Method and system for adaptive generation of assistance curve based on riding habit learning

Abstract

The application discloses a power assisting curve self-adaptive generation method and system based on riding habit learning, relates to the field of intelligent control of electric power assisting bicycles, and comprises the following steps: establishing a riding habit library based on historical riding data of target riders, and storing a pedaling force habit center and a pedaling frequency habit center representing long-term pedaling behavior characteristics of the riders; collecting multi-source real-time sensing data such as pedal torque, pedaling frequency, vehicle driving state and battery state in real time; constructing a comfort envelope model through the pedaling force habit center and the pedaling frequency habit center, and calculating a comfort coefficient; generating comprehensive control indexes in combination with the comfort coefficient, the vehicle driving state and the battery state data, dynamically adjusting the shape parameters of a basic power assisting function, generating a self-adaptive power assisting curve matched with a current riding scene, and then determining a motor power assisting torque instruction and outputting the motor power assisting torque instruction to a motor controller. In this way, the individualized coordination and coupling of man-machine torque aiming at natural pedaling force performance of the riding user habits are realized.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology for electric-assisted bicycles, and in particular to an adaptive generation method and system for assist curves based on riding habit learning. Background Technology

[0002] Electric-assist bicycles provide users with a more relaxed riding experience through a combination of human pedaling and motor assistance. To determine the motor's output intensity, the control system typically needs to access sensor information such as pedaling force / cadence, speed, gradient, motor current, and battery charge, and convert this information into motor torque or current commands, thus forming what is known as the "assist curve / assist strategy."

[0003] Currently, most mainstream electric-assist bicycles offer several "assist modes / gears" with preset fixed assist curves (e.g., commuting, leisure, exercise, hill climbing), or use a threshold and ratio stacking strategy to amplify the human torque by a fixed factor. Users manually select from a limited set of modes, and the system outputs the desired result based on a template. However, this static control method cannot perceive the individual differences in rider's physical characteristics and is difficult to adapt to complex dynamic road conditions. This leads to problems such as insufficient assist when starting on inclines and excessive assist when riding on flat roads, seriously affecting riding comfort. Summary of the Invention

[0004] This application provides a method, system, storage medium, computer program product, and electronic device for adaptive generation of assist curves based on cycling habit learning, in order to at least solve the problem that the assist control strategy in the current related technology lacks the ability to adapt to individual differences of cyclists and dynamic cycling scenarios, resulting in a mismatch between assist output and actual needs.

[0005] In a first aspect, embodiments of this application provide an adaptive assist curve generation method based on cycling habit learning. The method includes: establishing a cycling habit database based on the historical cycling data of a target cyclist; the cycling habit database storing a pedal force habit center and a cadence habit center characterizing the long-term pedaling behavior of the cyclist; collecting multi-source real-time sensor data of an electric-assisted bicycle during a real-time control cycle; the multi-source real-time sensor data including pedal torque, cadence, vehicle driving status data, and battery status data; constructing a comfort envelope model based on the pedal force habit center and the cadence habit center, calculating a comfort coefficient based on the deviation of the pedal torque and the cadence from the comfort envelope model; generating a comprehensive control index by combining the comfort coefficient, the vehicle driving status data, and the battery status data, and dynamically adjusting the morphological parameters of a preset basic assist function using the comprehensive control index to generate an adaptive assist curve matching the current cycling scenario; substituting the pedal torque into the dynamically adjusted adaptive assist curve to determine the motor assist torque command, and outputting it to the motor controller to drive the motor to output assist.

[0006] Secondly, embodiments of this application provide an adaptive generation system for assist curves based on cycling habit learning. The system includes: a habit database construction unit, used to establish a cycling habit database based on the historical cycling data of a target cyclist; the cycling habit database stores pedal force habit centers and cadence habit centers characterizing the long-term pedaling behavior of a cyclist; a multi-source data acquisition unit, used to acquire multi-source real-time sensor data of the electric-assisted bicycle during a real-time control cycle; the multi-source real-time sensor data includes pedal torque, cadence, vehicle driving status data, and battery status data; and a comfort calculation unit, used to construct a... A comfort envelope model calculates a comfort coefficient based on the deviation of the pedal torque and cadence from the comfort envelope model. A dynamic modulation unit for the assist curve is used to combine the comfort coefficient, vehicle driving status data, and battery status data to generate a comprehensive control index, and to dynamically adjust the morphological parameters of a preset basic assist function using the comprehensive control index to generate an adaptive assist curve that matches the current riding scenario. An assist torque command output unit is used to substitute the pedal torque into the dynamically adjusted adaptive assist curve to determine the motor assist torque command, and output it to the motor controller to drive the motor to output assist.

[0007] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the assist curve adaptive generation method based on cycling habit learning according to any embodiment of the present application.

[0008] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, characterized in that, when the program is executed by a processor, it implements the steps of the adaptive generation method for assist curve based on cycling habit learning according to any embodiment of this application.

[0009] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the adaptive generation method for assist curve based on cycling habit learning according to any embodiment of this application.

[0010] The adaptive generation method and system for assist curves based on cycling habit learning provided in this application can achieve at least the following technical effects: (1) An “individual benchmark” is formed based on the rider’s historical pedaling behavior, and a comfort envelope that can constrain the pedaling force-cadence coordination relationship is constructed based on this benchmark, so that the assist adjustment is transformed from a uniform template-driven relationship to an “individual behavior reference-driven relationship”. Since the real-time pedal torque and cadence are mapped into this envelope and quantified into a comfort coefficient by the degree of deviation, the control system obtains feedback that can directly reflect the current pedaling state relative to the individual comfort range. This allows for the automatic formation of differentiated assist responses among different riders, and suppresses the drift of pedaling experience when the same rider’s state fluctuates (such as fatigue or rhythm changes), thereby improving the consistency and controllability of pedaling rhythm and assist response.

[0011] (2) The comfort coefficient is coupled with the vehicle driving state and battery state to generate a comprehensive control index, and the shape parameters of the basic assist function are dynamically shaped accordingly to achieve synchronous self-adaptation of the assist curve with operating conditions and energy constraints. Since the adjustment object is the curve shape parameter rather than discrete mode switching or fixed multiple amplification, the assist output can change continuously with the scenario and maintain the consistency of the command generation link, thereby reducing the risk of sudden output changes during the switching of operating conditions such as starting, climbing, and cruising; at the same time, after the battery state is included in the same index system, the assist intensity and change rate can be controlled to adjust when the energy capacity changes, thereby improving the output stability and experience consistency of the whole vehicle under different energy conditions.

[0012] This technical solution constructs a comfort envelope based on individual riding habits, and integrates comfort feedback with vehicle / battery status to dynamically adjust the assist curve parameters. This enables the assist control to continuously adapt to individual differences and dynamic operating conditions. Consequently, the assist torque command can better match the current riding scenario and stably approach the rider's comfortable pedaling range, achieving a more consistent, smoother, and more controllable assist experience. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart is shown as an example of an adaptive generation method for assist curves based on cycling habit learning according to an embodiment of this application; Figure 2 A diagram illustrating the operational mechanism of the adaptive generation method for assist curves based on cycling habit learning according to an embodiment of this application is shown. Figure 3 This diagram illustrates a comparison of the output characteristics of the adaptive boost curve provided in this embodiment of the invention with those of the traditional fixed-rate boost mode. Figure 4 This paper presents a heat map showing the difference in assist output between the adaptive assist strategy and the traditional balance mode under different combinations of pedal force and slope provided in the embodiments of this application. Figure 5 This paper presents a time-domain response comparison diagram of adaptive assist torque and cumulative battery energy consumption during simulated complex road conditions riding, as provided in an embodiment of this application. Figure 6 A structural block diagram of an example of an adaptive generation system for assist curves based on cycling habit learning, according to an embodiment of this application, is shown. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] It should be noted that current electric assist control solutions primarily focus on meeting basic power assistance needs, but still have limitations in terms of personalized adaptation, multi-dimensional information fusion, and intelligent energy management. Specifically:

[0017] Early solutions for controlling assist intensity often relied on Hall effect sensors or simple speed sensors. Some studies indicate that traditional triggering methods typically start the motor when the magnetic element in the bottom bracket passes the Hall element, causing a sudden step change in current. This discontinuous control not only easily causes mechanical shock to the internal reduction gears, affecting the lifespan of the transmission system, but also, because the system cannot accurately sense changes in the rider's cadence, the motor output often exhibits intermittent characteristics, making it difficult to provide a smooth riding experience. Although some improved solutions have begun to incorporate multi-sensor data, optimizing the control logic by integrating speed, cadence, pedal torque, and gradient information—for example, setting different assist ratios based on the user's selected economy, balance, or boost mode—the core logic remains a linear calculation based on instantaneous sensor values ​​and a preset fixed ratio, lacking the ability to deeply learn and adaptively adjust to the rider's long-term riding habits.

[0018] In terms of cycling intention recognition, some related technologies attempt to infer the rider's acceleration needs using kinematic parameters such as the difference in rotational speed between the front and rear wheels. These solutions typically determine acceleration intention by monitoring changes in the flywheel's rotational acceleration relative to the front wheel, thereby increasing the assist output. While this method is structurally simple and effectively prevents false triggering from a stationary position, its limitations lie in its inverse reasoning from kinematic results. It cannot directly obtain the magnitude of the rider's pedaling force or their real-time physical condition. Furthermore, it fails to fully integrate multi-dimensional constraints such as road gradient and remaining battery power for comprehensive decision-making, making accurate intention tracking difficult.

[0019] In terms of intelligent and personalized experiences, current commercial products generally adopt preset, template-based assist modes. For example, systems typically offer multiple fixed modes such as commuting, leisure, exercise, or hill climbing for users to manually switch between. While these modes enrich riding options to some extent, such as providing continuous strong assist in commuting mode or intermittent weak assist in exercise mode, they are essentially still based on a static logical template of "gear-time-gradient." This kind of "one-size-fits-all" control strategy ignores the significant individual differences among cyclists in terms of age, physical fitness, pedaling frequency preferences, and force application habits, and cannot maintain optimal riding comfort and efficiency in dynamically changing road conditions.

[0020] Furthermore, with increasing attention to healthy cycling, some scholars have begun to explore closed-loop control systems based on physiological signal feedback. For example, some studies have proposed using e-bikes as rehabilitation training platforms, monitoring heart rate in real time and adjusting motor output using proportional-derivative (PD) control algorithms to maintain the cyclist's heart rate within a specific medical prescription range. However, the initial design intent of such systems primarily serves medical rehabilitation purposes, with the control objective being the steady-state control of physiological indicators, rather than comfort optimization based on the cyclist's natural pedaling habits. They also lack the ability to analyze and remember long-term cycling behavior characteristics. Meanwhile, in the field of energy management, although there are studies on energy optimization based on route planning aimed at eliminating cyclists' "range anxiety," these studies mostly focus on distance-level energy allocation strategies, failing to consider the cyclist's micro-pedaling habits (such as force distribution and cadence preference) as core factors in energy management. This makes it difficult to achieve refined optimization of range while ensuring personalized comfort.

[0021] In summary, current technologies in the field of electric assist control mainly face challenges such as a lack of personalized learning, insufficient depth of multi-source information fusion, and the difficulty in balancing energy management and comfort. Most systems remain at the stage of rule-based control based on single or limited sensor signals, failing to construct dynamic assist curves that can sense and adapt to the rider's long-term habits and real-time status. This results in difficulties achieving an ideal balance between riding comfort, power responsiveness, and energy efficiency in complex and ever-changing human-vehicle-road environments.

[0022] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0023] The technical solutions in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information, comply with relevant laws and regulations and do not violate public order and good morals.

[0024] Figure 1 A flowchart illustrating an example of an adaptive generation method for assist curves based on cycling habit learning according to an embodiment of this application is shown.

[0025] Regarding the execution subject of the method in the embodiments of this application, it can be any controller or processor with computing or processing capabilities, such as a microprocessor, digital signal processor (DSP) or field-programmable gate array (FPGA), which implements the various steps of the method by executing program code stored on a computer-readable medium.

[0026] In some examples, the controller or processor can be integrated into the main control unit, smart terminal (such as mobile phone, watch, vehicle display) or dedicated power assist control device of the electric-assist bicycle through software, hardware or a combination of software and hardware. The control and computing module can exchange data with various sensors of the electric-assist bicycle (such as pedal torque sensor, cadence sensor, speed sensor, slope sensor, etc.) through the wireless communication module and control the motor.

[0027] Furthermore, the types of terminals or electronic devices can be diverse, including but not limited to smartphones, in-vehicle displays, smartwatches, computers, or dedicated bicycle controllers, and these devices can be customized to meet the needs of different cyclists.

[0028] like Figure 1 As shown, in step S110, a cycling habit database is established based on the target cyclist's historical cycling data. The cycling habit database stores pedal force habit centers and cadence habit centers that characterize the cyclist's long-term pedaling behavior.

[0029] In some implementations, the system first acquires historical cycling data accumulated by the target cyclist over a period of time through an onboard storage unit or a cloud server. This data accurately records the cyclist's natural pedaling performance under different road conditions and physical fitness levels. Subsequently, the processor uses statistical analysis or machine learning algorithms to extract features from this unstructured time-series data, identifying the cyclist's most frequently used pedal torque and cadence ranges.

[0030] By analyzing long-term behavioral patterns based on a cycling habit database, the system can abstract characteristic values ​​from discrete historical samples that represent the cyclist's "most comfortable" or "most habitual" force exertion state, namely the pedaling force habit center and cadence habit center. These two center values ​​are not merely arithmetic means, but rather characterize the biomechanical balance point that the cyclist tends to maintain unconsciously. Solidifying and storing these characteristic parameters in the cycling habit database provides a personalized reference benchmark for subsequent real-time calculations of comfort deviations, thus avoiding the "one-size-fits-all" control experience caused by the universal template used in traditional e-bikes.

[0031] In step S120, during the real-time control cycle, multi-source real-time sensor data of the electric-assist bicycle is collected. The multi-source real-time sensor data includes pedal torque, cadence, vehicle driving status data, and battery status data.

[0032] Here, during the real-time control cycle, the system collects multi-source real-time sensor data from various sensors on the e-bike, and uses an embedded control unit to synchronously collect signals from various sensors via a high-speed bus. More specifically, torque and cadence sensors capture the rider's instantaneous force and pedaling frequency in real time, directly reflecting the rider's input intention and exercise intensity; the inertial measurement unit (IMU) or wheel speed sensor provides the vehicle's speed, acceleration, and attitude information to calculate the vehicle's driving status data (such as whether it is currently uphill, starting, or cruising); simultaneously, the battery management system (BMS) uploads battery status data such as remaining charge, voltage, and temperature. Through the collaborative work of these multi-source data, the system can comprehensively understand the rider's current pedaling status and vehicle dynamics, and continuously optimize the assist output based on real-time data.

[0033] In step S130, a comfort envelope model is constructed based on the pedal force habit center and the cadence habit center, and the comfort coefficient is calculated according to the deviation of pedal torque and cadence in the comfort envelope model.

[0034] Here, the system transforms the user's "personalized habits" into quantifiable "mathematical evaluation indicators." By accessing the pedaling force and cadence habit centers in the cycling habit database, and using these as anchor points, a comfort envelope model is constructed in a two-dimensional plane or multi-dimensional feature space. This comfort envelope model can define the rider's cadence and pedaling force distribution using a Gaussian distribution or other mathematical models to reflect the rider's most frequently used cadence and pedaling force ranges. By comparing the deviation between real-time data and the habit centers, a comfort coefficient is calculated, indicating whether the current pedaling state is within the rider's comfort range. Specifically, when cadence and pedaling force deviate significantly from the habit centers, the comfort coefficient decreases, and the system needs to adjust the assist output to restore a comfortable state. Conversely, when cadence and pedaling force are close to the habit centers, the comfort coefficient is higher, indicating that the current state is close to the rider's long-term habits (i.e., in a comfortable state).

[0035] In step S140, a comprehensive control index is generated by combining the comfort coefficient, vehicle driving status data and battery status data. The morphological parameters of the preset basic assist function are dynamically adjusted using the comprehensive control index to generate an adaptive assist curve that matches the current riding scenario.

[0036] In this embodiment, the system does not adjust solely based on comfort level. Instead, it weights and fuses the comfort coefficient with vehicle driving conditions (such as slope resistance) and battery conditions (such as range anxiety) to generate a comprehensive control index that balances human-machine comfort and system constraints. Using this index, the system dynamically corrects the morphological parameters of the preset basic assist function (i.e., the mathematical model describing the mapping relationship between input torque and output torque).

[0037] It should be noted that the correction mechanism in this embodiment is not a simple gear shift, but rather a continuous and smooth reshaping of the assist curve's slope (response sensitivity), inflection point (intervention timing), or amplitude (maximum output). For example, when the overall indicators show that the rider feels strained and the battery is fully charged, the system automatically steepens the assist curve, allowing even a smaller pedaling force to trigger a larger motor output; conversely, it flattens the curve. Through this dynamic adjustment, the system generates an adaptive assist curve specific to the current moment and scenario, achieving a technological upgrade in assist strategy from "static lookup and coarse mode switching" to "dynamic generation and fine-tuning of morphological characteristics."

[0038] In step S150, the pedal torque is substituted into the dynamically adjusted adaptive assist curve to determine the motor assist torque command, and then output to the motor controller to drive the motor to output assist.

[0039] In some implementations, the system uses the real-time collected pedal torque as an input variable, substituting it into the adaptive assist curve updated and generated according to the real-time control cycle (or the current control cycle). Through function mapping, it calculates the theoretically required motor assist torque command for the current moment. This command is then sent to the motor controller (such as an FOC controller), which adjusts the switching state of the power devices, precisely controls the motor current and phase, and drives the motor to output the corresponding auxiliary torque. This ensures that the final driving force acting on the wheels reflects the rider's pedal input intention, while also incorporating the auxiliary torque calculated through adaptive assist. This maintains good consistency between the motor output torque and the rider's pedal torque in terms of amplitude and trend, thereby achieving personalized and coordinated coupling of human-machine torque to reflect the natural pedaling force performance oriented towards the cycling user's habits.

[0040] Regarding the implementation details of real-time acquisition of multi-source sensing parameters in step S120, in some examples of embodiments of this application, the pedal torque sequence and pedal frequency sequence are acquired at a preset sampling frequency by deploying torque sensors and pedal frequency sensors, and the three-axis acceleration signal and three-axis angular velocity signal of the vehicle body are acquired by inertial measurement unit.

[0041] In this embodiment, the real-time acquisition system constructs a comprehensive perception of the human-vehicle-road system through a heterogeneous sensor array. First, torque sensors and cadence sensors deployed at the central axis position are used at a preset high sampling frequency (e.g., Real-time acquisition of pedal torque sequence and cadence sequence High-frequency sampling can fully capture the rider's dead points and peak force characteristics within a single pedaling cycle, recording the rider's load changes at every instant.

[0042] Meanwhile, the onboard inertial measurement unit (IMU) integrates a three-axis accelerometer and a three-axis gyroscope, outputting three-axis acceleration signals in the vehicle coordinate system in real time. and triaxial angular velocity signals Among them, the acceleration signal reflects the linear motion of the vehicle body and the gravitational component, while the angular velocity signal reflects the rotational dynamics of the vehicle body.

[0043] Then, the attitude calculation algorithm is used to perform multi-sensor fusion processing on the three-axis acceleration signal and the three-axis angular velocity signal, and zero-point drift compensation is performed to calculate the vehicle pitch angle, which is then converted into the road slope in the vehicle driving state data.

[0044] It should be noted that, in order to address the physical limitation that a single accelerometer cannot distinguish between "motion acceleration" and "gravity component" when a vehicle is accelerating or decelerating, this embodiment uses an attitude calculation algorithm (preferably using complementary filtering or Kalman filtering) to fuse acceleration and angular velocity.

[0045] Specifically, the high dynamic characteristics of the gyroscope are used to respond to rapid attitude changes, while the low-frequency characteristics of the accelerometer are used to correct the integral drift of the gyroscope. Vehicle pitch angle. The solution logic can be represented by the following complementary filtering model:

[0046] Equation (1) In the formula, This is a weighting factor (e.g., 0.98). To control the periodic sampling interval, This represents the angular velocity around the horizontal axis. Through this fusion algorithm, the system can filter out road vibration noise and decouple motion acceleration, thereby obtaining accurate road slope information (i.e.,...). Characterized by an uphill slope, (Characterizing downhill slopes). Therefore, by introducing a fusion algorithm of acceleration and angular velocity (complementary filtering formula), the accuracy of slope perception is significantly improved.

[0047] Subsequently, based on the timestamp of the pedal torque sequence, timing alignment was performed on the pedal frequency sequence and road slope, and matched with the battery state of charge and motor temperature uploaded by the battery management system to construct multi-source real-time sensor data with a unified time base.

[0048] It should be noted that due to phase differences in the sampling clocks of different sensors, the system performs strict timing alignment. Specifically, the system uses the timestamps of the pedal torque sequence... Using the baseline as the principal axis, the tread frequency data sampled at low frequencies and the calculated slope data Perform linear interpolation or nearest neighbor matching to synchronize it to time.

[0049] Based on this, the system obtains the battery state of charge from the battery management system (BMS) via the CAN bus. and motor temperature Ultimately, a multi-source real-time sensing feature vector with a unified time base is constructed. : Equation (2) This feature vector can be directly used by subsequent comfort calculation and decision-making assistance modules.

[0050] Furthermore, anomaly detection based on physical thresholds is performed on multi-source real-time sensor data. If at least one sensor data is missing or out of bounds, a degradation control strategy is triggered, using the effective value of the previous control cycle or a preset safety value to replace the current abnormal data, so as to ensure the continuity of the assist torque command.

[0051] In this embodiment, to ensure the system's robustness in the event of sensor failure or extreme environments, the system performs anomaly detection based on physical thresholds on multi-source real-time sensor data. Specifically, the system presets the effective range for each physical quantity. For example, the effective torque range is [0, 150] Nm, and the effective SOC range is [0, 100]%.

[0052] When any sensor data is detected Missing data (e.g., packet loss) or out-of-bounds values ​​(e.g.) When this occurs, the system immediately triggers a degradation control strategy, utilizing the effective value from the previous control cycle or a preset safety value. By replacing the current abnormal data, the continuity of the generated assist torque command is ensured, avoiding riding hazards caused by sudden changes in motor output due to sensor signal jumps, and greatly improving the reliability of the system.

[0053] Regarding the implementation details of establishing the cycling habit database in step S110, in some examples of the embodiments of this application, historical cycling data is parsed into a continuous time-series data stream, and the time-series data stream is divided into multiple independent cycling sample segments based on a preset time window or preset cycling event. Pedal torque sample sequence and cadence sample sequence are extracted from each cycling sample segment.

[0054] In some implementations, the establishment of a cycling habit database begins with the cleaning and reconstruction of unstructured historical data. This involves first parsing the raw historical data in memory into a continuous time-series data stream, and then introducing valid cycling events as segmentation logic. Specifically, the processor sets a preset time window (e.g., 5 minutes) or a physical state-based event trigger (e.g., continuous pedaling for more than 30 seconds with an average speed greater than 5 km / h) to segment the continuous data stream into several independent cycling sample segments, excluding traffic light waiting or gliding states. For each sample segment The system extracts the pedal torque sample sequences respectively. With cadence sample sequence This process eliminates invalid, static, or low-quality data, ensuring that subsequent statistical analysis focuses solely on the cyclist's actual behavioral characteristics during active exertion, thus improving the signal-to-noise ratio of habit recognition.

[0055] Then, statistical distribution analysis was performed on the pedal torque sample sequence and the pedal frequency sample sequence respectively. The expected value representing the central tendency of the data was extracted as the pedal force habit center and the pedal frequency habit center, and the standard deviation or variance representing the fluctuation range of the data was extracted as the pedal force dispersion parameter and the pedal frequency dispersion parameter.

[0056] Here, for the extracted pedal torque and cadence sample sequences, the system performs statistical distribution analysis to quantify the rider's behavioral patterns, focusing primarily on two core statistical dimensions: central tendency and dispersion. Specifically, the system calculates the expected value (or mean) of the sample sequence as the center of pedaling habit. With cadence habit center It represents the "comfortable power output point" and "comfortable rhythm" that cyclists subconsciously tend to maintain; at the same time, the standard deviation of the sample sequence is calculated as a parameter of pedal force dispersion. With cadence dispersion parameter This characterizes a cyclist's "tolerance range" or "fluctuation habit" to changes in pedal force and cadence. Through the parameterized description of these two dimensions, the system concretizes the abstract "cycling habit" into numerical features that can be processed by a computer, providing the necessary mathematical benchmark for constructing a personalized comfort envelope.

[0057] Furthermore, a parameter update model based on the time decay forgetting factor is constructed. The statistical characteristics of cycling sample segments at different time points are weighted and fused to update the pedal force habit center, cadence habit center, pedal force dispersion parameter, and cadence dispersion parameter. The cycling sample segments closer to the current time have a higher weight contribution in the update, so as to achieve adaptive tracking of the cyclist's recent habit changes. The updated pedal force habit center, cadence habit center, pedal force dispersion parameter, and cadence dispersion parameter are synchronously stored in the cycling habit database.

[0058] Here, in order to address the issue of user habits drifting over time (such as physical fitness growth and seasonal changes), an online update strategy based on a time decay forgetting factor is introduced. Instead of statically storing historical averages, the system constructs a recursive update model with dynamic weights.

[0059] Specifically, for any statistical characteristic parameter (It can represent) When the first Statistical characteristics of a new cycling sample segment At that time, the system utilizes the forgetting factor Perform the following weighted merge update: Equation (3) In the formula, The historical values ​​stored in the current habit library, This is the updated value. Forgetting factor. This determines the system's "memory strength" of historical data: The smaller the value, the more sensitive the system is to new data and the faster it forgets old data. Through this mechanism, sample segments closer to the current time have a higher weight contribution in the update, thus achieving adaptive tracking of recent changes in cyclists' habits. This prevents the "rigidity" of the profile model, allowing the assistance strategy to dynamically evolve along with the user's growth or change of status, always maintaining the best match with the user's current state.

[0060] After completing the weighted fusion calculation, the system will update the four core parameters, namely the pedaling habit center. cadence habit center , pedal force dispersion parameter and cadence dispersion parameters The system synchronously writes the cycling habit database to non-volatile memory, overwriting older parameter records. This updated habit database serves as the latest benchmark for comfort calculations in the next control cycle, ensuring the system retains the most up-to-date user feedback even after a power outage and restart.

[0061] Regarding the implementation details of the above-mentioned statistical distribution analysis, in some examples of the embodiments of this application, a Gaussian kernel function is used as a smoothing operator for the pedal torque sample sequence. With cadence sample sequence Construct continuous probability density functions respectively: Equation (4) In the formula, This represents the total number of samples in the sample sequence. and Let represent the independent variables of the probability density functions for pedal torque and cadence, respectively. and They represent the first in the sequence. The first pedal torque sample value and the first One cadence sample value and These are the bandwidth parameters for pedal torque and cadence, respectively. For Gaussian kernel function, and Let represent the probability density functions of pedal torque and pedal frequency, respectively.

[0062] In this embodiment, to accurately capture the non-parametric behavioral characteristics of cyclists, the system abandons the simple arithmetic mean method and instead adopts the non-parametric kernel density estimation (KDE) method. Specifically, using the mathematical model shown in equation (4), a Gaussian kernel function is used... As a smoothing operator, it transforms the discrete pedal torque sample sequence and cadence sample sequence Transform into a continuous probability density function and .

[0063] Here, each historical sample point or They are all considered to be the center of a miniature Gaussian distribution, with bandwidth parameters and This determines the "fatness" (i.e., smoothness) of these micro-distributions. In equation (4), by... By linearly superimposing and normalizing a miniature Gaussian distribution, a "probabilistic topographic map" that can truly reflect the density distribution of the data is constructed. This effectively smooths out random noise in the sample while preserving the multimodal characteristics of the data (e.g., cyclists may have two commonly used cadence ranges), and can realistically restore the distribution of cyclists' behavioral habits.

[0064] Then, the constructed probability density function is integrated to solve for the first-order raw moments and the second-order central moments: Equation (5) Equation (6) In the formula, and They represent the calculated mathematical expectations, and These represent the calculated standard deviations.

[0065] Here, after constructing a continuous probability density function, the system needs to collapse it into key feature parameters that can be used to control the algorithm. This process is achieved by performing the integral operations shown in equations (5) and (6).

[0066] Specifically, the system first solves for the first-order raw moment (i.e., the mathematical expectation) of the probability density function, by considering the variable values ​​over the entire domain. or ) and its corresponding probability density value ( or Integrate the product of the two products to calculate the "center of gravity" of the data, i.e., the center of pedaling habit. and cadence habit center .

[0067] Subsequently, based on the determined center value, the system further solves for the second-order central moments (i.e., variance) and takes the square root to obtain the standard deviation, which is the square of the deviation in equations (5) and (6). The weighted integration is then performed. This step physically quantifies the dispersion or fluctuation range of the cyclist's behavior data. In practical embedded processor implementations, the above integration operation is usually approximated using a discretized Riemann Sum or numerical integration algorithm, thus achieving a mathematical definition of the cyclist's "comfort center" and "comfort bandwidth" with extremely low computational cost.

[0068] Furthermore, the calculated results and The centers of pedal force habituation and pedal frequency habituation were determined respectively, and the calculated values ​​were... and These are respectively defined as pedal force dispersion parameter and pedal frequency dispersion parameter.

[0069] Here, the system performs physical semantic mapping on the statistical feature values ​​calculated above. The calculated mathematical expectation... and These were directly identified as the center of pedaling force habit and the center of cadence habit, representing the most natural and energy-efficient points of force application and frequency for cyclists during long-term riding; the calculated standard deviation... and These parameters, known as pedal force dispersion and pedal frequency dispersion, define the "flexible boundary" of changes in cyclist habits.

[0070] Through the embodiments of this application, complex historical behavior data is successfully compressed into four simplified control parameters, providing a precise mathematical basis for the subsequent construction of a personalized two-dimensional Gaussian comfort envelope model, and ensuring that the power assist control strategy can closely match the rider's real physiological characteristics.

[0071] Regarding the implementation details of calculating the comfort coefficient in step S130, in some examples of embodiments of this application, the system models the cyclist's "comfort zone" as a probability distribution space. Specifically, it calls four core parameters stored in the cycling habit library, namely, pedal force habit center. cadence habit center , pedal force dispersion parameter and cadence dispersion parameters , construct The geometric center (i.e., the peak point of the probability density) and with and A two-dimensional Gaussian membership function model is used for the distribution scale parameter (i.e., the parameter controlling the width of the mountain peak), and this model is defined as the comfort envelope model. Geometrically, this comfort envelope model forms a surface resembling a "mountain peak," with the apex corresponding to the cyclist's most comfortable state. The steepness of the slope is determined by the dispersion parameter: the greater the dispersion, the stronger the cyclist's adaptability to changes in that dimension, and the gentler the "comfort slope"; conversely, the smaller the dispersion, the steeper the slope. Thus, a soft-boundary comfort zone is constructed using statistical properties, avoiding the jumps in judgment caused by traditional hard thresholds (such as "cadence > 80 is comfortable").

[0072] Then, obtain the real-time pedal torque for the current control cycle. With real-time cadence Real-time pedal torque With real-time cadence Substitute into a two-dimensional Gaussian membership function model to solve for the comfort coefficient. : Equation (7) In the formula, It is an exponential function; comfort coefficient It is used to quantify the degree to which the current pedaling state belongs to the cyclist's long-term comfort zone, and its value range is (0, 1]. The closer the value is to 1, the more the current state is in line with the cyclist's habits.

[0073] In equation (7), the numerator term and The squared Euclidean distance of the current state relative to the habit center was calculated; the denominator term... and It serves as a standardization and weighting mechanism, scaling the deviation based on the rider's habit dispersion—for dispersion In smaller dimensions (where habits are very fixed), the same absolute deviation leads to a larger calculated value, thus not only eliminating the dimensional differences between torque and frequency but also reflecting the different weights of the impact of different dimensions on comfort; using an exponential function The monotonically decreasing property of the above-mentioned standardized deviation values ​​maps them to the interval (0, 1]. That is, when the deviation is 0, As the deviation increases, the exponential term tends towards negative infinity. It rapidly decays towards 0.

[0074] It should be understood that the calculated comfort coefficient It is a dimensionless normalized scalar that intuitively quantifies the degree to which the current pedaling state belongs to the cyclist's long-term comfort zone. When When the value is close to 1, it indicates that the current pedaling force and cadence are almost entirely within the rider's habitual center, meaning the rider is in the most relaxed and natural riding state; when When the value of is significantly reduced (e.g., less than 0.5), it indicates that the current state deviates severely from the habitual envelope (e.g., being forced to pedal heavily at low frequencies while climbing, or idling at high frequencies on flat ground), suggesting that the cyclist may be in a state of discomfort, fatigue, or abnormal exertion. Therefore, by employing a quantization method based on probabilistic distance, the complex bivariate physiological state is successfully compressed into a simple control variable.

[0075] Regarding the implementation details of generating the comprehensive control index in step S140, in some examples of the embodiments of this application, based on the comfort envelope model, the standardized pedal force deviation and standardized pedal frequency deviation, which constitute the basis for calculating the comfort coefficient, are extracted. Road slope and battery state of charge are introduced to construct a multidimensional linear weighted model including habit deviation, slope compensation, and battery charge constraint terms to calculate the comprehensive control index. : Equation (8) In the formula, For road slope, The battery is in its state of charge. Indicates the real-time slope weight. Indicates real-time battery power weight; and These represent the pedal torque weight and cadence weight, respectively. They are determined based on the relative magnitudes of the pedal torque stability index and cadence stability index in the target cyclist's historical riding data. The stability index is used to characterize the steady-state degree of the corresponding pedal feature dimension.

[0076] In this embodiment, a multidimensional linear weighted model as shown in equation (8) is constructed to fuse perceptual data from different dimensions into a unified decision variable. This model first calculates the standardized tread force deviation based on the comfort envelope parameters in step S130. Deviation from standardized cadence .

[0077] Here, the Z-Score standardization concept from statistics is utilized, along with the dispersion parameter. The dimensional difference between pedal torque (Nm) and cadence (RPM) is eliminated, making their deviations numerically comparable. Subsequently, the model incorporates a road gradient term. With power constraints Among them, the power constraint term utilizes This value indicates the "level of battery depletion," and the lower the battery level, the larger the value.

[0078] By performing a linear weighted summation using equation (8), the system generates comprehensive control indices. This indicator is a highly integrated dimensionless scalar that represents the "comprehensive urgency of the need for assist adjustment" under current operating conditions: it includes the degree to which the rider deviates from their comfort zone, as well as the power requirements for overcoming slope resistance and the energy-saving constraints under low battery conditions.

[0079] It should be noted that in equation (8), the weight of the pedal force is... With cadence weight The settings are not fixed but depend on the rider's historical stability. Specifically, the system extracts the stability index (i.e., steady-state percentage or the reciprocal of fluctuation) of pedal torque and cadence from historical data. If the rider is very stable in pedal force control (high stability index), it means they are very sensitive to changes in pedal force, and any slight deviation in pedal force should be taken seriously by the system, thus assigning a higher value. Conversely, if the cyclist's cadence fluctuates (low stability index), the system will appropriately reduce the cadence. To avoid erratic fluctuations in cadence that could affect control indicators. Unnecessary oscillations occur; therefore, by adopting a stability-based weighting mechanism, the control system is ensured to understand the feedback dimensions that the rider cares about most.

[0080] Preferably, to resolve the conflict between range anxiety and power performance, the system executes a dynamic weighted scheduling strategy based on energy state. Specifically, during the calculation process, the dynamic weighted scheduling strategy based on energy state is executed: a low battery threshold is set. (e.g., 20%), real-time monitoring ,when At that time, calculate the energy regulation gain. And perform nonlinear correction on the weighting coefficients:

[0081] Equation (9) Equation (10) In the formula, It is an exponential function. The energy sensitivity coefficient, and These are the preset base weights for electricity consumption and slope, respectively.

[0082] An exponential function is introduced in the calculation of the energy regulation gain in equation (9). Due to the characteristics of the exponential function, as... Further below the threshold, This will result in explosive growth. Furthermore, in equation (10), the system utilizes this gain to couple and adjust the weights: on the one hand, it significantly amplifies the power weight. This led to "energy conservation constraints" quickly becoming the dominant control indicator. The core factor; on the other hand, through Synchronous suppression of gradient weight means that the system actively ignores some gradient compensation requests when the battery is low. In this way, through the nonlinear control logic of "one rise and one fall", the mathematical model perfectly interprets the physical principle of "sacrificing some gradient comfort in order to maintain range" for the intelligent decision-making process.

[0083] In some cases, when detected season and keep and This is to ensure the continuity of the weight scheduling strategy at the threshold.

[0084] It should be noted that, in order to prevent control strategies from being implemented... A sudden change occurs at the critical point when sufficient power is detected ( When ), the system forces an order. This ensures that each weight remains at a preset base value (i.e., This ensures the mathematical continuity of the weight scheduling strategy at the threshold.

[0085] In addition, the standardized pedal force deviation Deviation from standardized cadence The positive and negative signs are used to characterize the direction of deviation of the real-time pedaling state from the habit center, so that the comprehensive control index It carries both deviation magnitude information and deviation direction information.

[0086] It is worth noting that in the calculation During the process, the system retains the sign of the standardized deviation. Positive signs (e.g.) A positive sign indicates that the cyclist's current pedaling weight is too heavy or the cadence is too fast, while a negative sign indicates that the pedaling weight is too light or the cadence is too slow. This directional information is not absoluteized during the weighting process but is fully preserved in the comprehensive control index. This allows subsequent adjustments to the assist function to not only detect "how much deviation there is," but also "in which direction the deviation is occurring," thus achieving refined adaptive control in each direction.

[0087] Regarding the implementation details of generating the adaptive assist curve in step S140, in some examples of embodiments of this application, a parameterized Logistic function is constructed as the mathematical model of the basic assist function, and the comprehensive control index is integrated. Mapped to motor assist coefficient : Equation (11) In the formula, and These are the real-time minimum assist coefficient and the real-time maximum assist coefficient, respectively, and satisfy the following conditions: ; Let be the slope gain parameter, and satisfy . ; This refers to the inflection point offset parameter.

[0088] In this embodiment, the system abandons the traditional linear amplification or piecewise lookup table, and instead constructs a parameterized Logistic function (S-shaped function) as the mathematical model of the basic assist function. In this mathematical model, the Logistic function naturally has the nonlinear characteristics of "gradual start, linear enhancement in the middle, and high load saturation", which is very consistent with the physiological perception characteristics of human muscle exertion.

[0089] In equation (11), and The dynamic range of the assist coefficient is defined, corresponding to the minimum and maximum assist forces, respectively; the denominator of the exponent term contains... The slope gain parameter, in its physical sense, determines the "steepness" of the boost curve or the response sensitivity, i.e. The larger the value, the more drastic the change in the contribution caused by a change in the unit indicator. This is the inflection point offset parameter, which determines the timing of the intervention of the assist curve on the horizontal axis, i.e. The smaller the value, the more significant the assistance provided by the system at a lower overall performance level. Thus, the complex assistance strategy can be abstracted into a set of dynamically adjustable morphological parameters.

[0090] Then, in order to achieve an adaptive effect that "matches the current cycling scenario," the system executes a multi-dimensional parameter modulation strategy based on comfort and energy state, and based on the comfort coefficient... With battery state of charge Dynamic adjustments are made to the morphological parameters: On the one hand, when the amount of comfort deviation is detected When increasing, increase the slope gain parameter. and / or reduce the inflection point offset parameter This is to improve the sensitivity of the adaptive assist curve to changes in the overall control index.

[0091] Here, the aim is to perform morphological reshaping; when the calculated comfort deviation... An increase in slope gain indicates that the cyclist is currently in an uncomfortable zone (e.g., overexertion). In this case, the system increases the slope gain parameter through algorithmic logic. and / or reduce the inflection point offset parameter This improves the responsiveness of the assist curve and allows for earlier intervention, enabling riders to achieve greater motor response with less effort and quickly return to their comfort zone.

[0092] On the other hand, when the battery state of charge is detected The power level drops below a preset threshold and / or a power weight is detected. When the power is increased, the real-time maximum assist coefficient is dynamically reduced. This helps to compress the upper limit of the output in the high power consumption range.

[0093] Here, the aim is to enforce boundary constraints; when a battery is detected... Reduced to below a preset threshold, or the energy weight calculated in the previous steps. When significantly enhanced, the system dynamically reduces the real-time maximum assist coefficient. This adjustment compresses the output limit in the high-power range without changing the assist in the low-load range, thereby enforcing an energy-saving strategy while ensuring basic riding performance.

[0094] Furthermore, an adaptive boost curve is determined based on the dynamically adjusted morphological parameters, and a continuity constraint is applied to the morphological parameters to ensure a smooth transition of the generated boost coefficient in the time domain; wherein, the continuity constraint is used to independently limit the rate of change of the morphological parameters in adjacent control cycles: Equation (12) In the formula, To control the cycle length, For the first The values ​​of each morphological parameter during the real-time control cycle. This is the preset maximum change step size threshold for the corresponding parameter; where, It remains unchanged during the real-time control cycle.

[0095] It should be noted that, since the above parameter modulation is based on real-time sampling data, in order to prevent drastic changes in the shape of the assist curve due to sensor noise or sudden changes in road conditions (which would cause a jerking sensation in the motor output), the system imposes strict continuity constraints on the shape parameters. Specifically, the system uses equation (12) to adjust the adjacent control cycles (time intervals) The rate of change of the morphological parameters is subject to independent limiting.

[0096] In equation (12), the parameter value at the current time is calculated. Compared with the parameter value at the previous time step The absolute value of the difference is determined, and this difference is required to not exceed a preset maximum change step size threshold. This constraint covers the slope. ,inflection point and upper limit of amplitude This ensures that the adaptive assist curve is a smooth, gradual transition rather than an abrupt change in the time domain, thus guaranteeing the smoothness of the motor's output torque, eliminating mechanical shock, and improving the riding experience. It should be noted that the real-time minimum assist coefficient... It is not subject to this constraint because it serves as the lower limit of the motor output, ensuring that the rider always receives at least a basic level of assistance. It also reduces the complexity of system calculations and improves control stability.

[0097] As a preferred embodiment of this application, the method of this embodiment further includes an online self-learning update step for parameters, which is used to continuously optimize the assist response characteristics based on the real-time pedaling behavior feedback of the cyclist, so as to give the control system the ability to "evolve itself".

[0098] Specifically, first, a target comfort level representing the ideal load state of the cyclist is set. And calculate the comfort coefficient in each control cycle. Comfort error between target comfort : Equation (13) Here, the system sets a target comfort level. (It can be set to 1.0 or a value close to 1.0), which represents the rider's most ideal and least strenuous physiological load state. Within each control cycle, the real-time comfort coefficient is calculated using equation (13). Comfort error between the target value and the target value .

[0099] like This indicates that the current level of comfort is insufficient (e.g., the rider feels strained), and the system needs to improve the responsiveness of the assist; if This indicates that the current parameters are already optimal, and the system should maintain its current state. By introducing this error signal, it becomes the driving force for subsequent parameter iterations, transforming the original open-loop parameter adjustment into a closed-loop negative feedback control system.

[0100] Then, based on comfort error For slope gain parameter With inflection point offset parameter Perform error-driven online iterative updates to gradually bring the parameter combination closer to the target response characteristics that match the cyclist's current physical condition: Equation (14) Equation (15) In the formula, This represents the slope gain parameter updated in the next control cycle. This indicates the inflection point offset parameter after the next control cycle update. and These are the learning rates for the slope gain parameter and the inflection point offset parameter, respectively.

[0101] In equation (14), the slope gain parameter is... The system performs error-driven gradient descent updates. In this formula, the update term consists of three parts: the learning rate... Controlling convergence speed, error The decision was made to adjust the direction, and the item This reflects the derivative characteristic of the Logistic function. Specifically, only when the comprehensive control index... Deviation from inflection point When changing the slope Only then will it have a significant impact on the output; if Located precisely near the inflection point, changing the slope has a negligible impact on the function value. Therefore, we introduce... This product term automatically weights the update magnitude based on the current operating point's position, increasing the update magnitude when the operating point is in the non-linear sensitive region of the curve and decreasing it when it is in the central linear region. This ensures that the slope adjustment always occurs at the moment that most significantly affects the boost output characteristics, thereby improving the algorithm's convergence efficiency.

[0102] Regarding the inflection point offset parameter The system performs a reverse iterative update as shown in equation (15). The negative sign in the formula... This reveals its physical regulation logic: when comfort error When the value is positive (i.e., the user feels discomfort / exhaustion), the system needs to reduce... At this point, shifting the center of the Logistic curve to the left (towards lower indicators) lowers the threshold for intervention, meaning that a smaller overall control index is required. A slight decrease can trigger a larger boost output. Conversely, if comfort is excessive (e.g., ... If the value is negative, although the actual upper limit of comfort is 1, a brief overshoot is allowed internally by the algorithm (through negative feedback callbacks), the system will increase. This delays intervention. Thus, through incremental updates over cycles, the system can dynamically identify the most suitable intervention time for the cyclist's current physical condition.

[0103] In this embodiment, by employing the aforementioned online self-learning update steps, a "personalized and constantly evolving" control effect is achieved. Unlike traditional fixed-parameter PID control, this solution can automatically compensate for model mismatch issues caused by rider fatigue, changes in road conditions, or long-term drift in personal habits as riding time progresses. The system no longer passively executes preset commands but gradually converges to an optimal parameter combination that provides the most comfort for the rider through continuous trial and error and correction, thereby greatly improving the human-machine collaboration and intelligence level of the electric-assist bicycle.

[0104] Regarding the implementation details of determining the motor assist torque command in step S150, in some examples of embodiments of this application, the real-time pedal torque is... Substitute the dynamically adjusted adaptive assist curve and use the motor assist coefficient. Calculate the unlimited raw assist torque command : Equation (16) Here, instruction synthesis is performed first. In equation (16), the motor assist coefficient at the current moment is used. As a magnification factor, for human input Apply gain. The result is... This represents the "ideal assist value" that the system expects to output based on the rider's habits and environmental needs. However, this value has not been verified by the physical capabilities of the motor and may exceed the safety limits of the hardware. Therefore, it exists only as an intermediate variable.

[0105] Then, by monitoring the motor temperature and battery state of charge in real time, a dynamic torque saturation threshold is constructed to define the motor's output capability. The original assist torque command is clamped using a dynamic torque saturation threshold to obtain a limited assist torque command. : Equation (17) Equation (18) In the formula, This refers to the rated peak torque of the motor. This is the thermal protection attenuation coefficient. The low power attenuation coefficient is 1 when the corresponding state is in the safe range, and monotonically decreases to the preset lower limit when approaching the safe limit threshold, so as to dynamically compress the dynamic torque saturation threshold.

[0106] It should be noted that, in order to prevent the motor from overheating or the battery from over-discharging, the system has constructed a dynamically changing safety ceiling, namely the dynamic torque saturation threshold. The threshold is calculated using equation (17), where... The physical peak torque (fixed value) is determined by the motor hardware, while and These are the thermal protection attenuation coefficient and the low-charge attenuation coefficient, respectively. These two coefficients embody a product-type derating strategy, meaning that both coefficients are 1 under normal conditions, making... equal When the motor temperature approaches the thermal protection threshold or the battery charge approaches the cutoff voltage, the corresponding coefficient monotonically decreases (e.g., from 1 to 0.5 or 0). This ensures that even if any safety indicator alarms, the total output capacity remains constant. It will be compressed, thus physically limiting the system's maximum power output.

[0107] Subsequently, a limited-amplitude assist torque command is generated through the dual amplitude clamping process of equation (18). Among them, the inner layer Used to implement non-negative constraints, it can filter out negative values ​​caused by torque sensor zero-point drift or calculation errors, preventing motor reversal or resistance torque in non-feedback braking mode; the outer layer Used to enforce safety envelope constraints, ensuring that the final issued command will never exceed the dynamic torque saturation threshold allowed at the current moment.

[0108] Furthermore, the limited assist torque command A torque change rate limit is applied to ensure that the torque increment between adjacent control cycles does not exceed a preset mechanical flexibility threshold, thereby eliminating the mechanical shock caused by torque step jumps, and the assist torque command after the change rate limit is determined as the motor assist torque command.

[0109] It should be noted that, although While the amplitude is safe, drastic jumps in the time domain could still damage the mechanical structure. Therefore, the system ultimately... A torque change rate limit is applied. In some implementations, the system calculates the difference between the current cycle command and the previous cycle command. This limits the torque command to a preset mechanical flexibility threshold range. By "beveling" the rectangular or step-wave torque command, the gear impact noise and mechanical shock caused by sudden torque changes are eliminated, ensuring a smooth and linear power output to the wheel end, thereby improving the lifespan of transmission components and the overall riding comfort.

[0110] Figure 2The diagram illustrates the operational mechanism of the adaptive generation method for assist curve based on cycling habit learning according to an embodiment of this application, which demonstrates the complete closed-loop control process from multi-source data input to assist torque command output.

[0111] like Figure 2 As shown, the system first collects real-time data from multiple sources, such as pedal torque, cadence, vehicle driving status (speed / gradient), and battery status, through the sensor array on the left, and transmits these data in parallel to the processing link. The habit learning module builds a habit profile (i.e., cycling habit library) containing long-term pedaling characteristics based on historical data. Then, the comfort range estimation module combines real-time data to build a comfort envelope model and calculate the comfort coefficient.

[0112] Based on this, the adaptive assist curve generator, as the core decision-making hub, integrates the comfort assessment results with the constraint strategies from the top-level energy management and safety control module (such as weight adjustment and safety limit under low battery conditions) to dynamically generate an assist curve that is highly matched with the current riding scenario. Finally, it calculates the precise assist torque command to drive the motor, and realizes closed-loop iteration and update of parameters through the data logger.

[0113] To verify the effectiveness of the method proposed in this application, simulation experiments were conducted for comparative analysis. Three sets of experiments were designed: comparison of assist curves, heatmap of assist difference, and simulated riding process, to evaluate the shape of the adaptive assist curve, its adaptability to different road conditions, and its combined impact on riding comfort and battery consumption, respectively. Since actual hardware verification requires a significant amount of time, an electric-assist bicycle model was built in a simulation environment, simulating factors such as sensor noise, road gradient, and battery degradation.

[0114] Figure 3 The diagram shows a comparison of the output characteristics of the adaptive assist curve provided by the embodiment of the present invention and the traditional fixed-rate assist mode. The horizontal axis in the diagram represents the rider's input pedal torque (N·m), and the vertical axis represents the auxiliary torque output by the motor (N·m).

[0115] like Figure 3 As shown, the dashed line group represents the fixed assist modes commonly found in current related technologies (including 0.5x damping / economy mode, 1.0x balance mode, and 1.5x reinforcement mode). Their assist output and pedal force have a strictly fixed linear ratio, and cannot be flexibly adjusted according to the user's condition. In contrast, the assist curve (colored solid line group) generated by the method in this application's embodiment exhibits a significant Logistic S-shaped nonlinear characteristic: it maintains a smooth output in the low pedal force range to prevent initial surge and save energy; in the comfort range close to the center of habit, it provides strong assist through a high-slope, rapid response; and in the high pedal force range, it is limited by the dynamic torque saturation threshold (…). It tends to become smooth and saturated.

[0116] Furthermore, comparing "Phase 1 (Initial Habit, solid blue line)" and "Phase 2 (Adapted Habit, solid green line)" reveals that, through the online self-learning update step, the morphological parameters of the basic assist function have undergone adaptive evolution (e.g., the slope gain parameter increased from 0.08 to 0.12, and the inflection point offset parameter decreased from 50 N·m to 40 N·m). This results in the evolved green curve shifting to the left and becoming steeper compared to the initial blue curve, indicating that the system has successfully learned that the cyclist needs earlier assistance intervention and a more sensitive response characteristic in the current physical condition, thus achieving a dynamic leap from a general template to a personalized strategy.

[0117] Figure 4 The diagram shows a heatmap illustrating the difference in assist output between the adaptive assist strategy and the traditional balance mode under different pedal force and slope combinations provided in this application embodiment. The horizontal axis represents pedal torque (N·m), and the vertical axis represents road slope (0-0.20, i.e. 0%-20%). The color intensity of the heatmap represents the difference between the assist torque calculated by the algorithm of this invention and the output torque of the traditional fixed ratio (1 times pedal force) mode (warm red indicates higher assist in this solution, and cool blue indicates lower assist in this solution).

[0118] like Figure 4 As shown, in the upper left region with low pedal force and low slope (blue cool zone), the assist output of this invention is slightly lower than that of the traditional linear mode. This is because the Logistic basic assist function has a smooth characteristic in the initial segment, designed to prevent start-up lurch and optimize energy efficiency under low load. However, in the middle and lower regions with medium to high pedal force and high slope (red warm zone), the assist output of this invention is significantly higher than that of the traditional mode. This is because the comprehensive control index... A positive slope compensation term was introduced. This allows the system to keenly sense uphill resistance and automatically increase motor output, thereby effectively reducing the rider's physical load when climbing hills. The result directly verifies that the method of the embodiments of this application can dynamically generate an adaptive assist curve based on the multidimensional coupling relationship between pedaling force and slope.

[0119] Figure 5This diagram illustrates the time-domain response comparison between adaptive assist torque and cumulative battery energy consumption during a simulated complex road condition riding experience, as provided in an embodiment of this application. The simulation experiment depicted a 1000s riding process involving mixed road conditions including flat roads, inclines, and downhill sections. In the diagram, the horizontal axis represents time (s), the left vertical axis represents torque (N·m), and the right vertical axis represents cumulative battery consumption (unit energy). The red solid line represents the motor assist output using the baseline method (i.e., the traditional fixed-rate balancing mode); the blue solid line represents the adaptive assist torque command generated using the method described in this paper (i.e., the method of this invention); and the orange and green solid lines represent the cumulative battery consumption curves of the baseline method and the method described in this paper, respectively.

[0120] By comparing the curve shapes, the significant dynamic adjustment advantages of the method in the embodiments of this application can be observed: First, regarding the dynamics of the output, the blue curve generated by the method in this application's embodiments exhibits significant adaptability to different operating conditions. During periods of higher pedaling force or uphill driving (e.g., the t=400s-500s interval), the slope compensation term contributes to the improved performance. With the high sensitivity response of the Logistic curve, the blue curve rises rapidly and is significantly higher than the red baseline method auxiliary curve, indicating that the system can actively enhance power supply to reduce the rider's climbing burden; while in the stage of reduced pedaling force or downhill (such as the t=600s-700s interval), the blue curve falls rapidly and is lower than the baseline method, reflecting the system's energy-saving strategy of avoiding ineffective assistance.

[0121] Secondly, regarding the balance between energy and comfort, although in the high-load range, in order to ensure riding comfort (to increase the comfort coefficient)... The increased instantaneous power output (maintaining a higher level) leads to a slightly faster energy consumption increase in the climbing section of the proposed method (green energy consumption curve). However, due to the effective energy suppression implemented by the system in the low-load range, the overall energy utilization efficiency is more in line with the principle of on-demand allocation. This result intuitively demonstrates that the HAACG (Habitual Adaptive Assist Curve Generation) method proposed in this application can break through the limitations of traditional fixed templates and achieve the intelligent control objective of "strong assistance under heavy load and weak assistance under light load" in complex dynamic road conditions.

[0122] Overall, the experimental results strongly validate the significant advantages of the proposed HAACG method in multidimensional performance. By statistically learning the probability distribution characteristics of rider pedaling force and cadence, it achieves adaptive evolution of the assist curve morphology parameters (especially inflection points and slopes) according to user habits, precisely meeting the personalized comfort needs of different users. Simultaneously, the system deeply integrates multi-source sensor information such as pedaling force, cadence, road gradient, and battery state of charge. While ensuring riding smoothness and responsiveness, it utilizes a dynamic weight scheduling strategy to achieve intelligent optimization of energy consumption. Combined with a strict torque saturation threshold and rate of change limitation mechanism, it effectively improves the safety and handling stability of the entire vehicle under complex road conditions.

[0123] This study systematically addresses the problems of inconsistency and low utilization of multi-source information in existing electric power assist control strategies. Specifically, by constructing a Gaussian comfort envelope model and an adaptive Logistic mapping function, this method successfully transforms discrete, nonlinear riding habits into continuously adjustable control variables. Furthermore, it utilizes comfort error to drive online closed-loop parameter updates, fundamentally overcoming the limitations of traditional fixed-template control and achieving deep collaboration within the "human-vehicle-road" system.

[0124] In the future, intelligent upgrades can be achieved by combining deep learning models, such as exploring the use of deep neural networks (DNN) or recurrent neural networks (RNN) to uncover deeper temporal cycling patterns. In addition, swarm intelligence and cloud collaboration can be carried out, combining IoT technology to build a cloud habit database, perform cluster analysis on user groups, and achieve rapid cold start matching of parameters for new users.

[0125] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0126] Figure 6 A structural block diagram of an example of an adaptive generation system for assist curves based on cycling habit learning, according to an embodiment of this application, is shown.

[0127] like Figure 6As shown, the assist curve adaptive generation system 600 based on cycling habit learning includes a habit library construction unit 610, a multi-source data acquisition unit 620, a comfort calculation unit 630, an assist curve dynamic modulation unit 640, and an assist torque command output unit 650.

[0128] The habit database construction unit 610 is used to build a cycling habit database based on the target cyclist's historical cycling data; the cycling habit database stores pedal force habit centers and cadence habit centers that characterize the cyclist's long-term pedaling behavior.

[0129] The multi-source data acquisition unit 620 is used to acquire multi-source real-time sensor data of the electric-assisted bicycle during the real-time control cycle; the multi-source real-time sensor data includes pedal torque, cadence, vehicle driving status data and battery status data.

[0130] The comfort calculation unit 630 is used to construct a comfort envelope model based on the pedal force habit center and the pedal frequency habit center, and to calculate the comfort coefficient according to the degree of deviation of the pedal torque and the pedal frequency in the comfort envelope model.

[0131] The assist curve dynamic modulation unit 640 is used to combine the comfort coefficient, the vehicle driving status data and the battery status data to generate a comprehensive control index, and to use the comprehensive control index to dynamically adjust the shape parameters of the preset basic assist function in order to generate an adaptive assist curve that matches the current riding scenario.

[0132] The assist torque command output unit 650 is used to substitute the pedal torque into the dynamically adjusted adaptive assist curve to determine the motor assist torque command, and output it to the motor controller to drive the motor to output assist.

[0133] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. The execution instructions can be read and executed by an electronic device (including but not limited to a computer, server, or network device) to perform the steps of any of the above-described methods for adaptive generation of assist curves based on cycling habit learning.

[0134] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described methods for adaptive generation of assist curves based on cycling habit learning.

[0135] In some embodiments, this application also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of an adaptive generation method for assist curves based on cycling habit learning.

[0136] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0137] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0138] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. 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 embodiment according to actual needs.

[0139] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for adaptive generation of assist curves based on cycling habit learning, characterized in that, The method includes: A cycling habit database is established based on the historical cycling data of the target cyclist; the cycling habit database stores pedal force habit centers and cadence habit centers that characterize the long-term pedaling behavior of the cyclist. During the real-time control cycle, multi-source real-time sensor data of the electric-assist bicycle is collected; the multi-source real-time sensor data includes pedal torque, cadence, vehicle driving status data and battery status data; A comfort envelope model is constructed based on the pedal force habit center and the pedal frequency habit center, and a comfort coefficient is calculated based on the deviation of the pedal torque and the pedal frequency from the comfort envelope model. The comfort coefficient, vehicle driving status data, and battery status data are combined to generate a comprehensive control index, and the shape parameters of the preset basic assist function are dynamically adjusted using the comprehensive control index to generate an adaptive assist curve that matches the current riding scenario. The pedal torque is substituted into the dynamically adjusted adaptive assist curve to determine the motor assist torque command, which is then output to the motor controller to drive the motor to output assist.

2. The method according to claim 1, characterized in that, The process of collecting multi-source real-time sensor data from the electric-assisted bicycle during the real-time control cycle includes: The pedal torque and pedal frequency sequences are collected at a preset sampling frequency by deploying torque sensors and pedal frequency sensors, and the three-axis acceleration and three-axis angular velocity signals of the vehicle body are collected by an inertial measurement unit. The attitude calculation algorithm is used to perform multi-sensor fusion processing on the triaxial acceleration signal and the triaxial angular velocity signal, and zero-point drift compensation is performed to calculate the vehicle pitch angle, and the vehicle pitch angle is converted into the road slope in the vehicle driving state data. Based on the timestamp of the pedal torque sequence, the pedal frequency sequence and the road slope are time-aligned, and matched with the battery state of charge and motor temperature uploaded by the battery management system to construct multi-source real-time sensor data with a unified time base. Anomaly detection based on physical thresholds is performed on the multi-source real-time sensor data. If at least one sensor data is missing or out of bounds, a degradation control strategy is triggered, and the effective value of the previous control cycle or a preset safety value is used to replace the current abnormal data to ensure the continuity of the assist torque command.

3. The method according to claim 1, characterized in that, The establishment of a cycling habit database based on the target cyclist's historical cycling data includes: The historical cycling data is parsed into a continuous time-series data stream, and the time-series data stream is divided into multiple independent cycling sample segments based on a preset time window or a preset cycling event. Pedal torque sample sequence and cadence sample sequence are extracted from each cycling sample segment. Statistical distribution analysis was performed on the pedal torque sample sequence and the pedal frequency sample sequence respectively. The expected value representing the central tendency of the data was extracted as the pedal force habit center and the pedal frequency habit center, and the standard deviation or variance representing the data fluctuation range was extracted as the pedal force dispersion parameter and the pedal frequency dispersion parameter. A parameter update model based on the time decay forgetting factor is constructed. The statistical features of the cycling sample segments at different time points are weighted and fused to update the pedal force habit center, the cadence habit center, the pedal force dispersion parameter, and the cadence dispersion parameter. The cycling sample segments closer to the current time have a higher weight contribution in the update, so as to achieve adaptive tracking of the cyclist's recent habit changes. The updated pedal force habit center, the cadence habit center, the pedal force dispersion parameter, and the cadence dispersion parameter are synchronously stored in the cycling habit database.

4. The method according to claim 3, characterized in that, The statistical distribution analysis performed on the pedal torque sample sequence and the pedal frequency sample sequence respectively includes: Using the Gaussian kernel function as a smoothing operator, the pedal torque sample sequence is processed. With cadence sample sequence Construct continuous probability density functions respectively: ， In the formula, This represents the total number of samples in the sample sequence. and Let represent the independent variables of the probability density functions for pedal torque and cadence, respectively. and They represent the first in the sequence. The first pedal torque sample value and the first One cadence sample value and These are the bandwidth parameters for pedal torque and cadence, respectively. For Gaussian kernel function, and Let these represent the probability density functions of pedal torque and pedal frequency, respectively. Perform integration on the constructed probability density function to solve for the first-order raw moments and the second-order central moments: ， ， In the formula, and They represent the calculated mathematical expectations, and They represent the calculated standard deviations; The calculated results and The centers of pedal force habituation and pedal frequency habituation were determined respectively, and the calculated values ​​were... and These are respectively defined as pedal force dispersion parameter and pedal frequency dispersion parameter.

5. The method according to claim 4, characterized in that, The process of constructing a comfort envelope model based on the pedal force habit center and the pedal frequency habit center, and calculating the comfort coefficient according to the deviation of the pedal torque and the pedal frequency from the comfort envelope model, includes: Call the pedaling habit center cadence habit center , pedal force dispersion parameter and cadence dispersion parameters , construct With geometric center and and A two-dimensional Gaussian membership function model with distribution scale parameters is used, and the two-dimensional Gaussian membership function model is determined as the comfort envelope model; Get the real-time pedal torque for the current control cycle With real-time cadence The real-time pedal torque With the real-time cadence Substitute the two-dimensional Gaussian membership function model to solve for the comfort coefficient. : ， In the formula, It is an exponential function; the comfort coefficient It is used to quantify the degree to which the current pedaling state belongs to the cyclist's long-term comfort zone, and its value range is (0, 1]. The closer the value is to 1, the more the current state is in line with the cyclist's habits.

6. The method according to claim 5, characterized in that, The process of generating a comprehensive control index by combining the comfort coefficient, the vehicle driving status data, and the battery status data includes: Based on the aforementioned comfort envelope model, standardized pedal force deviation and standardized pedal frequency deviation, which form the basis for calculating the comfort coefficient, are extracted. Road slope and battery state of charge are then introduced to construct a multidimensional linear weighted model that includes habit deviation, slope compensation, and battery constraint terms to calculate the comprehensive control index. : ， In the formula, For road slope, The battery is in its state of charge. Indicates the real-time slope weight. Indicates real-time battery power weight; and These represent the pedal torque weight and cadence weight, respectively. They are determined based on the relative magnitudes of the pedal torque stability index and cadence stability index in the target cyclist's historical riding data. The stability index is used to characterize the steady-state degree of the corresponding pedal characteristic dimension. A dynamic weighted scheduling strategy based on energy state is executed during the calculation process: a low energy threshold is set. Real-time monitoring ,when At that time, calculate the energy regulation gain. And perform nonlinear correction on the weighting coefficients: ， ， In the formula, It is an exponential function. The energy sensitivity coefficient, and These are the preset base weights for electricity consumption and slope, respectively; When detected season and keep and This ensures the continuity of the weight scheduling strategy at the threshold. Among them, the standardized pedal force deviation Deviation from standardized cadence The positive and negative signs are used to characterize the direction of deviation of the real-time pedaling state from the habit center, so that the comprehensive control index It carries both deviation magnitude information and deviation direction information.

7. The method according to claim 6, characterized in that, The step of dynamically adjusting the morphological parameters of the preset basic assist function using the comprehensive control index to generate an adaptive assist curve that matches the current riding scenario includes: A parameterized Logistic function is constructed as the mathematical model of the basic assist function, and the comprehensive control index is used. Mapped to motor assist coefficient : ， In the formula, and These are the real-time minimum assist coefficient and the real-time maximum assist coefficient, respectively, and satisfy the following conditions: ; Let be the slope gain parameter, and satisfy . ; This refers to the inflection point offset parameter; Execute a multi-dimensional parameter modulation strategy, and based on the comfort coefficient With the battery state of charge Dynamic adjustments are made to the morphological parameters: On the one hand, when the amount of comfort deviation is detected When the slope gain parameter is increased, the slope gain parameter is increased. and / or reduce the inflection point offset parameter This is to improve the sensitivity of the adaptive assist curve to changes in the comprehensive control index; wherein, ; On the other hand, when the battery state of charge is detected The power level drops below a preset threshold and / or a power weight is detected. When the value increases, the real-time maximum assist coefficient is dynamically adjusted downwards. This helps to compress the upper limit of the power output in the high-power range; The adaptive boost curve is determined based on the dynamically adjusted morphological parameters, and a continuity constraint is applied to the morphological parameters to ensure a smooth transition of the generated boost coefficient in the time domain; wherein, the continuity constraint is used to independently limit the rate of change of the morphological parameters in adjacent control cycles: ， In the formula, To control the cycle length, For the first The values ​​of each morphological parameter during the real-time control cycle. This is the preset maximum change step size threshold for the corresponding parameter; where, It remains unchanged during the real-time control cycle.

8. The method according to claim 7, characterized in that, The step of substituting the pedal torque into the dynamically adjusted adaptive assist curve to determine the motor assist torque command includes: Real-time pedal torque Substitute the dynamically adjusted adaptive assist curve, and use the motor assist coefficient. Calculate the unlimited raw assist torque command : ， Real-time monitoring of motor temperature and battery state of charge is used to construct a dynamic torque saturation threshold for defining the motor's output capability. The original assist torque command is then subjected to amplitude clamping processing using the dynamic torque saturation threshold to obtain a limited assist torque command. : ， ， In the formula, This refers to the rated peak torque of the motor. This is the thermal protection attenuation coefficient. The attenuation coefficient is a low power decay coefficient; the attenuation coefficient is 1 when the corresponding state is in the safe range, and monotonically decreases to a preset lower limit when approaching the safe limit threshold, so as to dynamically compress the dynamic torque saturation threshold. The limiting assist torque command A torque change rate limit is applied to ensure that the torque increment between adjacent control cycles does not exceed a preset mechanical flexibility threshold, thereby eliminating the mechanical shock caused by torque step jumps, and the assist torque command after the change rate limit is determined as the motor assist torque command.

9. The method according to claim 7, characterized in that, The method also includes an online self-learning update step for parameters, used to continuously optimize the assist response characteristics based on the cyclist's real-time pedaling behavior feedback, specifically including: Set a target comfort level that represents the ideal load state of the cyclist. And calculate the comfort coefficient in each control cycle. Comfort error between the target comfort level and the target comfort level : ， Based on the comfort error For the slope gain parameter With the inflection point offset parameter Perform error-driven online iterative updates to gradually bring the parameter combination closer to the target response characteristics that match the cyclist's current physical condition: ， ， In the formula, This represents the slope gain parameter updated in the next control cycle. This indicates the inflection point offset parameter after the next control cycle update. and These are the learning rates for the slope gain parameter and the inflection point offset parameter, respectively.

10. A system for adaptive generation of assist curves based on cycling habit learning, characterized in that, The system includes: The habit database construction unit is used to build a cycling habit database based on the target cyclist's historical cycling data; the cycling habit database stores pedal force habit centers and cadence habit centers that characterize the cyclist's long-term pedaling behavior. The multi-source data acquisition unit is used to acquire multi-source real-time sensor data of the electric-assist bicycle during the real-time control cycle; the multi-source real-time sensor data includes pedal torque, cadence, vehicle driving status data and battery status data; The comfort calculation unit is used to construct a comfort envelope model based on the pedal force habit center and the pedal frequency habit center, and to calculate the comfort coefficient according to the degree of deviation of the pedal torque and the pedal frequency in the comfort envelope model; The assist curve dynamic modulation unit is used to combine the comfort coefficient, the vehicle driving status data and the battery status data to generate a comprehensive control index, and use the comprehensive control index to dynamically adjust the shape parameters of the preset basic assist function to generate an adaptive assist curve that matches the current riding scenario. The assist torque command output unit is used to substitute the pedal torque into the dynamically adjusted adaptive assist curve to determine the motor assist torque command, and output it to the motor controller to drive the motor to output assist.

Images