Thermal runaway early warning system and method

Abstract

The embodiment of the invention provides a thermal runaway early warning system and method, and the method comprises the steps: firstly obtaining the working condition data of a multi-mode vehicle-mounted battery in real time through a vehicle-end vehicle-mounted edge calculation module during the execution of the system, carrying out the thermal runaway risk assessment through combining with a preset time sequence convolutional neural network, and outputting a thermal runaway analysis request to a cloud if the existence of a thermal runaway risk is determined. Meanwhile, the pile end obtains pile end collaborative sensing data representing the external environment state of the vehicle-mounted battery in real time, when a thermal runaway analysis module of the cloud side responds to a thermal runaway analysis request, thermal runaway risk analysis is completed by combining the two types of data, and the thermal runaway probability and a prediction time window representing the remaining duration of the battery from thermal runaway are output. And finally, the thermal runaway protection module of the cloud performs strategy analysis in combination with the current charging scene, the thermal runaway probability and the prediction time window, generates an adaptive charging protection strategy and issues the adaptive charging protection strategy to the vehicle end, so that the accuracy of thermal runaway early warning is effectively improved, and the vehicle safety in the fast charging scene is ensured.

Description

Technical Field

[0001] This application relates to the field of thermal runaway early warning technology, and in particular to a thermal runaway early warning system and method. Background Technology

[0002] With the development of fast charging technology for new energy vehicles, the charging speed has increased significantly, but so has the risk of battery thermal runaway. Current battery thermal runaway warning schemes for fast charging in new energy vehicles rely solely on sensor data such as voltage, temperature, and surface temperature, using fixed thresholds or simple models to determine thermal runaway. However, for early thermal runaway issues caused by internal short circuits or lithium plating during fast charging, the current detection window is typically less than 60 seconds, resulting in significant warning lag and making false alarms highly susceptible to factors such as sensor noise and cell inconsistencies. Summary of the Invention

[0003] To address the aforementioned issues and improve the accuracy of thermal runaway warning during fast charging of batteries in new energy vehicles, this application provides a thermal runaway warning system and method.

[0004] The embodiments of this application disclose the following technical solutions: In a first aspect, embodiments of this application provide a thermal runaway early warning system, the system comprising: a vehicle-mounted terminal, a charging pile terminal, and a cloud-based terminal; the vehicle-mounted terminal includes: an on-board edge computing module; the charging pile terminal is used to acquire charging pile terminal collaborative sensing data in real time, the charging pile terminal collaborative sensing data being used to characterize the external environmental state of the on-board battery when the vehicle-mounted terminal is executing the charging process; the cloud-based terminal includes: a thermal runaway analysis module and a thermal runaway protection module; The vehicle-mounted edge computing module is used to perform thermal runaway risk assessment based on real-time acquired multimodal vehicle battery operating condition data and a preset temporal convolutional neural network, and output a thermal runaway analysis request to the cloud if thermal runaway risk is confirmed. The thermal runaway analysis module is used to respond to the thermal runaway analysis request, perform thermal runaway risk analysis based on the operating condition feature fusion vector and the pile-end collaborative sensing data, and obtain the thermal runaway probability and prediction time window; the prediction time window is used to characterize the remaining time of the vehicle battery before the triggering of thermal runaway; the operating condition feature fusion vector is obtained by fusing the multimodal vehicle battery operating condition data; The thermal runaway protection module is used to perform strategy analysis based on the current charging scenario, the thermal runaway probability, and the prediction time window to obtain a charging protection strategy, and then send the charging protection strategy to the vehicle.

[0005] In one possible implementation, the in-vehicle edge computing module is specifically used for: Based on the preset temporal convolutional neural network, a risk score is calculated according to the multimodal vehicle battery operating condition data to obtain a thermal runaway risk score. If the thermal runaway risk score is not less than the risk score threshold, it is determined that the thermal runaway risk exists. If the thermal runaway risk score is less than the risk score threshold, it is determined that there is no risk of thermal runaway.

[0006] In one possible implementation, the preset temporal convolutional neural network comprises multiple parallel TCN channels, and each TCN channel is composed of multiple stacked causal dilated convolutional blocks; the vehicle-mounted edge computing module includes a risk scoring unit, which is specifically used for: Based on the data modes corresponding to each of the TCN channels, feature map transformation is performed on the multimodal vehicle battery operating condition data to obtain multiple operating condition feature maps. Based on the preset attention weights corresponding to each of the TCN channels, feature fusion processing is performed on each of the working condition feature maps to obtain the working condition feature fusion vector. The thermal runaway risk score is determined based on the Sigmoid activation function and the fusion vector of the operating condition features.

[0007] In one possible implementation, the thermal runaway analysis module is specifically used for: Based on the Transformer encoder, feature fusion processing is performed on the working condition feature fusion vector and the pile end collaborative sensing data to obtain cross-mode fusion features; Based on a preset spatiotemporal graph neural network, thermal runaway risk is predicted according to the cross-modal fusion features, and the thermal runaway probability and the prediction time window are obtained.

[0008] In one possible implementation, the preset spatiotemporal graph neural network includes a spatiotemporal graph convolutional layer, a pooling layer, and an output layer; the thermal runaway analysis module includes a risk prediction unit, which is specifically used for: Each cell or module in the vehicle battery is taken as a graph node. The spatiotemporal graph convolution layer is used to perform spatial graph convolution processing on each graph node, and the features between each graph node and its corresponding neighboring nodes are aggregated to obtain multiple node spatial aggregated features. The initial features of the graph nodes are determined based on the cross-module fusion features. as well as, The spatiotemporal graph convolutional layer performs temporal convolution processing on each graph node, and performs convolution processing on the time series features of each graph node to obtain multiple node temporal dynamic features. Through the pooling layer, global average pooling is performed on the spatial aggregated features of multiple nodes and the temporal dynamic features of multiple nodes to obtain global pooled features for the vehicle battery. The output layer is used to perform a fully connected network mapping on the global pooling features to obtain the thermal runaway probability and the prediction time window; The node spatial aggregation feature is used to characterize the diffusion effect of the graph node when thermal runaway is triggered; the node temporal dynamic feature is used to characterize the evolution trend of the graph node's features when thermal runaway is triggered.

[0009] In one possible implementation, the vehicle end includes: an on-board battery cooling module; the current charging scenario includes: a first charging scenario, a second charging scenario, and a third charging scenario, wherein the charging power of the third charging scenario is greater than that of the second charging scenario, and the charging power of the second charging scenario is greater than that of the first charging scenario; the thermal runaway protection module is specifically used for: Based on the thermal runaway probability and the prediction time window, a hazard index is assessed to obtain the thermal runaway hazard index. When the current charging scenario is the first charging scenario, and the thermal runaway risk index is less than the second index threshold and greater than the first index threshold, the charging power of the vehicle is reduced in a stepwise manner, and the working power of the vehicle battery cooling module is adjusted to half of the rated power. If the current charging scenario is the second charging scenario, and the thermal runaway risk index is less than the second index threshold and greater than the first index threshold, the charging current at the vehicle end is reduced to the current limit, and the operating power of the vehicle cooling module is adjusted to the rated power. If the current charging scenario is the third charging scenario, and the thermal runaway risk index is less than the second index threshold and greater than the first index threshold, determine the current ambient temperature of the vehicle. When the current ambient temperature is below the temperature threshold, the vehicle battery is preheated. Once the vehicle battery has been preheated to the target temperature, the charging current at the vehicle end is reduced to the current limit, and the operating power of the vehicle cooling module is adjusted to the rated power.

[0010] In one possible implementation, the vehicle terminal further includes: a data storage module; the data storage module is specifically used for: Acquire thermal runaway data in the event of thermal runaway; The thermal runaway data is written into a preset blockchain.

[0011] Secondly, embodiments of this application provide a thermal runaway early warning method, applicable to any possible thermal runaway early warning system in the first aspect; the method includes: The vehicle edge computing module performs thermal runaway risk assessment based on real-time acquired multimodal vehicle battery operating condition data and a preset temporal convolutional neural network, and outputs a thermal runaway analysis request to the cloud if thermal runaway risk is confirmed. The thermal runaway analysis module, in response to the thermal runaway analysis request, performs thermal runaway risk analysis based on the operating condition feature fusion vector and the pile-end collaborative sensing data to obtain the thermal runaway probability and prediction time window; the prediction time window is used to characterize the remaining time of the vehicle battery before triggering thermal runaway; the operating condition feature fusion vector is obtained by fusing the multimodal vehicle battery operating condition data; The thermal runaway protection module performs strategy analysis based on the current charging scenario, the thermal runaway probability, and the prediction time window to obtain a charging protection strategy, which is then sent to the vehicle.

[0012] In one possible implementation, the assessment of thermal runaway risk based on real-time acquired multimodal vehicle battery operating condition data and a preset temporal convolutional neural network includes: Based on the preset temporal convolutional neural network, a risk score is calculated according to the multimodal vehicle battery operating condition data to obtain a thermal runaway risk score. If the thermal runaway risk score is not less than the risk score threshold, it is determined that the thermal runaway risk exists. If the thermal runaway risk score is less than the risk score threshold, it is determined that there is no risk of thermal runaway.

[0013] In one possible implementation, the preset temporal convolutional neural network includes multiple parallel TCN channels, and each TCN channel is composed of multiple stacked causal dilated convolutional blocks. The step of calculating a thermal runaway risk score based on the preset temporal convolutional neural network and the multimodal vehicle battery operating condition data includes: Based on the data modes corresponding to each of the TCN channels, feature map transformation is performed on the multimodal vehicle battery operating condition data to obtain multiple operating condition feature maps. Based on the preset attention weights corresponding to each of the TCN channels, feature fusion processing is performed on each of the working condition feature maps to obtain the working condition feature fusion vector. The thermal runaway risk score is determined based on the Sigmoid activation function and the fusion vector of the operating condition features.

[0014] Compared to existing technologies, this application offers the following advantages: This application provides a thermal runaway early warning system and method. During system execution, the vehicle-mounted edge computing module first acquires multimodal vehicle battery operating condition data in real time. It then combines this data with a preset temporal convolutional neural network to conduct thermal runaway risk assessment. This approach is not limited to the traditional method of relying solely on voltage and temperature sensor data. It effectively avoids false alarms caused by sensor noise and cell inconsistencies from both data dimension and assessment model perspectives. If a thermal runaway risk is confirmed, a thermal runaway analysis request is sent to the cloud. Simultaneously, the charging pile acquires real-time collaborative sensing data characterizing the external environmental state of the vehicle battery. This data, combined with the multimodal data from the vehicle, forms an operating condition feature fusion vector, which serves as the basis for cloud analysis. This achieves multi-source data fusion analysis, making the basis for thermal runaway risk judgment more comprehensive. The cloud-configured thermal runaway analysis module responds to the thermal runaway analysis request, combines the two types of data to complete the thermal runaway risk analysis, and outputs the thermal runaway probability and a predicted time window characterizing the remaining time before thermal runaway. This overcomes the limitation of a window period of less than sixty seconds in existing detection methods. Finally, the cloud-based thermal runaway protection module performs strategy analysis based on the current charging scenario, thermal runaway probability, and prediction time window, generates an appropriate charging protection strategy, and sends it to the vehicle. This allows thermal runaway warning and protection to form a complete system, thereby effectively improving the accuracy of thermal runaway warning and ensuring vehicle safety in fast charging scenarios. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the structure of a thermal runaway early warning system provided in an embodiment of this application; Figure 2 A schematic diagram of a process for calculating a thermal runaway risk score is provided for an embodiment of this application; Figure 3 This is a flowchart illustrating a method for predicting thermal runaway risk provided in an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and accompanying drawings. It should be particularly noted that the embodiments described in this application are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] As described earlier, with the development of fast charging technology for new energy vehicles, the charging speed has increased significantly, but so has the risk of battery thermal runaway. Current battery thermal runaway warning schemes for fast charging in new energy vehicles rely solely on sensor data such as voltage, temperature, and surface temperature, using fixed thresholds or simple models to determine thermal runaway. However, for early thermal runaway issues caused by internal short circuits or lithium plating during fast charging, the current detection window is typically less than 60 seconds, resulting in significant warning lag and a high susceptibility to false alarms due to sensor noise and cell inconsistencies.

[0020] Based on this, this application provides a thermal runaway early warning system and method. During system execution, the vehicle-side edge computing module first acquires multimodal vehicle battery operating condition data in real time, and then combines this data with a preset temporal convolutional neural network to conduct thermal runaway risk assessment. This approach is not limited to the traditional method of relying solely on voltage and temperature sensor data, but effectively avoids false alarms caused by sensor noise and cell inconsistencies from both data dimension and assessment model perspectives. If a thermal runaway risk is confirmed, a thermal runaway analysis request is sent to the cloud. Simultaneously, the charging pile acquires real-time collaborative sensing data characterizing the external environmental state of the vehicle battery. This data, combined with the multimodal data from the vehicle side to form an operating condition feature fusion vector, serves as the basis for cloud analysis, achieving multi-source data fusion analysis and providing a more comprehensive basis for judging thermal runaway risk.

[0021] Specifically, the charging pile-side collaborative sensing data focuses on external battery monitoring dimensions, including non-contact data collected by the charging pile-side infrared thermal imager and acoustic data captured by the charging pile-side acoustic microphone array. The non-contact data collected by the charging pile-side infrared thermal imager can include temperature distribution data at the charging interface contact point, overall temperature field data of the battery pack's outer surface, and data on the location of abnormal temperature hotspots and temperature difference changes on the outer surface. The acoustic data captured by the charging pile-side acoustic microphone array can include broadband sound signals emitted by the battery pack during charging, spectral characteristic data of abnormal vibrations or noises, and data on the intensity and frequency variation trends of sound signals. Simultaneously, the charging pile itself detects charging interaction and environmental data, such as contact pressure data of the charging interface, real-time environmental parameters around the charging pile including temperature, humidity, and atmospheric pressure, real-time changes in current, voltage, and power output from the charging pile to the vehicle, and temperature monitoring data of the charging cable. This data comprehensively reflects the external state of the battery during charging from multiple external dimensions, including heat, sound, environment, and charging interaction, providing crucial charging pile-side sensing data for cloud-based thermal runaway analysis.

[0022] The cloud-based thermal runaway analysis module responds to thermal runaway analysis requests, combining two types of data to perform thermal runaway risk analysis. It outputs the thermal runaway probability and a predicted time window representing the remaining time before thermal runaway, thus overcoming the limitation of less than sixty seconds in the original detection method. Finally, the cloud-based thermal runaway protection module performs strategy analysis based on the current charging scenario, thermal runaway probability, and predicted time window, generating an appropriate charging protection strategy and distributing it to the vehicle. This creates a complete system for thermal runaway warning and protection, effectively improving the accuracy of thermal runaway warnings and ensuring vehicle safety in fast-charging scenarios.

[0023] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0024] See Figure 1This figure is a schematic diagram of a thermal runaway early warning system provided in an embodiment of this application. As shown in the figure, the system in this embodiment includes a vehicle-side terminal, a charging pile terminal, and a cloud-based terminal. The vehicle-side terminal includes an onboard edge computing module 100, which is a hardware carrier for preliminary thermal risk assessment of the vehicle, used to determine whether the vehicle has a thermal runaway risk based on real-time battery data during the charging process. The charging pile terminal's function is to acquire real-time collaborative sensing data, which is used to accurately characterize the external environmental state of the vehicle battery during the charging process, supplementing the overall thermal runaway risk analysis with key collaborative sensing information from the perspective of the battery's external environment, serving as an important reference in addition to the vehicle-side data. The cloud-based terminal, as the core analysis and decision-making layer of this system, includes a thermal runaway analysis module 200 and a thermal runaway protection module 300. The thermal runaway analysis module responds to thermal runaway analysis requests from the vehicle-side terminal, conducting professional thermal runaway risk analysis based on multi-source data. The thermal runaway protection module is responsible for formulating appropriate charging protection strategies based on the analysis results and the actual charging scenario, providing direct protection execution basis for the vehicle-side terminal. Thus, the vehicle-side, pile-side, and cloud-side systems work together to form the operational framework for the thermal runaway early warning system in this embodiment.

[0025] Specifically, the vehicle-mounted edge computing module 100 is used to perform thermal runaway risk assessment based on real-time acquired multimodal vehicle battery operating condition data and a preset temporal convolutional neural network, and outputs a thermal runaway analysis request to the cloud if a thermal runaway risk is confirmed.

[0026] When the vehicle begins the fast charging process, the on-board edge computing module (i.e., edge computing unit) deployed on the vehicle acquires multimodal on-board battery operating condition data in real time during the real-time charging process. Based on this data and a preset temporal convolutional neural network, it determines whether there is a risk of thermal runaway in the on-board battery. When it is determined that there is a risk of thermal runaway in the on-board battery, it outputs a thermal runaway analysis request to the cloud to request a more in-depth thermal runaway analysis, and then obtains the action instructions that the vehicle needs to execute to deal with the risk of thermal runaway.

[0027] Among them, multimodal vehicle battery operating condition data refers to real-time battery operating status data collected from multiple physical dimensions such as electricity, heat, gas, sound, and pressure during the fast charging process of new energy vehicles, covering core battery charging and discharging parameters and fault precursor characteristic signals. For example, electrical data includes parameters such as real-time voltage of individual battery cells and modules, total charging current, remaining capacity, battery health status, and fast charging rate, which can reflect the electrical operating conditions of battery charging and discharging. Thermal data includes cell surface temperature, inter-module temperature difference, internal temperature gradient of battery pack, and temperature rise rate detected by distributed thermocouples and fiber optic grating sensors, to capture changes in battery thermal state; gas data covers the concentration values ​​and concentration surge slopes of early characteristic gases of thermal runaway such as VOCs (Volatile Organic Compounds), hydrogen, carbon monoxide, and hydrogen fluoride in the battery pack; pressure data includes micro-pressure difference inside the battery pack, pressure wave spectrum characteristics caused by gas generation, and dynamic changes in internal pressure of the cells. Multiple modalities of on-board battery operating condition data can comprehensively characterize battery operating conditions from the perspective of multiple physics fields, providing multi-dimensional underlying data support for thermal runaway risk assessment.

[0028] Specifically, the vehicle-mounted edge computing module mainly assesses the risk of thermal runaway through the following three steps: Step 1: Based on the preset temporal convolutional neural network, calculate the risk score according to the multimodal vehicle battery operating condition data to obtain the thermal runaway risk score.

[0029] Firstly, during the real-time acquisition of multimodal vehicle battery operating condition data by the vehicle edge computing module, a thermal runaway risk score needs to be calculated using a pre-set temporal convolutional neural network (TCN) to determine whether the vehicle battery is at risk of thermal runaway. In this system, the pre-set temporal convolutional neural network includes multiple parallel TCN channels, each composed of multiple stacked causal dilated convolutional blocks. It is a dedicated lightweight model designed to meet the real-time inference requirements of automotive-grade chips. The design of multiple parallel TCN channels matches the heterogeneity of multimodal data. Different channels correspond to battery operating condition data of different modes such as temperature, gas, pressure, and acoustics, enabling independent temporal feature extraction of each modality. This avoids feature interference between different modal signals and captures the time dependence and dynamic change patterns of each modality. The multiple causal dilated convolutional blocks stacked within each TCN channel are the core units for the network to extract temporal features. The causal nature ensures the logical rationality of the temporal analysis and avoids interference from future data on the feature extraction at the current moment. The stacking of multiple convolutional blocks, combined with the exponentially increasing dilation coefficient, also enables the model to efficiently capture long-distance temporal dependencies in the data, thereby deeply mining the early features of thermal runaway hidden in multimodal data.

[0030] See Figure 2 Regarding the calculation method of the thermal runaway risk score in this step, the figure is a schematic diagram of a process for calculating the thermal runaway risk score provided by an embodiment of this application. The calculation process of the thermal runaway risk score in this system is implemented by a risk scoring unit set in the system, which specifically implements the calculation of the thermal runaway risk score through the following steps: S101: Based on the data modes corresponding to each of the TCN channels, feature map transformation is performed on the multi-modal vehicle battery operating condition data to obtain multiple operating condition feature maps.

[0031] Based on the correspondence between each TCN channel and data mode, feature map transformation processing is performed on multimodal vehicle battery operating condition data to obtain multiple operating condition feature maps. Different TCN channels correspond to a data mode such as temperature, gas, pressure, and acoustics. Multimodal vehicle battery operating condition data are assigned to the corresponding TCN channels according to the mode type. Each channel, based on its own network structure designed for that mode, performs time-series feature extraction and dimensional transformation on the received raw operating condition data, transforming the original data sequence into an operating condition feature map that can reflect the core features of that mode of data.

[0032] S102: Based on the preset attention weights corresponding to each of the TCN channels, perform feature fusion processing on each of the working condition feature maps to obtain the working condition feature fusion vector.

[0033] After feature map transformation, the risk scoring unit performs feature fusion processing on all operating condition feature maps based on the preset attention weights corresponding to each TCN channel, thereby generating an operating condition feature fusion vector. The preset attention weights are set according to the degree of correlation between different data modes and the risk of thermal runaway. For modes highly correlated with early precursors of thermal runaway, such as hydrogen concentration and temperature gradient, their corresponding TCN channels are assigned higher attention weights, while modes with relatively weaker correlations receive lower weights. During the fusion process, the feature maps of each operating condition are weighted and integrated according to their corresponding weights to achieve the aggregation of multimodal features.

[0034] S103: Determine the thermal runaway risk score based on the Sigmoid activation function and the fusion vector of the operating condition features.

[0035] Finally, the final thermal runaway risk score is determined by combining the Sigmoid activation function with the fusion vector of operating condition features. The risk scoring unit inputs the integrated fusion vector of operating condition features into a fully connected computing network to complete feature calculations. The result is then normalized using the Sigmoid activation function, which precisely maps the result to a value range of 0 to 1. The value within this range is the final thermal runaway risk score; a higher value indicates a higher degree of thermal runaway risk for the vehicle battery. The 0-1 score range provides an intuitive and unified standard for judging the risk level, serving as the basis for the vehicle to determine whether a thermal runaway risk exists. When the score reaches a preset risk score threshold, the vehicle sends a thermal runaway analysis request to the cloud. The use of the Sigmoid activation function makes the score results more comparable and definitive, adapting to the business need for the vehicle to quickly complete the initial assessment of thermal runaway risk locally.

[0036] Step 2: If the thermal runaway risk score is not less than the risk score threshold, then the existence of the thermal runaway risk is confirmed. Step 3: If the thermal runaway risk score is less than the risk score threshold, it is determined that there is no thermal runaway risk.

[0037] Based on this, after calculating the thermal runaway risk score, the system determines whether the vehicle battery is at risk of thermal runaway. During the determination process, if the calculated thermal runaway risk score reaches or exceeds a preset risk score threshold, it is directly determined that the current vehicle battery is at risk of thermal runaway. This determination immediately triggers subsequent operations on the vehicle side, sending a thermal runaway analysis request to the cloud. This propels the system from the initial local judgment stage on the vehicle side to the in-depth analysis stage on the cloud, leveraging the cloud's stronger computing power and multi-source data fusion analysis capabilities to analyze the thermal runaway risk of the vehicle battery. Conversely, if the actual thermal runaway risk score is lower than the preset risk score threshold, it is determined that the battery is not at risk of thermal runaway. In this case, the vehicle side will not send an analysis request to the cloud, maintaining the normal fast charging process for new energy vehicles. Simultaneously, it will continue to collect multi-modal vehicle battery operating condition data in real time and perform risk score calculation and determination, achieving dynamic monitoring of the battery's operating condition.

[0038] The thermal runaway analysis module 200 is used to respond to the thermal runaway analysis request, perform thermal runaway risk analysis based on the operating condition feature fusion vector and the pile-end collaborative sensing data, and obtain the thermal runaway probability and prediction time window; the prediction time window is used to characterize the remaining time of the vehicle battery before the triggering of thermal runaway; the operating condition feature fusion vector is obtained by fusing the multimodal vehicle battery operating condition data.

[0039] The thermal runaway analysis module, a key component for in-depth thermal runaway analysis in the cloud, is triggered by responding to thermal runaway analysis requests from the vehicle. This module conducts thermal runaway risk analysis based on two types of multi-source data: operating condition feature fusion vectors and charging pile collaborative sensing data. The operating condition feature fusion vectors are obtained by fusing multimodal on-board battery operating condition data collected from the vehicle. These vectors incorporate multi-dimensional operating condition characteristics such as electrical, thermal, gas, acoustic, and pressure characteristics during battery operation, providing crucial data reflecting the battery's internal state. The charging pile collaborative sensing data, acquired in real-time from the charging pile, focuses on the external environmental conditions during battery charging, supplementing the risk analysis with external environmental data. The combination of these two types of data overcomes the limitations of single-data-dimensional analysis, enabling collaborative analysis of the battery's internal and external states. After completing the thermal runaway risk analysis, the module outputs the thermal runaway probability and prediction time window.

[0040] Specifically, the thermal runaway analysis module performs the thermal runaway risk analysis process mainly through the following two steps: Step 1: Based on the Transformer encoder, perform feature fusion processing on the working condition feature fusion vector and the pile end collaborative sensing data to obtain cross-mode fusion features.

[0041] The first step in the thermal runaway analysis module's thermal runaway risk analysis is to perform cross-modal feature fusion processing based on the fusion vector of the vehicle-side operating conditions and the collaborative sensing data from the pile-side, obtained by the Transformer encoder, ultimately yielding cross-modal fused features. First, the fused vector of multi-modal operating conditions features and the collaborative sensing data from the pile-side (including infrared thermal imaging and acoustic monitoring after feature extraction) are mapped to a unified feature dimension through linear projection layers. Simultaneously, learnable modal type embeddings are added to both types of data to distinguish information sources, and sinusoidal position encoding is added to the time-series features to address the issue of inconsistent dimensions in heterogeneous data. Then, the preprocessed multi-modal feature sequence is input into the Transformer encoder stack. Utilizing the multi-head self-attention mechanism of each layer of the encoder, the intrinsic correlation between the vehicle and pile-side data is captured, such as the correlation strength between vehicle-side battery temperature features and pile-side thermal imaging temperature field data, and between vehicle-side acoustic features and pile-side abnormal noise signals. Finally, a feedforward neural network performs a nonlinear transformation on the interacting features. Through iterative processing by multi-layer encoders, the features of vehicle-pile data are fully and deeply fused, ultimately outputting cross-mode fused features with rich contextual information and high feature correlation.

[0042] Step 2: Based on the preset spatiotemporal graph neural network, perform thermal runaway risk prediction according to the cross-modal fusion features to obtain the thermal runaway probability and the prediction time window.

[0043] Within this system, the pre-defined spatiotemporal graph neural network includes spatiotemporal graph convolutional layers, pooling layers, and an output layer. The process of predicting thermal runaway risk in this step requires reliance on these layers within the pre-defined spatiotemporal graph neural network. (See also...) Figure 3 This figure is a flowchart illustrating a thermal runaway risk prediction method provided in an embodiment of this application. The figure shows the specific method for performing thermal runaway risk prediction in this step, and the process is implemented through a risk prediction unit within this system, specifically including the following steps: S201: Each cell or module in the vehicle battery is taken as a graph node, and the spatiotemporal graph convolution layer is used to perform spatial graph convolution processing on each graph node to aggregate the features between each graph node and its corresponding neighboring nodes to obtain multiple node spatial aggregation features; the initial features of the graph node are determined based on the cross-modal fusion features.

[0044] In step S201, the basic construction of the graph network is first completed based on the physical structure of the vehicle battery pack. Each cell or module within the battery is independently set as a graph node, with each node corresponding to a specific battery unit within the battery pack. Simultaneously, all graph nodes are assigned initial features determined based on cross-modal fusion features. These initial features integrate information from vehicle-side multimodal battery operating condition data and pile-side collaborative sensing data, ensuring that the initial features of each node accurately reflect the actual operating state of the corresponding cell or module. After completing the graph node construction and initial feature assignment, spatial graph convolution processing is performed on each graph node using a preset spatiotemporal graph convolutional layer. Based on the actual physical connection and thermal coupling relationships between cells and modules within the battery pack, the neighboring nodes corresponding to each graph node are identified. Then, through a feature aggregation algorithm of spatial graph convolution, the initial features of each graph node are fused with the features of neighboring nodes to generate corresponding node spatial aggregation features for each graph node. The entire implementation process conforms to the actual physical layout of the vehicle battery pack and the spatial propagation characteristics of thermal runaway, ensuring that subsequent feature processing aligns with actual risk evolution patterns.

[0045] Among these features, the node spatial aggregation feature, as the core output of this step, conforms to the risk propagation law of vehicle batteries under thermal runaway. The characteristic of the node spatial aggregation feature lies in its accurate characterization of the diffusion effect of the corresponding node in the graph when thermal runaway is triggered. It reflects how the risk characteristics of a single cell or module are transmitted to adjacent cells or modules when thermal anomalies occur, filling the gap that single node features cannot reflect spatial correlation, and providing a basis for subsequent prediction of the spatial diffusion path of thermal runaway. At the same time, this feature is generated based on cross-module fusion features of multi-source data from vehicles and charging piles, inheriting the comprehensiveness of multi-source data. It includes not only the electrical, thermal, and gaseous operating characteristics of the cell itself, but also the external environmental characteristics sensed by the charging pile, allowing the aggregated feature to reflect the risk status of the node and its surroundings from an internal and external perspective.

[0046] S202: Perform temporal convolution processing on each graph node through the spatiotemporal graph convolution layer, and perform convolution processing on the time series features of each graph node to obtain multiple node temporal dynamic features.

[0047] The acquisition of node spatial aggregation features in step S202 and step S201 is executed in parallel. Simultaneously, each graph node undergoes temporal convolution processing through the spatiotemporal graph convolutional layer within the pre-defined spatiotemporal graph neural network. Each graph node possesses its own time-series features that continuously change during the charging process. These features are generated based on cross-modal fusion features that integrate multi-source data from both the vehicle and the charging pile, including dynamic changes in indicators such as voltage, temperature, and gas concentration at different time steps. The spatiotemporal graph convolutional layer performs temporal convolution processing separately for each graph node's time-series features. By performing sliding operations on the time-series data through the convolution kernel, it captures the feature correlations and change patterns of the same node at different time steps, such as the continuous upward trend of temperature, the sudden increase slope of characteristic gas concentration, and periodic fluctuations in pressure—dynamic features in the time dimension. Each graph node undergoes this process to complete the deep extraction of its own time-series features, ultimately generating corresponding node-specific dynamic time-series features for each node. The entire implementation focuses on the evolution trajectory of a single node's own features in the time dimension, allowing the model to move beyond static analysis at a single time point and achieve dynamic tracking of battery thermal runaway features.

[0048] The role of node time-series dynamic features is to characterize the evolution trend of various features of the corresponding graph node during the thermal runaway process. It reflects the entire feature change process of the cell or module gradually developing from normal charging conditions to an abnormal state, until it may trigger thermal runaway. This allows for the prediction of the future development direction of node features and provides a time dimension basis for predicting the thermal runaway time window. At the same time, this feature retains the change trajectory of various features of graph nodes at different charging time steps, which is consistent with the gradual development of thermal runaway precursors. It can effectively identify those slowly accumulating abnormal feature changes and avoid over-responding to sudden but unrelated instantaneous data.

[0049] S203: Through the pooling layer, global average pooling is performed on the spatial aggregated features of multiple nodes and the temporal dynamic features of multiple nodes to obtain global pooled features for the vehicle battery.

[0050] Furthermore, the pooling layer performs global average pooling on the previously obtained spatial aggregation features and temporal dynamic features of multiple nodes to obtain global pooled features for the vehicle battery. These two types of input features are node-level features corresponding to each cell or module within the battery pack, respectively carrying the spatial diffusion effect and temporal evolution trend of thermal runaway. They are dispersed local features and cannot be directly used for overall thermal runaway risk assessment of the battery pack. The global average pooling operation of the pooling layer performs overall average calculations on the spatial aggregation features and temporal dynamic features of all cells or modules. While retaining core risk feature information, it integrates and reduces the dimensionality of massive node-level features, eliminating interference caused by individual feature differences between different nodes. It refines the dispersed local spatiotemporal features into global features that can characterize the entire vehicle battery pack, elevating features from the local level of cells and modules to the global level of the battery pack. The resulting global pooled features are a unified feature carrier integrating risk information in both spatial and temporal dimensions, providing comprehensive feature basis for the overall risk prediction of the subsequent output layer.

[0051] S204: Through the output layer, the global pooling features are mapped using a fully connected network to obtain the thermal runaway probability and the prediction time window.

[0052] Finally, the output layer takes the globally pooled features as input and maps them through a fully connected network to obtain the thermal runaway probability and prediction time window. The fully connected network in the output layer contains multiple layers of interconnected neurons that perform multiple rounds of nonlinear transformations and precise mapping of the feature dimensions on the globally pooled features. Corresponding mapping paths are constructed for the two prediction targets: the thermal runaway probability and the prediction time window. This transforms the abstract globally pooled features into concrete, quantifiable warning results. Specifically, the thermal runaway probability is mapped to a scalar value between 0 and 1, intuitively reflecting the likelihood of the entire vehicle battery experiencing thermal runaway. The prediction time window is mapped to a specific positive value, representing the remaining time from the current state of the battery to triggering thermal runaway. This mapping process establishes a correlation between the global spatiotemporal features of the battery pack and actual warning indicators. The two quantifiable output results eliminate vague risk judgments and achieve accurate thermal runaway warnings.

[0053] The thermal runaway protection module 300 is used to perform strategy analysis based on the current charging scenario, the thermal runaway probability, and the prediction time window to obtain a charging protection strategy, and then send the charging protection strategy to the vehicle.

[0054] Finally, after determining the probability of thermal runaway and the prediction time window, the cloud-based thermal runaway protection module conducts a comprehensive risk level and intervention feasibility assessment based on the power characteristics of the charging scenario, the probability of thermal runaway, and the remaining duration of the prediction time window. Combining the protection requirements of different fast-charging scenarios and battery operating characteristics, it performs targeted strategy analysis and formulates appropriate charging protection strategies. These strategies cover the adjustment methods of charging power and current, the power adjustment of the on-board battery cooling module, and specific protective actions such as battery preheating in low-temperature environments. After the strategy is formulated, the module sends the charging protection strategy to the vehicle in real time. Upon receiving the strategy, the vehicle immediately executes the corresponding protective actions. Simultaneously, the module also continuously acquires real-time data from the vehicle and charging pile, adjusting the protection strategy according to the dynamic changes in battery conditions and thermal runaway risk to ensure the safety of the vehicle's fast charging process.

[0055] In this embodiment, the charging scenario is divided into a first charging scenario, a second charging scenario, and a third charging scenario based on the vehicle-side charging power. The charging power of the third charging scenario is greater than that of the second charging scenario, and the charging power of the second charging scenario is greater than that of the first charging scenario. In the charging protection strategy determined by the thermal runaway protection module, in addition to limiting the vehicle-side charging power and charging current, it is also necessary to simultaneously control the vehicle-side on-board battery cooling module. In this embodiment, the process of the thermal runaway protection module determining the charging protection strategy is implemented through the following six steps: Step 1: Evaluate the hazard index based on the thermal runaway probability and the prediction time window to obtain the thermal runaway hazard index.

[0056] First, Xu combined the probability of thermal runaway with the predicted time window to complete the hazard index assessment, obtaining a thermal runaway hazard index that comprehensively characterizes the degree and urgency of thermal runaway risk. The hazard index assessment uses a preset algorithm to fuse the two indicators, combining the probability level with the length of the time window. It considers both the likelihood of the risk occurring and the urgency of the thermal runaway risk triggering. The final thermal runaway hazard index is a comprehensive quantitative result of the current thermal runaway risk, which can more accurately and comprehensively reflect the actual risk status of the battery, providing a unified risk judgment basis for subsequent development of targeted protection strategies for different charging scenarios.

[0057] Step 2: When the current charging scenario is the first charging scenario, and the thermal runaway risk index is less than the second index threshold and greater than the first index threshold, the charging power of the vehicle is reduced in a stepwise manner, and the working power of the vehicle battery cooling module is adjusted to half of the rated power.

[0058] After completing the hazard index assessment, for the current charging scenario (Scenario 1) where the thermal runaway hazard index is between the first and second index thresholds (low to medium risk), the system will execute corresponding protective actions. This involves a stepped reduction in the vehicle's charging power and adjusting the onboard battery cooling module's operating power to half of its rated power. This protection strategy is well-suited to the low-power characteristics of Scenario 1, while also considering safety and charging experience under low to medium risk conditions. Scenario 1 is a low-power charging scenario, often used for home or destination fast charging. Users have higher requirements for the continuity and user experience of the charging process, making aggressive interruptions or significant power reductions unsuitable. The stepped reduction in charging power avoids sudden power drops that could impact battery cells and charging equipment, allowing the charging power to decrease smoothly. This suppresses battery temperature rise, reduces the risk of thermal runaway, and maintains charging continuity to the greatest extent possible. Adjusting the cooling module to half of its rated power is a reasonable temperature control strategy based on low to medium risk. It can suppress the continuous rise in battery temperature through moderate cooling, addressing potential thermal runaway risks from a thermal management perspective, while avoiding unnecessary energy consumption due to excessive cooling power, thus achieving a balance between safety protection and system energy consumption.

[0059] Step 3: When the current charging scenario is the second charging scenario, and the thermal runaway risk index is less than the second index threshold and greater than the first index threshold, the charging current at the vehicle end is reduced to the current limit, and the operating power of the vehicle cooling module is adjusted to the rated power.

[0060] The second charging scenario is a medium-power charging scenario, with charging power between the first low-power and third high-power scenarios. The corresponding thermal runaway risk index is between the first and second index thresholds, representing a medium-risk state requiring active intervention but not extreme measures. Therefore, the design of the protective actions balances the effectiveness of risk control with adaptability to the characteristics of medium-power charging conditions. Under the matching conditions of this scenario and risk level, the system will execute two core protective operations: First, directly reducing the vehicle-side charging current to the current limit. Compared to the stepped power reduction in the first scenario, directly limiting the current allows for more precise and rapid control of the battery's charging input power, reducing heat generation during battery charging and discharging, and preventing the continuous accumulation of heat load under medium-power charging that could lead to thermal runaway. Simultaneously, the current limit setting ensures that the charging process will not experience equipment shock due to sudden stops or significant current reductions. Second, adjusting the operating power of the on-board cooling module to its rated power maximizes the cooling system's heat dissipation efficiency, quickly removing accumulated heat from the battery and suppressing the continuous rise in battery temperature through active cooling, forming a synergistic protection with the current-limited heat generation control.

[0061] Step 4: If the current charging scenario is the third charging scenario, and the thermal runaway risk index is less than the second index threshold and greater than the first index threshold, determine the current ambient temperature of the vehicle. Step 5: If the current ambient temperature is lower than the temperature threshold, preheat the vehicle battery.

[0062] The third charging scenario primarily involves high-power fast charging for commercial vehicles. Under high-power charging, the battery heat generation rate is significantly higher than in the previous two scenarios, and low temperatures cause a sharp increase in battery internal resistance. Directly implementing current control and cooling operations could easily lead to battery performance degradation or even exacerbate internal heat loss. Therefore, this protection strategy adds an ambient temperature detection and battery preheating step before the core current control and cooling process. Specifically, step four first determines the current ambient temperature at the vehicle end. This operation serves as the basis for determining whether to initiate preheating, ensuring that protection actions are not solely matched to risk levels but are more aligned with actual environmental conditions. Step five, when the ambient temperature is detected to be below a preset temperature threshold, performs targeted preheating of the vehicle battery. Through temperature regulation, the battery is gradually warmed to a suitable charging range, reducing internal resistance and establishing a foundation for subsequent current control and cooling operations, avoiding secondary damage to the battery caused by direct intervention at low temperatures. This approach effectively balances the high-power characteristics of the third charging scenario with the battery's operating characteristics in low-temperature environments, making the protection strategy more scientific and adaptable.

[0063] Step 6: After the vehicle battery is preheated to the target temperature, the charging current at the vehicle end is reduced to the current limit, and the operating power of the vehicle cooling module is adjusted to the rated power.

[0064] Step six is ​​the protection execution phase after completing the pre-temperature adaptation. This step requires that the vehicle battery be preheated to the target temperature before formally executing the operations of reducing the vehicle-side charging current to the current limit and adjusting the operating power of the vehicle cooling module to the rated power. This sequential execution logic ensures that the medium-risk protection in the third charging scenario effectively prevents thermal runaway risk while protecting battery performance. Reducing the charging current to the limit directly reduces the heat generation rate of high-power charging, quickly curbing the accumulation of heat load inside the battery and addressing the core risk of rapid heat generation in the third charging scenario. The vehicle cooling module operates at its rated power, maximizing heat dissipation efficiency and quickly removing the heat generated by the battery during high-power charging. This, combined with the current-limiting operation, effectively mitigates the potential for thermal runaway under medium-risk conditions from two dimensions. The entire protection process first performs adaptive preheating based on the ambient temperature, and then executes the core current control and cooling actions, effectively avoiding the risk of thermal runaway while minimizing additional battery damage caused by the protection operations.

[0065] In one possible implementation, the vehicle-mounted terminal of this application embodiment is further provided with a data storage module. This data storage module is used to write the relevant data of the thermal runaway event into a preset blockchain when thermal runaway is triggered in the vehicle-mounted terminal. Specifically, the data storage module mainly realizes the writing of thermal runaway data into the preset blockchain through the following two steps: Step 1: Acquire thermal runaway data in the event of thermal runaway.

[0066] Thermal runaway data encompasses key data triggering the entire thermal runaway process. Specifically, it includes raw multimodal onboard battery operating condition data collected in real-time by the vehicle, thermal runaway risk scores and related analysis data generated by the onboard edge computing module, thermal runaway probability, prediction time windows, and charging protection strategies distributed from the cloud, as well as the execution process and results of these protection strategies. It also includes collaborative sensing data synchronized with the charging pile and external data such as ambient temperature and charging scenario at the time of thermal runaway. The data storage module automatically extracts the above data from various units on the vehicle, including the sensor array, edge computing module, and cooling module, as well as the data transmission channel between the vehicle and the charging pile / cloud at the point of thermal runaway triggering. Through preset data acquisition rules, the data is integrated and verified to ensure that the acquired thermal runaway data accurately reconstructs the entire thermal runaway process, providing a reliable data source for subsequent blockchain-based evidence storage.

[0067] Step 2: Write the thermal runaway data into a preset blockchain.

[0068] The pre-defined blockchain network in this system includes automakers, battery manufacturers, charging operators, insurance companies, and relevant regulatory agencies as nodes. When writing data, the data storage module performs hash encryption on the thermal runaway data and adds a timestamp signature according to the blockchain's notarization rules before uploading it to the blockchain network to complete distributed ledger recording. This ensures data security and accurately records the timeline of thermal runaway occurrences through timestamps.

[0069] This application provides a thermal runaway early warning system. During system execution, the vehicle-mounted edge computing module first acquires multimodal vehicle battery operating condition data in real time. This data, combined with a pre-set temporal convolutional neural network, is used to assess thermal runaway risk. This approach goes beyond the traditional reliance on voltage and temperature sensor data, effectively avoiding false alarms caused by sensor noise and cell inconsistencies from both a data dimension and assessment model perspective. If a thermal runaway risk is confirmed, a thermal runaway analysis request is sent to the cloud. Simultaneously, the charging pile acquires real-time collaborative sensing data characterizing the external environmental state of the vehicle battery. This data, combined with the multimodal data from the vehicle, forms a condition feature fusion vector, which serves as the basis for cloud analysis. This multi-source data fusion analysis provides a more comprehensive basis for judging thermal runaway risk. The cloud-configured thermal runaway analysis module responds to the thermal runaway analysis request, combining the two types of data to complete the thermal runaway risk analysis. It outputs the thermal runaway probability and a predicted time window characterizing the remaining time before thermal runaway, thus overcoming the limitation of a window period of less than sixty seconds in traditional detection methods. Finally, the cloud-based thermal runaway protection module performs strategy analysis based on the current charging scenario, thermal runaway probability, and prediction time window, generates an appropriate charging protection strategy, and sends it to the vehicle. This allows thermal runaway warning and protection to form a complete system, thereby effectively improving the accuracy of thermal runaway warning and ensuring vehicle safety in fast charging scenarios.

[0070] Furthermore, this application embodiment also provides a thermal runaway early warning method, which is applied to the thermal runaway early warning system in any of the above embodiments; the method includes: The vehicle edge computing module performs thermal runaway risk assessment based on real-time acquired multimodal vehicle battery operating condition data and a preset temporal convolutional neural network, and outputs a thermal runaway analysis request to the cloud if thermal runaway risk is confirmed. The thermal runaway analysis module, in response to the thermal runaway analysis request, performs thermal runaway risk analysis based on the operating condition feature fusion vector and the pile-end collaborative sensing data to obtain the thermal runaway probability and prediction time window; the prediction time window is used to characterize the remaining time of the vehicle battery before triggering thermal runaway; the operating condition feature fusion vector is obtained by fusing the multimodal vehicle battery operating condition data; The thermal runaway protection module performs strategy analysis based on the current charging scenario, the thermal runaway probability, and the prediction time window to obtain a charging protection strategy, which is then sent to the vehicle.

[0071] In one possible implementation, the assessment of thermal runaway risk based on real-time acquired multimodal vehicle battery operating condition data and a preset temporal convolutional neural network includes: Based on the preset temporal convolutional neural network, a risk score is calculated according to the multimodal vehicle battery operating condition data to obtain a thermal runaway risk score. If the thermal runaway risk score is not less than the risk score threshold, it is determined that the thermal runaway risk exists. If the thermal runaway risk score is less than the risk score threshold, it is determined that there is no risk of thermal runaway.

[0072] In one possible implementation, the preset temporal convolutional neural network includes multiple parallel TCN channels, and each TCN channel is composed of multiple stacked causal dilated convolutional blocks. The step of calculating a thermal runaway risk score based on the preset temporal convolutional neural network and the multimodal vehicle battery operating condition data includes: Based on the data modes corresponding to each of the TCN channels, feature map transformation is performed on the multimodal vehicle battery operating condition data to obtain multiple operating condition feature maps. Based on the preset attention weights corresponding to each of the TCN channels, feature fusion processing is performed on each of the working condition feature maps to obtain the working condition feature fusion vector. The thermal runaway risk score is determined based on the Sigmoid activation function and the fusion vector of the operating condition features.

[0073] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the system and method embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The system and method embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0074] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A thermal runaway early warning system, characterized in that, The system includes: a vehicle-side terminal, a charging pile terminal, and a cloud-based terminal; the vehicle-side terminal includes: an on-board edge computing module; the charging pile terminal is used to acquire charging pile terminal collaborative sensing data in real time, and the charging pile terminal collaborative sensing data is used to characterize the external environmental state of the on-board battery when the vehicle-side terminal is performing the charging process; the cloud-based terminal includes: a thermal runaway analysis module and a thermal runaway protection module. The vehicle-mounted edge computing module is used to perform thermal runaway risk assessment based on real-time acquired multimodal vehicle battery operating condition data and a preset temporal convolutional neural network, and output a thermal runaway analysis request to the cloud if thermal runaway risk is confirmed. The thermal runaway analysis module is used to respond to the thermal runaway analysis request, perform thermal runaway risk analysis based on the operating condition feature fusion vector and the pile-end collaborative sensing data, and obtain the thermal runaway probability and prediction time window; the prediction time window is used to characterize the remaining time of the vehicle battery before the triggering of thermal runaway; the operating condition feature fusion vector is obtained by fusing the multimodal vehicle battery operating condition data; The thermal runaway protection module is used to perform strategy analysis based on the current charging scenario, the thermal runaway probability, and the prediction time window to obtain a charging protection strategy, and then send the charging protection strategy to the vehicle.

2. The system according to claim 1, characterized in that, The vehicle-mounted edge computing module is specifically used for: Based on the preset temporal convolutional neural network, a risk score is calculated according to the multimodal vehicle battery operating condition data to obtain a thermal runaway risk score. If the thermal runaway risk score is not less than the risk score threshold, it is determined that the thermal runaway risk exists. If the thermal runaway risk score is less than the risk score threshold, it is determined that there is no risk of thermal runaway.

3. The system according to claim 2, characterized in that, The preset temporal convolutional neural network comprises multiple parallel TCN channels, and each TCN channel is composed of multiple stacked causal dilated convolutional blocks; the vehicle-mounted edge computing module includes a risk scoring unit, which is specifically used for: Based on the data modes corresponding to each of the TCN channels, feature map transformation is performed on the multimodal vehicle battery operating condition data to obtain multiple operating condition feature maps. Based on the preset attention weights corresponding to each of the TCN channels, feature fusion processing is performed on each of the working condition feature maps to obtain the working condition feature fusion vector. The thermal runaway risk score is determined based on the Sigmoid activation function and the fusion vector of the operating condition features.

4. The system according to claim 1, characterized in that, The thermal runaway analysis module is specifically used for: Based on the Transformer encoder, feature fusion processing is performed on the working condition feature fusion vector and the pile end collaborative sensing data to obtain cross-mode fusion features; Based on a preset spatiotemporal graph neural network, thermal runaway risk is predicted according to the cross-modal fusion features, and the thermal runaway probability and the prediction time window are obtained.

5. The system according to claim 4, characterized in that, The preset spatiotemporal graph neural network includes a spatiotemporal graph convolutional layer, a pooling layer, and an output layer; the thermal runaway analysis module includes a risk prediction unit, which is specifically used for: Each cell or module in the vehicle battery is taken as a graph node. The spatiotemporal graph convolution layer is used to perform spatial graph convolution processing on each graph node, and the features between each graph node and its corresponding neighboring nodes are aggregated to obtain multiple node spatial aggregate features. The initial features of the graph nodes are determined based on the cross-modal fusion features; as well as, The spatiotemporal graph convolutional layer performs temporal convolution processing on each graph node, and performs convolution processing on the time series features of each graph node to obtain multiple node temporal dynamic features. Through the pooling layer, global average pooling is performed on the spatial aggregated features of multiple nodes and the temporal dynamic features of multiple nodes to obtain global pooled features for the vehicle battery. The output layer is used to perform a fully connected network mapping on the global pooling features to obtain the thermal runaway probability and the prediction time window; The node spatial aggregation feature is used to characterize the diffusion effect of the graph nodes when thermal runaway is triggered. The node temporal dynamic features are used to characterize the evolution trend of the graph node in triggering thermal runaway.

6. The system according to claim 1, characterized in that, The vehicle-side component includes: an on-board battery cooling module; the current charging scenario includes: a first charging scenario, a second charging scenario, and a third charging scenario, wherein the charging power of the third charging scenario is greater than that of the second charging scenario, and the charging power of the second charging scenario is greater than that of the first charging scenario; the thermal runaway protection module is specifically used for: Based on the thermal runaway probability and the prediction time window, a hazard index is assessed to obtain the thermal runaway hazard index. When the current charging scenario is the first charging scenario, and the thermal runaway risk index is less than the second index threshold and greater than the first index threshold, the charging power of the vehicle is reduced in a stepwise manner, and the working power of the vehicle battery cooling module is adjusted to half of the rated power. If the current charging scenario is the second charging scenario, and the thermal runaway risk index is less than the second index threshold and greater than the first index threshold, the charging current at the vehicle end is reduced to the current limit, and the operating power of the vehicle cooling module is adjusted to the rated power. If the current charging scenario is the third charging scenario, and the thermal runaway risk index is less than the second index threshold and greater than the first index threshold, determine the current ambient temperature of the vehicle. When the current ambient temperature is below the temperature threshold, the vehicle battery is preheated. Once the vehicle battery has been preheated to the target temperature, the charging current at the vehicle end is reduced to the current limit, and the operating power of the vehicle cooling module is adjusted to the rated power.

7. The system according to claim 1, characterized in that, The vehicle terminal also includes: a data storage module; the data storage module is specifically used for: Acquire thermal runaway data in the event of thermal runaway; The thermal runaway data is written into a preset blockchain.

8. A method for early warning of thermal runaway, characterized in that, Applied to the thermal runaway early warning system as described in any one of claims 1-7; the method includes: The vehicle edge computing module performs thermal runaway risk assessment based on real-time acquired multimodal vehicle battery operating condition data and a preset temporal convolutional neural network, and outputs a thermal runaway analysis request to the cloud if thermal runaway risk is confirmed. The thermal runaway analysis module, in response to the thermal runaway analysis request, performs thermal runaway risk analysis based on the operating condition feature fusion vector and the pile-end collaborative sensing data to obtain the thermal runaway probability and prediction time window; the prediction time window is used to characterize the remaining time of the vehicle battery before triggering thermal runaway; the operating condition feature fusion vector is obtained by fusing the multimodal vehicle battery operating condition data; The thermal runaway protection module performs strategy analysis based on the current charging scenario, the thermal runaway probability, and the prediction time window to obtain a charging protection strategy, which is then sent to the vehicle.

9. The method according to claim 8, characterized in that, The thermal runaway risk assessment based on real-time acquired multimodal vehicle battery operating condition data and a preset temporal convolutional neural network includes: Based on the preset temporal convolutional neural network, a risk score is calculated according to the multimodal vehicle battery operating condition data to obtain a thermal runaway risk score. If the thermal runaway risk score is not less than the risk score threshold, it is determined that the thermal runaway risk exists. If the thermal runaway risk score is less than the risk score threshold, it is determined that there is no risk of thermal runaway.

10. The method according to claim 9, characterized in that, The preset temporal convolutional neural network includes multiple parallel TCN channels, and each TCN channel is composed of multiple stacked causal dilated convolutional blocks. The step of calculating a thermal runaway risk score based on the preset temporal convolutional neural network and the multimodal vehicle battery operating condition data includes: Based on the data modes corresponding to each of the TCN channels, feature map transformation is performed on the multimodal vehicle battery operating condition data to obtain multiple operating condition feature maps. Based on the preset attention weights corresponding to each of the TCN channels, feature fusion processing is performed on each of the working condition feature maps to obtain the working condition feature fusion vector. The thermal runaway risk score is determined based on the Sigmoid activation function and the fusion vector of the operating condition features.

Images