A power battery fire warning system based on high-precision temperature sensor

Abstract

The present invention discloses a power battery fire warning system based on a high-precision temperature sensor, which relates to the field of battery fire warning, including a high-precision temperature sensor, a real-time monitoring module, a video recognition and monitoring module, a voice alarm module and a wireless transmission module. The real-time monitoring model is realized by I 2 The C / SPI interface is connected to a high-precision temperature sensor to collect and analyze battery temperature data in real time; the video recognition and monitoring module uses ResNet and diffusion models for video recognition, captures the real scene of the battery and identifies the battery status; when the battery temperature reaches the preset ignition point alarm range, the system automatically issues a voice alarm to prompt relevant personnel to take emergency measures; the alarm information and monitoring data are wirelessly transmitted to mobile phones and monitoring platforms for remote monitoring and management. This system not only optimizes battery performance and life, but also provides a comprehensive, efficient and reliable solution for the safe management of power batteries through its innovative integrated technology.

Description

Technical Field

[0001] The present invention relates to the field of power battery fire warning, and in particular to a power battery fire warning system based on a high-precision temperature sensor. Background Art

[0002] During transportation or long-term storage, power batteries may catch fire due to excessive temperature or other factors. Traditional monitoring methods have the following main problems:

[0003] Untimely warning: Existing fire warning systems usually rely on smoke detectors or temperature detectors, but these devices have a slow response time and often only sound the alarm when the fire has begun to spread, which cannot provide enough time for emergency response.

[0004] Low accuracy: Traditional temperature monitoring equipment has low accuracy and cannot accurately measure temperature changes on the battery surface, which makes it impossible for the system to issue early warning when the temperature gradually rises but has not yet reached a dangerous level.

[0005] Real-time monitoring is difficult: Existing technologies make it difficult to achieve real-time and accurate monitoring of battery status, especially during large-scale battery stacking or transportation. Due to the large number of equipment and wide monitoring range, it is difficult to achieve comprehensive real-time monitoring.

[0006] Alarm triggering lag: Due to the limited accuracy and response time of monitoring equipment, when the temperature approaches the battery ignition point, traditional systems often fail to issue an alarm in time, resulting in an inability to effectively prevent fires.

[0007] These problems lead to great safety risks in the actual application of the existing power battery monitoring system. With the widespread application of power batteries in electric vehicles, energy storage systems and other fields, there is an urgent need for a system that can monitor the battery temperature in real time and accurately and give an alarm in time before the fire hazard occurs. Summary of the invention

[0008] To solve the above problems, the present application provides a power battery fire warning system based on a high-precision temperature sensor, with the goal of improving the reliability and accuracy of the system, aiming to be able to efficiently handle potential safety risks and bring innovative technological advances to the field of battery safety management.

[0009] In this application, a power battery fire warning system based on a high-precision temperature sensor is provided, comprising:

[0010] High-precision temperature sensor;

[0011] Real-time monitoring module, the real-time monitoring model is implemented by I 2 The C / SPI interface is connected to a high-precision temperature sensor to collect and analyze battery temperature data in real time;

[0012] A video recognition and monitoring module, which uses ResNet and diffusion models to perform video recognition, capture the real scene of the battery and identify the battery status;

[0013] Voice alarm module: when the battery temperature reaches the preset ignition point alarm range, the system automatically issues a voice alarm to prompt relevant personnel to take emergency measures;

[0014] The wireless transmission module transmits alarm information and monitoring data wirelessly to mobile phones and monitoring platforms for remote monitoring and management.

[0015] Preferably, the preparation method of the high-precision temperature sensor is:

[0016] Screening of temperature sensing materials;

[0017] The integrated circuit part of the temperature sensor is designed using CMOS technology, and the integrated circuit part includes a temperature sensing element and a signal processing circuit;

[0018] A wafer is obtained by manufacturing a sensor circuit on a silicon wafer through a semiconductor manufacturing process;

[0019] Cutting the manufactured wafer into a number of individual chips;

[0020] Using miniaturized LFCSP packaging, a single chip is packaged into an easy-to-use sensor module;

[0021] Packaged sensors are tested and calibrated to ensure accuracy over the specified temperature range.

[0022] Preferably, the high-precision sensor uses I 2 C (Inter-Integrated Circuit) peripheral device interface and communication protocol, data transmission with peripheral devices;

[0023] The high-precision sensor is installed on the battery surface in a button manner to ensure the accuracy of temperature monitoring.

[0024] Preferably, the diffusion model adopted by the video recognition and monitoring module includes a data processing layer, a ResNet feature extraction layer, an inverse denoising layer, a video recognition and state classification layer, and a monitoring and feedback layer;

[0025] The diffusion model uses a ResNet denoising model, including an Encoder part and a Decoder part;

[0026] The ResNet model in the Encoder part compresses the image size, and the Decoder part restores the image size.

[0027] Among them, the monitoring and feedback layer includes a real-time monitoring system, which compares and evaluates the output of the model with the actual measurement value, and adjusts the model parameters based on the feedback to continuously optimize the model parameters.

[0028] Preferably, the data processing layer includes a data collection unit and a data preprocessing unit;

[0029] The data collection unit is the input layer, and is used to collect temperature data of lithium batteries in different states, record the charging and discharging current changes of lithium batteries under various working conditions, collect voltage changes of batteries at different time points, and record the temperature and humidity external conditions in the transportation environment;

[0030] Convert the data collected by the collection unit into a series of multi-dimensional images as the input layer;

[0031] The data preprocessing unit is the preprocessing layer. The data preprocessing unit normalizes all data within a unified range for standardization, applies a filtering algorithm to remove high-frequency noise in the data, and identifies and annotates abnormal situations in the data through data analysis and expert annotation.

[0032] Preferably, the ResNet feature extraction layer, i.e., the forward diffusion process, includes the following contents:

[0033] Noise is gradually added to the standardized battery state data to form a variety of noise state data sets ranging from low noise to high noise;

[0034] Noise is added to the standardized battery state data at multiple stages to simulate the gradual degradation process of the battery state and generate data with multiple noise levels.

[0035] Preferably, the ResNet feature extraction layer is constructed by stacking residual modules of the ResNet architecture, wherein the residual module includes:

[0036] Token embedding module: The token embedding module integrates additional data or metadata into the processing flow;

[0037] Position encoding module: The position encoding module is used to add context about the location of features in the image;

[0038] Temporal Cross-Attention Module: The temporal cross-attention module focuses on the changes between consecutive frames to predict and refine the next frame in the sequence based on the previous frames;

[0039] Spatial downsampling / upsampling module: The spatial downsampling / upsampling module adjusts the resolution of the image during processing.

[0040] Preferably, the reverse denoising layer, i.e., the diffusion layer, includes the following contents:

[0041] The diffusion model architecture based on deep learning ResNet is used as the core model of the reverse denoising layer.

[0042] By comparing the original data and the noisy data, the core model is trained to gradually recover from the high-noise state to the low-noise state and restore the original data;

[0043] Set the loss function for the training process;

[0044] The expression of the loss function is:

[0045]

[0046] In the expression, at inference time, the denoiser learned is iteratively applied, ∈ θ (x t ,t) is the noise predicted by the model, from the normally distributed variable x t Start reconstructing x0∶=x.

[0047] Preferably, the video recognition and state classification layer, i.e., the output layer, includes the following contents:

[0048] Generate new battery status data from the noisy data using the trained diffusion model;

[0049] The actual collected battery data is input into the diffusion model, and its denoising and feature extraction capabilities are used to identify and classify the battery status in real time;

[0050] By identifying the results, the health status of the battery can be monitored in real time to prevent potential safety risks.

[0051] Preferably, the specific processing process of the diffusion model includes the following steps:

[0052] The data point x0 extracted from the basic data distribution x0~q(x) is converted into a prior distribution;

[0053] Convert to a standard Gaussian distribution within T steps Where I is the identity matrix;

[0054] Given a variance schedule

[0055] The sampling expression at any time step t is:

[0056]

[0057] Among the samples Iterate with active step size t, where and The reverse process q(x t |x t-1 ) by the neural network p θ (x t |x t-1 ) and parameterized θ to determine;

[0058] The reverse operation generates new samples x*~q(x);

[0059] The new sample expression at any time step t is:

[0060]

[0061] where μ θ is the predicted mean,

[0062] When the condition meets the preset range of x0, by comparing the original data and the noise data, the diffusion model is trained to learn to remove noise from the noise data and restore the original data.

[0063] In summary, compared with traditional technologies, the power battery fire warning system based on a high-precision temperature sensor of the present invention involves a diffusion model. The diffusion model has broad application prospects in lithium battery status identification. Through data collection, preprocessing, model training and real-time monitoring, it can effectively improve the accuracy and reliability of battery status monitoring, and provide strong protection for preventing safety accidents such as fires. The diffusion model can efficiently extract key information from the battery status and improve recognition accuracy by learning the complex characteristics of the data. It has strong recovery capabilities for noise data and can maintain high recognition performance in a noisy environment. In addition, by identifying the battery status in real time, it can promptly detect and handle abnormal situations to ensure the safety of the battery during transportation.

[0064] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of a power battery fire warning system based on a high-precision temperature sensor of the present invention;

[0066] Figure 2 This is a schematic diagram of the structure of a digital temperature sensor ADT7420 chip of a power battery fire warning system based on a high-precision temperature sensor of the present invention;

[0067] Figure 3 This is a diffusion model architecture diagram of a power battery fire warning system based on a high-precision temperature sensor in the present invention. DETAILED DESCRIPTION

[0068] The technical solution of the present invention is further described below by means of the accompanying drawings and embodiments. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​described in these embodiments do not limit the scope of the present application.

[0069] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present application, its application, or uses.

[0070] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0071] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0072] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0073] like Figure 1 As shown, the present invention provides a power battery fire warning system based on a high-precision temperature sensor, showing the overall architecture of the system of the present invention, including the connection relationship between the temperature sensor, the alarm, the wireless transmission system, the video monitoring platform, and the data acquisition and analysis system.

[0074] High-precision temperature sensor, the high-precision temperature sensor mentioned in the present invention adopts the ADT7420 digital temperature sensor of Analog Devices. The sensor model is ADT7420, with an accuracy of ±0.2°C (within the range of -10°C to 85°C), and in a wider temperature range, the accuracy meets the requirement of ±0.5°C. The detection range of this system is: -40°C to 150°C.

[0075] Real-time monitoring module, the real-time monitoring model is implemented by I 2 The C / SPI interface is connected to a high-precision temperature sensor to collect and analyze battery temperature data in real time;

[0076] Video recognition and monitoring module, the video recognition and monitoring module uses ResNet and diffusion model for video recognition, captures the real scene of the battery and identifies the battery status. Through the diffusion model for video recognition, the system can capture the real scene of the battery in real time, analyze the battery status, and identify fire precursors (such as smoke, abnormal temperature rise, etc.). If the temperature rise is close to the set threshold of 5°C, the alarm is activated and the battery cooling process is started.

[0077] The diffusion model uses deep learning algorithms to process video data, which can restore clear images from noisy videos and improve recognition accuracy.

[0078] The system combines temperature data with video recognition results to make a comprehensive judgment and improve the reliability of the early warning.

[0079] The data processing module can integrate and analyze data from different sources to identify potential fire hazards.

[0080] Voice alarm module, when the battery temperature reaches the preset ignition point alarm range (5 degrees before the ignition point), the system automatically issues a voice alarm to prompt relevant personnel to take emergency measures. When the battery temperature is close to the ignition point (5 degrees before the ignition point), the system can promptly trigger the alarm to issue a voice alarm to remind relevant personnel to take emergency measures, such as cooling the battery.

[0081] The alarm information includes the current temperature, the cause of the alarm and the recommended response measures, ensuring that on-site personnel can respond quickly to avoid fire.

[0082] The wireless transmission module transmits the alarm information and monitoring data to the mobile phone and monitoring platform wirelessly, which is convenient for remote monitoring and management. The system transmits the monitoring data and alarm information to the mobile phone or remote monitoring platform through the wireless transmission module, which is convenient for real-time monitoring and management.

[0083] Supports multiple wireless communication protocols, such as Wi-Fi, Bluetooth, etc., to ensure the stability and real-time performance of data transmission, making it convenient for remote managers to obtain monitoring information in a timely manner.

[0084] Furthermore, the preparation method of the ADT7420 high-precision temperature sensor is as follows:

[0085] Screen the temperature sensing materials and select high-precision temperature sensing materials, such as thermal resistor materials, and prepare them in combination with semiconductor technology.

[0086] The integrated circuit part of the temperature sensor is designed using CMOS technology, and the integrated circuit part includes a temperature sensing element and a signal processing circuit;

[0087] A wafer is obtained by manufacturing a sensor circuit on a silicon wafer through a semiconductor manufacturing process;

[0088] Cutting the manufactured wafer into a number of individual chips;

[0089] Using miniaturized LFCSP packaging, a single chip is packaged into an easy-to-use sensor module;

[0090] Packaged sensors are tested and calibrated to ensure accuracy over the specified temperature range.

[0091] Preferably, the high-precision sensor uses I 2 C (Inter-Integrated Circuit) peripheral device interface and communication protocol, data transmission with peripheral devices.

[0092] The chip circuit schematic diagram of the ADT7420 digital temperature sensor is as follows Figure 2 As shown, the following is a description of the specific interface and communication protocol:

[0093] (1) Interface pins:

[0094] SCL (serial clock line): I 2 C bus clock signal, generated and controlled by the master device.

[0095] SDA (Serial Data Line): I used for bidirectional data transmission 2 C-bus data line.

[0096] GND (ground): power ground.

[0097] VDD (power supply): sensor power input, typical operating voltage is 2.7V to 5.5V.

[0098] (2)I 2 C communication protocol:

[0099] Address assignment: ADT7420 has a 7-bit fixed address, and different I 2 C address to allow multiple sensors to be on the same I 2 C bus.

[0100] Start and stop conditions: I 2 C communication is initiated by a start condition (S) and terminated by a stop condition (P). The start condition is when SDA changes from high to low during the SCL high level period, and the stop condition is when SDA changes from low to high during the SCL high level period.

[0101] Data transmission: 8 bits of data are transmitted each time, and the receiver sends an acknowledgment bit (ACK) after each byte is transmitted.

[0102] Reading data: After the master device sends a read command, the ADT7420 returns 16-bit temperature data, with the high byte and low byte respectively contained in two consecutive 8-bit transmissions.

[0103] (3) Register Operation

[0104] Configuration register: By writing to the configuration register, the user can set the sensor's operating mode, resolution and other parameters.

[0105] Temperature data register: The user obtains the current temperature measurement value by reading the temperature data register.

[0106] (4) Sample code

[0107] The following is the use of I 2 Example code for reading ADT7420 temperature data using C protocol (based on Arduino platform):

[0108]

[0109]

[0110] like Figure 2 As shown, the schematic diagram of the temperature sensor chip and its connection schematic diagram with the monitoring system, the ADT7420 digital temperature sensor circuit schematic description:

[0111] (1) ADT7420 sensor:

[0112] VDD: Connect to the power supply (e.g. 3.3V or 5V).

[0113] GND: Connect to ground.

[0114] SCL(Serial Clock Line): Connect to the corresponding SCL pin of the microcontroller.

[0115] SDA(Serial Data Line): Connect to the corresponding SDA pin of the microcontroller.

[0116] ADD0: This pin can be connected to VDD or GND to set the I 2 C address.

[0117] (2) Microcontroller:

[0118] VDD: Connect to the power supply (same voltage as the sensor).

[0119] GND: Connect to ground.

[0120] SCL: Connect to the SCL pin of ADT7420.

[0121] SDA: Connect to the SDA pin of ADT7420.

[0122] (3) Pull-up resistor:

[0123] SCL Line: 4.7k ohm resistor connected between SCL and VDD.

[0124] SDA Line: 4.7k ohm resistor connected between SDA and VDD.

[0125] (4) Decoupling capacitor:

[0126] A 0.1uF capacitor is connected between VDD and GND near the ADT7420 sensor to filter out noise.

[0127] (5) Connection method:

[0128] VDD(ADT7420) to VDD(microcontroller): Connect the sensor’s power pin to the microcontroller’s power pin.

[0129] GND(ADT7420) to GND(microcontroller): Connect the ground pin of the sensor to the ground pin of the microcontroller.

[0130] SCL (ADT7420) to SCL (microcontroller): Connect the serial clock line of the sensor to the serial clock line of the microcontroller.

[0131] SDA (ADT7420) to SDA (microcontroller): Connect the serial data line of the sensor to the serial data line of the microcontroller.

[0132] (6) Other components:

[0133] Pull-up resistors: Connect a 4.7k ohm resistor between the SCL line and VDD, and another 4.7k ohm resistor between the SDA line and VDD.

[0134] Decoupling capacitor: Place a 0.1uF capacitor between VDD and GND near the ADT7420 sensor.

[0135] ADD0 pin: This pin can be connected to VDD or GND to configure the I 2 C address.

[0136] High-precision sensors are installed on the battery surface in a button manner to ensure accurate temperature monitoring.

[0137] The button-style temperature sensor is designed for easy installation and replacement and is suitable for various battery types and installation environments.

[0138] The system has a simple structure and is easy to maintain. The sensors, alarms and wireless modules are highly integrated, which reduces maintenance costs and difficulties.

[0139] Furthermore, the diffusion model adopted by the video recognition and monitoring module includes a data processing layer, a ResNet feature extraction layer, an inverse denoising layer, a video recognition and state classification layer, and a monitoring and feedback layer;

[0140] like Figure 3As shown, the specific process of video recognition using the diffusion model is demonstrated, including data collection and preprocessing, forward diffusion process, reverse denoising process, and detailed steps of video recognition and state classification. The diffusion model describes an advanced artificial intelligence (AI) network model processing method for monitoring and analyzing temperature distribution in battery systems, where each step is designed to improve the accuracy and reliability of the detected temperature field.

[0141] The diffusion model uses a ResNet denoising model, including an Encoder part and a Decoder part;

[0142] The ResNet model in the Encoder part compresses the image size, and the Decoder part restores the image size.

[0143] ResNet is used as the core model for predicting noise in the diffusion model, which is generally divided into two parts: Encoder and Decoder. In the Encoder part, the ResNet model will gradually compress the size of the image; in the Decoder part, it will gradually restore the size of the image. At the same time, "residual connection" is used between the Encoder and Deocder to ensure that the Decoder part does not lose the information of the previous steps when reasoning and restoring the image information. Conv operations are performed on the compression processing of images between different layers. The model layer operations used, such as convolutional layers and tensor operations, will be analyzed in the forward propagation process monitoring of the model by writing corresponding analysis codes to analyze the model complexity. The results of using the diffusion model for recognition and classification tasks can be displayed in real time in the training information visualization module.

[0144] Furthermore, the data processing layer includes a data collection unit and a data preprocessing unit, which are used to collect temperature data, video data and other relevant information of the power battery under different states, and perform standardization and denoising processing.

[0145] The data collection unit is the input layer, and is used to collect temperature data of lithium batteries in different states, record the charging and discharging current changes of lithium batteries under various working conditions, collect voltage changes of batteries at different time points, and record the temperature and humidity external conditions in the transportation environment;

[0146] Convert the data collected by the collection unit into a series of multi-dimensional images as the input layer;

[0147] Taking the collected data as input layer, the system receives a series of multi-dimensional images representing the temperature distribution of the battery at different time points. These time series data provide the model with rich dynamic temperature information, which is the basis for accurate prediction.

[0148] The data preprocessing unit is the preprocessing layer. The data preprocessing unit normalizes all data within a unified range to eliminate the impact between different dimensions. Filtering algorithms (such as Kalman filtering, low-pass filtering, etc.) are applied to remove high-frequency noise in the data, and abnormal conditions in the data, such as overheating, overcharging, over-discharging, etc., are identified and labeled through data analysis and expert annotation.

[0149] This step corresponds to the preprocessing layer in the model, which uses advanced noise processing techniques to perform preliminary processing on the input data. This step is critical because it prepares the data to meet the requirements of the deep learning model and ensures the quality and usability of the data.

[0150] Furthermore, the ResNet feature extraction layer, i.e., the forward diffusion process, is a process of gradually adding noise to the preprocessed data, simulating the gradual degradation process of the data, and generating data sets with various noise levels, including the following:

[0151] (1) Noise addition: Noise is gradually added to the standardized battery state data to form a variety of noise state data sets ranging from low noise to high noise;

[0152] (2) Multi-stage generation: Noise is added to the standardized battery status data at multiple stages to simulate the gradual degradation process of the battery status and generate data with multiple noise levels.

[0153] Further, the ResNet feature extraction layer uses the stacking of multiple residual modules of the ResNet architecture to construct a deep feature extraction network, where the residual module includes:

[0154] Token embedding module: The token embedding module integrates additional data or metadata into the processing flow;

[0155] Position encoding module: The position encoding module is used to add context about the location of features in the image;

[0156] Temporal Cross-Attention Module: The temporal cross-attention module focuses on the changes between consecutive frames to predict and refine the next frame in the sequence based on the previous frames;

[0157] Spatial downsampling / upsampling module: The spatial downsampling / upsampling module adjusts the resolution of the image during processing.

[0158] Each residual module contains several convolutional layers, which can not only capture subtle changes in temperature data, but also prevent information loss during training through residual connections, effectively learn complex patterns and features in temperature data, and enhance the learning depth and effect of the model. The module includes batch normalization and ReLU activation functions, which are key technologies for optimizing network performance and stability. Spatial convolution and cross attention: These processes may involve extracting and focusing on specific features in the image frame, enhancing relevant details while suppressing noise.

[0159] Furthermore, the reverse denoising layer, i.e., the diffusion layer, is the process of training the diffusion model to gradually recover from a high-noise state to a low-noise state and finally restore the original data, which includes the following:

[0160] (1) Model architecture: The diffusion model architecture based on deep learning ResNet is used as the core model of the reverse denoising layer.

[0161] (2) Denoising training: By comparing the original data and the noisy data, the core model is trained to gradually recover from a high-noise state to a low-noise state and restore the original data;

[0162] Set a loss function for the training process. During the training process, a loss function (such as mean square error loss) is used to measure the difference between the restored data and the original data.

[0163] The expression of the loss function is:

[0164]

[0165] In the expression, at inference time, the denoiser learned is iteratively applied, ∈ θ (x t ,t) is the noise predicted by the model, from the normally distributed variable x t Start reconstructing x0∶=x.

[0166] The function of this layer is to remove noise and restore the temperature field. The denoising diffusion process is used to simulate the dynamic changes of temperature distribution. This process removes image noise through repeated iterative reverse inference, gradually clarifies and refines the image of temperature distribution, with the aim of clearly restoring the true distribution of the temperature field and effectively restoring the accurate temperature field from noisy data. This denoising diffusion process is the core of the model, which can effectively extract useful information from messy data.

[0167] Furthermore, the video recognition and status classification layer, i.e., the output layer, uses the trained diffusion model to recognize and classify the real-time video data and accurately analyze the battery status, including the following:

[0168] (1) New data generation: Using the trained diffusion model, new battery status data is generated from the noise data to simulate various possible battery operating conditions.

[0169] (2) Status identification and classification: The actual collected battery data is input into the diffusion model, and its denoising and feature extraction capabilities are used to identify and classify the battery status in real time. The identification results can be used to monitor the health status of the battery in real time and prevent potential safety risks.

[0170] The processed temperature field image is output, which clearly shows the temperature distribution of the battery. This layer not only outputs visual results, but also classifies the battery according to the set temperature threshold to determine whether the battery status is dangerous or safe.

[0171] Monitoring and feedback layer: The model includes a real-time monitoring system that compares the model's output with actual measurements. This allows users to evaluate the model's performance and further adjust the model parameters based on feedback, continuously optimizing the model parameters and improving the model's accuracy and response speed.

[0172] Furthermore, the specific processing process of the diffusion model includes the following steps:

[0173] The diffusion model is used to train the reverse random process, which transforms the data point x0 (e.g., a battery temperature field image) extracted from the basic data distribution x0~q(x) into a prior distribution;

[0174] Usually converted to a standard Gaussian distribution within T steps Where I is the identity matrix, which can be formally understood as a stationary Markov chain with Gaussian transitions;

[0175] It is given by the variance timetable Given parameter values;

[0176] The sampling expression at any time step t is:

[0177]

[0178] Among the samples Iterate with active step size t, where and We assume that the inverse process q(x t |x t-1 ) by the neural network p θ (x t |x t-1 ) and parameterized θ to determine;

[0179] In order to generate new samples, run the reverse operation to generate new samples x*~q(x), and run the reverse Markov chain to achieve the goal;

[0180] The new sample expression at any time step t is:

[0181]

[0182] where μ θ is the mean of the forecasts, and we set the covariance to be purely time-dependent:

[0183] This model is usually trained by maximizing the variational lower bound of the log-likelihood. When the conditions meet the preset range of x0, by comparing the original data and the noisy data, the diffusion model is trained to learn to remove noise from the noisy data and restore the original data.

[0184] The diffusion model mentioned in this invention integrates deep learning and physical models, combining the deep learning capabilities of ResNet with the physical diffusion process, providing a novel perspective for temperature monitoring. Through dynamic simulation and iterative reverse inference processes, noise can be effectively removed while maintaining data integrity. The system can automatically adjust according to real-time data feedback to improve the accuracy of prediction and response speed.

[0185] In summary, this system based on ResNet and diffusion model not only improves the accuracy of battery temperature monitoring, but also achieves rapid response and accurate prediction of potential dangerous conditions by deeply learning and analyzing the dynamic changes of temperature data. In addition, the design of the entire system takes into account both efficiency and practicality, which is very suitable for deployment in practical applications and provides strong technical support for battery safety management.

[0186] In summary, the power battery fire warning system of the present invention has significant advantages in high-precision temperature monitoring, real-time warning, video recognition, data fusion, wireless transmission and system robustness. It can effectively prevent the occurrence of fire and ensure the safe use of batteries, and has broad application prospects.

[0187] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A power battery fire warning system based on a high-precision temperature sensor, characterized in that: include: High-precision temperature sensor; A real-time monitoring module is provided. 2 The C / SPI interface is connected to a high-precision temperature sensor to collect and analyze battery temperature data in real time; A video recognition and monitoring module, which uses ResNet and diffusion models to perform video recognition, capture the real scene of the battery and identify the battery status; The diffusion model adopted by the video recognition and monitoring module includes a data processing layer, a ResNet feature extraction layer, an inverse denoising layer, a video recognition and state classification layer, and a monitoring and feedback layer; Voice alarm module: when the battery temperature reaches the preset ignition point alarm range, the system automatically issues a voice alarm to prompt relevant personnel to take emergency measures; Wireless transmission module, which transmits alarm information and monitoring data to mobile phones and monitoring platforms wirelessly, facilitating remote monitoring and management; The data processing layer includes a data collection unit and a data preprocessing unit; The data collection unit is the input layer, which is used to collect temperature data of lithium batteries in different states, record the charging and discharging current changes of lithium batteries under various working conditions, collect voltage changes of batteries at different time points, and record temperature and humidity external conditions in the transportation environment; Convert the data collected by the collection unit into a series of multi-dimensional images as the input layer; The data preprocessing unit is the preprocessing layer. The data preprocessing unit normalizes all data to a unified range for standardization, applies a filtering algorithm to remove high-frequency noise in the data, and identifies and annotates abnormal situations in the data through data analysis and expert annotation; The inverse denoising layer, i.e. the diffusion layer, includes the following: The diffusion model architecture based on deep learning ResNet is used as the core model of the reverse denoising layer. By comparing the original data and the noisy data, the core model is trained to gradually recover from the high-noise state to the low-noise state and restore the original data; Set the loss function for the training process; The expression of the loss function is: ； In the expression, at inference time, the learned denoiser is iteratively applied, is the noise predicted by the model, from the normally distributed variable Start refactoring , indicating that based on the parameters The neural network model at time step For the current noise data Predicted noise; is the random noise added to the original data during the forward diffusion process; To represent the original, unnoised input data, It is the temperature, current and voltage information of lithium batteries under different working conditions collected by the data collection unit; In the forward diffusion process, at the time step The noisy data under Add a certain degree of noise to get; is the time step or level in the diffusion process; Represents the overall loss value of the model, which is used to measure the gap between the noise predicted by the model during the inverse denoising process and the actual noise added; The video recognition process of the diffusion model includes the following steps: From the basic data distribution The data points extracted from Convert to prior distribution; Convert to a standard Gaussian distribution within T steps ,in is the identity matrix; Given a variance schedule ; The sampling expression at any time step t is: ； ； Among them, X t is the data sample of the current step in the diffusion process, X t-1 represents the data sample of the previous step in the diffusion process, α t For the time step The proportion of original data information retained, β t is the time step The noise variance scheduling parameter is in the range of (0,1) and controls the intensity of the noise added in this step; The sample x t = Iterate with a step size of t, where and , the reverse process Neural Networks ; Run in reverse to generate new samples ; The new sample expression at any time step t is: ； ； in is the predicted mean, ; When the conditions are met When the range is preset, by comparing the original data and the noise data, the diffusion model is trained to learn to remove noise from the noise data and restore the original data.

2. A power battery fire warning system based on a high-precision temperature sensor according to claim 1, characterized in that: The preparation method of the high-precision temperature sensor is as follows: Screening of temperature sensing materials; The integrated circuit part of the temperature sensor is designed using CMOS technology, and the integrated circuit part includes a temperature sensing element and a signal processing circuit; A wafer is obtained by manufacturing a sensor circuit on a silicon wafer through a semiconductor manufacturing process; Cutting the manufactured wafer into a number of individual chips; Using miniaturized LFCSP packaging, a single chip is packaged into an easy-to-use sensor module; Packaged sensors are tested and calibrated to ensure accuracy over the specified temperature range.

3. A power battery fire warning system based on a high-precision temperature sensor according to claim 2, characterized in that: The high-precision temperature sensor uses an I²C (Inter-Integrated Circuit) peripheral device interface and communication protocol to transmit data with the peripheral device; The high-precision temperature sensor is installed on the battery surface in a button manner to ensure the accuracy of temperature monitoring.

4. A power battery fire warning system based on a high-precision temperature sensor according to claim 1, characterized in that: The diffusion model uses a ResNet denoising model, including an Encoder part and a Decoder part; Among them, the ResNet model compresses the size of the image in the Encoder part and restores the size of the image in the Decoder part; Among them, the data processing layer and the ResNet feature extraction layer are forward diffusion processes, the reverse denoising layer is a reverse diffusion process, and the video recognition and state classification layer and the monitoring and feedback layer are output processes and feedback processes; Among them, the monitoring and feedback layer includes a real-time monitoring system, which compares and evaluates the output of the model with the actual measurement value, and adjusts the model parameters based on feedback to continuously optimize the model parameters.

5. A power battery fire warning system based on a high-precision temperature sensor according to claim 4, characterized in that: The ResNet feature extraction layer, i.e., the forward diffusion process, includes the following: Noise is gradually added to the standardized battery state data to form a variety of noise state data sets ranging from low noise to high noise; Noise is added to the standardized battery state data at multiple stages to simulate the gradual degradation process of the battery state and generate data with multiple noise levels.

6. A power battery fire warning system based on a high-precision temperature sensor according to claim 4, characterized in that: The ResNet feature extraction layer is constructed by stacking residual modules of the ResNet architecture, wherein the residual modules include: Token embedding module: The token embedding module integrates additional data or metadata into the processing flow; Position encoding module: The position encoding module is used to add context about the location of features in the image; Temporal Cross-Attention Module: The temporal cross-attention module focuses on the changes between consecutive frames to predict and refine the next frame in the sequence based on the previous frames; Spatial downsampling / upsampling module: The spatial downsampling / upsampling module adjusts the resolution of the image during processing.

7. A power battery fire warning system based on a high-precision temperature sensor according to claim 4, characterized in that: The video recognition and state classification layer, i.e., the output layer, includes the following contents: Generate new battery status data from the noisy data using the trained diffusion model; The actual collected battery data is input into the diffusion model, and its denoising and feature extraction capabilities are used to identify and classify the battery status in real time; By identifying the results, the health status of the battery can be monitored in real time to prevent potential safety risks.

Images