Battery pack thermal runaway risk level determination method, device and electronic equipment

Abstract

The application discloses a battery pack thermal runaway risk level determination method and device and electronic equipment. The method comprises the following steps: obtaining running data of a battery pack in a preset time period, and dividing the running data according to two division strategies respectively to obtain a first data sequence and a second data sequence of the battery pack, wherein the first data sequence comprises a plurality of first data segments, the second data sequence comprises a plurality of second data segments, the first number of the plurality of first data segments is less than the second number of the plurality of second data segments; determining a plurality of first loss values based on the first data sequence; determining a plurality of second loss values based on the second data sequence; determining a plurality of comprehensive loss values based on the plurality of first loss values and the plurality of second loss values; and determining a thermal runaway risk level of the battery pack based on the plurality of comprehensive loss values. The application solves the technical problem of inaccurate determination result of the thermal runaway risk level of the battery pack in the related art.

Description

Technical Field

[0001] This application relates to the field of battery safety technology, and more specifically, to a method, apparatus, and electronic device for determining the thermal runaway risk level of a battery pack. Background Technology

[0002] Lithium-ion battery thermal runaway is a core safety hazard in new energy applications such as electric vehicles and energy storage power stations. The resulting high temperatures, combustion, and explosions can easily cause equipment damage and personal injury. However, in actual operating conditions, the online time of lithium-ion battery packs is uncontrollable due to factors such as usage scenarios and maintenance requirements, leading to significant differences in the length of collected time-series data sequences. Single models have strict requirements on the length of the input sequence and cannot adapt to data from all battery packs. Therefore, related technologies suffer from inaccurate determination of the thermal runaway risk level of battery packs.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a method, apparatus, and electronic device for determining the thermal runaway risk level of a battery pack, in order to at least solve the technical problem of inaccurate determination results of the thermal runaway risk level of a battery pack in related technologies.

[0005] According to one aspect of the embodiments of this application, a method for determining the thermal runaway risk level of a battery pack is provided, comprising: acquiring operational data of the battery pack over a preset time period, and segmenting the operational data according to two segmentation strategies to obtain a first data sequence and a second data sequence of the battery pack, wherein the first data sequence includes multiple first data segments, the second data sequence includes multiple second data segments, and the first number of the multiple first data segments is less than the second number of the multiple second data segments; determining multiple first loss values ​​based on the first data sequence, wherein the multiple first loss values ​​correspond one-to-one with the multiple first data segments; determining multiple second loss values ​​based on the second data sequence, wherein the multiple second loss values ​​correspond one-to-one with the multiple second data segments; determining multiple comprehensive loss values ​​based on the multiple first loss values ​​and the multiple second loss values, wherein the multiple comprehensive loss values ​​correspond one-to-one with the multiple second data segments; and determining the thermal runaway risk level of the battery pack based on the multiple comprehensive loss values. By fusing the first loss values ​​and the second loss values, a collaborative assessment of battery pack state anomalies at different time scales is achieved, improving the accuracy of the determination result of the thermal runaway risk level of the battery pack.

[0006] Optionally, based on the first data sequence, multiple first loss values ​​are determined, including: based on multiple first data segments, determining the first target features corresponding to each of the multiple first data segments, and the first labels corresponding to each of the multiple first data segments; based on the first target features corresponding to each of the multiple first data segments, and the first labels corresponding to each of the multiple first data segments, using a first prediction model and a first loss function corresponding to the first prediction model, to obtain multiple first loss values. By constructing a first prediction model based on TimesNet and combining it with the first loss function, the long-term degradation trend and periodicity of the battery pack's long-term operating data are accurately captured, laying the foundation for determining the subsequent thermal runaway risk level.

[0007] Optionally, based on the second data sequence, multiple second loss values ​​are determined, including: determining the second target features corresponding to each of the multiple second data segments, and the second labels corresponding to each of the multiple second data segments; and using the second prediction model and the second loss function corresponding to the second prediction model to obtain multiple second loss values ​​based on the second target features and the second labels corresponding to the multiple second data segments. By constructing a second prediction model based on TCN and combining it with the second loss function, accurate capture of local burst features in short-time series data of the battery pack is achieved, laying the foundation for subsequent determination of thermal runaway risk levels.

[0008] Optionally, multiple comprehensive loss values ​​are determined based on multiple first loss values ​​and multiple second loss values, including: aligning and supplementing multiple first loss values ​​and multiple second loss values ​​with multiple second loss values ​​as a benchmark to obtain multiple loss value results, wherein each loss value result corresponds one-to-one with multiple second data segments, and each loss value result includes a second loss value and an aligned loss value, wherein the aligned loss value is a first loss value or a preset value; determining multiple gating factors based on the second target features corresponding to the multiple second data segments respectively, wherein each gating factor corresponds one-to-one with the multiple second data segments; obtaining multiple first weights based on the multiple gating factors and multiple aligned loss values, wherein each aligned loss value is determined based on the multiple loss value results, and each first weight corresponds one-to-one with the multiple aligned loss values; obtaining multiple second weights based on the multiple gating factors and multiple second loss values, wherein each second weight corresponds one-to-one with the multiple second loss values; and obtaining multiple comprehensive loss values ​​based on the multiple aligned loss values, the multiple first weights corresponding to the multiple aligned loss values, the multiple second loss values, and the multiple second weights corresponding to the multiple second loss values. By introducing a dynamic gating mechanism based on voltage change rate and temperature difference change rate, and adaptively weighting the fusion of loss values, the fusion problem after aligning data segments at different time scales is solved, and the accuracy of the thermal runaway risk level determination is improved.

[0009] Optionally, the thermal runaway risk level of the battery pack is determined based on multiple comprehensive loss values, including: dividing the multiple comprehensive loss values ​​into subsets corresponding to multiple local time windows using a sliding time window approach; for any local time window, determining a third target feature of any local time window based on multiple first comprehensive loss values ​​included in the subset of loss values ​​of any local time window; obtaining third target features corresponding to multiple local time windows by determining the third target features of any local time window; and determining the thermal runaway risk level based on the third target features corresponding to multiple local time windows and multiple comprehensive loss values. By introducing a sliding window to perform time-series trend analysis on the comprehensive loss values ​​and extracting features such as mean and slope to characterize the evolution of the battery pack state, the error caused by single-moment judgment is effectively reduced, and the accuracy and stability of the thermal runaway risk level determination results are improved.

[0010] Optionally, the thermal runaway risk level is determined based on the third target features corresponding to multiple local time windows and multiple comprehensive loss values. This includes: determining a target comprehensive loss value based on multiple comprehensive loss values, wherein the target comprehensive loss value is the comprehensive loss value corresponding to the last second data segment in the second data sequence; determining the thermal runaway trend of the battery pack within a preset time period based on the third target features corresponding to multiple local time windows; and determining the thermal runaway risk level based on the target comprehensive loss value and the thermal runaway trend. By combining the comprehensive loss value of the last second data segment (reflecting the degree of instantaneous anomaly) with the thermal runaway trend within the preset time period (reflecting the state evolution pattern), it is possible to effectively distinguish between instantaneous disturbances and actual thermal runaway precursors, thereby improving the accuracy and rationality of the thermal runaway risk level determination results.

[0011] Optionally, when the thermal runaway trend includes stable, slowly rising, rapidly rising, and steeply rising, the thermal runaway risk level is determined based on the target comprehensive loss value and the thermal runaway trend. This includes: if the target comprehensive loss value is greater than a preset thermal runaway threshold, and the thermal runaway trend is either rapidly rising or steeply rising, the thermal runaway risk level is determined to be high risk; if the target comprehensive loss value is greater than the preset thermal runaway threshold, and the thermal runaway trend is either stable or slowly rising, the thermal runaway risk level is determined to be low risk; if the target comprehensive loss value is less than or equal to the preset thermal runaway threshold, and the thermal runaway trend is either rapidly rising or steeply rising, the thermal runaway risk level is determined to be low risk; if the target comprehensive loss value is less than or equal to the preset thermal runaway threshold, and the thermal runaway trend is either stable or slowly rising, the thermal runaway risk level is determined to be no risk. By constructing a dual-dimensional risk assessment mechanism combining "instantaneous anomaly degree" and "long-term evolution trend," the shortcomings of single-threshold early warning, which is susceptible to instantaneous fluctuations, are effectively overcome, achieving accurate determination and early identification of the battery pack thermal runaway risk level.

[0012] According to another aspect of the embodiments of this application, a device for determining the thermal runaway risk level of a battery pack is provided, comprising: a data acquisition module, configured to acquire operating data of the battery pack during a preset time period, and to segment the operating data according to two segmentation strategies to obtain a first data sequence and a second data sequence of the battery pack, wherein the first data sequence includes a plurality of first data segments, the second data sequence includes a plurality of second data segments, and the first number of the plurality of first data segments is less than the second number of the plurality of second data segments; a first determination module, configured to determine a plurality of first loss values ​​based on the first data sequence, wherein the plurality of first loss values ​​correspond one-to-one with the plurality of first data segments; a second determination module, configured to determine a plurality of second loss values ​​based on the second data sequence, wherein the plurality of second loss values ​​correspond one-to-one with the plurality of second data segments; a third determination module, configured to determine a plurality of comprehensive loss values ​​based on the plurality of first loss values ​​and the plurality of second loss values, wherein the plurality of comprehensive loss values ​​correspond one-to-one with the plurality of second data segments; and a fourth determination module, configured to determine the thermal runaway risk level of the battery pack based on the plurality of comprehensive loss values.

[0013] According to another aspect of the embodiments of this application, a non-volatile storage medium is provided, which stores multiple instructions adapted for a method for determining the thermal runaway risk level of a battery pack, wherein any one of the instructions is loaded by a processor.

[0014] According to another aspect of the embodiments of this application, an electronic device is provided, including: one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any one of the following methods for determining the thermal runaway risk level of a battery pack.

[0015] According to another aspect of the embodiments of this application, a computer program product is provided, which, when executed on a data processing device, is adapted to perform the steps of a method for determining the thermal runaway risk level of a battery pack.

[0016] In this embodiment, by acquiring the operating data of the battery pack over a preset time period and segmenting the operating data according to two segmentation strategies, a first data sequence and a second data sequence of the battery pack are obtained. The first data sequence includes multiple first data segments, and the second data sequence includes multiple second data segments. The first number of the multiple first data segments is less than the second number of the multiple second data segments. Based on the first data sequence, multiple first loss values ​​are determined, wherein each of the multiple first loss values ​​corresponds one-to-one with a multiple of the multiple first data segments. Based on the second data sequence, multiple second loss values ​​are determined, wherein each of the multiple second loss values ​​corresponds one-to-one with a multiple of the multiple second data segments. Based on the multiple first loss values ​​and the multiple second loss values, multiple comprehensive loss values ​​are determined, wherein each of the multiple comprehensive loss values ​​corresponds one-to-one with a multiple of the multiple second data segments. Based on the multiple comprehensive loss values, the thermal runaway risk level of the battery pack is determined. The goal is to determine the thermal runaway risk level of the battery pack by fusing multiple first loss values ​​and multiple second loss values ​​determined based on the battery pack's operating data to obtain multiple comprehensive loss values. This improves the accuracy of the determination of the thermal runaway risk level of the battery pack and solves the technical problem of inaccurate determination of the thermal runaway risk level of the battery pack in related technologies. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0018] Figure 1 This is a flowchart of a method for determining the thermal runaway risk level of a battery pack according to an embodiment of this application;

[0019] Figure 2 This is a flowchart of an optional method for determining the thermal runaway risk level of a battery pack according to an embodiment of this application;

[0020] Figure 3 This is a schematic diagram of an optional first thermal runaway detection result provided according to an embodiment of this application;

[0021] Figure 4 This is an optional dynamic gating weight adaptive change curve provided according to an embodiment of this application;

[0022] Figure 5This is a schematic diagram of an optional second thermal runaway detection result provided according to an embodiment of this application;

[0023] Figure 6 This is a schematic diagram of a device for determining the thermal runaway risk level of a battery pack according to an embodiment of this application;

[0024] Figure 7 This is a structural diagram of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application 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 this application 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 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.

[0027] It should be noted that the information collected in this application (including but not limited to preset thermal runaway thresholds) and data (including but not limited to battery pack operating data over preset time periods) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with relevant laws, regulations, and standards, and necessary confidentiality measures have been taken. This process does not violate public order and good morals, and corresponding access points are provided for users to choose whether to authorize or refuse. For example, interfaces are set up between this system and relevant users or organizations, providing users with corresponding access points to choose whether to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0028] According to the embodiments of the present application, a method embodiment of a method for determining a thermal runaway risk level of a battery pack is provided. It should be noted that the steps shown in the flow diagram of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flow diagram, in some cases, the steps shown or described can be executed in an order different from that presented herein.

[0029] Figure 1 is a flow chart of a method for determining a thermal runaway risk level of a battery pack provided according to the embodiments of the present application, as Figure 1 shown, the method includes the following steps:

[0030] Step S102: acquiring operation data of the battery pack in a preset time period, and splitting the operation data respectively according to two splitting strategies to obtain a first data sequence and a second data sequence of the battery pack, wherein the first data sequence includes a plurality of first data segments, the second data sequence includes a plurality of second data segments, and a first number of the plurality of first data segments is smaller than a second number of the plurality of second data segments;

[0031] It can be understood that a BMS (Battery Management System) can be used to acquire operation data of the battery pack in the preset time period, such as the minimum voltage, maximum temperature, state of charge (SOC), charging and discharging current of the battery pack. The first splitting strategy of the two splitting strategies is: the time span of the data sequence of the obtained first data segment is ≥ TL, the sampling step of the first data segment is tL, and the number of data points in the data sequence of the first data segment is ≥ NL; the second splitting strategy of the two splitting strategies is: the time span of the data sequence of the second data segment is ≥ TS (TS < TL), the sampling step of the second data segment is tS (tS ≤ tL), and the number of data points in the data sequence of the second data segment is ≥ NS (NS < NL). By setting differentiated splitting strategies, two-track splitting of the same operation data is realized, which lays a foundation for the subsequent determination of a first loss value and a second loss value.

[0032] Optionally, the operation data of the lithium battery pack can be collected in real time by the BMS and stored in a cloud platform database. The collected operation data includes but is not limited to the minimum voltage, maximum temperature, state of charge (SOC), charging and discharging current, etc. Meanwhile, the actual online duration and original operation data timestamp of each battery pack are recorded to provide a basis for data splitting.

[0033] Optionally, the collected original operation data can be preprocessed to remove duplicate data, missing data and obviously wrong data frames; for data frames with abnormal jumps in characteristics such as voltage and temperature, a sliding window mean filtering method can be used for smoothing processing.

[0034] Step S104: Based on the first data sequence, determine multiple first loss values, wherein each of the multiple first loss values ​​corresponds one-to-one with a multiple first data segment;

[0035] It is understandable that, based on the first data sequence, a first prediction model based on TimesNet (Times Series Transformer) is used to obtain multiple first loss values ​​that correspond one-to-one with multiple first data segments.

[0036] In one optional embodiment, determining multiple first loss values ​​based on a first data sequence includes: determining first target features corresponding to the multiple first data segments and first labels corresponding to the multiple first data segments; and obtaining multiple first loss values ​​by using a first prediction model and a first loss function corresponding to the first prediction model based on the first target features corresponding to the multiple first data segments and the first labels corresponding to the multiple first data segments.

[0037] It can be understood that by deriving multi-dimensional time-series features from the first data segment, multiple first derived features corresponding to each first data segment are obtained, such as voltage change rate, temperature difference change rate, current accumulation value, SOC change rate, and the mean / variance of voltage / temperature. Combined with the operational data of the first data segment, first target features corresponding to each first data segment are constructed. Simultaneously, the minimum voltage and maximum temperature of the battery pack corresponding to multiple data points in each first data segment are determined as the first label for that first data segment. After normalizing the first target features corresponding to each of the multiple first data segments, they are input into a first prediction model based on TimesNet, resulting in multiple first prediction results corresponding one-to-one with each of the multiple first data segments, including predicted minimum voltage and predicted maximum temperature. Based on the multiple first prediction results and the first labels corresponding to each of the multiple first data segments, multiple first loss values ​​corresponding one-to-one with each of the multiple first data segments are calculated using the first loss function corresponding to the first prediction model. By constructing a first prediction model based on TimesNet and combining it with a first loss function, we can accurately capture the long-term degradation trend and periodic patterns of battery pack long-term operation data, laying the foundation for the subsequent determination of thermal runaway risk level.

[0038] Optionally, the directional modeling of the first prediction model based on TimesNet and the construction of the first loss function can be as follows:

[0039] Input settings: Only the first target features corresponding to the first data sequence that meets the long sequence segmentation rules are input into the TimesNet model (i.e., the first prediction model based on TimesNet) for targeted modeling;

[0040] Model Construction: The TimesNet model is constructed based on the Generalized Trend Seasonal Decomposition (GTS) mechanism, specifically as follows:

[0041] Periodic embedding: The inherent periodicity of the first target feature is extracted using Fast Fourier Transform (FFT) to capture the cyclic characteristics of the battery pack under different seasons and different charge and discharge cycles;

[0042] Trend reinforcement: During backpropagation, the gradient of the trend term is amplified to ensure that the model can accurately capture the long-term degradation trend of the battery pack's State of Health (SOH).

[0043] Loss Function and Training: The first loss function of the TimesNet model is constructed using Trend-Weighted Mean Squared Error (TW-MSE), specifically as follows:

[0044] A smooth trend mask is designed. This mask automatically reduces the weight of points with sharp fluctuations by calculating the variance of the first derivative of the predicted values ​​at adjacent time steps, thereby improving the ability to capture long-term stable degradation features. The first loss function formula is: ,in, This is a trend mask matrix; the greater the fluctuation, the lower the mask value. Based on the first tag vector, including minimum voltage and maximum temperature, The first prediction result vector includes the predicted minimum voltage and the predicted maximum temperature. This represents the number of data points included in the first data segment.

[0045] The training set is constructed using historical first target features (i.e., historical first data sequence) based on historical long-sequence data of normal battery pack operating conditions. The AdamW optimizer is used for training, and a learning rate decay strategy is set to prevent overfitting.

[0046] Output results: After training, predict the battery packs that meet the long sequence segmentation rules and output the TimesNet first loss value lossT corresponding to batteryCode, forming the TimesNet directional first loss value dataset (i.e., a dataset composed of multiple first loss values, with batteryCode as the primary key).

[0047] Step S106: Based on the second data sequence, determine multiple second loss values, wherein each of the multiple second loss values ​​corresponds one-to-one with a multiple second data segment;

[0048] It is understandable that, based on the second data sequence, a second prediction model based on TCN (Temporal Convolutional Network) is used to obtain multiple second loss values ​​that correspond one-to-one with multiple second data segments.

[0049] In one optional embodiment, determining multiple second loss values ​​based on a second data sequence includes: determining second target features corresponding to each of the multiple second data segments and second labels corresponding to each of the multiple second data segments; and obtaining multiple second loss values ​​by using a second prediction model and a second loss function corresponding to the second prediction model based on the second target features corresponding to each of the multiple second data segments and the second labels corresponding to each of the multiple second data segments.

[0050] It is understandable that by deriving multi-dimensional time-series features from the second data segment, multiple second-derived features corresponding to each second data segment are obtained, such as voltage change rate, temperature difference change rate, current accumulation value, SOC change rate, and the mean / variance of voltage / temperature. Combined with the operational data of the second data segment, second target features corresponding to each second data segment are constructed. Simultaneously, the minimum voltage and maximum temperature of the battery pack corresponding to multiple data points in each second data segment are determined as the second label for that second data segment. After normalizing the second target features corresponding to the multiple second data segments, they are input into the TCN-based second prediction model, resulting in multiple second prediction results corresponding one-to-one with the multiple second data segments, including predicted minimum voltage and predicted maximum temperature. Based on the multiple second prediction results and the second labels corresponding to the multiple second data segments, multiple second loss values ​​corresponding one-to-one with the multiple second data segments are calculated using the second loss function corresponding to the second prediction model. By constructing the TCN-based second prediction model and combining it with the second loss function, accurate capture of local burst features in the short-time-series data of the battery pack is achieved, laying the foundation for subsequent determination of the thermal runaway risk level.

[0051] Optionally, the same derivation rules are applied to the first data segment and the second data segment to obtain derived features, so as to ensure the consistency of the feature dimensions of the first target feature and the second target feature.

[0052] Optionally, the generated first and second target features can be normalized respectively. Simultaneously, correlation analysis and variance screening can be used for feature dimensionality reduction to eliminate redundant features with low correlation to battery thermal runaway and variances approaching zero. Since the prediction targets of both the first and second prediction models are the minimum voltage and maximum temperature of the battery pack, first labels for the long sequence data (i.e., the first data sequence) and second labels for the short sequence data (i.e., the second data sequence) can be constructed based on the minimum voltage and maximum temperature data from the long and short sequence data, respectively, for use in the subsequent calculation of the first and second loss values.

[0053] Optionally, the modeling of the second prediction model based on TCN and the construction of the second loss function can be as follows:

[0054] Input settings: Input the second target features corresponding to the second data sequence of all battery packs after preprocessing into the TCN model (i.e., the second prediction model based on TCN) to achieve full coverage processing of TCN.

[0055] Model Construction: Construct a TCN model with dilated convolution and residual connections, specifically as follows:

[0056] To address the characteristics of short sequence data, an initial hole rate of 1 is set and increased exponentially with the number of network layers, in order to maximize the receptive field within a limited short sequence length and capture long-distance correlations in local time series data.

[0057] Loss Function and Training: Design a mutation-awareness and physical mechanism-weighted loss function (Mutual AwarenessMean Absolute Error with Adaptive Residual Reweighting, MA-MAE-ARR) (i.e., the second loss function), specifically as follows:

[0058] A temperature-sensitive factor is introduced, where the chemical reaction rate doubles for every 10°C increase in temperature. Therefore, an exponential penalty weight is introduced into the prediction of the maximum temperature error term. The specific formula for the second loss function is as follows: ,in, The reference temperature is represented by λ, and the thermal sensitivity coefficient is represented by λ. The number of data points included in the second data segment. The maximum temperature in the second label. The predicted maximum temperature in the second prediction result. This is the voltage error weighting coefficient. The minimum voltage in the second label. This is the predicted minimum voltage in the second prediction result.

[0059] The training set is constructed using historical second target features (i.e., historical second data sequences) based on historical short-sequence data of normal battery pack operating conditions. The Adam optimizer is used for training, and an early stopping mechanism is introduced to prevent overfitting.

[0060] Output results: After training, predictions are made for all battery packs that meet the short sequence segmentation rules, and the TCN second loss value lossC corresponding to each batteryCode is output, forming the TCN full second loss value dataset (i.e., a dataset composed of multiple second loss values, with batteryCode as the primary key).

[0061] Step S108: Based on multiple first loss values ​​and multiple second loss values, determine multiple comprehensive loss values, wherein each of the multiple comprehensive loss values ​​corresponds one-to-one with a multiple second data segment;

[0062] It is understandable that multiple first loss values ​​and multiple second loss values ​​are fused to obtain multiple comprehensive loss values ​​that correspond one-to-one with multiple second data segments. By fusing the first loss values ​​and the second loss values, a collaborative assessment of battery pack state anomalies at different time scales is achieved, improving the accuracy of the determination of the battery pack thermal runaway risk level.

[0063] In one optional embodiment, determining multiple comprehensive loss values ​​based on multiple first loss values ​​and multiple second loss values ​​includes: aligning and supplementing multiple first loss values ​​and multiple second loss values ​​with multiple second loss values ​​as a benchmark to obtain multiple loss value results, wherein each loss value result corresponds one-to-one with multiple second data segments, and each loss value result includes a second loss value and an aligned loss value, wherein the aligned loss value is a first loss value or a preset value; determining multiple gating factors based on the second target features corresponding to the multiple second data segments respectively, wherein each gating factor corresponds one-to-one with the multiple second data segments; obtaining multiple first weights based on the multiple gating factors and multiple aligned loss values, wherein the multiple aligned loss values ​​are determined based on the multiple loss value results, and the multiple first weights correspond one-to-one with the multiple aligned loss values; obtaining multiple second weights based on the multiple gating factors and multiple second loss values, wherein the multiple second weights correspond one-to-one with the multiple second loss values; and obtaining multiple comprehensive loss values ​​based on the multiple aligned loss values, the multiple first weights corresponding to the multiple aligned loss values, the multiple second loss values, and the multiple second weights corresponding to the multiple second loss values.

[0064] It is understandable that, since the number of first data segments is less than the number of second data segments, in order to subsequently fuse multiple first and second loss values, it is necessary to first align and supplement them. Using multiple second loss values ​​as a benchmark, the multiple first and second loss values ​​are aligned and supplemented to obtain multiple loss value results. Alignment refers to aligning the timestamps corresponding to the first data segments and the timestamps corresponding to the second data segments in terms of time; supplementation refers to supplementing the first loss values ​​of second data segments that cannot be aligned with the first data segments, thereby obtaining multiple loss value results that correspond one-to-one with the multiple second data segments. For example, if the first second data segment in the second data sequence cannot find a first data segment to align with, then the corresponding second loss value also cannot find a corresponding first loss value. In this case, the first loss value in the loss value result corresponding to the second data segment is filled with a preset value (e.g., 0) to facilitate subsequent loss value fusion. Multiple gating factors are determined based on the voltage change rate and temperature difference change rate in the second target features corresponding to the multiple second data segments. Multiple first weights are obtained based on multiple gating factors and multiple alignment loss values. Simultaneously, multiple second weights are obtained based on multiple gating factors and multiple second loss values. Finally, multiple comprehensive loss values ​​are obtained by weighted summation of the alignment loss values, the first weights, the second loss values, and the second weights. By introducing a dynamic gating mechanism based on voltage and temperature change rates, and adaptively adjusting the fusion weights of the loss values, the fusion problem after aligning data segments at different time scales is solved, improving the accuracy of the thermal runaway risk level determination results.

[0065] Optionally, the TimesNet oriented first loss value dataset can be left-connected based on the full TCN second loss value dataset to align multiple first loss values ​​and multiple second loss values. At the same time, missing first loss values ​​can be supplemented according to a preset value (e.g., 0).

[0066] Optionally, a gating function G is introduced in the weight calculation, and the calculation formula is as follows: ,in, For the Sigmoid function, and For adjustment coefficients, The voltage change rate is the second target feature. The temperature change rate is the second objective feature. As the voltage / temperature change rate increases, the gating factor G approaches 1.

[0067] Optionally, the fusion formula is:

[0068] in, This is the fusion loss value (i.e., the overall loss value). The first weight is determined based on the gating factor G and the alignment loss value (if a first loss value exists, this value is the first loss value; otherwise, this value is a preset value of 0). This leads to the final result = ) Definitely. The second weight is determined based on the gating factor G and the second loss value. Yes, that's for sure. It can also be set based on experience. =0.6, =0.4.

[0069] Step S110: Determine the thermal runaway risk level of the battery pack based on multiple comprehensive loss values.

[0070] It is understandable that by determining the thermal runaway risk level of the battery pack based on the comprehensive loss value, thermal runaway early warning of the battery pack can be achieved based on the long-term and short-term performance change trends of the battery pack, thereby improving the rationality and accuracy of the thermal runaway risk level determination results.

[0071] In one optional embodiment, determining the thermal runaway risk level of the battery pack based on multiple comprehensive loss values ​​includes: dividing the multiple comprehensive loss values ​​using a sliding time window method to obtain multiple subsets of loss values ​​corresponding to each local time window; for any local time window, determining a third target feature of any local time window based on multiple first comprehensive loss values ​​included in the subset of loss values ​​of any local time window; obtaining third target features corresponding to each local time window by using the method of determining the third target feature of any local time window; and determining the thermal runaway risk level based on the third target features corresponding to each local time window and the multiple comprehensive loss values.

[0072] It is understandable that a sliding time window approach is used to divide multiple comprehensive loss values, resulting in multiple subsets of loss values ​​corresponding to local time windows. For any local time window, based on the multiple first comprehensive loss values ​​included in the subset of loss values ​​of that local time window, a third target feature is determined, such as mean, slope, volatility, and duration of rise. By determining the third target feature of any local time window, the third target features corresponding to multiple local time windows are obtained. Combined with multiple comprehensive loss values, the thermal runaway risk level of the battery pack is determined. By introducing a sliding window to perform time-series trend analysis on the comprehensive loss values ​​and extracting features such as mean and slope to characterize the evolution of the battery pack state, the error caused by single-moment judgment is effectively reduced, improving the accuracy and stability of the thermal runaway risk level determination results.

[0073] In one optional embodiment, the thermal runaway risk level is determined based on the third target features corresponding to multiple local time windows and multiple comprehensive loss values, including: determining a target comprehensive loss value based on multiple comprehensive loss values, wherein the target comprehensive loss value is the comprehensive loss value corresponding to the last second data segment in the second data sequence; determining the thermal runaway trend of the battery pack within a preset time period based on the third target features corresponding to multiple local time windows; and determining the thermal runaway risk level based on the target comprehensive loss value and the thermal runaway trend.

[0074] It is understandable that the comprehensive loss value corresponding to the last second data segment in the second data sequence is determined as the target comprehensive loss value. Based on the third target characteristics corresponding to multiple local time windows, the thermal runaway trend of the battery pack within a preset time period is determined, such as stable, slowly rising, rapidly rising, or steeply rising. Based on the target comprehensive loss value and the thermal runaway trend, the thermal runaway risk level of the battery pack is comprehensively determined. By combining the comprehensive loss value of the last second data segment (reflecting the degree of instantaneous anomaly) with the thermal runaway trend within the preset time period (reflecting the state evolution pattern), it is possible to effectively distinguish between instantaneous disturbances and actual thermal runaway precursors, improving the accuracy and rationality of the thermal runaway risk level determination results.

[0075] Alternatively, a sliding time window method combined with piecewise polynomial fitting can be used to evaluate the battery pack. Trend modeling is performed on the time series data, with the size of the local time window set according to the battery pack application scenario. Core features such as the mean, slope, volatility, and duration of the rise in the comprehensive loss value within each local time window are extracted to obtain the third target feature for each local time window. Based on the third target features corresponding to multiple local time windows, the thermal runaway trend of the battery pack is determined and divided into four levels: stable, slowly rising, rapidly rising, and steeply rising.

[0076] In one optional embodiment, when the thermal runaway trend includes a stable, slowly rising, rapidly rising, and steeply rising state, the thermal runaway risk level is determined based on the target comprehensive loss value and the thermal runaway trend, including: if the target comprehensive loss value is greater than a preset thermal runaway threshold and the thermal runaway trend is either rapidly rising or steeply rising, the thermal runaway risk level is determined to be high risk; if the target comprehensive loss value is greater than the preset thermal runaway threshold and the thermal runaway trend is either stable or slowly rising, the thermal runaway risk level is determined to be low risk; if the target comprehensive loss value is less than or equal to the preset thermal runaway threshold and the thermal runaway trend is either rapidly rising or steeply rising, the thermal runaway risk level is determined to be low risk; if the target comprehensive loss value is less than or equal to the preset thermal runaway threshold and the thermal runaway trend is either stable or slowly rising, the thermal runaway risk level is determined to be no risk.

[0077] It is understandable that thermal runaway trends can range from steady, slow, rapid, to steep. If the target comprehensive loss value is greater than the preset thermal runaway threshold, and the thermal runaway trend is either rapidly increasing or steeply increasing, it indicates that the battery pack is in a severely abnormal state and deteriorating at an extremely rapid rate, with a very high risk of thermal runaway. The thermal runaway risk level of the battery pack is determined to be high risk. If the target comprehensive loss value is greater than the preset thermal runaway threshold, and the thermal runaway trend is either stable or slowly increasing, it indicates that the battery pack is abnormal but its state is relatively stable or deteriorating slowly, posing a potential risk but not an emergency. The thermal runaway risk level of the battery pack is determined to be low risk. If the target comprehensive loss value is less than or equal to the preset thermal runaway threshold, and the thermal runaway trend is either rapidly increasing or steeply increasing, it indicates that the battery pack appears to be in a normal state but is deteriorating rapidly, posing a delayed or sudden risk. The thermal runaway risk level of the battery pack is determined to be low risk. If the target comprehensive loss value is less than or equal to the preset thermal runaway threshold, and the thermal runaway trend is either stable or slowly increasing, it indicates that the battery pack is in a normal state and its long-term evolution is stable, with no obvious abnormal characteristics detected. The thermal runaway risk level of the battery pack is determined to be no risk. By constructing a dual-dimensional risk assessment mechanism that combines "instantaneous anomaly level" and "long-term evolution trend", the shortcomings of single threshold early warning are easily affected by instantaneous fluctuations, and the accurate determination and early identification of the risk level of battery pack thermal runaway are achieved.

[0078] Alternatively, " The dual judgment logic of "current value (i.e., target comprehensive loss value) + long-term thermal runaway trend" enables accurate classification of thermal runaway risk. The specific rules are as follows:

[0079] High risk: >Th1 (preset thermal runaway threshold) and the long-term thermal runaway trend level is rapidly increasing / sharply increasing;

[0080] Low risk: >Th1 or long-term thermal runaway trend level is rapidly rising / sharply rising (only one of them is met).

[0081] No risk: ≤Th1 and the long-term thermal runaway trend level is stable / slowly rising.

[0082] Through the above steps S102 to S110, the goal of obtaining multiple comprehensive loss values ​​by fusing multiple first loss values ​​and multiple second loss values ​​determined based on the battery pack's operating data can be achieved, thereby determining the thermal runaway risk level of the battery pack. This achieves the technical effect of improving the accuracy of the determination result of the thermal runaway risk level of the battery pack, and solves the technical problem of inaccurate determination result of the thermal runaway risk level of the battery pack in related technologies.

[0083] Based on the above embodiments and optional embodiments, this application proposes an implementation method for determining the thermal runaway risk level of a battery pack. This optional implementation method can be understood as a lithium battery pack thermal runaway early warning method based on the fusion of TimesNet and TCN, which aims to solve the technical problem that the uncontrollable online time of lithium battery packs leads to inconsistent time series lengths and insufficient early warning coverage of a single model, and achieve full coverage collaborative early warning of long / short series models.

[0084] To address the issue that single models have strict requirements on input sequence length and cannot adapt to all battery pack operating data, a lithium battery pack thermal runaway early warning method based on the fusion of TimesNet and TCN is proposed. This method involves doubly splitting the same segment of operating data, directing the long sequence (first data sequence) that meets the requirements into TimesNet modeling, and the short sequence (second data sequence) into TCN modeling. The results from both parts are then weighted to output a comprehensive loss value. Meanwhile, TimesNet is adapted to long time-series sequence processing, accurately capturing the long-term periodic patterns and state evolution trends of battery pack operating data. It has strong modeling capabilities for the long-term variation characteristics of core indicators such as voltage and temperature, and can uncover potential long-term precursors of thermal runaway, providing accurate judgment criteria for trend anomalies in battery pack status. TCN, on the other hand, has high adaptability to short time-series sequences, possessing strong local feature and sudden change capture capabilities, and can effectively extract sudden precursor features of thermal runaway such as voltage drops and temperature rises within a short period of time. The lithium battery pack thermal runaway early warning method based on the fusion of TimesNet and TCN leverages TimesNet's advantage in capturing trends in long sequences and TCN's strong adaptability to short sequences to enhance early warning. It breaks through many technical limitations of traditional single algorithms and achieves full coverage and high-precision thermal runaway early warning for battery packs.

[0085] Figure 2 This is a flowchart of an optional method for determining the thermal runaway risk level of a battery pack according to an embodiment of this application, such as... Figure 2 As shown, the steps of the lithium battery pack thermal runaway early warning method based on the fusion of TimesNet and TCN include:

[0086] Step S1: Collect and clean battery pack operation data, and establish long / short sequence dual segmentation rules.

[0087] Step S11, data acquisition.

[0088] The BMS collects real-time operational data from the lithium battery packs and stores it in the cloud platform database. The collected operational data includes, but is not limited to, minimum voltage, maximum temperature, state of charge (SOC), and charge / discharge current. It also records the actual online time of each battery pack and the timestamp of the original operational data, providing a basis for data segmentation.

[0089] Step S12: data cleaning.

[0090] Preprocess the collected original operation data, remove duplicate data, missing data and obviously wrong data frames; perform smoothing processing on data frames with abnormal jumps of features such as voltage and temperature by adopting a sliding window mean filtering method.

[0091] Step S13: establishment of long / short sequence dual-splitting rules.

[0092] According to the rigid length requirement of the TimesNet model for input sequences, a long sequence splitting rule (i.e., the first splitting strategy) is set: the time span of the data sequence of the first data segment is ≥ TL, the sampling step of the first data segment is tL, and the number of data points in the data sequence of the first data segment is ≥ NL. A sequence meeting this rule is a valid long data segment sequence that can be input into the TimesNet model;

[0093] According to the adaptation characteristic of the TCN model for short sequences, a short sequence splitting rule (i.e., the second splitting strategy) is set: the time span of the data sequence of the second data segment is ≥ TS (TS < TL), the sampling step of the second data segment is tS (tS ≤ tL), and the number of data points in the data sequence of the second data segment is ≥ NS (NS < NL). A sequence meeting this rule is a valid short data segment sequence that can be input into the TCN model;

[0094] Specify the requirements for dual-splitting for the same battery pack: for the original operation data of each battery pack, perform splitting respectively according to the above long and short sequence splitting rules at the same time, generate corresponding long sequence data (i.e., the first data sequence) for those meeting the long sequence rule, and generate corresponding short sequence data (i.e., the second data sequence) for those meeting the short sequence rule, so as to achieve the splitting effect of "one source, two sequences".

[0095] Step S2: multi-dimensional feature derivation and preprocessing.

[0096] Step S21: feature derivation.

[0097] Perform multi-dimensional temporal feature derivation based on the split operation data, including but not limited to voltage change rate, temperature difference change rate, current cumulative value, SOC change rate, sliding window mean / variance of voltage / temperature, etc., to obtain a first derived feature corresponding to the first data sequence and a second derived feature corresponding to the second data sequence. By combining the operation data of the first data sequence and the second data sequence, a first target feature corresponding to the first data sequence and a second target feature corresponding to the second data sequence are constructed. Wherein, the long and short sequence data adopt the same derivation rule to ensure the consistency of feature dimensions.

[0098] Step S22: feature preprocessing.

[0099] The generated first and second target features are normalized respectively; correlation analysis and variance screening are used to reduce the dimensionality of the features and remove redundant features with low correlation to battery thermal runaway and variance close to 0; the prediction targets of the model are the minimum voltage and maximum temperature of the battery pack. Therefore, based on the minimum voltage and maximum temperature data in the long and short sequence data, the first label of the long sequence data and the second label of the short sequence data are constructed respectively for the subsequent calculation of the first and second loss values.

[0100] Step S3: TCN full coverage modeling + TimesNet targeted modeling.

[0101] Step S31: Oriented modeling and loss value output of the first prediction model based on TimesNet.

[0102] Input settings: Only the first target features corresponding to the first data sequence that meets the long sequence segmentation rules are input into the TimesNet model (i.e., the first prediction model based on TimesNet) for targeted modeling;

[0103] Model Construction: The TimesNet model is constructed based on the Generalized Trend Seasonal Decomposition (GTS) mechanism, specifically as follows:

[0104] Periodic embedding: The inherent periodicity of the first target feature is extracted using Fast Fourier Transform (FFT) to capture the cyclic characteristics of the battery pack under different seasons and different charge and discharge cycles;

[0105] Trend reinforcement: During backpropagation, the gradient of the trend term is amplified to ensure that the model can accurately capture the long-term degradation trend of the battery pack's State of Health (SOH).

[0106] Loss Function and Training: The first loss function of the TimesNet model is constructed using Trend-Weighted Mean Squared Error (TW-MSE), specifically as follows:

[0107] A smooth trend mask is designed. This mask automatically reduces the weight of points with sharp fluctuations by calculating the variance of the first derivative of the predicted values ​​at adjacent time steps, thereby improving the ability to capture long-term stable degradation features. The first loss function formula is: ,in, This is a trend mask matrix; the greater the fluctuation, the lower the mask value. Based on the first tag vector, including minimum voltage and maximum temperature, The first prediction result vector includes the predicted minimum voltage and the predicted maximum temperature. This represents the number of data points included in the first data segment.

[0108] The training set is constructed using historical first target features (i.e., historical first data sequence) based on historical long-sequence data of normal battery pack operating conditions. The AdamW optimizer is used for training, and a learning rate decay strategy is set to prevent overfitting.

[0109] Output results: After training, predict the battery packs that meet the long sequence segmentation rules and output the TimesNet first loss value lossT corresponding to batteryCode, forming the TimesNet directional first loss value dataset (i.e., a dataset composed of multiple first loss values, with batteryCode as the primary key).

[0110] Step S32: Modeling and outputting the loss value of the second prediction model based on TCN.

[0111] Input settings: Input the second target features corresponding to the second data sequence of all battery packs after preprocessing into the TCN model (i.e., the second prediction model based on TCN) to achieve full coverage processing of TCN.

[0112] Model Construction: Construct a TCN model with dilated convolution and residual connections, specifically as follows:

[0113] To address the characteristics of short sequence data, an initial hole rate of 1 is set and increased exponentially with the number of network layers, in order to maximize the receptive field within a limited short sequence length and capture long-distance correlations in local time series data.

[0114] Loss Function and Training: Design a mutation-awareness and physical mechanism-weighted loss function (Mutual AwarenessMean Absolute Error with Adaptive Residual Reweighting, MA-MAE-ARR) (i.e., the second loss function), specifically as follows:

[0115] A temperature-sensitive factor is introduced, where the chemical reaction rate doubles for every 10°C increase in temperature. Therefore, an exponential penalty weight is introduced into the prediction of the maximum temperature error term. The specific formula for the second loss function is as follows: ,in, The reference temperature is represented by λ, and the thermal sensitivity coefficient is represented by λ. The number of data points included in the second data segment. The maximum temperature in the second label. The predicted maximum temperature in the second prediction result. This is the voltage error weighting coefficient. The minimum voltage in the second label. This is the predicted minimum voltage in the second prediction result.

[0116] The training set is constructed using historical second target features (i.e., historical second data sequences) based on historical short-sequence data of normal battery pack operating conditions. The Adam optimizer is used for training, and an early stopping mechanism is introduced to prevent overfitting.

[0117] Output results: After training, predictions are made for all battery packs that meet the short sequence segmentation rules, and the TCN second loss value lossC corresponding to each batteryCode is output, forming the TCN full second loss value dataset (i.e., a dataset composed of multiple second loss values, with batteryCode as the primary key).

[0118] Step S4: Fusion of full battery pack loss values ​​based on dynamic gating mechanism.

[0119] Step S41: Align and stitch the dual-path loss values.

[0120] Using the full TCN second loss value dataset as a baseline, a left connection is performed on the TimesNet directed first loss value dataset.

[0121] Step S42, dynamic weight calculation.

[0122] The weight calculation introduces a gating function G, and the calculation formula is as follows: ,in, For the Sigmoid function, and For adjustment coefficients, The voltage change rate is the second target characteristic. The temperature change rate is the second objective feature. As the voltage / temperature change rate increases, the gating factor G approaches 1.

[0123] Step S43: Calculate and output the final comprehensive loss value.

[0124] The fusion formula is:

[0125] in, This is the fusion loss value (i.e., the overall loss value). The first weight is determined based on the gating factor G and the alignment loss value (if a first loss value exists, this value is the first loss value; otherwise, this value is a preset value of 0). This leads to the final result = ) Definitely. The second weight is determined based on the gating factor G and the second loss value. Yes, that's for sure. It can also be set based on experience. =0.6, =0.4.

[0126] Missing value handling: For battery packs with only short-sequence data containing lossC (e.g., the total runtime of the battery pack is less than TL, making it impossible to generate long-sequence data), output directly. = And marked as "TCN single-modal mode"; for battery packs with dual-model outputs, the output is dynamically gated and weighted. .

[0127] Output: Output the final overall loss value corresponding to the full batteryCode. and the dynamic weight parameters at that moment. and This results in a dataset of comprehensive loss values ​​for the entire battery pack that possesses adaptive capabilities.

[0128] Step S5: Long-term fusion loss value trend analysis.

[0129] The sliding time window method combined with piecewise polynomial fitting was used to study the battery pack. Trend modeling is performed on the time series data, with the size of the local time window set according to the battery pack application scenario. Core features such as the mean, slope, volatility, and duration of the rise in the comprehensive loss value within each local time window are extracted to obtain the third target feature for each local time window. Based on the third target features corresponding to multiple local time windows, the thermal runaway trend of the battery pack is determined and divided into four levels: stable, slowly rising, rapidly rising, and steeply rising.

[0130] Step S6: Determine the risk of thermal runaway and output the battery pack list.

[0131] Step S61: Set the dual-layer early warning threshold.

[0132] Based on long-term trend analysis of historical thermal runaway data of lithium battery packs accumulated on the cloud platform, a two-layer early warning threshold for thermal runaway risk assessment is set, both based on batteryCode. The core criterion for judgment:

[0133] Single-value threshold (i.e., preset thermal runaway threshold) Th1: The absolute value threshold of the fusion loss value, reflecting the degree of abnormality of the current state of the battery pack;

[0134] Trend Level Threshold Th2: The level threshold of the long-term thermal runaway trend of the comprehensive loss value, reflecting the abnormal evolution trend of the battery pack state.

[0135] Step S62, Dual Risk Assessment Rule.

[0136] use" The dual judgment logic of "current value (i.e., target comprehensive loss value) + long-term thermal runaway trend" enables accurate classification of thermal runaway risk. The specific rules are as follows:

[0137] High risk: >Th1 and the long-term thermal runaway trend level is rapidly increasing / sharply increasing;

[0138] Low risk: >Th1 or long-term thermal runaway trend level is rapidly rising / sharply rising (only one of them is met).

[0139] No risk: ≤Th1 and the long-term thermal runaway trend level is stable / slowly rising.

[0140] Step S63: Output the risk list.

[0141] Based on the thermal runaway risk level assessment results, a list of lithium battery packs at risk of thermal runaway is generated. The list includes the battery code, state of health (SOH), and online duration. , , Core information includes long-term thermal runaway trend level, thermal runaway risk level, minimum voltage / maximum temperature values, and the first / second prediction results of the model.

[0142] Step S7: Contextualized self-learning update of multi-dimensional early warning thresholds.

[0143] Based on the continuously accumulated battery pack fault data, normal operating condition data, early warning result verification data, and after-sales maintenance data on the cloud platform, an incremental learning method is used to perform scenario-based self-learning updates on the dual-layer early warning thresholds Th1 and Th2, ensuring that the thresholds always adapt to actual operating conditions.

[0144] 1) Scenario Classification: The battery pack is classified in multiple dimensions according to the application scenario (passenger car / commercial vehicle / energy storage power station), season (high temperature / low temperature / normal temperature), and battery health status (SOH≥80% / 60%≤SOH<80% / SOH<60%), and a special warning threshold is set for different categories;

[0145] 2) Incremental update: When the amount of new valid data in a certain category reaches the set threshold, the threshold model is retrained based on statistical analysis and trend fitting, and the Th1 and Th2 of that category are updated.

[0146] 3) Failure data removal: Periodically remove historical failure data that is out of time range or has undergone significant changes in operating conditions to ensure the effectiveness and timeliness of threshold training data.

[0147] Figure 3 This is a schematic diagram of an optional first thermal runaway detection result provided according to an embodiment of this application, such as... Figure 3 As shown, the thermal runaway detection results of traditional models, thermal runaway detection results based on a single TCN, thermal runaway detection results based on a single TimesNet, and thermal runaway detection results of a lithium battery pack thermal runaway early warning method based on a fusion of TimesNet and TCN are compared. The detection results include precision, thermal runaway fault recall rate, and F1 score. Figure 3 It can be seen that the thermal runaway detection results of the lithium battery pack thermal runaway early warning method based on the fusion of TimesNet and TCN are significantly better than those of the other three methods.

[0148] Figure 4 This is an optional dynamic gating weight adaptive change curve provided according to an embodiment of this application, such as... Figure 4 As shown, the relationship between the TCN weights (i.e., the second weights) and the TimesNet weights (i.e., the first weights) and the state of charge of the battery pack is illustrated after the gating mechanism is adopted.

[0149] Figure 5 This is a schematic diagram of an optional second thermal runaway detection result provided according to an embodiment of this application, such as... Figure 5 As shown, the thermal runaway detection results of a traditional single model, a thermal runaway detection result based on fixed-weight dual-model fusion, and a thermal runaway early warning method for lithium battery packs based on the fusion of TimesNet and TCN (corresponding to "physical constraints + dynamic gating") are compared. The detection results include thermal runaway fault coverage and thermal runaway false alarm rate. Figure 5 It can be seen that the thermal runaway fault coverage of the lithium battery pack thermal runaway early warning method based on the fusion of TimesNet and TCN is higher than that of the other two methods, while the false alarm rate of thermal runaway is significantly lower than that of the other two methods.

[0150] based on Figure 3 and Figure 5 It can be seen that the TCN and TimesNet fusion architecture constructed by the above optional implementation methods can simultaneously capture short sequence mutation features and long sequence trend features, with a detection F1 value of 93.6%; relying on physical sensing loss and dynamic gating mechanism, the thermal runaway fault coverage rate is increased to 92.3%, and the thermal runaway false alarm rate is controlled within 8.7%, with the overall effect being far superior to the existing technical solutions.

[0151] The above optional implementation methods achieve at least the following effects:

[0152] (1) Achieving comprehensive decoupling and blind-zone-free coverage of long and short time series features, significantly improving detection accuracy under complex conditions: By constructing a dual-track modeling architecture of TCN and TimesNet, the technical challenge of traditional single models being unable to simultaneously address local mutations and long-term trends is effectively solved. Utilizing the causal convolution and dilation mechanism of TCN, local mutation features such as instantaneous and drastic temperature rises in short sequences are accurately captured; simultaneously, the ability of TimesNet to convert one-dimensional time series into two-dimensional tensors is utilized to deeply explore the multi-periodicity and periodic variation patterns in long sequences. This clearly defined architecture achieves seamless coverage across all scenarios, filling the warning blind spots of traditional long-period models under short-sequence conditions, and avoiding misjudgments of long-term performance degradation by short-sequence models.

[0153] (2) The deep integration of physical mechanism constraints and dynamic gating mechanism significantly reduces the false alarm rate and enhances the robustness of the model: It breaks through the limitations of traditional pure data-driven models and solves the problem of high false alarm rate under complex working conditions by introducing physical mechanisms and adaptive weight allocation. Physical mechanism constraints are introduced into the TCN loss function (second loss function) so that the model follows the physical laws of equipment operation (such as the physical correlation between voltage and temperature) during the training phase, thereby having extremely high sensitivity and recognition accuracy to anomalies such as sudden and drastic temperature rise. The designed dynamic gating mechanism can dynamically calculate and allocate the weight parameters (first weight and second weight) of the two models according to the voltage / temperature change rate. When the working conditions fluctuate drastically, it automatically favors the mutation detection of TCN and the trend prediction of TimesNet during the stable period. This adaptive fusion mechanism effectively avoids the failure risk of a single model in a specific scenario. After verification by production projects, the fault coverage rate can reach more than 90%, and the error rate of the full data is controlled within 10%.

[0154] (3) A closed-loop self-learning system of "perception-decision-evolution" was constructed, which solved the performance drift problem after long-term model operation: a scenario-based threshold self-learning update mechanism was established, which gave the early warning system the ability to continuously evolve and significantly improved its engineering applicability. The system can automatically identify the current working scenario based on the fusion results and the dominant state of the model, and update the early warning threshold in reverse. This closed-loop mechanism enables the model to adapt to the data distribution drift caused by equipment aging and environmental changes.

[0155] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0156] This embodiment also provides a device for determining the thermal runaway risk level of a battery pack. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that performs a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementations, or a combination of software and hardware, are also possible and contemplated.

[0157] According to an embodiment of this application, an apparatus embodiment for implementing a method for determining the thermal runaway risk level of a battery pack is also provided. Figure 6 This is a schematic diagram of a device for determining the thermal runaway risk level of a battery pack according to an embodiment of this application, as shown below. Figure 6 As shown, the above-mentioned device for determining the thermal runaway risk level of the battery pack includes a data acquisition module 602, a first determination module 604, a second determination module 606, a third determination module 608, and a fourth determination module 610. The device will be described below.

[0158] The data acquisition module 602 is used to acquire the operating data of the battery pack during a preset time period, and to segment the operating data according to two segmentation strategies to obtain a first data sequence and a second data sequence of the battery pack. The first data sequence includes multiple first data segments, and the second data sequence includes multiple second data segments. The first number of the multiple first data segments is less than the second number of the multiple second data segments.

[0159] The first determining module 604 is connected to the data acquisition module 602 and is used to determine multiple first loss values ​​based on the first data sequence, wherein the multiple first loss values ​​correspond one-to-one with multiple first data segments;

[0160] The second determining module 606 is connected to the first determining module 604 and is used to determine multiple second loss values ​​based on the second data sequence, wherein the multiple second loss values ​​correspond one-to-one with multiple second data segments;

[0161] The third determining module 608, connected to the second determining module 606, is used to determine multiple comprehensive loss values ​​based on multiple first loss values ​​and multiple second loss values, wherein the multiple comprehensive loss values ​​correspond one-to-one with multiple second data segments;

[0162] The fourth determination module 610, connected to the third determination module 608, is used to determine the thermal runaway risk level of the battery pack based on multiple comprehensive loss values.

[0163] The embodiment of this application provides a device for determining the thermal runaway risk level of a battery pack. By setting a data acquisition module 602, a first determination module 604, a second determination module 606, a third determination module 608, and a fourth determination module 610, the device achieves the purpose of determining the thermal runaway risk level of the battery pack by fusing multiple first loss values ​​and multiple second loss values ​​determined based on the operating data of the battery pack to obtain multiple comprehensive loss values. This improves the accuracy of the determination result of the thermal runaway risk level of the battery pack and solves the technical problem of inaccurate determination result of the thermal runaway risk level of the battery pack in related technologies.

[0164] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0165] It should be noted that the data acquisition module 602, the first determining module 604, the second determining module 606, the third determining module 608, and the fourth determining module 610 mentioned above correspond to steps S102 to S110 in the embodiments. The instances and application scenarios implemented by the above modules and their corresponding steps are the same, but they are not limited to the content disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can run in a computer terminal.

[0166] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0167] The aforementioned battery pack thermal runaway risk level determination device may further include a processor and a memory. The data acquisition module 602, the first determination module 604, the second determination module 606, the third determination module 608, the fourth determination module 610, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0168] The processor contains a core that retrieves the corresponding program unit from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0169] This application provides a non-volatile storage medium storing a program that, when executed by a processor, implements a method for determining the thermal runaway risk level of a battery pack.

[0170] This application provides an electronic device. Figure 7 This is a structural diagram of an electronic device provided according to an embodiment of this application. For example... Figure 7 As shown, the electronic device may include: one or more ( Figure 7 (Only one is shown) A processor 702, a memory 704, a memory controller, and a peripheral interface, wherein the peripheral interface is connected to an RF module, an audio module, and a display. The electronic device includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: acquiring operating data of the battery pack over a preset time period, and segmenting the operating data according to two segmentation strategies to obtain a first data sequence and a second data sequence of the battery pack, wherein the first data sequence includes multiple first data segments, and the second data sequence includes multiple second data segments, the first number of the multiple first data segments being less than the second number of the multiple second data segments; determining multiple first loss values ​​based on the first data sequence, wherein each of the multiple first loss values ​​corresponds one-to-one with each of the multiple first data segments; determining multiple second loss values ​​based on the second data sequence, wherein each of the multiple second loss values ​​corresponds one-to-one with each of the multiple second data segments; determining multiple comprehensive loss values ​​based on the multiple first loss values ​​and the multiple second loss values, wherein each of the multiple comprehensive loss values ​​corresponds one-to-one with each of the multiple second data segments; and determining the thermal runaway risk level of the battery pack based on the multiple comprehensive loss values. The device in this document may be a server, PC, etc.

[0171] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program with the following method steps: acquiring operating data of a battery pack over a preset time period, and segmenting the operating data according to two segmentation strategies to obtain a first data sequence and a second data sequence of the battery pack, wherein the first data sequence includes multiple first data segments, the second data sequence includes multiple second data segments, and the first number of the multiple first data segments is less than the second number of the multiple second data segments; determining multiple first loss values ​​based on the first data sequence, wherein the multiple first loss values ​​correspond one-to-one with the multiple first data segments; determining multiple second loss values ​​based on the second data sequence, wherein the multiple second loss values ​​correspond one-to-one with the multiple second data segments; determining multiple comprehensive loss values ​​based on the multiple first loss values ​​and the multiple second loss values, wherein the multiple comprehensive loss values ​​correspond one-to-one with the multiple second data segments; and determining the thermal runaway risk level of the battery pack based on the multiple comprehensive loss values.

[0172] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0173] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0174] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0175] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0176] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0177] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0178] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0179] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0180] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0181] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for determining the thermal runaway risk level of a battery pack, characterized in that, include: The battery pack's operating data during a preset time period is acquired, and the operating data is segmented according to two segmentation strategies to obtain a first data sequence and a second data sequence of the battery pack. The first data sequence includes multiple first data segments, and the second data sequence includes multiple second data segments. The first number of the multiple first data segments is less than the second number of the multiple second data segments. Based on the first data sequence, a plurality of first loss values ​​are determined, wherein the plurality of first loss values ​​correspond one-to-one with the plurality of first data segments; Based on the second data sequence, a plurality of second loss values ​​are determined, wherein the plurality of second loss values ​​correspond one-to-one with the plurality of second data segments; Based on the plurality of first loss values ​​and the plurality of second loss values, a plurality of comprehensive loss values ​​are determined, wherein the plurality of comprehensive loss values ​​correspond one-to-one with the plurality of second data segments; Based on the multiple comprehensive loss values, the thermal runaway risk level of the battery pack is determined.

2. The method according to claim 1, characterized in that, The determination of multiple first loss values ​​based on the first data sequence includes: Based on the plurality of first data segments, determine the first target features corresponding to the plurality of first data segments respectively, and the first labels corresponding to the plurality of first data segments respectively; Based on the first target features corresponding to the plurality of first data segments and the first labels corresponding to the plurality of first data segments, the plurality of first loss values ​​are obtained by using a first prediction model and a first loss function corresponding to the first prediction model.

3. The method according to claim 1, characterized in that, The determination of multiple second loss values ​​based on the second data sequence includes: Based on the plurality of second data segments, determine the second target features corresponding to the plurality of second data segments respectively, and the second labels corresponding to the plurality of second data segments respectively; Based on the second target features corresponding to the plurality of second data segments and the second labels corresponding to the plurality of second data segments, the plurality of second loss values ​​are obtained by using the second prediction model and the second loss function corresponding to the second prediction model.

4. The method according to claim 1, characterized in that, The determination of multiple comprehensive loss values ​​based on the multiple first loss values ​​and the multiple second loss values ​​includes: Based on the plurality of second loss values, the plurality of first loss values ​​and the plurality of second loss values ​​are aligned and supplemented to obtain a plurality of loss value results, wherein the plurality of loss value results correspond one-to-one with the plurality of second data segments, and each loss value result includes a second loss value and an alignment loss value, wherein the alignment loss value is a first loss value or a preset value; Based on the second target features corresponding to the multiple second data segments respectively, multiple gate control factors are determined, wherein the multiple gate control factors correspond one-to-one with the multiple second data segments; Based on the multiple gating factors and multiple alignment loss values, multiple first weights are obtained, wherein the multiple alignment loss values ​​are determined based on the results of the multiple loss values, and the multiple first weights correspond one-to-one with the multiple alignment loss values; Based on the plurality of gating factors and the plurality of second loss values, a plurality of second weights are obtained, wherein the plurality of second weights correspond one-to-one with the plurality of second loss values; Based on the plurality of alignment loss values, the plurality of first weights corresponding to the plurality of alignment loss values, the plurality of second loss values, and the plurality of second weights corresponding to the plurality of second loss values, the plurality of comprehensive loss values ​​are obtained.

5. The method according to any one of claims 1 to 4, characterized in that, The determination of the thermal runaway risk level of the battery pack based on the multiple comprehensive loss values ​​includes: The multiple comprehensive loss values ​​are divided using a sliding time window approach to obtain multiple subsets of loss values ​​corresponding to local time windows; For any local time window among multiple local time windows, the third target feature of the local time window is determined based on multiple first comprehensive loss values ​​included in the loss value subset of the local time window. The third target features corresponding to each of the plurality of local time windows are obtained by determining the third target features of any local time window. The thermal runaway risk level is determined based on the third target features corresponding to the multiple local time windows and the multiple comprehensive loss values.

6. The method according to claim 5, characterized in that, The determination of the thermal runaway risk level based on the third target features corresponding to the multiple local time windows and the multiple comprehensive loss values ​​includes: Based on the multiple comprehensive loss values, a target comprehensive loss value is determined, wherein the target comprehensive loss value is the comprehensive loss value corresponding to the last second data segment in the second data sequence; Based on the third target features corresponding to the multiple local time windows, the thermal runaway trend of the battery pack in the preset time period is determined; Based on the target comprehensive loss value and the thermal runaway trend, the thermal runaway risk level is determined.

7. The method according to claim 6, characterized in that, When the thermal runaway trend includes a stable, slowly rising, rapidly rising, and steeply rising pattern, determining the thermal runaway risk level based on the target comprehensive loss value and the thermal runaway trend includes: If the target comprehensive loss value is greater than the preset thermal runaway threshold, and the thermal runaway trend is either rapidly increasing or steeply increasing, the thermal runaway risk level is determined to be high risk. If the target comprehensive loss value is greater than the preset thermal runaway threshold, and the thermal runaway trend is either stable or slowly rising, the thermal runaway risk level is determined to be low risk. If the target comprehensive loss value is less than or equal to the preset thermal runaway threshold, and the thermal runaway trend is either rapidly increasing or steeply increasing, the thermal runaway risk level is determined to be low risk. If the target comprehensive loss value is less than or equal to the preset thermal runaway threshold, and the thermal runaway trend is either stable or slowly increasing, the thermal runaway risk level is determined to be no risk.

8. A device for determining the thermal runaway risk level of a battery pack, characterized in that, include: The data acquisition module is used to acquire the operating data of the battery pack during a preset time period, and to segment the operating data according to two segmentation strategies to obtain a first data sequence and a second data sequence of the battery pack. The first data sequence includes multiple first data segments, and the second data sequence includes multiple second data segments. The first number of the multiple first data segments is less than the second number of the multiple second data segments. The first determining module is used to determine a plurality of first loss values ​​based on the first data sequence, wherein the plurality of first loss values ​​correspond one-to-one with the plurality of first data segments; The second determining module is used to determine a plurality of second loss values ​​based on the second data sequence, wherein the plurality of second loss values ​​correspond one-to-one with the plurality of second data segments; The third determining module is used to determine multiple comprehensive loss values ​​based on the multiple first loss values ​​and the multiple second loss values, wherein the multiple comprehensive loss values ​​correspond one-to-one with the multiple second data segments; The fourth determining module is used to determine the thermal runaway risk level of the battery pack based on the multiple comprehensive loss values.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores multiple instructions, which are adapted to be loaded by a processor and executed by the method for determining the thermal runaway risk level of the battery pack according to any one of claims 1 to 7.

10. An electronic device, characterized in that, include: One or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method for determining the thermal runaway risk level of a battery pack according to any one of claims 1 to 7.

Images