Electric power-assisted bicycle controller, intelligent control method and electric power-assisted bicycle

Abstract

The invention provides an electric power-assisted bicycle controller, an intelligent control method and an electric power-assisted bicycle. The electric power-assisted bicycle controller comprises a multi-mode sensing layer, a control layer and a control layer, wherein a bicycle body dynamic sensor and a rider state sensor are integrated in the multi-mode sensing layer; the data processing layer is used for primarily processing the first data and the second data; the decision control layer is used for carrying out correlation analysis and identifying a shaking source as a first type of shaking caused by an uneven road surface or a second type of shaking caused by a state change of a rider; and the execution layer is used for executing a first preset control strategy corresponding to the first type of jitter or executing a second preset control strategy corresponding to the second type of jitter. By accurately identifying and differentially processing jitters from different sources, signal noise can be effectively filtered and smooth assistance output can be provided aiming at road bumping, and instability of a vehicle body due to sudden change of power is avoided; stable assistance can be actively provided for self-state changes such as fatigue of a rider, and the riding risk caused by poor physical strength is prevented.

Description

Technical Field

[0001] This invention belongs to the field of controller technology, specifically relating to an electric-assist bicycle controller, an intelligent control method, and an electric-assist bicycle. Background Technology

[0002] Existing electric-assist bicycles typically use torque and speed sensors to sense the rider's pedaling intentions and provide auxiliary power accordingly. This control method is essentially a passive response mechanism. The controller provides feedback to the rider's actions based on a preset assistance mode, and its power adjustment often lags behind the rider's actual needs and real-time changes in road conditions, resulting in a poor riding experience and low energy utilization efficiency.

[0003] During riding, torque sensors receive various vibration signals. These signals originate from complex sources, ranging from uneven road surfaces (such as gravel roads or potholes) causing bike jolting to changes in the rider's own condition (such as fatigue causing body swaying or unstable pedaling rhythm). To improve the riding experience, some existing technologies attempt to adjust control strategies by sensing external parameters such as road surface roughness, aiming for more stable vehicle control. However, these technologies have a fundamental flaw in processing vibration signals: they cannot effectively distinguish vibrations from different sources. They typically treat all vibrations as changes in the rider's intention or uniform road condition feedback, thus triggering unreasonable and unintended assist outputs. For example, on bumpy roads, the controller may incorrectly provide intermittent "bouncing" assist, exacerbating bike instability; when the rider sways due to fatigue, the controller fails to recognize their true state and provide effective stabilization assistance. This confusion and misinterpretation of vibration signal sources not only severely impacts riding comfort but also poses potential safety hazards. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide an electric-assist bicycle controller, an intelligent control method, and an electric-assist bicycle.

[0005] An electric-assist bicycle controller according to the present invention includes: Multimodal perception layer: integrates a vehicle dynamic sensor that can collect first data characterizing the vibration state of the electric-assist bicycle body, and a rider state sensor that can collect second data characterizing the rider's body posture and / or pedaling behavior. Data processing layer: performs preliminary processing on the first and second data; Decision control layer: Performs correlation analysis on the first and second data processed by the data processing layer to identify the source of vibration as either the first type of vibration caused by uneven road surface or the second type of vibration caused by changes in the rider's own state; Execution layer: Executes the first preset control strategy corresponding to the first type of jitter, or executes the second preset control strategy corresponding to the second type of jitter.

[0006] In a preferred embodiment, the vehicle dynamics sensor measures three linear accelerations and three angular velocities of the electric-assisted bicycle in real time; Among them, the vertical acceleration data constitutes the first data.

[0007] In a preferred embodiment, the cyclist state sensor includes a cycling posture sensor and a multi-dimensional torque sensor; The riding posture sensor monitors the frequency and amplitude of the rider's body swaying. The multidimensional torque sensor measures the magnitude of the cyclist's pedaling torque; The second set of data consists of data collected by cycling posture sensors and multi-dimensional torque sensors that characterize the cyclist's body posture or pedaling behavior.

[0008] In a preferred embodiment, when performing correlation analysis, the decision control layer compares the time domain and frequency domain characteristics of the first data and the second data to determine the correlation between the first data and the second data. If the correlation between the first and second data is weak, the current vibration is determined to be a type I vibration caused by uneven road surface. If the first and second data are strongly correlated, the current vibration is determined to be a second type of vibration caused by changes in the rider's own state.

[0009] In a preferred embodiment, the first preset control strategy includes: performing adaptive low-pass filtering on the first data and / or the second data; And / or, when an impact is anticipated based on the analysis of the first data, the motor output torque may be locked in advance or released smoothly.

[0010] In a preferred embodiment, the second preset control strategy includes: Automatically switch between assist modes; And / or, provide adaptive assist RPM based on the rider's historical smooth cadence.

[0011] In a preferred embodiment, a heart rate sensor is also integrated into the multimodal sensing layer to help determine the cyclist's physiological fatigue state.

[0012] According to the present invention, an intelligent control method for an electric-assisted bicycle is provided, the control method comprising: Acquire first data characterizing the vibration state of the electric-assist bicycle body, collected by the vehicle body dynamic sensor; Acquire second data, collected by cyclist status sensors, that characterizes the cyclist's body posture or pedaling behavior; Based on the correlation analysis of the first and second data, the source of vibration is identified as either the first type of vibration caused by uneven road surface or the second type of vibration caused by changes in the rider's own state. When the first type of jitter is identified, a first preset control strategy aimed at smoothing the motor-assisted output is executed; When the vibration is identified as the second type, a second preset control strategy designed to assist the rider in restoring stable riding is executed.

[0013] In a preferred embodiment, the multimodal perception layer also integrates a global positioning system module for environmental perception, a forward-looking camera, and a millimeter-wave radar, and high-precision map data is pre-loaded into the controller's memory. Real-time perception of environmental and / or geographical information; based on the perceived environmental and / or geographical information, prediction of future riding conditions through a preset path gradient prediction model; and advance adjustment of assist output and / or energy recovery strategies.

[0014] According to the present invention, an electric-assisted bicycle also includes a frame and a motor.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By accurately identifying and differentiating vibrations from different sources, this invention fundamentally solves the problem of "false assistance" caused by misreading vibration signals in the prior art; for road bumps, it can effectively filter signal noise and provide smooth assistance output, avoiding instability of the vehicle due to sudden changes in power; for changes in the rider's own state such as fatigue, it can actively provide stability assistance to prevent riding risks caused by physical exhaustion.

[0016] 2. This invention combines predictive control, which not only handles current vibrations but also predicts road conditions ahead and plans assist curves and energy strategies in advance, eliminating the response lag and jerking of traditional controllers and bringing a smarter riding experience.

[0017] 3. The control strategy of this invention can adaptively cope with various cycling scenarios such as flat urban roads, mountain slopes, and gravel and bumpy roads, meeting the needs of various users from daily commuting to off-road sports, and is applicable to a wide range of scenarios. Attached Figure Description

[0018] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating the first type of jitter in this invention; Figure 2 The flowchart illustrates the second type of jitter in this invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0020] Example 1 An electric-assist bicycle controller according to the present invention includes: a multimodal sensing layer: integrating a vehicle dynamic sensor capable of collecting first data characterizing the vibration state of the electric-assist bicycle body, and a rider state sensor capable of collecting second data characterizing the rider's body posture and / or pedaling behavior; a data processing layer: performing preliminary processing on the first and second data; a decision control layer: performing correlation analysis on the first and second data processed by the data processing layer, identifying the source of vibration as either a first type of vibration caused by uneven road surface, or a second type of vibration caused by changes in the rider's own state; and an execution layer: executing a first preset control strategy corresponding to the first type of vibration, or executing a second preset control strategy corresponding to the second type of vibration.

[0021] As the core of the intelligent control system, this controller can be implemented as an integrated microcontroller or a system composed of multiple cooperating processors. Logically, the controller can be divided into a multimodal perception layer, a data processing layer, a decision control layer, and an execution layer. From a hardware implementation perspective, the controller includes at least one processor and a memory connected to the processor. The memory stores computer program instructions, which, when executed by the processor, are used to implement the control methods described below.

[0022] The electric-assisted bicycle described in this embodiment, in addition to the conventional frame, wheels, transmission system and battery, is also equipped with the controller and a series of sensors and actuators connected to it.

[0023] The vehicle dynamics sensor measures three linear accelerations and three angular velocities of the e-bike in real time. The vertical acceleration data constitutes the first set of data. Specifically, the controller's multimodal perception layer integrates multiple sensors to achieve comprehensive perception of the vehicle, rider, and environment. In one feasible embodiment of this application, key sensors for accurate identification of vibration sources include a vehicle dynamics sensor and a rider state sensor.

[0024] In one feasible implementation, the vehicle dynamics sensor is a six-axis inertial measurement unit (IMU) mounted at a critical location on the frame (e.g., near the vehicle's center of gravity or seat tube). This six-axis IMU is capable of measuring the three linear accelerations (ax, ay, az) and three angular velocities (gx, gy, gz) of the e-bike in real time at a high sampling frequency of at least 100 Hz. It should be noted that the vertical acceleration data measured by the six-axis IMU (e.g., z-axis acceleration) is crucial information characterizing the vibration state of the vehicle body due to uneven road surfaces, and constitutes what is referred to as "first data" in this application.

[0025] In one feasible implementation, the cyclist state sensor includes a cycling posture sensor and a multidimensional torque sensor. The cycling posture sensor monitors the frequency and amplitude of the cyclist's body sway. The multidimensional torque sensor measures the magnitude of the cyclist's pedaling torque. The data collected by the cycling posture sensor and the multidimensional torque sensor, which characterize the cyclist's body posture or pedaling behavior, together constitute the second data. Specifically, the cyclist state sensor includes a cycling posture sensor mounted under the seatpost or saddle, and a multidimensional torque sensor integrated into the bottom bracket or crank. The cycling posture sensor monitors changes in the cyclist's upper body posture, particularly the frequency and amplitude of body sway. The multidimensional torque sensor not only measures the magnitude of the cyclist's pedaling torque but also analyzes the smoothness, consistency, and torque consistency of the left and right legs. The data collected by these two types of sensors, which characterize the cyclist's body posture or pedaling behavior, together constitute the second data referred to in this application.

[0026] The following describes the specific working process of this embodiment when dealing with bumpy roads. Assume the cyclist is riding on an unpaved road covered with gravel.

[0027] like Figure 1 As shown, the controller continuously acquires sensing data from the multimodal sensing layer through its input interface. The six-axis inertial measurement unit sends the acquired high-frequency acceleration data stream to the controller. The signal processing module in the controller's data processing layer performs preliminary processing on this raw data, such as filtering out some high-frequency electrical noise through digital filters and focusing on the vertical acceleration component. At the same time, the riding posture sensor and the multi-dimensional torque sensor also synchronously acquire data and send the data stream characterizing the cyclist's upper body posture and pedaling behavior to the controller.

[0028] Subsequently, during the correlation analysis, the decision control layer compares the time-domain and frequency-domain characteristics of the first and second data to determine the correlation between them. If the correlation between the first and second data is weak, the current vibration is determined to be a type I vibration caused by uneven road surface. If the correlation between the first and second data is strong, the current vibration is determined to be a type II vibration caused by changes in the cyclist's own state.

[0029] The source of vibration is identified. The data processing layer transmits the processed first data (vertical vibration data of the vehicle body) and second data (cyclist posture and pedaling data) to the vibration identification unit in the decision control layer. The core task of the vibration identification unit is to perform correlation analysis. Understandably, in this scenario, due to the road surface being composed of a large amount of gravel, the wheels will produce high-frequency, low-amplitude bounces, which will be directly reflected in the first data collected by the six-axis inertial measurement unit, manifesting as dense, sharp peak fluctuations in the vertical acceleration signal. Correspondingly, because the cyclist consciously maintains body stability to cope with the bumps, their upper body does not exhibit large-amplitude swaying synchronized with the vehicle body vibration, so the second data collected by the riding posture sensor appears relatively stable. In addition, to maintain forward momentum, the cyclist's pedaling rhythm and force are also kept within a relatively stable range, so the second data collected by the multi-dimensional torque sensor does not show significant anomalies.

[0030] The vibration recognition unit, by comparing the time and frequency domain characteristics of the first and second data, found that the first data exhibited significant high-frequency vibration characteristics, while the second data did not show a strong correlation within the same time window. Specifically, the power spectral density of the vehicle's vertical vibration had a significant peak in a certain high-frequency range (e.g., 10-30 Hz), while the dominant frequency of the rider's body swaying was much lower than this, and the variance of the pedaling torque was also within the normal range. Based on this, the judgment condition of "the first data fluctuating significantly while the second data is relatively stable" was met, and the vibration recognition unit determined that the current vibration source was the first type of vibration caused by uneven road surface.

[0031] Once the jitter is identified as Type I jitter, the control strategy unit in the decision control layer immediately invokes and executes a first preset control strategy. The first preset control strategy includes: performing adaptive low-pass filtering on the first data and / or the second data; and / or, locking or gently releasing the motor output torque in advance when an impact is predicted based on the analysis of the first data.

[0032] In this embodiment, the first preset control strategy aims to smooth the motor assist output to avoid unnecessary power fluctuations caused by torque sensor signal contamination due to road bumps. As a preferred implementation, this strategy can specifically involve activating a bump filter. That is, the control strategy unit instructs the signal processing module to perform an adaptive low-pass filter on the raw torque signal output from the multi-dimensional torque sensor. The cutoff frequency of this filter is adaptively adjusted, for example, dynamically set based on the vibration intensity (such as vibration energy or root mean square value) detected by the six-axis inertial measurement unit. When the vibration is severe, the cutoff frequency can be appropriately lowered to more effectively attenuate high-frequency noise; when the vibration weakens, the cutoff frequency can be appropriately raised to ensure a rapid response to the rider's true pedaling intention. After filtering, the high-frequency jitter components of the torque signal are effectively removed, making it smoother and thus better reflecting the rider's continuous pedaling intention.

[0033] Finally, at the execution layer, the smoothed torque signal is sent to the motor controller. Based on this smoothed signal and the current assist mode, the motor controller calculates the target assist torque and controls the motor to output this torque stably and continuously.

[0034] As can be seen, throughout the entire control process, although the input inertial measurement unit signal exhibits high-frequency vibration, the final output motor torque curve remains smooth and continuous because the system correctly identifies the source of the vibration and executes the first preset control strategy. The resulting benefit is that riders will not experience the jerky or bouncing sensations caused by wheel movement on bumpy roads; the vehicle's tracking is improved, and riding comfort and safety are significantly enhanced.

[0035] As an optional implementation, the first preset control strategy can also include a torque locking mechanism based on impact prediction. When the six-axis inertial measurement unit detects a specific acceleration waveform that foreshadows a severe impact (such as approaching a speed bump or falling into a pothole), the vibration recognition unit can predict the impact in advance. At this time, the control strategy unit can instruct the motor controller to temporarily lock the current motor output torque for a very short time (e.g., tens of milliseconds) before the impact occurs, or release the torque at a very smooth rate. This avoids sudden changes in motor output due to drastic changes in pedaling torque at the moment the wheel is suspended in the air or collides with an obstacle, thereby preventing the vehicle from experiencing a sudden forward lurch or dragging sensation, further improving riding stability under extremely bumpy road conditions.

[0036] The present invention also provides an intelligent control method for electric-assisted bicycles, based on the above-mentioned electric-assisted bicycle controller, the control method comprising: Acquire first data characterizing the vibration state of the electric-assist bicycle body, collected by the vehicle body dynamic sensor; Acquire second data, collected by cyclist status sensors, that characterizes the cyclist's body posture or pedaling behavior; Based on the correlation analysis of the first and second data, the source of vibration is identified as either the first type of vibration caused by uneven road surface or the second type of vibration caused by changes in the rider's own state. When the first type of jitter is identified, a first preset control strategy aimed at smoothing the motor-assisted output is executed; When the vibration is identified as the second type, a second preset control strategy designed to assist the rider in restoring stable riding is executed.

[0037] The present invention also provides an electric-assisted bicycle, which further includes a frame and a motor.

[0038] Example 2 like Figure 2 As shown in Embodiment 1, this embodiment describes in detail how the system identifies and executes a corresponding second preset control strategy when the vibration originates from changes in the rider's own state (i.e., the second type of vibration) to actively assist the rider and ensure riding safety. The hardware architecture of the electric-assist bicycle and its controller used in this embodiment is basically the same as that in Embodiment 1.

[0039] Suppose a cyclist is engaged in a long-distance, high-intensity ride. After several hours, their physical strength begins to decline significantly, and they enter a state of fatigue.

[0040] Similar to the previous embodiments, the control flow begins with the acquisition of multimodal sensing data. The controller continuously acquires real-time data from the six-axis inertial measurement unit, the riding posture sensor, and the multi-dimensional torque sensor.

[0041] In a more preferred embodiment, a heart rate sensor is also integrated into the multimodal sensing layer to assist in determining the rider's physiological fatigue state. To more accurately determine the rider's fatigue state, an additional heart rate sensor is integrated into the multimodal sensing layer of this embodiment, for example, integrated into the handlebar grip or connected wirelessly to a wearable device worn by the rider.

[0042] The jitter detection unit performs in-depth correlation analysis on the collected multimodal data. In this scenario, cyclists may struggle to maintain a stable riding posture due to insufficient core muscle strength, resulting in involuntary, asynchronous swaying of the upper body that is out of sync with the pedaling rhythm. This swaying is typically low-frequency (e.g., 0.5-2 Hz) and large-amplitude. Therefore, the second data collected by the riding posture sensor clearly captures this characteristic. Simultaneously, fatigue also leads to deformed pedaling motions, uneven force application, and decreased coordination between the left and right legs. This is reflected in the second data collected by the multidimensional torque sensor, manifesting as a decrease in the regularity and consistency of the pedaling torque signal; the signal waveform is no longer a smooth sine wave but instead exhibits many spikes and irregular fluctuations. The swaying of the bicycle is also captured by the six-axis inertial measurement unit, and its first data may also show low-frequency vibrations similar to the body's swaying frequency. In addition, data from the heart rate sensor can show that the cyclist's heart rate remains at a high level (e.g., more than 85% of their maximum heart rate), and heart rate variability may decrease accordingly. These are typical indicators of physiological fatigue.

[0043] The vibration recognition unit comprehensively analyzes this information: the first data (vehicle vibration) and the rider's posture swaying in the second data show a strong correlation in both time and frequency. Simultaneously, the pedaling behavior in the second data also exhibits significant irregularity. This, combined with the physiological fatigue state corroborated by heart rate data, leads the vibration recognition unit to determine that the current vibration is not caused by the road surface, but rather by a second type of vibration caused by changes in the rider's own state (fatigue).

[0044] When the identified vibration is classified as Type II, a second preset control strategy is executed. In this case, the control strategy unit's objective is no longer to filter the signal, but to actively intervene and assist the rider in restoring stability. The system will execute the second preset control strategy. The second preset control strategy includes: automatically switching the assist mode; and / or providing an adaptive assist speed based on the rider's historical stable cadence.

[0045] As a preferred implementation, the second preset control strategy may include automatic switching of the assist mode. For example, if the system detects fatigue and shaking in the rider in "Sport Mode" (lower assist ratio, more sensitive response), the control strategy unit will immediately generate a command to prompt the rider through the human-machine interface (such as a display screen) that "fatigue has been detected, switch to Comfort Mode," and automatically switch the assist mode to a preset "Comfort ECO Mode" with a higher assist ratio and a smoother response curve. In this new mode, the motor controller will provide greater assist torque for the same pedaling input, thereby significantly reducing the rider's physical burden. This mode typically has a first preset assist ratio and a first response curve, characterized by a smoother and more linear assist output, which helps the rider relax their body and regain a stable riding posture.

[0046] As an alternative or supplementary implementation, the second preset control strategy can also provide an adaptive auxiliary speed. When the vibration recognition unit determines that the vibration is caused by the rider's disordered pedaling rhythm, the control strategy unit can refer to the rider's historical stable cadence to provide guidance. Specifically, the controller's memory records and learns the rider's cadence data under normal riding conditions (this data also comes from the second data collected by the multi-dimensional torque sensor), and through statistical analysis (such as calculating the mean or mode of historical cadence data), a "historical stable cadence," such as 85 rpm, is derived. When irregular pedaling is detected, the control strategy unit instructs the motor controller to provide a weak but stable adaptive auxiliary speed, with the target value set around 85 rpm. In other words, the motor plays a role in "guiding" the rider to pedal at a more efficient and stable rhythm. The rider will feel a gentle guiding force on the pedals, thus unconsciously adjusting their pedaling rhythm to match the motor's auxiliary rhythm, thereby helping them restore efficient and stable pedaling behavior.

[0047] By implementing the second preset control strategy described above, the rider's physical exertion is reduced, body swaying decreases, pedaling rhythm becomes more stable, and the overall stability of the vehicle is significantly enhanced. This not only improves the comfort of long-distance riding, but more importantly, it effectively avoids safety risks such as loss of control and crashes that may result from exhaustion or decreased attention, demonstrating the system's active safety assistance capabilities.

[0048] Example 3 Based on Embodiment 1 or Embodiment 2, the present invention also provides an intelligent control method for electric-assisted bicycles. The multimodal perception layer integrates a global positioning system module for environmental perception, a forward-looking camera, and a millimeter-wave radar, and high-precision map data is pre-loaded into the controller's memory. The method senses environmental and / or geographical information in real time, and based on the sensed environmental and / or geographical information, predicts future riding conditions using a preset path gradient prediction model, and adjusts the power assist output and / or energy recovery strategy in advance.

[0049] This embodiment aims to demonstrate how the jitter differentiation control of this application can be integrated with predictive control functions to cope with complex riding scenarios involving multiple elements, thereby achieving a higher level of intelligent control. In addition to the sensors described in Embodiments 1 and 2, the electric-assist bicycle controller of this embodiment also deeply integrates a global positioning system module for environmental perception, a forward-looking camera, and millimeter-wave radar in its multimodal perception layer, and high-precision map data is pre-loaded in the controller's memory.

[0050] In this embodiment, the control strategy unit in the decision control layer not only includes the logic for jitter-differentiated control but also integrates a predictive control algorithm module. Simultaneously, the energy optimization unit also participates in the decision-making process, responsible for planning the overall energy usage strategy.

[0051] To better understand the technical solution of this application, imagine a complex cycling scenario: an electric-assisted bicycle is about to pass through a series of small-radius S-shaped curves, followed by a 300-meter-long undulating road section with a surface composed of a mixture of cement blocks and soil, which includes two small uphill and downhill sections.

[0052] When the vehicle is about 50 meters from the entrance to the S-curve, the predictive control function begins to intervene.

[0053] Real-time perception and prediction: The controller obtains its precise current location through the Global Positioning System (GPS) module and matches it with high-precision map data to predict that it will enter a series of curves 50 meters ahead. Simultaneously, a forward-looking camera helps identify the curvature of the curves and road surface boundaries, while millimeter-wave radar detects whether there are other vehicles or obstacles within the curves. Based on this perceived environmental and geographical information, the decision-making and control layer predicts the curvature and slope changes of the entire future path (e.g., 500 meters ahead) using a pre-set path slope prediction model (which can be based on map elevation data or visual slope estimation), forming a "predictive spatiotemporal window."

[0054] Predictive Control Execution (Cornering Phase): Based on the prediction results, the control strategy unit determines that a cornering condition requiring high stability is about to begin. To ensure safe and smooth cornering, the system executes the cornering control strategy in advance. Specifically, before entering the corner, the motor output torque is linearly reduced. For example, the torque output coefficient of the current assist mode is smoothly reduced to 30% of the base mode. This avoids excessive power generated by the rider's unintentional pedaling in the corner, which could lead to rear wheel slippage or vehicle tilting, thereby improving cornering stability and safety.

[0055] Shaking control and predictive control work together (uneven road section after exiting a curve): When the vehicle successfully passes through the S-curve and enters the 300-meter-long uneven road section, the wheels begin to run over the uneven cement blocks and dirt road surface.

[0056] Shake Recognition and Processing: The six-axis inertial measurement unit immediately detects a severe, high-frequency vertical vibration signal (first data). Simultaneously, data from the riding posture sensor and multi-dimensional torque sensor show that the rider's posture and pedaling behavior remain stable (second data). Based on this uncorrelated characteristic, the shake recognition unit immediately classifies the shake as Type I shake caused by uneven road surface. The control strategy unit then activates the first preset control strategy, namely the "bump filter" described in Example 1, to adaptively low-pass filter the torque sensor signal, ensuring that the motor controller receives a smooth torque command, thereby outputting continuous and stable assist.

[0057] Energy Optimization and Predictive Control: Accordingly, the energy optimization unit executes an energy optimization strategy of "moderate uphill assist and active downhill recovery" based on the slope information of the undulating road ahead in the "prediction time window". Approximately 20-50 meters before entering the first small uphill section, the system gradually increases torque reserve, allowing the motor to enter its efficient operating range in advance, helping the rider ascend the hill easily and avoiding energy waste from sudden exertion at the bottom of the hill. According to experimental data, this energy optimization for long uphill sections can result in a 5-10% increase in range. When entering a downhill section, the system automatically adjusts the recovery intensity of the energy recovery module 142 according to the slope, achieving "smart recovery" and maximizing the conversion of potential energy into electrical energy stored back in the battery.

[0058] Throughout the complex riding experience, differentiated vibration control for road bumps and predictive control and energy optimization strategies based on future road conditions work in parallel and seamlessly collaborate. Vibration control ensures riding comfort and stability at the micro level, while predictive control plans the optimal power output and energy management scheme at the macro level. Ultimately, riders experience stable and controllable riding when cornering, no jerking on bumpy roads, smooth and natural power delivery when going uphill and downhill, and significantly optimized energy consumption throughout the trip. Experimental data shows that under similar urban start-stop and undulating road conditions, the comprehensive predictive energy optimization strategy can increase the range by 10% to 15%. This fully demonstrates the significant advantages of this application in achieving a balance between safety, comfort, and energy efficiency.

[0059] Example 4 Based on Embodiment 1, Embodiment 2, or Embodiment 3, this embodiment provides another optional implementation of the jitter source identification method, namely, using a pre-trained machine learning model to replace the rule-based and threshold-based judgment logic in the aforementioned embodiments. The hardware configuration of this embodiment is the same as that of Embodiment 1 and Embodiment 2. The key difference is that a machine learning classifier is embedded in the jitter identification unit in the decision control layer of the controller.

[0060] In one specific implementation of this application, the machine learning model can be a support vector machine classifier, a gradient boosting decision tree, a random forest, or a lightweight neural network model. This model needs to be trained offline. The training dataset can be constructed as follows: Test subjects of different weights and cycling styles are invited to ride on various typical road surfaces (such as smooth asphalt roads, gravel roads, and cobblestone roads), and they are guided to simulate fatigue states (such as body swaying and unstable pedaling) at certain stages. Simultaneously, the multimodal sensing system described in this application records complete sensor data, and professionals manually label each data segment with its actual source of vibration ("road surface vibration," "cyclist vibration," or "no vibration").

[0061] In the actual workflow of the controller, the jitter identification process may include the following steps: Data acquisition and preprocessing: Similar to Example 1, the controller acquires data in real time from the six-axis inertial measurement unit, riding posture sensor and multi-dimensional torque sensor.

[0062] Feature Extraction: The signal processing module or a dedicated feature extraction module in the data processing layer processes the acquired time-series data, extracting a series of key features that can effectively distinguish different jitter sources, forming a high-dimensional feature vector. These features may include, but are not limited to: Features from the first data (inertial measurement unit signal): the root mean square value, peak-to-peak value, and zero-crossing rate of the vertical acceleration signal within a specific time window; the energy proportion of the power spectral density obtained through fast Fourier transform in different frequency bands (e.g., low-frequency band 0-5Hz, high-frequency band 10-30Hz). Features from the second data (cyclist state signal): the dominant frequency, amplitude, and variance of the cyclist's upper body sway signal; the smoothness index of the pedaling torque signal (e.g., the energy of the first or second derivative of the signal), the left and right torque balance coefficient, and the stability of the pedal frequency (e.g., the standard deviation of the pedal frequency). Cross-modal features: the cross-correlation coefficient between the vertical vibration signal of the vehicle body and the body sway signal.

[0063] Model Inference and Classification: The feature vector is fed as input into a pre-trained machine learning model (e.g., a support vector machine classifier) ​​embedded within the shake detection unit. The classifier computes a classification result on the input feature vector, indicating the most likely source of the shake. For example, an output label "0" represents type I shake (road surface shake), and an output label "1" represents type II shake (cyclist shake). The model can also simultaneously output a confidence score, representing its degree of confidence in the classification result.

[0064] Differential control is implemented: Subsequent control strategy units determine whether to implement a first or second preset control strategy based on the classification results (and optional confidence levels) output by the machine learning model. For example, when the model output is "Type I jitter" and the confidence level is higher than a certain threshold (such as 0.8), the system implements a turbulence filtering strategy.

[0065] Understandably, models capable of automatically learning more complex and subtle nonlinear patterns from large amounts of data typically achieve higher recognition accuracy. Simultaneously, the models exhibit greater robustness, better adapting to cyclists of varying weights, heights, and riding styles, as well as more diverse road surface types, since these differences are already reflected in the training data. Furthermore, the controller in this application supports online learning or federated learning mechanisms, meaning the model can be continuously fine-tuned and optimized using newly collected data during user operation, achieving true personalized adaptation and resulting in increasingly accurate shake recognition.

[0066] In summary, by employing a machine learning model to identify the source of vibration, the technical solution of this application has been further improved in terms of intelligence and adaptability, providing users with a more reliable and personalized intelligent riding experience.

[0067] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0068] In the description of this application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0069] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A controller for an electric-assisted bicycle, characterized in that, include: Multimodal perception layer: integrates a vehicle dynamic sensor that can collect first data characterizing the vibration state of the electric-assist bicycle body, and a rider state sensor that can collect second data characterizing the rider's body posture and / or pedaling behavior. Data processing layer: performs preliminary processing on the first and second data; Decision control layer: Performs correlation analysis on the first and second data processed by the data processing layer to identify the source of vibration as either the first type of vibration caused by uneven road surface or the second type of vibration caused by changes in the rider's own state; Execution layer: Executes the first preset control strategy corresponding to the first type of jitter, or executes the second preset control strategy corresponding to the second type of jitter.

2. The electric-assist bicycle controller according to claim 1, characterized in that, The vehicle dynamic sensor measures the three linear accelerations and three angular velocities of the electric-assisted bicycle in real time; Among them, the vertical acceleration data constitutes the first data.

3. The electric-assist bicycle controller according to claim 1, characterized in that, The cyclist status sensor includes a cycling posture sensor and a multi-dimensional torque sensor; The riding posture sensor monitors the frequency and amplitude of the rider's body swaying. The multidimensional torque sensor measures the magnitude of the cyclist's pedaling torque; The second set of data consists of data collected by cycling posture sensors and multi-dimensional torque sensors that characterize the cyclist's body posture or pedaling behavior.

4. The electric-assist bicycle controller according to claim 1, characterized in that, When performing correlation analysis, the decision control layer compares the time domain and frequency domain characteristics of the first data and the second data to determine the correlation between the first data and the second data. If the correlation between the first and second data is weak, the current vibration is determined to be a type I vibration caused by uneven road surface. If the first and second data are strongly correlated, the current vibration is determined to be a second type of vibration caused by changes in the rider's own state.

5. The electric-assist bicycle controller according to claim 1, characterized in that, The first preset control strategy includes: performing adaptive low-pass filtering on the first data and / or the second data; And / or, when an impact is anticipated based on the analysis of the first data, the motor output torque may be locked in advance or released smoothly.

6. The electric-assist bicycle controller according to claim 1, characterized in that, The second preset control strategy includes: Automatically switch between assist modes; And / or, provide adaptive assist RPM based on the rider's historical smooth cadence.

7. The electric-assist bicycle controller according to claim 1, characterized in that, A heart rate sensor is also integrated into the multimodal sensing layer to help determine the cyclist's physiological fatigue state.

8. A method for intelligent control of an electric-assisted bicycle, characterized in that, The electric-assisted bicycle controller according to any one of claims 1 to 7, the control method includes: Acquire first data characterizing the vibration state of the electric-assist bicycle body, collected by the vehicle body dynamic sensor; Acquire second data, collected by cyclist status sensors, that characterizes the cyclist's body posture or pedaling behavior; Based on the correlation analysis of the first and second data, the source of vibration is identified as either the first type of vibration caused by uneven road surface or the second type of vibration caused by changes in the rider's own state. When the first type of jitter is identified, a first preset control strategy aimed at smoothing the motor-assisted output is executed; When the vibration is identified as the second type, a second preset control strategy designed to assist the rider in restoring stable riding is executed.

9. The intelligent control method for electric-assisted bicycles according to claim 8, characterized in that, The multimodal perception layer also integrates a global positioning system module for environmental perception, a forward-looking camera, and a millimeter-wave radar, and high-precision map data is pre-loaded into the controller's memory. Real-time perception of environmental and / or geographical information; based on the perceived environmental and / or geographical information, prediction of future riding conditions through a preset path gradient prediction model; and advance adjustment of assist output and / or energy recovery strategies.

10. An electric-assisted bicycle, characterized in that, The electric-assist bicycle controller according to any one of claims 1 to 7 further includes a frame and a motor.

Images