Lithium battery energy storage system thermal energy management optimization method and system based on deep learning

Abstract

The application discloses a lithium battery energy storage system thermal energy management optimization method and system based on deep learning, and relates to the technical field of energy storage systems. The method comprises the following steps: constructing a contact boundary data set; dividing the service stage based on the contact boundary data set to form a contact pressure space atlas; converting the contact pressure space atlas into a contact thermal resistance space redistribution according to the calibration relationship between the pressure and the contact thermal resistance in the thermal pad parameters; determining the cell heat generation distribution of each service stage based on the operating condition data; predicting the temperature field distribution of each preset candidate flow channel layout in each service stage; identifying the hot spot sensitive area based on the temperature field distribution; performing life weighted evaluation on each preset candidate flow channel layout according to the duration of each service stage, the safety risk and the economic life; and performing multi-objective collaborative optimization on the main flow channel, the auxiliary flow channel and the adhesion reinforcement area of the liquid cooling plate. The application improves the temperature control uniformity of the lithium battery energy storage system during the service process.

Description

Technical Field

[0001] This application relates to the field of energy storage system technology, and more specifically, to a method and system for optimizing thermal energy management of lithium battery energy storage systems based on deep learning. Background Technology

[0002] Lithium-ion battery energy storage systems, with their advantages of high energy density and long cycle life, have been widely used in scenarios such as grid connection of new energy power generation, grid frequency regulation and peak shaving, and industrial and commercial energy storage. Energy storage systems consist of a large number of battery cells cascaded through modules and battery clusters, continuously generating heat during long-term charge-discharge cycles. Temperature is a key factor affecting the consistency and safety of battery cell performance; excessive temperature differences between cells can accelerate localized aging, exacerbate inconsistent capacity decay, and in severe cases, may even induce thermal runaway. Therefore, thermal management systems are of great significance for ensuring the safe operation of energy storage lithium-ion battery systems and extending their service life.

[0003] Currently, the mainstream thermal management methods for lithium-ion battery energy storage systems include air cooling and liquid cooling. Among them, liquid cooling technology has become an important choice for large-capacity energy storage systems due to its advantages such as high heat exchange efficiency and good temperature control accuracy. As the core heat exchange component between the liquid cooling system and the battery cell, the layout design of the liquid cooling plate directly affects the temperature distribution and cooling uniformity inside the module. However, existing liquid cooling plate layout optimization methods are usually based on thermal simulation design of the initial geometric dimensions of the battery cell and stable assembly state, and the optimization targets are mostly indicators such as the maximum temperature, maximum temperature difference, and pressure drop in the initial state. In actual long-term service, factors such as cyclic expansion of the battery cell, compression creep of the thermal pad, and decay of preload force will cause spatial redistribution of the contact pressure between the liquid cooling plate and the battery cell, which will lead to changes in contact thermal resistance. This may cause areas with good cooling effect in the initial state to have insufficient cooling capacity in the later stages of service.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a method and system for optimizing thermal energy management of lithium battery energy storage systems based on deep learning, in order to solve the above-mentioned technical problems.

[0006] This application provides a deep learning-based method for optimizing thermal management of lithium battery energy storage systems, including: S1, acquiring initial cell dimensions, cell cyclic expansion characteristics, thermal pad parameters, assembly deviations, preload, and operating condition data to construct a contact boundary dataset; S2, dividing the service stage based on the contact boundary dataset, determining the cell expansion, thermal pad compression, and preload attenuation in each service stage, and establishing a contact pressure spatial mapping of the liquid cooling plate contact surface based on the assembly deviations to form a contact pressure spatial map; S3, converting the contact pressure spatial map into a contact thermal resistance spatial redistribution based on the calibration relationship between pressure and contact thermal resistance in the thermal pad parameters; S4, based on... The operating condition data determines the cell heat distribution at each service stage; the contact thermal resistance spatial redistribution, the cell heat distribution, the operating condition data, and multiple preset candidate flow channel layouts are input into a trained deep learning model to predict the temperature field distribution of each preset candidate flow channel layout at each service stage; S5, based on the temperature field distribution, hotspot sensitive areas are identified, and each preset candidate flow channel layout is evaluated with lifespan weighting according to the duration, safety risk, and economic lifespan of each service stage to determine the optimal candidate layout; S6, based on the optimal candidate layout and the hotspot sensitive areas, multi-objective collaborative optimization is performed on the main flow channel, auxiliary flow channel, and bonding reinforcement area of ​​the liquid cooling plate.

[0007] This application provides a deep learning-based thermal management optimization system for lithium battery energy storage systems, including: a dataset construction module for acquiring initial cell dimensions, cell cyclic expansion characteristics, thermal pad parameters, assembly deviations, preload, and operating condition data to construct a contact boundary dataset; a map construction module for dividing the service stage based on the contact boundary dataset, determining the cell expansion, thermal pad compression, and preload attenuation in each service stage, and establishing a contact pressure spatial mapping of the liquid cooling plate contact surface based on the assembly deviations to form a contact pressure spatial map; a conversion module for converting the contact pressure spatial map into a contact thermal resistance spatial redistribution based on the calibration relationship between pressure and contact thermal resistance in the thermal pad parameters; and a prediction module for using... The system determines the cell heat distribution at each service stage based on the operating condition data; it inputs the contact thermal resistance spatial redistribution, the cell heat distribution, the operating condition data, and multiple preset candidate flow channel layouts into a trained deep learning model to predict the temperature field distribution of each preset candidate flow channel layout at each service stage; an optimal layout determination module is used to identify hotspot sensitive areas based on the temperature field distribution, and to perform a life-weighted evaluation of each preset candidate flow channel layout according to the duration, safety risk, and economic life of each service stage to determine the optimal candidate layout; a multi-objective optimization module is used to perform multi-objective collaborative optimization of the main flow channel, auxiliary flow channel, and bonding reinforcement area of ​​the liquid cooling plate based on the optimal candidate layout and the hotspot sensitive areas.

[0008] Based on the embodiments provided in this application, by acquiring data on initial cell dimensions, cyclic expansion characteristics, thermal pad parameters, assembly deviations, preload, and operating conditions, and constructing a contact boundary dataset, a data foundation is provided for the changes in contact state during long-term service. This avoids the problem of insufficient data support caused by prior art relying solely on the initial geometric dimensions and stable assembly state of the cell for design. By dividing the service stage based on the contact boundary dataset and determining the cell expansion, thermal pad compression, and preload attenuation in each stage, and combining assembly deviations to establish a spatial mapping of contact pressure on the liquid cooling plate contact surface and form a spectrum, the evolution law of contact pressure over service time can be determined. This transforms the change in contact state between the liquid cooling plate and the cell from an unpredictable gradual process into an analyzable staged spectrum, thereby solving the problem of design benchmark inaccuracies caused by prior art neglecting long-term changes in contact state. By transforming the spatial spectrum of contact pressure into a spatial redistribution of contact thermal resistance based on the calibration relationship between pressure and contact thermal resistance in the parameters of the thermal pad, the impact of changes in contact state on heat transfer performance can be evaluated. This provides calculable boundary conditions for identifying areas where heat transfer capacity declines in the later stages of service, thus solving the problem of hot spot prediction bias caused by assuming a fixed value for contact thermal resistance in existing technologies.

[0009] By inputting contact thermal resistance spatial redistribution, cell heat distribution, operating condition data, and candidate flow channel layouts into a trained deep learning model to predict the temperature field distribution of each candidate layout at each service stage, the temperature prediction model can reflect the impact of contact state degradation on temperature distribution during long-term service. This solves the problem of existing thermal simulation methods failing to predict hot spots later due to evaluating cooling effects only based on the initial state. By identifying hot spot-sensitive areas based on temperature field distribution and performing a lifetime-weighted evaluation of each candidate layout according to the duration of each service stage, safety risks, and economic lifespan, the optimal candidate layout is determined. This ensures that the determination of the optimal layout no longer depends solely on the cooling effect in the initial state, but comprehensively considers the duration of each stage, safety risks, and economic lifespan. This solves the problem of existing technologies mistakenly selecting layouts that perform best in the initial state but fail to cover hot spots later in the lifespan as the optimal solution. By performing multi-objective collaborative optimization of the main flow channel, auxiliary flow channel, and bonding reinforcement area of ​​the liquid cooling plate based on the optimal candidate layout and hotspot sensitive areas, the flow channel layout design directly targets the real hotspot areas that appear at each stage of the entire life cycle. This solves the problem that the existing flow channel layout is only designed based on the initial uniform bonding state, resulting in insufficient cooling capacity in the later stages of service. Attached Figure Description

[0010] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart of an optional deep learning-based thermal management optimization method for lithium battery energy storage systems according to an embodiment of this application; Figure 2 This is a flowchart of an optional method for generating spatial redistribution of contact thermal resistance according to an embodiment of this application; Figure 3 This is a flowchart illustrating an optional method for identifying newly added hotspot sensitive locations according to an embodiment of this application; Figure 4 This is a flowchart illustrating an optional division of labor for processing hotspot identification results according to an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application.

[0011] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0012] 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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0013] With the rapid development of new energy power generation grid connection, peak shaving and valley filling, frequency and voltage regulation, and industrial and commercial energy storage scenarios, lithium battery energy storage systems are evolving towards larger capacity, higher integration, longer lifespan, and higher safety. Energy storage batteries typically consist of a large number of square cells, battery modules, and battery clusters, which continuously generate heat during long-term charge-discharge cycles. If this heat cannot be dissipated in a timely and uniform manner, it can easily lead to increased temperature differences between cells, inconsistent capacity decay, accelerated localized aging, and in severe cases, may even induce thermal runaway and its propagation. Therefore, thermal management systems have become a crucial element in the design of the safety, reliability, and economic lifespan of energy storage lithium battery systems.

[0014] Currently, common thermal management methods for energy storage lithium battery systems include natural heat dissipation, air cooling, liquid cooling, and phase change material-assisted heat dissipation. Compared to air cooling, liquid cooling thermal management has advantages such as high heat exchange efficiency, good temperature control accuracy, and high system integration, and has gradually become an important thermal management method for large-capacity energy storage systems. As the main heat exchange component between the liquid cooling system and the battery cell, the installation position, flow channel structure, coverage area, heat exchange area, and contact state with the battery cell of the liquid cooling plate directly affect the internal temperature distribution and cooling uniformity of the module.

[0015] Existing liquid cooling plate layout optimization methods typically rely on thermal simulations based on the initial geometry of the battery cell, uniform thermal pad thickness, stable assembly pressure, and fixed contact thermal resistance. Optimization targets include maximum temperature, maximum temperature difference, pressure drop, and pump power consumption to determine the liquid cooling plate location and flow channel distribution. However, during actual long-term service, the battery cell expands due to cyclic aging, the thermal pad undergoes compressive creep and reduced rebound capacity, the preload of end plates, straps, or bolts gradually decreases, and module assembly tolerances further amplify localized fit differences. These factors cause spatial migration of the contact pressure between the liquid cooling plate and the battery cell, leading to a redistribution of contact thermal resistance. This results in insufficient cooling capacity, hot spot coverage failure, or localized heat exchange bottlenecks in areas that were initially effective in the early stages of the battery's lifespan.

[0016] Meanwhile, the operational data of energy storage lithium battery systems exhibits characteristics of multi-source heterogeneity, spatiotemporal coupling, and nonlinear variation. Relying solely on traditional thermal simulations or empirical rules makes it difficult to identify hotspot-sensitive areas in the later stages of the battery's lifespan. Deep learning demonstrates strong capabilities in temperature field prediction, anomaly identification, multi-source data fusion, and intelligent optimization decision-making, providing a new technical approach for optimizing the thermal management of energy storage systems. Therefore, it is necessary to propose a thermal management optimization method and system that integrates full-lifespan contact boundary modeling, contact thermal resistance redistribution analysis, deep learning temperature field prediction, and liquid cooling plate layout optimization to improve the long-term operational safety and lifespan consistency of energy storage lithium battery systems.

[0017] According to one aspect of the embodiments of this application, such as Figure 1 As shown, this application provides a deep learning-based method for optimizing thermal management of lithium battery energy storage systems, including: S1. Obtain the initial size of the battery cell, the cyclic expansion characteristics of the battery cell, the parameters of the thermal pad, the assembly deviation, the preload force and the operating condition data, and construct the contact boundary dataset; In this embodiment, data on initial cell dimensions, cyclic expansion characteristics, thermal pad material parameters, assembly deviations, liquid cooling plate installation boundaries, end plate preload, and operating conditions are collected to construct a full-lifetime contact boundary basic dataset. This full-lifetime contact boundary basic dataset can also be called a contact boundary dataset, which provides a data foundation for subsequent thermal management optimization.

[0018] S2, based on the contact boundary dataset, divide the service stages, determine the cell expansion, thermal pad compression and preload attenuation of each service stage, and establish a contact pressure spatial mapping of the liquid cooling plate contact surface in combination with assembly deviations to form a contact pressure spatial map. In this embodiment, a mapping relationship is established between cell expansion, decreased resilience of the thermal pad, attenuation of preload, and spatial distribution of contact pressure, forming a spatial spectrum of contact pressure on the liquid cooling plate contact surface under different service stages. This spatial spectrum of contact pressure, which is the result of the above-mentioned spatial mapping of contact pressure, can reflect the process by which the bonding state between the liquid cooling plate and the cell gradually evolves from initial uniform bonding to local weak bonding, local high-pressure bonding, or local detachment.

[0019] S3, based on the calibration relationship between pressure and contact thermal resistance in the thermal pad parameters, the contact pressure spatial spectrum is transformed into a contact thermal resistance spatial redistribution. In this embodiment, based on the calibration relationship between the thermal pad pressure and the contact thermal resistance, the change in contact pressure is converted into a spatial redistribution of contact thermal resistance; this spatial redistribution of contact thermal resistance can also be referred to as the aforementioned spatial redistribution of contact thermal resistance. This step can identify key areas where heat exchange capacity declines in the later stages of the lifespan.

[0020] S4. Determine the cell heat distribution at each service stage based on operating condition data; input the contact thermal resistance spatial redistribution, cell heat distribution, operating condition data, and multiple preset candidate flow channel layouts into the trained deep learning model to predict the temperature field distribution of each preset candidate flow channel layout at each service stage. In this embodiment, by combining cell heat distribution, operating conditions and historical temperature data, a deep learning model is used to predict temperature field changes and hotspot sensitive areas during the life cycle; wherein, the temperature field changes during the life cycle refer to the temperature field distribution of each preset candidate flow channel layout at each service stage, and the prediction results provide a basis for subsequent liquid cooling plate layout optimization.

[0021] S5 identifies hotspot sensitive areas based on temperature field distribution, and performs a life-weighted evaluation of each preset candidate flow channel layout based on the duration, safety risks and economic life of each service stage to determine the optimal candidate layout. In this embodiment, S5 can be further refined into two sub-steps: hotspot sensitive area identification and lifespan weighted evaluation. First, the contact thermal resistance redistribution results at different service stages are analyzed in conjunction with the cell heating distribution to identify areas prone to concentrated temperature rise under long-term service conditions, thus identifying hotspot sensitive areas. Then, for each candidate liquid cooling plate layout, its cooling effect is calculated at the initial, mid-term, and late-term service stages, and a weighted evaluation is performed based on the duration of each stage, safety risk weight, and economic lifespan impact weight to determine the optimal candidate layout. The safety risk weight and economic lifespan impact weight can also be referred to as the aforementioned safety risk and economic lifespan, respectively.

[0022] S6 performs multi-objective collaborative optimization of the main flow channel, auxiliary flow channel and bonding reinforcement area of ​​the liquid cooling plate based on the optimal candidate layout and hotspot sensitive areas.

[0023] In this embodiment, based on the lifetime-weighted cooling effectiveness evaluation results, the main flow channel, auxiliary flow channel, ineffective flow channel reduction region, and bonding reinforcement region of the liquid cooling plate are optimized to generate thermal management optimization results for long-term operation. The aforementioned ineffective flow channel reduction region refers to the flow channel region with low heat transfer contribution and large pressure drop loss during long-term operation, and can be considered as one of the factors in multi-objective collaborative optimization. Through this optimization, the liquid-cooled thermal management system is transformed from "initial static design" to "full-life-cycle dynamic optimization design".

[0024] Furthermore, S1 includes: The acquired data on initial cell dimensions, cell cyclic expansion characteristics, thermal pad parameters, assembly deviations, preload, and operating conditions are cleaned, normalized, synchronized in time, and processed in spatial coordinates to construct a contact boundary dataset. The initial dimensions of the battery cell include: initial thickness, length, width, and material thermal parameters. The cyclic expansion characteristics of the battery cell include: cyclic expansion curve, heating characteristics under different states of charge and health, degree of cell expansion, and direction of cell expansion; The parameters of the thermal pad include: initial thickness, compressive modulus, creep characteristics, resilience, thermal conductivity and the calibration relationship between pressure and contact thermal resistance, as well as the degree of compression aging of the thermal pad. Assembly deviations include: module assembly tolerances, liquid cooling plate mounting boundaries, and contact surface geometric parameters; Preload includes: end plate preload, strap tension, and bolt tightening force; Operating data includes: current, voltage, SOC, SOH, cell cycle count, capacity retention, internal resistance change, temperature, liquid cooling flow rate, inlet and outlet temperature difference, coolant pressure, ambient temperature, and charge / discharge rate during the operation of the energy storage system.

[0025] Specifically, the data collected includes the initial thickness, length, width, material thermal parameters, cyclic expansion curves, and heating characteristics under different states of charge and health conditions of the battery cells; the initial thickness, compressive modulus, creep characteristics, resilience, thermal conductivity, and pressure-thermal resistance calibration data of the thermal pads; end plate preload, strap tension, bolt tightening force, assembly tolerances, liquid cooling plate installation boundaries, and contact surface geometric parameters during module assembly; and operational data such as current, voltage, SOC, SOH, temperature, liquid cooling flow rate, inlet and outlet temperature difference, coolant pressure, ambient temperature, and charge / discharge rate during the energy storage system's operation. Through data cleaning, normalization, time synchronization, and spatial coordinate unification, a basic dataset suitable for thermal management optimization is formed, which is the aforementioned contact boundary dataset. The thermal pad compression aging degree is used to characterize the performance degradation of the thermal pad under long-term compression, serving as one of the criteria for classifying service stages.

[0026] Data cleaning includes removing outliers caused by abnormal sensor fluctuations, interpolating and repairing missing values, and removing duplicate records. Normalization includes unifying current, voltage, temperature, pressure, and dimensional parameters to the same order of magnitude, such as using Min-Max normalization or Z-score standardization. Time synchronization includes aligning the data from different acquisition systems based on the cell cycle count or timestamp to ensure that data from different service stages can be correlated. Spatial coordinate processing includes establishing a unified spatial coordinate system based on the liquid cooling plate contact surface, mapping cell expansion, thermal pad compression, preload attenuation, and assembly deviation to the corresponding spatial positions on the liquid cooling plate contact surface, allowing various parameters at the same spatial position to be superimposed for calculation. Through the above processing, a basic dataset suitable for thermal management optimization is formed, which is the aforementioned contact boundary dataset. Among them, the thermal pad compression aging degree is used to characterize the performance degradation of the thermal pad under long-term compression, serving as one of the criteria for classifying service stages. The specific processing steps of data cleaning, normalization, time synchronization, and spatial coordinate processing are all well-known technologies in this field, and will not be described in detail here.

[0027] Furthermore, S2 includes: Based on the number of cell cycles, capacity retention rate, internal resistance change, cell expansion degree, and thermal pad compression aging degree, the service process of energy storage battery modules is divided into the initial service stage, the mid-term service stage, and the late service stage. For each stage of service, The cell expansion amount during this service stage is obtained by interpolating the cyclic expansion curve and the number of cell cycles. Based on the initial thickness, length and width of the cell, as well as the direction of cell expansion, the cell expansion amount during this service stage is allocated to the corresponding spatial position of the liquid cooling plate contact surface. Based on the compression modulus and creep characteristics in the parameters of the thermal pad, and combined with the preload of the service stage, the compression amount of the thermal pad in the service stage is determined, and the compression amount of the thermal pad is allocated to the corresponding spatial position of the liquid cooling plate contact surface. The amount of preload decay during this service stage is determined based on the preload and preload decay law, and the amount of preload decay during this service stage is allocated to the corresponding spatial position of the liquid cooling plate contact surface. Based on the contribution of cell expansion, thermal pad compression, preload attenuation, and assembly deviation at the same spatial location, the contact pressure value at that spatial location is determined. The contact pressure spatial map is formed by the contact pressure values ​​at each spatial location. The contact pressure spatial map is used to characterize the process by which the bonding state between the liquid cooling plate and the battery cell evolves from the initial uniform bonding to local weak bonding, local high-pressure bonding, or local detachment.

[0028] Specifically, the module's service life is divided into several stages, including initial, mid-term, and late-term. These stages can also be referred to as the initial service stage, mid-term service stage, and late-term service stage, respectively. For each service stage, the cell thickness expansion, thermal pad compression creep, end plate preload attenuation, and assembly deviation are mapped onto the liquid cooling plate contact surface to obtain the contact pressure distribution results at different spatial locations. The cell thickness expansion, thermal pad compression creep, and end plate preload attenuation can also be referred to as the cell expansion contribution, thermal pad compression contribution, and preload attenuation contribution, respectively.

[0029] For example, the cell expansion amount during the service stage can be obtained by interpolation based on the cyclic expansion curve and the number of cell cycles. This expansion amount is then allocated to the corresponding spatial position on the liquid cooling plate contact surface based on the initial thickness, length, and width of the cell, as well as the expansion direction (e.g., expansion in the thickness direction). Similarly, the thermal pad compression amount during the service stage is determined based on the compressive modulus and creep characteristics of the thermal pad, combined with the preload during that stage, and allocated to the corresponding spatial position on the liquid cooling plate contact surface. Finally, the preload decay amount during the service stage is determined based on the preload and its decay law, and allocated to the corresponding spatial position on the liquid cooling plate contact surface. The preload decay law can be obtained, for example, through stress relaxation tests on the end plate or straps, or determined based on the material creep characteristics combined with finite element simulation. The specific methods for obtaining these laws are well-known and will not be elaborated upon in this embodiment. Different methods for determining decay laws are not unique and do not affect the implementation of this embodiment.

[0030] Based on the aforementioned factors at the same spatial location, the contact pressure value at that location is determined, and a contact pressure spatial map is formed from the contact pressure values ​​at each spatial location. This contact pressure spatial map, which is the result of the aforementioned contact pressure spatial mapping, reflects the gradual evolution of the bonding state between the liquid cooling plate and the battery cell from an initial uniform bonding. Specifically, areas with contact pressure values ​​below the designed lower limit of contact pressure correspond to localized weak bonding, indicating poor contact and reduced heat exchange capacity in that area; areas with contact pressure values ​​above the designed upper limit of contact pressure correspond to localized high-pressure bonding, indicating excessive compression and over-compression of the thermal pad in that area; areas with contact pressure values ​​approaching zero correspond to localized detachment, indicating that effective contact has been lost in that area. The aforementioned upper and lower limits of the designed contact pressure are determined by those skilled in the art based on the characteristics of the thermal pad material, battery cell specifications, and assembly process requirements. The threshold values ​​can be adjusted accordingly for different material systems, which will not be elaborated further in this embodiment.

[0031] Furthermore, the calibration relationship between pressure and contact thermal resistance uses the thermal pad compression and contact pressure as independent variables and contact thermal resistance as the dependent variable; S3 includes: For each spatial location on the contact surface of the liquid cooling plate, the contact thermal resistance value at that spatial location is determined based on the contact pressure value and the compression amount of the thermal pad at that spatial location. The spatial redistribution of contact thermal resistance is determined based on the contact thermal resistance values ​​at each spatial location.

[0032] Specifically, the relationship between the contact thermal resistance of the thermal pad under different compression ratios and contact pressures is obtained through material calibration tests, finite element simulations, or empirical fitting. The thermal pad compression ratio corresponds to the thermal pad compression amount, both reflecting the degree of compression deformation. The calibration relationship can be defined with the thermal pad compression amount and contact pressure as independent variables and contact thermal resistance as the dependent variable. A pressure-thermal resistance conversion model is established based on this calibration relationship; that is, a mapping calculation model that takes the contact pressure value and the thermal pad compression amount as inputs and the contact thermal resistance value as the output. This pressure-thermal resistance conversion model can be, for example, an empirical formula obtained by fitting calibration data, a lookup table model, or a trained neural network model. Its specific form is well-known technology and will not be elaborated upon in this embodiment. The specific modeling method of this pressure-thermal resistance conversion model is not unique and does not affect the implementation of this embodiment.

[0033] The contact pressure spatial map is input into the pressure-thermal resistance conversion model. For each spatial location on the liquid cooler plate contact surface, the contact thermal resistance value at that location is determined based on the contact pressure value and the compression of the thermal pad. Then, based on the contact thermal resistance values ​​at each spatial location, the spatial redistribution of contact thermal resistance is determined. For areas with low contact pressure, reduced thermal pad resilience, or large assembly deviations, the contact thermal resistance increases accordingly; for areas where pressure is well maintained and the fit is sufficient, the contact thermal resistance remains at a low level. This identifies key areas where heat exchange capacity declines in the later stages of the product's lifespan.

[0034] Building upon the above-described determination of the spatial redistribution of contact thermal resistance based on the forward calibration relationship, this embodiment further provides an implementation method for online reverse correction of the spatial redistribution of contact thermal resistance using measured temperature data from the energy storage system during operation, in order to further compensate for the cumulative errors caused by material parameter drift and individual variability during long-term operation. This implementation method can form a data closed loop with the forward calculation results of step S3 above, enabling the contact boundary model to maintain dynamic updating capability throughout the entire life cycle of the energy storage system.

[0035] Specifically, temperature sensor arrays are arranged on the surface of each module of the energy storage system. For example, four thermocouples are arranged on the large surface of each cell, located at the one-quarter and three-quarters positions near the positive and negative terminals along the length of the cell, and symmetrically on both sides along the width. The battery management system continuously collects and stores the measured temperature data from these temperature sensors during operation.

[0036] When the system reaches a certain service stage, the following inversion correction steps are performed: The first step is parameterized dimensionality reduction of the spatial distribution of contact thermal resistance. The contact surface between the liquid cooling plate and the battery cell is divided into a uniform spatial grid. The spatial redistribution results of contact thermal resistance obtained by the above-mentioned forward calibration method in multiple historical calculated service phases of the energy storage system are collected as a sample set. These historical sample sets are subjected to intrinsic orthogonal decomposition to extract a few dominant spatial distribution modes. Analysis shows that the spatial variation of contact thermal resistance on the contact surface is usually dominated by 3 to 5 basic modes, such as the overall uniform degradation mode, the edge gradual detachment mode, the high voltage maintenance mode in the central region, and the diagonal non-uniform degradation mode. Each mode corresponds to a spatial basis function, which exhibits a specific distribution shape in space. The spatial distribution of contact thermal resistance to be corrected in the current service phase is represented as a weighted combination of these spatial basis functions, and the weight coefficients are the inversion parameters to be determined. In this way, the contact thermal resistance value that originally needed to be determined independently at each grid node is transformed into only needing to determine a few weight coefficients, which significantly reduces the complexity and difficulty of the inversion problem.

[0037] The second step involves constructing and iteratively solving a constrained optimization problem. Using the few weight coefficients obtained in the first step as optimization variables, a constrained optimization problem is constructed. The reconstructed spatial distribution of contact thermal resistance under the current weight coefficients is input into the trained deep learning model (i.e., the deep learning model in S4 above), yielding the model's output temperature prediction. This temperature prediction is then compared point-by-point with the measured temperature values ​​from the temperature sensor array. The overall deviation between the two is used as the optimization objective, and the weight coefficients are iteratively adjusted to gradually reduce this deviation.

[0038] The following constraints are applied during the iteration process: Non-negativity constraint: the contact thermal resistance value at any grid node obtained by inversion must not be lower than the physical lower limit corresponding to the thermal conduction of the air gap, so as to avoid the occurrence of physically meaningless excessively low thermal resistance; Spatial continuity constraint: the change in contact thermal resistance between adjacent grid nodes must not exceed the preset gradient threshold. If a certain iteration causes a drastic change in a local area, a smoothing penalty term is added to the optimization objective to force the inversion result to tend to a spatially continuous distribution; Monotonically increasing constraint: the contact thermal resistance value obtained by inversion at each spatial location in the current service stage must not be lower than the contact thermal resistance value determined at that location in the previous service stage. This constraint reflects the irreversibility of the aging and deterioration of the contact interface and prevents the non-physical phenomenon that the contact state improves with running time from appearing in the inversion result.

[0039] An optimization algorithm suitable for handling boundary constraints is used for iterative solution. The iteration terminates when the overall deviation between the predicted and measured temperature values ​​converges to a stable level and the adjustment of the weight coefficients is less than a preset threshold, thus obtaining the optimal weight coefficients.

[0040] The third step is to reconstruct and update the spatial redistribution of contact thermal resistance. Using the optimal weighting coefficients obtained in the second step, the spatial basis functions extracted in the first step are weighted and combined to reconstruct the updated spatial distribution of contact thermal resistance. This distribution satisfies both the constraints of measured temperature data and the constraints of physical feasibility and irreversible aging.

[0041] The fourth step is to update the contact pressure space map through inverse mapping. Based on the pressure and contact thermal resistance calibration relationship of the thermal pad parameters in step S3, and given the current thermal pad compression during service, the updated contact pressure value is calculated point-by-point from the updated contact thermal resistance space distribution using a lookup table inverse mapping method. Specifically, for each grid node on the contact surface, the thermal pad compression at that node is fixed, and the contact pressure value corresponding to the updated contact thermal resistance value of that node is found in the calibration data table; if the updated contact thermal resistance value exceeds the original range of the calibration table, the corresponding contact pressure value is determined using a logarithmic linear extrapolation method. The updated contact pressure values ​​of all grid nodes on the contact surface are combined to form the updated contact pressure space map. This map replaces the original contact pressure space map obtained through forward calculation in step S2 and is used in the subsequent step S4 to predict the temperature field distribution for the next time period or the next service stage.

[0042] Compared to forward calculation methods that rely solely on factory calibration parameters, this embodiment uses measured temperature data during operation to reverse-correct the spatial distribution of contact thermal resistance. This enables the contact boundary model to adaptively track the actual aging path, significantly reducing the long-term prediction error of the contact boundary model. Simultaneously, intrinsic orthogonal decomposition is used for parameterization and dimensionality reduction, reducing the number of variables to be solved in the inversion problem from hundreds of grid nodes to at least a few weight coefficients. This allows the previously difficult-to-solve high-dimensional ill-posed inversion problem to be completed in real-time under the limited computing resources of the energy storage system, with the time required for a single correction meeting the requirements for online updates.

[0043] This embodiment combines the positive calibration results of step S3 with the deep learning prediction capabilities of step S4, significantly improving the accuracy of the contact boundary model throughout the entire lifecycle.

[0044] The specific mathematical implementation of the aforementioned intrinsic orthogonal decomposition and constrained optimization algorithm is a well-known technology and will not be elaborated upon in this embodiment. This inversion correction method is merely a further calculation method in one embodiment of the present invention and is not the only one. Other dimensionality reduction inversion methods that satisfy the constraints of nonnegativity, continuity, and monotonically increasing can also be used for online correction of the spatial distribution of contact thermal resistance, and the specific algorithm selection does not affect the implementation of this embodiment.

[0045] In one specific implementation, such as Figure 2As shown, based on the cell cycle count, capacity retention, internal resistance change, cell expansion degree, and thermal pad compression aging degree, the service life of the energy storage battery module is divided into initial service stage, mid-term service stage, and late service stage. For each service stage, the cell expansion amount is obtained by interpolation based on the cycle expansion curve and cell cycle count. Then, based on the initial cell thickness, length, and width, as well as the cell expansion direction, the cell expansion amount is allocated to the corresponding spatial position on the liquid cooling plate contact surface. Based on the compressive modulus and creep characteristics of the thermal pad parameters, combined with the preload for that service stage, the thermal pad compression amount is determined and allocated to the corresponding spatial position on the liquid cooling plate contact surface. Finally, based on the preload and preload decay law, the preload decay amount for that service stage is determined and allocated to the corresponding spatial position on the liquid cooling plate contact surface.

[0046] Based on the contributions of cell expansion, thermal pad compression, preload attenuation, and assembly deviation at the same spatial location, the contact pressure value at that location is determined. A contact pressure spatial map is formed from the contact pressure values ​​at each spatial location. This map characterizes the process by which the bonding state between the liquid cooling plate and the cell evolves from an initial uniform bonding to localized weak bonding, localized high-pressure bonding, or localized delamination. Specifically, areas with contact pressure values ​​below the design lower limit correspond to localized weak bonding, indicating poor contact and reduced heat exchange capacity in that area; areas with contact pressure values ​​above the design upper limit correspond to localized high-pressure bonding, indicating excessive compression and over-compression of the thermal pad in that area; and areas with contact pressure values ​​approaching zero correspond to localized delamination, indicating a loss of effective contact in that area.

[0047] The calibration relationship between pressure and contact thermal resistance uses the thermal pad compression and contact pressure as independent variables and contact thermal resistance as the dependent variable. For each spatial location on the liquid cooling plate contact surface, the contact thermal resistance value at that location is determined based on the contact pressure and thermal pad compression. Based on the contact thermal resistance values ​​at each spatial location, a spatial redistribution of contact thermal resistance is determined. For areas with lower contact pressure, reduced thermal pad resilience, or larger assembly deviations, the contact thermal resistance increases accordingly; for areas where pressure is maintained well and the fit is sufficient, the contact thermal resistance remains at a lower level. This identifies key areas where heat exchange capacity declines later in the product's lifespan.

[0048] Furthermore, the predictive architecture of deep learning models includes spatial feature branches and temporal feature branches; The spatial feature branch uses convolutional neural networks, U-Net, or graph neural networks to extract the spatial relationship between the liquid cooling plate contact surface, the flow channel coverage area, and the cell temperature field. The time feature branch uses a recurrent neural network, LSTM or Transformer structure to learn the effects of charging and discharging conditions on the cell temperature field over time, the effects of cooling flow rate on the cell temperature field over time, the effects of ambient temperature on the cell temperature field over time, and the effects of cell aging state on the cell temperature field over time. The output of the trained deep learning model includes the temperature field distribution, maximum temperature, maximum temperature difference, hot spot probability, hot spot duration, and cooling effectiveness evaluation results for each preset candidate flow channel layout in each service stage.

[0049] Specifically, the model uses cell heat distribution, contact thermal resistance spatial map, candidate flow channel layout of liquid cooling plate, operating condition data, and historical temperature data as inputs to establish a deep learning model that integrates spatial and temporal features. Spatial features can be extracted using convolutional neural networks, U-Net, or graph neural networks to extract the spatial coupling relationship between the liquid cooling plate contact surface, the flow channel coverage area, and the cell temperature field; temporal features can be learned using recurrent neural networks, LSTM, or Transformer structures to study the effects of charging and discharging conditions, cooling flow rate, ambient temperature, and cell aging state on the temperature field over time.

[0050] The heat distribution of the battery cell can be determined, for example, based on the heat characteristics of the cell under different states of charge and health, combined with the current and charge / discharge rate under the current operating conditions. Historical temperature data includes measured temperature field data of the modules collected during the operation of the energy storage system, which is used to train and validate the deep learning model. The training process of the deep learning model includes: pairing the above input data with the corresponding measured temperature field data to construct a training sample set; iteratively adjusting the model parameters by minimizing the error between the predicted temperature field and the measured temperature field until the model converges. This training process is a well-known technique in the field of deep learning, and will not be elaborated here. The specific training strategy and hyperparameter settings are not unique and do not affect the implementation of this embodiment.

[0051] The model output includes module temperature field distribution, maximum temperature, maximum temperature difference, hotspot probability, hotspot duration, and cooling effectiveness evaluation results. The module temperature field distribution refers to the temperature field distribution of each preset candidate flow channel layout at each service stage. Hotspot probability indicates the likelihood of an abnormal temperature rise at a specific location under specific operating conditions. Hotspot duration indicates the duration for which the temperature in the hotspot area exceeds the safety threshold. The cooling effectiveness evaluation results are used to determine the overall effect of the current flow channel layout on temperature control.

[0052] Building upon the input of multiple preset candidate flow channel layouts into a deep learning model to predict temperature field distribution, this embodiment further reduces the complexity of constructing the candidate flow channel layout library and improves the matching accuracy between flow channel layouts and contact state evolution during service stages. Specifically, it provides an implementation method that encodes some candidate flow channel layouts into lifetime-responsive parameterized models. This implementation method allows a single parameterized model to cover flow channel morphology changes across multiple service stages, reducing the number of discrete layouts that the deep learning model needs to traverse in S4. Simultaneously, it enables the lifetime-weighted evaluation in S5 to provide a more refined assessment of the progressive evolution characteristics of flow channel geometry.

[0053] Specifically, when constructing the candidate flow channel layout library, in addition to using a fixed discrete flow channel layout scheme, at least some of the candidate flow channel layouts are encoded using a lifetime-response-type parameterized model.

[0054] In this parametric model, the geometric parameters of the main flow channel of the liquid cooling plate remain fixed, covering the hot-spot sensitive areas identified as hot-spot sensitive areas in each service stage in S5, to ensure stable cooling of continuously high-heat areas throughout the entire life cycle. The geometric parameters of the auxiliary flow channel are encoded as a continuous function of the service stage, specifically including: the width of the auxiliary flow channel increases exponentially with the service stage, corresponding to the basic width in the initial stage, the maximum width in the later stage, and a linear or exponential transition law in the middle stage; the coverage area of ​​the auxiliary flow channel expands outward exponentially with the service stage, with zero coverage or only covering high-risk areas in the initial stage, and expanding to cover the newly added hot-spot sensitive areas in the later stage of the life cycle identified by cross-stage comparison in S5; the curvature of the auxiliary flow channel direction is adjusted exponentially with the service stage, with a relatively straight direction in the initial stage to ensure low flow resistance, and an adaptive curvature in the later stage based on the spatial distribution of newly added hot-spot sensitive areas in the later stage of the life cycle to improve local heat transfer.

[0055] When the aforementioned lifetime-response parameterized model is input into the deep learning model trained in S4, the model calculates the auxiliary flow channel geometry corresponding to each service stage, thereby predicting the temperature field distribution for each service stage. In the lifetime-weighted evaluation in S5 and the multi-objective collaborative optimization in S6, the optimization object is no longer a discrete flow channel layout scheme, but rather the coefficients of the aforementioned continuous function. By optimizing these coefficients, the lifetime-weighted comprehensive score is maximized.

[0056] Compared to a fixed candidate library approach, the lifetime-response parametric model can express the evolution trend of the flow channel layout with service stages using only a small number of continuous coefficients, reducing the number of candidate layouts and lowering the training and inference burden of the deep learning model. Furthermore, because the auxiliary flow channel geometry is continuously coupled with the service stage, the optimization results can more precisely match the gradual process of contact state evolution, avoiding the step-like performance fluctuations caused by discrete layout switching, and enabling a smooth transition in the thermal management capability of the liquid cooler plate with the service stage.

[0057] This embodiment upgrades the prediction input of step S4 from discrete geometric enumeration to a continuous parameterized family, which echoes the progressive evolution characteristics of the contact state in steps S2-S3, improving the consistency and economy of the flow channel design throughout the entire life cycle.

[0058] The specific form and coefficient optimization method of the aforementioned continuous function are well-known technologies and will not be elaborated upon in this embodiment. This parameterized encoding method is merely a further calculation method in one embodiment of the present invention and is not the only one. Other parameterized expressions that can reflect the monotonic or segmented evolution of the flow channel geometry with the service stage can also be used for encoding candidate flow channel layouts, and their specific function forms do not affect the implementation of this embodiment.

[0059] Furthermore, based on the temperature field distribution, hotspot sensitive areas are identified, including: In each service phase, the spatial locations where the contact thermal resistance value exceeds the preset thermal resistance threshold corresponding to that service phase and the cell heat value exceeds the preset heat generation threshold corresponding to that service phase are marked as hot spot sensitive locations for that service phase. The hotspot sensitive locations marked at each service stage will be compared across stages to select spatial locations that were not marked as hotspot sensitive locations in the early or middle service stages but were marked as hotspot sensitive locations in the later service stages, which will be added as hotspot sensitive locations in the later stages of the service life. The hotspot sensitive areas for each service phase are formed by connecting and aggregating adjacent hotspot sensitive locations. The hotspot sensitive areas for the later stages of service are formed by connecting and aggregating adjacent locations among the newly added hotspot sensitive locations in the later stages of service.

[0060] Specifically, the redistribution results of contact thermal resistance at different service stages are coupled with the cell heat distribution for analysis to identify areas prone to concentrated temperature rise under long-term service conditions. If a region simultaneously exhibits high heat generation, increased contact thermal resistance, insufficient liquid cooling channel coverage, or decreased coolant heat exchange capacity, it is identified as a hotspot-sensitive area. The aforementioned conditions, such as "high heat generation and increased contact thermal resistance," can be achieved, for example, by marking locations where both the contact thermal resistance value and the cell heat generation value exceed a preset thermal resistance threshold corresponding to that service stage as hotspot-sensitive locations for that service stage. The preset thermal resistance threshold and preset heat generation threshold are determined by those skilled in the art based on cell safety operation requirements and thermal management design standards. These thresholds can be adjusted accordingly for different cell systems, and will not be elaborated further in this embodiment.

[0061] This identification result not only considers the high-temperature regions in the initial thermal simulation, but also focuses on newly added hot spots caused by cell expansion, thermal pad creep, and preload decay in the later stages of the service life; these newly added hot spots can also be referred to as newly added hot spot sensitive locations in the later stages of the service life. Adjacent hot spot sensitive locations within each service stage are connected and aggregated to form the corresponding hot spot sensitive region for that service stage, and adjacent locations among the newly added hot spot sensitive locations in the later stages of the service life are connected and aggregated to form the newly added hot spot sensitive region in the later stages of the service life.

[0062] Based on the identification of hotspot-sensitive areas at each service stage and newly added hotspot-sensitive areas in the later stages of service life, this embodiment also provides a pre-processing implementation method based on thermal propagation network analysis to further avoid the problem of inter-regional heat migration caused by isolated cooling enhancement of individual hotspot areas. This implementation method is executed before the multi-objective collaborative optimization in step S6, and is used to identify implicit thermal coupling paths between hotspot-sensitive areas, so that the flow channel optimization and bonding enhancement in step S6 not only target isolated hotspots, but also block the thermal crosstalk channels between hotspots as a whole.

[0063] Specifically, after completing the hotspot sensitive area identification in S5, the following heat propagation network analysis steps are performed: The first step is to construct a heat propagation network. The hotspot sensitive areas identified in step S5 are used as network nodes. Based on the temperature field distribution predicted in step S4, the direction of the temperature gradient between every two adjacent hotspot sensitive areas is analyzed, and this direction is used as the network edge direction connecting these two network nodes. The heat flux density in this direction is calculated and used as the weight of the network edge; the greater the heat flux density, the stronger the thermal coupling between the two hotspot sensitive areas, and the easier it is for heat to be transferred along this direction.

[0064] The second step is to identify the critical paths for heat propagation. The minimum spanning tree algorithm is used to process the heat propagation network to find the tree structure that connects all hotspot sensitive areas with the minimum total weight. In this tree structure, the edge with the largest weight is the critical path for heat propagation connecting multiple hotspot sensitive areas. Alternatively, the maximum flow algorithm can be used, taking the hotspot sensitive area with the highest heat generation as the source and the area with the lowest temperature as the sink, to calculate the maximum heat flow path in the network; the edges traversed by this path are the critical paths for heat propagation.

[0065] The third step is the collaborative optimization of the thermal blocking structure. In the multi-objective collaborative optimization process of step S6, in addition to optimizing the flow channel geometry parameters and the bonding reinforcement area, a thermal blocking design for the critical heat propagation path is added. Specifically, this includes: adjusting the direction of the auxiliary flow channel in the liquid cooling plate area traversed by the critical heat propagation path to allow the coolant to flow preferentially through this path, reducing the temperature gradient along the path and thus weakening the thermal crosstalk intensity; or locally adding heat insulation pads or thinning the thickness of the thermal conductive pads between the battery cell and the liquid cooling plate along the critical heat propagation path to increase the thermal resistance along this path and block heat transfer along the path; or setting bonding reinforcement areas on both sides of the critical heat propagation path to enhance the heat exchange capacity between the two sides and the liquid cooling plate by increasing the local clamping force, directly conducting heat away instead of crosstalking to other hot spots along the path. The above thermal blocking design parameters are added as new continuous geometric parameters in the S6 multi-objective collaborative optimization process and participate in the iterative optimization.

[0066] By adding heat propagation network analysis before optimizing the S6 flow channel, hidden thermal coupling paths that are difficult to detect from the high-value distribution of the temperature field can be identified. In subsequent multi-objective collaborative optimization, thermal blocking structures are set up for these critical paths, which can significantly improve the overall temperature uniformity of the module. This avoids the problem of overcooling a certain hot spot sensitive area, causing heat to migrate to another area and exacerbating thermal imbalance, thereby improving the systematicness and synergy of thermal management throughout the entire life cycle.

[0067] The specific implementations of the minimum spanning tree algorithm and the maximum flow algorithm described above are well-known technologies and will not be elaborated upon in this embodiment. This heat propagation network analysis method is merely a further calculation method in one embodiment of the present invention and is not the only one. Other graph analysis methods that can identify key coupling paths in the heat propagation network can also be used for the collaborative optimization of heat-blocking structures, and the specific algorithm selection does not affect the implementation of this embodiment.

[0068] Connectivity aggregation of adjacent hotspot sensitive locations can be achieved, for example, by using a region growing algorithm or a connected component labeling algorithm based on spatial adjacency. The specific algorithms are well-known technologies in the field of image processing, and will not be elaborated here. The choice of different connectivity aggregation methods is not unique and does not affect the implementation of this embodiment.

[0069] Furthermore, based on the duration, safety risks, and economic life of each service phase, a life-weighted evaluation is performed on each preset candidate flow channel layout to determine the optimal candidate layout, including: For each preset candidate flow channel layout, the cooling effect index of the preset candidate flow channel layout is calculated in the initial service stage, the mid-service stage and the late service stage. The duration of each service phase is determined based on its duration. Combined with the preset safety risk coefficient and economic life coefficient, the cooling effect index of each service phase is weighted and calculated to obtain the life weighted comprehensive score of each preset candidate flow channel layout. By comparing the lifetime-weighted comprehensive scores of each preset candidate flow channel layout, the preset candidate flow channel layout with the highest lifetime-weighted comprehensive score is determined as the optimal candidate layout.

[0070] Specifically, for each candidate liquid cooling plate layout, its cooling effect is calculated in the initial, mid-term, and late-term service stages. Cooling effect indicators include, for example, maximum temperature, maximum temperature difference, hot spot coverage, contact thermal resistance sensitive area coverage, cooling pressure drop, pump power consumption, flow channel utilization, and structural manufacturability. The duration of each service stage is determined based on its duration percentage. Combined with preset safety risk coefficients and economic life coefficients, the cooling effect indicators for each service stage are weighted and calculated to obtain a life-weighted comprehensive score. The safety risk weight and economic life impact weight can also be referred to as the aforementioned safety risk coefficient and economic life coefficient, respectively.

[0071] The safety risk coefficient reflects the relative weight of thermal runaway risk at different service stages. For example, in the later service stages, due to the higher degree of cell aging, the safety risk coefficient can be increased accordingly. The economic life coefficient reflects the relative weight of the impact of different service stages on the overall economic life of the system. For example, in the mid-term service stage, which lasts longer, the economic life coefficient can be increased accordingly. The specific values ​​of the above coefficients are determined by those skilled in the art based on the safety level requirements and economic objectives of the energy storage system, and will not be elaborated here.

[0072] By using a lifetime-weighted evaluation, liquid cooling plate layouts that perform well initially but fail to cover hot spots later in their lifespan are prevented from being mistakenly selected as the optimal solution. The lifetime-weighted comprehensive score of each preset candidate flow channel layout is compared, and the preset candidate flow channel layout with the highest lifetime-weighted comprehensive score is determined as the optimal candidate layout.

[0073] Based on determining the optimal candidate layout using static weighted summation, this embodiment also provides an implementation method that models the flow channel layout selection as a multi-stage dynamic decision-making process across service phases. This implementation method is executed after the life-weighted evaluation in S5 and before the multi-objective collaborative optimization in S6. It is used to perform full-lifecycle dynamic verification and sequential optimization of the optimal candidate layout determined in S5, ensuring that the optimal layout not only performs well in the independent scores of each stage, but also remains globally optimal under the physical evolution laws between stages.

[0074] Specifically, the selection process for the liquid cooling plate flow channel layout is modeled as a multi-stage dynamic decision-making process spanning service phases. The following explanation uses the initial, mid-term, and late-term service phases of S3 as examples.

[0075] The first step is to define state variables and decision variables. State variables are represented by a low-dimensional representation of the contact pressure space map at the start of each service stage. Specifically, the contact pressure space map obtained in S2 is compressed into a few coefficients using intrinsic orthogonal decomposition; these coefficients together constitute the state description of the current stage. Decision variables are the layout adjustment parameters of the auxiliary flow channels in each service stage. The geometric parameters of the main flow channel of the liquid cooling plate remain fixed throughout its lifespan, covering the hot-spot sensitive areas marked as hot-spot sensitive locations in each service stage as described in S5 above. Decision variables for the auxiliary flow channels include whether they are activated, the location of the covered area, and the size of the coverage area. To reduce the complexity of the decision space, the cell contact surface is pre-divided into several candidate sub-regions, each sub-region corresponding to an optional auxiliary flow channel deployment location.

[0076] The second step is to establish a contact state evolution model. This involves establishing state transition relationships to describe how the current flow channel layout decisions and the current contact state jointly determine the contact state in the next stage. The state transition relationship includes the following physical processes: Preload decay accumulation: Based on the S2 preload decay law, the preload loss at the end of the current stage is calculated and subtracted from the initial preload of the current stage to obtain the usable residual preload for the next stage; Cell expansion accumulation: Based on the S2 cyclic expansion curve, the cell expansion increment of the current stage is interpolated from the number of cycles, and added to the current accumulated expansion to obtain the total cell expansion for the next stage; Thermal pad creep accumulation: Based on the S2 thermal pad creep characteristics, the compression increment of the thermal pad in the current stage is calculated. This compression increment depends on the contact pressure level of the current stage, reflecting the material characteristic that the higher the stress level, the faster the creep rate; Flow channel structure stiffness feedback: When a decision is made to add or adjust an auxiliary flow channel in a certain area of ​​the current stage, the local thickness or structural form of the liquid cooling plate in that area changes, leading to a change in local structural stiffness. The contact surface pressure redistribution effect caused by this stiffness change is calculated through structural stiffness analysis. The above four physical effects are superimposed to obtain the contact pressure spatial map of the next stage, and then projected in a low dimension to obtain the state variables of the next stage.

[0077] The third step is to define the stage evaluation indicators and cross-stage constraints. For each service stage, the S5 cooling effect index is used as the stage evaluation basis. In the multi-stage decision-making process, the following cross-stage physical constraints are applied: preload conservation constraint, the residual preload in the final service stage must be equal to the initial preload minus the sum of the preload decay in the initial and middle stages, to prevent non-physical situations where the preload is generated or disappears out of thin air in the optimization results; irreversible expansion constraint, the total expansion of the cell increases monotonically with the service stage; irreversible creep constraint, the compression of the thermal pad increases monotonically with the service stage; temperature safety constraint, the predicted maximum temperature in any service stage must not exceed the cell's safe temperature threshold.

[0078] The fourth step is to solve for the optimal decision sequence in reverse recursion. Starting from the end-of-service stage, the optimal decision is solved by progressively recursively solving for the previous stage. For the end-of-service stage, given a state, all feasible auxiliary flow channel layout decisions are traversed, and the decision that optimizes the cooling effect of this stage and satisfies the temperature safety constraints is selected and recorded as the optimal value. For the mid-service stage, given a state, all feasible decisions are traversed. For each candidate decision, the cooling effect of the decision on the current stage is calculated, and the optimal value corresponding to the next stage state predicted by the state transition relationship is added. The decision that maximizes the sum of the two and satisfies all cross-stage physical constraints is selected. For the early-service stage, all feasible decisions are traversed in the same way. The sum of the cooling effect of the current stage and the predicted optimal value of the next stage is calculated, and the initial decision that optimizes the overall cooling effect over the entire life cycle is selected. During the recursive process, if a candidate decision causes subsequent stages to fail to meet the temperature safety constraints or violate the irreversible constraint of expansion and creep, the candidate decision is determined to be infeasible and directly eliminated.

[0079] The fifth step is to extract the optimal flow channel layout sequence. After the reverse recursion is completed, starting from the known initial state of the initial service stage, the optimal decision for each stage is determined by forward lookup, resulting in the optimal flow channel layout decision sequence for the initial, mid-term, and late-term service stages. If the liquid cooling plate flow channel structure is fixed and cannot be adjusted, the static layout that provides the most comprehensive coverage of the hotspot sensitive areas in each stage is taken as the final design scheme. This final design scheme can be used as the initial input for S6 multi-objective collaborative optimization. If the liquid cooling plate is equipped with adjustable valves or a flow distribution mechanism, the on / off state of the auxiliary flow channel or the flow distribution can be switched according to this sequence in each stage during actual operation.

[0080] This embodiment models the state transition relationships between stages using dynamic programming. Because the state transition relationships incorporate feedback on the flow channel structure stiffness, cumulative preload decay, cumulative cell expansion, and cumulative thermal pad creep, the optimized flow channel layout sequence avoids the common problem in static weighted methods where initial performance is good but later hot spots appear due to deteriorating contact conditions. Simultaneously, the introduction of cross-stage physical constraints ensures that the optimization results are physically self-consistent throughout the entire lifecycle, preventing physically infeasible solutions from arising from excessive preload consumption or excessive thermal pad compression in the early stages, leading to complete contact failure later.

[0081] The specific algorithm implementations of the dynamic programming and state transitions described above are well-known technologies and will not be elaborated upon in this embodiment. This multi-stage dynamic decision-making method is merely a further calculation method in one embodiment of the present invention and is not the only one. Other sequential decision-making methods capable of handling cross-stage state transitions and physical constraint recursion can also be used for full-lifecycle flow channel layout optimization, and their specific solution strategies do not affect the implementation of this embodiment.

[0082] In one specific implementation, such as Figure 3 As shown, in each service stage, the spatial locations where the contact thermal resistance value exceeds the preset thermal resistance threshold corresponding to that service stage and the cell heating value exceeds the preset heating threshold corresponding to that service stage are marked as hot spot sensitive locations for that service stage.

[0083] A cross-stage comparison will be conducted on hotspot sensitive locations marked at each service stage. Spatial locations that were not marked as hotspot sensitive locations in the early or mid-service stages but are marked as such in the later service stages will be identified as newly added hotspot sensitive locations in the later stages of the service life. The core of this cross-stage comparison is that if a hotspot sensitive location has already been marked in the early or mid-service stages, it indicates that the location is a continuous problem throughout the entire service life and can be covered by the main flow channel; if the location was not marked in the early or mid-service stages and only appears in the later stages, it is considered a newly added hotspot sensitive location in the later stages and requires targeted compensation by auxiliary flow channels and bonding reinforcement structures.

[0084] The hotspot sensitive areas are formed by connecting and aggregating adjacent hotspot sensitive locations within each service stage, and by connecting and aggregating adjacent locations among newly added hotspot sensitive locations in the later stages of the service life, forming newly added hotspot sensitive areas in the later stages of the service life. This identification result not only considers the high temperature regions in the initial thermal simulation, but also focuses on the newly added hotspot regions caused by cell expansion, thermal pad creep, and preload decay in the later stages of the service life.

[0085] Furthermore, the main channel of the liquid cooling plate covers the hot-spot sensitive area formed by the spatial location marked as a hot-spot sensitive location throughout the entire service phase; The auxiliary flow channel is used to compensate for newly added hot and sensitive areas in the later stages of the service life; the location, direction and coverage of the auxiliary flow channel are determined according to the spatial distribution of newly added hot and sensitive areas in the later stages of the service life.

[0086] Specifically, based on the life-weighted cooling effectiveness evaluation results, the main flow channel, auxiliary flow channels, ineffective flow channel reduction areas, and bonding reinforcement areas of the liquid cooling plate are determined. The main flow channel prioritizes covering areas with continuous high heat generation or high-risk hot spots throughout the entire lifespan; these areas refer to spatially defined hot spot sensitive locations marked as hot spot sensitive positions throughout all service phases. Auxiliary flow channels are used to compensate for newly added hot spot sensitive areas due to increased contact thermal resistance in the later stages of the lifespan. Their location, direction, and coverage are determined based on the spatial distribution of these newly added hot spot sensitive areas. For example, if newly added hot spot sensitive areas are located at the module edge, the auxiliary flow channels can be arranged in a serpentine pattern along that edge area or the local flow channel density can be increased; if the newly added hot spot sensitive areas are discretely distributed, the auxiliary flow channels can use branched flow channels for multi-point coverage. Ineffective flow channel reduction areas are flow channel areas with low heat transfer contribution and large pressure drop losses during long-term operation. Bonding reinforcement areas can improve contact conditions by increasing local clamping structures, optimizing thermal pad thickness, or adding support structures.

[0087] Furthermore, multi-objective collaborative optimization includes: Based on the optimal candidate layout, the continuous geometric parameters are iteratively optimized using a multi-objective optimization algorithm. The continuous geometric parameters include the channel width, channel depth, channel curvature and channel spacing of the main channel and auxiliary channels of the liquid cooling plate, as well as the local pressing structure dimensions of the bonding reinforcement area. The objective function of the multi-objective optimization algorithm consists of multiple optimization sub-objectives, including: lifetime weighted comprehensive score, hot spot sensitive area coverage, hot spot sensitive area spatial distribution density, temperature uniformity, cooling efficiency, pressure drop, pump power consumption, manufacturing cost, and structural reliability. A Pareto optimal solution set is obtained through a multi-objective optimization algorithm to weigh the trade-offs among the various optimization sub-objectives.

[0088] Specifically, based on the optimal candidate layout, continuous geometric parameters are iteratively optimized using a multi-objective optimization algorithm. This multi-objective optimization algorithm can employ, for example, a multi-objective genetic algorithm, Bayesian optimization, reinforcement learning, or a deep learning surrogate model. The specific algorithm selection is well-known and will not be elaborated upon in this embodiment. The choice of different algorithms does not affect the implementation of this embodiment. The objective function consists of multiple optimization sub-objectives, including: lifetime-weighted comprehensive score, hotspot sensitive area coverage, hotspot sensitive area spatial distribution density, temperature uniformity, cooling efficiency, pressure drop, pump power consumption, manufacturing cost, and structural reliability, aiming to achieve a balance among these sub-objectives.

[0089] Continuous geometric parameters include the channel width, depth, curvature, and spacing of the main and auxiliary channels of the liquid cooling plate, as well as the dimensions of the local pressing structures in the bonding reinforcement area. The dimensions of the local pressing structures in the bonding reinforcement area include geometric parameters such as the height, width, or distribution density of the local pressing structures. Hotspot sensitive area coverage represents the effective coverage ratio of the identified hotspot sensitive areas by the channel layout; the spatial distribution density of hotspot sensitive areas reflects the concentration or dispersion of hotspot sensitive areas in the module space, used to guide fine-tuning of the channel layout.

[0090] A Pareto optimal solution set is obtained through a multi-objective optimization algorithm to weigh the trade-offs among various optimization sub-objectives and generate thermal management optimization results oriented towards long-term operation. The Pareto optimal solution set refers to the set of solutions from which other sub-objectives cannot be further improved without sacrificing any of the sub-objectives; those skilled in the art can select the final implementation scheme from this solution set according to actual engineering needs.

[0091] In one specific implementation, such as Figure 4 As shown, the main flow channel of the liquid cooling plate covers the hot-spot sensitive areas formed by spatial locations marked as hot-spot sensitive positions throughout the entire service life. That is, the main flow channel prioritizes covering hot-spot areas that experience continuous high heat generation or high risk throughout the entire lifespan. Auxiliary flow channels are used to compensate for newly added hot-spot sensitive areas later in the lifespan. The location, direction, and coverage of the auxiliary flow channels are determined based on the spatial distribution of these newly added hot-spot sensitive areas. For example, if a newly added hot-spot sensitive area is located at the edge of the module, the auxiliary flow channels can be arranged in a serpentine pattern along that edge area or the local flow channel density can be increased; if the newly added hot-spot sensitive areas are discretely distributed, the auxiliary flow channels can use branched flow channels for multi-point coverage. A bonding reinforcement area is set in the newly added hot-spot sensitive area later in the lifespan. This area's contact condition is improved and the interfacial contact thermal resistance is reduced by increasing the local pressing structure, optimizing the thermal pad thickness, or adding support structures.

[0092] After determining the main flow channel, auxiliary flow channels, and the bonding reinforcement region, a multi-objective optimization algorithm is used to iteratively optimize the continuous geometric parameters based on the optimal candidate layout. The continuous geometric parameters include the flow channel width, flow channel depth, flow channel curvature, and flow channel spacing of the main and auxiliary flow channels of the liquid cooling plate, as well as the dimensions of the local pressing structure in the bonding reinforcement region. The objective function of the multi-objective optimization algorithm consists of multiple optimization sub-objectives, including lifetime-weighted comprehensive score, hotspot sensitive area coverage, hotspot sensitive area spatial distribution density, temperature uniformity, cooling efficiency, pressure drop, pump power consumption, manufacturing cost, and structural reliability. The Pareto optimal solution set is obtained through the multi-objective optimization algorithm to balance the various optimization sub-objectives. Ineffective flow channel reduction regions are those with low heat transfer contribution and large pressure drop losses during long-term operation; these flow channels are removed or reduced in these regions.

[0093] According to another aspect of the embodiments of this application, a deep learning-based thermal management optimization system for lithium battery energy storage systems is also provided, comprising: a dataset construction module, used to acquire initial cell size, cell cyclic expansion characteristics, thermal pad parameters, assembly deviation, preload, and operating condition data to construct a contact boundary dataset; a map construction module, used to divide the service stages based on the contact boundary dataset, determine the cell expansion, thermal pad compression, and preload attenuation in each service stage, and establish a contact pressure space mapping of the liquid cooling plate contact surface in combination with assembly deviations to form a contact pressure space map; and a conversion module, used to convert the contact pressure space map into a contact thermal resistance space map according to the calibration relationship between pressure and contact thermal resistance in the thermal pad parameters. The system includes a redistribution and prediction module, which determines the cell heat distribution at each service stage based on operating condition data. It inputs the contact thermal resistance spatial redistribution, cell heat distribution, operating condition data, and multiple preset candidate flow channel layouts into a trained deep learning model to predict the temperature field distribution of each preset candidate flow channel layout at each service stage. An optimal layout determination module identifies hotspot-sensitive areas based on the temperature field distribution and performs a lifespan-weighted evaluation of each preset candidate flow channel layout based on the duration of each service stage, safety risks, and economic lifespan to determine the optimal candidate layout. A multi-objective optimization module performs multi-objective collaborative optimization of the main flow channel, auxiliary flow channel, and bonding reinforcement area of ​​the liquid cooling plate based on the optimal candidate layout and hotspot-sensitive areas.

[0094] It should be noted that the embodiments implemented on the thermal energy management optimization system side of the lithium battery energy storage system based on deep learning in this application can be referenced with the embodiments implemented on the thermal energy management optimization method side of the lithium battery energy storage system based on deep learning, and will not be described in detail here.

[0095] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described deep learning-based thermal energy management optimization method for lithium battery energy storage systems is also provided. This electronic device may be... Figure 5 The terminal device or server shown. This embodiment uses this electronic device as an example of a server. Figure 5 As shown, the electronic device includes a memory 402, a processor 404, and a transmission device 406. The memory 402 stores a computer program, and the processor 404 is configured to execute the steps of any of the above method embodiments through the computer program.

[0096] Optionally, in this embodiment, the aforementioned electronic device may be located in at least one of a plurality of network devices in a computer network.

[0097] Optionally, the transmission device 406 is used to receive or send data via a network. Specific examples of the network described above may include wired and wireless networks. In one example, the transmission device 406 includes a Network Interface Controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 406 is a Radio Frequency (RF) module used to communicate with the Internet wirelessly. Furthermore, the electronic device also includes a display 408 and a connection bus 410, which connects the various module components within the electronic device.

[0098] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A deep learning-based optimization method for thermal energy management of lithium battery energy storage systems, characterized in that, include: S1. Obtain the initial size of the battery cell, the cyclic expansion characteristics of the battery cell, the parameters of the thermal pad, the assembly deviation, the preload force and the operating condition data, and construct the contact boundary dataset; S2, based on the contact boundary dataset, divide the service stage, determine the cell expansion, thermal pad compression and preload attenuation of each service stage, and establish a contact pressure spatial mapping of the liquid cooling plate contact surface in combination with the assembly deviation to form a contact pressure spatial map. S3, based on the calibration relationship between pressure and contact thermal resistance in the parameters of the thermal pad, the contact pressure spatial map is converted into a contact thermal resistance spatial redistribution. S4. Determine the cell heat distribution at each service stage based on the operating condition data; input the contact thermal resistance spatial redistribution, the cell heat distribution, the operating condition data, and multiple preset candidate flow channel layouts into a trained deep learning model to predict the temperature field distribution of each preset candidate flow channel layout at each service stage. S5. Based on the temperature field distribution, identify hotspot sensitive areas, and perform a life-weighted evaluation of each preset candidate flow channel layout according to the duration, safety risks and economic life of each service stage, so as to determine the optimal candidate layout. S6. Based on the optimal candidate layout and the hotspot sensitive area, perform multi-objective collaborative optimization on the main flow channel, auxiliary flow channel and bonding reinforcement area of ​​the liquid cooling plate.

2. The method for optimizing thermal energy management of a lithium battery energy storage system based on deep learning according to claim 1, characterized in that, S1 includes: The obtained initial dimensions of the battery cell, the cyclic expansion characteristics of the battery cell, the parameters of the thermal pad, the assembly deviation, the preload force, and the operating condition data are subjected to data cleaning, normalization, time synchronization, and spatial coordinate processing to construct the contact boundary dataset. The initial dimensions of the battery cell include: initial thickness, length, width, and material thermal parameters; The cell cyclic expansion characteristics include: cyclic expansion curve, heating characteristics under different states of charge and health, degree of cell expansion, and direction of cell expansion; The parameters of the thermal pad include: initial thickness, compressive modulus, creep characteristics, resilience, thermal conductivity and the calibrated relationship between pressure and contact thermal resistance, as well as the degree of compression aging of the thermal pad. The assembly deviations include: module assembly tolerances, liquid cooling plate mounting boundaries, and contact surface geometric parameters; The preload force includes: end plate preload force, strap tension, and bolt tightening force; The operating condition data includes: current, voltage, SOC, SOH, cell cycle count, capacity retention rate, internal resistance change, temperature, liquid cooling flow rate, inlet and outlet temperature difference, coolant pressure, ambient temperature, and charge / discharge rate during the operation of the energy storage system.

3. The method for optimizing thermal energy management of lithium battery energy storage systems based on deep learning according to claim 2, characterized in that, S2 include: Based on the cell cycle count, capacity retention rate, internal resistance change, cell expansion degree, and thermal pad compression aging degree, the service life of the energy storage battery module is divided into the initial service stage, the mid-term service stage, and the late service stage. For each stage of service, The cell expansion amount during this service stage is obtained by interpolation based on the cyclic expansion curve and the number of cell cycles. Based on the initial thickness, length and width of the cell, and the expansion direction of the cell, the cell expansion amount during this service stage is allocated to the corresponding spatial position of the liquid cooling plate contact surface. Based on the compression modulus and creep characteristics in the parameters of the thermal pad, and combined with the preload of the service stage, the compression amount of the thermal pad in the service stage is determined, and the compression amount of the thermal pad is allocated to the corresponding spatial position of the liquid cooling plate contact surface. The preload attenuation amount for this service stage is determined based on the preload and preload attenuation law, and the preload attenuation amount for this service stage is allocated to the corresponding spatial position of the liquid cooling plate contact surface. Based on the contribution of cell expansion, thermal pad compression, preload attenuation, and assembly deviation at the same spatial location, the contact pressure value at that spatial location is determined. The contact pressure spatial map is formed by the contact pressure values ​​at each spatial location. The contact pressure spatial map is used to characterize the process by which the bonding state between the liquid cooling plate and the battery cell evolves from an initial uniform bonding to a local weak bonding, a local high-pressure bonding, or a local detachment.

4. The method for optimizing thermal energy management of a lithium battery energy storage system based on deep learning according to claim 3, characterized in that, The calibration relationship between pressure and contact thermal resistance uses the thermal pad compression and contact pressure as independent variables and contact thermal resistance as the dependent variable; S3 includes: For each spatial location on the contact surface of the liquid cooling plate, the contact thermal resistance value at that spatial location is determined based on the contact pressure value and the compression amount of the thermal pad at that spatial location. The spatial redistribution of contact thermal resistance is determined based on the contact thermal resistance values ​​at each spatial location.

5. The method for optimizing thermal energy management of a lithium battery energy storage system based on deep learning according to claim 1, characterized in that, The prediction architecture of the deep learning model includes a spatial feature branch and a temporal feature branch; The spatial feature branch uses a convolutional neural network, U-Net, or graph neural network to extract the spatial relationship between the liquid cooling plate contact surface, the flow channel coverage area, and the cell temperature field. The time feature branch adopts a recurrent neural network, LSTM or Transformer structure to learn the influence of charging and discharging conditions on the cell temperature field over time, the influence of cooling flow rate on the cell temperature field over time, the influence of ambient temperature on the cell temperature field over time, and the influence of cell aging state on the cell temperature field over time. The output of the trained deep learning model includes the temperature field distribution, maximum temperature, maximum temperature difference, hot spot probability, hot spot duration, and cooling effectiveness evaluation results for each preset candidate flow channel layout in each service stage.

6. The method for optimizing thermal energy management of a lithium battery energy storage system based on deep learning according to claim 5, characterized in that, Identifying hotspot-sensitive areas based on the temperature field distribution includes: In each service phase, the spatial locations where the contact thermal resistance value exceeds the preset thermal resistance threshold corresponding to that service phase and the cell heat value exceeds the preset heat generation threshold corresponding to that service phase are marked as hot spot sensitive locations for that service phase. The hotspot sensitive locations marked at each service stage will be compared across stages to select spatial locations that were not marked as hotspot sensitive locations in the early or middle service stages but were marked as hotspot sensitive locations in the later service stages, which will be added as hotspot sensitive locations in the later stages of the service life. The hotspot sensitive locations within each service phase are connected and aggregated to form the hotspot sensitive area of ​​the corresponding service phase. The newly added hotspot sensitive locations in the later stage of the service life are connected and aggregated to form the newly added hotspot sensitive area in the later stage of the service life.

7. The method for optimizing thermal energy management of a lithium battery energy storage system based on deep learning according to claim 1, characterized in that, Based on the duration, safety risks, and economic life of each service phase, a life-weighted evaluation of each preset candidate flow channel layout is performed to determine the optimal candidate layout, including: For each preset candidate flow channel layout, the cooling effect index of the preset candidate flow channel layout is calculated in the initial service stage, the mid-service stage and the late service stage. The duration of each service phase is determined based on its duration. Combined with the preset safety risk coefficient and economic life coefficient, the cooling effect index of each service phase is weighted and calculated to obtain the life weighted comprehensive score of each preset candidate flow channel layout. By comparing the lifetime-weighted comprehensive scores of each preset candidate flow channel layout, the preset candidate flow channel layout with the highest lifetime-weighted comprehensive score is determined as the optimal candidate layout.

8. The method for optimizing thermal energy management of a lithium battery energy storage system based on deep learning according to claim 7, characterized in that, The main channel of the liquid cooling plate covers the hot spot sensitive area formed by the spatial location marked as a hot spot sensitive location throughout the entire service phase; The auxiliary flow channel is used to compensate for the newly added hotspot sensitive areas in the later stage of the lifespan; the location, direction and coverage of the auxiliary flow channel are determined according to the spatial distribution of the newly added hotspot sensitive areas in the later stage of the lifespan.

9. The method for optimizing thermal energy management of a lithium battery energy storage system based on deep learning according to claim 8, characterized in that, The multi-objective collaborative optimization includes: Based on the optimal candidate layout, the continuous geometric parameters are iteratively optimized using a multi-objective optimization algorithm. The continuous geometric parameters include the channel width, channel depth, channel curvature, and channel spacing of the main channel and auxiliary channel of the liquid cooling plate, as well as the local pressing structure dimensions of the bonding reinforcement area. The objective function of the multi-objective optimization algorithm consists of multiple optimization sub-objectives, including: lifetime weighted comprehensive score, hotspot sensitive area coverage, hotspot sensitive area spatial distribution density, temperature uniformity, cooling efficiency, pressure drop, pump power consumption, manufacturing cost, and structural reliability. The Pareto optimal solution set is obtained through the multi-objective optimization algorithm to weigh the trade-offs among the various optimization sub-objectives.

10. A method for optimizing thermal energy management of lithium battery energy storage systems based on deep learning, characterized in that: include: The dataset construction module is used to acquire data on initial cell dimensions, cell cyclic expansion characteristics, thermal pad parameters, assembly deviations, preload, and operating conditions to construct a contact boundary dataset. The map construction module is used to divide the service stages based on the contact boundary dataset, determine the cell expansion, thermal pad compression and preload attenuation of each service stage, and establish a contact pressure space mapping of the liquid cooling plate contact surface in combination with the assembly deviation to form a contact pressure space map. The conversion module is used to convert the contact pressure spatial map into a contact thermal resistance spatial redistribution based on the calibration relationship between pressure and contact thermal resistance in the parameters of the thermal pad. The prediction module is used to determine the cell heat distribution at each service stage based on the operating condition data; the contact thermal resistance spatial redistribution, the cell heat distribution, the operating condition data, and multiple preset candidate flow channel layouts are input into a trained deep learning model to predict the temperature field distribution of each preset candidate flow channel layout at each service stage. The optimal layout determination module is used to identify hotspot sensitive areas based on the temperature field distribution, and to perform a life-weighted evaluation of each preset candidate flow channel layout according to the duration, safety risks and economic life of each service stage, so as to determine the optimal candidate layout. The multi-objective optimization module is used to perform multi-objective collaborative optimization of the main flow channel, auxiliary flow channel and bonding reinforcement area of ​​the liquid cooling plate based on the optimal candidate layout and the hot spot sensitive area.

Images