A battery core temperature online estimation and prediction method, system, device and storage medium based on hybrid TSDM-AMBO-GRU

Abstract

The application discloses a kind of battery core temperature online estimation and prediction method, system, equipment and storage medium based on mixed TSDM-AMBO-GRU, applied to lithium ion battery core temperature estimation and prediction field, comprising: obtaining battery operating data, and utilizing TSDM to create synthesis of battery operating data, as training set with battery operating data together;Training set is input to BO-GRU-AM framework, and the battery core temperature prediction model based on BO-GRU-AM is obtained by training;Input test set to battery core temperature prediction model, and obtain battery core temperature prediction result.The application not only can effectively overcome the problem caused by uneven distribution of temperature sensors and limited data in lithium ion battery system, and can realize real-time, efficient and accurate battery core temperature prediction under various temperatures and complex conditions, provide reliable fault warning and real-time monitoring support for BMS in practical application.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery core temperature estimation and prediction, and particularly to a method, system, device, and storage medium for online estimation and prediction of battery core temperature based on a hybrid TSDM-AMBO-GRU. Background Technology

[0002] Developing renewable energy is a key strategy for addressing energy security and mitigating the environmental crisis. Lithium-ion batteries (LiB) are widely used in electric vehicles and energy storage systems due to their high power density, excellent energy density, long lifespan, and environmental friendliness. To ensure the safe, efficient, and stable operation of battery systems, meticulous management of lithium-ion batteries is essential, especially considering their temperature sensitivity. For example, high temperatures can trigger thermal runaway during charging and discharging, increasing safety risks, while low temperatures slow down internal electrochemical reactions, reducing the energy density and power output of lithium-ion batteries. Importantly, both extremes accelerate battery aging; low temperatures can lead to lithium deposition, while high temperatures can induce the growth of a solid electrolyte interfacial film. Therefore, employing advanced thermal management technologies to precisely maintain battery temperature within the optimal range is crucial for the lifespan of the battery system. In this regard, real-time and precise monitoring and control of battery temperature is essential for the battery management system (BMS).

[0003] Most BMS (Body Management Systems) focus on thermal safety by tracking surface temperature, which is easily achievable. However, the difference between core temperature and surface temperature in LiB (Liquid Biological Bodies) can exceed 10°C. Therefore, accurate estimation of core temperature becomes particularly important. Currently, some researchers have inserted temperature sensors into LiBs to measure their core temperature. While these methods can accurately capture core temperature, they are costly and challenging to implement. Therefore, exploring non-destructive methods to estimate the core temperature of LiBs has become a key focus and challenge in current LiB management research. Non-destructive methods for estimating core temperature mainly fall into three categories: electrochemical impedance spectroscopy (EIS)-based methods, numerical thermal modeling-based methods, and artificial intelligence-based methods. EIS-based methods determine the core temperature by analyzing the EIS characteristics of LiBs and extracting relevant features (such as the imaginary and real parts of impedance, phase shift, and phase angle at specific frequencies). However, the requirement for additional excitation hardware and the need to stop charging and discharging for impedance measurement hinder the real-time implementation of EIS-based methods in traditional LiB systems. Regarding estimations based on numerical thermal models, some researchers employ finite element analysis or partial differential equations with boundary conditions to simulate the thermodynamic processes within the battery, including heat generation, accumulation, and dissipation. While these methods offer high accuracy, their simulation complexity and computational cost are high, making them unsuitable for real-time estimation.

[0004] Currently, artificial intelligence is leading a technological revolution, and various methods are being researched to estimate the state of LiB (Li-ion batteries). Battery Management Systems (BMS) can effectively measure parameters such as current, voltage, surface temperature, and ambient temperature, thus facilitating black-box modeling to correlate input and output parameters; this modeling is particularly suitable for unknown or complex physical systems. Notably, in multi-state estimation of lithium-ion batteries involving state of charge, state of health, remaining lifespan prediction, and capacity prediction, data-driven methods can achieve accuracy comparable to or even exceeding that of physical models. Regarding core temperature estimation of lithium-ion batteries, Liu et al. proposed a linear neural network model that characterizes the nonlinear behavior of battery core temperature within the temperature range of 20℃–40℃. Li et al. used a deep learning method featuring a simple cyclic cell to estimate the core temperature of lithium-ion batteries under dynamic conditions, demonstrating that this cell can estimate core temperatures at 5℃, 15℃, and 45℃. While AI-based methods can estimate the core temperature of lithium-ion batteries by directly measuring external parameters such as voltage, current, ambient temperature, and surface temperature without in-depth analysis of the complex internal reaction mechanisms and electrochemical processes of lithium-ion batteries, these methods typically rely on large datasets for model training. In real-world scenarios, especially in systems containing hundreds or thousands of batteries, the typical average sensor-to-battery ratio is approximately 1:10, making it nearly impossible to equip each battery cell with a temperature sensor. Therefore, these methods often face challenges related to small sample sizes. Furthermore, lithium-ion batteries operate under highly variable and complex conditions, including challenging environments and wide-ranging temperature fluctuations. Current methods often fail to meet the requirements for real-time response and long-term predictive capabilities, making it difficult to predict potential failures or performance degradation.

[0005] Therefore, how to provide a battery core temperature online estimation and prediction method based on hybrid TSDM-AMBO-GRU that can effectively overcome the challenges caused by uneven distribution of temperature sensors and limited data in lithium-ion battery systems, and can achieve real-time, efficient and accurate prediction of battery core temperature under wide temperature range and complex conditions, and provide reliable fault warning and real-time monitoring support for BMS in practical applications, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention proposes a method, system, device and storage medium for online estimation and prediction of battery core temperature based on hybrid TSDM-AMBO-GRU.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for online estimation and prediction of battery core temperature based on a hybrid TSDM-AMBO-GRU includes:

[0009] Step 1: Acquire battery operating data and use TSDM to create and synthesize the battery operating data, which together with the battery operating data will serve as the training set;

[0010] Step 2: Input the training set into the BO-GRU-AM framework to train a battery core temperature prediction model based on BO-GRU-AM.

[0011] Step 3: Input the test set into the battery core temperature prediction model to obtain the battery core temperature prediction results.

[0012] Optionally, in step 1, battery operating data includes: current, voltage, ambient temperature, surface temperature, and core temperature.

[0013] Optionally, in step 1, the battery operating data is created and synthesized using TSDM, specifically as follows:

[0014] Define the dimensions of the input data and the number of noise steps to control the degree of data perturbation;

[0015] A three-layer fully connected network is used to extract and map data features, while integrating Dropout and BatchNorm1d technologies;

[0016] Forward diffusion: The model gradually adds noise to the input data, causing the data to gradually deviate from the original distribution, simulating the dynamic behavior of the battery system within a preset temperature range;

[0017] Backdiffusion: The model uses fully connected layers and activation functions to restore the data with added noise perturbation based on forward diffusion to its original state;

[0018] TSDM iteratively executes a loop of data loading, noise addition, and noise reduction to complete the creation and synthesis of battery operating data.

[0019] Optionally, in step 2, the training process of the battery core temperature prediction model based on BO-GRU-AM is as follows:

[0020] The BO algorithm constructs a proxy probability model to approximate the relationship between the objective function and hyperparameters, and iteratively updates it to efficiently search for the optimal combination of hyperparameters in the parameter space.

[0021] After the hyperparameters are adjusted by the BO algorithm, the data is fed into GRU-AM. The GRU network uses a gating mechanism to handle long-term dependencies in time series data, while AM ​​simulates human attention, focusing on important data, capturing positional relationships and assigning dynamic weight parameters. Finally, a battery core temperature prediction model based on BO-GRU-AM is trained.

[0022] This invention also provides a battery core temperature online estimation and prediction system based on a hybrid TSDM-AMBO-GRU, utilizing a hybrid TSDM-AMBO-GRU-based method for online estimation and prediction of battery core temperature, comprising:

[0023] Training set acquisition module: used to acquire battery operating data and use TSDM to create and synthesize the battery operating data, which together with the battery operating data is used as the training set;

[0024] Model training module: used to input the training set into the BO-GRU-AM framework to train a battery core temperature prediction model based on BO-GRU-AM.

[0025] The result prediction module is used to input the test set into the battery core temperature prediction model and obtain the battery core temperature prediction result.

[0026] The present invention also provides an electronic device, comprising:

[0027] Memory, used to store computer programs;

[0028] The processor is used to implement a method for online estimation and prediction of battery core temperature based on a hybrid TSDM-AMBO-GRU when executing a computer program.

[0029] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a method for online estimation and prediction of battery core temperature based on a hybrid TSDM-AMBO-GRU.

[0030] As can be seen from the above technical solutions, compared with the prior art, the present invention proposes a method, system, device and storage medium for online estimation and prediction of battery core temperature based on hybrid TSDM-AMBO-GRU. First, by using the TSDM system to systematically add and remove noise, the dynamic behavior of the battery system within a temperature range of -10℃ to 55℃ is simulated, generating synthetic data containing four important features: current, voltage, surface temperature, and core temperature. This effectively solves the problems caused by uneven distribution of temperature sensors and limited data in lithium-ion battery systems, improving the accuracy and robustness of the model under complex conditions. Second, the BO algorithm is used to construct a surrogate probability model to search for the optimal hyperparameter combination, achieving the most efficient network configuration, improving model training efficiency and performance, and reducing resource waste. Finally, GRU-AM ensures accurate core temperature estimation, improving model performance and saving computational resources and time. Specifically, the GRU network uses a gating mechanism to handle long-term dependencies in time-series data, overcoming the gradient vanishing or exploding problem of traditional recurrent neural networks (RNNs) when processing long sequences, and can accurately learn the complex relationship between battery operating data and battery core temperature. AM simulates human attention, focusing on important data to capture positional relationships and assigning dynamic weight parameters to overcome the problem of poor feature extraction caused by traditional neural networks treating all input data equally. The final trained model is a scalable, robust, and widely applicable battery core temperature prediction model based on BO-GRU-AM. Experiments demonstrate that the model provided by this invention not only achieves significant accuracy in estimating core temperatures under various environmental conditions but also performs excellently in prediction tasks, particularly for short-term forecasts. In the thermal warning mechanism, the model of this invention not only provides single-step temperature estimation but also risk assessment based on prediction results at multiple future time steps. Through this multi-step prediction capability, future temperature changes can be predicted and combined with safety thresholds to issue thermal warnings, providing real-time safety warnings for the battery management system, meeting real-time requirements, helping to maintain the safety and stability of the battery system, and extending battery life. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0033] Figure 2This diagram illustrates the core temperature and corresponding error indices estimated by the hybrid TSDM-AMBO-GRU model and the traditional AMBO-GRU model (without TSDM) of UDDS at an ambient temperature of -10℃.

[0034] Figure 3 This diagram illustrates the core temperature and corresponding error indices estimated by the hybrid TSDM-AMBO-GRU model and the traditional AMBO-GRU model (without TSDM) of UDDS at an ambient temperature of 0°C, according to the present invention.

[0035] Figure 4 This diagram illustrates the core temperature and corresponding error indices estimated by the hybrid TSDM-AMBO-GRU model and the traditional AMBO-GRU model (without TSDM) of UDDS at an ambient temperature of 15°C, according to the present invention.

[0036] Figure 5 This diagram illustrates the core temperature and corresponding error indices estimated by the hybrid TSDM-AMBO-GRU model and the traditional AMBO-GRU model (without TSDM) of UDDS at an ambient temperature of 45°C, according to the present invention.

[0037] Figure 6 This is a schematic diagram comparing the core temperature estimation performance of the model disclosed in this invention with other algorithms. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1:

[0040] Embodiment 1 of this invention discloses an online estimation and prediction method for battery core temperature based on a hybrid TSDM-AMBO-GRU, such as... Figure 1 As shown, it includes:

[0041] Step 1: Acquire battery operating data and use TSDM to create and synthesize the battery operating data, which together with the battery operating data will serve as the training set.

[0042] Battery operating data includes key parameters such as current, voltage, ambient temperature, surface temperature, and core temperature. These parameters are the foundational information for subsequent analysis and model training, closely related to the battery's operating state and temperature changes, and are crucial for accurately estimating the battery's core temperature.

[0043] In real-world battery operation scenarios, obtaining a large amount of real-world data under different operating conditions is challenging. TSDM can effectively increase data diversity, expand the training dataset, compensate for data insufficiency, enable the model to learn a wider range of data features, improve generalization ability, and better cope with complex real-world operating conditions. To enhance data diversity and expand the training dataset, TSDM was used to create and synthesize battery operating data, which, together with actual battery operating data, served as the training set, while dynamic operating data from real-world scenarios was used as the test set.

[0044] The creation and synthesis of battery operating data using TSDM is as follows:

[0045] Define the dimensions of the input data and the number of noise steps to control the degree of data perturbation;

[0046] A three-layer fully connected network is used to extract and map data features, while integrating Dropout and BatchNorm1d techniques to effectively alleviate overfitting and enhance training stability.

[0047] Forward diffusion: The model gradually adds noise to the input data, causing the data to gradually deviate from the original distribution, simulating the dynamic behavior of the battery system within a preset temperature range (in this embodiment of the invention, the preset temperature range is selected as -10℃ to 55℃).

[0048] Backdiffusion: The model uses fully connected layers and activation functions to restore the data with added noise perturbation based on forward diffusion to its original state;

[0049] Throughout the training process, TSDM iteratively executes loops of data loading, noise addition, and denoising to create and synthesize battery operating data. This model not only helps to accurately reconstruct real battery data but also generates representative synthetic data. TSDM can significantly improve the accuracy and robustness of battery core temperature estimation under complex conditions, providing crucial support for achieving effective battery system performance under challenging conditions.

[0050] By systematically adding and removing noise using TSDM, the dynamic behavior of the battery system in the temperature range of -10℃ to 55℃ is simulated, generating synthetic data that includes four important features: current, voltage, surface temperature, and core temperature. This effectively solves the problems caused by uneven distribution of temperature sensors and limited data in lithium-ion battery systems.

[0051] Step 2: Input the training set into the BO-GRU-AM framework to train a battery core temperature prediction model based on BO-GRU-AM.

[0052] The training process of the battery core temperature prediction model based on BO-GRU-AM is as follows:

[0053] The Bayesian Optimization (BO) algorithm approximates the relationship between the objective function and hyperparameters by constructing a surrogate probability model and iteratively updating it to efficiently search for the optimal hyperparameter combination in the parameter space, thereby achieving the most efficient network configuration. Specifically, the BO algorithm sets the loss function of the deep learning model as the objective function f(x) and optimizes hyperparameters such as hidden dimension (hidden_dim), number of layers (num_layer), and learning rate. First, a dataset is created and a surrogate probability model is fitted. Then, optimized hyperparameters x[hidden_dim, num_layer, learning rate] are selected from the parameter space. The surrogate model is iteratively improved to reduce the impact of noise and performance limitations, resulting in robust and effective results.

[0054] Deep learning hyperparameter optimization is challenging due to the high-dimensional search space, and traditional methods are inefficient. The BO algorithm is used to automatically adjust hyperparameters. By utilizing a surrogate probability model to effectively approximate the objective function, it selects the evaluation point with the greatest potential, reduces the number of evaluations, achieves the most efficient network configuration, improves model training efficiency and performance, and reduces resource waste.

[0055] The data, after hyperparameter adjustment using the BO algorithm, enters GRU-AM. The GRU (Gated Recurrent Unit) network utilizes a gating mechanism to handle long-term dependencies in time-series data, overcoming the gradient vanishing or exploding problems of traditional recurrent neural networks (RNNs) when processing long sequences. It focuses on capturing long-term dependencies in time series data, making it very suitable for data with time-series characteristics, such as battery operation data, and can accurately learn the complex relationship between battery operation data and battery core temperature. AM (Attention Mechanism) is integrated into the GRU network. By simulating human attention, it focuses on important data, improves feature extraction efficiency, captures positional relationships, and assigns dynamic weight parameters. This overcomes the problem of poor feature extraction caused by traditional neural networks treating all input data equally. Finally, a scalable, robust, and widely applicable battery core temperature prediction model based on BO-GRU-AM is trained.

[0056] Step 3: Input the test set into the battery core temperature prediction model to obtain the battery core temperature prediction results.

[0057] Using dynamic operating data from real-world scenarios as a test set, the performance of the battery core temperature prediction model in practical applications is examined. It accurately estimates the core temperature of the battery under various dynamic conditions, especially over a wide temperature range, providing important support for the battery management system. This helps maintain safety and stability, and by accurately predicting the temperature in advance, measures can be taken to avoid problems such as battery overheating, thus extending battery life.

[0058] Example 2:

[0059] Embodiment 2 of this invention discloses a specific application test and performance verification of a battery core temperature online estimation and prediction method based on a hybrid TSDM-AMBO-GRU, as follows:

[0060] 18650 lithium-ion batteries were used. To ensure the safety of LiB at low temperatures, the charging cutoff voltage was configured to 4.2V, while the discharging cutoff voltage was set to 3V to avoid over-discharge. The experimental platform included battery charge / discharge testing equipment, a constant temperature chamber, a computer host, and a data acquisition system. The battery testing equipment was used to (a) precisely control the charging and discharging process of LiB and (b) record parameters such as voltage, current, and temperature. The constant temperature chamber simulated the operating conditions of LiB under different ambient temperatures, ensuring the dataset had broad applicability and representativeness. The computer host was responsible for signal control and data storage, managing the overall operation of the system. The data acquisition system was responsible for recording and storing data in real time during the experiment, ensuring the integrity and accuracy of the data.

[0061] To verify the applicability and robustness of the battery core temperature estimation method disclosed in this invention under a wide range of temperatures and actual operating conditions, this invention uses real urban cycle operating condition (UDDS) data as the target dataset for model validation.

[0062] Furthermore, to reduce the impact of data noise on the test results, accelerate model convergence, and improve prediction accuracy, this invention implements standardization as a preprocessing technique. Standardization is crucial for training; without it, the model may fail to converge. A linear function (min-max normalization) is used to scale the original data to the range [0,1].

[0063] To evaluate the estimation results, this invention uses root mean square error (RMSE), mean absolute error (MAE), and maximum error (MAXE) to compare the actual data with the data predicted by the proposed mixture model. All model implementations are equipped with... Core TM The model was executed on a computer with an i7-8700 CPU @ 3.20GHz processor and Windows 10 operating system. The proposed model was built using Python 3.7.16.

[0064] To evaluate the core temperature estimation performance of the hybrid TSDM-AMBO-GRU model disclosed in this invention, core temperature estimation results obtained under Dynamic Driving Cycles (UDDS) over a wide temperature range (-10°C to 55°C) were compared. Experiments compared and verified the estimation results of the TSDM-AMBO-GRU model with those of the AMBO-GRU model that did not use TSDM-generated data.

[0065] like Figures 2-5 As shown, the core temperature and corresponding error indices estimated by the proposed hybrid TSDM-AMBO-GRU model and the traditional AMBO-GRU model (without TSDM) under different ambient temperatures and driving cycles are compared. Figure 2 , Figure 3 , Figure 4 , Figure 5 The core temperatures and corresponding error indices estimated by the hybrid TSDM-AMBO-GRU model and the traditional AMBO-GRU model (without TSDM) for UDDS at ambient temperatures of -10℃, 0℃, 15℃, and 45℃ are shown in Table 1. The RMSE, MAE, and MAXE values ​​of the models are listed. To verify the robustness of the model disclosed in this invention, five tests were conducted for each operating condition at different ambient temperatures, and the mean error and 95% confidence interval were calculated.

[0066] The results show that integrating TSDM into the conventional data generation model can significantly reduce estimation errors: the maximum RMSE, MAE, and MAXE are reduced by approximately 36.4%, 43.0%, and 31.9%, respectively. Under the UDDS condition at 5°C, the average MAE decreased from 0.1452°C to 0.0568°C (approximately 60.9%), and under the UDDS condition at 45°C, the average MAXE decreased from 0.3189°C to 0.1777°C, a reduction of approximately 44.3%.

[0067] These results demonstrate that when using TSDM for data generation, the BO-GRU-AM model can effectively capture the time dependence in time series data, thereby improving the accuracy of core temperature estimation, while maintaining robust performance under various ambient temperatures and complex dynamic driving conditions.

[0068] Table 1 shows the nuclear temperature estimation results of the proposed method and the unused data amplification over a wide temperature range.

[0069]

[0070]

[0071] To evaluate the effectiveness of the model disclosed in this invention for core temperature estimation, its performance was compared with that of 10 algorithms categorized into three classes. The first class includes traditional models, including bimodal thermal models (TM) and TMs with unscented Kalman filters (UKF). The second class includes deep learning methods without TSDM, including recurrent neural networks (RNNs), long short-term memory (LSTMs), GRUs, and AM-GRUs. The third class includes TSDM-enhanced deep learning techniques, including TSDM-RNNs, TSDM-LSTMs, and TSDM-GRUs.

[0072] All experiments were conducted on the same dataset under dynamic operating conditions at a constant ambient temperature of 35°C. Figure 6 As shown, the core temperature estimation performance of the model disclosed in this invention is compared with that of other algorithms, and the error distribution is represented by a box plot. Table 2 lists the performance metrics, including RMSE, MAE, MAXE, and training and testing time.

[0073] Table 2 Comparison of results from different nuclear temperature estimation methods

[0074] method RMSE (°C) MAE (°C) MAXE (°C) Training time (s) Test time (s) TM 0.2958 0.2579 0.6790 / 0.6741 TM-UKF 0.1790 0.1590 0.3977 / 3.2068 RNN 0.1788 0.1412 0.7573 48.2847 0.2016 LSTM 0.1249 0.0954 0.5575 68 0.2523 GRU 0.1172 0.0938 0.4555 62.0549 0.2254 AM-GRU 0.0902 0.0703 0.5013 74.4790 0.2713 TSDM-RNN 0.0977 0.0715 0.3881 92.9798 0.2094 TSDM-LSTM 0.0843 0.0615 0.3605 129.1379 0.2324 TSDM-GRU 0.0579 0.0429 0.3782 121.1797 0.2424 TSDM-AM-GRU 0.0455 0.0343 0.3168 146.1626 0.2631

[0075] Compared to other core temperature estimation methods, the model disclosed in this invention yields superior estimation results, with an RMSE of 0.0455°C, a MAE of 0.0343°C, and a MAXE of 0.3168°C. These results represent a significant improvement over conventional methods. For example, compared to the conventional TM model, the RMSE of the model disclosed in this invention is reduced by approximately 84.6% (from 0.2958°C to 0.0455°C), and the MAE is reduced by approximately 86.7% (from 0.2579°C to 0.0343°C). Similarly, compared to the standard GRU model, the RMSE of the model disclosed in this invention is reduced by approximately 61.2% (from 0.1172°C to 0.0455°C), and the MAE is reduced by 63.4% (from 0.0938°C to 0.0343°C). Figure 6 As shown, among all evaluation methods, the TSDM-AM-GRU model proposed in this invention exhibits the most concentrated error distribution. Its box plot height is significantly smaller, indicating a reduced error variance and more accurate predictions.

[0076] In terms of computational efficiency, the TSDM-AM-GRU has training and testing times of 146.16 and 0.2631 seconds, respectively. Although the training time is longer than some benchmark models, the testing time is very short, 0.0615 seconds faster than the fastest RNN model, fully meeting the real-time requirements of BMS. Furthermore, since retraining is not required, the trained model can be directly deployed on devices with limited computing resources.

[0077] The core temperature estimation model introduced in this invention also demonstrates predictive capability for future time intervals. This multi-step prediction capability enables the system to predict future temperature changes and, in conjunction with a predefined safety threshold, determine whether a thermal warning should be issued. Experiments were conducted under DST conditions at different initial ambient temperatures, predicting the next 10, 30, 60, and 120 seconds. This method predicts temperature changes over several future time intervals to assess whether the temperature exceeds a predetermined safety threshold, thereby providing real-time safety alerts for the LiB system. To verify the robustness of the proposed method, five tests were performed for each operating condition at different ambient temperatures, and the mean error and 95% confidence interval were recorded. The core temperature prediction experimental results are shown in Table 3.

[0078] Table 3 shows the core temperature advance prediction results at different temperatures using the hybrid model proposed in this invention.

[0079]

[0080]

[0081] The maximum and minimum RMSE values ​​of the model disclosed in this invention are 0.8504℃±0.0052℃ and 0.0640℃±0.0061℃, respectively; the maximum and minimum MAE values ​​are 0.7130℃±0.0562℃ and 0.0488℃±0.0053℃, respectively; and the maximum and minimum MAXE values ​​are 1.9892℃±0.0871℃ and 0.2859℃±0.0267℃, respectively. It is noteworthy that the RMSE for predicting the core temperature increases with the extension of the prediction period. Furthermore, the proposed hybrid TSDM-AMBO-GRU method exhibits high prediction accuracy under different ambient temperature conditions, especially for short-term predictions of 10 and 30 seconds. For example, at -10℃, the RMSE for the 10-second prediction of the disclosed model is 0.1929℃, significantly higher than the RMSE for the 120-second prediction (0.8504℃).

[0082] In summary, the proposed hybrid TSDM-AMBO-GRU model not only achieves significant accuracy in estimating core temperatures under various environmental conditions but also performs excellently in forecasting tasks, particularly for short-term predictions. In thermal warning mechanisms, the model of this invention provides not only single-step temperature estimation but also risk assessment based on predictions at multiple future time steps. If the predicted temperature exceeds a critical threshold at any given future time point, the system will immediately trigger a thermal warning, thus providing proactive safety measures. This multi-step prediction can make a significant contribution to fault alarms and temperature monitoring within the BMS.

[0083] Example 3:

[0084] Embodiment 3 of the present invention discloses a battery core temperature online estimation and prediction system based on a hybrid TSDM-AMBO-GRU, comprising:

[0085] Training set acquisition module: used to acquire battery operating data and use TSDM to create and synthesize the battery operating data, which together with the battery operating data is used as the training set;

[0086] Model training module: used to input the training set into the BO-GRU-AM framework to train a battery core temperature prediction model based on BO-GRU-AM.

[0087] The result prediction module is used to input the test set into the battery core temperature prediction model and obtain the battery core temperature prediction result.

[0088] Example 4:

[0089] Embodiment 4 of the present invention discloses an electronic device, comprising:

[0090] Memory, used to store computer programs;

[0091] The processor is used to implement a method for online estimation and prediction of battery core temperature based on a hybrid TSDM-AMBO-GRU when executing a computer program.

[0092] Example 5:

[0093] Embodiment 5 of the present invention discloses a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of an online estimation and prediction method for battery core temperature based on a hybrid TSDM-AMBO-GRU.

[0094] This invention discloses a method, system, device, and storage medium for online estimation and prediction of battery core temperature based on a hybrid TSDM-AMBO-GRU. First, by using the TSDM system to systematically add and remove noise, the dynamic behavior of the battery system within a temperature range of -10℃ to 55℃ is simulated, generating synthetic data containing four important features: current, voltage, surface temperature, and core temperature. This effectively solves the problems caused by uneven distribution of temperature sensors and limited data in lithium-ion battery systems, improving the accuracy and robustness of the model under complex conditions. Second, the BO algorithm is used to construct a surrogate probability model to search for the optimal hyperparameter combination, achieving the most efficient network configuration, improving model training efficiency and performance, and reducing resource waste. Finally, GRU-AM ensures accurate core temperature estimation, improving model performance and saving computational resources and time. Specifically, the GRU network uses a gating mechanism to handle long-term dependencies in time-series data, overcoming the gradient vanishing or exploding problem of traditional recurrent neural networks (RNNs) when processing long sequences, and can accurately learn the complex relationship between battery operating data and battery core temperature. AM simulates human attention, focusing on important data to capture positional relationships and assigning dynamic weight parameters to overcome the problem of poor feature extraction caused by traditional neural networks treating all input data equally. The final trained model is a scalable, robust, and widely applicable battery core temperature prediction model based on BO-GRU-AM. Experiments demonstrate that the model provided by this invention not only achieves significant accuracy in estimating core temperatures under various environmental conditions but also performs excellently in prediction tasks, particularly for short-term forecasts. In the thermal warning mechanism, the model of this invention not only provides single-step temperature estimation but also risk assessment based on prediction results at multiple future time steps. Through this multi-step prediction capability, future temperature changes can be predicted and combined with safety thresholds to issue thermal warnings, providing real-time safety warnings for the battery management system, meeting real-time requirements, helping to maintain the safety and stability of the battery system, and extending battery life.

[0095] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0096] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for online estimation and prediction of battery core temperature based on a hybrid TSDM-AMBO-GRU, characterized in that, include: Step 1: Acquire battery operating data and use TSDM to create and synthesize the battery operating data, which together with the battery operating data serves as the training set; Step 2: Input the training set into the BO-GRU-AM framework to train a battery core temperature prediction model based on BO-GRU-AM. Step 3: Input the test set into the battery core temperature prediction model to obtain the battery core temperature prediction results.

2. The method for online estimation and prediction of battery core temperature based on a hybrid TSDM-AMBO-GRU according to claim 1, characterized in that, In step 1, the battery operating data includes: current, voltage, ambient temperature, surface temperature, and core temperature.

3. The method for online estimation and prediction of battery core temperature based on a hybrid TSDM-AMBO-GRU according to claim 1, characterized in that, In step 1, the battery operating data is created and synthesized using TSDM, specifically as follows: Define the dimensions of the input data and the number of noise steps to control the degree of data perturbation; A three-layer fully connected network is used to extract and map data features, while integrating Dropout and BatchNorm1d technologies; Forward diffusion: The model gradually adds noise to the input data, causing the data to gradually deviate from the original distribution, simulating the dynamic behavior of the battery system within a preset temperature range; Backdiffusion: The model uses fully connected layers and activation functions to restore the data with added noise perturbation based on the forward diffusion to its original state; TSDM iteratively executes a loop of data loading, noise addition, and noise reduction to complete the creation and synthesis of the battery operating data.

4. The method for online estimation and prediction of battery core temperature based on a hybrid TSDM-AMBO-GRU according to claim 1, characterized in that, Step 2, the training process of the battery core temperature prediction model based on BO-GRU-AM, is as follows: The BO algorithm constructs a proxy probability model to approximate the relationship between the objective function and hyperparameters, and iteratively updates it to efficiently search for the optimal combination of hyperparameters in the parameter space. After the hyperparameters are adjusted by the BO algorithm, the data is fed into GRU-AM. The GRU network uses a gating mechanism to handle long-term dependencies in time series data, while AM ​​simulates human attention, focusing on important data, capturing positional relationships and assigning dynamic weight parameters. Finally, a battery core temperature prediction model based on BO-GRU-AM is trained.

5. A battery core temperature online estimation and prediction system based on a hybrid TSDM-AMBO-GRU, utilizing the battery core temperature online estimation and prediction method based on any one of claims 1-4, characterized in that, include: Training set acquisition module: used to acquire battery operating data and use TSDM to create and synthesize the battery operating data, which together with the battery operating data is used as the training set; Model training module: used to input the training set into the BO-GRU-AM framework to train a battery core temperature prediction model based on BO-GRU-AM; Result prediction module: used to input the test set into the battery core temperature prediction model to obtain the battery core temperature prediction result.

6. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the online estimation and prediction method for battery core temperature based on a hybrid TSDM-AMBO-GRU as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the online estimation and prediction method for battery core temperature based on a hybrid TSDM-AMBO-GRU as described in any one of claims 1-4.

Images