A battery thermal runaway management method and system

Abstract

This invention discloses a battery thermal runaway management method and system. The method is applied to a battery thermal runaway management system. Using multimodal sensors, when triggering conditions are met, multimodal physical state data of a target battery cluster in the energy storage compartment is collected. Using an edge computing device, a state feature vector is determined based on the multimodal physical state data. Based on the state feature vector and a pre-built rule base, a target transmission strategy and a first monitoring data packet are determined. The first monitoring data packet, including the state feature vector and an initial thermal runaway warning level, is uploaded to the management platform based on the target transmission strategy. The management platform determines the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet, and determines a thermal runaway management instruction based on the target thermal runaway warning level. The thermal runaway management instruction is then sent to the corresponding instruction execution object. This method improves the accuracy, timeliness, and effectiveness of thermal runaway management.

Description

Technical Field

[0001] This invention relates to the field of battery safety technology, and in particular to a battery thermal runaway management method and system. Background Technology

[0002] With the continuous growth of new energy installed capacity, electrochemical energy storage systems play a crucial role in grid frequency regulation and peak shaving. Lithium-ion batteries have become the mainstream energy storage component due to their high energy density and long cycle life, but their thermal runaway risk remains a bottleneck restricting large-scale application. Thermal runaway refers to the process in which an exothermic reaction inside the battery leads to a rapid increase in temperature, eventually causing a fire or even an explosion. Once it occurs, it often results in serious economic losses and safety accidents.

[0003] Existing battery thermal runaway early warning methods typically upload all raw data directly to the cloud or a backend platform for analysis. However, this cloud-based approach results in large data transmission volumes and high latency, leading to low decision-making efficiency and response delays, thus failing to meet the high real-time requirements for rapid thermal runaway assessment. Furthermore, existing methods often rely solely on temperature sensor data for thermal runaway monitoring; however, this approach cannot reflect early-stage microscopic changes within the battery cluster, resulting in delayed warnings and low accuracy. Summary of the Invention

[0004] This invention provides a battery thermal runaway management method and system to solve the problems of insufficient real-time performance and low accuracy of thermal runaway early warning in existing methods.

[0005] In a first aspect, embodiments of the present invention provide a battery thermal runaway management method, applied to a battery thermal runaway management system, the battery thermal runaway management system including at least one set of multimodal sensors deployed inside an energy storage compartment, an edge computing device connected to the multimodal sensors, a management platform connected to the edge computing device, and an instruction execution object connected to the management platform; The method includes: The multimodal sensor collects multimodal physical state data of the target battery cluster in the energy storage compartment when the triggering conditions are met. The edge computing device determines a state feature vector based on the multimodal physical state data, and determines a target transmission strategy and a first monitoring data packet based on the state feature vector and a pre-built rule base. The first monitoring data packet is then uploaded to the management platform based on the target transmission strategy. The first monitoring data packet includes a state feature vector and an initial thermal runaway warning level. The management platform determines the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet, and determines the thermal runaway management instruction based on the target thermal runaway warning level, and sends the thermal runaway management instruction to the corresponding instruction execution object.

[0006] Secondly, embodiments of the present invention provide a battery thermal runaway management system, the battery thermal runaway management system including at least one set of multimodal sensors deployed inside an energy storage compartment, an edge computing device connected to the multimodal sensors, a management platform connected to the edge computing device, and an instruction execution object connected to the management platform; The multimodal sensor is used to collect multimodal physical state data of the target battery cluster in the energy storage compartment when the triggering conditions are met. The edge computing device is used to determine a state feature vector based on the multimodal physical state data, and to determine a target transmission strategy and a first monitoring data packet based on the state feature vector and a pre-built rule base. Based on the target transmission strategy, the first monitoring data packet is uploaded to the management platform. The first monitoring data packet includes a state feature vector and an initial thermal runaway warning level. The management platform is used to determine the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet, and to determine the thermal runaway management instruction based on the target thermal runaway warning level, and to send the thermal runaway management instruction to the corresponding instruction execution object.

[0007] The technical solution of this invention is applied to a battery thermal runaway management system. Using multimodal sensors, when triggering conditions are met, multimodal physical state data of a target battery cluster in the energy storage compartment is collected. Using edge computing devices, a state feature vector is determined based on the multimodal physical state data. A target transmission strategy and a first monitoring data packet are determined based on the state feature vector and a pre-built rule base. The first monitoring data packet, including the state feature vector and an initial thermal runaway warning level, is uploaded to the management platform based on the target transmission strategy. The management platform determines the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet, and determines a thermal runaway management instruction based on the target thermal runaway warning level. The thermal runaway management instruction is then sent to the corresponding instruction execution object. Using this method, multimodal physical state data of the target battery cluster is collected through multimodal sensors, which can comprehensively and multidimensionally reflect the actual operating state of the battery cluster. Through cross-validation of multi-source information, the accuracy, reliability, and timeliness of thermal runaway warnings are effectively improved. By extracting state feature vectors locally through edge computing devices and performing lightweight inference based on a rule base, and determining the target transmission strategy based on the initial thermal runaway warning level obtained from the inference, redundant data transmission can be reduced, communication pressure can be lowered, and system response speed can be improved. At the same time, the computational load on the management platform can be alleviated, enabling rapid identification of early thermal runaway states. The management platform determines the target thermal runaway warning level based on the first monitoring data packet uploaded from the edge and generates corresponding thermal runaway management instructions, enabling accurate decision-making and graded handling. The instructions are then sent to the instruction execution objects to complete closed-loop control, allowing thermal runaway anomalies to be responded to and handled in a timely and effective manner, thereby improving the operational safety and availability of the battery system.

[0008] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart of a battery thermal runaway management method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a battery thermal runaway management system provided in an embodiment of the present invention. Detailed Implementation

[0011] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0012] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0013] It is understood that before using the technical methods disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0014] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose, based on the prompt message, whether to provide personal information to the software or hardware such as the electronic device, application, server, or storage medium performing the operations of this disclosed technology.

[0015] As an optional but non-limiting implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0016] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0017] It should be noted that existing battery thermal runaway monitoring and early warning methods still have the following shortcomings. First, existing methods often integrate two or three types of sensors, such as temperature sensors and gas sensors, failing to comprehensively cover the various physicochemical signals of battery failure (such as acoustic emission, partial discharge, infrared thermal imaging, etc.), resulting in a high risk of missed warnings. Second, existing methods rely heavily on cloud analysis, lacking real-time capability. Third, existing methods cannot spatially locate the thermal runaway source, making it difficult to guide precise and rapid response. Fourth, existing methods often employ simple threshold combinations or post-fusion, failing to fully utilize the complementarity and correlation between different modal sensor data, resulting in low early warning accuracy.

[0018] Based on this, embodiments of the present invention provide a battery thermal runaway management method. Figure 1 This is a flowchart illustrating a battery thermal runaway management method provided in an embodiment of the present invention. This embodiment is applicable to scenarios involving monitoring, early warning, and location-based handling of thermal runaway in energy storage batteries. The method can be applied to a battery thermal runaway management system, which includes at least one set of multimodal sensors deployed inside the energy storage compartment, an edge computing device connected to the multimodal sensors, a management platform connected to the edge computing device, and an instruction execution object connected to the management platform.

[0019] The multimodal sensor can be understood as a group of sensors capable of acquiring multiple signals (such as image signals and sound signals), and a group of multimodal sensors can monitor a battery cluster stored in the energy storage compartment. For example, the energy storage compartment can be a battery cabinet for storing battery clusters. The edge computing device can be considered a device capable of rapidly processing data locally. For example, the edge computing device can be an industrial-grade ruggedized industrial control computer with built-in embedded program modules, which can be used for data acquisition and control, data preprocessing, feature extraction, and data uploading, capable of handling millisecond-level real-time data access, synchronization, noise reduction, feature modeling, and local lightweight early warning. Optionally, all embedded program modules use a real-time Linux kernel, and the embedded program modules communicate with each other through shared memory and RT FIFO. The management platform can be considered a digital twin platform or backend server privately deployed on a local server cluster in the substation, used to make the final thermal runaway risk judgment and send instructions to instruct the corresponding execution objects to implement corresponding disposal measures. For example, the instruction execution objects can include the battery management system and fire suppression system deployed inside the energy storage compartment.

[0020] like Figure 1 As shown, the battery thermal runaway management method provided in this embodiment of the invention may specifically include: S101. Using a multimodal sensor, when the triggering conditions are met, collect multimodal physical state data of the target battery cluster in the energy storage compartment.

[0021] The target battery cluster can be considered as a battery cluster that may experience thermal runaway, as monitored by the multimodal sensor. Multimodal physical state data can be understood as various state data related to battery failure collected by the multimodal sensor, which can be used to reflect the macroscopic physical changes of the battery cluster (such as heat generation) and the microscopic physical changes occurring inside the battery cluster (such as changes in internal microstructure, loose electrical connections, and electrolyte leakage).

[0022] In this embodiment, each sensor in a group of multimodal sensors, when the current triggering conditions are met, such as receiving a data acquisition command sent by an edge computing device, reaching a preset acquisition period, and / or receiving a trigger signal from a sensor associated with a preset hardware trigger chain, begins to acquire the physical state data of the target battery cluster in the energy storage compartment.

[0023] As an optional embodiment, the multimodal sensor includes a flexible ultrasonic sensor, a high-frequency current sensor, a temperature sensor, an infrared thermal imager, and a flue gas monitoring sensor; correspondingly, the multimodal physical state data includes flexible ultrasonic signals, high-frequency current signals, temperature sequences, infrared image streams, and flue gas data.

[0024] Optionally, the flexible ultrasonic sensor can be connected to an edge computing device (such as an industrial PC) via a coaxial cable and a multi-core coaxial connector (such as the LEMO-FGG.0B.304 connector). After being pre-amplified, the signal is input to the edge computing device's PCIe high-speed acquisition card as an LVDS differential signal. The transmission protocol is a custom binary frame structure, supporting continuous streaming acquisition at ≥1MS / s. A high-frequency current sensor can be connected to a dedicated 250MS / s high-speed oscilloscope module via shielded twisted-pair cable, and then directly connected to the industrial PC's memory via a PCIe x4 Gen3 bus, using the IEEE 1588v2 PTP protocol for nanosecond-level hardware timestamp alignment. The temperature sensor can be a CCS temperature sensor, which can be connected to the edge computing device via a CAN FD bus. The physical layer uses an ISO 11898-2 high-speed CAN transceiver, and the data frame format follows the SAE J1939-71 extended protocol, with each frame containing 8 bytes of payload and a baud rate of 2Mbps.

[0025] Optionally, the infrared thermal imager can connect to the edge computing device via Gigabit Ethernet, using the GenICamv3.3 protocol stack, with the image stream pushed in real time via UDP multicast, each frame containing The sensor includes a pixel radiometric matrix (16 bits / pixel) and an embedded timestamp. The flue gas monitoring sensor can include four sensors, each sharing an RS-485 bus to connect to an edge computing device. It uses the Modbus RTU protocol, polling every 2 seconds and returning a 4-byte floating-point concentration value.

[0026] The technical solution described in this embodiment utilizes flexible ultrasonic sensors to collect flexible ultrasonic signals, enabling the monitoring of microstructural changes within the battery cluster. It also employs high-frequency current sensors to collect high-frequency current signals, allowing the monitoring of electrical faults causing partial discharge within the battery cluster. Furthermore, it uses temperature sensors to collect temperature data and construct temperature sequences, enabling the monitoring of thermal anomalies within the battery cluster. Finally, it utilizes infrared thermal imagers and smoke monitoring sensors to detect leaks of dust, gas, etc. Based on this, the collection of multimodal physical state data provides rich and reliable evidence for subsequent assessment of battery thermal runaway risks, improving the reliability and accuracy of the assessment. Moreover, by providing signals or data related to early battery failure, it increases the lead time for early warning, avoiding delayed warnings and providing effective support for emergency response.

[0027] Based on the above optional embodiments, as an optional implementation method, before collecting multimodal physical state data of the target battery cluster in the energy storage compartment using the multimodal sensor when the triggering conditions are currently met, the following optimization can be further included: The edge computing device broadcasts a first message to the multimodal sensor and a second message to the multimodal sensor after sending the first message; the multimodal sensor determines the master-slave clock offset based on the first message and the second message, and generates a globally unified timeline based on the master-slave clock offset; The edge computing device determines the baseline sampling rate and sets up a hardware trigger chain, which includes the rising edge of the high-frequency current sensor triggering synchronous sampling of the flexible ultrasonic sensor and the interruption of the infrared image stream triggering synchronous sampling of the flue gas monitoring sensor.

[0028] The first message can be understood as a message containing a synchronization command, used to trigger all sensors to start timing to establish a unified time reference. For example, the first message could be an IEEE 1588v2 PTP Sync message. The second message can be considered as a message containing the precise timestamp of the first message, used to compensate for network transmission delays and eliminate time errors. For example, the second message could be a Follow_Up message. For example, the reference sampling rate could be selected as 250 megasamples per second (MS / s).

[0029] In this embodiment, the edge computing device may include acquisition and control related programs to achieve deterministic acquisition scheduling and hardware-level time synchronization of multi-source heterogeneous sensors (multimodal sensors). Specifically, during the initialization phase, the edge computing device can load drivers for each sensor, such as LVDS acquisition card drivers, CAN FD controller drivers, GenICam SDK and / or Modbus libraries, and read the sensor firmware version and calibration parameters. Then, the edge computing device can broadcast a first message to all sensors to trigger the synchronization process, followed by sending a second message. Each sensor in the multimodal sensor group determines the master-slave clock offset between the master and slave clocks based on the receiving timestamp of the first message and the precise transmission timestamp of the first message carried in the second message. Based on this master-slave clock offset, the local clock is corrected, enabling all sensors and the edge computing device to form a globally unified time axis UTC+ns. This provides support for subsequent hardware-level nanosecond time synchronization of multimodal physical state data, thereby ensuring the accuracy of thermal runaway risk judgment and handling. It should be noted that the clock of the edge computing device is the master clock, and the clocks of each sensor are slave clocks.

[0030] Optionally, the physical state data collected by each sensor can uniformly carry a PTP timestamp, and each physical state data can be written into the circular buffer of the edge computing device by sensor type and channel. The capacity of the circular buffer can be greater than 2GB and supports resume transmission after network interruption.

[0031] S102. Using edge computing devices, determine the state feature vector based on multimodal physical state data, and determine the target transmission strategy and the first monitoring data packet based on the state feature vector and the pre-built rule base. Based on the target transmission strategy, upload the first monitoring data packet to the management platform. The first monitoring data packet includes the state feature vector and the initial thermal runaway early warning level.

[0032] The state feature vector can be considered as a vector obtained after feature extraction processing of multimodal physical state data. It contains key information related to battery thermal runaway or battery failure and has the characteristics of physical meaning, strong robustness, and controllable dimensionality. The rule base can be considered as a database pre-built based on expert knowledge, storing rules for judging thermal runaway warning levels. The target transmission strategy can be understood as the data transmission strategy corresponding to the initial thermal runaway warning level. For example, the target transmission strategy may include the transmission frequency, and may also include the transmission frequency and the compression and encryption strategy used for transmission. The first monitoring data packet can be considered as a data packet with transmission requirements, used to provide data support and rapid judgment basis for the management platform to finally make a thermal runaway risk judgment. The initial thermal runaway warning level can be understood as the thermal runaway warning level determined by the edge computing device locally through rapid preliminary thermal runaway risk judgment.

[0033] In this embodiment, determining the state feature vector based on multimodal physical state data may include: extracting features from the multimodal physical state data to obtain a state feature vector; or preprocessing the multimodal physical state data and extracting features from the preprocessed multimodal physical state data to obtain a state feature vector. Determining the target transmission strategy and the first monitoring data packet based on the state feature vector and a pre-built rule base may involve: quickly judging the state feature vector using the rule base to determine the initial thermal runaway warning level of the target battery cluster, and determining the target transmission strategy and the first monitoring data packet based on the initial thermal runaway warning level. For example, the target transmission strategy may include a compression encryption strategy and a transmission frequency. The method for uploading the first monitoring data packet to the management platform based on the target transmission strategy can be as follows: The first monitoring data packet is compressed according to the compression and encryption strategy in the target transmission strategy to obtain a structured compressed monitoring data packet. Then, various monitoring data in the compressed monitoring data packet, such as the initial thermal runaway warning level, timestamp, sensor device ID, state feature vector, and / or multimodal physical state data, are uploaded to the management platform's database via HTTP or MQTT according to the transmission frequency corresponding to the target transmission strategy. Optionally, the database may include high-performance time-series databases InfluxDB and TimescaleDB.

[0034] Optionally, the target transmission strategy may also include a network outage handling strategy. Specifically, upon detecting a network outage, the system automatically switches to a 4G LTE Cat.12 backup link and uploads cached data in FIFO queue order to ensure timing integrity.

[0035] S103. Through the management platform, determine the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet, and determine the thermal runaway management instruction based on the target thermal runaway warning level, and send the thermal runaway management instruction to the corresponding instruction execution object.

[0036] The target thermal runaway warning level can be considered the final thermal runaway warning level determined by the management platform. It may be the same as or different from the initial thermal runaway warning level, but it is more accurate. Thermal runaway management instructions can be understood as instructions used to instruct the object to initiate and execute corresponding emergency measures.

[0037] In this embodiment, after receiving the first monitoring data packet sent by the edge computing device, the management platform obtains the initial thermal runaway warning level in the first monitoring data packet and determines the target thermal runaway warning level corresponding to the target battery cluster based on the initial thermal runaway warning level. The method for determining the target thermal runaway warning level based on the initial thermal runaway warning level can be: setting the initial thermal runaway warning level as the target thermal runaway warning level; or the method can be: inputting the initial thermal runaway warning level into a trained fault identification model, and determining the target thermal runaway warning level based on the identification result output by the fault identification model and the corresponding confidence level.

[0038] In this embodiment, after determining the target thermal runaway warning level, corresponding thermal runaway management instructions can be generated based on the target thermal runaway warning level. For example, the target thermal runaway warning level can include Level 1 Attention, Level 2 Warning, and Level 3 Alarm. When the target thermal runaway warning level is Level 1 Attention, the thermal runaway management instructions can include the target thermal runaway warning level, record the event, and generate a monitoring prompt. The management platform or a functional module within the management platform can execute the thermal runaway management instructions. When the target thermal runaway warning level is Level 2 Warning, the thermal runaway management instructions can include the target thermal runaway warning level and generate a warning prompt to instruct maintenance personnel to conduct on-site inspections. The thermal runaway management instructions can be executed by the management platform or a functional module within the management platform. When the target thermal runaway warning level is Level 3 Alarm, the thermal runaway management instructions can include the target thermal runaway warning level and the location information of the thermal runaway source. The thermal runaway management instructions can be sent to the battery management system and fire suppression system to initiate emergency measures such as disconnecting fuses, activating heat dissipation, and activating water mist spraying.

[0039] The battery thermal runaway management method provided in this invention is applied to a battery thermal runaway management system. It uses multimodal sensors to collect multimodal physical state data of a target battery cluster in the energy storage compartment when triggering conditions are met. Using an edge computing device, it determines a state feature vector based on the multimodal physical state data, and determines a target transmission strategy and a first monitoring data packet based on the state feature vector and a pre-built rule base. The first monitoring data packet, including the state feature vector and an initial thermal runaway warning level, is uploaded to the management platform based on the target transmission strategy. The management platform determines the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet, and determines a thermal runaway management instruction based on the target thermal runaway warning level, sending the thermal runaway management instruction to the corresponding instruction execution object. Using this method, the multimodal physical state data of the target battery cluster collected by multimodal sensors can comprehensively and multidimensionally reflect the actual operating state of the battery cluster. Through cross-validation of multi-source information, the accuracy, reliability, and timeliness of thermal runaway warnings are effectively improved. By extracting state feature vectors locally through edge computing devices and performing lightweight inference based on a rule base, and determining the target transmission strategy based on the initial thermal runaway warning level obtained from the inference, redundant data transmission can be reduced, communication pressure can be lowered, and system response speed can be improved. At the same time, the computational load on the management platform can be alleviated, enabling rapid identification of early thermal runaway states. The management platform determines the target thermal runaway warning level based on the first monitoring data packet uploaded from the edge and generates corresponding thermal runaway management instructions, enabling accurate decision-making and graded handling. The instructions are then sent to the instruction execution objects to complete closed-loop control, allowing thermal runaway anomalies to be responded to and handled in a timely and effective manner, thereby improving the operational safety and availability of the battery system.

[0040] As a first optional embodiment of the present invention, based on the above embodiments, the determination of the state feature vector according to the multimodal physical state data can be specified as the following steps: a1) Preprocess the multimodal physical state data to obtain intermediate physical state data.

[0041] Intermediate physical state data can be considered as higher-quality data obtained after preprocessing multimodal physical state data.

[0042] In this embodiment, real-time, low-overhead, and configurable signal cleanup is performed at the edge, providing high-quality input for subsequent feature extraction. For example, the preprocessing of the multimodal physical state data may include: First, for flexible ultrasound signals, a Daubechies-8 wavelet basis (db8) 5-layer decomposition is used to perform soft thresholding denoising on the detail coefficients d1–d3, with the threshold... Where N is the number of sampling points, The noise standard deviation is estimated based on the detail coefficient d1. After reconstruction, effective acoustic emission events with amplitudes greater than 50mV and durations greater than 0.5μs are retained. Second, for high-frequency current signals, a fourth-order Butterworth bandpass filter (1–50 MHz) with a center frequency of 10 MHz is designed to suppress switching noise and power frequency interference; the envelope is extracted using Hilbert transform, and events with amplitudes less than 50mV are removed. The pseudo-discharge pulse, in which, This represents the standard deviation of the envelope signal noise floor. Third, for the temperature sequence, sliding window midpoint filtering and least-squares fitting-based digital filtering (such as Savitzky-Golay smoothing filtering) are applied to eliminate glitches caused by abrupt changes in contact thermal resistance; the temperature rise rate is calculated using the forward difference method, which can be specifically expressed as: ;in, The change in temperature; The time interval can be a 10ms window. Fourth, for the infrared image stream, perform non-uniformity correction, superimpose adaptive histogram equalization to enhance low-contrast areas, and then perform morphological opening operations. Elliptic kernels are used to eliminate dust noise. Fifth, for flue gas data, first-order hysteresis filtering is applied to CO / H2 / VOCs concentrations to suppress ventilation disturbances; smoke optical density is smoothed using an exponentially weighted moving average.

[0043] Following the example described above, outlier handling can also be performed. For instance, dynamic discrimination can be performed on the physical state data of all modalities. The most recent 1000 samples are calculated in a rolling manner. mean with standard deviation Remove The data points are filled with linear interpolation. Next, normalization can be performed. For example, Min-Max normalization is used to output floating-point numbers in the range [0,1].

[0044] b1) Extract features from the intermediate physical state data to obtain a state feature vector.

[0045] The state feature vector includes ultrasonic features for each effective acoustic emission event, current features for each partial discharge pulse, temperature features for multiple temperature channels of the target battery cluster, thermal imaging features for each infrared image frame, and flue gas monitoring features for multiple gases of the target battery cluster. The ultrasonic features include voltage amplitude, amplitude rise time, effective amplitude duration, ring count, center frequency, and peak frequency; the current features include current amplitude, current phase, number of discharges, and spectral energy distribution; the temperature features include temperature value, temperature rise rate, and temperature difference; the thermal imaging features include maximum temperature, hotspot area, and temperature gradient; and the flue gas monitoring features include gas concentration, concentration change rate, and smoke optical density.

[0046] In this embodiment, feature extraction is performed on the preprocessed intermediate physical state data to obtain various state features. These state features are then integrated to obtain a physically meaningful, robust, and dimensionally controllable state feature vector. For example, for each effective acoustic emission event in a flexible ultrasonic signal, the extracted state features include voltage amplitude, amplitude rise time, effective amplitude duration, ring count, center frequency, and peak frequency. Voltage amplitude can be understood as the peak voltage; amplitude rise time can be considered the time required for the amplitude to rise from 10% to 90%; effective amplitude duration can be understood as the duration for the amplitude to be greater than 50% of the peak value; ring count can be considered as the number of zero-crossing points; center frequency and peak frequency can be understood as the frequency corresponding to the centroid of the power spectral density and the frequency corresponding to the maximum value (kHz), respectively. For each partial discharge pulse in a high-frequency current signal, the extracted state features include current amplitude, current phase, number of discharges, and spectral energy distribution. The current amplitude can be considered as the peak current; the current phase can be considered as the angle relative to the power frequency voltage cycle; the number of discharges can be understood as the number of pulses in each power frequency cycle; the spectral energy distribution can be understood as the energy proportion in the 0–20 MHz frequency band after the fast Fourier transform, divided into 2 MHz segments (10 dimensions in total).

[0047] For example, for each of the 12 temperature channels in the temperature sequence of the target battery cluster, the extracted state features include temperature value, temperature rise rate, and temperature difference. The temperature value can be understood as the average of the 12 temperatures; the temperature rise rate can be considered as the maximum temperature rise rate of the 12 channels; and the temperature difference can be understood as the difference between the maximum and minimum temperature values ​​of the 12 channels. For each frame in the infrared image stream of the target battery cluster, the extracted state features include the highest temperature, hotspot area, and temperature gradient. The highest temperature can be understood as the pixel (e.g., ...) The maximum radiation temperature in the data is considered; the hotspot area can be considered as the number of connected pixels whose temperature is greater than the sum of the mean and twice the standard deviation of the temperature; the temperature gradient can be considered as the average horizontal / vertical gradient magnitude calculated by the Sobel operator. For the four gases in the flue gas data of the target battery cluster, the extracted state features include gas concentration, concentration change rate, and smoke optical density. Gas concentration can be considered as the current concentration value of CO / H2 / VOCs / smoke; the concentration change rate can be understood as the slope of the linear fitting of gas concentration within a sliding window (e.g., 60s); smoke optical density can be considered as the extinction coefficient inverted by the Beer-Lambert law.

[0048] The above-described technical solution in this embodiment completes the preprocessing and state feature extraction of multimodal physical state data locally through edge computing devices, avoiding long-distance transmission of large amounts of raw high-resolution data, significantly reducing communication bandwidth consumption and transmission latency, improving system response speed, and sinking basic computing tasks to the edge side, effectively reducing the computing load of the management platform. At the same time, local feature extraction can avoid monitoring failures caused by network latency and packet loss, enhancing the operational stability and reliability of the system in complex industrial scenarios, and providing solid support for early and rapid identification of thermal runaway.

[0049] As a second optional embodiment of the present invention, based on the above embodiments, the steps of determining the target transmission strategy and the first monitoring data packet according to the state feature vector and the pre-built rule base can be optimized as follows: a2) Determine the initial thermal runaway warning level based on the state feature vector and the rule base, generate a first monitoring data packet based on the initial thermal runaway warning level and monitoring data, and generate a warning level label corresponding to the initial thermal runaway warning level. The monitoring data includes the state feature vector, the multimodal physical state data, infrared image frames and / or ultrasonic event waveform segments.

[0050] In this embodiment, the edge computing device uses the extracted state feature vector and compares it with preset rules in the rule base to quickly determine the current thermal runaway risk level of the target battery cluster locally, and generates an initial thermal runaway warning level. For example, if the rule base determines that the state feature vector corresponding to a single sensor is abnormal, such as an increase in ultrasonic events, the initial thermal runaway warning level can be determined as Level 1 concern; if the rule base determines that the state feature vector corresponding to two or more sensors is abnormal simultaneously, or the state feature vector triggers the rule corresponding to Level 2 warning in the rule base, the initial thermal runaway warning level is determined as Level 2 warning; if the state feature vector triggers the rule corresponding to Level 3 alarm in the rule base, the initial thermal runaway warning level is determined as Level 3 alarm; if none of the state features in the state feature vector are abnormal, the initial thermal runaway warning level is determined as normal.

[0051] For example, the rules corresponding to a level-two early warning in the rule base may include: if the temperature rise rate of a certain battery cell is greater than 5℃ / min and the infrared hotspot area of ​​that region is greater than 10cm². 2 If the CO concentration is greater than 50 ppm and the ultrasonic event energy is greater than 1000, a level 2 warning will be triggered. The rule for a level 3 alarm in the rule base can include triggering a level 3 warning if the CO concentration is greater than 50 ppm and the ultrasonic event energy is greater than 1000.

[0052] In this embodiment, the method for generating the first monitoring data packet based on the initial thermal runaway warning level and monitoring data can be as follows: determine the monitoring data according to the initial thermal runaway warning level, and integrate the initial thermal runaway warning level and monitoring data to generate the first monitoring data packet. For example, the method for determining the monitoring data according to the initial thermal runaway warning level can be as follows: if the initial thermal runaway warning level is normal, then the monitoring data includes a state feature vector and an infrared image frame; if the initial thermal runaway warning level is Level 1 concern, then the monitoring data includes a state feature vector and an ultrasonic event waveform segment; if the initial thermal runaway warning level is Level 2 warning or Level 3 alarm, then the monitoring data includes multimodal physical state data. Optionally, the monitoring data may also include a timestamp and the sensor's device ID. Optionally, the infrared image frame may be an infrared thumbnail.

[0053] For example, if the initial thermal runaway warning level is Normal or Level 1 Concern, the generated warning level label can be Normal or P2; if the initial thermal runaway warning level is Level 2 Warning, the generated warning level label can be High or P1; if the initial thermal runaway warning level is Level 3 Alarm, the generated warning level label can be Critical or P0.

[0054] b2) Based on the warning level label, determine the compression encryption strategy and the corresponding transmission frequency of the first monitoring data packet, and generate a target transmission strategy based on the compression encryption strategy and the transmission frequency.

[0055] In this embodiment, different target transmission strategies can be determined based on different warning level labels. When the warning level label represents a normal state or a low-risk level of concern, a transmission strategy of high compression, ordinary encryption, and long time interval can be adopted. When the risk level represented by the warning level label increases, a transmission strategy of low compression, strong encryption, and short time interval can be adopted accordingly.

[0056] For example, the first monitoring data packet labeled P0 or P1 is compressed using open-source lossless high-speed compression, such as Zstandard compression.

[0057] For example, if the warning level label is P2, the transmission frequency corresponding to the first monitoring data packet is determined by: uploading one state feature vector every 5 seconds and one infrared image frame every 30 seconds; if the warning level label is P1, the transmission frequency corresponding to the first monitoring data packet is determined by: uploading one state feature vector every 1 second and one ultrasonic event waveform segment every 5 seconds; if the warning level label is P0, the transmission frequency corresponding to the first monitoring data packet is determined by: enabling TCP fast retransmission and QUIC protocol, and continuously uploading multimodal physical state data in real time at the millisecond level.

[0058] The technical solution described in this embodiment determines the initial thermal runaway warning level by performing rule-based thermal runaway risk judgment at the edge, thereby reducing the computational load, improving the response speed, and ensuring the real-time nature of the judgment and subsequent handling. Based on the initial thermal runaway warning level, the first monitoring data packet and warning level label are determined, and the target transmission strategy is determined according to the warning level label. This can reduce the amount of data transmission and save bandwidth resources in low-risk conditions, and improve the transmission frequency and data integrity in high-risk conditions. While ensuring the real-time nature of the warning, the system resource utilization is optimized, and secure, reliable, and bandwidth-adaptive data transmission from the edge to the platform is achieved.

[0059] As a third optional embodiment of the present invention, the thermal runaway early warning level includes Level 1 concern, Level 2 early warning and Level 3 alarm, and the thermal runaway early warning level includes initial thermal runaway early warning level and target thermal runaway early warning level.

[0060] Accordingly, the determination of the target thermal runaway early warning level corresponding to the target battery cluster based on the first monitoring data packet can be specified as follows: a3) Obtain the initial thermal runaway warning level from the first monitoring data packet. If the initial thermal runaway warning level is a level three alarm, then use the initial thermal runaway warning level as the target thermal runaway warning level.

[0061] b3) If the initial thermal runaway warning level is Level 1 concern or Level 2 warning, then the state feature vector in the first monitoring data packet is input into the fault identification model to obtain the identification result and corresponding confidence level output by the fault identification model.

[0062] The fault identification model can be understood as a trained model capable of identifying whether a battery cluster has experienced thermal runaway. During training, the fault identification model fully learns the complementarity and correlation between physical state data of different modes. The training sample data for the fault identification model can be derived from historical thermal runaway experiments and simulations, and the label categories corresponding to the training sample data can include normal and thermal runaway.

[0063] As one implementation, the step of inputting the state feature vector in the first monitoring data packet into a fault identification model to obtain the identification result and corresponding confidence level output by the fault identification model includes: inputting the state feature vector into a random forest classifier included in the fault identification model to obtain a first confidence level for each fault category output by the random forest classifier; inputting the state feature vector into a deep neural network classifier included in the fault identification model to obtain a second confidence level for each fault category output by the deep neural network classifier; for each fault category, performing a weighted average of the first confidence level and the second confidence level corresponding to the fault category to obtain a third confidence level; determining the fault category corresponding to the largest third confidence level among all the third confidence levels as the identification result, and determining the largest third confidence level as the confidence level corresponding to the identification result.

[0064] In this embodiment, the fault identification model includes a random forest classifier and a deep neural network classifier. The deep neural network classifier can employ a three-layer fully connected network (128-64-32) to further extract deep features. The state feature vector is input into the random forest classifier and the deep neural network classifier in the fault identification model, respectively. The random forest classifier processes the high-dimensional features and outputs the probability distribution of various fault types, i.e., the first confidence score of each fault category. Fault categories include normal and thermal runaway. The deep neural network classifier performs deep feature extraction and mapping on the state feature vector to obtain the second confidence score of each fault category. Then, for each fault category, the first and second confidence scores output by the two classifiers (branches) are weighted and averaged to obtain the third confidence score of the fault category. Finally, the third confidence scores of each fault category are numerically compared to determine the maximum third confidence score. The fault category corresponding to the maximum third confidence score is then determined as the identification result, and the maximum third confidence score is set as the confidence score corresponding to the identification result.

[0065] c3) If the identification result is thermal runaway and the confidence level is greater than the preset threshold, determine the target thermal runaway warning level as a level three alarm; otherwise, use the initial thermal runaway warning level as the target thermal runaway warning level.

[0066] For example, the preset threshold can be 0.8.

[0067] The technical solution described in this embodiment, by classifying the initial thermal runaway warning level, directly uses the level 3 alarm as the target thermal runaway warning level when the edge side determines it. This avoids redundant model calculations, significantly shortens decision-making delays, and enables rapid response in high-risk scenarios, saving valuable time for emergency handling. When the edge side determines it to be a low-to-medium risk level such as Level 1 concern or Level 2 warning, the state feature vector is input into the fault identification model for refined identification. This leverages the model's high accuracy to further improve warning reliability and reduce false alarms and missed alarms. Furthermore, the target thermal runaway warning level is only determined to be a Level 3 alarm when the identification result is thermal runaway and the confidence level meets the standard; otherwise, the initial thermal runaway warning level is used. This ensures that potential risks are not overlooked and avoids unfounded escalation of levels, thereby improving decision-making efficiency while ensuring warning accuracy. This achieves the optimized goal of rapid handling of high-risk situations and accurate identification of low-risk situations.

[0068] As a fourth optional embodiment of the present invention, multiple flexible ultrasonic sensors are arranged in a planar array within the energy storage chamber. Correspondingly, the process of determining a thermal runaway management command based on the target thermal runaway early warning level and sending the thermal runaway management command to the corresponding command execution object can be specified as follows: When the target thermal runaway early warning level is level three, the arrival time difference is determined based on the flexible ultrasonic signal collected by the flexible ultrasonic sensor array combined with the generalized cross-correlation-phase transformation method. The distance difference is determined based on the arrival time difference and the sound wave propagation speed. Under the preset three-dimensional geometric constraints inside the energy storage cabin, a hyperbolic equation system is constructed based on the distance difference. The hyperbolic equation system is solved by the least squares method to obtain the location information of the thermal runaway source. Based on the location information of the thermal runaway source and the target thermal runaway early warning level, the thermal runaway management command is generated and sent to the battery management system and the fire protection system.

[0069] In this context, the thermal runaway source can be understood as the target battery cluster or a specific battery within the target battery cluster that is about to experience or has already experienced a thermal runaway failure.

[0070] In this embodiment, when the target thermal runaway warning level is level three, the system needs to immediately locate the thermal runaway source. At this time, a flexible ultrasonic sensor array can be formed by multiple flexible ultrasonic sensors arranged in a planar array within the energy storage chamber. This array collects flexible ultrasonic signals emitted by the battery due to precursors of thermal runaway such as short circuits, gas generation, microcracks, and partial discharge. Then, the arrival time difference of the same ultrasonic signal to different flexible ultrasonic sensors in the array is determined using the generalized cross-correlation-phase transform method. The product of the arrival time difference and the sound wave propagation speed is determined as the distance difference, used to characterize the distance difference between the thermal runaway source and each flexible ultrasonic sensor. Because the battery clusters are installed on fixed shelves in a fixed chamber, the height coordinates are fixed, thus simplifying three-dimensional spatial positioning to two-dimensional planar coordinate positioning. Each distance difference corresponds to a hyperbola, and the thermal runaway source must lie on this line. The hyperbolic equations are formed by combining the hyperbolic equations, and the least squares method is used to solve these equations to obtain the two-dimensional intersection coordinates of each hyperbola, which is the location information of the thermal runaway source.

[0071] In this embodiment, after determining the location information of the thermal runaway source, the location information of the thermal runaway source and the target thermal runaway warning level are integrated to generate a thermal runaway management command. The thermal runaway management command is then sent to the battery management system (BMS) and the fire protection system in 8-byte binary frame format via a flexible data rate controller area network (CAN FD) and a hard-wired dual channel, enabling them to perform corresponding emergency response based on the thermal runaway management command.

[0072] The above-described technical solution in this embodiment enables rapid and accurate location of the thermal runaway source, and allows for targeted and precise handling of thermal runaway faults, improving handling efficiency and reducing the losses and impacts caused by battery thermal runaway.

[0073] As one implementation, after sending the thermal runaway management command to the battery management system and the fire suppression system, further optimizations can be made, including: The battery management system receives the thermal runaway management command and initiates an emergency protocol based on the target thermal runaway warning level in the thermal runaway management command. The faulty battery cluster is determined based on the location information of the thermal runaway source in the thermal runaway management command, and electrical isolation and enhanced heat dissipation operations are performed on the faulty battery cluster. The fire protection system receives the thermal runaway management command and activates the water mist spraying device based on the target thermal runaway warning level in the thermal runaway management command. The water mist spraying device includes a pan-tilt unit equipped with directional nozzles. Based on the location information of the thermal runaway source in the thermal runaway management command, determine the target azimuth angle and / or target pitch angle of the gimbal, drive the gimbal to rotate to the target azimuth angle and / or target pitch angle, and open the directional nozzle until the preset spray duration is reached.

[0074] In this embodiment, after receiving a thermal runaway management command, the battery management system parses the command, activates an emergency protocol based on the target thermal runaway warning level obtained from the parsing, and determines the battery cluster number corresponding to the location information of the thermal runaway source by querying a mapping table or other means, based on the parsed location information of the thermal runaway source. The battery cluster corresponding to the battery cluster number is then identified as the faulty battery cluster. A faulty battery cluster can be understood as a target battery cluster that is about to experience a thermal runaway fault or is currently experiencing a thermal runaway fault.

[0075] For example, the method for performing electrical isolation and enhanced heat dissipation operations on the faulty battery cluster can be as follows: disconnect the DC-side fuse of the faulty battery cluster within 50 ms and activate the corresponding liquid-cooled plate solenoid valve. Simultaneously, a log of the thermal runaway fault can be generated and stored or reported to the station control system. The log may include the time of occurrence of the thermal runaway fault, the energy storage compartment number, the faulty battery cluster number, and the execution status of the handling action (such as whether the fuse has tripped or the solenoid valve has opened).

[0076] In this embodiment, after receiving a thermal runaway management command, the fire protection system analyzes the command and activates the water mist spraying device according to the analyzed target thermal runaway warning level. Based on the analyzed location information of the thermal runaway source, the system drives the pan-tilt unit in the water mist spraying device to rotate to the target azimuth and / or target pitch angle corresponding to the location information. It can be understood that after the pan-tilt unit rotates to its designated position, the directional nozzles in the water mist spraying device can precisely extinguish the fire in the area where the thermal runaway source is located. At this time, the directional nozzles are activated until the preset spraying duration is reached.

[0077] Optionally, after the spraying is completed, a fire action log can be generated and stored. The fire action log may include the actual spray volume, the location information of the thermal runaway source, and the response time of the consumption system.

[0078] The above-described technical solution in this embodiment, by having the battery management system locate the faulty battery cluster based on the location information of the thermal runaway source, performs targeted electrical isolation and enhanced heat dissipation, can quickly cut off the fault circuit, suppress the spread of temperature rise, and prevent the fault from affecting non-faulty battery clusters, thereby improving the overall safety and operational reliability of the system; by having the fire protection system calculate the pan-tilt angle based on the location information and drive the directional nozzles, it can efficiently achieve precise spraying of water mist towards the thermal runaway source. Compared with full-area spraying, it can reduce water consumption, reduce secondary damage to equipment and waste of resources while ensuring fire extinguishing effect; by rotating the pan-tilt to achieve dynamic alignment of the nozzle azimuth and pitch angle, it can further improve the coverage accuracy of cooling and fire extinguishing, ensure effective suppression of thermal runaway development within the preset spraying time, and maximize the safety of the energy storage compartment.

[0079] Figure 2 This is a schematic diagram of a battery thermal runaway management system provided in an embodiment of the present invention. Figure 2 As shown, the system includes: at least one set of multimodal sensors 21 deployed inside the energy storage compartment; an edge computing device 22 connected to the multimodal sensors 21; a management platform 23 connected to the edge computing device; and an instruction execution object 24 connected to the management platform 23. The multimodal sensor 21 is used to collect multimodal physical state data of the target battery cluster in the energy storage compartment when the triggering conditions are met. The edge computing device 22 is used to determine a state feature vector based on the multimodal physical state data, and to determine a target transmission strategy and a first monitoring data packet based on the state feature vector and a pre-built rule base. Based on the target transmission strategy, the first monitoring data packet is uploaded to the management platform 23. The first monitoring data packet includes a state feature vector and an initial thermal runaway warning level. The management platform 23 is used to determine the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet, and to determine the thermal runaway management instruction based on the target thermal runaway warning level, and to send the thermal runaway management instruction to the corresponding instruction execution object 24.

[0080] The battery thermal runaway management system provided in this invention collects multimodal physical state data of a target battery cluster in an energy storage compartment using multimodal sensors when triggering conditions are met. Using an edge computing device, it determines a state feature vector based on the multimodal physical state data, and determines a target transmission strategy and a first monitoring data packet based on the state feature vector and a pre-built rule base. The first monitoring data packet, including the state feature vector and an initial thermal runaway warning level, is uploaded to the management platform based on the target transmission strategy. The management platform determines the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet, and determines a thermal runaway management instruction based on the target thermal runaway warning level, sending the thermal runaway management instruction to the corresponding instruction execution object. This system, by collecting multimodal physical state data of the target battery cluster using multimodal sensors, can comprehensively and multidimensionally reflect the actual operating state of the battery cluster. Through cross-validation of multi-source information, it effectively improves the accuracy, reliability, and timeliness of thermal runaway warnings. By extracting state feature vectors locally through edge computing devices and performing lightweight inference based on a rule base, and determining the target transmission strategy based on the initial thermal runaway warning level obtained from the inference, redundant data transmission can be reduced, communication pressure can be lowered, and system response speed can be improved. At the same time, the computational load on the management platform can be alleviated, enabling rapid identification of early thermal runaway states. The management platform determines the target thermal runaway warning level based on the first monitoring data packet uploaded from the edge and generates corresponding thermal runaway management instructions, enabling accurate decision-making and graded handling. The instructions are then sent to the instruction execution objects to complete closed-loop control, allowing thermal runaway anomalies to be responded to and handled in a timely and effective manner, thereby improving the operational safety and availability of the battery system.

[0081] Furthermore, the multimodal sensor 21 includes a flexible ultrasonic sensor, a high-frequency current sensor, a temperature sensor, an infrared thermal imager, and a flue gas monitoring sensor; The multimodal physical state data includes flexible ultrasonic signals, high-frequency current signals, temperature sequences, infrared image streams, and flue gas data.

[0082] Furthermore, before the multimodal physical state data of the target battery cluster in the energy storage compartment is collected by the multimodal sensor 21 when the triggering condition is met, the edge computing device 22 can specifically be used to broadcast a first message to the multimodal sensor 21, and after sending the first message, broadcast a second message to the multimodal sensor 21 to receive feedback from the multimodal sensor 21. The multimodal sensor 21 can be used to determine the master-slave clock offset based on the first message and the second message, and generate a globally unified time axis based on the master-slave clock offset. The edge computing device 22 can also be used to determine the reference sampling rate and set up a hardware trigger chain, which includes the rising edge of the high-frequency current sensor triggering the flexible ultrasonic sensor to sample synchronously, and the interruption of the infrared image stream triggering the flue gas monitoring sensor to sample synchronously.

[0083] Furthermore, the edge computing device 22 can also be used for: The multimodal physical state data is preprocessed to obtain intermediate physical state data; Feature extraction is performed on the intermediate physical state data to obtain a state feature vector; The state feature vector includes ultrasonic features for each valid acoustic emission event, current features for each partial discharge pulse, temperature features for multiple temperatures of the target battery cluster, thermal imaging features for each infrared image frame, and flue gas monitoring features for multiple gases of the target battery cluster. The ultrasonic features include voltage amplitude, amplitude rise time, effective amplitude duration, ring count, center frequency, and peak frequency; the current features include current amplitude, current phase, number of discharges, and spectral energy distribution; the temperature features include temperature value, temperature rise rate, and temperature difference; the thermal imaging features include maximum temperature, hot spot area, and temperature gradient; and the flue gas monitoring features include gas concentration, concentration change rate, and smoke optical density.

[0084] Furthermore, the edge computing device 22 can also be used for: The initial thermal runaway warning level is determined based on the state feature vector and the rule base. A first monitoring data packet is generated based on the initial thermal runaway warning level and the monitoring data. A warning level label corresponding to the initial thermal runaway warning level is also generated. The monitoring data includes the state feature vector, the multimodal physical state data, infrared image frames and / or ultrasonic event waveform segments. Based on the warning level label, a compression encryption strategy and the corresponding transmission frequency of the first monitoring data packet are determined, and a target transmission strategy is generated based on the compression encryption strategy and the transmission frequency.

[0085] Furthermore, the thermal runaway early warning levels include Level 1 Concern, Level 2 Early Warning, and Level 3 Alarm; the thermal runaway early warning levels also include Initial Thermal Runaway Early Warning Level and Target Thermal Runaway Early Warning Level. The management platform 23 may specifically include: The first decision module is used to obtain the initial thermal runaway warning level from the first monitoring data packet. If the initial thermal runaway warning level is a level three alarm, then the initial thermal runaway warning level is used as the target thermal runaway warning level. The fault identification module is used to input the state feature vector in the first monitoring data packet into the fault identification model if the initial thermal runaway warning level is Level 1 concern or Level 2 warning, so as to obtain the identification result and corresponding confidence level output by the fault identification model. The second decision module is used to determine the target thermal runaway warning level as a level three alarm if the identification result is thermal runaway and the confidence level is greater than a preset threshold; otherwise, the initial thermal runaway warning level is used as the target thermal runaway warning level.

[0086] Furthermore, the fault identification module can specifically be used for: The state feature vector is input into the random forest classifier included in the fault identification model to obtain the first confidence level of each fault category output by the random forest classifier. The state feature vector is input into the deep neural network classifier included in the fault identification model to obtain the second confidence level of each fault category output by the deep neural network classifier; For each fault category, the first confidence level and the second confidence level corresponding to the fault category are weighted and averaged to obtain the third confidence level; The fault category corresponding to the largest third confidence level among the various third confidence levels is determined as the identification result, and the largest third confidence level is determined as the confidence level corresponding to the identification result.

[0087] Furthermore, multiple flexible ultrasonic sensors are arranged in a planar array inside the energy storage compartment; The management platform 23 can also be used for: When the target thermal runaway early warning level is level three, the arrival time difference is determined based on the flexible ultrasonic signal collected by the flexible ultrasonic sensor array combined with the generalized cross-correlation-phase transformation method. The distance difference is determined based on the arrival time difference and the sound wave propagation speed. Under the preset three-dimensional geometric constraints inside the energy storage cabin, a hyperbolic equation system is constructed based on the distance difference. The hyperbolic equation system is solved by the least squares method to obtain the location information of the thermal runaway source. Based on the location information of the thermal runaway source and the target thermal runaway early warning level, the thermal runaway management command is generated and sent to the battery management system and the fire protection system.

[0088] Furthermore, the battery management system can be used to receive the thermal runaway management instruction after the management platform 23 sends the thermal runaway management instruction to the battery management system, and to initiate an emergency protocol based on the target thermal runaway warning level in the thermal runaway management instruction; determine the faulty battery cluster according to the location information of the thermal runaway source in the thermal runaway management instruction, and perform electrical isolation and enhanced heat dissipation operations on the faulty battery cluster; Furthermore, the fire protection system can be used to receive the thermal runaway management instruction after the management platform 23 sends the thermal runaway management instruction to the fire protection system, and to activate the water mist spraying device based on the target thermal runaway warning level in the thermal runaway management instruction. The water mist spraying device includes a pan-tilt unit equipped with directional nozzles. Based on the location information of the thermal runaway source in the thermal runaway management command, determine the target azimuth angle and / or target pitch angle of the gimbal, drive the gimbal to rotate to the target azimuth angle and / or target pitch angle, and open the directional nozzle until the preset spray duration is reached.

[0089] The battery thermal runaway management system provided in this embodiment of the invention can execute the battery thermal runaway management method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0090] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0091] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A battery thermal runaway management method, characterized in that, An application is made in a battery thermal runaway management system, which includes at least one set of multimodal sensors deployed inside an energy storage compartment, an edge computing device connected to the multimodal sensors, a management platform connected to the edge computing device, and an instruction execution object connected to the management platform. The method includes: The multimodal sensor collects multimodal physical state data of the target battery cluster in the energy storage compartment when the triggering conditions are met. The edge computing device determines a state feature vector based on the multimodal physical state data, and determines a target transmission strategy and a first monitoring data packet based on the state feature vector and a pre-built rule base. The first monitoring data packet is then uploaded to the management platform based on the target transmission strategy. The first monitoring data packet includes a state feature vector and an initial thermal runaway warning level. The management platform determines the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet, and determines the thermal runaway management instruction based on the target thermal runaway warning level, and sends the thermal runaway management instruction to the corresponding instruction execution object.

2. The method according to claim 1, characterized in that, The multimodal sensor includes a flexible ultrasonic sensor, a high-frequency current sensor, a temperature sensor, an infrared thermal imager, and a flue gas monitoring sensor. The multimodal physical state data includes flexible ultrasonic signals, high-frequency current signals, temperature sequences, infrared image streams, and flue gas data.

3. The method according to claim 2, characterized in that, Before acquiring multimodal physical state data of the target battery cluster in the energy storage compartment using the multimodal sensor when the triggering conditions are met, the process further includes: The edge computing device broadcasts a first message to the multimodal sensor, and then broadcasts a second message to the multimodal sensor after sending the first message; Using the multimodal sensor, the master-slave clock offset is determined based on the first message and the second message, and a globally unified time axis is generated based on the master-slave clock offset; The edge computing device determines the baseline sampling rate and sets up a hardware trigger chain, which includes the rising edge of the high-frequency current sensor triggering synchronous sampling of the flexible ultrasonic sensor and the interruption of the infrared image stream triggering synchronous sampling of the flue gas monitoring sensor.

4. The method according to claim 1, characterized in that, Determining the state feature vector based on the multimodal physical state data includes: The multimodal physical state data is preprocessed to obtain intermediate physical state data; Feature extraction is performed on the intermediate physical state data to obtain a state feature vector; The state feature vector includes ultrasonic features for each valid acoustic emission event, current features for each partial discharge pulse, temperature features for multiple temperatures of the target battery cluster, thermal imaging features for each infrared image frame, and flue gas monitoring features for multiple gases of the target battery cluster. The ultrasonic features include voltage amplitude, amplitude rise time, effective amplitude duration, ring count, center frequency, and peak frequency; the current features include current amplitude, current phase, number of discharges, and spectral energy distribution; the temperature features include temperature value, temperature rise rate, and temperature difference; the thermal imaging features include maximum temperature, hot spot area, and temperature gradient; and the flue gas monitoring features include gas concentration, concentration change rate, and smoke optical density.

5. The method according to claim 1, characterized in that, The step of determining the target transmission strategy and the first monitoring data packet based on the state feature vector and the pre-built rule base includes: The initial thermal runaway warning level is determined based on the state feature vector and the rule base. A first monitoring data packet is generated based on the initial thermal runaway warning level and the monitoring data. A warning level label corresponding to the initial thermal runaway warning level is also generated. The monitoring data includes the state feature vector, the multimodal physical state data, infrared image frames and / or ultrasonic event waveform segments. Based on the warning level label, a compression encryption strategy and the corresponding transmission frequency of the first monitoring data packet are determined, and a target transmission strategy is generated based on the compression encryption strategy and the transmission frequency.

6. The method according to claim 1, characterized in that, The thermal runaway warning levels include Level 1 Attention, Level 2 Warning and Level 3 Alarm, and the thermal runaway warning levels include Initial Thermal Runaway Warning Level and Target Thermal Runaway Warning Level; The step of determining the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet includes: The initial thermal runaway warning level is obtained from the first monitoring data packet. If the initial thermal runaway warning level is a level three alarm, the initial thermal runaway warning level is taken as the target thermal runaway warning level. If the initial thermal runaway warning level is Level 1 concern or Level 2 warning, then the state feature vector in the first monitoring data packet is input into the fault identification model to obtain the identification result and corresponding confidence level output by the fault identification model. If the identification result is thermal runaway and the confidence level is greater than a preset threshold, the target thermal runaway warning level is determined to be a Level 3 alarm; otherwise, the initial thermal runaway warning level is used as the target thermal runaway warning level.

7. The method according to claim 6, characterized in that, The step of inputting the state feature vector from the first monitoring data packet into the fault identification model to obtain the identification result and corresponding confidence level output by the fault identification model includes: The state feature vector is input into the random forest classifier included in the fault identification model to obtain the first confidence level of each fault category output by the random forest classifier. The state feature vector is input into the deep neural network classifier included in the fault identification model to obtain the second confidence level of each fault category output by the deep neural network classifier; For each fault category, the first confidence level and the second confidence level corresponding to the fault category are weighted and averaged to obtain the third confidence level; The fault category corresponding to the largest third confidence level among the various third confidence levels is determined as the identification result, and the largest third confidence level is determined as the confidence level corresponding to the identification result.

8. The method according to claim 1, characterized in that, The energy storage chamber contains multiple flexible ultrasonic sensors arranged in a planar array; the process of determining thermal runaway management instructions based on the target thermal runaway early warning level and sending the thermal runaway management instructions to the corresponding instruction execution object includes: When the target thermal runaway early warning level is level three, the arrival time difference is determined based on the flexible ultrasonic signal collected by the flexible ultrasonic sensor array combined with the generalized cross-correlation-phase transformation method. The distance difference is determined based on the arrival time difference and the sound wave propagation speed. Under the preset three-dimensional geometric constraints inside the energy storage cabin, a hyperbolic equation system is constructed based on the distance difference. The hyperbolic equation system is solved by the least squares method to obtain the location information of the thermal runaway source. Based on the location information of the thermal runaway source and the target thermal runaway early warning level, the thermal runaway management command is generated and sent to the battery management system and the fire protection system.

9. The method according to claim 8, characterized in that, After sending the thermal runaway management command to the battery management system and the fire suppression system, the system further includes: The battery management system receives the thermal runaway management command and initiates an emergency protocol based on the target thermal runaway warning level in the thermal runaway management command. The faulty battery cluster is determined based on the location information of the thermal runaway source in the thermal runaway management command, and electrical isolation and enhanced heat dissipation operations are performed on the faulty battery cluster. The fire protection system receives the thermal runaway management command and activates the water mist spraying device based on the target thermal runaway warning level in the thermal runaway management command. The water mist spraying device includes a pan-tilt unit equipped with directional nozzles. Based on the location information of the thermal runaway source in the thermal runaway management command, determine the target azimuth angle and / or target pitch angle of the gimbal, drive the gimbal to rotate to the target azimuth angle and / or target pitch angle, and open the directional nozzle until the preset spray duration is reached.

10. A battery thermal runaway management system, characterized in that, The battery thermal runaway management system includes at least one set of multimodal sensors deployed inside the energy storage compartment, an edge computing device connected to the multimodal sensors, a management platform connected to the edge computing device, and an instruction execution object connected to the management platform. The multimodal sensor is used to collect multimodal physical state data of the target battery cluster in the energy storage compartment when the triggering conditions are met. The edge computing device is used to determine a state feature vector based on the multimodal physical state data, and to determine a target transmission strategy and a first monitoring data packet based on the state feature vector and a pre-built rule base. Based on the target transmission strategy, the first monitoring data packet is uploaded to the management platform. The first monitoring data packet includes a state feature vector and an initial thermal runaway warning level. The management platform is used to determine the target thermal runaway warning level corresponding to the target battery cluster based on the first monitoring data packet, and to determine the thermal runaway management instruction based on the target thermal runaway warning level, and to send the thermal runaway management instruction to the corresponding instruction execution object.

Images