Battery cluster peak power estimation method and system and medium
By combining the meta-learning framework with electrothermal coupling physical equations, the accuracy and safety issues of battery cluster peak power estimation are solved, achieving rapid adaptation and efficient battery cluster power estimation, and significantly improving the system's safety and regulation performance.
Patent Information
- Application Number
- CN202511691585.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-18
- Publication Date
- 2026-02-13
AI Technical Summary
Existing peak power estimation methods for battery clusters are difficult to apply to complex topologies, resulting in inaccurate power estimation results. Furthermore, traditional methods suffer from severe model drift when batteries age and the environment changes, affecting system safety and control performance.
A three-stage training framework based on meta-learning is adopted, which combines convolutional layers and fully connected layers to build the model. The model is trained through support sets, query sets and fine-tuning sets, and the electrothermal coupling physical equation is introduced as a constraint to achieve rapid model adaptation and cross-task generalization.
Under small sample conditions, a fast, accurate and safe estimation of the peak power of battery clusters was achieved with an average error of less than 1.3%, which is significantly better than traditional methods. It meets the real-time requirements of power grid frequency regulation and effectively suppresses the risks of model drift and thermal runaway.
Smart Images
Figure CN121522475A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of battery management, in particular to a battery cluster peak power estimation method, system and medium. BACKGROUND
[0002] With the deepening of global energy transformation, the energy storage system taking lithium-ion battery as the core has become a key technology to support the stable grid connection of renewable energy. Accurate estimation of the peak power of the battery cluster is crucial to ensure system safety and improve control performance. However, existing estimation methods are mostly focused on the battery module level, which is difficult to directly apply to the battery cluster composed of a large number of modules in series and parallel, resulting in a mismatch between the estimated power and the actual output capacity, which may easily cause safety problems such as over-discharge or over-charge.
[0003] In recent years, physical information neural network (PINN) has been introduced into the field of battery modeling, but it still needs to collect a large amount of data for the target battery in single battery modeling, which cannot be directly migrated to battery clusters with different topologies. At the same time, model-agnostic meta-learning (MAML) has not been effectively applied in cluster-level power estimation, resulting in long deployment period and large sample demand, which restricts the engineering promotion. In addition, the peak power of the battery is strongly coupled with temperature and aging state, and traditional data-driven models are prone to model drift when the working conditions change or the battery ages, which requires expensive re-calibration to restore performance.
[0004] Therefore, there is an urgent need in the art for a battery cluster peak power estimation method that can achieve fast, accurate and safe estimation under small sample conditions, to systematically solve the key problems of poor model transferability, weak environmental adaptability and insufficient estimation accuracy in the whole life cycle. SUMMARY
[0005] The purpose of the present application is to provide a battery cluster peak power estimation method that can achieve fast, accurate and safe estimation of the peak power of energy storage battery clusters.
[0006] To solve the above technical problems, an embodiment of the present application provides a battery cluster peak power estimation method, comprising the following steps: Obtain time series data sets of the voltage, current and temperature of each battery module in the battery cluster, the time series data sets being divided into a support set, a query set and a fine-tuning set; construct a model for estimating the maximum available charge and discharge power of the battery module, the model being composed of convolutional layers and fully connected layers; Use the support set to perform inner loop training on the model, and perform gradient descent optimization on all parameters of the model by minimizing the mean square error loss of the power prediction value and the true value, to obtain task adaptation parameters; The task adaptation parameters are loaded into the model, and the model is trained in the outer loop using the query set. By minimizing the composite loss function composed of the mean square error and the residual of the electrothermal coupling equation, the meta-initialization parameters of the model are optimized by backpropagation to obtain meta-initialization parameters with cross-task generalization ability. The initialization parameters are loaded into the model, and the model is trained using a fine-tuning set. The parameters of the convolutional layers are kept constant, and the weights and biases of the fully connected layers are optimized by minimizing the composite loss function to obtain the trained battery module maximum available charge and discharge power estimation model, which is used to estimate the maximum available charge and discharge power of the battery module. The maximum available charge and discharge power of each battery module is fused to obtain the peak power of the battery cluster.
[0007] In some optional embodiments, a safety correction for the peak power of the battery cluster is also included, comprising the following steps: Real-time monitoring of the rate of change of battery temperature over time; If the rate of change exceeds a preset threshold, the peak power of the battery cluster is multiplied by a preset margin coefficient for safety correction.
[0008] In some optional embodiments, the step of optimizing all model parameters by gradient descent by minimizing the mean squared error loss between the predicted and actual power values specifically includes: The formula for mean squared error loss is as follows: ; In the formula, The mean squared error loss is a scalar value used to quantify the model's performance on the support set. On the predictive performance; Initialize parameters for the meta-initialization; To support the set; To represent the support set The total number of samples included; Indicates the support set Each data point in Perform subsequent error calculations and sum the calculation results for all data points; The input consists of time-series data of voltage, current, and temperature. The power prediction value output by the model; The power label represents the actual value of the data. The gradient descent optimization formula is as follows: ; In the formula, Adapt parameters for the task; The learning rate for the inner loop, taking values... ; For parameters The gradient; This is the mean square error loss; To support the set; Use the initialization parameters for the meta-initialization.
[0009] In some optional embodiments, the backpropagation optimization of the meta-initialization parameters of the model by minimizing the composite loss function formed by the mean square error and the residual of the electrothermal coupling equation specifically includes: The electrothermal coupling equation is as follows: ; ; In the formula, This is the physical residual loss; For query set; Adapt parameters for the task; The number of samples in the query set; The residuals of the electrothermal coupling equation; The power prediction value output by the model; For battery heat capacity; R is the rate of temperature change; I is the current; R is the battery internal resistance; T is the battery temperature. Ambient temperature; Battery thermal resistance; The input consists of time-series data of voltage, current, and temperature. The formula for the composite loss function is as follows: ; In the formula, The meta-loss function; The weighting coefficients for the physical law residuals; Adapt parameters for the task; For query set; This represents the physical residual loss.
[0010] The formula for updating the meta-initial parameters using stochastic gradient descent is as follows: ; In the formula, The learning rate for the outer loop; Initialize parameters for the meta-initialization. Meta-loss function For meta-parameters The gradient; This is the meta-loss function.
[0011] In some optional embodiments, the process of fusing the maximum available charge and discharge power of each battery module to obtain the peak power of the battery cluster specifically includes the following steps: The maximum available charge and discharge power of each battery module is aligned and fused along a unified time axis to generate a time series of the maximum available charge and discharge power for each battery module. The charge and discharge power time series is processed by a Long Short-Term Memory (LSTM) network to capture the dynamic dependence of the power of a single battery module over time, and outputs the hidden state that integrates the full time-series information of a single battery module. The hidden state of all battery modules is processed by a Transformer encoder module, and the power interaction and constraint relationship between different battery modules is captured by a self-attention mechanism, outputting a feature matrix that integrates global correlation information. The feature matrix is weighted and aggregated by an attention mechanism, tilting the weights toward the weakest battery module that has the greatest power limitation on the entire battery cluster. The decoder maps the aggregated features to the peak power of the battery cluster.
[0012] Embodiments of the present invention also provide a computer device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described battery cluster peak power estimation method.
[0013] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when run by a processor, is capable of executing the above-described battery cluster peak power estimation method.
[0014] The battery cluster peak power estimation method provided by this invention has at least the following beneficial effects: This invention proposes a three-stage training framework based on meta-learning. It effectively transfers pre-trained basic model knowledge to battery cluster scenarios through a transfer learning mechanism, and systematically introduces electrothermal coupling physical equations as constraints during multi-stage model training. This method requires only a small amount of cluster-level data, such as support sets, query sets, and fine-tuning sets, to achieve rapid model adaptation, effectively solving the problems of poor model transferability and inaccurate estimation caused by the complex topology and scarce samples of traditional methods. In four-dimensional operating condition tests covering temperature, rate capability, SOC, and SOH, the average error is only 1.3%, significantly outperforming traditional HPPC methods and purely data-driven models. Attached Figure Description
[0015] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:
[0016] Figure 1 This is a flowchart of a battery cluster peak power estimation method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the migration-PINN model structure provided according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a battery cluster peak power estimation method according to an embodiment of the present invention; Figure 4 This is a schematic diagram of an edge computing terminal hardware architecture provided according to an embodiment of the present invention. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0018] One embodiment of the present invention relates to a method for estimating the peak power of a battery cluster. The implementation details of the method for estimating the peak power of a battery cluster in this embodiment are described in detail below. The following content is only for the convenience of understanding the implementation details and is not necessary for implementing this solution.
[0019] The specific process of the battery cluster peak power estimation method in this embodiment can be described as follows: Figure 1 As shown, it includes: Step 101: Obtain the time series dataset of voltage, current and temperature of each battery module in the battery cluster. The time series dataset is divided into a support set, a query set and a fine-tuning set. Construct a model to estimate the maximum available charge and discharge power of the battery module. The model consists of convolutional layers and fully connected layers. The battery cluster interface board is used to synchronously collect voltage, current and temperature data of each battery module in the battery cluster; Step 102: Use the support set to train the model in the inner loop. By minimizing the mean squared error loss between the predicted power value and the true value, perform gradient descent optimization on all parameters of the model to obtain the task-fit parameters. The formula for mean squared error loss is as follows: ; In the formula, The mean squared error loss is a scalar value used to quantify the model's performance on the support set. On the predictive performance; Initialize parameters for the meta-initialization; To support the set; To represent the support set The total number of samples included; Indicates the support set Each data point in Perform subsequent error calculations and sum the calculation results for all data points; The input consists of time-series data of voltage, current, and temperature. The power prediction value output by the model; The power label represents the actual value of the data. The gradient descent optimization formula is as follows: ; In the formula, Adapt parameters for the task; The learning rate for the inner loop, taking values... ; For parameters The gradient; This is the mean square error loss; To support the set; Use the initialization parameters for the meta-initialization.
[0020] Step 103: Load the task adaptation parameters into the model, use the query set to train the model in the outer loop, optimize the meta-initialization parameters of the model by minimizing the composite loss function composed of the mean square error and the residual of the electrothermal coupling equation, and obtain meta-initialization parameters with cross-task generalization ability. The electrothermal coupling equation is as follows: ; ; In the formula, This is the physical residual loss; For query set; Adapt parameters for the task; The number of samples in the query set; The residuals of the electrothermal coupling equation; The power prediction value output by the model; For battery heat capacity; R is the rate of temperature change; I is the current; R is the battery internal resistance; T is the battery temperature. Ambient temperature; Battery thermal resistance; The input consists of time-series data of voltage, current, and temperature. The formula for the composite loss function is as follows: ; In the formula, The meta-loss function; The weighting coefficients for the physical law residuals; Adapt parameters for the task; For query set; This represents the physical residual loss.
[0021] The formula for updating the meta-initial parameters using stochastic gradient descent is as follows: ; In the formula, The learning rate for the outer loop; Initialize parameters for the meta-initialization. Meta-loss function For meta-parameters The gradient; This is the meta-loss function.
[0022] Step 104: Load the initialization parameters into the model, train the model using the fine-tuning set, keep the convolutional layer parameters unchanged, optimize the weights and biases of the fully connected layer by minimizing the composite loss function, and obtain the trained battery module maximum available charge and discharge power estimation model, which is used to estimate the maximum available charge and discharge power of the battery module; fuse the maximum available charge and discharge power of each battery module to obtain the peak power of the battery cluster. The peak power estimation model for battery clusters is structured as a migration-PINN model, as shown in the diagram below. Figure 2 As shown, the forward propagation path in the diagram begins at the input layer, which receives multi-channel voltage, current, and temperature time-series data. The data then enters a transfer learning structure containing (A) a pre-trained pedestal model pre-trained on a single dataset using Meta-Learning (MAML) with frozen weights, and (B) a task-specific layer (fine-tuning task head) fine-tuned on a small amount of cluster-level data. The predicted value Y_pred output by this structure enters the PINN loss function calculation module. This module contains two parallel computation paths: the first path calculates the data-driven loss (Loss_data), which is the mean squared error (MSE) between the predicted value Y_pred and the true value Y_true; the second path calculates the physical information loss (Loss_phys), which substitutes Y_pred into a known physical equation (such as a first-order RC electrothermal coupling equation) and calculates the L2 squared of its residual. Finally, the total loss function (Loss_total) is obtained by weighted summation of these two losses (with weight λ). The total loss is fed into an optimizer (such as Adam), and the calculated gradients are used for backpropagation to update the trainable weights of the fine-tuning task head.
[0023] Battery cluster peak power estimation method as follows Figure 3As shown, the process in the diagram starts from (Start) and proceeds to step (1) data acquisition, synchronously acquiring the original time-series data of voltage, current, and temperature of each module. In step (2) data preprocessing, the CEEMDAN-KMeans-VMD framework is used to denoise the original data to obtain clean time-series data. In step (3) model inference, the clean data is input into the loaded migration-PINN model, and the internal state of the battery is output. Next, in step (4) multi-scale decoding, the decoder calculates the preliminary cluster-level SOP (SOP_pack) based on the internal state output by the model. Step (5) safety boundary check is then performed, which is a decision node where the system determines whether the battery's temperature change rate is greater than a safety threshold (e.g., 2°C / min). If the result is "yes", the process proceeds to step (6a), applying a safety margin (e.g., multiplying by 0.8) to SOP_pack and generating a warning signal to obtain the reduced SOP_final. If the result is "no", the process proceeds to step (6b), maintaining the original value, i.e., SOP_final equals SOP_pack. Finally, in the output of step (7), the final SOP_final is reported to the Energy Management System (EMS), and the process ends at (End).
[0024] The edge computing terminal hardware architecture designed in this invention is as follows: Figure 4 As shown, this architecture is specifically designed for the efficient, real-time execution of complex algorithms involved in the battery cluster peak power estimation (SOP) model, such as the Physical Information Neural Network (PINN) and the Model Independent Meta-Learning Framework (MAML). This ensures the model completes a single inference within 6 ms, meeting the stringent response speed requirements of scenarios like power grid frequency regulation. In this architecture, the STM32H7 series MCU acts as the main controller, responsible for scheduling the entire SOP estimation process; the FPGA acts as a coprocessor, accelerating computationally intensive tasks such as neural networks to handle high parallel loads. The system is equipped with Flash memory to store pre-trained model weights and related algorithm firmware, ensuring offline operation capabilities; RAM provides high-speed caching support for real-time battery data (voltage, current, temperature) and intermediate variables required for model operation. Regarding data interfaces, the terminal acquires battery cluster sensor signals in real-time via high-speed synchronous interfaces such as SPI / CAN as input to the SOP model; the final calculation results are reported to the Energy Management System (EMS) via a communication module (supporting Ethernet / 4G) to achieve real-time control of battery power. This module also supports receiving OTA updates from the cloud, facilitating remote model iteration and optimization. The entire terminal is powered by a power management unit (PMU) to ensure continuous and stable operation in complex industrial environments.
[0025] To enable those skilled in the art to fully understand and replicate this invention, the applicant conducted a system verification experiment covering four-dimensional operating conditions: temperature, rate, SOC, and SOH, on a standard battery cluster testing platform (Arbin BT-5HC, 80 channels, ±5 V / ±300 A). The experiment used 80 lithium iron phosphate battery clusters, each with a 15-parallel, 48-string structure and a rated capacity of 280 Ah. Each cluster underwent five cycles of hybrid pulse power characteristic (HPPC), accumulating 400 valid test sequences. The environmental operating conditions covered temperatures from -10 °C to 45 °C (5 °C step), pulse rates from 0.5 C to 3 C (0.5 C step), SOC from 20% to 90%, and SOH from 85% to 100%. To evaluate the model's extrapolation capability, an additional 30 sequences were collected under extreme conditions (-20 °C, 4.5 C, 10% SOC) without prior training.
[0026] Regarding error assessment, the formula for Mean Error (MAE) is as follows: MAE = (1 / N)·Σ|SOP_pred − SOP_ref| / SOP_ref ×100 %; In the formula, SOP_ref is the reference measured value; the extrapolation error is calculated in the same way as MAE, but SOP_ref is taken from the measured data under extreme conditions; the inference delay is defined as the time required from the completion of voltage, current and temperature sampling to the output of SOP_final, excluding communication overhead. The results are shown in Table 1.
[0027] Table 1 Results Analysis; The complete solution of this invention achieves MAE of 1.3% ± 0.2% under 400 coverage conditions, with a maximum single error of only 2.3%; even under extreme extrapolation conditions, it still maintains an error of 1.5%, which is significantly lower than the 15.3% of the closest prior art (Zhang 2022), proving that the PINN physical residual term effectively suppresses model drift.
[0028] The ablation test results clearly verified the necessity of the core design of this invention: after removing the physical constraint terms in the Physical Information Neural Network (PINN), the extrapolation error of the model under extreme conditions significantly increased from 1.5% to 5.1%, indicating that the embedding of physical laws is an indispensable technical feature for ensuring the model's generalization ability. Furthermore, if the Meta-Learning (MAML) pre-training stage is removed, the extrapolation error continues to rise to 6.7%, which confirms the key role of the meta-learning mechanism in achieving rapid model adaptation and cross-condition generalization. In addition, the method exhibits an inference latency of 6 ms ± 0.5 ms in actual deployment, fully meeting the real-time requirements of typical applications such as power grid frequency regulation for system response times of less than 10 ms, demonstrating its good engineering applicability. Security verification is shown in Table 2.
[0029] Table 2 Security Verification; Safety verification results demonstrate that the safety module introduced in this invention plays a crucial role in suppressing thermal runaway. In 120 3C pulse test sequences with a temperature rise rate exceeding 2 °C / min, the control group without the safety module experienced 14 instances of thermal runaway, a probability of 11.7%. However, after using a linear interpolation margin coefficient k∈[0.7, 0.9] to limit the state of operation (SOP), the number of thermal runaways decreased to 5, with a probability of only 4.2%, effectively reducing the risk by 62.4%. Simultaneously, this protection strategy only resulted in an average SOP reduction of 26%, indicating that while effectively improving the system's safety level, it also effectively maintained the battery's usable power capacity.
[0030] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the protection scope of this invention. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, without changing the core design of the algorithm and process, are also within the protection scope of this invention.
[0031] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.
[0032] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0033] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing the present invention, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of the present invention.
Claims
1. A method for estimating the peak power of a battery cluster, characterized in that, The method includes: Obtain time-series datasets of voltage, current, and temperature of each battery module in the battery cluster. The time-series datasets are divided into a support set, a query set, and a fine-tuning set. Construct a model for estimating the maximum available charge and discharge power of the battery module. The model consists of convolutional layers and fully connected layers. The model is trained in an inner loop using a support set. By minimizing the mean squared error loss between the predicted and actual power values, gradient descent is used to optimize all model parameters to obtain task-fit parameters. The task adaptation parameters are loaded into the model, and the model is trained in the outer loop using the query set. By minimizing the composite loss function composed of the mean square error and the residual of the electrothermal coupling equation, the meta-initialization parameters of the model are optimized by backpropagation to obtain meta-initialization parameters with cross-task generalization ability. The initialization parameters are loaded into the model, and the model is trained using a fine-tuning set. The parameters of the convolutional layers are kept constant, and the weights and biases of the fully connected layers are optimized by minimizing the composite loss function to obtain the trained battery module maximum available charge and discharge power estimation model, which is used to estimate the maximum available charge and discharge power of the battery module. The maximum available charge and discharge power of each battery module is fused to obtain the peak power of the battery cluster.
2. The battery cluster peak power estimation method as described in claim 1, characterized in that, It also includes safety correction for peak power of battery clusters, including the following steps: Real-time monitoring of the rate of change of battery temperature over time; If the rate of change exceeds a preset threshold, the peak power of the battery cluster is multiplied by a preset margin coefficient for safety correction.
3. The battery cluster peak power estimation method as described in claim 1, characterized in that, The step of optimizing all model parameters by gradient descent by minimizing the mean squared error loss between the predicted and actual power values specifically includes: The formula for mean squared error loss is as follows: ； In the formula, The mean squared error loss is a scalar value used to quantify the model's performance on the support set. On the predictive performance; Initialize parameters for the meta-initialization; To support the set; To represent the support set The total number of samples included; Indicates the support set Each data point in Perform subsequent error calculations and sum the calculation results for all data points; The input consists of time-series data of voltage, current, and temperature. The power prediction value output by the model; The power label represents the actual value of the data. The gradient descent optimization formula is as follows: ； In the formula, Adapt parameters for the task; The learning rate for the inner loop, taking values... ; For parameters The gradient; This is the mean square error loss; To support the set; Use the initialization parameters for the meta-initialization.
4. The battery cluster peak power estimation method as described in claim 1, characterized in that, The meta-initialization parameters of the backpropagation optimization model, which are formed by minimizing the composite loss function consisting of the mean square error and the residual of the electrothermal coupling equation, specifically include: The electrothermal coupling equation is as follows: ； ； In the formula, This is the physical residual loss; For query set; Adapt parameters for the task; The number of samples in the query set; The residuals of the electrothermal coupling equation; The power prediction value output by the model; For battery heat capacity; R is the rate of temperature change; I is the current; R is the battery internal resistance; T is the battery temperature. Ambient temperature; Battery thermal resistance; The input consists of time-series data of voltage, current, and temperature. The formula for the composite loss function is as follows: ； In the formula, The meta-loss function; The weighting coefficients for the physical law residuals; Adapt parameters for the task; For query set; This represents the physical residual loss.
5. The formula for updating the initial parameters of the meta-initializers using stochastic gradient descent is as follows: ； In the formula, The learning rate for the outer loop; Initialize parameters for the meta-initialization. Meta-loss function For meta-parameters The gradient; This is the meta-loss function.
6. The battery cluster peak power estimation method as described in claim 1, characterized in that, The process of integrating the maximum available charge and discharge power of each battery module to obtain the peak power of the battery cluster includes the following steps: The maximum available charge and discharge power of each battery module is aligned and fused along a unified time axis to generate a time series of the maximum available charge and discharge power for each battery module. The charge and discharge power time series is processed by a Long Short-Term Memory (LSTM) network to capture the dynamic dependence of the power of a single battery module over time, and outputs the hidden state that integrates the full time-series information of a single battery module. The hidden state of all battery modules is processed by a Transformer encoder module, and the power interaction and constraint relationship between different battery modules is captured by a self-attention mechanism, outputting a feature matrix that integrates global correlation information. The feature matrix is weighted and aggregated by an attention mechanism, tilting the weights toward the weakest battery module that has the greatest power limitation on the entire battery cluster. The decoder maps the aggregated features to the peak power of the battery cluster.
7. A computer system, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the battery cluster peak power estimation method as described in any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, is capable of performing the battery cluster peak power estimation method as defined in any one of claims 1 to 5.