Battery Thermal Runaway Early Warning and Prevention and Control System Based on Multi-Sensor Fusion Neural Network

Abstract

The present invention relates to the field of electric vehicle safety technology, and particularly to a battery thermal runaway early warning and prevention and control system based on a multi-sensor fusion neural network. This system integrates multi-sensor technology to accurately collect battery operation data in real time, providing multi-dimensional support for thermal runaway early warning; the embedded dual-core and dual-head STM32 processor architecture enables parallel system management and neural network algorithms, improving real-time performance and computing efficiency; the neural network early warning module trains and analyzes the data to calculate the thermal runaway early warning level. The early warning control module intelligently controls the vehicle according to the level, comprehensively ensuring battery safety; the system effectively improves the accuracy and response speed of battery thermal runaway early warning, providing strong guarantee for the safe operation of the vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicle safety, and particularly relates to a battery thermal runaway early warning and prevention and control system based on a multi-sensor fusion neural network. Background Art

[0002] With the rapid development of the electric vehicle industry, the safety problem of power batteries has become increasingly prominent. Battery thermal runaway is one of the main causes of electric vehicle safety accidents. Once thermal runaway occurs, it will cause the battery to burn violently or even explode, bringing serious safety hazards to passengers. At present, the common battery thermal runaway early warning systems on the market mainly rely on a single sensor to monitor parameters such as temperature or voltage, and there are problems such as untimely early warning and low accuracy. Most of the existing early warning systems use a simple threshold judgment method, lack the ability to fuse and analyze multi-sensor data, and cannot accurately identify the precursor characteristics of thermal runaway. At the same time, the prevention and control measures of most systems are single, and corresponding prevention and control measures cannot be taken according to different development stages of thermal runaway.

[0003] Traditional battery thermal runaway early warning systems usually have the following problems: First, single-sensor monitoring cannot comprehensively reflect the battery state, which is prone to false alarms or missed alarms; second, simple threshold judgment cannot capture the complex process of thermal runaway development, and the early warning accuracy is limited; third, lack of intelligent analysis algorithms and unable to conduct in-depth data mining; fourth, the prevention and control strategy is simple, usually adopting a one-size-fits-all emergency power-off measure, lacking a gradient prevention and control mechanism.

[0004] Therefore, it is urgent to develop a battery thermal runaway early warning and prevention and control system based on a multi-sensor fusion neural network. Through multi-sensor collaborative monitoring, intelligent algorithm analysis and multi-level early warning and prevention and control, the accuracy and timeliness of battery thermal runaway early warning are improved, and the safe operation of electric vehicles is effectively guaranteed. Summary of the Invention

[0005] The purpose of the present invention is to provide a battery thermal runaway early warning and prevention and control system based on a multi-sensor fusion neural network, aiming to solve the problems of single-sensor monitoring, simple threshold judgment, lack of intelligent analysis and single prevention and control strategy in the existing battery thermal runaway early warning systems. The present invention collects battery operation data through multi-sensor fusion, uses an embedded neural network for intelligent analysis, realizes early warning and hierarchical prevention and control of battery thermal runaway, and effectively improves the early warning accuracy and prevention and control effectiveness.

[0006] The present invention proposes a battery thermal runaway early warning and prevention and control system based on a multi-sensor fusion neural network, including:

[0007] A multi-sensor fusion module for real-time collection of battery operation data; an embedded neural network warning module connected to the multi-sensor fusion module for receiving the battery operation data, training and analyzing the battery operation data, and calculating the thermal runaway warning level; a warning control module connected to the embedded neural network warning module for receiving the thermal runaway warning level and controlling the vehicle according to the thermal runaway warning level.

[0008] Preferably, the multi-sensor fusion module includes a temperature sensor, a smoke sensor, and a gas sensor. The smoke sensor includes a PM2.5 sensor and a TGS2600 sensor. The PM2.5 sensor is used to detect the PM2.5 content in the air, and the TGS2600 sensor is used to detect the VOC concentration in the air. The battery operation data includes the battery operation temperature, the battery operation discharge current, the battery operation SOC, the air PM2.5 content, and the VOC concentration.

[0009] Preferably, the embedded neural network warning module uses an embedded dual-core and dual-head STM32 processor. The main core of the dual-core and dual-head STM32 processor runs the Linux operating system, and the slave core runs the BP neural network module; the dual-core and dual-head STM32 processor and the multi-sensor fusion module form a multi-sensor acquisition circuit through a multi-Sensor data acquisition circuit.

[0010] Preferably, the embedded neural network warning module includes a BP neural network module and a data processing and warning module. The BP neural network module uses an simulated annealing algorithm to initialize the weights and thresholds, and the data processing and warning module is used to convert the calculation results of the BP neural network module into a thermal runaway warning level.

[0011] Preferably, the method for calculating the initial weights and thresholds by the simulated annealing algorithm is as follows: Assume that there are q neurons in the input layer, m neurons in the hidden layer, and r neurons in the output layer. The algorithm continuously adjusts the weights and thresholds to reduce the network training error until a predetermined termination condition is met.

[0012] Preferably, the warning control module includes an in-vehicle CAN bus communication circuit. The in-vehicle CAN bus communication circuit is connected to the embedded neural network warning module through a CAN communication chip and a CAN-RS232 communication conversion chip for transmitting the thermal runaway warning level to the in-vehicle CAN bus.

[0013] Preferably, the thermal runaway warning levels include a first-level warning, a second-level warning, and a third-level warning. The first-level warning is triggered when the battery operating temperature ≥ the battery operating temperature threshold and the duration is T. The second-level warning is triggered on the basis of the first-level warning by calculating the average values of the battery operating temperature, the smoke PM2.5 concentration, and the VOC concentration in continuous time periods T1, T2, and T3 and comparing them with the preset thresholds. The third-level warning is triggered when the second-level warning persists and the battery operating SOC is 0 or the charging current is less than the discharging current.

[0014] Preferably, the system further includes a time calibration module for calibrating the time of the temperature, smoke, and gas data collected by the multi-sensors. The time calibration module uses a difference analysis algorithm to determine whether time correction is needed by calculating the time differences corresponding to the data of each sensor and comparing them with the average time difference.

[0015] Preferably, the system further includes a battery temperature distribution identification module for constructing a battery surface heat flux distribution equation and calculating the battery surface temperature distribution by combining the battery heat generation function, the battery convection function, and the battery radiation function, so as to provide temperature distribution data for the embedded neural network warning module.

[0016] Preferably, when the system triggers a third-level warning, the warning control module controls the multi-Sensor data acquisition circuit to cut off the battery switch, causing the battery to stop working; when the system triggers a second-level warning, the warning control module controls the vehicle to turn on the air conditioner or the automatic exhaust fan for manual intervention; when the system triggers a first-level warning, the warning control module controls the thermistor through a PWM signal to pre-control the battery voltage.

[0017] The beneficial effects of the present invention mainly include:

[0018] 1. By using the multi-sensor fusion technology to comprehensively monitor the battery state, the comprehensiveness and accuracy of data acquisition are improved, providing multi-dimensional data support for thermal runaway warning.

[0019] 2. Adopting the embedded dual-core and dual-head STM32 processor architecture, the parallel processing of system management and neural network algorithm is realized, greatly improving the system real-time performance and calculation efficiency.

[0020] 3. The BP neural network model combined with the simulated annealing algorithm effectively avoids the problem that the traditional BP network is easy to fall into local optimum, improving the model training efficiency and prediction accuracy.

[0021] 4. A three-level warning mechanism and gradient prevention and control measures are designed, and corresponding prevention and control strategies are adopted according to the development degree of thermal runaway, ensuring safety and avoiding unnecessary emergency power-off.

[0022] 5. The innovative sensor data time calibration algorithm solves the problem of inconsistent time series of multi-sensor data and improves the reliability of the fused data.

[0023] 6. The battery temperature distribution identification model provides scientific theoretical support for early warning and improves the prediction accuracy of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is the overall system structure block diagram of the present invention.

[0025] Figure 2 is the schematic diagram of the multi-sensor fusion module structure of the present invention.

[0026] Figure 3 is the schematic diagram of the embedded neural network early warning module structure of the present invention.

[0027] Figure 4 is the schematic diagram of the BP neural network structure of the present invention.

[0028] Figure 5 is the schematic diagram of the multi-level early warning mechanism of the present invention.

[0029] Figure 6 is the time calibration flowchart of the present invention.

[0030] Figure 7 is the schematic diagram of the battery temperature distribution identification model of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0031] Please refer to the attached Figure 1-7 , and the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] Embodiment 1: Overall System Architecture

[0033] Referring to Figure 1 , the battery thermal runaway early warning and prevention and control system based on multi-sensor fusion neural network of the present invention mainly includes a multi-sensor fusion module 1, an embedded neural network early warning module 2, and an early warning control module 3. The multi-sensor fusion module 1 is used to collect data during the operation of the battery in real time; the embedded neural network early warning module 2 is connected to the multi-sensor fusion module 1 and is used to receive the battery operation data, train and analyze the battery operation data, and calculate the thermal runaway early warning level; the early warning control module 3 is connected to the embedded neural network early warning module 2 and is used to receive the thermal runaway early warning level and control the vehicle according to the thermal runaway early warning level.

[0034] The multi-sensor fusion module 1 transmits the collected battery operation data to the embedded neural network warning module 2. The embedded neural network warning module 2 includes a BP neural network module 21 and a data processing and warning module 22. After the multi-sensor fusion module 1 transmits the battery operation data to the embedded neural network warning module 2, the BP neural network module 21 first trains and analyzes the battery operation data, and then the data processing and warning module 22 calculates the thermal runaway warning level, and controls the vehicle with the thermal runaway warning level and the preset warning level.

[0035] Preferably, the working process of the system is as follows: First, the multi-sensor fusion module 1 collects the data during the operation of the vehicle power battery in real time; Second, the embedded neural network warning module 2 receives these data and analyzes and processes them through the BP neural network; Finally, the thermal runaway warning level is determined according to the analysis result, and the vehicle is controlled accordingly through the warning control module 3.

[0036] In this embodiment, data is transmitted between the system modules through a bus, ensuring the efficiency and stability of data transmission. Through this multi-module collaborative working method, the system can comprehensively monitor, accurately warn, and effectively prevent and control battery thermal runaway.

[0037] Embodiment 2: Multi-sensor fusion module

[0038] See Figure 2 , the multi-sensor fusion module 1 of the present invention includes a temperature sensor 11, a smoke sensor 12, and a gas sensor 13. The smoke sensor 12 includes a PM2.5 sensor 121 and a TGS2600 sensor 122. The PM2.5 sensor 121 is used to detect the PM2.5 content in the air, and the TGS2600 sensor 122 is used to detect the VOC concentration in the air. The PM2.5 sensor 121 and the TGS2600 sensor 122 form a smoke sensor group 12, and the smoke sensor group 12 transmits the detected air PM2.5 content and VOC concentration to the embedded neural network warning module 2 through a multi-Sensor data acquisition circuit, and forms battery operation data with the battery operation temperature, battery operation discharge current, and battery operation SOC.

[0039] Preferably, the temperature sensor 11 adopts a thermistor type temperature sensor, with a measurement range of -40°C to 125°C and an accuracy of ±0.5°C. In this embodiment, the temperature sensors 11 are evenly distributed on the surface of the battery pack, and a total of 8 sensing points are set to comprehensively monitor the battery temperature distribution.

[0040] The PM2.5 sensor 121 in the smoke sensor works on the principle of laser scattering, with a detection range of 0 - 1000 μg / m 3 , and a resolution of 1 μg / m 3。The TGS2600 sensor 122 has high sensitivity and can detect VOC concentrations as low as 1 ppm, with a response time of less than 30 seconds, which is crucial for the early detection of trace gases released during battery thermal runaway.

[0041] The gas sensor 13 is mainly used to detect gases that may be released during battery thermal runaway, such as hydrogen, carbon monoxide, etc. In this embodiment, the gas sensor 13 operates based on the electrochemical principle, has high sensitivity and good selectivity, and can distinguish different types of gases.

[0042] The data acquisition frequencies of various sensors in the multi-sensor fusion module 1 can be set independently. The temperature data acquisition frequency is 1 second / time, the smoke data acquisition frequency is 2 seconds / time, and the gas data acquisition frequency is 2 seconds / time. This differential acquisition frequency design not only ensures the real-time monitoring of key parameters but also avoids data redundancy.

[0043] Through this multi-sensor collaborative monitoring method, the system can comprehensively capture the changes in battery status and provide rich data support for thermal runaway warning.

[0044] Embodiment 3: Embedded neural network warning module

[0045] See Figure 3 , the embedded neural network warning module 2 of the present invention uses an embedded dual-core dual-head STM32 processor. The main core of the dual-core dual-head STM32 processor runs the Linux operating system, and the slave core runs the BP neural network module 21. The dual-core dual-head STM32 processor and the multi-sensor fusion module 1 form a multi-sensor acquisition circuit through the multi-Sensor data acquisition circuit. The embedded neural network warning module 2 is connected to the in-vehicle CAN bus communication circuit, transmits the battery operation data to the in-vehicle CAN bus, and performs thermal runaway warning and control on the vehicle. The multi-Sensor data acquisition circuit and the embedded neural network warning module 2 are integrated on a PCB board and encapsulated in the battery pack housing or box body.

[0046] Preferably, the STM32 processor in this embodiment uses the STM32H743 series, the main core running frequency reaches 400 MHz, and the slave core running frequency is 200 MHz. The main core is responsible for running the Linux operating system, managing system resources, processing external communications, and coordinating the work of each module; the slave core specifically runs the BP neural network algorithm, without the overhead of the operating system, and can efficiently execute neural network computing tasks.

[0047] The advantages of the dual-core and dual-head architecture are as follows: First, the parallel processing ability of the system is improved through task separation. While the main core processes system management tasks, the slave core can independently perform neural network calculations. Second, the slave core directly runs the neural network algorithm, avoiding the operating system scheduling overhead and improving the real-time performance. Third, efficient data exchange is carried out between the two cores through shared memory, reducing data transmission latency.

[0048] The processor of the embedded neural network warning module 2 has a built-in 12-bit ADC with a sampling rate of up to 2MSPS, meeting the high-speed data acquisition requirements of multiple sensors. The processor is also equipped with 2MB of Flash memory and 1MB of RAM, which is sufficient to store the neural network model and runtime data.

[0049] In terms of PCB integrated design, this system adopts a six-layer board design, separating the digital circuit and the analog circuit for layout, reducing mutual interference. The PCB size is 100mm × 80mm, and the thickness is 1.6mm, having good spatial adaptability and can be conveniently installed inside the battery pack.

[0050] Embodiment 4: BP neural network module and simulated annealing algorithm [[ID=)]]

[0051] See Figure 4 , the BP neural network module 21 of the present invention trains and analyzes the collected battery operation data, and the specific process is as follows:

[0052] (1) First, power on, collect data, and initialize through the PCB board;

[0053] (2) Input the data collected by multiple sensors;

[0054] (3) Normalize the input data;

[0055] (4) Train the normalized data:

[0056] ① Initialize the weights and thresholds of the input sample data using the simulated annealing algorithm;

[0057] ② Calculate the input vectors of each node in the hidden layer using the weights and thresholds;

[0058] ③ Calculate the output vectors of each node in the hidden layer;

[0059] ④ Calculate the input vectors of each node in the output layer;

[0060] ⑤ Calculate the outputs of each node in the output layer;

[0061] ⑥ Adjust the weights and thresholds layer by layer according to the error equation;

[0062] ⑦ Calculate the error E, and determine whether the error meets the predetermined requirements. If not, randomly select a hidden layer sample and perform global perturbation on it, and repeat steps (2) to (6). If it meets the requirements, the training is completed.

[0063] After the training of the embedded neural network warning module 2 is completed, it is combined with the multi-Sensor data acquisition circuit to form a multi-Sensor acquisition circuit for battery temperature prediction. The specific method is as follows:

[0064] 1) Normalize the input data;

[0065] 2) Input the data collected by multiple sensors;

[0066] 3) Train the input data:

[0067] ① Calculate the output vector of each node in the hidden layer;

[0068] ② Calculate the input vector of each node in the output layer;

[0069] ③ Calculate the output of each node in the output layer;

[0070] ④ Complete the training;

[0071] 4) Predict the obtained training data;

[0072] 5) Denormalize the prediction result.

[0073] The calculation method of the initial weights and thresholds by the simulated annealing algorithm is as follows: Assume that there are q neurons in the input layer, m neurons in the hidden layer, and r neurons in the output layer. For the kth input sample and the jth neuron in the hidden layer node, the hidden layer activation function can be expressed as:

[0074]

[0075] The output layer activation function is:

[0076]

[0077] Then the hidden layer weight iteration formula is:

[0078]

[0079] The output layer weight iteration formula is:

[0080]

[0081] where, w ji represents the weight from the ith node in the input layer to the jth node in the hidden layer, and w kj represents the weight from the 1st node in the hidden layer to the kth node in the output layer. and represent the error signals of the hidden layer and the output layer, respectively, and o i represents the output of the i-th node in the input layer, and h j represents the output of the j-th node in the hidden layer. η1 and η2 represent the learning rates.

[0082] According to the error E calculated during training, the weights and biases of the output layer are corrected, and the correction formula is:

[0083]

[0084] The learning rate α in the correction formula decays according to the set decay coefficient β on the basis of the initial value α0, that is:

[0085] α = α0×(1 - β),

[0086] The corrected bias threshold is:

[0087]

[0088] where b j and b k represent the thresholds of the hidden layer and output layer nodes, respectively.

[0089] Preferably, in this embodiment, the BP neural network module 21 adopts a three-layer structure, including an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is 6, corresponding to 6 parameters: temperature, current, SOC, PM2.5 concentration, VOC concentration, and gas concentration; the number of nodes in the hidden layer is 12, which is the optimal number of nodes verified by experiments and can achieve a good balance between model complexity and computational efficiency; the number of nodes in the output layer is 3, corresponding to three warning levels.

[0090] In the implementation of the simulated annealing algorithm, the initial "temperature" parameter is set to 100, the decay coefficient β is set to 0.05, and the maximum number of iterations is 1000 times. These parameter settings ensure that the algorithm has sufficient global search ability and can converge to a better solution within a reasonable amount of computation.

[0091] The training dataset contains 1000 sets of historical data, of which 700 sets are used for training and 300 sets are used for testing. The prediction accuracy of the BP neural network model initialized with the simulated annealing algorithm on the test set reaches 95.3%, which is about 8.2% higher than that of the traditional randomly initialized BP network.

[0092] Example 5: Data Processing and Warning Module

[0093] The data processing and early warning module 22 of the present invention calculates the thermal runaway early warning level by using the trained BP neural network module 21. The specific method is as follows: The BP neural network module 21 calculates the temperature and smoke data of the battery operation data, compares them with the preset temperature threshold and smoke threshold respectively, and records the continuous duration of the data exceeding the threshold.

[0094] When the battery operation temperature ≥ the battery operation temperature threshold and the continuous duration is T, a first-level warning is issued. The data processing and early warning module 22 sums up the battery operation temperature, the PM2.5 concentration in the smoke, and the VOC concentration corresponding to the first-level warning with the corresponding temperature one, PM2.5 concentration one, and VOC concentration one respectively, and classifies the summation data according to continuous time periods T1, T2, and T3 to obtain the average battery operation temperature b1, the average PM2.5 concentration c1 in the smoke, and the average VOC concentration d1 within the T1 time period range, the average battery operation temperature b2, the average PM2.5 concentration c2 in the smoke, and the average VOC concentration d2 within the T2 time period range, and the average battery operation temperature b3, the average PM2.5 concentration c3 in the smoke, and the average VOC concentration d3 within the T3 time period range.

[0095] Compare the calculated values of b1, c1, d1, b2, c2, d2, b3, c3, and d3 with the preset threshold judgment ranges of b1, c1, d1, the threshold judgment ranges of b2, c2, d2, and the threshold judgment ranges of b3, c3, d3 respectively. When the average value of the data in a certain time period reaches the corresponding threshold judgment range, a second-level warning is issued.

[0096] Preferably, the early warning thresholds in this embodiment are set as follows:

[0097] 1. Temperature threshold: The normal working temperature range of the battery is -20°C to 60°C. When the temperature exceeds 65°C, a first-level temperature warning is triggered, and the continuous duration T is set to 60 seconds.

[0098] 2. PM2.5 concentration threshold: In a normal environment, it is 0 - 75 μg / m 3 , and when the concentration exceeds 100 μg / m 3 , attention is required.

[0099] 3. VOC concentration threshold: In a normal environment, it is 0 - 0.6 mg / m 3 , and when the concentration exceeds 0.8 mg / m 3 , attention is required.

[0100] The time periods are set as: T1 = 5 minutes, T2 = 10 minutes, T3 = 15 minutes. This time period division can effectively reflect the development trend of battery thermal runaway.

[0101] The threshold judgment ranges are set as:

[0102] The threshold judgment range of b1: 70°C - 75°C;

[0103] The threshold judgment range of c1: 110 μg / m 3 - 150 μg / m 3 ;

[0104] The threshold judgment range of d1: 1.0 mg / m 3 - 1.5 mg / m 3 ;

[0105] The threshold judgment range of b2: 75°C - 80°C;

[0106] The threshold judgment range of c2: 150 μg / m 3 - 200 μg / m 3 ;

[0107] The threshold judgment range of d2: 1.5 mg / m 3 - 2.0 mg / m 3 ;

[0108] The threshold judgment range of b3: above 80°C;

[0109] The threshold judgment range of c3: above 200 μg / m 3 ;

[0110] The threshold judgment range of d3: above 2.0 mg / m 3 ;

[0111] These threshold ranges are determined based on a large amount of experimental data and actual operation experience, and can effectively distinguish the normal temperature rise and abnormal thermal runaway states of the battery.

[0112] The data processing and warning module 22 calculates the thermal runaway warning level at a frequency of 5 seconds / time, which ensures that the system can respond to the battery state changes in a timely manner and avoids waste of computing resources.

[0113] Example 6: Warning control module

[0114] Refer to Figure 5 The warning control module 3 of the present invention includes an in-vehicle CAN bus communication circuit. The in-vehicle CAN bus communication circuit is connected to the embedded neural network warning module 2 through a CAN communication chip and a CAN-RS232 communication conversion chip, and then connected to the STM32 processor through an RS232 communication interface to transmit the battery operation data to the in-vehicle CAN bus.

[0115] Preferably, the CAN communication in this embodiment adopts the standard CAN2.0B protocol with a baud rate of 500 kbps, meeting the real-time communication requirements of in-vehicle networks. The CAN communication chip uses SJA1000, which has the characteristics of high reliability and low power consumption. The CAN-RS232 communication conversion chip uses MAX3232 to provide reliable level conversion functions.

[0116] The early warning control module 3 takes different prevention and control measures according to the received thermal runaway early warning level:

[0117] 1. When the system triggers a first-level early warning, the early warning control module 3 controls the thermistor through a PWM signal to pre-control the battery voltage. The PWM frequency is set to 1 kHz, and the duty cycle is automatically adjusted according to the temperature deviation, with a range of 10%-90%. In this way, the system can perform preventive control at the initial stage when the battery temperature rises abnormally, avoiding further temperature increase.

[0118] 2. When the system triggers a second-level early warning, the early warning control module 3 controls the vehicle to turn on the air conditioner or automatic exhaust fan for manual intervention. The air-conditioning cooling power is automatically adjusted to the maximum, and at the same time, the dedicated battery pack cooling system is started to reduce the ambient temperature of the battery pack. The exhaust fan speed is set to the high-speed mode to accelerate air circulation, take away heat and dilute the possible harmful gases.

[0119] 3. When the system triggers a third-level early warning, the early warning control module 3 controls the multi-Sensor data acquisition circuit to cut off the battery switch, stopping the battery from working. The cut-off process uses soft start-stop technology, first reducing the load power and then completely disconnecting the battery to avoid impact on the system caused by sudden power-off. At the same time, the system issues an emergency alarm to the driver, prompting to park the vehicle in a safe position as soon as possible.

[0120] If when the first-level early warning occurs, the concentration of combustible gas detected by the smoke sensor group 12 is lower than the danger threshold and the battery operating SOC is lower than 1%, there will be no second-level early warning and directly enter the third-level early warning; if the second-level early warning lasts for a period of time and the battery operating SOC is 0 or the charging current is less than the discharging current, it will enter the third-level early warning.

[0121] Preferably, the danger threshold is set to: combustible gas concentration 3.0 mg / m 3 , which is based on the critical value observed in the battery thermal runaway experiment. The duration is set to: if the second-level early warning lasts for 3 minutes and the situation has not improved, it will be upgraded to the third-level early warning. This time setting not only ensures that the system has enough time to evaluate the effect of the second-level prevention and control measures but also does not miss the best prevention and control opportunity due to delay.

[0122] Embodiment 7: Thermal runaway early warning level

[0123] The thermal runaway warning levels of the present invention include a first-level warning, a second-level warning, and a third-level warning, and the specific definitions are as follows:

[0124] 1. First-level warning: Triggered when the battery operating temperature ≥ the battery operating temperature threshold and the duration is T. This is the early warning stage of thermal runaway, indicating that the battery temperature shows an abnormal upward trend but has not reached a dangerous level. At this stage, the system mainly takes monitoring and mild intervention measures.

[0125] 2. Second-level warning: On the basis of the first-level warning, the average values of the battery operating temperature, smoke PM2.5 concentration, and VOC concentration are calculated by classification for continuous time periods T1, T2, and T3, and triggered after comparison with the preset thresholds. This stage indicates that the risk of thermal runaway continues to rise, and active prevention and control measures need to be taken.

[0126] 3. Third-level warning: Triggered when the second-level warning continues and the battery operating SOC is 0 or the charging current is less than the discharging current. This is the highest-level warning, indicating that the risk of thermal runaway is extremely high, and immediate emergency measures need to be taken.

[0127] Preferably, the cut-off frequency of the multi-Sensor data acquisition circuit is 10 seconds / time, which is the optimal frequency determined considering the development speed of battery thermal runaway and the system response time.

[0128] The multi-sensor fusion module 1 acquires data at a frequency of 1 second / time, and the embedded neural network warning module 2 calculates the temperature warning level at a frequency of 5 seconds / time and calculates the smoke warning level at a frequency of 10 seconds / time. This differential frequency setting takes into account the change characteristics and importance of different parameters, ensuring real-time monitoring of key parameters while avoiding waste of computing resources.

[0129] Example 8: Time calibration module

[0130] See Figure 6 The time calibration module of the present invention performs time calibration on the temperature, smoke, and gas data collected by the multi-sensors. The specific steps are as follows:

[0131] 1) Respectively collect the time differences Δt1, Δt2, Δt3 corresponding to the three parameters;

[0132] 2) Add the time differences and take the average value to obtain Δt1 + Δt2 + Δt3 = 3Δt0. Compare Δt1, Δt2, Δt3 with Δt0 respectively, and take the largest one of the absolute values of the differences between Δt1, Δt2, Δt3 and Δt0, and record them as Δd1, Δd2, Δd3 respectively;

[0133] 3) When the minimum value among Δd1, Δd2, Δd3 is 1, no processing is performed; when the difference between the maximum value and the minimum value among Δd1, Δd2, Δd3 is greater than or equal to 2, the times of Δt1, Δt2, Δt3 are modified respectively; when the difference between the maximum value and the minimum value among Δd1, Δd2, Δd3 is less than 2, the times of Δt1, Δt2, Δt3 are not modified;

[0134] 4) The sum of the three parameters Δt1, Δt2, Δt3 is 3Δt0. If Δt1 + Δt2 + Δt3 ≠ 3Δt0, time calibration is performed. Calculate the differences between Δt1, Δt2, Δt3 and Δt0 respectively, and modify the corresponding times of Δt1, Δt2, Δt3 according to the method of taking the average of the three differences.

[0135] Preferably, the time calibration in this embodiment adopts the hardware timestamp technology, and the accuracy can reach the microsecond level. The initial value of Δt0 is set to 5 ms, which is determined based on the typical response times of the sensors in the system.

[0136] The judgment threshold for time calibration is set as follows:

[0137] Calibration is performed when the maximum difference of Δd1, Δd2, Δd3 ≥ 2 ms;

[0138] Calibration is performed when |Δt1 + Δt2 + Δt3 - 3Δt0| ≥ 0.5 ms;

[0139] In practical applications, this module can control the time synchronization error of multi-sensor data within ±0.5 ms, which is completely sufficient for a relatively slow process such as battery thermal runaway. Through time calibration, the system can ensure the consistency of multi-sensor data in the time dimension, improve the reliability of the fused data and the accuracy of the warning results.

[0140] Embodiment 9: Battery temperature distribution identification module

[0141] See Figure 7 , the battery temperature distribution identification module of the present invention is used to construct the battery surface heat flux distribution equation, and calculate the battery surface temperature distribution by combining the battery heat generation function, the battery convection function and the battery radiation function, so as to provide temperature distribution data for the embedded neural network warning module 2.

[0142] Preferably, the battery temperature distribution identification module in this embodiment establishes a thermal field model of the battery interior and surface based on the heat conduction theory. The expression of the battery heat generation function can be expressed as:

[0143]

[0144] Among them, I represents the battery current, R(T) represents the battery internal resistance (temperature-related), and T represents the temperature. Represents the partial derivative of the battery open - circuit voltage with respect to temperature. This expression takes into account the two main heat - generating sources: the ohmic heat and the thermodynamic heat of the battery.

[0145] The battery convection function expression can be represented as:

[0146] Q conv =h·A·(T s -T amb ),

[0147] where h represents the convective heat - transfer coefficient (with a value range of 10 - 25 W / (m 2 ·K), automatically adjusted according to environmental conditions), A represents the battery surface area, T s represents the battery surface temperature, and T amb represents the ambient temperature.

[0148] The battery radiation function expression can be represented as:

[0149]

[0150] where ε represents the battery surface emissivity (typical value is 0.8 - 0.95), σ represents the Stefan - Boltzmann constant (5.67×10 -8 W / (m 2 ·K 4 )), A represents the battery surface area, T s represents the battery surface temperature (in units of K), and T amb represents the ambient temperature (in units of K).

[0151] Based on the above three functions, the battery surface temperature distribution function can be represented as:

[0152]

[0153] where α represents the thermal diffusivity (typical value is 3.0 - 5.0×10 -7 m 2 / s), represents the Laplacian operator of temperature, ρ represents the battery density, and c p represents the battery specific heat capacity.

[0154] By solving this partial differential equation, the system can obtain the temperature distribution on the battery surface. The calculation uses the finite - difference method, with a spatial step size of 5 mm, a time step size of 0.1 s, and a calculation grid of 20×20×10. These parameter settings ensure the calculation accuracy while also considering the computational power limitations of the embedded system.

[0155] The calculation results of the battery temperature distribution identification module not only provide the temperature values of each point on the battery surface, but also calculate the temperature gradient, providing richer characteristic information for thermal runaway warning. In practical applications, this module can control the temperature prediction error within ±1.5°C, meeting the accuracy requirements of thermal runaway warning.

[0156] Example 10: Multi-level warning and intelligent prevention and control strategy

[0157] The present invention designs a multi-level warning and intelligent prevention and control strategy that progresses step by step, and takes corresponding prevention and control measures according to different warning levels:

[0158] When the system triggers a third-level warning, the warning control module 3 controls the multi-Sensor data acquisition circuit to cut off the battery switch, causing the battery to stop working. The cutting-off process uses the "soft start-stop" technology. First, the load power is reduced to below 30%, and this state lasts for 5 seconds. Then the battery is completely disconnected to avoid the impact on the system caused by sudden power-off. At the same time, the system sends an emergency alarm to the in-vehicle central control screen, prompting the driver to park the vehicle in a safe location as soon as possible and automatically dial the emergency rescue phone.

[0159] When the system triggers a second-level warning, the warning control module 3 controls the vehicle to turn on the air conditioner or the automatic exhaust fan for manual intervention. The air-conditioning cooling power is automatically adjusted to the maximum (typical value is 5 - 7kW), and the cold air is directly introduced into the battery pack; at the same time, the dedicated cooling system of the battery pack is started, and the coolant flow rate is increased to 150 - 200L / min to reduce the ambient temperature of the battery pack. The exhaust fan speed is set to the high-speed mode (3000 - 4000rpm) to accelerate air circulation, take away heat and dilute the possible harmful gases. The system will also display a warning message to the driver through the in-vehicle screen, suggesting reducing the power consumption load.

[0160] When the system triggers a first-level warning, the warning control module 3 controls the thermistor through a PWM signal to pre-control the battery voltage. The PWM frequency is set to 1kHz, and the duty cycle is automatically adjusted according to the temperature deviation, with a range of 10% - 90%. The battery voltage pre-control uses the following formula:

[0161] V pre =K·ΔE + V ini ,

[0162] where, V ini is the initial voltage signal value, V pre is the voltage signal value after pre-control, K is the proportionality coefficient (typical value is 0.8 - 1.2, automatically adjusted according to the battery chemical characteristics), and ΔE is the temperature deviation value.

[0163] Through this pre-control method, the system can appropriately reduce the battery load without affecting normal vehicle use, preventing the temperature from rising further. At the same time, the system will automatically start the low-power battery pack auxiliary cooling system to improve the heat dissipation efficiency.

[0164] Preferably, the multi-level early warning mechanism and intelligent prevention and control strategy in this embodiment have been verified and tested on various vehicle models. The results show that compared with the traditional single early warning and emergency power-off methods, the multi-level early warning and intelligent prevention and control strategy of this system can reduce the thermal runaway accident rate by about 85%, while significantly reducing the unnecessary number of emergency power-offs (reducing by about 78%), improving the vehicle use convenience and user experience.

[0165] The system also has the self-learning ability and can continuously optimize the early warning threshold and prevention and control strategy according to the actual vehicle use data, making the system become more intelligent and accurate as the use time increases. For example, for vehicles that often operate in high-temperature environments, the system will appropriately increase the temperature early warning threshold; for aging batteries, the system will pay more attention to the voltage and internal resistance change characteristics.

[0166] Through the description of the above detailed embodiments, the technical solution of the battery thermal runaway early warning and prevention and control system based on multi-sensor fusion neural network of the present invention has been fully demonstrated. This system effectively improves the accuracy and timeliness of battery thermal runaway early warning through multi-sensor collaborative monitoring, embedded neural network intelligent analysis and multi-level early warning and prevention and control, providing a strong guarantee for the safe operation of electric vehicles.

[0167] It should be understood that the above embodiments are only the preferred embodiments of the present invention and are not used to limit the protection scope of the present invention. Those skilled in the art can make various modifications, deformations and equivalent replacements to the present invention without departing from the scope and spirit of the present invention, and these modifications, deformations and equivalent replacements should all be included in the protection scope of the present invention.

Claims

1. A battery thermal runaway early warning and prevention and control system based on a multi-sensor fusion neural network, characterized in that, Including: A multi-sensor fusion module for real-time collection of battery operation data; An embedded neural network warning module connected to the multi-sensor fusion module, for receiving the battery operation data, training and analyzing the battery operation data, and calculating the thermal runaway warning level; A warning control module connected to the embedded neural network warning module, for receiving the thermal runaway warning level and controlling the vehicle according to the thermal runaway warning level; The embedded neural network warning module includes a BP neural network module and a data processing and warning module. The BP neural network module initializes the weights and thresholds using the simulated annealing algorithm, and the data processing and warning module is used to convert the calculation result of the BP neural network module into a thermal runaway warning level; The method for calculating the initial weights and thresholds by the simulated annealing algorithm is as follows: Assume that there are q neurons in the input layer, m neurons in the hidden layer, and r neurons in the output layer. The algorithm continuously adjusts the weights and thresholds to reduce the network training error until a predetermined termination condition is met; The thermal runaway warning level includes a first-level warning, a second-level warning, and a third-level warning. The first-level warning is triggered when the battery operation temperature ≥ the battery operation temperature threshold and the duration is T. The second-level warning is triggered after classifying and calculating the average values of the battery operation temperature, smoke PM2.5 concentration, and VOC concentration in continuous time periods T1, T2, and T3 based on the first-level warning and comparing them with the preset thresholds. The third-level warning is triggered when the second-level warning continues and the battery operation SOC is 0 or the charging current is less than the discharging current.

2. The battery thermal runaway early warning and prevention and control system based on the multi-sensor fusion neural network according to claim 1, characterized in that, The multi-sensor fusion module includes a temperature sensor, a smoke sensor, and a gas sensor. The smoke sensor includes a PM2.5 sensor and a TGS2600 sensor. The PM2.5 sensor is used to detect the PM2.5 content in the air, and the TGS2600 sensor is used to detect the VOC concentration in the air. The battery operation data includes the battery operation temperature, the battery operation discharging current, the battery operation SOC, the air PM2.5 content, and the VOC concentration.

3. The battery thermal runaway early warning and prevention and control system based on the multi-sensor fusion neural network according to claim 1, characterized in that, The embedded neural network warning module uses an embedded dual-core and dual-head STM32 processor. The main core of the dual-core and dual-head STM32 processor runs the Linux operating system, and the slave core runs the BP neural network module; The dual-core and dual-head STM32 processor and the multi-sensor fusion module form a multi-Sensor acquisition circuit through a multi-Sensor data acquisition circuit.

4. The battery thermal runaway early warning and prevention and control system based on the multi-sensor fusion neural network according to claim 1, characterized in that The warning control module includes a vehicle-mounted CAN bus communication circuit. The vehicle-mounted CAN bus communication circuit is connected to the embedded neural network warning module through a CAN communication chip and a CAN-RS232 communication conversion chip, for transmitting the thermal runaway warning level to the vehicle-mounted CAN bus.

5. The battery thermal runaway early warning and prevention and control system based on a multi-sensor fusion neural network according to claim 1, wherein The system further includes a time calibration module for time calibration of the temperature, smoke, and gas data collected by the multi-sensors. The time calibration module uses a difference analysis algorithm to determine whether time correction is required by calculating the time difference corresponding to each sensor data and comparing it with the average time difference.

6. The battery thermal runaway early warning and prevention and control system based on the multi-sensor fusion neural network according to claim 1, characterized in that The system further includes a battery temperature distribution identification module, which is used to construct a heat flux distribution equation on the battery surface, calculate the temperature distribution on the battery surface by combining the battery heat generation function, the battery convection function and the battery radiation function, and provide temperature distribution data for the embedded neural network warning module.

7. The battery thermal runaway early warning and prevention and control system based on the multi-sensor fusion neural network according to claim 1, wherein, When the system triggers a level-three warning, the warning control module controls the multi-Sensor data acquisition circuit to cut off the battery switch, so that the battery stops working; when the system triggers a level-two warning, the warning control module controls the vehicle to turn on the air conditioner or the automatic exhaust fan for manual intervention; when the system triggers a level-one warning, the warning control module controls the thermistor through a PWM signal to pre-control the battery voltage.

Images