution. Finally, the weighted fused feature tensor is globally summed to generate a dimensionality-compressed pattern recognition feature vector.

[0148] Global average pooling transforms a high-dimensional tensor into a low-dimensional vector by calculating the average value of features at each spatial location of the fused feature tensor, thus simplifying the subsequent calculation of attention weights.

[0149] (5) The pattern recognition feature vector is classified by a preset fully connected classifier, and the current cycling mode category is output.

[0150] For example, a pre-defined fully connected classifier is invoked, and the pattern recognition feature vector is input into the classifier's input layer. The pattern recognition feature vector is passed from the input layer to the hidden layer, where linear transformations and non-linear activation functions are used to further abstract and filter the features, extracting more discriminative pattern features. The output of the hidden layer is passed to the output layer, where a softmax activation function is used to map the pattern features to probability values ​​corresponding to various cycling modes. The cycling mode with the highest probability value is selected as the classification result, and the current cycling mode category is output.

[0151] The pre-defined fully connected classifier is trained using a large amount of sample data labeled with cycling mode categories, and its internal weight parameters have been optimized to accurately distinguish between different cycling modes, such as flat road cruising mode, mountain climbing mode, and urban commuting mode.

[0152] In one embodiment, the motion parameters and environmental parameters of the bicycle are collected in real time using a multimodal sensor array, including:

[0153] (1) Obtain wheel speed pulse signals through speed sensors.

[0154] For example, a speed sensor is installed on a rotating part of the bicycle wheel or at a corresponding position on the frame to ensure that the sensor can capture physical signals during the wheel's rotation in real time. The speed sensor generates a pulse signal every time the wheel rotates by a specific angle, using electromagnetic induction or photoelectric induction principles, and continuously collects this pulse signal to obtain the wheel speed pulse signal.

[0155] The speed sensor should be installed away from areas of severe wheel vibration to ensure the stability and accuracy of the collected wheel speed pulse signal, which can be used to calculate the bicycle's riding speed.

[0156] (2) The change in resistance of the crank strain gauge is collected by a pressure sensor.

[0157] For example, a pressure sensor is connected to a strain gauge on a bicycle crank, ensuring the strain gauge is in close contact with the crank surface. When the user pedals the crank, the crank undergoes a slight deformation, causing a change in the resistance value of the strain gauge. The pressure sensor monitors and records the change in the strain gauge resistance value in real time, converts the resistance change information into an acquireable data form through signal conversion, and then acquires the amount of change in crank strain gauge resistance.

[0158] Among them, the strain gauge needs to be selected according to the material and stress characteristics of the crank to ensure that it can sensitively reflect the deformation of the crank under different pedaling forces. The change in resistance of the crank strain gauge can be used to indirectly judge the magnitude of the user's pedaling force.

[0159] (3) Measure the Euler angle and angular velocity of the vehicle body using an attitude sensor.

[0160] For example, the attitude sensor is fixed to a stable part of the bicycle frame, ensuring that the sensor's measurement axis is aligned with the bicycle's coordinate system. The attitude sensor incorporates a gyroscope and an accelerometer. The gyroscope measures the angular velocity of the bicycle's rotation around each coordinate axis in real time, while the accelerometer, combined with gravitational acceleration information, calculates the Euler angles of the bicycle in three-dimensional space. The measurement data is processed through a data fusion algorithm within the sensor to obtain the bicycle's Euler angles and angular velocity.

[0161] (4) Detect obstacle distance information through environmental perception sensors.

[0162] For example, an environmental sensing sensor is mounted on the front of the bicycle handlebars or frame, ensuring that the sensor's detection direction is towards the area in front of the bicycle's path. The environmental sensing sensor emits a detection signal and receives the reflected signal after it encounters an obstacle. Based on the time difference or phase difference between signal transmission and reception, combined with the signal propagation speed, it calculates the distance between the obstacle and the bicycle. This detection process is continuously executed, and obstacle distance information is detected.

[0163] The environmental perception sensor's detection range needs to cover the main area along the bicycle's direction of travel, while filtering out interference signals from non-obstacles such as the ground and small stones. Obstacle distance information can be used to help identify potential obstacles in the cycling environment. Environmental perception sensors include ultrasonic sensors and lidar sensors.

[0164] (5) Motion parameters include wheel speed pulse signal, crank strain gauge resistance change, vehicle body Euler angle and angular velocity; environmental parameters include obstacle distance information.

[0165] For example, the collected data is categorized and organized as follows: the wheel speed pulse signals obtained by the speed sensor, the crank strain gauge resistance changes obtained by the pressure sensor, and the Euler angles and angular velocities of the vehicle body measured by the attitude sensor are grouped into one category and collectively referred to as motion parameters; the obstacle distance information detected by the environmental perception sensor is grouped into a separate category and referred to as environmental parameters. Through this categorization and organization, motion parameters and environmental parameters are clearly distinguished and obtained.

[0166] The classification and sorting process must retain the collection timestamps of each parameter to ensure the correspondence between motion parameters and environmental parameters in the time dimension, providing a data foundation for subsequent multi-parameter fusion analysis.

[0167] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0168] In one embodiment, such as Figure 3 As shown, this application also provides a bicycle gear adaptive decision-making system 300 based on multimodal sensor cooperative processing, the system 300 including:

[0169] The multi-source sensing module 301 is used to collect the bicycle's motion parameters and environmental parameters in real time through a multi-modal sensor array;

[0170] The data normalization module 302 is used to perform spatiotemporal alignment and noise filtering on motion parameters and environmental parameters, and generate standardized feature vectors.

[0171] The pattern recognition module 303 is used to perform cycling pattern recognition processing on the standardized feature vector through a lightweight neural network model and output the current cycling pattern category.

[0172] The strategy decision module 304 is used to perform adaptive strategy matching processing based on the preset gear configuration template library and the current riding mode category to generate adaptive gear decision instructions.

[0173] The execution control module 305 is used to execute the gear switching operation corresponding to the adaptive gear decision command through a preset closed-loop control unit.

[0174] Specifically, the multi-source sensing module 301 uses a multi-modal sensor array to collect wheel speed pulse signals, crank strain gauge resistance changes, vehicle body Euler angles and angular velocities, and obstacle distance information, respectively. These data are then aggregated and converted into digital signals to obtain the bicycle's motion and environmental parameters. The motion parameters include wheel speed pulse signals, crank strain gauge resistance changes, and vehicle body Euler angles and angular velocities, while the environmental parameters include obstacle distance information.

[0175] The data normalization module 302 acquires motion parameters and environmental parameters, performs spatiotemporal alignment based on timestamps, filters noise signals using filtering algorithms, standardizes parameters to unify numerical scales, and generates standardized feature vectors.

[0176] The pattern recognition module 303 inputs the standardized feature vector into the lightweight neural network model, and sequentially completes multimodal feature decoupling, temporal dependency capture, cross-modal space modeling, and dynamic weighted aggregation. Through a fully connected classifier, it makes a decision and outputs the current cycling mode category.

[0177] The strategy decision module 304 calls the preset gear configuration template library, matches the current riding mode category to obtain the initial gear strategy, optimizes the gear switching threshold by combining the user's historical riding data, optimizes the gear interval ratio parameter according to the mechanical characteristics of the bicycle, updates the strategy and converts it into a control signal to generate an adaptive gear decision command.

[0178] The execution control module 305 inputs the adaptive gear decision command into the closed-loop control unit, parses the command, generates a drive signal sequence, and schedules it into a segmented execution command. When the gear shifting action is executed, vibration feedback signals are collected, and the drive signals are corrected online until the gear shifting operation corresponding to the adaptive gear decision command is completed.

[0179] The strategy decision module 304 is also used for:

[0180] Call the gear configuration template in the preset gear configuration template library. The gear configuration template includes gear switching threshold and gear interval ratio parameters.

[0181] Based on the current riding mode category, a dynamic strategy is matched to generate an initial strategy matching result;

[0182] The gear shifting threshold is dynamically optimized based on the user's historical riding data to generate an optimized gear shifting threshold.

[0183] The gear interval ratio parameter is dynamically optimized based on the mechanical characteristics of the bicycle to generate the optimized gear interval ratio parameter.

[0184] The initial strategy matching result is updated based on the optimized gear shifting threshold and the optimized gear interval ratio parameter, and the strategy matching result is generated.

[0185] Adaptive gear selection instructions are generated based on the strategy matching results.

[0186] The strategy decision module 304 is also used for:

[0187] Extract cycling mode distribution characteristics and gear efficiency indicators from users' historical cycling data to generate cycling habit analysis data;

[0188] Based on cycling habit analysis data, an adaptive optimization strategy is matched to generate threshold adjustment parameters;

[0189] Using the following formula, the gear shifting threshold is dynamically and iteratively adjusted based on the threshold adjustment parameters through an incremental learning mechanism to generate an optimized gear shifting threshold:

[0190]

[0191] in, This indicates the optimized gear shift threshold. Indicates the initial gear shift threshold. Indicates the incremental learning rate. This represents the number of historical cycling data samples. Indicates the first Time decay weights for each sample This indicates the efficiency index of gear positions. This indicates a pattern matching indicator function.

[0192] The strategy decision module 304 is also used for:

[0193] Extract real-time mechanical state parameters of bicycle transmission to generate a mechanical characteristic dataset;

[0194] Based on the preset transmission efficiency optimization model, the mechanical characteristic dataset is dynamically analyzed and processed to generate gear ratio adjustment coefficients.

[0195] Using the following formula, based on the gear ratio adjustment coefficient, an adaptive smoothing algorithm is used to incrementally optimize the gear interval ratio parameter, generating the optimized gear interval ratio parameter:

[0196]

[0197] in, This indicates the optimized gear interval ratio parameter. This represents the initial gear interval ratio parameter vector. This represents element-wise multiplication. Indicates the adaptive smoothing coefficient. This represents the transmission efficiency loss function. Indicates the mechanical wear factor. Indicates the start time for optimization. Indicates the optimization termination time. Represents a time variable. This indicates the gear interval ratio parameter.

[0198] The execution control module 305 is also used for:

[0199] Based on the closed-loop control unit, the adaptive gear decision command is parsed into target gear parameters and switching trajectory constraints;

[0200] A drive signal sequence matching the target gear parameters is generated based on a dynamic sliding diaphragm control algorithm;

[0201] A progressive execution strategy is used to perform time-series scheduling on the driving signal sequence to generate segmented execution instructions;

[0202] Based on the switching trajectory constraints, the gear shifting action corresponding to the segmented execution command is executed, and the vibration feedback signal of the mechanical transmission system is collected in real time.

[0203] The drive signal sequence is corrected online based on the vibration feedback signal until the gear shifting operation is completed.

[0204] The pattern recognition module 303 is also used for:

[0205] Multimodal feature decoupling is performed on the standardized feature vectors to generate velocity feature subspace, pressure feature subspace, attitude feature subspace and environment feature subspace;

[0206] By using a bidirectional long short-term memory network, the temporal dependencies in each feature subspace of velocity, pressure, attitude and environment are captured respectively, and temporal enhanced feature vectors are generated.

[0207] Spatial correlation modeling of temporal augmentation feature vectors across modal dimensions is performed using 3D convolutional kernels to generate fused vector feature tensors;

[0208] Dynamically weighted and aggregated fused feature tensors based on an attention weight allocation mechanism to generate pattern recognition feature vectors;

[0209] The current cycling mode category is output by classifying the pattern recognition feature vectors using a pre-defined fully connected classifier.

[0210] The multi-source sensing module 301 is also used for:

[0211] Wheel speed pulse signals are acquired using a speed sensor;

[0212] The change in resistance of the crank strain gauge is collected using a pressure sensor;

[0213] The Euler angle and angular velocity of the vehicle body are measured using an attitude sensor;

[0214] Detects obstacle distance information using environmental perception sensors;

[0215] Motion parameters include wheel speed pulse signals, crank strain gauge resistance changes, vehicle body Euler angles, and angular velocities; environmental parameters include obstacle distance information.

[0216] In one embodiment, a matrix of strain gauges is symmetrically arranged on both sides of a bicycle crank as a carrier. Based on the principle of resistance strain, the strain gauges can detect the force and minute deformation of the crank during the rider's pedaling process in real time, converting mechanical strain into resistance change signals to ensure dynamic capture of subtle changes in pedaling pressure. A six-axis IMU module is integrated near the bottom bracket of the crank. This module includes a three-axis gyroscope and a three-axis accelerometer, which can accurately measure the angular velocity and linear acceleration during crank rotation and simultaneously acquire related data on cadence, crank motion trajectory, and vehicle posture. In addition, the signals from the strain gauge group and the six-axis IMU module are connected to the crank PCB control board via wires. The control board is equipped with signal conditioning circuitry and a data preprocessing unit to filter, amplify, and convert the raw signals collected by the sensors into analog and digital data. Then, the processed pressure data, cadence data, and posture data are wirelessly transmitted to the central main controller via Bluetooth Low Energy protocol.

[0217] In real-world cycling scenarios, the pressure-sensing crank module can collect multi-dimensional data in real time, including the rider's pedaling pressure distribution, cadence changes, and crank rotation angle. After receiving this data, the central controller combines it with speed data from the speed sensor and uses a lightweight neural network model deployed on the central controller to perform cycling mode recognition and gear selection analysis. When the system detects increased pedaling pressure and decreased cadence, it triggers the gear ratio switching system to perform fully automatic gear shifting, adjusting to a more suitable gear to reduce the rider's energy expenditure. Simultaneously, this data is also synchronized to a WeChat mini-program, displaying the rider's pedaling efficiency and energy distribution through a data visualization interface. This data also provides support for a smart cycling advisor system to generate personalized training plans and dietary guidance.

[0218] By deeply integrating multiple sensors with the bicycle crank, not only is high-precision and high-frequency collection of cycling data achieved, breaking the functional limitations of the traditional crank as only a transmission component, but it also provides core data sources for the system's safety warnings, intelligent shifting, and personalized health management, effectively improving the safety, efficiency, and intelligent experience of cycling.

[0219] In one embodiment, the bicycle's front and rear lights serve as an integrated carrier, embedding a high-sensitivity millimeter-wave radar module within the lights. This radar module can scan the bicycle's surroundings in real time, detecting moving or stationary obstacles within its range, and simultaneously capturing environmental information such as the relative distance and trajectory of surrounding objects to the bicycle. The lights themselves employ a light source structure that conforms to lighting standards, offering multiple lighting modes including daytime, nighttime, rain, and fog modes. They can automatically adjust the light color and brightness according to ambient light and riding scenario, balancing basic lighting needs with scene adaptation requirements.

[0220] In terms of hardware connectivity, the millimeter-wave radar module and the headlight control circuit are connected to the system's central main controller, forming a data interaction and control linkage mechanism. During riding, the radar module continuously collects data on the surrounding environment and transmits the data to the central main controller. The main controller analyzes the data using a preset safety warning algorithm to identify potential dangerous targets (such as approaching vehicles, pedestrians, road obstacles, etc.). When a dangerous target is detected entering the warning range, the central main controller simultaneously sends a control command to the headlights, triggering the headlights' audible and visual warning function—transmitting visual warning signals through LED breathing flashing (different flashing frequencies correspond to different danger levels), while simultaneously triggering an audible reminder through the headlights' built-in buzzer. This dual warning ensures that the rider is aware of the risk in a timely manner.

[0221] Furthermore, in nighttime riding mode, when the radar module detects pedestrians or non-motorized vehicles ahead, the headlights automatically dim the high beams and switch to low beams to avoid glare. In rainy or foggy weather, the radar module uses an adaptive threshold algorithm to eliminate rain and fog interference, ensuring obstacle detection accuracy. Simultaneously, the headlights switch to a high-penetration lighting mode to improve visibility and environmental awareness. This integrated design of radar and headlights not only simplifies the device's installation structure and avoids the clutter of multiple components, but also achieves seamless coordination between environmental perception and safety warnings, significantly improving safety and reliability in various riding scenarios.

[0222] In one embodiment, an inductive high-precision proximity switch is used as the core sensing component. The proximity switch is installed at the end of the bicycle's rear fork, with its detection end corresponding to the brake disc and maintaining a preset detection distance from the perforated metal plate on the brake disc. The proximity switch operates based on the principle of electromagnetic induction. When the bicycle wheel rotates, the brake disc rotates synchronously with the wheel, and the perforated metal plate on the brake disc periodically approaches or moves away from the detection end of the proximity switch. When the metal plate approaches, the internal magnetic field of the proximity switch changes, triggering the switch to output an electrical pulse signal. When the metal plate moves away, the magnetic field returns to normal, and the pulse signal terminates. Through this process, the proximity switch generates a pulse signal for every specific angle the wheel rotates, achieving non-contact acquisition of the speed signal.

[0223] The pulse signals acquired by the proximity switch are transmitted to the central main controller (based on an ESP32S3 microcontroller) via wires. The speed calculation module inside the main controller processes the pulse signals, including registering an interrupt callback function for the proximity switch signal reading pin and capturing the rising and falling edges of the pulse signals in real time. In a loop, the time difference between two adjacent pulse signals is calculated. If the current time difference is greater than the previously calculated time difference, the instantaneous speed is calculated based on the current time difference; otherwise, the previous time difference is used to ensure the stability of the speed calculation. The calculated speed value is then converted to a unit that conforms to the system's data standard.

[0224] In real-world cycling scenarios, this high-precision proximity switch speed measurement system can output stable speed data in real time. Firstly, the speed data is synchronized to a WeChat mini-program, displaying real-time speed, average speed, and speed trend through a data visualization interface, allowing users to intuitively view their riding status. Secondly, the speed data is fused with cadence and posture data collected by the intelligent pressure-sensing crank module and input into a lightweight neural network model (LSTM+CNN hybrid network). This provides key parameters for the system to identify riding modes such as climbing, cruising, and sprinting, while also providing speed-based information for the gear ratio switching system—when the system detects a continuous increase in speed and stable cadence, it triggers a fully automatic gear shift, adjusting to a higher gear to improve riding efficiency. Furthermore, this speed measurement system has strong anti-interference capabilities, operating stably in complex environments such as rain and sandstorms, ensuring the accuracy and continuity of speed data collection and providing fundamental data support for the stable operation of the entire intelligent cycling system.

[0225] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0226] In one embodiment, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0227] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0228] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A bicycle gear adaptive decision-making method based on multimodal sensor collaborative processing, characterized in that, The method includes: The bicycle's motion parameters and environmental parameters are collected in real time using a multimodal sensor array. The motion parameters and environmental parameters are spatiotemporally aligned and noise filtered to generate a standardized feature vector. The standardized feature vector is processed by a lightweight neural network model to identify the cycling mode, and the current cycling mode category is output. Based on the preset gear configuration template library and the current riding mode category, an adaptive strategy matching process is performed to generate an adaptive gear decision instruction. The gear shifting operation corresponding to the adaptive gear decision command is executed by a preset closed-loop control unit.

2. The bicycle gear adaptive decision-making method based on multimodal sensor collaborative processing according to claim 1, characterized in that, The adaptive strategy matching process based on the preset gear configuration template library and the current riding mode category to generate adaptive gear decision instructions includes: Call the gear configuration template in the preset gear configuration template library. The gear configuration template includes a gear switching threshold and a gear interval ratio parameter. Based on the current riding mode category, a dynamic strategy is matched to generate an initial strategy matching result; The gear shifting threshold is dynamically optimized based on the user's historical riding data to generate an optimized gear shifting threshold. The gear interval ratio parameter is dynamically optimized based on the mechanical characteristics of the bicycle to generate the optimized gear interval ratio parameter. The initial strategy matching result is updated based on the optimized gear shifting threshold and the optimized gear interval ratio parameter to generate a strategy matching result. The adaptive gear selection instruction is generated based on the strategy matching result.

3. The bicycle gear adaptive decision-making method based on multimodal sensor collaborative processing according to claim 2, characterized in that, The step of dynamically optimizing the gear shifting threshold based on the user's historical riding data to generate an optimized gear shifting threshold includes: The cycling mode distribution characteristics and gear efficiency indicators are extracted from the user's historical cycling data to generate cycling habit analysis data. Based on the cycling habit analysis data, an adaptive optimization strategy is matched to generate threshold adjustment parameters; Using the following formula, based on the threshold adjustment parameters, the gear shifting threshold is dynamically and iteratively adjusted through an incremental learning mechanism to generate the optimized gear shifting threshold: in, This indicates the optimized gear shift threshold. Indicates the initial gear shift threshold. Indicates the incremental learning rate. This represents the number of historical cycling data samples. Indicates the first Time decay weights for each sample This indicates the efficiency index of gear positions. This indicates a pattern matching indicator function.

4. The bicycle gear adaptive decision-making method based on multimodal sensor collaborative processing according to claim 2, characterized in that, The dynamic optimization of the gear interval ratio parameters based on the mechanical characteristics of the bicycle to generate optimized gear interval ratio parameters includes: Extract real-time mechanical state parameters of bicycle transmission to generate a mechanical characteristic dataset; The mechanical characteristic dataset is dynamically analyzed and processed based on a preset transmission efficiency optimization model to generate gear ratio adjustment coefficients. Using the following formula, based on the gear ratio adjustment coefficient, the gear interval ratio parameter is incrementally optimized using an adaptive smoothing algorithm to generate the optimized gear interval ratio parameter: in, This indicates the optimized gear interval ratio parameter. This represents the initial gear interval ratio parameter vector. This represents element-wise multiplication. Indicates the adaptive smoothing coefficient. This represents the transmission efficiency loss function. Indicates the mechanical wear factor. Indicates the start time for optimization. Indicates the optimization termination time. Represents a time variable. This indicates the gear interval ratio parameter.

5. The bicycle gear adaptive decision-making method based on multimodal sensor collaborative processing according to claim 1, characterized in that, The step of executing the gear shifting operation corresponding to the adaptive gear shift decision command through a preset closed-loop control unit includes: Based on the closed-loop control unit, the adaptive gear decision command is parsed into target gear parameters and switching trajectory constraints. A drive signal sequence matching the target gear parameters is generated based on a dynamic sliding diaphragm control algorithm; The driving signal sequence is processed by a progressive execution strategy to generate segmented execution instructions. Based on the switching trajectory constraints, the gear shifting action corresponding to the segmented execution command is executed, and the vibration feedback signal of the mechanical transmission system is collected in real time. The drive signal sequence is corrected online based on the vibration feedback signal until the gear shifting operation is completed.

6. The bicycle gear adaptive decision-making method based on multimodal sensor collaborative processing according to claim 1, characterized in that, The process of using a lightweight neural network model to perform cycling mode recognition on the standardized feature vector and outputting the current cycling mode category includes: The standardized feature vectors are subjected to multimodal feature decoupling processing to generate velocity feature subspace, pressure feature subspace, attitude feature subspace and environment feature subspace; A bidirectional long short-term memory network is used to capture the temporal dependencies in each feature subspace of the velocity feature subspace, the pressure feature subspace, the attitude feature subspace, and the environment feature subspace, respectively, and generate temporally enhanced feature vectors. Spatial correlation modeling of the temporal enhancement feature vector is performed on the cross-modal dimension using a three-dimensional convolutional kernel to generate a fused vector feature tensor; The fused feature tensor is dynamically weighted and aggregated based on an attention weight allocation mechanism to generate a pattern recognition feature vector. The pattern recognition feature vector is classified using a pre-defined fully connected classifier, and the current cycling mode category is output.

7. The bicycle gear adaptive decision-making method based on multimodal sensor collaborative processing according to claim 1, characterized in that, The method of real-time acquisition of bicycle motion parameters and environmental parameters through a multimodal sensor array includes: Wheel speed pulse signals are acquired using a speed sensor; The change in resistance of the crank strain gauge is collected using a pressure sensor; The Euler angle and angular velocity of the vehicle body are measured using an attitude sensor; Detects obstacle distance information using environmental perception sensors; The motion parameters include the wheel speed pulse signal, the change in resistance of the crank strain gauge, the Euler angle of the vehicle body, and the angular velocity; the environmental parameters include the obstacle distance information.

8. A bicycle gear adaptive decision-making system based on multimodal sensor collaborative processing, characterized in that, The system includes: The multi-source sensing module is used to collect the bicycle's motion parameters and environmental parameters in real time through a multi-modal sensor array; The data normalization module is used to perform spatiotemporal alignment and noise filtering on the motion parameters and the environmental parameters to generate standardized feature vectors. The pattern recognition module is used to perform cycling pattern recognition processing on the standardized feature vector using a lightweight neural network model and output the current cycling pattern category. The strategy decision module is used to perform adaptive strategy matching processing based on the preset gear configuration template library and the current riding mode category to generate adaptive gear decision instructions. The execution control module is used to execute the gear switching operation corresponding to the adaptive gear decision command through a preset closed-loop control unit.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the bicycle gear adaptive decision-making method based on multimodal sensor cooperative processing as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the bicycle gear adaptive decision-making method based on multimodal sensor cooperative processing as described in any one of claims 1 to 7.

Images