Data-driven electric bicycle battery fire hazard detection and protection method

Abstract

The invention relates to the technical field of electric bicycles, and discloses a data-driven electric bicycle battery fire hazard detection and protection method, which comprises the following steps: S1, multi-modal data acquisition and fusion, S2, data preprocessing and feature extraction, S3, data-driven intelligent detection: classifying normal and abnormal states through a supervised learning model, S4, hierarchical protection strategy, and S5, data-driven intelligent detection. S5, cloud data optimization and model continuous learning; S6, model data updating; S7, intelligent linkage and long-term optimization; S8, cloud risk data acquisition; and low, middle and high three-level protection mechanisms are adopted. Through dynamic risk assessment, charging and discharging power is limited in a medium risk, a charger is automatically disconnected, a user is prompted to stop operation, active protection is triggered in a high risk, and sound-light alarm and remote notification are triggered at the same time, so that accidents are prevented from further spreading.

Description

Technical Field

[0001] The present invention relates to the technical field of electric bicycles, and in particular to a data-driven electric bicycle battery fire hazard detection and protection method. Background Art

[0002] An electric bicycle (electric vehicle or electric bike for short) is a vehicle that combines a traditional bicycle with an electric drive system. It is powered by a battery and usually has an electric power assist function, which can help riders reduce manpower consumption and improve riding comfort and efficiency.

[0003] The battery of an electric bicycle is one of its core components, directly affecting the performance indicators of the electric bicycle, such as endurance, service life, and charging speed. Currently, the batteries of electric bicycles are mainly divided into several types, the most common of which are lithium batteries and lead-acid batteries.

[0004] Most existing protective measures are triggered by a single action, such as simply cutting off power or issuing a warning. There is a lack of a tiered response to risk levels. During high-risk processes, only reminders are given without any other means of protection, resulting in excessive or insufficient response during the hidden danger stage, resulting in limited protective effects. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a data-driven electric bicycle battery fire hazard detection and protection method to solve the problem that only power is cut off or warning is issued, there is a lack of tiered response to risk levels, and excessive or insufficient response may occur in the hazard stage, resulting in limited protection effect.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a data-driven electric bicycle battery fire hazard detection and protection method, comprising the following steps:

[0007] S1. Multimodal data acquisition and fusion: Sensors collect battery operating data and external environmental parameters to form a multimodal data system. High-frequency core data and low-frequency environmental and user behavior data are then collected for long-term trend analysis.

[0008] S2. Data preprocessing and feature extraction: First, the collected data is preliminarily processed. The processed data is used to generate abnormal data in the simulation scenario to enhance the robustness of the model to extreme working conditions. The key features are extracted using time series analysis methods.

[0009] S3, data-driven intelligent detection, uses supervised learning models to classify normal and abnormal states, uses time series models to predict future failure trends, and uses unsupervised learning methods to identify unknown hidden danger patterns. The detection system is designed as a multi-level architecture, and edge computing is performed by locally deploying lightweight models.

[0010] S4. Grading protection strategy: Design a grading protection mechanism based on the fire hazard risk level. The grading protection mechanism is divided into low-risk, medium-risk and high-risk strategies, and the strategy is systematically implemented based on actual usage;

[0011] S5, cloud data optimization and model continuous learning, uploads operating data to the cloud platform for big data analysis, and the cloud uses complex deep learning models to train global data;

[0012] S6, model data update, generates extreme scenario data through simulation technology, and regularly pushes the optimized model to edge devices through the cloud for dynamic update of local detection capabilities;

[0013] S7, Intelligent Linkage and Long-Term Optimization: Through edge computing and cloud collaboration, the battery's operating status is monitored in real time and corresponding protection mechanisms are triggered. The system establishes a digital twin model of the battery and uses the collected data in combination with the cloud to iteratively optimize the detection algorithm.

[0014] S8. Cloud-based risk data collection: Based on the collaboration between edge computing and the cloud, a cloud-based risk database is built. By stripping normal data, only abnormal data with potential risks is collected and stored. When a battery accident occurs, the system automatically triggers data collection and uploads the relevant data to the cloud database in real time.

[0015] S9 interruption protection mechanism: When the system detects that the uploaded risk data matches the abnormal pattern in the historical data, it automatically interrupts the normal operation of the battery and activates the hierarchical protection strategy and protection mechanism, and feeds the relevant risk data back to the cloud for model update and optimization.

[0016] Preferably, in S1, the sensors include temperature sensors, voltage and current sensors, gas sensors, vibration sensors and environmental sensors; the operating data include voltage, current, temperature and internal resistance of the battery's internal state; the external environmental parameters include temperature, humidity and pressure; the multimodal data system is used to collect and integrate battery operating data and external environmental parameters from a variety of sensors, unifying these data from different sources and types into a comprehensive data system; the high-frequency core data collection includes temperature, voltage and current; the low-frequency environment and user behavior data collection includes riding time and charging time; the multimodal data collection and fusion is used to integrate multi-source data into a unified time series data set through time synchronization and feature fusion technology, so as to reduce data blind spots for battery status.

[0017] Preferably, in S2, the preliminary processing includes cleaning and denoising, outlier processing and standardization, the abnormal data includes overcharging, short circuit or collision, the key features include temperature change rate, voltage fluctuation pattern and current load characteristics, the extreme working conditions include high humidity environment, immersion or soaking, high altitude or low air pressure, the time sequence includes time sequence, dynamic variability and dependence, and the key features are used to provide high-value input for subsequent detection.

[0018] Preferably, in S3, the supervised learning model includes random forest or XGBoost, the time series model includes LSTM, GRU, the unsupervised learning method includes Autoencoder, the multi-level architecture includes first-level rule detection to quickly screen obvious anomalies, and second-level intelligent detection to capture complex hidden dangers through in-depth analysis. The lightweight model includes TinyML, and data-driven intelligent detection is used to perform edge computing through locally deployed lightweight models to reduce data transmission delays and ensure real-time performance and accuracy.

[0019] Preferably, in S4, the low risk includes a slight voltage fluctuation or a slight temperature increase, the medium risk includes a rapid temperature increase or a large current fluctuation, and the high risk includes detecting a thermal runaway signal or a gas leak. Active protection includes cutting off the battery main circuit, isolating the thermal runaway battery cell, and releasing the internal pressure through a pressure relief valve. The strategy of the graded protection mechanism is that the low-risk system sends an early warning message to the user and recommends adjusting the charging method or riding behavior. The medium-risk system limits the battery charging and discharging power, automatically disconnects the charger, or prompts the user to stop riding. At high risk, the system triggers active protection, which is used to prevent the further spread of fire or explosion, and triggers sound and light alarms and remote notifications.

[0020] Preferably, S5, the complex deep learning model includes Transformer, the long-term risks include battery aging or battery bulging, and cloud data optimization and model continuous learning are used to improve the ability to predict long-term risks.

[0021] Preferably, S6, the extreme scenario data includes temperature limit scenarios, electrical limit scenarios, mechanical limit scenarios, environmental limit scenarios and user operation limit scenarios, and the edge device includes an embedded microcontroller, a user mobile phone or a fire alarm.

[0022] Preferably, S7, user behavior data includes riding patterns and charging habits. The establishment of a battery digital twin model refers to the battery pack of an electric bicycle or the batteries of multiple vehicles. The battery digital twin model is used to virtually display the operating status. Intelligent linkage and long-term optimization are used to combine data to continuously improve the adaptability to new hidden dangers, and closed-loop management from data collection to protection execution.

[0023] Preferably, the risk cloud database is used to store data related to battery risks, and the accidents include temperature anomalies and severe fluctuations in battery voltage. The cloud database is used for historical data analysis, identification of potential risk patterns, and further optimization of risk prediction models.

[0024] Preferably, the protection mechanism includes automatic power off, switching to backup power, and alarm notification to the user. The hierarchical protection strategy and protection mechanism are used to ensure that the battery can be dealt with in a timely manner when risks occur, avoiding serious accidents such as fire.

[0025] The present invention provides a data-driven electric bicycle battery fire hazard detection and protection method. It has the following beneficial effects:

[0026] 1. The present invention adopts a three-level protection mechanism of low, medium and high. Through dynamic risk assessment, the system can send early warning information and suggest adjustment operations when the risk is low. When the risk is medium, the charging and discharging power is limited, the charger is automatically disconnected and the user is prompted to stop the operation. When the risk is high, active protection is triggered, including cutting off the main battery circuit, isolating the thermal runaway unit and depressurizing, and at the same time triggering sound and light alarms and remote notifications to prevent the accident from spreading further.

[0027] 2. The present invention constructs a unified time series data system through multimodal data collection of temperature, voltage, current, vibration, gas and environmental sensors. Combined with time synchronization and feature fusion technology, it accurately captures key features such as temperature change rate, voltage fluctuation pattern, current load characteristics and extreme scenario data such as high humidity, water immersion, low air pressure, etc., which reduces the monitoring blind spots that may be caused by a single data source during use, and provides more accurate information and data for subsequent hidden danger detection.

[0028] 3. The present invention establishes a digital twin model of batteries to virtually display the operating status of batteries of a single vehicle or multiple vehicles. By combining user behavior data, riding patterns, charging habits and big data analysis, the system can continuously optimize the detection and protection models and adapt to new hidden danger scenarios such as unknown temperature fluctuation patterns. During use, it achieves the effect of closed-loop management from data collection to protection execution.

[0029] 4. The present invention compares uploaded risk data with historical abnormal patterns. When similar data is found, the battery protection mechanism is automatically triggered to achieve immediate protection of the battery when encountering anomalies, preventing safety issues such as fire. This mechanism can effectively prevent the battery from fault expansion in high-risk situations and protect the lives and property of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a flow chart of the data-driven electric bicycle battery fire hazard detection and protection method of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] Please see the attached Figure 1 The embodiment of the present invention provides a data-driven electric bicycle battery fire hazard detection and protection method, comprising the following steps:

[0033] S1. Multimodal data acquisition and fusion: Sensors collect battery operating data and external environmental parameters to form a multimodal data system. High-frequency core data and low-frequency environmental and user behavior data are then collected for long-term trend analysis.

[0034] S2. Data preprocessing and feature extraction: First, the collected data is preliminarily processed. The processed data is used to generate abnormal data in the simulation scenario to enhance the robustness of the model to extreme working conditions. The key features are extracted using time series analysis methods.

[0035] S3, data-driven intelligent detection, uses supervised learning models to classify normal and abnormal states, uses time series models to predict future failure trends, and uses unsupervised learning methods to identify unknown hidden danger patterns. The detection system is designed as a multi-level architecture, and edge computing is performed by locally deploying lightweight models.

[0036] S4. Grading protection strategy: Design a grading protection mechanism based on the fire hazard risk level. The grading protection mechanism is divided into low-risk, medium-risk and high-risk strategies, and the strategy is systematically implemented based on actual usage;

[0037] S5, cloud data optimization and model continuous learning, uploads operating data to the cloud platform for big data analysis, and the cloud uses complex deep learning models to train global data;

[0038] S6. Model data update: Generate extreme scenario data through simulation technology, and regularly push the optimized model to edge devices through the cloud for dynamic update of local detection capabilities.

[0039] S7, Intelligent Linkage and Long-Term Optimization: Through edge computing and cloud collaboration, the battery's operating status is monitored in real time and corresponding protection mechanisms are triggered. The system establishes a digital twin model of the battery and uses the collected data in combination with the cloud to iteratively optimize the detection algorithm.

[0040] S8. Cloud-based risk data collection: Based on the collaboration between edge computing and the cloud, a cloud-based risk database is built. By stripping normal data, only abnormal data with potential risks is collected and stored. When a battery accident occurs, the system automatically triggers data collection and uploads the relevant data to the cloud database in real time.

[0041] S9 interruption protection mechanism: When the system detects that the uploaded risk data matches the abnormal pattern in the historical data, it automatically interrupts the normal operation of the battery and activates the hierarchical protection strategy and protection mechanism, and feeds the relevant risk data back to the cloud for model update and optimization.

[0042] In S1, sensors include temperature sensors, voltage and current sensors, gas sensors, vibration sensors and environmental sensors. Operation data include voltage, current, temperature and internal resistance of the battery's internal state. External environmental parameters include temperature, humidity and pressure. The multimodal data system is used to collect and integrate battery operation data and external environmental parameters from multiple sensors, and unify these data from different sources and types into a comprehensive data system. High-frequency core data collection includes temperature, voltage and current. Low-frequency environment and user behavior data collection includes riding time and charging time. Multimodal data collection and fusion are used to integrate multi-source data into a unified time series data set through time synchronization and feature fusion technology, which is used to reduce data blind spots for battery status.

[0043] Specifically, by integrating temperature sensors, voltage and current sensors, gas sensors, vibration sensors, and environmental sensors, the system collects real-time battery operating data such as voltage, current, temperature, and internal resistance, as well as external environmental parameters such as temperature, humidity, and pressure. Changes in battery temperature and voltage can reflect risks such as overcharging, overdischarge, or thermal runaway, while environmental parameters are used to assess the impact of external conditions on battery performance. The multimodal data system collects core data such as temperature, voltage, and current at high frequencies, and environmental and user behavior data such as riding time and charging time at low frequencies, ensuring dynamic monitoring and trend analysis of key signals.

[0044] Data is aligned through time synchronization technology, and feature fusion technology extracts key features of multi-source information to form a unified time series data set that comprehensively reflects the battery's operating status. During use, this not only reduces the blind spots of a single data source, providing high-quality data support for subsequent hidden danger detection and prevention, but also accurately identifies and predicts potential risks, effectively improving the safety and reliability of the battery system.

[0045] In the process of multimodal data acquisition and fusion, the present invention adopts a multi-source data fusion method based on the fusion of time synchronization and attention mechanism. First, a unified clock system is used to stamp high-precision timestamps for the data collected by various sensors (including temperature, voltage, current, gas, vibration and environmental sensors), and a sliding window and interpolation algorithm are used to resample the asynchronous sampled data to achieve the alignment of multimodal data in the time dimension. Secondly, in the feature fusion stage, a multi-level fusion strategy is used to integrate the collected data, and the key indicators of different sensors (such as temperature change rate, voltage fluctuation, humidity and water content) are initially combined by early feature splicing. The system combines the sensor data (equal) into a unified feature vector. In combination with the data collection scenario, a weighted fusion algorithm based on the Transformer attention mechanism is introduced to dynamically adjust the fusion weight of each sensor feature according to different working conditions. This enables the system to autonomously identify the data source that is most sensitive to fault trends in the current state, enhancing the model's ability to respond to complex abnormal patterns. Finally, the above fusion technology is used to construct a unified time series dataset as the input of the subsequent intelligent detection model. This effectively avoids the monitoring blind spots caused by a single data source, data imbalance, or different sampling frequencies in traditional methods, and significantly improves the system's perception accuracy and robustness of battery fire hazards.

[0046] The formula of the basic attention mechanism is: For the input feature matrix Where n is the number of time steps and d is the feature dimension. The attention calculation in Transformer is as follows:

[0047]

[0048] in:

[0049] Q=XW Q : Query vector (Query)

[0050] K=XW K :Key vector (Key)

[0051] V=XW V : Value vector (Value)

[0052] W Q ,W K ,W V is the trainable weight matrix

[0053] d k is the dimension of the key vector (used for scaling)

[0054] Softmax is used to normalize weights;

[0055] For data from different sensors (such as temperature, current, voltage, humidity, etc.), they can be concatenated or encoded into a feature sequence: X = [x温度 ,x 电压 ,x 电流 ,x 气体 ,x 振动 ,x 环境 ].

[0056] In S2, preliminary processing includes cleaning and denoising, outlier processing and standardization. Abnormal data includes overcharging, short circuit or collision. Key features include temperature change rate, voltage fluctuation pattern and current load characteristics. Extreme working conditions include high humidity environment, immersion or soaking, high altitude or low air pressure. Time sequence includes time sequence, dynamic variability and dependency. Key features are used to provide high-value input for subsequent detection.

[0057] Specifically, based on multimodal data collection, the present invention further improves data quality and effectiveness through data preprocessing and key feature extraction, providing high-value input for subsequent detection and analysis. First, the collected raw data is cleaned and denoised to eliminate noise caused by sensor errors or environmental interference. Outlier processing is performed to identify and eliminate abnormal data caused by extreme operating conditions such as overcharging, short circuits, and collisions. Finally, through standardization processing, the dimensions of each data dimension are unified to ensure that subsequent models can effectively analyze different types of data.

[0058] Data that has undergone preprocessing and feature extraction has higher accuracy and usability. During the data preprocessing and feature extraction process, only conventional data cleaning, standardization, and outlier removal strategies are initially used, and multiple enhancement mechanisms are introduced to improve the system's ability to perceive extreme battery operating conditions and early hidden dangers. First, an abnormal sample enhancement method based on battery thermal-electric coupling simulation and historical data distribution is proposed to simulate and generate data samples under extreme conditions such as flooding, short circuits, and overcharging to fill in the blind spots in real-world operating data. Second, a sliding window statistics + dynamic threshold algorithm is used to intelligently identify sudden anomalies by real-time statistically analyzing the feature change trend and standard deviation variation within the window, thereby improving the recognition sensitivity to non-stationary signals (such as sudden temperature changes and voltage drops). Third, time series features such as time lag features, local change rates, and sliding mean are introduced in the feature extraction stage to enhance the model's ability to model the dependence of key indicators over time. These processing methods jointly improve the effectiveness and robustness of multimodal data, providing a solid data foundation for subsequent risk trend prediction and intelligent detection. They not only provide high-value input for subsequent hidden danger detection and prediction models, but also significantly improve the system's sensitivity and accuracy in identifying hidden dangers, and effectively respond to various complex battery operating conditions and potential risks.

[0059] In S3, supervised learning models include random forest or XGBoost, time series models include LSTM and GRU, unsupervised learning methods include Autoencoder, multi-level architecture includes first-level rule detection to quickly screen obvious anomalies, second-level intelligent detection to capture complex hidden dangers through in-depth analysis, lightweight models include TinyML, and data-driven intelligent detection is used to perform edge computing through locally deployed lightweight models to reduce data transmission delays and ensure real-time performance and accuracy.

[0060] Specifically, data-driven intelligent detection technology enables real-time monitoring of battery operating status and identification of potential hazards. The intelligent detection system adopts a multi-level architecture to efficiently handle anomaly detection tasks of varying complexity. In the first-level rule-based detection, pre-set rules such as temperature thresholds and upper and lower voltage limits are used to quickly screen for obvious anomalies, including directly visible hazards such as overcharging or short circuits. In the second-level intelligent detection, a deep analysis model is used to conduct in-depth analysis of complex or hidden problems, such as early signs of thermal runaway, to capture complex potential hazard patterns.

[0061] The detection system uses supervised learning models such as random forests or XGBoost to classify labeled data and identify normal and abnormal operating conditions. It also combines time series models such as LSTM or GRU to analyze the dynamic trends of battery operating data and predict possible abnormal conditions in the future. It also introduces unsupervised learning methods such as Autoencoder to detect unknown hidden dangers or new faults by learning normal operating modes.

[0062] The basic model structures used, such as random forest, XGBoost, LSTM, GRU and Autoencoder, belong to the existing model framework, but in the model structure design and input feature construction, the Autoencoder model introduces a local attention mechanism and a residual reconstruction module to improve the sensitivity of unknown hidden danger identification and reconstruction robustness; the time series model adopts a CNN-LSTM hybrid structure to realize multi-scale trend modeling and early prediction of battery operating status; the supervised learning model introduces high-dimensional feature construction and a hierarchical labeling mechanism to improve the model generalization ability; in edge computing deployment, the present invention quantizes and trims the model, and optimizes the model structure and reasoning performance based on the TinyML framework to achieve lightweight operation of the battery safety detection model on the microcontroller platform, significantly improving the system's real-time response capability and energy efficiency performance. In addition, the Autoencoder model introduces a local attention module at the encoding end to focus on high-risk features, and introduces residual connections and dynamic error weighting mechanisms at the decoding end to enhance the model's sensitivity to potential abnormal patterns and support high-fidelity reconstruction of multi-dimensional time series data.

[0063] S4 medium and low risks include slight voltage fluctuations or slight temperature increases, medium risks include rapid temperature increases or large current fluctuations, and high risks include detecting thermal runaway signals or gas leaks. Active protection includes cutting off the battery's main circuit, isolating thermal runaway battery cells, and releasing internal pressure through a pressure relief valve. The strategy of the graded protection mechanism is that the low-risk system sends early warning information to the user and recommends adjusting the charging method or riding behavior. The medium-risk system limits the battery charging and discharging power, automatically disconnects the charger, or prompts the user to stop riding. At high risk, the system triggers active protection, which is used to prevent the further spread of fire or explosion, and at the same time triggers sound and light alarms and remote notifications.

[0064] Specifically, by conducting a risk assessment on the battery operating status, the risks are divided into three levels: low risk, medium risk, and high risk, and corresponding protection strategies are adopted to minimize the occurrence and impact of accidents;

[0065] Low-risk scenarios include minor voltage fluctuations or temperature increases. In these cases, the system will send a warning message to the user, alerting them to potential hazards and suggesting adjustments to charging methods or riding behaviors, such as reducing charging current or avoiding high-load riding. This strategy aims to prevent potential hazards from worsening while not significantly interfering with normal use.

[0066] Medium-risk scenarios include rapid temperature increases or large current fluctuations, which may indicate that the battery is in an abnormal state, but not yet at a dangerous level. For medium-risk scenarios, the system will actively limit the battery's charge and discharge power to avoid high-current discharge or overheating during charging. At the same time, the system can automatically disconnect the charger to stop charging and, if necessary, prompt the user to stop riding through a user device such as a mobile phone app to reduce the risk of developing into a higher level.

[0067] High-risk scenarios include detecting thermal runaway signals or gas leaks. At this point, the system will immediately trigger active protection mechanisms to minimize the occurrence and spread of fire or explosion. Active protection measures include: quickly disconnecting the battery's main circuit to interrupt current flow, isolating the thermal runaway battery cell to prevent the spread of thermal runaway, or releasing internal battery pressure through a pressure relief valve to reduce the risk of explosion. At the same time, the system will trigger audible and visual alarms to attract the attention of the user and nearby personnel, and remotely notify the management platform for emergency response.

[0068] Dynamic responses are implemented based on risk levels, with corresponding safety measures in place for everything from minor hazards to serious incidents. Low-risk situations prioritize prompts and suggestions to minimize user interference; medium-risk situations implement restrictive protection to limit the potential for further expansion; and high-risk situations utilize proactive protection to completely eliminate the source of danger and notify relevant parties for rapid intervention. This effectively prevents serious incidents such as fires and explosions during use, significantly improving battery system safety and user experience.

[0069] S5. Complex deep learning models include Transformer. Long-term risks include battery aging or battery bulging. Cloud data optimization and continuous model learning are used to improve the ability to predict long-term risks.

[0070] Specifically, through cloud-based data optimization and continuous model learning technology, the system's ability to identify and predict long-term risks is enhanced. The cloud platform integrates operational data from multiple edge devices, optimizes detection algorithms and prediction models through big data analysis and deep learning model training, and regularly pushes updated models to edge devices to achieve dynamic upgrades for the entire system.

[0071] The existing Transformer model was optimized in terms of model architecture and learning mechanism, forming a deep learning framework suitable for long-term risk prediction of e-bike batteries. The optimization steps are as follows: First, to address the unstructured and multi-dimensional heterogeneous characteristics of battery operating data, the model introduces a multi-channel embedding layer and an improved position encoding mechanism to enhance the modeling of time series features of parameters such as temperature, voltage, current, and internal resistance. Second, the weighted heat map output by the attention mechanism is combined to visualize the contribution of different features to the prediction results, improving the model's interpretability. Finally, the cloud system adopts a continuous learning mechanism, extracting high-value samples from edge device feedback data for incremental training each cycle, or completing local model fine-tuning based on the federated learning protocol. This allows the model to dynamically adapt over time and usage scenarios, effectively enhancing the ability to identify long-term risk states such as battery aging and bulging.

[0072] S6. Extreme scenario data includes temperature extreme scenarios, electrical extreme scenarios, mechanical extreme scenarios, environmental extreme scenarios, and user operation extreme scenarios. Edge devices include embedded microcontrollers, user mobile phones, or fire alarms.

[0073] Specifically, by combining extreme scenario data with edge devices, the system enhances its ability to detect battery hazards under complex operating conditions and ensures timely response and protective measures in high-risk scenarios. The system relies on real-time collection and analysis of multimodal data, and through the deployment of edge devices, it implements distributed computing and immediate warnings, fully covering the potential hazards of extreme scenarios.

[0074] Temperature extreme scenarios: High temperature scenarios, such as ambient temperatures exceeding 50°C, may cause battery thermal runaway. Low temperature scenarios, such as ambient temperatures below -20°C, may lead to reduced chemical reaction efficiency or lithium deposition. Both scenarios require real-time temperature data and temperature change rate identification.

[0075] Electrical extreme scenarios: Overcharge or over-discharge voltage exceeds the safe range, short-circuit abnormal current rises sharply, and high-current charging and discharging, such as the high current caused by fast charging equipment, may cause internal heat accumulation or battery damage. Voltage and current fluctuation data detection is required;

[0076] Mechanical extreme scenarios: Collision, puncture, and long-term vibration driving under complex road conditions may cause damage to the internal structure of the battery or short circuit. Monitoring relies on vibration data and mechanical stress sensors;

[0077] Extreme environmental scenarios: High humidity may cause short circuits, water immersion or soaking may damage battery insulation, and high altitudes or low air pressure may affect battery heat dissipation efficiency. These scenarios require a comprehensive assessment based on humidity, air pressure, and ambient temperature data.

[0078] User operation extreme scenarios: This includes internal faults caused by improper charging, such as using a low-quality charger, continuous overloading with high power output for a long time, or deep self-discharge due to long-term idleness. This is analyzed through comprehensive analysis of user behavior data, such as charging time, riding mode, and battery status.

[0079] By combining extreme scenario data with edge devices, this invention can accurately identify battery hazards in a variety of complex scenarios and achieve millisecond-level responses through a distributed computing architecture. A low-power embedded microcontroller ensures real-time processing efficiency, while the user's mobile phone improves interactivity and usability. The fire alarm enhances the warning effect of high-risk scenarios, significantly improving system safety and reliability and reducing the risk of battery accidents.

[0080] S7. User behavior data includes riding patterns and charging habits. Establishing a battery digital twin model refers to the battery pack of an electric bicycle or the batteries of multiple vehicles. The battery digital twin model is used to virtually display the operating status. Intelligent linkage and long-term optimization are used to combine data to continuously improve the adaptability to new hidden dangers, and achieve closed-loop management from data collection to protection execution.

[0081] Specifically, by establishing a battery digital twin model and intelligent linkage mechanism, a closed-loop management system from data collection to protection execution is achieved, ensuring real-time monitoring, risk prediction, and dynamic optimization of the battery system. The battery digital twin model can not only virtually display the battery's operating status, but also conduct in-depth analysis based on user behavior data, providing important support for the identification and prevention of new hidden dangers.

[0082] Riding Mode: This includes riding speed, acceleration or deceleration frequency, high-load riding duration, etc., reflecting the dynamic impact of user behavior on battery load. Frequent high-load acceleration may cause battery temperature to rise or abnormal current fluctuations.

[0083] Charging habits: These include the user's charging frequency, charging time, and type of charging device. These behaviors may affect the health of the battery. Prolonged overcharging or using non-standard chargers may cause overcharging, shortening battery life or causing potential safety hazards.

[0084] The battery digital twin model is a virtualized digital model based on the physical battery. It can be used for the battery pack of an e-bike or multiple vehicles. Through real-time data collection and virtual modeling, the digital twin model dynamically displays the operating status of the battery, intuitively presenting key operating parameters such as temperature distribution, voltage status, and load conditions.

[0085] During use, a closed-loop management is achieved from data collection of user behavior and sensor data to virtual display of digital twin models and then to intelligent linkage of protection execution. The system can not only display the operating status of the battery in real time, but also continuously optimize the protection strategy, enhance the ability to identify and adapt to new hidden dangers, and provide comprehensive protection for the safety, reliability and operation efficiency of the battery system.

[0086] The risk cloud database is used to store data related to battery risks, including abnormal temperatures and severe fluctuations in battery voltage. The cloud database is used to analyze historical data, identify potential risk patterns, and further optimize risk prediction models.

[0087] Specifically, through technical means such as multimodal data collection and fusion, data preprocessing and feature extraction, and data-driven intelligent detection, and further through the application of cloud databases to enhance the system's ability to identify and protect against battery fire hazards, the database is specifically used to store data related to battery risks, mainly including abnormal data that may cause fires during battery operation, such as abnormal temperatures, drastic fluctuations in battery voltage, etc. By collecting and storing these abnormal data at a high frequency, the system can update and analyze the operating status of the battery in real time and capture any possible risk patterns. In the event of a battery accident, the system will automatically collect relevant abnormal data and upload it to the cloud database for storage. Through this data, the cloud platform can conduct in-depth analysis of historical data, identify potential risk patterns in battery operation, and thus provide accurate predictions for battery safety;

[0088] The cloud database also enables large-scale data analysis and pattern recognition, continuously accumulating and updating data during battery operation. This data is then optimized through big data analysis and complex deep learning models to predict battery risk. By analyzing this collected data, the cloud system can not only identify the battery's current state but also predict potential failure trends. Further optimized predictive models can help the system accurately anticipate potential future risks and, combined with locally deployed edge computing models, enable effective protective measures to be implemented in advance.

[0089] Protection mechanisms include automatic power-off, switching to backup power, and alarm notification to users. Hierarchical protection strategies and protection mechanisms are used to ensure that batteries are dealt with in a timely manner when risks arise, avoiding serious accidents such as fire.

[0090] Specifically, first, sensors continuously monitor the battery's operating status, collecting battery operating data such as battery temperature, voltage, current, and other key parameters, and combining them with external environmental data to form multimodal data. Using an intelligent detection system, the battery status is monitored in real time through data analysis and risk prediction models. When risk signals such as abnormal temperature or severe voltage fluctuations are detected, the system automatically triggers the preset protection mechanism.

[0091] Secondly, the system activates a tiered protection strategy based on the risk level. For low-risk situations, the system first issues a warning and notifies the user. For medium-risk situations, the system activates battery protection mode, automatically disconnecting the battery power supply. For high-risk situations, the system immediately switches to backup power, issues an emergency alarm, and notifies the user via a mobile app or other communication methods, ensuring they can take further action.

[0092] Finally, all abnormal events and protective actions are uploaded to the cloud database in real time. The system then conducts risk assessments and optimizes models based on historical data to improve future risk warning and protection capabilities. This protection mechanism enables timely and effective response measures when battery risks occur, preventing accidents such as fires and thus ensuring the safe operation of e-bikes.

[0093] During use, data is uploaded to the cloud, and then normal data is eliminated from the data collected from the cloud, and risk data is extracted to build a risk database. When the electric bicycle detects the same data during charging or riding, an early warning is issued. When the data matches the database completely, a hierarchical protection step is taken to interrupt the battery operation.

[0094] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A data-driven electric bicycle battery fire hazard detection and protection method, characterized in that: The following steps are involved: S1. Multimodal data acquisition and fusion: Sensors collect battery operating data and external environmental parameters to form a multimodal data system. High-frequency core data and low-frequency environmental and user behavior data are then collected for long-term trend analysis. S2. Data preprocessing and feature extraction: First, the collected data is preliminarily processed. The processed data is used to generate abnormal data in the simulation scenario to enhance the robustness of the model to extreme working conditions. The key features are extracted using time series analysis methods. S3, data-driven intelligent detection, uses supervised learning models to classify normal and abnormal states, uses time series models to predict future failure trends, and uses unsupervised learning methods to identify unknown hidden danger patterns. The detection system is designed as a multi-level architecture, and edge computing is performed by locally deploying lightweight models. S4. Grading protection strategy: Design a grading protection mechanism based on the fire hazard risk level. The grading protection mechanism is divided into low-risk, medium-risk and high-risk strategies, and the strategy is systematically implemented based on actual usage; S5, cloud data optimization and model continuous learning, uploads operating data to the cloud platform for big data analysis, and the cloud uses complex deep learning models to train global data; S6, model data update, generates extreme scenario data through simulation technology, and regularly pushes the optimized model to edge devices through the cloud for dynamic update of local detection capabilities; S7, Intelligent Linkage and Long-Term Optimization: Through edge computing and cloud collaboration, the battery's operating status is monitored in real time and corresponding protection mechanisms are triggered. The system establishes a digital twin model of the battery and uses the collected data in combination with the cloud to iteratively optimize the detection algorithm. S8. Cloud-based risk data collection: Based on the collaboration between edge computing and the cloud, a cloud-based risk database is built. By stripping normal data, only abnormal data with potential risks is collected and stored. When a battery accident occurs, the system automatically triggers data collection and uploads the relevant data to the cloud database in real time. S9 interruption protection mechanism: When the system detects that the uploaded risk data matches the abnormal pattern in the historical data, it automatically interrupts the normal operation of the battery and activates the hierarchical protection strategy and protection mechanism, and feeds the relevant risk data back to the cloud for model update and optimization.

2. The data-driven electric bicycle battery fire hazard detection and protection method according to claim 1 is characterized in that: In S1, the sensors include temperature sensors, voltage and current sensors, gas sensors, vibration sensors and environmental sensors. The operating data include the voltage, current, temperature and internal resistance of the battery's internal state. The external environmental parameters include temperature, humidity and pressure. The multimodal data system is used to collect and integrate battery operating data and external environmental parameters from multiple sensors, and unify these data from different sources and types into a comprehensive data system. The high-frequency core data collection includes temperature, voltage and current. The low-frequency environment and user behavior data collection includes riding time and charging time. The multimodal data collection and fusion is used to integrate multi-source data into a unified time series data set through time synchronization and feature fusion technology, which is used to reduce data blind spots for battery status.

3. The data-driven electric bicycle battery fire hazard detection and protection method according to claim 1 is characterized in that: In S2, the preliminary processing includes cleaning and denoising, outlier processing and standardization. The abnormal data includes overcharging, short circuit or collision. The key features include temperature change rate, voltage fluctuation pattern and current load characteristics. The extreme working conditions include high humidity environment, immersion or soaking, high altitude or low air pressure. The time sequence includes time sequence, dynamic variability and dependence. The key features are used to provide high-value input for subsequent detection.

4. The data-driven electric bicycle battery fire hazard detection and protection method according to claim 1 is characterized in that: In S3, the supervised learning models include random forest or XGBoost, the time series models include LSTM and GRU, the unsupervised learning methods include Autoencoder, the multi-level architecture includes first-level rule detection to quickly screen obvious anomalies, and second-level intelligent detection to capture complex hidden dangers through in-depth analysis. The lightweight model includes TinyML, and data-driven intelligent detection is used to perform edge computing through locally deployed lightweight models to reduce data transmission delays and ensure real-time performance and accuracy.

5. The data-driven electric bicycle battery fire hazard detection and protection method according to claim 1 is characterized in that: In S4, the low risk includes a slight voltage fluctuation or a slight temperature increase, the medium risk includes a rapid temperature increase or a large current fluctuation, and the high risk includes detecting a thermal runaway signal or a gas leak. Active protection includes cutting off the battery main circuit, isolating the thermal runaway battery cell, and releasing the internal pressure through the pressure relief valve. The strategy of the graded protection mechanism is that the low-risk system sends an early warning message to the user and recommends adjusting the charging method or riding behavior. The medium-risk system limits the battery charging and discharging power, automatically disconnects the charger, or prompts the user to stop riding. At high risk, the system triggers active protection, which is used to prevent the further spread of fire or explosion, and at the same time triggers sound and light alarms and remote notifications.

6. The data-driven electric bicycle battery fire hazard detection and protection method according to claim 1 is characterized in that: S5. The complex deep learning model includes Transformer, and the long-term risks include battery aging or battery bulging. Cloud data optimization and continuous model learning are used to improve the ability to predict long-term risks.

7. The data-driven electric bicycle battery fire hazard detection and protection method according to claim 1 is characterized in that: S6. The extreme scenario data includes temperature extreme scenarios, electrical extreme scenarios, mechanical extreme scenarios, environmental extreme scenarios and user operation extreme scenarios. The edge device includes an embedded microcontroller, a user mobile phone or a fire alarm.

8. The data-driven electric bicycle battery fire hazard detection and protection method according to claim 1 is characterized in that: S7. User behavior data includes riding patterns and charging habits. The establishment of a battery digital twin model refers to the battery pack of an electric bicycle or the batteries of multiple vehicles. The battery digital twin model is used to virtually display the operating status. Intelligent linkage and long-term optimization are used to combine data to continuously improve the adaptability to new hidden dangers, and achieve closed-loop management from data collection to protection execution.

9. The data-driven electric bicycle battery fire hazard detection and protection method according to claim 1, characterized in that: The risk cloud database is used to store data related to battery risks. The accidents include temperature anomalies and severe fluctuations in battery voltage. The cloud database is used to analyze historical data, identify potential risk patterns, and further optimize risk prediction models.

10. The data-driven electric bicycle battery fire hazard detection and protection method according to claim 1, characterized in that: The protection mechanism includes automatic power off, switching to backup power, and alarm notification to the user. The hierarchical protection strategy and protection mechanism are used to ensure that the battery can be dealt with in a timely manner when risks occur, avoiding serious accidents such as fire.

Images