A method for monitoring battery fire in energy vehicles based on big data

Abstract

The present invention relates to the technical field of new energy vehicles, and discloses a method for monitoring battery fires in energy vehicles based on big data, including: step 1, arranging a multimodal sensor network at the battery unit, module and system level, wherein the sensors include temperature sensors, voltage sensors, current sensors, gas sensors, optical fiber sensors and vibration sensors to collect raw data related to temperature, voltage, current, gas concentration, heat distribution and vibration during battery operation. By arranging a multimodal sensor network, including temperature, voltage, current, gas, optical fiber and vibration sensors, and combining a dynamic sampling frequency adjustment strategy, the granularity and sensitivity of data collection are intelligently adjusted according to the battery operating conditions, and comprehensive feature data is generated to achieve full coverage and accurate monitoring of the battery operating status, and obtain the effect of multi-dimensional and high-quality monitoring data, providing a solid data foundation for subsequent analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy vehicles, and in particular to a method for monitoring battery fires in energy vehicles based on big data. Background Art

[0002] In recent years, with the rapid development of the new energy vehicle industry, lithium batteries have been increasingly used as core energy storage devices. However, lithium batteries are prone to thermal runaway due to overcharging, over-discharging, short circuits or external environmental factors under high energy density, high power output and complex working conditions, which can cause fires or even explosions. Especially in extreme conditions such as vehicle driving, charging or collision, battery fire accidents are characterized by strong suddenness and rapid spread, posing a serious threat to human and vehicle safety and public trust.

[0003] Traditional battery monitoring systems are mostly based on single sensor technology or threshold monitoring methods, which can monitor limited battery status parameters, such as temperature or voltage fluctuations, and cannot fully cover complex battery operating conditions. At the same time, they mostly use static thresholds as early warning mechanisms and cannot cope with complex failure modes of batteries under dynamic conditions. In addition, existing fault response strategies mostly rely on fixed rules, which reduce reaction speed and accuracy and cannot promptly control the risk of fire spread.

[0004] In recent years, with the widespread application of big data and artificial intelligence technologies, monitoring methods based on multimodal data fusion, dynamic prediction and intelligent response have gradually become a research hotspot. Through multi-dimensional data collection and analysis, the method can comprehensively describe the operating status of the battery and accurately predict faults. However, existing research has problems such as unreasonable sensor layout, low data processing efficiency, and insufficient adaptability of prediction models to complex working conditions.

[0005] Therefore, those skilled in the art provide an energy vehicle battery fire monitoring method based on big data to solve the above-mentioned problems. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention provides an energy vehicle battery fire monitoring method based on big data to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for monitoring battery fires in energy vehicles based on big data, comprising:

[0008] Step 1: Arrange a multimodal sensor network at the battery cell, module and system level. The sensors include temperature sensors, voltage sensors, current sensors, gas sensors, optical fiber sensors and vibration sensors to collect raw data related to temperature, voltage, current, gas concentration, heat distribution and vibration during battery operation.

[0009] Step 2: Perform real-time signal processing on the raw data collected in step 1, filter out the noise of the sensor data and smooth the state to obtain noise-free multimodal data. The signal processing includes removing high-frequency interference signals in the temperature, voltage and gas concentration data;

[0010] Step 3: extract features from the noise-free multimodal data obtained in step 2, classify and process the data according to the source and type of the sensor data, and fuse the temperature data, voltage data, gas concentration data and vibration data in a weighted manner to generate corresponding comprehensive feature data. The comprehensive feature data includes the battery temperature, battery voltage fluctuation, gas concentration abnormality and vibration impact degree after weighted fusion;

[0011] Step 4: Based on the comprehensive feature data generated in step 3, a preliminary analysis is performed in the edge computing device to detect the data of temperature rise, voltage fluctuation and gas concentration mutation, and the detection results are classified according to the preset threshold range to generate real-time analysis results;

[0012] Step 5: Upload the real-time analysis results generated in step 4 to the distributed storage system, and store the original data collected in step 1 and the comprehensive feature data generated in step 3 in time segments. The distributed storage system associates the real-time analysis results with the historical data for subsequent data analysis and call;

[0013] Step 6: Based on the real-time analysis results and historical data in step 5, the prediction model is used to analyze the future trend of the battery operation status, predict the changes in battery temperature, voltage and gas concentration in a short period of time, and compare the prediction results with the real-time analysis results to confirm potential abnormalities;

[0014] Step 7: Calculate the risk score of the battery based on the comparison result between the prediction result and the real-time analysis result in step 6. The risk score is based on a comprehensive evaluation of temperature change, voltage fluctuation and abnormal gas concentration to generate a score value representing the current battery operation safety status.

[0015] Step 8: Based on the risk score calculated in step 7, the battery operating status is graded and corresponding response measures are triggered. The response measures include implementing battery cooling, power-off protection and fire alarm functions according to the risk level. At the same time, real-time operating status information is pushed to users and maintenance platforms through the Internet of Vehicles.

[0016] Preferably, the arrangement of the multimodal sensors in step 1 is an optimized arrangement, and the optimization goal is to minimize the distance between the sensor arrangement points and cover the high-risk areas. The optimization goal formula is:

[0017]

[0018] Among them, pi and p j represents the layout location of the i-th and j-th sensors, and n is the total number of sensors.

[0019] Preferably, the signal processing process in step 2 includes a Kalman filter algorithm, and the filtering formula is:

[0020] x k+1 =Ax k +Bu k +w k ,

[0021] z k =Hx k +v k ,

[0022] Among them, x k is the state vector, z k is the observed value, A is the state transfer matrix, H is the observation matrix, B is the control input matrix, u k is the control input, w k and v k are process noise and observation noise.

[0023] Preferably, the process noise w in the Kalman filter algorithm is k and observation noise v k The distribution of satisfies the Gaussian distribution:

[0024] w k ~N(0,Q),

[0025] v k ~N(0,R),

[0026] Among them, Q and R are the process noise covariance matrix and the observation noise covariance matrix, and the values ​​of Q and R are obtained by fitting historical data.

[0027] Preferably, the sensor weight w in the multimodal data weighted fusion process in step 3 is i Dynamically adjust the weight based on the historical accuracy of the sensor. The weight adjustment formula is:

[0028]

[0029] Among them, Acc(i) is the accuracy of the i-th sensor in the historical data, n is the total number of sensors, and Acc(j) is the accuracy of the j-th sensor in the historical data.

[0030] Preferably, the preliminary analysis of the edge computing device in step 4 adopts a logistic regression model, and the formula for calculating the abnormal probability P(y=1|X) of the model is:

[0031]

[0032] Among them, X is the input feature vector, β is the weight vector of the logistic regression model, e is a natural constant, and β T It is the dot product of the weight vector β and the input feature vector X.

[0033] Preferably, in step 5, the distributed storage system partitions and stores the data according to time T and sensor type S, and the data partitioning rule is:

[0034]

[0035] Where D represents the data set, D t.s Represents a subset of data at time t and sensor type s.

[0036] Preferably, the real-time data and historical data in the distributed storage system are stored according to a cold-hot separation strategy, the storage medium for the real-time data is a solid-state hard disk, and the storage medium for the historical data is a mechanical hard disk.

[0037] Preferably, the prediction model in step 6 adopts a long short-term memory network, wherein the hidden state update formula of the time series prediction process is:

[0038] h t =σ(W h h t-1 +W x x t +b h ),

[0039] Among them, h t is the hidden state, W h and W x is the hidden layer weight matrix and the input weight matrix, b h is the bias term, x t is the input feature and σ is the activation function.

[0040] Preferably, the calculation formula of the risk score R in step 7 is:

[0041] R=w1T+w2ΔV+w3C gas ,

[0042] Where T is the battery temperature, ΔV is the battery voltage change, C gas is the abnormal value of gas concentration, w1, w2 and w3 are the corresponding weighting coefficients.

[0043] The present invention provides a method for monitoring battery fires in energy vehicles based on big data. It has the following beneficial effects:

[0044] 1. The present invention arranges a multimodal sensor network, including temperature, voltage, current, gas, optical fiber and vibration sensors, and combines a dynamic sampling frequency adjustment strategy to intelligently adjust the granularity and sensitivity of data collection according to the battery operating conditions, generate comprehensive feature data, achieve all-round coverage and accurate monitoring of the battery operating status, obtain multi-dimensional, high-quality monitoring data, and provide a solid data foundation for subsequent analysis.

[0045] 2. The present invention adopts digital twin modeling technology to simulate the internal chemical and thermal evolution process of the battery in real time based on the thermodynamic model, and combines real-time sensor data for dynamic correction to achieve synchronous updating of the virtual model and the actual battery status, thereby obtaining a high-precision prediction effect on the complex failure mode of thermal runaway, and providing effective support for the early warning of battery fires.

[0046] 3. The present invention introduces a long short-term memory network to perform time series prediction on battery parameters, capture the dynamic change trends of temperature, voltage and gas concentration, realize dynamic forward-looking prediction of battery operating status, obtain early perception of potential risks, and buy time for abnormal intervention.

[0047] 4. The present invention designs a multi-level risk assessment and response mechanism, calculates a comprehensive risk score based on temperature, voltage changes and gas concentration, and triggers graded response measures such as power outages, cooling and fire extinguishing according to the risk level, thereby achieving accurate and efficient emergency handling, significantly reducing the risk of fire spread and protecting user safety, and improving the reliability of emergency response. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0049] In order to make the technical personnel in the technical field understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in combination with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a partial embodiment of the present invention, not a complete embodiment. Based on the embodiment of the present invention, other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0050] The present invention is described in detail below in conjunction with the accompanying drawings:

[0051] Example:

[0052] Please see attached Figure 1 The embodiment of the present invention provides a method for monitoring battery fire of energy vehicles based on big data, comprising:

[0053] Step 1: Arrange a multimodal sensor network at the battery cell, module and system level. The sensors include temperature sensors, voltage sensors, current sensors, gas sensors, optical fiber sensors and vibration sensors to collect raw data related to temperature, voltage, current, gas concentration, heat distribution and vibration during battery operation.

[0054] Step 2: Perform real-time signal processing on the raw data collected in step 1, filter out the noise of the sensor data and smooth the state to obtain noise-free multimodal data. The signal processing includes removing high-frequency interference signals in the temperature, voltage and gas concentration data;

[0055] Step 3: extract features from the noise-free multimodal data obtained in step 2, classify and process the data according to the source and type of the sensor data, and fuse the temperature data, voltage data, gas concentration data and vibration data in a weighted manner to generate corresponding comprehensive feature data. The comprehensive feature data includes the battery temperature, battery voltage fluctuation, gas concentration abnormality and vibration impact degree after weighted fusion;

[0056] Step 4: Based on the comprehensive feature data generated in step 3, a preliminary analysis is performed in the edge computing device to detect the data of temperature rise, voltage fluctuation and gas concentration mutation, and the detection results are classified according to the preset threshold range to generate real-time analysis results;

[0057] Step 5: Upload the real-time analysis results generated in step 4 to the distributed storage system, and store the original data collected in step 1 and the comprehensive feature data generated in step 3 in time segments. The distributed storage system associates the real-time analysis results with the historical data for subsequent data analysis and call;

[0058] Step 6: Based on the real-time analysis results and historical data in step 5, the prediction model is used to analyze the future trend of the battery operation status, predict the changes in battery temperature, voltage and gas concentration in a short period of time, and compare the prediction results with the real-time analysis results to confirm potential abnormalities;

[0059] Step 7: Calculate the risk score of the battery based on the comparison result between the prediction result and the real-time analysis result in step 6. The risk score is based on a comprehensive evaluation of temperature change, voltage fluctuation and abnormal gas concentration to generate a score value representing the current battery operation safety status.

[0060] Step 8: Based on the risk score calculated in step 7, the battery operating status is graded and corresponding response measures are triggered. The response measures include implementing battery cooling, power-off protection and fire alarm functions according to the risk level. At the same time, real-time operating status information is pushed to users and maintenance platforms through the Internet of Vehicles.

[0061] Step 1 Benefits: By arranging temperature, voltage, current, gas, optical fiber and vibration multi-modal sensors, all-round monitoring of the battery operating status can be achieved, covering multiple physical and chemical signals of battery heat, gas and vibration, and can capture potential signals of battery abnormalities caused by thermal runaway, short circuit or external impact, providing a multi-dimensional high-quality data source for subsequent analysis;

[0062] Step 2 Benefits: By denoising and smoothing the raw data, high-frequency interference signals in the sensor data are eliminated, the accuracy and stability of the data are improved, and the processed noise-free multimodal data lays a solid foundation for anomaly detection and feature extraction. At the same time, it avoids misjudgment or missed judgment caused by data noise;

[0063] Step 3 Benefits: By classifying and weighted fusion of noise-free data, the generated comprehensive feature data can effectively characterize the current operating status of the battery, including temperature, voltage fluctuations, abnormal gas concentration and vibration effects. Multimodal fusion improves the reliability and information density of the data, making subsequent analysis more targeted and accurate.

[0064] Step 4 Benefits: Preliminary analysis of comprehensive feature data in edge computing devices can quickly detect abnormal conditions such as temperature rise, voltage fluctuation, and sudden change in gas concentration, classify battery operating status in real time, generate analysis results, and localized processing reduces data transmission delays and improves the system's real-time response capabilities.

[0065] Step 5 Benefits: The distributed storage system can realize the time-segmented storage and associated management of real-time data and historical data, providing an efficient storage architecture for data tracing, comparison and analysis. At the same time, associated storage improves data accessibility and query efficiency, providing rich data support for prediction models and risk assessment.

[0066] Benefits of step 6: Using real-time analysis results and historical data, the prediction model can be used to analyze the future changes in battery temperature, voltage, and gas concentration, and identify potential anomalies in advance. The comparison between prediction and real-time analysis can enhance the system's fault perception ability and provide a more accurate basis for risk scoring and response decisions.

[0067] Step 7 Benefits: By calculating risk scores based on temperature changes, voltage fluctuations, and abnormal gas concentrations, the current safety status of the battery is quantified, providing a scientific basis for subsequent risk grading and response. The scoring system enables the monitoring system to dynamically adjust the warning threshold and response strategy, improving the intelligence level of the system;

[0068] Benefits of step 8: By grading risk scores, response measures such as cooling, power outages and fire alarms are triggered, and real-time status information is pushed to users and maintenance platforms through the Internet of Vehicles, realizing a closed loop of the entire process from monitoring to emergency response. The multi-level response mechanism effectively reduces the risk of fire spread and improves the safety of users and vehicles.

[0069] In summary, the steps of the present invention are closely linked to each other, forming a complete monitoring process from data collection to predictive analysis, and then to risk assessment and intelligent response. The multimodal sensor network provides a comprehensive data foundation, real-time signal processing and data fusion enhance the quality and credibility of the data, distributed storage and edge computing ensure the real-time and efficiency of the system, and risk scoring and multi-level response mechanisms significantly improve the intelligence and safety of the monitoring system. It can monitor battery fires comprehensively, accurately and efficiently, effectively reduce the probability of battery thermal runaway and fire accidents, and provide reliable protection for the safety management of new energy vehicles.

[0070] The layout of the multimodal sensors in step 1 is optimized. The optimization goal is to minimize the distance between sensor layout points and cover high-risk areas. The optimization goal formula is:

[0071]

[0072] Among them, p i and p j represents the layout location of the i-th and j-th sensors, and n is the total number of sensors.

[0073] By optimizing the layout of sensors, the goal is to minimize the distance between sensor layout points while ensuring coverage of high-risk areas. The optimized layout plan can enable sensors to cover key areas more concentratedly, improving the sensitivity and accuracy of monitoring potential anomalies.

[0074] When the total number of sensors is fixed, redundant coverage can be reduced by optimizing the layout, reducing the proportion of invalid monitoring areas and achieving efficient use of resources;

[0075] By minimizing the distance between deployment points, data collection between sensors is more uniform, reducing the risk of missed fault detection due to monitoring blind spots or uneven monitoring density;

[0076] The optimized layout strategy can be dynamically adjusted according to the specific physical structure of the battery to adapt to battery packs of different sizes and shapes, thereby improving the applicability and flexibility of the solution;

[0077] The sensor layout covering high-risk areas reduces the system's monitoring blind spots in key areas, thereby enhancing the robustness and adaptability of the monitoring system, especially under dynamic conditions.

[0078] Through this optimized layout, the monitoring system can achieve accurate and comprehensive monitoring of the battery operating status, significantly improve monitoring efficiency and sensitivity, and provide more reliable basic support for subsequent abnormal detection and response.

[0079] The signal processing process in step 2 includes the Kalman filter algorithm, and the filtering formula is:

[0080] x k+1 =Ax k +Bu k +w k ,

[0081] z k =Hx k +v k ,

[0082] Among them, x k is the state vector, z k is the observed value, A is the state transfer matrix, H is the observation matrix, B is the control input matrix, u k is the control input, w k and v k are process noise and observation noise;

[0083] Process noise w in Kalman filter algorithm k and observation noise v k The distribution of satisfies the Gaussian distribution:

[0084] w k ~N(0,Q),

[0085] v k ~N(0,R),

[0086] Among them, Q and R are the process noise covariance matrix and the observation noise covariance matrix, and the values ​​of Q and R are obtained by fitting historical data.

[0087] By introducing the Kalman filter algorithm in the signal processing process, based on the optimal design of parameters such as the state transfer matrix, observation matrix, control input matrix, and the Gaussian distribution assumption of process noise and observation noise, the recursive update characteristics of the Kalman filter are utilized to efficiently suppress the high-frequency noise in multimodal sensor data in scenarios with high real-time requirements, thereby obtaining smooth and accurate signal output, significantly improving the availability of the original data.

[0088] Kalman filtering can filter out sensor noise and can accurately estimate the current state of the system through the recursive update formula of the state vector. In particular, when the sensor data is interfered with or partially lost, it can provide a more accurate estimate based on the historical state.

[0089] Through the design of the state transfer matrix A and the observation matrix H, the Kalman filter can adapt to the complex correlation characteristics of multimodal sensor data and provide a highly robust processing framework for the joint analysis of multiple parameters such as temperature, voltage, and gas concentration.

[0090] The Kalman filter uses a recursive form to update the state estimate. Compared with non-recursive algorithms, it has low computational complexity and can achieve real-time processing in resource-constrained edge computing devices, providing low-latency data support for subsequent analysis.

[0091] By combining state transfer and observation equations, Kalman filtering can enhance the correlation between multimodal sensor data, avoid error accumulation caused by inconsistent data distribution, and lay a solid foundation for subsequent feature extraction and risk analysis.

[0092] In summary, the application of Kalman filtering algorithm in signal processing can effectively filter out sensor noise, smooth data, dynamically adjust state estimation values, enhance the adaptability and accuracy of multimodal data, provide accurate input data for the monitoring system, and improve the sensitivity and reliability of anomaly detection.

[0093] Sensor weight w in the multimodal data weighted fusion process in step 3 i Dynamically adjust the weight based on the historical accuracy of the sensor. The weight adjustment formula is:

[0094]

[0095] Among them, Acc(i) is the accuracy of the i-th sensor in the historical data, n is the total number of sensors, and Acc(j) is the accuracy of the j-th sensor in the historical data.

[0096] Dynamically adjust sensor weights so that the weight contributed by sensors with high historical accuracy is increased, weakening the negative impact of low-accuracy sensors on comprehensive feature data, thereby generating more reliable comprehensive feature data and reducing the false alarm rate of anomaly detection.

[0097] By dynamically adjusting the weights, it can adapt to the diverse needs of different operating conditions. If a certain type of sensor performs well under specific conditions, the weight will automatically increase, thereby optimizing the data fusion results.

[0098] Different types of sensors have different sensitivities and applicability in battery operation monitoring. The weight adjustment formula makes full use of complementarity to make the integrated feature data after fusion more comprehensive and can more accurately characterize the current state of the battery.

[0099] The low-accuracy sensor has a lower weight in the historical data, which means that the impact of its noise on the comprehensive feature data is effectively suppressed. Compared with the traditional fixed weight fusion method, this method significantly improves the anti-interference ability.

[0100] In summary, the method of dynamically adjusting sensor weights based on historical accuracy in this step makes full use of the complementarity of multimodal data and the performance differences of sensors. The generated comprehensive feature data is more reliable, comprehensive and adaptive, which enhances the robustness of data processing and provides high-quality basic data for subsequent anomaly detection and risk assessment.

[0101] The preliminary analysis of the edge computing device in step 4 uses a logistic regression model, and the model calculates the abnormal probability P(y=1|X) as follows:

[0102]

[0103] Among them, X is the input feature vector, β is the weight vector of the logistic regression model, e is a natural constant, and β T It is the dot product of the weight vector β and the input feature vector X.

[0104] The logistic regression model runs on edge computing devices with low computational complexity, quickly analyzes multimodal comprehensive feature data, and calculates the probability of anomalies in real time, ensuring that the system can promptly perceive and respond to potential battery anomalies, significantly reducing the time delay for data transmission to the cloud.

[0105] The logistic regression model calculates the abnormal probability based on the dot product of the input feature vector X and the weight vector, which can effectively distinguish between normal and abnormal states. This model is suitable for binary classification of battery operating status and has strong interpretability and flexibility.

[0106] The model's weight vector N7 is dynamically trained and updated based on historical data and actual application needs, thereby improving the model's classification performance under different operating conditions and enhancing the monitoring system's adaptability to battery anomalies.

[0107] In summary, the logistic regression model is used for preliminary analysis of edge computing devices to achieve fast, accurate, and lightweight anomaly detection, which significantly improves the real-time performance, computing efficiency, and adaptability of the system. At the same time, it reduces data transmission and computing pressure, providing strong support for comprehensive battery fire monitoring.

[0108] In step 5, the distributed storage system partitions and stores the data according to time T and sensor type S. The data partitioning rule is:

[0109]

[0110] Where D represents the data set, D t.s A subset of data representing time t and sensor type s;

[0111] In the distributed storage system, real-time data and historical data are stored according to the hot-cold separation strategy. The storage medium for real-time data is a solid-state drive, and the storage medium for historical data is a mechanical hard drive.

[0112] The distributed storage system significantly improves data management efficiency and access performance through the partition storage rules of time and sensor type, combined with the hot and cold separation storage strategy, while reducing the storage cost of the system. This design supports the collaborative work of real-time and offline analysis, providing strong support for the long-term accumulation and efficient use of large-scale data.

[0113] The prediction model in step 6 adopts the long short-term memory network, where the hidden state update formula of the time series prediction process is:

[0114] h t =σ(W h h t-1 +W x x t +b h ),

[0115] Among them, h t is the hidden state, W h and W x is the hidden layer weight matrix and the input weight matrix, b h is the bias term, x t is the input feature and σ is the activation function.

[0116] The long short-term memory network model has shown significant advantages in the battery fire monitoring system with its powerful time series modeling capabilities and adaptability to nonlinear characteristics. It can accurately predict the future trend of battery operation status, effectively capture potential anomalies, provide a scientific basis for real-time warning and subsequent response, and greatly improve the intelligence level and reliability of the monitoring system.

[0117] The calculation formula for the risk score R in step 7 is:

[0118] R=w1T+w2ΔV+w3C gas ,

[0119] Where T is the battery temperature, ΔV is the battery voltage change, C gas is the abnormal value of gas concentration, w1, w2 and w3 are the corresponding weighting coefficients.

[0120] The risk scoring formula provides a quantitative risk assessment tool for battery safety status by comprehensively evaluating changes in temperature, voltage and gas concentration. The advantages of multi-factor comprehensive analysis, enhanced sensitivity and real-time support enable the system to more accurately identify potential risks, provide clear early warning and response basis, and significantly improve the intelligence level and reliability of the monitoring system.

[0121] Although 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 the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring battery fire in energy vehicles based on big data, characterized in that: include: Step 1: Arrange a multimodal sensor network at the battery cell, module and system level. The sensors include temperature sensors, voltage sensors, current sensors, gas sensors, optical fiber sensors and vibration sensors to collect raw data related to temperature, voltage, current, gas concentration, heat distribution and vibration during battery operation. Step 2: Perform real-time signal processing on the raw data collected in step 1, filter out the noise of the sensor data and smooth the state to obtain noise-free multimodal data. The signal processing includes removing high-frequency interference signals in the temperature, voltage and gas concentration data; Step 3: extract features from the noise-free multimodal data obtained in step 2, classify and process the data according to the source and type of the sensor data, and fuse the temperature data, voltage data, gas concentration data and vibration data in a weighted manner to generate corresponding comprehensive feature data. The comprehensive feature data includes the battery temperature, battery voltage fluctuation, gas concentration abnormality and vibration impact degree after weighted fusion; Step 4: Based on the comprehensive feature data generated in step 3, a preliminary analysis is performed in the edge computing device to detect the data of temperature rise, voltage fluctuation and gas concentration mutation, and the detection results are classified according to the preset threshold range to generate real-time analysis results; Step 5: Upload the real-time analysis results generated in step 4 to the distributed storage system, and store the original data collected in step 1 and the comprehensive feature data generated in step 3 in time segments. The distributed storage system associates the real-time analysis results with the historical data for subsequent data analysis and call; Step 6: Based on the real-time analysis results and historical data in step 5, the prediction model is used to analyze the future trend of the battery operation status, predict the changes in battery temperature, voltage and gas concentration in a short period of time, and compare the prediction results with the real-time analysis results to confirm potential abnormalities; Step 7: Calculate the risk score of the battery based on the comparison result between the prediction result and the real-time analysis result in step 6. The risk score is based on a comprehensive evaluation of temperature change, voltage fluctuation and abnormal gas concentration to generate a score value representing the current battery operation safety status. Step 8: Based on the risk score calculated in step 7, the battery operating status is graded and corresponding response measures are triggered. The response measures include implementing battery cooling, power-off protection and fire alarm functions according to the risk level. At the same time, real-time operating status information is pushed to users and maintenance platforms through the Internet of Vehicles.

2. According to the big data-based energy vehicle battery fire monitoring method of claim 1, it is characterized in that: The arrangement of the multimodal sensors in step 1 is an optimized arrangement. The optimization goal is to minimize the distance between sensor arrangement points and cover high-risk areas. The optimization goal formula is: Among them, p i and p j represents the layout location of the i-th and j-th sensors, and n is the total number of sensors.

3. The energy vehicle battery fire monitoring method based on big data according to claim 1 is characterized in that: The signal processing process in step 2 includes a Kalman filter algorithm, and the filter formula is: x k+1 =Ax k +Bu k +w k , z k =Hx k +v k , Among them, x k is the state vector, z k is the observed value, A is the state transfer matrix, H is the observation matrix, B is the control input matrix, u k is the control input, w k and v k are process noise and observation noise.

4. The method for monitoring energy vehicle battery fire based on big data according to claim 3 is characterized in that: The process noise w in the Kalman filter algorithm k and observation noise v k The distribution of satisfies the Gaussian distribution: Among them, Q and R are the process noise covariance matrix and the observation noise covariance matrix, and the values ​​of Q and R are obtained by fitting historical data.

5. The energy vehicle battery fire monitoring method based on big data according to claim 1 is characterized in that: The sensor weight w in the multimodal data weighted fusion process in step 3 i Dynamically adjust the weight based on the historical accuracy of the sensor. The weight adjustment formula is: Among them, Acc(i) is the accuracy of the i-th sensor in the historical data, n is the total number of sensors, and Acc(j) is the accuracy of the j-th sensor in the historical data.

6. The method for monitoring energy vehicle battery fire based on big data according to claim 1 is characterized in that: The preliminary analysis of the edge computing device in step 4 adopts a logistic regression model, and the formula for calculating the abnormal probability P(y=1|X) of the model is: Among them, X is the input feature vector, β is the weight vector of the logistic regression model, e is a natural constant, and β T It is the dot product of the weight vector β and the input feature vector X.

7. The method for monitoring energy vehicle battery fire based on big data according to claim 1, characterized in that: In step 5, the distributed storage system partitions and stores the data according to time T and sensor type S. The data partitioning rule is: Where D represents the data set, D t.s Represents a subset of data at time t and sensor type s.

8. The method for monitoring energy vehicle battery fire based on big data according to claim 7 is characterized in that: In the distributed storage system, real-time data and historical data are stored according to a cold-hot separation strategy. The storage medium for real-time data is a solid-state hard disk, and the storage medium for historical data is a mechanical hard disk.

9. The method for monitoring energy vehicle battery fire based on big data according to claim 1, characterized in that: The prediction model in step 6 adopts a long short-term memory network, wherein the hidden state update formula of the time series prediction process is: h t =σ(W h h t-1 +W x x t +b h ), Among them, h t is the hidden state, W h and W x is the hidden layer weight matrix and the input weight matrix, b h is the bias term, x t is the input feature and σ is the activation function.

10. The method for monitoring energy vehicle battery fire based on big data according to claim 1, characterized in that: The calculation formula of the risk score R in step 7 is: R=w1T+w2ΔV+w3C gas , Where T is the battery temperature, ΔV is the battery voltage change, C gas is the abnormal value of gas concentration, w1, w2 and w3 are the corresponding weighting coefficients.

Images