A control method and device for a lithium battery energy storage system

Abstract

This invention discloses a control method and apparatus for a lithium battery energy storage system, relating to the field of energy storage control. The method includes acquiring multiphysics data of the lithium battery energy storage system and extracting its features; inputting the extracted multiphysics data into a digital twin model to obtain the predicted results output by the digital twin model; generating an optimal control strategy under preset constraints, with system efficiency, battery balance, and waste heat recovery rate as optimization objectives; and performing topology switching control on the topology network according to the optimal control strategy. By acquiring multiphysics data, the obtained data perspectives are richer, resulting in more accurate results from the digital twin model. When a fault occurs in the lithium battery energy storage system predicted by the digital twin model, an optimization objective is generated, and the topology network is adjusted, thereby improving the safety of the energy storage system before the fault occurs.

Description

Technical Field

[0001] This invention relates to the field of energy storage control, and in particular to a control method and apparatus for a lithium battery energy storage system. Background Technology

[0002] Lithium-ion battery / supercapacitor hybrid energy storage systems are an emerging type of energy storage system for electric vehicles. Lithium-ion batteries possess high energy density and low power density, while supercapacitors, with their low energy density and high power density, effectively compensate for the shortcomings of lithium-ion batteries. Hybrid energy storage systems combining the characteristics of lithium-ion batteries and supercapacitors can effectively meet the power demands of the load. Energy management strategies are crucial for improving the efficiency of energy storage systems, enabling real-time allocation of load power. Traditional lithium-ion battery energy storage systems exhibit significant efficiency fluctuations across a wide load range, have a high proportion of energy consumption in the cooling system, low energy conversion efficiency, and a high false alarm rate for safety warnings. Summary of the Invention

[0003] The purpose of this invention is to provide a control method and apparatus for a lithium battery energy storage system. By acquiring multi-physics data, the obtained data perspectives are richer, resulting in more accurate output results from the digital twin model. After the digital twin model predicts a fault in the lithium battery energy storage system, an optimization target is generated, and the topology network is adjusted, thereby making adjustments before the fault occurs and improving the safety of the energy storage system.

[0004] To address the aforementioned technical problems, this invention provides a control method for a lithium battery energy storage system, comprising:

[0005] Acquire multi-physics data of a lithium battery, the multi-physics data including at least one of the following: voltage and current of the lithium battery in charge and discharge state, surface and internal temperature of the lithium battery, expansion stress of the lithium battery, and mechanical waves inside the lithium battery.

[0006] Feature extraction is performed on the multiphysics data;

[0007] The multiphysics data after feature extraction is input to obtain the prediction results output by the digital twin model. The prediction results include the safe time window, battery health status, and system efficiency under different topology modes.

[0008] Based on the multiphysics data after feature extraction and the prediction results, an optimal control strategy is generated under preset constraints with system efficiency, battery balance and waste heat recovery rate as optimization objectives.

[0009] According to the optimal control strategy, topology switching control is performed on the topology network of the lithium battery energy storage system. The topology network is a network formed by each lithium battery through a power conversion circuit. The topology network includes at least two of the following: star topology, ring topology, and hybrid topology.

[0010] On the other hand, acquiring multiphysics data of lithium batteries includes:

[0011] When the lithium battery is in a charging and discharging state, the voltage and current in the electrical circuit containing the lithium battery are obtained;

[0012] The temperature of the lithium battery is obtained from its surface and interior.

[0013] Obtain the expansion stress on the surface of the lithium battery;

[0014] Mechanical waves inside the lithium battery are acquired by collecting data from the electrode leads, the center of the lithium battery, and the edge heat dissipation surface.

[0015] On the other hand, based on the multiphysics data after feature extraction and the prediction results, under preset constraints, with system efficiency, battery balance, and waste heat recovery rate as optimization objectives, an optimal control strategy is generated, including:

[0016] The results of feature extraction and the prediction results of the digital twin model are input into the LSTM network to obtain the dynamic parameters of the battery after processing the time series data. The time series data includes the voltage ripple of the lithium battery in the charging and discharging state and the charging and discharging rate of the lithium battery.

[0017] The output of the LSTM network is represented as follows: , of which Weight matrix, For timing input data, This is the predicted output of the LSTM network;

[0018] The results of feature extraction, the prediction results of the digital twin model, and the output of the LSTM network are processed using the NSGA-III algorithm to generate Pareto solution sets. The non-time-series data includes the vibration intensity of mechanical waves inside the lithium battery and the heat dissipation flow rate of the fluid in the cooling system for dissipating heat from the lithium battery.

[0019] The input to the NSGA-III algorithm includes the prediction output of the LSTM network. and non-time-series environmental data , represented as ;

[0020] With the optimization objectives of maximizing system efficiency, optimizing battery balance, and maximizing waste heat recovery rate, and satisfying preset constraints, the optimal solution is selected from the Pareto solution set, and the optimal solution is used to generate the optimal control strategy.

[0021] On the other hand, it also includes:

[0022] Active cooling is achieved by controlling the phase change material embedded between the lithium batteries and the microchannel cold plate that contacts the power devices in the power conversion circuit of the lithium batteries;

[0023] Waste heat is recovered by controlling the thermoelectric generators that are arranged in a gradient along the heat dissipation airflow direction on the heat dissipation path of the lithium battery.

[0024] When performing a topology cut-off operation on a faulty lithium battery, the cooling system is pressurized and operated to cool the lithium battery.

[0025] Obtain the output voltage of the thermoelectric generator;

[0026] When the output voltage is greater than the first voltage threshold, the recovered electrical energy is fed back to the DC bus via the boost circuit;

[0027] When the output voltage is less than the second voltage threshold, the system switches to external auxiliary power supply to maintain the operation of the cooling system.

[0028] On the other hand, the construction process of the digital twin model includes:

[0029] A digital twin model is constructed based on the physical structure, operating conditions, and external environmental conditions of the lithium battery energy storage system.

[0030] The simulation results output by the digital twin model are compared with the measured values ​​of the actual lithium battery energy storage system.

[0031] When the deviation between the simulation results and the measured values ​​of the physical system exceeds the deviation threshold, the micro- and macro-parameters are corrected.

[0032] When the deviation between the simulation results and the measured values ​​of the physical system is lower than the deviation threshold, the construction of the digital twin model is considered complete.

[0033] On the other hand, according to the optimal control strategy, topology switching control is performed on the topology network of the lithium battery energy storage system, including:

[0034] The load fluctuation rate of the external load connected to the lithium battery is determined based on the collected voltage and current data.

[0035] The safety warning level of the lithium battery is determined based on the prediction results of multiphysics data and digital twin models;

[0036] Obtain the system efficiency of the lithium battery energy storage system predicted by the digital twin model under various topology modes;

[0037] The optimal topology pattern is determined based on the load volatility, security warning level, and predicted system efficiency.

[0038] When the load volatility exceeds the volatility threshold, the system efficiency is lower than the efficiency threshold, or the security warning level is raised, the topology network is controlled to switch between star topology, ring topology, and hybrid topology according to the optimal topology mode.

[0039] On the other hand, the safety warning level of the lithium battery is determined based on the prediction results of multiphysics data and digital twin models, including:

[0040] The electrochemical impedance spectroscopy of the lithium battery is determined based on the collected voltage and current data. When the risk of lithium plating is detected based on the electrochemical impedance spectroscopy, it is determined to be a level one warning.

[0041] Electrical faults are determined based on the mechanical waves inside the lithium battery. When a short circuit fault is detected, it is identified as a level two warning.

[0042] Based on the safety time window predicted by the digital twin model, when the safety time window is less than the preset safety time threshold, it is determined to be a level three warning.

[0043] On the other hand, electrical faults are determined based on the mechanical waves inside the lithium battery. When a short-circuit fault is detected, a level two warning is issued, including:

[0044] Bandpass filtering and wavelet thresholding are applied to the acoustic emission signal;

[0045] The filtered and denoised acoustic emission signal is segmented according to a preset time window, and the peak amplitude and root mean square value of the signal in each window are determined.

[0046] When the peak amplitude or root mean square value of the signal exceeds the normal operating threshold, the corresponding window is marked as a suspected abnormal window.

[0047] Extract feature parameters from the suspected abnormal window. The feature parameters include at least one of the following: peak factor, cumulative duration of peak amplitude, center frequency, characteristic frequency ratio, rising edge slope, and pulse count.

[0048] Input the feature parameters into the pre-trained recognition model;

[0049] When the identification model determines that a preset number of feature parameters meet the short-circuit threshold condition, it determines that a short-circuit fault exists and identifies it as a level two warning.

[0050] On the other hand, when the identification model determines that a short-circuit fault exists and identifies it as a level-two warning after determining that a preset number of feature parameters meet the short-circuit threshold condition, it further includes:

[0051] If the current topology is star, disconnect the electrical connection between the faulty lithium battery and the DC bus that charges and discharges the battery.

[0052] If the current topology is a ring, disconnect the electrical connection between the faulty lithium battery and the adjacent lithium battery, and enable the backup connection path to maintain the continuity of the ring loop.

[0053] After isolation is completed, the surface temperature, internal temperature and mechanical waves of the isolated lithium battery are determined. If the fault spreads to adjacent lithium batteries, a level three warning is triggered.

[0054] To address the aforementioned technical problems, the present invention also provides a control device for a lithium battery energy storage system, comprising:

[0055] Memory, used to store computer programs;

[0056] A processor is used to execute the computer program to implement the steps of the control method for the lithium battery energy storage system described above.

[0057] This application provides a control method and apparatus for a lithium battery energy storage system, relating to the field of energy storage control. The method includes acquiring multiphysics data of the lithium battery energy storage system and extracting its features; inputting the extracted multiphysics data into a digital twin model to obtain the prediction results output by the digital twin model; generating an optimal control strategy under preset constraints, with system efficiency, battery balance, and waste heat recovery rate as optimization objectives; and performing topology switching control on the topology network according to the optimal control strategy. By acquiring multiphysics data, the obtained data perspectives are richer, resulting in more accurate results from the digital twin model. After a fault is predicted in the lithium battery energy storage system by the digital twin model, an optimization objective is generated, and the topology network is adjusted, thereby improving the safety of the energy storage system before the fault occurs. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the prior art and embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 A flowchart of a control method for a lithium battery energy storage system provided by the present invention;

[0060] Figure 2This is a schematic diagram of the control device for a lithium battery energy storage system provided by the present invention. Detailed Implementation

[0061] The core of this invention is to provide a control method and device for a lithium battery energy storage system. By acquiring multi-physics data, the obtained data perspectives are richer, resulting in more accurate output results from the digital twin model. After the digital twin model predicts a fault in the lithium battery energy storage system, it generates an optimization target and adjusts the topology network, thereby making adjustments before the fault occurs and improving the safety of the energy storage system.

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Figure 1 A flowchart of a control method for a lithium battery energy storage system provided by the present invention is provided. The control method for the lithium battery energy storage system includes:

[0064] S11: Acquire multi-physics data of the lithium battery. The multi-physics data includes at least one of the following: voltage and current of the lithium battery under charging and discharging conditions, surface and internal temperature of the lithium battery, expansion stress of the lithium battery, and mechanical waves inside the lithium battery.

[0065] By acquiring multimodal data in real time and dynamically adjusting the topology, efficient energy conversion and millisecond-level operating condition response are achieved over a wide load range;

[0066] A multi-physics sensor array is deployed, employing high-speed synchronous sampling (<1μs) to achieve simultaneous acquisition of electrical, thermal, mechanical, and acoustic signals. A matrix-type bidirectional DC-DC module based on GaN devices (switching frequency ≥500kHz) supports dynamic switching between star and ring topologies. Built upon this GaN-based matrix-type bidirectional DC-DC module, the system achieves rapid and stable switching between two topology modes through the collaboration of hardware topology design and software decision-making mechanisms. Reconstruction trigger conditions include load volatility (>30%), efficiency threshold (<92%), and safety warning levels. Features (dV / dQ curves, impedance spectra) are extracted in real-time via FPGA, and a digital twin model is constructed in the cloud, linking microscopic battery aging with macroscopic topology efficiency. This cloud-based digital twin model serves as the core hub for system virtual mapping, real-time analysis, and predictive optimization. By replicating the physical entity, operating conditions, and environmental interaction characteristics of the lithium battery energy storage system at a 1:1 scale, it establishes a dynamic bridge between the microscopic battery aging mechanism and macroscopic topology efficiency, achieving a control upgrade from passive response to proactive prediction.

[0067] S12: Feature extraction from multiphysics data;

[0068] For voltage and current data, calculate the differential curve of voltage versus capacitance (dV / dQ curve) and extract the position, height and width of characteristic peaks; at the same time, by applying a small AC excitation and measuring the response, calculate the electrochemical impedance spectroscopy (EIS), extract the real and imaginary parts at different frequencies, and obtain characteristic parameters such as ohmic internal resistance and charge transfer impedance.

[0069] For temperature data, calculate the rate of temperature change, temperature gradient, and heat accumulation as thermal characteristic parameters.

[0070] For force signals (expansion stress), the stress change rate, peak stress and their corresponding time points are extracted to characterize the volume change and internal pressure accumulation of the battery.

[0071] For acoustic emission signals, the filtering, segmentation, and extraction methods of characteristic parameters (peak factor, cumulative duration of peak amplitude, center frequency, proportion of characteristic frequency, rising edge slope, pulse count) have been described in detail in the secondary early warning process. These features are also used as input for subsequent digital twin models.

[0072] S13: Input the multiphysics data after feature extraction into the digital twin model to obtain the prediction results output by the digital twin model. The prediction results include the safe time window, battery health status, and system efficiency under different topology modes.

[0073] The processor builds a digital twin model in the cloud. This model is a 1:1 virtual mapping of the lithium battery energy storage system, including the three-dimensional structure of the physical entity, electrical connections, thermal field distribution, mechanical response model, and environmental interaction interface.

[0074] The digital twin model receives feature extraction results from the edge layer (including dQ / dV curve features, EIS features, thermal features, mechanical features, and acoustic emission features), calculates the current state of health (SOH) through built-in battery aging models (such as SEI film growth kinetics model and lithium deposition model), and predicts future degradation trends.

[0075] The safe time window model, combined with a thermal runaway propagation model (based on the heat conduction equation and chemical reaction kinetics), simulates the remaining time from the current state to the triggering of thermal runaway. By comparing the model simulation results with measured temperature and voltage data, the model parameters are dynamically calibrated to ensure that the accuracy of the predicted safe time window meets engineering requirements.

[0076] The digital twin model embeds loss calculation sub-models for three topologies: star, ring, and hybrid. For the current operating conditions (load power, battery state of charge, temperature distribution), it simulates the conduction loss, switching loss, and circulating current loss under each of the three topologies, outputting predicted system efficiency values ​​for each topology mode as the basis for topology switching decisions.

[0077] S14: Based on the multiphysics data and prediction results after feature extraction, the optimal control strategy is generated under preset constraints with system efficiency, battery balance and waste heat recovery rate as optimization objectives.

[0078] A hybrid optimization model is adopted to simultaneously improve system efficiency, battery balance and safety threshold, achieving global optimal control under multiple constraints;

[0079] The extracted feature data (including dQ / dV curve features, electrochemical impedance spectroscopy features, temperature change rate, stress change rate, acoustic emission characteristic parameters, etc.) and the prediction results output by the S13 digital twin model (safe time window, battery health status, system efficiency under various topology modes) are jointly input into the hybrid optimization controller. This controller adopts an algorithm that combines NSGA-III and LSTM.

[0080] The optimization objectives are set as follows: maximizing system efficiency, maximizing battery balance (i.e., minimizing the difference in state of charge among individual battery cells), and maximizing waste heat recovery rate. Simultaneously, preset constraints must be met, including: voltage fluctuation amplitude not exceeding the rated value, battery temperature rise not exceeding safety limits, and lithium plating risk level not exceeding the permissible level. NSGA-III searches the solution space, generating a set of Pareto non-dominated solutions, and then selects an optimal compromise solution based on the current operating condition preferences (usually prioritizing safety, then considering efficiency). This optimal solution corresponds to a set of control parameters, including the target topology, switching frequency, pulse width modulation duty cycle, and cooling fan speed, which together form the optimal control strategy.

[0081] The objective function of the dynamic optimization model is: max[η, 1 / SOC difference, waste heat recovery rate], with constraints: voltage fluctuation ≤ ±1.5%, temperature rise ≤ 35K, and lithium plating risk level < 2.

[0082] S15: According to the optimal control strategy, perform topology switching control on the topology network of the lithium battery energy storage system. The topology network is a network formed by each lithium battery through a power conversion circuit. The topology network includes at least two of the following: star topology, ring topology, and hybrid topology.

[0083] Based on the generated optimal control strategy, switching commands are issued to the topology controller. The topology network consists of a matrix of bidirectional DC-DC modules based on GaN devices, and each lithium battery cell can be flexibly connected to different buses or loops through a switching matrix.

[0084] When the optimal control strategy indicates a switch to a star topology, the topology controller closes the connection switches between each lithium battery cell and the central bus, while simultaneously disconnecting the series switches between the cells, forming a radial structure with the central bus as the junction point. Star topologies are suitable for scenarios with low load fluctuations and high load balance.

[0085] When the optimal control strategy indicates a switch to a ring topology, the topology controller connects each lithium battery cell in series, forming a closed loop, and disconnects it from the central bus. Ring topology is suitable for high-power, high-efficiency transmission scenarios, but requires active suppression of circulating current.

[0086] When the optimal control strategy indicates a hybrid topology, the controller connects some battery cells to a star structure and the other part to a ring structure, achieving a compromise between balance and efficiency.

[0087] During the switching process, the processor monitors load volatility, current system efficiency, and safety warning levels in real time. If the load volatility exceeds a preset threshold, the system efficiency falls below a preset threshold, or the safety warning level increases, the optimal strategy is recalculated and a topology switch is executed. The entire switching process is completed via high-frequency switching action, taking extremely short time and capable of responding to power fluctuations on the order of seconds. After the switch is completed, the processor continues to collect multiphysics data, forming a closed-loop feedback mechanism to readjust the topology structure when operating conditions change.

[0088] This application provides a control method for a lithium battery energy storage system, relating to the field of energy storage control. The method includes collecting multi-physics field data of the lithium battery energy storage system and extracting its features; inputting the extracted multi-physics field data into a digital twin model to obtain the prediction results output by the digital twin model; generating an optimal control strategy under preset constraints, with system efficiency, battery balance, and waste heat recovery rate as optimization objectives; and performing topology switching control on the topology network according to the optimal control strategy. By acquiring multi-physics field data, the obtained data perspectives are richer, resulting in more accurate results from the digital twin model. When a fault occurs in the lithium battery energy storage system predicted by the digital twin model, an optimization objective is generated, and the topology network is adjusted, thereby enabling adjustments before the fault occurs and improving the safety of the energy storage system.

[0089] Based on the above embodiments:

[0090] In some embodiments, acquiring multiphysics data of a lithium battery includes:

[0091] When the lithium battery is charging or discharging, obtain the voltage and current in the electrical circuit containing the lithium battery.

[0092] The temperature of the lithium battery is obtained from its surface and interior.

[0093] Obtain the expansion stress on the surface of the lithium battery;

[0094] Mechanical waves inside a lithium battery are acquired at the electrode leads, the center of the lithium battery, and the edge heat dissipation surface.

[0095] Multimodal data real-time acquisition types:

[0096] Electrical signals: voltage and current (via voltage / current loop sensor);

[0097] Thermal signal: Battery surface and internal temperature (distributed temperature sensor);

[0098] Force signal: Battery expansion stress (piezoresistive sensor);

[0099] Acoustic signal: mechanical waves inside the battery (10–100kHz acoustic emission sensor).

[0100] The matrix-type bidirectional DC-DC module is based on the reconfigurable power layer of GaN devices and supports dynamic switching between star topology (for high-balance scenarios) and ring topology (for high-efficiency scenarios).

[0101] When the load volatility exceeds 30%, the system efficiency is less than 92%, or the safety warning level increases, a reconfiguration is triggered. The load volatility, warning level, and digital twin prediction results are integrated through a fuzzy comprehensive evaluation algorithm to output the optimal topology pattern.

[0102] In the case implementation, based on the electro-thermal-mechanical-acoustic multi-physics field sensor network (synchronous sampling accuracy <1μs) and edge-cloud collaborative computing, the anomaly detection response time is ≤200ms and the false alarm rate of safety warning is <5%. The cross-scale warning mechanism is as follows: Level 1 warning predicts the risk of lithium plating 30 minutes in advance through rapid EIS analysis (10kHz-1MHz band); Level 2 warning uses acoustic emission signals (10-100kHz band) to identify micro short circuit characteristics, and the false alarm rate of triggering topology isolation is reduced to 1%; Level 3 warning combines thermal runaway propagation model to accurately predict the safe time window (error <5 seconds), providing key data for emergency response.

[0103] In some embodiments, based on the multiphysics data and prediction results after feature extraction, an optimal control strategy is generated under preset constraints with system efficiency, battery balance, and waste heat recovery rate as optimization objectives, including:

[0104] The results of feature extraction and the prediction results of the digital twin model are input into the LSTM network to obtain the dynamic parameters of the battery after processing the time series data. The time series data includes the voltage ripple of the lithium battery under charging and discharging conditions and the charging and discharging rate of the lithium battery.

[0105] The output of the LSTM network is represented as follows: , of which Weight matrix, For timing input data, This is the predicted output of the LSTM network;

[0106] The results of feature extraction, the prediction results of the digital twin model, and the output of the LSTM network are processed using the NSGA-III algorithm to generate Pareto solution sets. The non-time-series data includes the vibration intensity of mechanical waves inside the lithium battery and the heat dissipation flow rate of the fluid in the cooling system for cooling the lithium battery.

[0107] The NSGA-III algorithm's input includes the predicted output of the LSTM network. and non-time-series environmental data , represented as ;

[0108] With the optimization objectives of maximizing system efficiency, optimizing battery balance, and maximizing waste heat recovery rate, and satisfying preset constraints, the optimal solution is selected from the Pareto solution set, and the optimal solution is used to generate the optimal control strategy.

[0109] LSTM processes time-series data (voltage ripple, charge / discharge rate) to predict short-term states, while NSGA-III optimizes non-time-series data (vibration intensity, heat dissipation flow) to generate Pareto solutions. In the case implementation, through the synergistic effect of dynamic topology reconfiguration technology and the NSGA-III-LSTM hybrid optimization algorithm, the system achieves an energy conversion efficiency of ≥94.5% within a 10%-100% load range, with efficiency fluctuations strictly controlled within ±0.7%. Furthermore, the GaN device matrix topology combined with 500kHz high-frequency switching control reduces conduction and switching losses, resulting in a significant improvement in efficiency across all operating conditions. Dynamic bus reconfiguration (star / ring mode switching) effectively suppresses circulating current losses, with SOC differences between battery clusters ≤3%, and equalization performance improved by 40%.

[0110] Data collected by the multiphysics sensor array is transmitted to the edge-cloud collaborative computing platform; the data processed by the edge-cloud collaborative computing platform (such as feature extraction results and digital twin model output) is input to the NSGA-III-LSTM hybrid optimization controller; the NSGA-III-LSTM hybrid optimization controller outputs control signals to the dynamically reconfigurable topology network according to the optimization results to realize topology switching (such as star / ring); at the same time, the gradient waste heat recovery device (TEG) feeds back the recovered electrical energy to the DC bus of the system, and the status of waste heat recovery (such as TEG output voltage) is also transmitted as an input signal to the thermal-electric collaborative control module, which may affect the decision of the optimization controller;

[0111] Dynamically reconfigurable topology network: Enables efficient energy conversion and dynamically adjusts the topology according to operating conditions; Multi-physics sensor array: Acquires multi-physics signals of the battery system in real time; Edge-cloud collaborative computing platform: Performs real-time feature extraction and digital twin modeling, providing data support for optimization and control; NSGA-III-LSTM hybrid optimization controller: Executes multi-objective optimization algorithms to generate optimal control strategies; Gradient waste heat recovery device: Recovers waste heat and converts it into electrical energy, improving system efficiency.

[0112] Cross-model rapid adaptation technology (based on transfer learning) enables the system to be deployed to different specifications of lithium battery packs in less than 24 hours (traditional solutions require 1-2 weeks); the edge computing layer data compression rate is >80%, reducing communication bandwidth requirements and supporting centralized monitoring of energy storage systems at the thousand-node level; the modular design supports flexible expansion of power capacity, and a single system can cover application scenarios from 50kWh to 1MWh.

[0113] In the case implementation: the switching between star / ring / hybrid modes of the dynamically reconfigurable topology is achieved in the following ways:

[0114] (a) A matrix-type bidirectional DC-DC module based on GaN devices, which achieves topology reconfiguration by changing the power switch state;

[0115] (b) The switching decision adopts a fuzzy comprehensive evaluation algorithm. The input parameters include load volatility α, safety warning level β, and digital twin prediction result γ. The specific steps include:

[0116] (i) Define the membership function:

[0117] 1. Load volatility α:

[0118] Low volatility: ;

[0119] High volatility: ;

[0120] 2. Safety warning level β:

[0121] ;

[0122] (ii) Output the optimal topology pattern by defuzzifying using the centroid method.

[0123] In some embodiments, it also includes:

[0124] Active cooling is achieved by controlling the phase change material embedded between lithium batteries and the microchannel cold plate that contacts the power devices in the power conversion circuit of the lithium battery.

[0125] Waste heat is recovered by controlling the thermoelectric generators that are arranged in a gradient along the heat dissipation airflow direction on the heat dissipation path of the lithium battery.

[0126] When performing a topology cut-off operation on a faulty lithium battery, the control cooling system is pressurized and used to cool the lithium battery.

[0127] Obtain the output voltage of the thermoelectric generator;

[0128] When the output voltage is greater than the first voltage threshold, the recovered electrical energy is fed back to the DC bus via the boost circuit;

[0129] When the output voltage is lower than the second voltage threshold, switch to external auxiliary power supply to maintain the operation of the cooling system.

[0130] The composite heat dissipation and energy recovery system includes an active cooling layer, a waste heat recovery layer, and a thermo-electric synergistic control module. The active cooling layer consists of phase change material (melting point 45℃) embedded between battery modules, with microchannel cold plates directly contacting the power devices. The waste heat recovery layer features thermoelectric generators (TEGs) arranged in a gradient along the heat dissipation airflow direction, recovering electrical energy and feeding it back to the DC bus via a Boost circuit (contributing 0.8-1.5% efficiency gain). The thermo-electric synergistic control system links the heat dissipation system power with topology reconfiguration; when the TEG output voltage > 5V, the bus feedback path is activated. The active cooling layer is embedded: the phase change material directly fills the gaps between battery modules, and the microchannel cold plates are in close contact with the power devices and battery module casing, achieving zero-distance heat dissipation.

[0131] The waste heat recovery layer is a convection type: the TEG array is arranged in a gradient along the heat dissipation airflow direction (from the cold plate heat dissipation end to the system air outlet), with the hot end facing the upstream of the heat dissipation airflow (high temperature side) and the cold end facing the downstream of the airflow (low temperature side), maximizing the utilization of temperature difference;

[0132] The control module is centralized: the collaborative controller is connected to the heat dissipation system, TEG recovery unit and topology controller through interfaces to realize centralized signal processing and linkage.

[0133] In some embodiments, the process of constructing a digital twin model includes:

[0134] A digital twin model is constructed based on the physical structure, operating conditions, and external environmental conditions of the lithium battery energy storage system.

[0135] The simulation results output by the digital twin model are compared with the measured values ​​of the actual lithium battery energy storage system.

[0136] When the deviation between the simulation results and the measured values ​​of the physical system exceeds the deviation threshold, the micro and macro parameters are corrected.

[0137] The construction of the digital twin model is considered complete when the deviation between the simulation results and the measured values ​​of the physical system is less than the deviation threshold.

[0138] In the case implementation, topology reconstruction took less than 10ms (compared to 50ms for traditional systems), supporting rapid mode switching in scenarios with second-level power fluctuations; the digital twin model synchronously correlates SEI film growth (microscopic) with system efficiency (macroscopic), enabling dynamic parameter calibration throughout the lifecycle, as detailed below:

[0139] 1. Calibration trigger conditions:

[0140] Real-time triggering: Every time the edge layer collects 100 sets of multi-physics field data (electric-thermal-mechanical-acoustic), it is simultaneously uploaded to the cloud digital twin model. If the deviation between the simulated system efficiency and battery internal resistance and the measured values ​​of the physical system exceeds 2%, real-time calibration is initiated.

[0141] Cycle triggering: Based on a fixed cycle (e.g., every 24 hours) or the number of battery cycles (e.g., every 100 cycles), combined with the battery state of health (SOH) assessment results, a full calibration is initiated to adapt to the cumulative effect of SEI film growth.

[0142] 2. Calibration core content:

[0143] Microscopic parameter calibration: Based on measured impedance spectrum, dV / dQ curve and other data, correct microscopic parameters such as SEI film thickness and growth rate in digital twin to ensure that the model is consistent with the actual aging state of physical battery.

[0144] Macroscopic parameter calibration: Based on the measured system efficiency values, adjust macroscopic parameters such as topology loss factor and power transmission efficiency coefficient to correct the loss calculation model under different topology modes.

[0145] Correlation coefficient calibration: Update the quantitative correlation coefficient of SEI film growth-system efficiency to adapt to the dynamic changes in the micro-macro coupling relationship during battery aging.

[0146] 3. Calibration execution logic:

[0147] Edge layer preprocessing: The collected multi-source data is filtered and denoised, and features are extracted and compressed (compression rate >80%) to reduce the cloud computing load.

[0148] Cloud-based model correction: Using algorithms such as Kalman filtering and least squares, the simulation results of the model are compared with the measured data of the physical system, and the micro and macro parameters and correlation coefficients are dynamically adjusted.

[0149] Calibration results are distributed: The calibrated model parameters are synchronized to the edge layer NSGA-III-LSTM hybrid optimization controller to guide the selection of topology mode and optimization of PWM strategy.

[0150] In some embodiments, topology switching control is performed on the topology network of the lithium battery energy storage system according to an optimal control strategy, including:

[0151] The load fluctuation rate of the external load connected to the lithium battery is determined based on the collected voltage and current data.

[0152] The safety warning level of lithium batteries is determined based on the prediction results of multiphysics data and digital twin models.

[0153] Obtain the system efficiency of lithium battery energy storage system predicted by digital twin model under various topology modes;

[0154] The optimal topology pattern is determined based on load volatility, security warning level, and predicted system efficiency.

[0155] When the load volatility exceeds the volatility threshold, the system efficiency is lower than the efficiency threshold, or the security warning level is raised, the control topology network switches between star topology, ring topology, and hybrid topology according to the optimal topology mode.

[0156] I. Indicator Preprocessing: Unifying Quantification Standards;

[0157] The raw data of the three core indicators are mapped to the [0,1] interval to eliminate heterogeneity differences and clarify the tendency of the indicators to correlate with the topological patterns:

[0158] Load volatility: ≤30% (stable) is quantified as 0.1-0.3 (leaning towards star pattern), 30%-60% (volatile) is quantified as 0.4-0.7 (leaning towards mixed pattern), >60% (violent) is quantified as 0.8-1.0 (leaning towards cyclical pattern).

[0159] Warning levels: Level 0 (no warning) is quantified as 0.1 (prone to star-shaped), Level 1 (lithium plating risk) is quantified as 0.4 (prone to star-shaped / mixed), Level 2 (micro-short circuit) is quantified as 0.7 (prone to toroidal), and Level 3 (precursor to thermal runaway) is quantified as 1.0 (forced toroidal).

[0160] Digital twin prediction results: Star-shaped optimization is 0.1, mixed optimization is 0.5, ring-shaped optimization is 0.9, and no significant difference is 0.4 (default star-shaped).

[0161] II. Fuzzification Processing: Constructing a Fit Degree Membership Function;

[0162] Using the triangular membership function, the quantized crisp values ​​are converted into fuzzy fit degrees (between 0 and 1, with higher values ​​indicating stronger fit) for three types of topologies: star, ring, and hybrid.

[0163] Star schema fit: 1 when the quantization value is ≤0.3 (fully fit), decreasing in the range of 0.3-0.5, and 0 when it is ≥0.5 (completely unfit).

[0164] Ring fit: 0 when the quantization value is ≤0.5, increases in the range of 0.5-0.8, and is 1 when it is ≥0.8 (perfect fit);

[0165] Mixed fit: The quantization value increases in the range of 0.3-0.5, is 1 when it is 0.5-0.6, decreases in the range of 0.6-0.8, and is 0 when it is ≤0.3 or ≥0.8.

[0166] III. Weight Allocation: Clarify the priority of indicators;

[0167] Based on the principles of prioritizing safety, balancing efficiency, and providing predictive support, the weights of the indicators are determined using the analytic hierarchy process (the sum of the weights is 1, and the overall weight of the algorithm is 4, which is reinforced through decision priority).

[0168] The early warning level has a weight of 0.4 (core safety), the load volatility has a weight of 0.35 (key to dynamic response), and the digital twin prediction result has a weight of 0.25 (assistance for full life cycle optimization).

[0169] IV. Fuzzy Synthesis: Calculating the overall fit;

[0170] The weighted average method is used to multiply the fuzzy fit of each index by its corresponding weight and then sum them to obtain the comprehensive fit of the three types of topologies:

[0171] Star-shaped overall fit = (early warning level weight × star-shaped fit) + (load volatility weight × star-shaped fit) + (digital twin prediction result weight × star-shaped fit);

[0172] The overall fit of ring-shaped and hybrid structures is calculated in the same way using the formula above.

[0173] V. Decision Output: Determine the optimal topology;

[0174] Output results according to the maximum membership principle, while setting constraint rules to ensure safety and stability:

[0175] Choose the topology pattern with the highest overall adaptability (star / ring / hybrid).

[0176] If the fit is equal: if the star and hybrid types are equal, choose the star type (efficiency priority); if the ring and hybrid types are equal, choose the ring type (safety priority).

[0177] Mandatory rules: When the warning level is 3, regardless of other indicators, directly output a ring topology; when the load is ≤20% and there is no warning, directly output a star topology to simplify the process.

[0178] VI. Execution and Calibration: Dynamic Optimization Decision Making;

[0179] The decision command is sent to the topology controller, the switching time is ≤10ms, and the system maintains stable operation for at least 100ms after the switch to avoid frequent switching.

[0180] The edge layer collects efficiency, voltage fluctuation, and temperature rise data in real time under the new topology. The cloud verifies the rationality of the decision through a digital twin model. Recalculation is triggered when the efficiency is less than 92% or the warning is upgraded.

[0181] The membership function parameters are optimized every 24 hours based on historical data to adapt to battery aging and changes in operating conditions.

[0182] In some embodiments, determining the safety warning level of a lithium battery based on prediction results from multiphysics data and a digital twin model includes:

[0183] The electrochemical impedance spectroscopy of the lithium battery is determined based on the collected voltage and current data. When the risk of lithium plating is detected by the electrochemical impedance spectroscopy, it is determined as a level one warning.

[0184] Electrical faults are identified based on the mechanical waves inside the lithium battery. When a short circuit fault is detected, it is designated as a level two warning.

[0185] Based on the safety time window predicted by the digital twin model, when the safety time window is less than the preset safety time threshold, it is determined to be a level three warning.

[0186] Level 1 Warning:

[0187] This edge computing-based EIS (Electronic Information System) rapid analysis (10kHz-1MHz) predicts lithium plating risk. Lithium plating in lithium batteries is the phenomenon where lithium ions deposit as metallic lithium on the negative electrode surface without being embedded in the crystal lattice. This directly alters the internal electrochemical characteristics of the battery, resulting in significant differences in the impedance spectrum in the 10kHz-1MHz mid-to-high frequency range. This provides a basis for risk identification. The prediction lead time for lithium plating risk is ≥30 minutes (a warning can be issued 30 minutes before significant lithium plating actually occurs when the battery enters the critical state of lithium plating). The false alarm rate for risk assessment is <1% (significantly reducing misjudgments caused by fluctuations in a single parameter through multi-dimensional verification of feature parameters and dynamic model calibration). A single analysis takes ≤150ms, fully adapting to the real-time requirements of edge computing and supporting continuous monitoring in scenarios with second-level power fluctuations.

[0188] Level 3 Warning:

[0189] By combining the thermal runaway propagation model with the digital twin to predict the safe time window, the maximum power operation of the cooling system is triggered, as follows:

[0190] (a) Ensuring the accuracy of safe time window predictions:

[0191] Model dynamic calibration: The digital twin model compares the inference results of the thermal runaway propagation model with the measured data of the physical system (such as temperature distribution and temperature rise rate) every 10 seconds, and uses the Kalman filter algorithm to correct the model parameters (such as thermal conductivity coefficient and heat generation rate coefficient) to ensure that the prediction error is less than 5 seconds.

[0192] Multi-model fusion validation: A simplified thermal runaway propagation model and a deep learning prediction model (trained based on historical thermal runaway case data) are introduced. The results of the three models are weighted and fused (with weights of 0.5, 0.3, and 0.2, respectively) to reduce the prediction bias of a single model.

[0193] (II) Reliability assurance of the cooling system at maximum power operation:

[0194] Redundancy design: Key components of the heat dissipation system (such as variable frequency pumps, fans, and cold plates) adopt a dual backup design. If the main component fails, the backup component will automatically start operating within 5ms to ensure uninterrupted heat dissipation power.

[0195] Thermal-electric co-protection: The thermal-electric co-control module monitors the power supply voltage and current of the heat dissipation system in real time to avoid overload during maximum power operation; when the TEG output voltage is <3V, it automatically switches to external auxiliary power supply to ensure continuous operation of the heat dissipation system.

[0196] Synergistic optimization of phase change material and cold plate: The contact area between the phase change material embedding position and the microchannel cold plate is optimized (contact area ≥90%) to ensure that the heat absorbed by the phase change material is quickly transferred to the cold plate and to prevent the phase change material from losing its cooling effect due to heat storage saturation.

[0197] (III) Rapid Response Guarantee:

[0198] Cloud computing power optimization: Using GPU clusters for parallel computing, the single simulation time of the thermal runaway propagation model and digital twin model is ≤200ms, meeting the real-time prediction requirements.

[0199] Communication link priority design: The communication channel for the heat dissipation system control commands is set to the highest priority to avoid conflicts with other data transmissions and ensure that the command issuance delay is ≤10ms.

[0200] Edge-cloud collaborative load reduction: The edge layer preprocesses the collected multi-source data in advance (such as outlier removal and feature extraction), with a data compression rate of >80%, reducing the cloud computing load and improving the inference speed.

[0201] In some embodiments, electrical faults are determined based on mechanical waves inside the lithium battery. When a short-circuit fault is detected, a level two warning is issued, including:

[0202] Bandpass filtering and wavelet thresholding are applied to the acoustic emission signal;

[0203] The filtered and denoised acoustic emission signal is segmented according to a preset time window, and the peak amplitude and root mean square value of the signal in each window are determined.

[0204] When the peak amplitude or root mean square value of a signal exceeds the threshold under normal operating conditions, the corresponding window is marked as a suspected abnormal window.

[0205] Extract feature parameters from suspected abnormal windows. Feature parameters include at least one of the following: peak factor, cumulative duration of peak amplitude, center frequency, characteristic frequency ratio, rising edge slope, and pulse count.

[0206] Input the feature parameters into the pre-trained recognition model;

[0207] When the identification model determines that a preset number of feature parameters meet the short-circuit threshold condition, a short-circuit fault is identified and a level-two warning is issued.

[0208] Level 2 warning:

[0209] The acoustic emission signal frequency band (10-100kHz) is used to identify micro-short circuit characteristics, and the topology controller is activated to start the isolation mode. The specific steps are as follows:

[0210] Step 1: Real-time acquisition and preprocessing of acoustic emission signals;

[0211] Distributed acoustic emission sensors (resonant frequency 50kHz, sensitivity ≥100mV / Pa) are deployed at key locations in each battery module (electrode leads, module center, and edge heat dissipation surface). The sensors are tightly bonded to the battery casing with a coupling agent to ensure attenuated transmission of mechanical wave signals. Combined with the system's high-speed synchronous sampling mechanism (sampling accuracy <1μs), the acoustic emission signals are continuously acquired. The sampling frequency is set to 500kHz (to satisfy the Nyquist sampling theorem for signals in the 10-100kHz frequency band and avoid signal distortion). The acquired data is transmitted to the edge layer FPGA in real time.

[0212] Filtering and denoising: An algorithm combining bandpass filtering (10-100kHz) and wavelet threshold denoising is used to filter out irrelevant signals such as environmental vibration (e.g., cooling fans, structural resonance), electromagnetic interference, etc., while retaining effective information of the target frequency band;

[0213] Signal segmentation and preliminary feature screening: The continuous signal is segmented into 10ms analysis windows. The peak amplitude and root mean square (RMS) of the signal within each window are calculated. If either parameter exceeds the normal operating threshold (calibrated using historical data, such as peak amplitude ≥ 200μV, RMS ≥ 50μV), it is marked as a suspected abnormal window and proceeds to the subsequent feature extraction process. If it does not exceed the threshold, the data in that window is discarded to reduce the computational load.

[0214] Step 2: Extraction and accurate identification of micro-short circuit feature parameters (executed by edge layer FPGA);

[0215] For suspected abnormal windows, three types of feature parameters strongly correlated with micro-short circuits are extracted, and accurate identification is achieved through multi-dimensional verification to avoid false positives:

[0216] Peak factor (peak value / root mean square value): The peak factor of a micro-short-circuit signal is ≥6 (≤3 under normal operating conditions), reflecting the pulse characteristics of the signal;

[0217] Peak amplitude cumulative duration: The duration of peak amplitude ≥150μV within the statistical window. This duration is ≥2ms during micro-short circuit (<0.5ms under normal operating conditions).

[0218] Center frequency: Calculate the power spectral density (PSD) of the signal. The center frequency of the micro-short-circuit signal is concentrated in the range of 30-80kHz (the center frequency under normal operating conditions has no fixed range and the energy is dispersed).

[0219] Characteristic frequency proportion: The proportion of signal energy in the 30-80kHz frequency band to the total energy in the 10-100kHz frequency band is statistically analyzed. During micro-short circuit, this proportion is ≥70% (under normal operating conditions, it is ≤40%).

[0220] Rising edge slope: Calculate the rising edge slope of the signal pulse (the rate at which the amplitude rises from 10% peak value to 90% peak value). The rising edge slope of the micro short-circuit signal is ≥100μV / μs (≤30μV / μs under normal operating conditions).

[0221] Pulse count: The number of pulses within the statistical window that meet the requirements of peak value ≥ 100μV and rising edge slope ≥ 50μV / μs. The pulse count is ≥ 3 during micro short circuit (≤ 1 under normal operating conditions).

[0222] The edge layer pre-stores a lightweight recognition model trained on a large amount of micro-short-circuit experimental data (using Naive Bayes or shallow neural networks, with inference time ≤20ms), and inputs the above 6 types of feature parameters into the model:

[0223] If ≥4 types of feature parameters meet the micro-short circuit threshold condition, and two consecutive analysis windows (total 20ms) identify it as a suspected micro-short circuit, then it is determined to be a confirmed micro-short circuit fault.

[0224] If 2-3 types of characteristic parameters meet the threshold conditions, it is judged as a suspected micro short circuit. The system will continue to monitor for the next 3 windows (30ms). If the characteristic parameters continue to meet the threshold conditions, it will be upgraded to a confirmed fault; otherwise, it will be judged as a false trigger.

[0225] Step 3: The linkage topology controller starts isolation mode (edge-topology layer collaborative execution);

[0226] Once the edge layer confirms a micro-short circuit fault, it immediately initiates the linkage response process, with a total response time of ≤50ms (including identification and topology switching):

[0227] The edge layer sends fault commands to the topology controller via a high-speed communication bus (such as CANFD, with a transmission rate of ≥2Mbps). The commands contain key information such as the faulty battery module number, the time of the fault, and the amplitude of characteristic parameters, ensuring that the topology controller can accurately locate the fault point.

[0228] Upon receiving the command, the topology controller immediately triggers the switching array action of the reconfigurable power layer: for a star topology, it disconnects the connection switch between the faulty battery module and the central bus while maintaining normal access for other modules to ensure uninterrupted power transmission of the system; for a ring topology, it disconnects the series switches on both sides of the faulty module to isolate the faulty module from the ring loop, while activating redundant interfaces to reconfigure the ring loop (e.g., switching a dual-ring topology to a single-ring topology) to prevent system shutdown due to loop breakage.

[0229] During isolation, the topology controller monitors the changes in bus voltage and current in real time and adjusts the PWM control strategy to ensure that voltage fluctuations are ≤±1.5% (meeting system constraints) and avoid power surges caused by isolation actions.

[0230] In some embodiments, after the identification model determines that a short-circuit fault exists and issues a level-two warning when it determines that a preset number of feature parameters meet the short-circuit threshold condition, the method further includes:

[0231] If the current topology is star, disconnect the electrical connection between the faulty lithium battery and the DC bus that charges and discharges the battery.

[0232] If the current topology is a ring, disconnect the electrical connection between the faulty lithium battery and the adjacent lithium battery, and enable the backup connection path to maintain the continuity of the ring loop.

[0233] After isolation is completed, the surface temperature, internal temperature and mechanical waves of the isolated lithium battery are determined. If the fault spreads to adjacent lithium batteries, a level three warning is triggered.

[0234] Level 2 Early Warning Linkage and Status Feedback:

[0235] The system activates a level-two warning with both sound and light, and simultaneously reports fault information to the cloud-based digital twin model, providing data support for subsequent fault analysis and battery maintenance.

[0236] After isolation is completed, the topology controller sends a signal to the edge layer indicating successful isolation. The edge layer continuously monitors the acoustic emission signal and temperature data of the isolation module. If the signal returns to normal and there is no abnormal temperature rise (temperature rise ≤ 5K / 10min), the isolation state is maintained. If the fault spreads (such as the adjacent module showing micro-short circuit characteristics), a three-level early warning linkage is triggered.

[0237] In summary, this high-efficiency energy conversion and intelligent monitoring system for lithium battery energy storage achieves an energy conversion efficiency of ≥94.5% within a 10%-100% load range through the synergistic effect of dynamic topology reconfiguration technology and the NSGA-III-LSTM hybrid optimization algorithm, with efficiency fluctuations strictly controlled within ±0.7%. Furthermore, the GaN device matrix topology combined with 500kHz high-frequency switching control reduces conduction and switching losses, resulting in a significant improvement in efficiency across all operating conditions. Dynamic bus reconfiguration (star / ring mode switching) effectively suppresses circulating current losses, with SOC differences between battery clusters ≤3%, and equalization performance improved by 40%.

[0238] Furthermore, based on an electro-thermal-mechanical-acoustic multi-physics sensor network (synchronous sampling accuracy <1μs) and edge-cloud collaborative computing, the anomaly detection response time is ≤200ms and the false alarm rate for safety warnings is <5%. A cross-scale warning mechanism is implemented: Level 1 warning uses rapid EIS analysis (10kHz-1MHz band) to predict lithium plating risk 30 minutes in advance; Level 2 warning utilizes acoustic emission signals (10-100kHz band) to identify micro-short circuit characteristics, reducing the false alarm rate for triggering topology isolation to 1%; Level 3 warning combines a thermal runaway propagation model to accurately predict the safe time window (error <5 seconds), providing crucial data for emergency response.

[0239] Furthermore, topology reconstruction takes less than 10ms (compared to 50ms for traditional systems), supporting rapid mode switching in scenarios with second-level power fluctuations; the digital twin model synchronously correlates SEI film growth (microscopic) and system efficiency (macroscopic), enabling dynamic parameter calibration throughout the lifecycle, with a battery state of health (SOH) prediction error of ≤2%; the NSGA-III-LSTM hybrid algorithm, through deep fusion of time-series prediction (LSTM) and multi-objective optimization (NSGA-III), improves Pareto solution set coverage by 35% and control strategy iteration efficiency by 40% under dynamic loads.

[0240] Figure 2 This is a schematic diagram of a control device for a lithium battery energy storage system provided by the present invention. The control device for the lithium battery energy storage system includes:

[0241] Memory 21 is used to store computer programs;

[0242] The processor 22 is used to execute computer programs to implement the steps of the control method for the lithium battery energy storage system described above.

[0243] The description of the control device for the lithium battery energy storage system provided in this application is similar to that in the above embodiments and will not be repeated here.

[0244] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0245] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0246] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A control method for a lithium battery energy storage system, characterized in that, include: Acquire multi-physics data of a lithium battery, the multi-physics data including at least one of the following: voltage and current of the lithium battery in charge and discharge state, surface and internal temperature of the lithium battery, expansion stress of the lithium battery, and mechanical waves inside the lithium battery. Feature extraction is performed on the multiphysics data; The multiphysics data after feature extraction is input into the digital twin model to obtain the prediction results output by the digital twin model. The prediction results include the safe time window, battery health status, and system efficiency under different topology modes. Based on the multiphysics data after feature extraction and the prediction results, an optimal control strategy is generated under preset constraints with system efficiency, battery balance and waste heat recovery rate as optimization objectives. According to the optimal control strategy, topology switching control is performed on the topology network of the lithium battery energy storage system. The topology network is a network formed by each lithium battery through a power conversion circuit. The topology network includes at least two of the following: star topology, ring topology, and hybrid topology.

2. The control method for the lithium battery energy storage system as described in claim 1, characterized in that, Acquire multiphysics data of lithium batteries, including: When the lithium battery is in a charging and discharging state, the voltage and current in the electrical circuit containing the lithium battery are obtained; The temperature of the lithium battery is obtained from its surface and interior. Obtain the expansion stress on the surface of the lithium battery; Mechanical waves inside the lithium battery are acquired by collecting data from the electrode leads, the center of the lithium battery, and the edge heat dissipation surface.

3. The control method for the lithium battery energy storage system as described in claim 1, characterized in that, Based on the multiphysics data after feature extraction and the prediction results, an optimal control strategy is generated under preset constraints, with system efficiency, battery balance, and waste heat recovery rate as optimization objectives. This strategy includes: The results of feature extraction and the prediction results of the digital twin model are input into the LSTM network to obtain the dynamic parameters of the battery after processing the time series data. The time series data includes the voltage ripple of the lithium battery in the charging and discharging state and the charging and discharging rate of the lithium battery. The output of the LSTM network is represented as follows: , of which Weight matrix, For timing input data, This is the predicted output of the LSTM network; The results of feature extraction, the prediction results of the digital twin model, and the output of the LSTM network are processed using the NSGA-III algorithm to generate Pareto solution sets. The non-time-series data includes the vibration intensity of mechanical waves inside the lithium battery and the heat dissipation flow rate of the fluid in the cooling system for dissipating heat from the lithium battery. The input to the NSGA-III algorithm includes the prediction output of the LSTM network. and non-time-series environmental data , represented as ; With the optimization objectives of maximizing system efficiency, optimizing battery balance, and maximizing waste heat recovery rate, and satisfying preset constraints, the optimal solution is selected from the Pareto solution set, and the optimal solution is used to generate the optimal control strategy.

4. The control method for the lithium battery energy storage system as described in claim 1, characterized in that, Also includes: Active cooling is achieved by controlling the phase change material embedded between the lithium batteries and the microchannel cold plate that contacts the power devices in the power conversion circuit of the lithium batteries; Waste heat is recovered by controlling the thermoelectric generators that are arranged in a gradient along the heat dissipation airflow direction on the heat dissipation path of the lithium battery. When performing a topology cut-off operation on a faulty lithium battery, the cooling system is pressurized and operated to cool the lithium battery. Obtain the output voltage of the thermoelectric generator; When the output voltage is greater than the first voltage threshold, the recovered electrical energy is fed back to the DC bus via the boost circuit; When the output voltage is less than the second voltage threshold, the system switches to external auxiliary power supply to maintain the operation of the cooling system.

5. The control method for the lithium battery energy storage system as described in claim 1, characterized in that, The construction process of the digital twin model includes: A digital twin model is constructed based on the physical structure, operating conditions, and external environmental conditions of the lithium battery energy storage system. The simulation results output by the digital twin model are compared with the measured values ​​of the actual lithium battery energy storage system. When the deviation between the simulation results and the measured values ​​of the physical system exceeds the deviation threshold, the micro- and macro-parameters are corrected. When the deviation between the simulation results and the measured values ​​of the physical system is lower than the deviation threshold, the construction of the digital twin model is considered complete.

6. The control method for a lithium battery energy storage system as described in any one of claims 1 to 5, characterized in that, According to the optimal control strategy, topology switching control is performed on the topology network of the lithium battery energy storage system, including: The load fluctuation rate of the external load connected to the lithium battery is determined based on the collected voltage and current data. The safety warning level of the lithium battery is determined based on the prediction results of multiphysics data and digital twin models; Obtain the system efficiency of the lithium battery energy storage system predicted by the digital twin model under various topology modes; The optimal topology pattern is determined based on the load volatility, security warning level, and predicted system efficiency. When the load volatility exceeds the volatility threshold, the system efficiency is lower than the efficiency threshold, or the security warning level is raised, the topology network is controlled to switch between star topology, ring topology, and hybrid topology according to the optimal topology mode.

7. The control method for the lithium battery energy storage system as described in claim 6, characterized in that, The safety warning level of the lithium battery is determined based on the prediction results of multiphysics data and digital twin models, including: The electrochemical impedance spectroscopy of the lithium battery is determined based on the collected voltage and current data. When the risk of lithium plating is detected based on the electrochemical impedance spectroscopy, it is determined to be a level one warning. Electrical faults are determined based on the mechanical waves inside the lithium battery. When a short circuit fault is detected, it is identified as a level two warning. Based on the safety time window predicted by the digital twin model, when the safety time window is less than the preset safety time threshold, it is determined to be a level three warning.

8. The control method for the lithium battery energy storage system as described in claim 7, characterized in that, Electrical faults are determined based on the mechanical waves inside the lithium battery. When a short circuit fault is detected, a level two warning is issued, including: Bandpass filtering and wavelet thresholding are applied to the acoustic emission signal; The filtered and denoised acoustic emission signal is segmented according to a preset time window, and the peak amplitude and root mean square value of the signal in each window are determined. When the peak amplitude or root mean square value of the signal exceeds the normal operating threshold, the corresponding window is marked as a suspected abnormal window. Extract feature parameters from the suspected abnormal window. The feature parameters include at least one of the following: peak factor, cumulative duration of peak amplitude, center frequency, characteristic frequency ratio, rising edge slope, and pulse count. Input the feature parameters into the pre-trained recognition model; When the identification model determines that a preset number of feature parameters meet the short-circuit threshold condition, it determines that a short-circuit fault exists and identifies it as a level two warning.

9. The control method for a lithium battery energy storage system as described in claim 8, characterized in that, When the identification model determines that a preset number of feature parameters meet the short-circuit threshold condition, it determines that a short-circuit fault exists and, after determining it as a level-two warning, it further includes: If the current topology is star, disconnect the electrical connection between the faulty lithium battery and the DC bus that charges and discharges the battery. If the current topology is a ring, disconnect the electrical connection between the faulty lithium battery and the adjacent lithium battery, and enable the backup connection path to maintain the continuity of the ring loop. After isolation is completed, the surface temperature, internal temperature and mechanical waves of the isolated lithium battery are determined. If the fault spreads to adjacent lithium batteries, a level three warning is triggered.

10. A control device for a lithium battery energy storage system, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the control method for the lithium battery energy storage system as described in any one of claims 1 to 9.

Images