Lithium battery thermal runaway real-time early warning method and system based on multi-source information fusion

Abstract

The invention discloses a lithium battery thermal runaway real-time early warning method and system based on multi-source information fusion, and relates to the technical field of thermal runaway early warning, and the method comprises the steps: obtaining lithium battery operation monitoring data in a lithium battery single body and a lithium battery module, the operation monitoring data comprises voltage, current, temperature, internal resistance, impedance characteristics, pressure and hydrogen gas sensing signals, and carrying out filtering noise reduction, abnormity elimination and time alignment processing on the lithium battery operation monitoring data; according to the invention, the risk assessment and fusion analysis module carries out fusion quantification on electrochemical abnormity, thermal evolution intensity and gas release trend, a continuous thermal runaway risk RSK is obtained through calculation, and the early warning judgment and grade output module realizes graded early warning output. According to the method, the operation state of the battery can be finely sensed and recognized in advance before thermal runaway occurs, so that the perspectiveness and accuracy of thermal safety monitoring of the lithium battery system are effectively improved, and the safety risk caused by sudden thermal runaway events is reduced.

Description

Technical Field

[0001] This invention relates to the field of thermal runaway early warning technology, specifically to a real-time early warning method and system for lithium battery thermal runaway based on multi-source information fusion. Background Technology

[0002] The lithium battery thermal runaway real-time early warning system is an intelligent monitoring and analysis system used to detect lithium battery safety risks in advance. Its core function is to identify dangerous signals that have begun to accumulate inside the battery before serious accidents such as fires or explosions occur. The system continuously collects various information during battery operation, such as changes in voltage, current, temperature, internal resistance, and trace gas release, to perceive the battery's "health status" from multiple perspectives. Under normal circumstances, these parameters change steadily, but when problems such as enhanced side reactions, material aging, or localized overheating occur inside the battery, various signals will show abnormal trends. The early warning system will comprehensively analyze these changes, rather than relying on a single temperature threshold, thereby judging risks earlier and more accurately. Existing real-time early warning systems for thermal runaway of lithium batteries are mostly based on monitoring the state of a single or a small number of operating parameters. They typically use temperature, voltage, or current as the main criteria and determine whether the battery is in an abnormal state by setting fixed thresholds or empirical rules. Such methods are relatively simple in engineering implementation, but they lack effective perception capabilities for hidden evolution processes such as electrochemical degradation, activation of side reactions, and early gas release within the cell. To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a real-time early warning method and system for lithium battery thermal runaway based on multi-source information fusion. By uniformly collecting, processing, and fusing the multi-source monitoring information generated during the operation of lithium battery cells and modules, a multi-dimensional state characterization mechanism that can simultaneously reflect electrochemical state, thermal evolution behavior, and gas release characteristics is constructed, thereby enabling early identification of thermal runaway risks before traditional temperature runaway.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a real-time early warning method and system for lithium battery thermal runaway based on multi-source information fusion, comprising: S1. In the lithium battery cell and module, acquire lithium battery operation monitoring data, including voltage, current, temperature, internal resistance, impedance characteristics, pressure and hydrogen gas sensing signals. S2. The lithium battery operation monitoring data is filtered and denoised, anomaly removed and time aligned, and state feature quantities characterizing voltage stability, temperature evolution characteristics, impedance change characteristics and gas release trend are extracted to form a multi-dimensional state feature vector. S3. Based on the multidimensional state feature vector, the activation behavior of side reactions inside the cell is judged by the data-driven model, and the thermal runaway abnormal evolution feature set is extracted. S4. Input the multidimensional state feature vector and the thermal runaway abnormal evolution feature set into the risk assessment model, and fuse and quantify the degree of electrochemical anomaly, thermal evolution intensity and gas release trend to calculate the thermal runaway risk RSK. S5. Compare the thermal runaway risk RSK with the set multi-level risk thresholds and provide graded early warning prompts.

[0005] The acquisition of lithium battery operation monitoring data includes: At the lithium battery cell level, the cell terminal voltage, current and surface temperature signals are collected, and the cell internal resistance and impedance characteristics are obtained based on the online impedance identification method. Collect the total voltage, branch current, and ambient temperature of the battery module at the battery module level; A gas sensor is placed inside the battery module to collect hydrogen gas sensing signals. By unifying the timestamps of individual cell terminal voltage, current, surface temperature signals, module total voltage, branch current, module ambient temperature, impedance characteristics, and hydrogen gas sensing signals, a lithium battery operation monitoring dataset is formed.

[0006] The preprocessing and feature construction of lithium battery operation monitoring data includes: S21. Perform joint processing of low-pass filtering and median filtering based on sliding time window on the lithium battery operation monitoring dataset; use statistical consistency criteria to remove outliers from data with missing segments, and reconstruct the time alignment of multi-source data through unified sampling period; S22. Calculate state characteristic quantities, including: voltage fluctuation index, voltage stability VOL, temperature rise rate characterizing thermal state evolution, temperature rise rate TMP, temperature change acceleration, temperature acceleration TAC, internal resistance change rate characterizing the degree of electrochemical degradation, internal resistance change RES, impedance trajectory offset, impedance offset IMP, gas concentration change rate, and gas change GAS. S23. The state feature quantities are vectorized and concatenated according to the feature dimension order to construct a multi-dimensional state feature vector.

[0007] The step of vectorizing and concatenating the state features according to their feature dimensions includes: the feature dimension order rule is arranged sequentially according to the logical hierarchy of electrochemical features, thermal features, and gas features. S231. Write the voltage stability VOL, internal resistance change RES, and impedance shift IMP into the electrochemical feature segment of the feature vector in a predetermined order. S232. Write the temperature rise rate (TMP) and temperature acceleration (TAC) characterizing thermal evolution behavior into the thermal feature sub-segment of the feature vector in sequence; S233. Write the gas change GAS, which characterizes the trend of side reaction gas release, into the gas feature sub-segment of the feature vector.

[0008] Building a data-driven model includes: A sample dataset containing different operating stages was obtained, wherein the sample dataset was labeled according to the normal stage, the degradation stage and the thermal runaway precursor stage, and multi-dimensional state feature vectors were obtained to assign side reaction activation labels. Normalization is performed on the multidimensional state feature vector, and the feature weight parameter WGT is introduced to adjust the contribution ratio of voltage stability VOL, temperature evolution TMP, impedance change IMP and gas release trend GAS in the model input. During training, the data-driven model parameters are iteratively updated by minimizing the loss function between the predicted output and the true label, where the thermal runaway risk RSK is calculated by the following formula: Wherein, FEAT_i is the i-th state feature component formed by splicing according to a preset dimension order, and WGT_i is the corresponding feature weight coefficient; when the thermal runaway risk RSK exceeds the preset activation threshold THR, the corresponding sample is determined to be in the side reaction activation state and included in the thermal runaway abnormal evolution feature set.

[0009] The sample dataset contains training samples of normal operation, abnormal evolution, and thermal runaway precursor states. The training sample set is labeled, and the data-driven model is trained offline using supervised learning, enabling the model to learn the mapping relationship between multidimensional state feature vectors and the degree of activation of side reactions.

[0010] The training of the risk assessment model includes: using multidimensional state feature vectors and thermal runaway abnormal evolution feature sets as training input data, constructing a training sample library containing normal intervals, precursor intervals and thermal runaway intervals, and assigning risk labels to each sample based on experimental trigger records and manual review; Perform sample windowing and class balancing on the training sample library to ensure that samples of different risk levels meet the preset ratio in the training batch; For each training sample, the multi-dimensional state feature vector and the abnormal evolution feature are concatenated according to the preset feature dimension order to form the training input vector, and normalization and scaling are performed on each feature dimension. A risk assessment network integrating classification and regression was selected as the model structure. The model parameters were iteratively updated by minimizing the loss function between the predicted risk and the risk label. Cross-validation and early stopping criteria were used to prevent overfitting. After training, the model output is calibrated with a threshold based on the error distribution of the validation set to form a risk index output scale.

[0011] Calculating the thermal runaway risk RSK includes: Using the multidimensional state feature vector and the thermal runaway abnormal evolution feature set as joint input, the joint input is divided into electrochemical abnormal feature subset, thermal evolution feature subset and gas release feature subset according to preset feature categories, and then input into the corresponding feature mapping function to obtain standardized risk feature quantity; Each feature subset is introduced with a weighting factor corresponding to its physical correlation, and the degree of electrochemical anomaly, the intensity of temperature rise rate change and gas release trend are weighted and fused to form a unified risk assessment input vector. The risk assessment input vector is fed into the risk assessment model, and the thermal runaway risk RSK is output.

[0012] Output thermal runaway warning status, including: The system receives the thermal runaway risk indicator (RIS). When the RIS is below the normal threshold (NOR), the lithium battery is determined to be in a safe operating state and a normal state label is output. When the RIS exceeds the first-level threshold (L1T) but does not reach the second-level threshold (L2T), the corresponding early thermal runaway warning level is output. When the RIS exceeds the second-level threshold (L2T) but does not reach the third-level threshold (L3T), a medium-level thermal runaway warning is output. When the RIS exceeds the third-level threshold (L3T), a high-level thermal runaway warning is output.

[0013] A real-time early warning system for thermal runaway of lithium batteries based on multi-source information fusion includes: The multi-source data acquisition module is used to collect voltage, current, temperature, impedance and gas data to form lithium battery operation monitoring data; The data preprocessing and feature construction module preprocesses the lithium battery operation monitoring data to obtain multi-dimensional state feature vectors; The side reaction activation identification module is used to analyze the evolution trend of potential side reactions inside the cell, identify abnormal evolution behaviors caused by electrochemical degradation and thermal anomalies, and extract the feature set of abnormal evolution of thermal runaway. The risk assessment and fusion analysis module fuses and quantifies the multidimensional state feature vector and the thermal runaway anomalous evolution feature set to calculate the thermal runaway risk RSK. The early warning determination and level output module is used to classify risk levels according to risk grading rules and output the corresponding thermal runaway early warning status.

[0014] This invention provides a real-time early warning method and system for lithium battery thermal runaway based on multi-source information fusion. Compared with existing technologies, it has the following advantages: This invention collects voltage, current, temperature, internal resistance, and impedance characteristics of lithium battery cells and modules during operation through a multi-source data acquisition module. A data preprocessing and feature construction module generates a multi-dimensional state feature vector. A side reaction activation identification module identifies potential side reaction activation behaviors and abnormal evolution trends within the cell. A risk assessment and fusion analysis module quantifies electrochemical anomalies, thermal evolution intensity, and gas release trends to calculate the continuous thermal runaway risk (RSK). A graded early warning output module provides tiered early warning output. This allows for refined perception and early identification of the battery's operating state before thermal runaway occurs, effectively improving the foresight and accuracy of lithium battery system thermal safety monitoring and reducing the safety risks associated with sudden thermal runaway events.

[0015] This invention acquires a training sample set containing normal, deterioration, and thermal runaway precursor stages, and combines multi-dimensional state feature vectors and thermal runaway abnormal evolution feature sets to perform offline training and threshold calibration of the risk assessment model. It can output corresponding early warning status labels according to different risk levels, improve the intelligence level of lithium battery system operation management, enhance the adaptability to thermal runaway risks under complex operating conditions and long-term aging conditions, and help improve the operational reliability of energy storage systems. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the process framework of the present invention; Figure 2 This is a schematic diagram of the system framework principle of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0018] Please see Figure 1 This application provides a real-time early warning method and system for lithium battery thermal runaway based on multi-source information fusion, including: S1. In the lithium battery cell and module, acquire lithium battery operation monitoring data, including voltage, current, temperature, internal resistance, impedance characteristics, pressure and hydrogen gas sensing signals. S2. The lithium battery operation monitoring data is filtered and denoised, anomaly removed and time aligned, and state feature quantities characterizing voltage stability, temperature evolution characteristics, impedance change characteristics and gas release trend are extracted to form a multi-dimensional state feature vector. S3. Based on the multidimensional state feature vector, the activation behavior of side reactions inside the cell is judged by the data-driven model, and the thermal runaway abnormal evolution feature set is extracted. S4. Input the multidimensional state feature vector and the thermal runaway abnormal evolution feature set into the risk assessment model, and fuse and quantify the degree of electrochemical anomaly, thermal evolution intensity and gas release trend to calculate the thermal runaway risk RSK. S5. Compare the thermal runaway risk RSK with the set multi-level risk thresholds and provide graded early warning prompts.

[0019] The acquisition of lithium battery operation monitoring data includes: At the lithium battery cell level, the cell terminal voltage, current and surface temperature signals are collected, and the cell internal resistance and impedance characteristics are obtained based on the online impedance identification method. Collect the total voltage, branch current, and ambient temperature of the battery module at the battery module level; A gas sensor is placed inside the battery module to collect hydrogen gas sensing signals. By unifying the timestamps of individual cell terminal voltage, current, surface temperature signals, module total voltage, branch current, module ambient temperature, impedance characteristics, and hydrogen gas sensing signals, a lithium battery operation monitoring dataset is formed.

[0020] In the specific implementation process, under the operating state of the lithium battery system, voltage acquisition channels are set at the positive and negative terminals of each cell at the individual cell level, and current sampling units are set in the corresponding circuits to obtain the cell charging and discharging current. At the same time, temperature sensors are deployed on the surface of the cell casing to collect the cell surface temperature information in real time. Based on the collected cell voltage, current and temperature data, perturbation excitation is periodically superimposed or the current fluctuation characteristics under natural operating conditions are utilized. An online impedance identification algorithm is used to update the parameters of the cell equivalent circuit model, thereby obtaining the cell internal resistance and impedance characteristics that reflect the changes in the cell's electrochemical state. At the battery module level, the overall voltage of the module and the current information of each branch are obtained by setting the module total voltage acquisition port and the branch current detection unit. Ambient temperature sensors are arranged inside or around the module to obtain the module ambient temperature, which is used to characterize the module-level heat distribution and load state. Gas sensing devices are arranged at key locations inside the battery module to detect characteristic gases that may be generated by electrolyte evaporation, SEI film decomposition or side reactions in real time, and to obtain hydrogen gas sensing signals that reflect the degree of gas release.

[0021] The preprocessing and feature construction of lithium battery operation monitoring data includes: S21. Perform joint processing of low-pass filtering and median filtering based on sliding time window on the lithium battery operation monitoring dataset; use statistical consistency criteria to remove outliers from data with missing segments, and reconstruct the time alignment of multi-source data through unified sampling period; S22. Calculate state characteristic quantities, including: voltage fluctuation index, voltage stability VOL, temperature rise rate characterizing thermal state evolution, temperature rise rate TMP, temperature change acceleration, temperature acceleration TAC, internal resistance change rate characterizing the degree of electrochemical degradation, internal resistance change RES, impedance trajectory offset, impedance offset IMP, gas concentration change rate, and gas change GAS. S23. The state feature quantities are vectorized and concatenated according to the feature dimension order to construct a multi-dimensional state feature vector.

[0022] The step of vectorizing and concatenating the state features according to their feature dimensions includes: the feature dimension order rule is arranged sequentially according to the logical hierarchy of electrochemical features, thermal features, and gas features. S231. Write the voltage stability VOL, internal resistance change RES, and impedance shift IMP into the electrochemical feature segment of the feature vector in a predetermined order. S232. Write the temperature rise rate (TMP) and temperature acceleration (TAC) characterizing thermal evolution behavior into the thermal feature sub-segment of the feature vector in sequence; S233. Write the gas change GAS, which characterizes the trend of side reaction gas release, into the gas feature sub-segment of the feature vector.

[0023] Building a data-driven model includes: A sample dataset containing different operating stages was obtained, wherein the sample dataset was labeled according to the normal stage, the degradation stage and the thermal runaway precursor stage, and multi-dimensional state feature vectors were obtained to assign side reaction activation labels. Normalization is performed on the multidimensional state feature vector, and the feature weight parameter WGT is introduced to adjust the contribution ratio of voltage stability VOL, temperature evolution TMP, impedance change IMP and gas release trend GAS in the model input. During training, the data-driven model parameters are iteratively updated by minimizing the loss function between the predicted output and the true label, where the thermal runaway risk RSK is calculated by the following formula: Wherein, FEAT_i is the i-th state feature component formed by splicing according to a preset dimension order, and WGT_i is the corresponding feature weight coefficient; when the thermal runaway risk RSK exceeds the preset activation threshold THR, the corresponding sample is determined to be in the side reaction activation state and included in the thermal runaway abnormal evolution feature set.

[0024] In a specific implementation, the data-driven model construction process can be achieved as follows: First, a sample dataset is constructed based on historical operating data, accelerated aging test data, and thermal runaway triggering test data. This sample dataset covers various operating conditions of lithium batteries during normal operation, performance degradation, and the pre-thermal runaway stage. Each operating sample is labeled with a stage based on test trigger records, expert annotation results, or accident retrospective analysis results. During sample processing, the multi-dimensional state feature vectors obtained in the previous steps are used as model input features. Each feature vector is assigned a side reaction activation label based on whether obvious side reaction signs appear, for the purpose of representing... The activation level of side reactions inside the battery cell is determined. Before model training, the multidimensional state feature vector is normalized to map feature components of different dimensions to a unified numerical range. A feature weight parameter, WGT, is introduced to adjust the relative contribution ratio of voltage stability (VOL), temperature evolution (TMP), impedance change (IMP), and gas release trend (GAS) in the model input, thereby strengthening the feature dimensions that are highly correlated with the activation of side reactions. During training, a discriminative data-driven model with the activation state of side reactions as the output is selected, and the model parameters are iteratively updated by minimizing the loss function between the model's predicted output and the true label.

[0025] Beneficial effects: This invention obtains a training sample set containing normal, deterioration, and thermal runaway precursor stages, and combines multi-dimensional state feature vectors and thermal runaway abnormal evolution feature sets to conduct offline training and threshold calibration of the risk assessment model. It can output corresponding early warning status labels according to different risk levels, improve the intelligence level of lithium battery system operation management, enhance the adaptability to thermal runaway risks under complex operating conditions and long-term aging conditions, and help improve the operational reliability of energy storage systems.

[0026] The sample dataset contains training samples of normal operation, abnormal evolution, and thermal runaway precursor states. The training sample set is labeled, and the data-driven model is trained offline using supervised learning, enabling the model to learn the mapping relationship between multidimensional state feature vectors and the degree of activation of side reactions.

[0027] The training of the risk assessment model includes: using multidimensional state feature vectors and thermal runaway abnormal evolution feature sets as training input data, constructing a training sample library containing normal intervals, precursor intervals and thermal runaway intervals, and assigning risk labels to each sample based on experimental trigger records and manual review; Perform sample windowing and class balancing on the training sample library to ensure that samples of different risk levels meet the preset ratio in the training batch; For each training sample, the multi-dimensional state feature vector and the abnormal evolution feature are concatenated according to the preset feature dimension order to form the training input vector, and normalization and scaling are performed on each feature dimension. A risk assessment network integrating classification and regression was selected as the model structure. The model parameters were iteratively updated by minimizing the loss function between the predicted risk and the risk label. Cross-validation and early stopping criteria were used to prevent overfitting. After training, the model output is calibrated with a threshold based on the error distribution of the validation set to form a risk index output scale.

[0028] In a specific implementation, the training process of the risk assessment model can be carried out as follows: First, a training input dataset is constructed based on the multi-dimensional state feature vector and the thermal runaway abnormal evolution feature set obtained in the aforementioned steps. This dataset covers the continuous operating states of lithium batteries in the normal operating range, the abnormal evolution precursor range, and the thermal runaway range. Combined with thermal runaway triggering experimental records, online monitoring alarm records, and manual review conclusions, each sample is assigned a corresponding risk level label to characterize the thermal runaway risk level of the cell or module at different stages. During sample construction, a sliding time window segmentation process is performed on the original time series data, dividing the continuous operating data into multiple time window samples. Simultaneously, oversampling, undersampling, or weight compensation methods are used to perform class balancing processing on samples of different risk levels, ensuring that samples of each risk level are balanced during training. The batches meet the preset proportion requirements to avoid class bias during model training. Subsequently, for each training sample, the multi-dimensional state feature vector is concatenated with the corresponding thermal runaway anomaly evolution features according to the preset feature dimension order to form a unified training input vector. Normalization and scaling are then performed on the concatenated features to eliminate the influence of different feature dimensions on the model convergence process. In terms of model structure selection, a risk assessment network with both classification and regression capabilities is used as the main body of the model, enabling the model to learn the discrimination boundaries of different risk intervals and output a continuous risk index. The model parameters are iteratively updated by minimizing the loss function between the model's predicted risk value and the sample's true risk label. Cross-validation mechanism and early stopping criterion are introduced during training to prevent model overfitting and improve generalization ability.

[0029] Calculating the thermal runaway risk RSK includes: Using the multidimensional state feature vector and the thermal runaway abnormal evolution feature set as joint input, the joint input is divided into electrochemical abnormal feature subset, thermal evolution feature subset and gas release feature subset according to preset feature categories, and then input into the corresponding feature mapping function to obtain standardized risk feature quantity; Each feature subset is introduced with a weighting factor corresponding to its physical correlation, and the degree of electrochemical anomaly, the intensity of temperature rise rate change and gas release trend are weighted and fused to form a unified risk assessment input vector. The risk assessment input vector is fed into the risk assessment model, and the thermal runaway risk RSK is output.

[0030] In a specific implementation, the calculation of the thermal runaway risk RSK can be carried out according to the following process: First, during real-time operation, the multi-dimensional state feature vector formed in the previous steps and the identified thermal runaway abnormal evolution feature set are used as joint input data. Feature classification rules are introduced to structurally divide the joint input. Specifically, data reflecting voltage stability, internal resistance change, and impedance shift characteristics are classified as the electrochemical abnormal feature subset, data reflecting temperature rise rate and temperature acceleration evolution characteristics are classified as the thermal evolution feature subset, and data reflecting gas concentration change trends are classified as the gas release feature subset. Subsequently, the electrochemical abnormal feature subset, thermal evolution feature subset, and gas release feature subset are respectively input to the corresponding... The feature mapping function performs nonlinear mapping and normalization on features with different physical dimensions and distribution characteristics to obtain standardized risk feature quantities with a unified numerical scale. On this basis, a weighting factor matching the degree of physical correlation of each feature subset is introduced to perform weighted fusion of electrochemical anomaly degree, thermal evolution intensity and gas release trend to construct a risk assessment input vector characterizing the current comprehensive instability risk state of the cell or battery module. Finally, the risk assessment input vector is fed into the risk assessment model that has been trained offline and has completed threshold calibration. The model performs comprehensive quantitative calculation of the coupling contribution of various risk features and outputs a continuous thermal runaway risk RSK.

[0031] Output thermal runaway warning status, including: The system receives the thermal runaway risk indicator (RIS). When the RIS is below the normal threshold (NOR), the lithium battery is determined to be in a safe operating state and a normal state label is output. When the RIS exceeds the first-level threshold (L1T) but does not reach the second-level threshold (L2T), the corresponding early thermal runaway warning level is output. When the RIS exceeds the second-level threshold (L2T) but does not reach the third-level threshold (L3T), a medium-level thermal runaway warning is output. When the RIS exceeds the third-level threshold (L3T), a high-level thermal runaway warning is output.

[0032] Beneficial effects: This invention collects voltage, current, temperature, internal resistance, and impedance characteristics of lithium battery cells and modules during operation through a multi-source data acquisition module. A data preprocessing and feature construction module forms a multi-dimensional state feature vector. A side reaction activation identification module identifies potential side reaction activation behaviors and abnormal evolution trends within the cell. A risk assessment and fusion analysis module quantifies electrochemical anomalies, thermal evolution intensity, and gas release trends to calculate the continuous thermal runaway risk (RSK). A graded early warning output module provides tiered early warning output. This allows for refined perception and early identification of the battery's operating state before thermal runaway occurs, effectively improving the foresight and accuracy of lithium battery system thermal safety monitoring and reducing the safety risks associated with sudden thermal runaway events.

[0033] A real-time early warning system for thermal runaway of lithium batteries based on multi-source information fusion includes: The multi-source data acquisition module is used to collect voltage, current, temperature, impedance and gas data to form lithium battery operation monitoring data; The data preprocessing and feature construction module preprocesses the lithium battery operation monitoring data to obtain multi-dimensional state feature vectors; The side reaction activation identification module is used to analyze the evolution trend of potential side reactions inside the cell, identify abnormal evolution behaviors caused by electrochemical degradation and thermal anomalies, and extract the feature set of abnormal evolution of thermal runaway. The risk assessment and fusion analysis module fuses and quantifies the multidimensional state feature vector and the thermal runaway anomalous evolution feature set to calculate the thermal runaway risk RSK. The early warning determination and level output module is used to classify risk levels according to risk grading rules and output the corresponding thermal runaway early warning status.

[0034] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0035] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0036] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A real-time early warning method for lithium battery thermal runaway based on multi-source information fusion, characterized in that, include: S1. In the lithium battery cell and module, acquire lithium battery operation monitoring data, including voltage, current, temperature, internal resistance, impedance characteristics, pressure and hydrogen gas sensing signals. S2. The lithium battery operation monitoring data is filtered and denoised, anomaly removed and time aligned, and state feature quantities characterizing voltage stability, temperature evolution characteristics, impedance change characteristics and gas release trend are extracted to form a multi-dimensional state feature vector. S3. Based on the multidimensional state feature vector, the activation behavior of side reactions inside the cell is judged by the data-driven model, and the thermal runaway abnormal evolution feature set is extracted. S4. Input the multidimensional state feature vector and the thermal runaway abnormal evolution feature set into the risk assessment model, and fuse and quantify the degree of electrochemical anomaly, thermal evolution intensity and gas release trend to calculate the thermal runaway risk RSK. S5. Compare the thermal runaway risk RSK with the set multi-level risk thresholds and provide graded early warning prompts.

2. The real-time early warning method for lithium battery thermal runaway based on multi-source information fusion according to claim 1, characterized in that, The acquisition of lithium battery operation monitoring data includes: At the lithium battery cell level, the cell terminal voltage, current and surface temperature signals are collected, and the cell internal resistance and impedance characteristics are obtained based on the online impedance identification method. Collect the total voltage, branch current, and ambient temperature of the battery module at the battery module level; A gas sensor is placed inside the battery module to collect hydrogen gas sensing signals. By unifying the timestamps of individual cell terminal voltage, current, surface temperature signals, module total voltage, branch current, module ambient temperature, impedance characteristics, and hydrogen gas sensing signals, a lithium battery operation monitoring dataset is formed.

3. The real-time early warning method for lithium battery thermal runaway based on multi-source information fusion according to claim 1, characterized in that, The preprocessing and feature construction of lithium battery operation monitoring data includes: S21. Perform joint processing of low-pass filtering and median filtering based on sliding time window on the lithium battery operation monitoring dataset; use statistical consistency criteria to remove outliers from data with missing segments, and reconstruct the time alignment of multi-source data through unified sampling period; S22. Calculate state characteristic quantities, including: voltage fluctuation index, voltage stability VOL, temperature rise rate characterizing thermal state evolution, temperature rise rate TMP, temperature change acceleration, temperature acceleration TAC, internal resistance change rate characterizing the degree of electrochemical degradation, internal resistance change RES, impedance trajectory offset, impedance offset IMP, gas concentration change rate, and gas change GAS. S23. The state feature quantities are vectorized and concatenated according to the feature dimension order to construct a multi-dimensional state feature vector.

4. The real-time early warning method for lithium battery thermal runaway based on multi-source information fusion according to claim 3, characterized in that, The step of vectorizing and concatenating the state features according to their feature dimensions includes: the feature dimension order rule is arranged sequentially according to the logical hierarchy of electrochemical features, thermal features, and gas features. S231. Write the voltage stability VOL, internal resistance change RES, and impedance shift IMP into the electrochemical feature segment of the feature vector in a predetermined order. S232. Write the temperature rise rate (TMP) and temperature acceleration (TAC) characterizing thermal evolution behavior into the thermal feature sub-segment of the feature vector in sequence; S233. Write the gas change GAS, which characterizes the trend of side reaction gas release, into the gas feature sub-segment of the feature vector.

5. The real-time early warning method for lithium battery thermal runaway based on multi-source information fusion according to claim 1, characterized in that, Building a data-driven model includes: A sample dataset containing different operating stages was obtained, wherein the sample dataset was labeled according to the normal stage, the degradation stage and the thermal runaway precursor stage, and multi-dimensional state feature vectors were obtained to assign side reaction activation labels. Normalization is performed on the multidimensional state feature vector, and the feature weight parameter WGT is introduced to adjust the contribution ratio of voltage stability VOL, temperature evolution TMP, impedance change IMP and gas release trend GAS in the model input. During training, the data-driven model parameters are iteratively updated by minimizing the loss function between the predicted output and the true label, where the thermal runaway risk RSK is calculated by the following formula: Wherein, FEAT_i is the i-th state feature component formed by splicing according to a preset dimension order, and WGT_i is the corresponding feature weight coefficient; when the thermal runaway risk RSK exceeds the preset activation threshold THR, the corresponding sample is determined to be in the side reaction activation state and included in the thermal runaway abnormal evolution feature set.

6. The real-time early warning method for lithium battery thermal runaway based on multi-source information fusion according to claim 1, characterized in that, The sample dataset contains training samples of normal operation, abnormal evolution, and thermal runaway precursor states. The training sample set is labeled, and the data-driven model is trained offline using supervised learning, enabling the model to learn the mapping relationship between multidimensional state feature vectors and the degree of activation of side reactions.

7. The real-time early warning method for lithium battery thermal runaway based on multi-source information fusion according to claim 1, characterized in that, The training of the risk assessment model includes: using multidimensional state feature vectors and thermal runaway abnormal evolution feature sets as training input data, constructing a training sample library containing normal intervals, precursor intervals and thermal runaway intervals, and assigning risk labels to each sample based on experimental trigger records and manual review; The training sample library is subjected to sample windowing and class balancing to ensure that samples of different risk levels meet the preset proportion in the training batch. For each training sample, the multi-dimensional state feature vector and the abnormal evolution feature are concatenated according to the preset feature dimension order to form the training input vector, and normalization and scaling are performed on each feature dimension. A risk assessment network integrating classification and regression was selected as the model structure. The model parameters were iteratively updated by minimizing the loss function between the predicted risk and the risk label. Cross-validation and early stopping criteria were used to prevent overfitting. After training, the model output is calibrated with a threshold based on the error distribution of the validation set to form a risk index output scale.

8. The real-time early warning method for lithium battery thermal runaway based on multi-source information fusion according to claim 1, characterized in that, Calculating the thermal runaway risk RSK includes: Using the multidimensional state feature vector and the thermal runaway abnormal evolution feature set as joint input, the joint input is divided into electrochemical abnormal feature subset, thermal evolution feature subset and gas release feature subset according to preset feature categories, and then input into the corresponding feature mapping function to obtain standardized risk feature quantity; Each feature subset is introduced with a weighting factor corresponding to its physical correlation, and the degree of electrochemical anomaly, the intensity of temperature rise rate change and gas release trend are weighted and fused to form a unified risk assessment input vector. The risk assessment input vector is fed into the risk assessment model, and the thermal runaway risk RSK is output.

9. The real-time early warning method for lithium battery thermal runaway based on multi-source information fusion according to claim 1, characterized in that, Output thermal runaway warning status, including: The system receives the thermal runaway risk indicator (RIS). When the RIS is below the normal threshold (NOR), the lithium battery is determined to be in a safe operating state and a normal state label is output. When the RIS exceeds the first-level threshold (L1T) but does not reach the second-level threshold (L2T), the corresponding early thermal runaway warning level is output. When the RIS exceeds the second-level threshold (L2T) but does not reach the third-level threshold (L3T), a medium-level thermal runaway warning is output. When the RIS exceeds the third-level threshold (L3T), a high-level thermal runaway warning is output.

10. A real-time early warning system for thermal runaway of lithium batteries based on multi-source information fusion, based on the real-time early warning method for thermal runaway of lithium batteries based on multi-source information fusion as described in any one of claims 1-9, characterized in that, include: The multi-source data acquisition module is used to collect voltage, current, temperature, impedance and gas data to form lithium battery operation monitoring data; The data preprocessing and feature construction module preprocesses the lithium battery operation monitoring data to obtain multi-dimensional state feature vectors; The side reaction activation identification module is used to analyze the evolution trend of potential side reactions inside the cell, identify abnormal evolution behaviors caused by electrochemical degradation and thermal anomalies, and extract the feature set of abnormal evolution of thermal runaway. The risk assessment and fusion analysis module fuses and quantifies the multidimensional state feature vector and the thermal runaway anomalous evolution feature set to calculate the thermal runaway risk RSK. The early warning determination and level output module is used to classify risk levels according to risk grading rules and output the corresponding thermal runaway early warning status.

Images