Energy storage battery cluster temperature comprehensive perception method based on hybrid expert model

Abstract

This invention discloses a comprehensive temperature sensing method for energy storage battery clusters based on a hybrid expert model. The method includes: acquiring full thermal-electrical data of the target battery cluster; using the acquired full thermal-electrical data as input, classifying all cells using a clustering algorithm, selecting the central cell of the cluster, and installing a temperature sensor at the central cell to form a sparse temperature measurement arrangement; processing the acquired data using a sliding window with a preset time window length to construct a model input feature matrix; constructing a multi-source time-series feature matrix and a T-MoE model, using a multi-gated, multi-task structure for classification output; training the T-MoE model based on the main loss function and auxiliary loss function; and outputting the temperature of unmeasured cells or the temperature distribution of the entire cluster cells based on the trained T-MoE model. This invention enables high-precision, stable, and scalable sensing of the temperature distribution of large-scale energy storage battery clusters under low-sensor configuration conditions.

Description

Technical Field

[0001] This invention relates to the field of energy storage battery temperature measurement technology, and in particular to a method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model. Background Technology

[0002] Currently, in large-scale battery energy storage systems, battery clusters are typically composed of multiple battery packs connected in series, with each pack containing a large number of lithium-ion cells. Batteries continuously generate heat during charging and discharging, and their temperature levels and distribution directly affect charge / discharge performance, capacity decay rate, cycle life, consistency, and system safety. Especially in energy storage power station scenarios, battery clusters are large in scale, have long operating cycles, and exhibit significant fluctuations in operating conditions. Failure to promptly monitor internal temperature changes within the battery cluster can easily lead to localized overheating, increased temperature differences, and accelerated performance degradation, potentially even inducing thermal runaway. Therefore, how to effectively, economically, and comprehensively sense the temperature of battery clusters has become a crucial technical challenge in the field of energy storage thermal management and safety monitoring.

[0003] Existing temperature sensing solutions mainly include the following: First, direct temperature measurement requires the deployment of many temperature sensors, making it difficult to balance cost and comprehensiveness; second, under sparse temperature measurement conditions, there is a lack of scientific methods for selecting representative measurement points, resulting in insufficient representativeness of the measurement points; third, traditional mechanistic models are highly dependent on parameters and difficult to apply online; and fourth, existing data-driven models are difficult to adapt to scenarios with high-dimensional input, complex nonlinear coupling, and differences in thermal behavior among different cells in large-scale energy storage battery clusters.

[0004] For the specific scenario of energy storage battery clusters, the aforementioned existing technologies can usually obtain electrical information such as the total cluster current and the voltage of all cells relatively easily. However, due to limitations such as cost, installation space, reliability, and ease of maintenance, temperature sensors can often only be installed on a small number of cells or a few key locations. Therefore, existing solutions generally face the following problems: insufficient representativeness of temperature measurement points, incomplete temperature perception of the entire cluster, insufficient ability to model complex thermal behavior, uneven utilization of experts leading to unstable training, and insufficient generalization ability across operating conditions. Summary of the Invention

[0005] The main objective of this invention is to provide a comprehensive temperature sensing method for energy storage battery clusters based on a hybrid expert model, in order to solve the technical problems in the prior art.

[0006] To achieve the above objectives, this invention provides a method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model, the method comprising the following steps: S10, during the development, testing or debugging phase of energy storage battery clusters, performs full thermal-electric data acquisition on the target battery cluster; S20, the collected full amount of thermal-electric data is used as input, a clustering algorithm is used to classify all the cells, and the center cell of the cluster is selected. A temperature sensor is installed at the center cell of the cluster to form a sparse temperature measurement arrangement scheme. S30, Perform sliding window processing on the collected data with a preset time window length to construct the model input feature matrix; S40, construct a multi-source temporal feature matrix and build a T-MoE model, and use a multi-gated multi-task structure for classification output; S50, The T-MoE model is trained based on the main loss function and the auxiliary loss function; S60, based on the trained T-MoE model, outputs the temperature of the unmeasured cells or the temperature distribution of the entire cell cluster.

[0007] Optionally, the full thermal-electric data includes: the temperature time series of each cell in the battery cluster during continuous operation, the voltage time series of each cell in the battery cluster, and the cluster-level current time series of the battery cluster.

[0008] Optionally, step S20 includes the following steps: S210, the collected temperature time series is used as input, the dynamic time warping distance is used to measure the similarity between the temperature rise trajectories of different cells, and the DTW-K clustering algorithm is used to classify all cells. S220, in each category, select the central cell that best represents the temperature evolution characteristics of that category as the cluster central cell; S230, in the actual engineering deployment phase, a temperature sensor is retained or installed at the selected cluster center cell to form a sparse temperature measurement arrangement scheme.

[0009] Optionally, step S30 includes the following steps: S310, Select input features; S320, during the online operation phase, the collected full amount of thermal-electric data is processed by a sliding window with a preset time window length to construct the model input feature matrix; S330, based on the input feature matrix of the model, calculate the temperature of the battery cell without sensors.

[0010] Optionally, step S40 includes the following steps: S410, construct a multi-source temporal feature matrix; S420, the constructed multi-source temporal feature matrix is ​​input into a preset feature embedding module; S430 projects the original input onto a high-dimensional feature space through linear mapping, gated linear units, and nonlinear activation functions to obtain a sequence representation suitable for deep network processing. S440, The time-series representation is input into the encoder to model the dependencies between different time steps and between different feature dimensions in order to construct a T-MoE model; S450, a multi-gated multi-task learning structure is set on the output side of the T-MoE model to define cell temperature estimation for different thermal behavior categories as different tasks.

[0011] Optionally, step S50 includes the following steps: S510 uses temperature estimation error as the main loss function; S520 uses differentiable load balancing loss or other loss forms that can reflect the degree of balance in expert utilization as auxiliary loss functions. S530, The T-MoE model is trained based on the main loss function and the auxiliary loss function.

[0012] Optionally, the expressions for the main loss function and the auxiliary loss function are as follows:

[0013]

[0014] In the formula: f i r represents the token share allocated to expert i. i It represents the route probability share assigned to expert i, II represents the indicator function, K represents the top k highest-scoring experts, q is the time step, and s is the number of steps. i，t The expert activation score is calculated, and L represents the final loss L,L. MSE L represents the loss associated with the estimated task. DLB This represents the differentiable load balancing loss.

[0015] Optionally, step S60 includes the following steps: S610, deploy the trained T-MoE model on the battery management system, edge controller, energy storage station server or cloud platform; The S620 collects real-time information on the temperature of representative measurement points, the voltage of all cells, and the cluster current during operation. S630, based on the representative measurement point temperature, all cell voltages and cluster current information, outputs the temperature of unmeasured cells or the temperature distribution of the entire cluster cells through the deployed T-MoE model.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: The comprehensive temperature sensing method for energy storage battery clusters based on a hybrid expert model provided in this application addresses the problem of insufficient representativeness of sparse temperature measurement in existing technologies by classifying all cells using a clustering algorithm and selecting the central cell of the cluster. It also solves the problem of incomplete temperature sensing of the entire cluster in existing technologies by constructing a T-MoE model. Furthermore, it addresses the difficulty in uniformly and accurately modeling complex thermal behaviors in existing technologies by using hybrid experts and a multi-gated, multi-task structure for classification output. Finally, it further improves the problems of unstable training and insufficient generalization ability across operating conditions in existing technologies through load balancing constraints and migration deployment mechanisms. Therefore, this invention can achieve high-precision, stable, and scalable sensing of the temperature distribution of large-scale energy storage battery clusters under low-sensor configuration conditions, demonstrating significant technological advancement and engineering application value. Attached Figure Description

[0017] Figure 1 A flowchart of a method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model, provided in an embodiment of this application; Figure 2 A schematic diagram of full thermal-electrical data based on a hybrid expert model provided for embodiments of this application; Figure 3 This is a schematic diagram of the structure of the T-MoE model provided in the embodiments of this application.

[0018] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the application. Rather, these embodiments are provided to make the disclosure more thorough and complete, and to fully convey the scope of the disclosure to those skilled in the art.

[0020] To address the aforementioned technical problems, embodiments of this application provide a method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model, such as... Figure 1 As shown, the method may include the following steps: S10 performs full thermal-electric data acquisition on the target battery cluster during the development, testing, or commissioning phase of the energy storage battery cluster.

[0021] The full thermal-electric data includes: the temperature time series of each cell in the battery cluster during continuous operation, the voltage time series of each cell in the battery cluster, and the cluster-level current time series of the battery cluster.

[0022] In this embodiment, relatively complete true temperature data is collected during the prototype stage for subsequent temperature behavior clustering and representative measurement point determination. Therefore, this embodiment can identify the actual thermal behavior categories inside the battery cluster from the data level, providing a basis for reducing the number of sensors while maintaining representativeness.

[0023] S20, the collected full amount of thermal-electric data is used as input, and a clustering algorithm is used to classify all the battery cells. The center battery cell of the cluster is selected, and a temperature sensor is installed at the center battery cell of the cluster to form a sparse temperature measurement arrangement scheme.

[0024] In an exemplary embodiment, step S20 may specifically include the following steps: S210, the collected temperature time series is used as input, the similarity between the temperature rise trajectories of different cells is measured by the dynamic time warping distance, and the DTW-K (Dynamic time warping and K-means, DTW-K) clustering algorithm is used to classify all cells.

[0025] The key to the DTW-K clustering algorithm lies in its cluster analysis of battery pack temperature data. Since battery temperature data is essentially time-series data, a clustering algorithm suitable for time-series data must be used. DTW-K is a dynamic time warping combined with k-means clustering method specifically designed for this type of data. Its core method enhances traditional time-series clustering by improving the distance metric and the cluster center update process.

[0026] S220, in each category, select the central cell that best represents the temperature evolution characteristics of that category as the cluster central cell; S230, in the actual engineering deployment phase, a temperature sensor is retained or installed at the selected cluster center cell to form a sparse temperature measurement arrangement scheme.

[0027] Specifically, the cluster center cell can be defined as the cell closest to the cluster center of the category, or as the representative cell with the best overall similarity within the category.

[0028] In addition, the number of temperature sensors corresponds to the number of cluster categories, that is, each thermal behavior category corresponds to at least one representative measurement point.

[0029] In a preferred embodiment, temperature sensors are placed only at a small number of representative cells, while the voltage of all cells and the current of the battery cluster continue to be collected.

[0030] Figure 2The voltage, temperature, and current curves of the battery cluster during a single charge-discharge cycle are presented. Throughout the charging and discharging process, electrochemical heat generation causes a continuous temperature rise. Upon reaching the cutoff voltage, the thermal management system rapidly cools the battery cells during the subsequent rest period. Since the cells within the battery pack operate uniformly, the overall temperature trend is consistent. However, the temperature trajectories are observed to diverge into two distinct groups (one exhibiting a larger temperature rise during charging and discharging, while the other shows a smaller rise). This bifurcation is attributed to uneven heat dissipation, resulting from differences in spatial layout and inconsistent thermal management performance, leading to a maximum temperature difference exceeding 5 degrees Celsius within the cluster.

[0031] This embodiment uses a "clustering first, point selection later, and sparse deployment" approach to enable a small number of sensors to have stronger information representation capabilities. Therefore, this invention can significantly reduce the number of sensors, reduce wiring and installation complexity, and reduce system costs and maintenance difficulty, while still maintaining high temperature estimation accuracy.

[0032] S30: Perform sliding window processing on the collected data with a preset time window length to construct the model input feature matrix.

[0033] In an exemplary embodiment, step S30 may specifically include the following steps: S310, Select input features; S320, during the online operation phase, the collected full amount of thermal-electric data is processed by a sliding window with a preset time window length to construct the model input feature matrix; S330, based on the input feature matrix of the model, calculate the temperature of the battery cell without sensors.

[0034] Specifically, the input features include: Temperature sequence T of a representative battery cell with a temperature sensor within the current time window. j The voltage sequence V of all cells within the battery cluster within the current time window. i The current sequence I of the battery cluster within the current time window. Among them, there are n cells with temperature sensors, m cells in the cluster, and mn cells without sensors.

[0035] These sequences are represented as: T j = [T j (t-q+1), T j (t-q+2), ..., T j (t), j [(1,n)],V i = [V i (t-q+1), V i (t-q+2), ..., Vi (t), i (1,m)], and I = [I(t-q+1), I(t-q+2), …, I(t)].

[0036] In the formula: T j This represents the temperature sequence of a representative battery cell with a temperature sensor within the current time window; V i I represents the voltage sequence of all cells in the battery cluster within the current time window; n represents the current sequence of the battery cluster within the current time window; m represents the number of cells with temperature sensors; mn represents the number of cells in the cluster; and mn represents the number of cells without sensors.

[0037] Each input sample X contains the temperature, voltage, and current within the current time window. The corresponding output sample Y represents the temperature of the battery cell that did not have a temperature sensor at the last step of that time window. Y = [T n+1 (t), T n+2 (t), ..., T m (t)] In the formula: Y represents the temperature of the battery cell without a temperature sensor in the last step of the time window, and m represents the number of cells in the cluster.

[0038] Specifically, the matrix is ​​formed along the time and feature dimensions. On the time axis, a sliding window WL (window length) of length q consecutive time steps is used to capture the temporal evolution of the signal. Training samples are obtained using a sliding window method with WL=q and step size s, segmenting the original time series into multiple overlapping samples. This method can capture the temporal evolution and spatial correlation between cells in the data representation. For example, when the sampling interval is 5 minutes and q=12, each window corresponds to 1 hour of historical data. Through this process, the raw sensor readings are transformed into structured temporal feature matrices, which represent both the spatial relationships between cells and their dynamic changes over time. Subsequently, these matrices are used as input to the T-MoE model to estimate the temperature of cells without sensors. Therefore, the battery pack temperature estimation task can be mathematically represented as: Y=f T-MoE (X) In the formula: f T-MoE This refers to the mapping relationship between the feature matrix X and the output (temperature of the remaining un-deployed sensors) Y, or the mapping relationship learned by training the T-MoE model.

[0039] This embodiment combines a small amount of real temperature information with all electrical quantities for joint modeling, and explicitly preserves the dynamic evolution process through a time window. Therefore, this embodiment can more fully explore the temporal coupling relationship between temperature, voltage and current, thereby enhancing the model's ability to express the dynamic temperature change process and improving the accuracy and robustness of the estimation results.

[0040] S40, construct a multi-source time series feature matrix and build a T-MoE model.

[0041] Specifically, such as Figure 3 As shown, the T-MoE model mainly includes an input feature matrix, an embedding layer, a Transformer encoder, and a task output layer. The input feature matrix consists of temperature sequences from sparse temperature measurement points, voltage sequences from all battery cells, and current sequences from the battery cluster. The input is first processed by the embedding layer to obtain a high-dimensional representation suitable for deep modeling. Subsequently, the feature representation is fed into a Transformer encoder with an M-layer structure. The first M-1 layers of the encoder sequentially include a multi-head self-attention module, a normalization module, and a MoE module. The MoE module scores multiple expert networks through a router and selects the Top-K experts to participate in the computation, achieving adaptive extraction of temporal features under different operating conditions and thermal behavior modes. The last layer of the encoder adopts an MMoE structure, which, based on a shared expert pool, sets up independent gating networks and task towers for different tasks, thereby achieving classified temperature estimation of battery cells with different thermal behavior categories. Finally, the model outputs the temperature results of battery cells without temperature sensors, thus achieving comprehensive perception of the temperature distribution of the energy storage battery cluster.

[0042] In an exemplary embodiment, step S40 may specifically include the following steps: S410, construct a multi-source temporal feature matrix; S420, the constructed multi-source temporal feature matrix is ​​input into a preset feature embedding module; S430 projects the original input onto a high-dimensional feature space through linear mapping, gated linear units, and nonlinear activation functions to obtain a sequence representation suitable for deep network processing. S440, The time-series representation is input into the encoder to model the dependencies between different time steps and between different feature dimensions in order to construct a T-MoE model; S450, a multi-gated multi-task learning structure is set on the output side of the T-MoE model, and the cell temperature estimation of different thermal behavior categories is defined as different tasks. Each task shares the underlying expert pool, but has an independent gating network and task output head, so as to establish differentiated mapping relationships for different categories of cells while sharing common knowledge.

[0043] For example, different temperature categories obtained from clustering can be assigned to different task heads, and each task head can output the temperature estimate of the untested cells in its category. Finally, the results of each task can be spliced ​​together to form the temperature estimate of the entire cluster.

[0044] In this exemplary embodiment, the specific steps for constructing the multi-source temporal feature matrix are as follows: First, the time series of temperature of representative cells, voltage of all cells, and current of the battery cluster are collected. Second, data from different sampling periods are synchronized to ensure all data correspond to a unified time axis. Then, using the current prediction time as a reference, a sliding time window is used to extract historical sequences of length q. The representative cell temperature, all cell voltage, and cluster current within this window are concatenated according to their feature dimensions to form an input feature matrix X, where the matrix dimension d = n + m + 1, where n is the number of representative cells and m is the number of cells. Finally, the actual temperature of cells without temperature sensors at the current time is used as the output label to form supervised learning samples. This method enables a unified representation of a small amount of real thermal information and the full amount of electrical information, preserving the dynamic evolution characteristics of the thermal and electrical states of the battery cluster, thus providing effective input for overall cluster temperature estimation.

[0045] This embodiment uses a multi-task output mechanism to treat cells of different thermal categories differently. Therefore, this embodiment can further learn the individual temperature rise patterns of various cells on the basis of sharing the overall thermal-electrical correlation knowledge, thereby improving the estimation accuracy of hot spots, edge areas and special thermal behavior areas.

[0046] Specifically, the T-MoE model is based on an encoder architecture, employs a multimodal extended architecture Transformer, and uses a multimodal extended attention mechanism head to achieve temperature detection for different types of batteries. (Reference) Figure 3 The encoder architecture consists of the following key components: an embedding layer, a Transformer encoder, and MMoE. The Transformer encoder has M layers and is able to extract deep features from temperature-voltage time series using MoE and MMoE mechanisms, effectively integrating cross-class knowledge, thereby enabling accurate estimation of the temperature of all clusters from a sparse subset of sensors.

[0047] The Transformer encoder includes an M-layer encoding unit, which can be set according to the task complexity, data size and deployment computing power. The value of M can be any positive integer from 2 to 6.

[0048] The Mixture of Expert (MoE) mechanism includes multiple expert sub-networks and a gated routing network. The gated routing network assigns weights to each expert based on the input features and selects at least one expert to participate in the feature extraction of the current sample, thereby achieving adaptive modeling of complex thermo-electric coupling relationships under different operating conditions and different thermal behavior modes.

[0049] The Multi-Gated Hybrid Expert Mechanism (MMoE) includes a shared expert pool, multiple task gating networks, and multiple task output heads. Different task gating networks weight and combine the shared expert pool to achieve common knowledge sharing and individual feature differentiation among cell temperature estimation tasks of different thermal behavior categories.

[0050] This embodiment employs a Transformer-based temporal coding structure to uniformly model the coupling relationship between temperature, voltage, and current over a long time span. Therefore, this embodiment can improve the ability to extract complex temporal features and provide a more effective intermediate representation for subsequent refined temperature estimation.

[0051] Furthermore, within the encoder, this embodiment replaces the traditional feedforward network with a hybrid expert network structure. The hybrid expert structure includes multiple expert sub-networks and corresponding gated routing networks. The gated routing network assigns weights to each expert based on the current input features, and preferably selects the experts with the highest scores to participate in the calculation of the current sample.

[0052] It is important to note that in this invention, "expert" refers to a sub-network unit in the hybrid expert model (T-MoE) used for mapping or feature extraction of input features under specific patterns. Each expert receives the same or partially the same input features and outputs corresponding intermediate feature representations. The router or gating network assigns weights to each expert based on the current input features, or selects one or more experts to participate in the calculation. Finally, the outputs of the selected experts are weighted and fused to obtain the output result of this layer.

[0053] Each expert subnetwork can learn the feature mapping patterns under different types of input patterns, for example: Thermoelectric relationship under different power fluctuation characteristics; Local temperature rise patterns of battery cells with different thermal behavior categories; Dynamic changes at different operational stages.

[0054] Most existing temperature estimation models use a single network to uniformly fit all input patterns. However, the thermal behavior of energy storage battery clusters is complex and diverse, making it difficult for a single network to account for all scenarios. This embodiment uses a hybrid expert structure to break down the complex problem into multiple relatively specialized sub-problems and uses a gating mechanism to call the most suitable expert as needed. Through this improvement, the present invention can significantly enhance the model's adaptability to complex nonlinear relationships and diverse operating conditions, thereby improving the problem of insufficient temperature estimation accuracy in large-scale energy storage scenarios.

[0055] S50, The T-MoE model is trained based on the main loss function and the auxiliary loss function.

[0056] In an exemplary embodiment, step S50 may specifically include the following steps: S510 uses temperature estimation error as the main loss function; S520 uses differentiable load balancing loss or other loss forms that can reflect the degree of balance in expert utilization as auxiliary loss functions. S530, The T-MoE model is trained based on the main loss function and the auxiliary loss function.

[0057] During the model training phase, this embodiment introduces an auxiliary loss function to constrain the balance of expert usage. This auxiliary loss is used to avoid a situation where a few experts are frequently called upon while other experts remain idle for a long time during training, thereby improving the common expert load imbalance problem in hybrid expert models.

[0058] This embodiment introduces a load balancing constraint to regularize expert routing. Through this constraint, this embodiment can improve the stability of model training, promote each expert to fully learn effective features, and ultimately improve the accuracy and consistency of temperature estimation.

[0059] Specifically, the expressions for the main loss function and the auxiliary loss function are as follows:

[0060]

[0061] In the formula: f i r represents the token share allocated to expert i. i It represents the route probability share assigned to expert i, II represents the indicator function, K represents the top k highest-scoring experts, q is the time step, and s is the number of steps. i，t The expert activation score is calculated, and L represents the final loss L,L. MSE L represents the loss associated with the estimated task. DLB This represents the differentiable load balancing loss.

[0062] S60, based on the trained T-MoE model, outputs the temperature of the unmeasured cells or the temperature distribution of the entire cell cluster.

[0063] In an exemplary embodiment, step S60 may specifically include the following steps: S610, deploy the trained T-MoE model on the battery management system, edge controller, energy storage station server or cloud platform; The S620 collects real-time information on the temperature of representative measurement points, the voltage of all cells, and the cluster current during operation. S630, based on the representative measurement point temperature, all cell voltages and cluster current information, outputs the temperature of unmeasured cells or the temperature distribution of the entire cluster cells through the deployed T-MoE model.

[0064] In this embodiment, for different battery clusters or different energy storage sites, transfer learning can be used to quickly adapt and deploy new objects by only fine-tuning the model's subsequent parameters, task header parameters, or a small number of adaptation layers.

[0065] The improvement of this embodiment over the prior art lies in: Existing models are often trained individually for a single object, exhibiting weak cross-cluster transfer capabilities and difficulty in generalizing to other similar battery clusters. This embodiment, however, provides a transfer deployment mechanism. Through this improvement, the present invention can reduce the training cost when deploying new battery clusters, thereby enhancing engineering promotion efficiency and applicability.

[0066] In the embodiments provided by this invention, by employing a clustering algorithm to classify all battery cells and selecting the central cell of the cluster, the problem of insufficient representativeness of sparse temperature measurement in existing technologies is solved; by constructing a T-MoE model, the problem of incomplete temperature sensing of the entire cluster in existing technologies is solved; by using hybrid experts and a multi-gated multi-task structure for classification output, the problem of difficulty in uniformly and accurately modeling complex thermal behaviors in existing technologies is solved; and by using load balancing constraints and migration deployment mechanisms, the problems of unstable training and insufficient generalization ability across operating conditions in existing technologies are further improved. Therefore, this invention can achieve high-precision, stable, and scalable sensing of the temperature distribution of large-scale energy storage battery clusters under low sensor configuration conditions, and has significant technological advancements and engineering application value.

[0067] In the description of this application, it should be noted that the terms "first", "second", and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0068] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0069] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0070] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0071] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0072] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. 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.

[0073] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the technical scope disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

[0074] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

Claims

1. A method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model, characterized in that, The method includes the following steps: S10, during the development, testing or debugging phase of energy storage battery clusters, performs full thermal-electric data acquisition on the target battery cluster; S20, the collected full amount of thermal-electric data is used as input, a clustering algorithm is used to classify all the cells, and the center cell of the cluster is selected. A temperature sensor is installed at the center cell of the cluster to form a sparse temperature measurement arrangement scheme. S30, Perform sliding window processing on the collected data with a preset time window length to construct the model input feature matrix; S40, construct a multi-source temporal feature matrix and build a T-MoE model, and use a multi-gated multi-task structure for classification output; S50, The T-MoE model is trained based on the main loss function and the auxiliary loss function; S60, based on the trained T-MoE model, outputs the temperature of the unmeasured cells or the temperature distribution of the entire cell cluster.

2. The method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model according to claim 1, characterized in that, The full thermal-electric data includes: the temperature time series of each cell in the battery cluster during continuous operation, the voltage time series of each cell in the battery cluster, and the cluster-level current time series of the battery cluster.

3. The method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model according to claim 2, characterized in that, Step S20 includes the following steps: S210, the collected temperature time series is used as input, the dynamic time warping distance is used to measure the similarity between the temperature rise trajectories of different cells, and the DTW-K clustering algorithm is used to classify all cells. S220, in each category, select the central cell that best represents the temperature evolution characteristics of that category as the cluster central cell; S230, in the actual engineering deployment phase, a temperature sensor is retained or installed at the selected cluster center cell to form a sparse temperature measurement arrangement scheme.

4. The method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model according to claim 1, characterized in that, Step S30 includes the following steps: S310, Select input features; S320, during the online operation phase, the collected full amount of thermal-electric data is processed by a sliding window with a preset time window length to construct the model input feature matrix; S330, based on the input feature matrix of the model, calculate the temperature of the battery cell without sensors.

5. The method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model according to claim 1, characterized in that, Step S40 includes the following steps: S410, construct a multi-source temporal feature matrix; S420, the constructed multi-source temporal feature matrix is ​​input into a preset feature embedding module; S430 projects the original input onto a high-dimensional feature space through linear mapping, gated linear units, and nonlinear activation functions to obtain a sequence representation suitable for deep network processing. S440, The time-series representation is input into the encoder to model the dependencies between different time steps and between different feature dimensions in order to construct a T-MoE model; S450, a multi-gated multi-task learning structure is set on the output side of the T-MoE model to define cell temperature estimation for different thermal behavior categories as different tasks.

6. The method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model according to claim 1, characterized in that, Step S50 includes the following steps: S510 uses temperature estimation error as the main loss function; S520 uses differentiable load balancing loss or other loss forms that can reflect the degree of balance in expert utilization as auxiliary loss functions. S530, The T-MoE model is trained based on the main loss function and the auxiliary loss function.

7. The method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model according to claim 6, characterized in that, The expressions for the main loss function and the auxiliary loss function are as follows: In the formula: f i r represents the token share allocated to expert i. i It represents the route probability share assigned to expert i, II represents the indicator function, K represents the top k highest-scoring experts, q is the time step, and s is the number of steps. i，t The expert activation score is calculated, and L represents the final loss L,L. MSE L represents the loss associated with the estimated task. DLB This represents the differentiable load balancing loss.

8. The method for comprehensive temperature sensing of energy storage battery clusters based on a hybrid expert model according to claim 1, characterized in that, Step S60 includes the following steps: S610, deploy the trained T-MoE model on the battery management system, edge controller, energy storage station server or cloud platform; The S620 collects real-time information on the temperature of representative measurement points, the voltage of all cells, and the cluster current during operation. S630, based on the representative measurement point temperature, all cell voltages and cluster current information, outputs the temperature of unmeasured cells or the temperature distribution of the entire cluster cells through the deployed T-MoE model.

Images