Battery energy storage cabinet protection system and method

Abstract

The invention relates to the technical field of electrochemical energy storage safety monitoring and battery intelligent management, in particular to a battery energy storage cabinet protection system and method, and the method comprises the steps: constructing and initializing a physical information neural network model, and obtaining pure state data in response to a high-rate charge-discharge operation state of a battery energy storage cabinet; inputting the pure state data into a physical information neural network model, and deducing to generate virtual temperature field data and chemical reaction rate field data in the blind area of the sensor; combining the virtual temperature field data, the chemical reaction rate field data and the pure state data to calculate and obtain electrochemical-thermal coupling lag entropy representing the internal damage degree of the system; based on a comparison result of the electrochemical-thermal coupling lag entropy and a preset risk level threshold, generating a self-adaptive hierarchical protection strategy; outputting a control signal of an energy storage cabinet execution mechanism according to the self-adaptive hierarchical protection strategy; according to the method, the detection robustness and the anti-interference capability of the system in the whole life cycle are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of electrochemical energy storage safety monitoring and intelligent battery management technology, specifically to a battery energy storage cabinet protection system and its method. Background Technology

[0002] Under the high-rate charge and discharge operation environment of the battery energy storage cabinet, the battery cluster exhibits complex electrochemical and thermodynamic coupling characteristics. Moreover, due to the limited confined space inside the cabinet, physical sensors cannot achieve full-area status monitoring. Existing protection schemes generally rely on limited surface point sensing data, which is difficult to penetrate and sense the heat accumulation and side reaction rates in the blind zone of the sensor. They are also highly susceptible to interference from dark current noise generated by non-uniform aging of the battery cells, leading to misjudgments. At the same time, due to the thermal island effect inside the battery, the surface monitoring data has a significant response lag relative to the actual internal state, making the traditional system insufficient in identifying the thermal runaway latency period. Furthermore, the one-size-fits-all power outage protection measures often used disrupt the continuity of power grid dispatch.

[0003] Therefore, how to achieve accurate reconstruction of multiphysics fields in the blind zone under high noise and sparse observation conditions, quantify the degree of thermal-electric coupling hysteresis, and implement hierarchical adaptive protection without interruption of operation has become an urgent technical problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a battery energy storage cabinet protection system and method, which avoids the lag in sensor blind zone thermal accumulation identification and the disruption to grid dispatch continuity caused by the one-size-fits-all shutdown protection of traditional battery management systems under high-rate operating conditions. Furthermore, it can reconstruct the internal multi-physics field through a physical information neural network, accurately identify latent risks, and achieve adaptive hierarchical protection. Specifically, the technical solution of this invention is as follows: A battery energy storage cabinet protection method includes: A multi-source sensor array is deployed in the battery clusters and cooling circuit inside the battery storage cabinet. The multi-source sensor array includes an electrical sensing unit and an environmental sensing unit. A physical information neural network model was constructed and initialized, which embedded electrochemical reaction kinetic equations and heat conduction partial differential equations. In response to the battery energy storage cabinet being in a high-rate charge / discharge operation state, a multi-physics field reconstruction and graded protection process is triggered, including: Step 1: Based on the multi-source heterogeneous data collected by the electrical sensing unit and the environmental sensing unit, the dark current disturbance noise caused by the difference in cell aging is removed by the non-uniform aging feature extraction algorithm to obtain pure state data. Step 2: Input the pure state data into the physical information neural network model, and through dynamic thermo-electrochemical field reconstruction processing, generate virtual temperature field data and chemical reaction rate field data within the sensor blind zone; Step 3: Combining virtual temperature field data, chemical reaction rate field data, and pure state data, the electrochemical-thermal coupling hysteresis entropy characterizing the degree of internal damage of the system is calculated through multidimensional coupling analysis. Step 4: Based on the comparison results between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold, an adaptive hierarchical protection strategy is generated. Step 5: Based on the adaptive hierarchical protection strategy, control signals for the energy storage cabinet actuators are output through hierarchical adjustment of electrical topology control and dielectric release control.

[0005] Preferably, the multi-source sensor array includes: Voltage and current sensors are installed at the battery module terminals and busbars; Temperature sensors are installed on the surface of the battery pack and at the inlet of the cooling channel; Combustible gas concentration sensor installed on the top of the cabinet.

[0006] Preferably, the dark current disturbance noise caused by differences in cell aging is removed by a non-uniform aging feature extraction algorithm, including: Obtain the historical charge-discharge curves and current voltage response data of each individual cell within the battery cluster; A benchmark consistency model is constructed based on historical charge and discharge curves, and the difference sequence between the current voltage response data and the output of the benchmark consistency model is calculated. Frequency domain analysis is performed on the difference sequence to identify fluctuation components with frequencies higher than the preset cutoff frequency as dark current disturbance noise, which are then filtered out from multi-source heterogeneous data.

[0007] Preferably, virtual temperature field data and chemical reaction rate field data within the sensor blind zone are generated through dynamic thermo-electrochemical field reconstruction processing, including: Pure state data is used as boundary condition constraints and input into the physical information neural network model; By using the partial differential equation of heat conduction in the physical information neural network model, the heat diffusion distribution in the area not covered by the sensor is calculated, and virtual temperature field data is generated. By utilizing the electrochemical reaction kinetic equations in the physical information neural network model, the activity level of side reactions inside the battery is calculated, generating chemical reaction rate field data.

[0008] Preferably, the electrochemical-thermal coupling hysteresis entropy characterizing the degree of internal damage in the system is calculated through multidimensional coupling analysis, including: The total amount of heat accumulation is obtained by performing time integration calculation on the virtual temperature field data; The difference between the peak time of the chemical reaction rate field data and the peak time of the voltage change in the pure state data is calculated to obtain the electrothermal response hysteresis time. Dimensionless normalization was performed on the total heat accumulation and the electrothermal response hysteresis time, respectively. Using preset weighting coefficients, the normalized total heat accumulation and the normalized electrothermal response lag time are weighted and summed to obtain the electrochemical-thermal coupling lag entropy.

[0009] Preferably, an adaptive hierarchical protection strategy is generated based on the comparison between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold, including: A first threshold, a second threshold, and a third threshold are preset, wherein the first threshold is less than the second threshold, and the second threshold is less than the third threshold; If the electrochemical-thermal coupling hysteresis entropy is greater than or equal to the first threshold and less than the second threshold, it is determined to be a latent risk and a first-level thermal suppression strategy is generated. The first-level thermal suppression strategy includes: maintaining the main loop closed and outputting instructions to adjust the opening of the liquid cooling flow channel distribution valve and bypass specific high-risk modules. If the electrochemical-thermal coupling hysteresis entropy is greater than or equal to the second threshold and less than the third threshold, it is determined to be a critical period risk, and a secondary thermal neutralization strategy is generated. The secondary thermal neutralization strategy includes: cutting off the local circuit and outputting instructions for the targeted release of trace amounts of inhibitor.

[0010] Preferably, the adaptive graded protection strategy generated based on the comparison between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold further includes: If the electrochemical-thermal coupling hysteresis entropy is greater than or equal to the third threshold, it is judged as a risk of outbreak, and a three-level blocking strategy is generated. The three-level blocking strategy includes: cutting off the main circuit of the system and outputting instructions for the release of the total flooding extinguishing medium and the activation of the explosion-proof pressure relief device.

[0011] A battery energy storage cabinet protection system, comprising: The data acquisition module is used to acquire multi-source heterogeneous data through a multi-source sensor array; The model building module is used to build physical information neural network models; The core processing module includes: The noise cleaning unit is used to remove dark current disturbance noise based on the non-uniform aging feature extraction algorithm to obtain clean state data. The field reconstruction unit is used to input pure state data into the physical information neural network model to deduce and generate virtual temperature field data and chemical reaction rate field data. The entropy calculation unit is used to combine virtual temperature field data, chemical reaction rate field data and pure state data to calculate the electrochemical-thermal coupling hysteresis entropy. The decision generation unit is used to generate an adaptive hierarchical protection strategy based on the comparison results between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold. The execution control unit is used to output control signals for the energy storage cabinet actuators according to the adaptive hierarchical protection strategy.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs a physical information neural network with embedded electrochemical and thermal conduction equations. It uses sparse electrical and environmental observation data as boundary constraints to deduce and generate virtual temperature fields and chemical reaction rate fields within the sensor blind zone. This breaks through the spatial limitations of physical sensor deployment, realizes the leap from single-point monitoring to full-domain field perception, can accurately capture the precursors of deep cell thermal runaway, and effectively solves the problem of not being able to fully cover monitoring in confined spaces.

[0013] 2. This invention introduces a non-uniform aging feature extraction algorithm, which effectively removes dark current disturbance noise caused by cell aging differences through frequency domain analysis technology, and obtains pure state data. This mechanism prevents the protection system from misjudging the fluctuations caused by normal aging as a precursor to a violent chemical reaction, ensuring that the data input to the model can truly reflect the main electrochemical state of the battery, and significantly improving the system's detection robustness and anti-interference ability throughout its entire life cycle.

[0014] 3. This invention proposes an electrochemical-thermal coupling hysteresis entropy index, which quantifies the total amount of heat accumulation and the electrothermal response hysteresis time through multidimensional coupling analysis. This index can keenly capture the response hysteresis caused by the thermal island effect inside the battery. It can reveal the potential risk of the thermoelectric balance inside the system being broken before the temperature value exceeds the standard, providing an accurate quantitative basis for latent period early warning and overcoming the shortcomings of traditional surface monitoring in identifying internal state hysteresis.

[0015] 4. This invention generates an adaptive hierarchical protection strategy based on risk entropy values, breaking the limitation of traditional protection that is either open or closed. During the latency and critical periods, by adjusting the liquid cooling flow, bypassing specific modules, or releasing trace amounts of inhibitors, the power output of the energy storage cabinet is maintained while ensuring safety. This refined management achieves a dynamic balance between safety and efficiency, avoiding the disruption of grid dispatch continuity caused by one-size-fits-all power outage protection. Attached Figure Description

[0016] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0018] Example 1: Please see Figure 1 A battery energy storage cabinet protection method includes: deploying a multi-source sensor array in the battery clusters and cooling circuit inside the battery energy storage cabinet, the multi-source sensor array including: electrical sensing units and environmental sensing units; constructing and initializing a physical information neural network model, the physical information neural network model embedding electrochemical reaction kinetic equations and heat conduction partial differential equations; and triggering a multi-physics field reconstruction and graded protection process in response to the battery energy storage cabinet being in a high-rate charge / discharge operation state, including: Step 1: Based on the multi-source heterogeneous data collected by the electrical sensing unit and the environmental sensing unit, the dark current disturbance noise caused by the difference in cell aging is removed by the non-uniform aging feature extraction algorithm to obtain pure state data. Step 2: Input the pure state data into the physical information neural network model, and through dynamic thermo-electrochemical field reconstruction processing, generate virtual temperature field data and chemical reaction rate field data within the sensor blind zone; Step 3: Combining virtual temperature field data, chemical reaction rate field data, and pure state data, the electrochemical-thermal coupling hysteresis entropy characterizing the degree of internal damage of the system is calculated through multidimensional coupling analysis. Step 4: Based on the comparison results between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold, an adaptive hierarchical protection strategy is generated. Step 5: Based on the adaptive hierarchical protection strategy, control signals for the energy storage cabinet actuators are output through hierarchical adjustment of electrical topology control and dielectric release control.

[0019] This embodiment details a digital twin protection logic driven by a physical information neural network (PINN), aiming to solve the problem of delayed identification of thermal accumulation in sensor blind zones by traditional BMS under high-rate charge and discharge conditions. The system deploys multi-source sensor arrays at key physical nodes in the battery cluster and cooling circuit to acquire spatially uneven and heterogeneous observations, providing sparse boundary conditions for subsequent field reconstruction. A PINN model with embedded electrochemical reaction kinetic equations and heat conduction partial differential equations is constructed and initialized. This model uses physical laws as prior knowledge to constrain the output of the neural network in data-scarce regions. In response to the current rate of the energy storage cabinet exceeding a preset threshold, such as 0.5C, and the duration meeting the trigger condition of the set duration, the system initiates a multi-physics field reconstruction process. In this process, the system performs non-uniform aging feature extraction, removes dark current disturbance noise caused by the difference in SOH of individual cells, and prevents false signal interference. The cleaned pure state data is mapped to the PINN model to generate virtual temperature field data and chemical reaction rate field data covering the sensor blind zone, realizing the leap from point monitoring to field sensing. Based on this, the electrochemical-thermal coupling hysteresis entropy is calculated through multidimensional coupling analysis to quantify the degree of asynchrony in the system's thermo-electric response; an adaptive hierarchical protection strategy is generated based on this entropy value to drive the actuator to perform topology control or directional medium release without interrupting power supply. This embodiment constructs a digital twin with see-through capabilities by embedding physical conservation laws into a neural network. It can accurately identify latent risks within sensor blind zones under high-risk steady-state conditions at high-rate operation. At the same time, it introduces electrochemical-thermal coupling hysteresis entropy as a core indicator to effectively quantify the response hysteresis caused by the thermal island effect inside the battery. This enables precise graded intervention before thermal runaway occurs, avoiding the disruption to grid dispatch continuity caused by traditional one-size-fits-all shutdown protection.

[0020] Example 2: The multi-source sensor array includes: voltage and current sensors deployed at the battery module terminals and busbars; temperature sensors deployed on the surface of the battery pack and at the cooling channel inlet; and combustible gas concentration sensors deployed on the top of the cabinet. This embodiment is a further specification of the multi-source sensor array deployment strategy in Embodiment 1. The deployment scheme is specifically designed to meet the strict constraint requirements of the PINN model on boundary conditions. The system collects current and voltage data of electrical connection nodes in real time through voltage and current sensors deployed at the battery module terminals and busbars. These data are used to calculate the Joule heat distribution generated by the contact resistance in real time, and serve as the input of the heat source term in the heat conduction equation. Data from temperature sensors deployed on the battery pack surface is used as the Dirichlet boundary condition for the thermal field. This is combined with temperature sensor data and flow velocity information from the cooling channel inlet to construct the thermal field boundary condition, i.e., the heat flux constraint. This is to address the reference temperature required for calculating the total heat accumulation in this embodiment. To address hardware support issues, the environmental sensing unit also includes an environmental temperature sensor located at the external air inlet of the energy storage cabinet or in the shaded area on the back of the cabinet, used to collect real-time environmental reference temperature data that is not affected by heat sources inside the cabinet. In addition, the system monitors characteristic gases such as carbon monoxide and volatile organic compounds by using combustible gas concentration sensors installed on the top of the cabinet. These gas concentration data are used as correction factors for the calculation of the chemical reaction rate field to correct the model's estimation of the intensity of side reactions. This embodiment provides precise physical constraint boundaries for the PINN model by deploying heterogeneous sensors at key thermoelectric nodes and fluid boundaries. This sparse but critical placement strategy enables the model to accurately invert the three-dimensional temperature field and reaction rate field inside the entire energy storage cabinet using a limited number of observation points, effectively solving the engineering problem of the limited space inside the energy storage cabinet making it difficult to provide full coverage wiring.

[0021] Example 3: The algorithm for extracting non-uniform aging features is used to remove dark current disturbance noise caused by differences in cell aging. This includes: obtaining the historical charge-discharge curves and current voltage response data of each individual cell in the battery cluster; constructing a benchmark consistency model based on the historical charge-discharge curves; calculating the difference sequence between the current voltage response data and the output of the benchmark consistency model; performing frequency domain analysis on the difference sequence; identifying fluctuation components with frequencies higher than the preset cutoff frequency as dark current disturbance noise; and filtering them out from multi-source heterogeneous data.

[0022] This embodiment further specifies the noise cleaning step in Embodiment 1; this step aims to eliminate interference from microscopic internal short circuits or polarization effects caused by inconsistent state of harmonics (SOH) of individual cells; the system obtains the historical charge-discharge curves (VQ curves) and current voltage response data of each individual cell in the battery cluster through the BMS data interface; based on the historical data, a benchmark consistency model is constructed using the least squares method. To clarify the mathematical structure of the model for easy computer solution, this embodiment uses a 6th-order polynomial fitting model: in, The current cumulative charge / discharge capacity, expressed in ampere-hours (Ah), is calculated using the following formula: , The fitting coefficients are obtained using the least squares method based on historical data. Taking a 280Ah lithium iron phosphate energy storage cell as an example, the typical range of fitting coefficient values ​​at a 0.5C rate is as follows: , , The absolute values ​​of the remaining higher-order coefficients are less than The model structure can approximate the nonlinear voltage plateau characteristics of the battery with minimal computational cost, while avoiding the gradient explosion problem of the exponential model in the low SOC region; the model characterizes the voltage response characteristics under ideal conditions without aging differences. The system calculates the difference sequence between the current voltage response data and the output of the benchmark consistency model, and performs a Fast Fourier Transform (FFT) on this difference sequence for frequency domain analysis; during this process, the system sets a preset cutoff frequency. This frequency is determined based on the time constant of the electrochemical reaction; the response is that a frequency higher than [a certain value] exists in the difference sequence. The system identifies the fluctuation components as dark current disturbance noise caused by micro-short circuits or contact instability, filters them out from the raw data, and finally outputs clean state data. This embodiment effectively eliminates false high-frequency signals caused by non-uniform aging through frequency domain feature separation technology, preventing the protection system from misjudging normal fluctuations caused by aging as precursors to violent chemical reactions. This ensures that the data input into the PINN model can truly reflect the main electrochemical state of the battery, significantly improving the robustness of the model throughout its entire life cycle.

[0023] Example 4: Through dynamic thermo-electrochemical field reconstruction processing, virtual temperature field data and chemical reaction rate field data within the sensor blind zone are generated. This includes: inputting pure state data as boundary condition constraints into a physical information neural network model; using the partial differential equation of heat conduction in the physical information neural network model to calculate the heat diffusion distribution in the sensor-uncovered area, generating virtual temperature field data; and using the electrochemical reaction kinetic equation in the physical information neural network model to calculate the activity level of side reactions inside the battery, generating chemical reaction rate field data. This embodiment is a further specification of the physical field reconstruction steps in Embodiment 1. This step utilizes the generalization ability of the PINN model to solve the problem of the sensor blind zone being invisible. The system uses the pure state data obtained in the preceding steps, including the temperature of known points... ,Voltage , which serve as the boundary and initial conditions for training the PINN model; The model utilizes the partial differential equation of heat conduction to calculate the heat diffusion distribution in areas not covered by the sensor, generating virtual temperature field data. The formula is as follows: in, The Hamiltonian operator represents the spatial divergence or gradient operation. This is virtual temperature field data, sourced from model output, physically representing the temperature distribution in the blind zone, and measured in K. The density is derived from a preset material library and its physical meaning is the density of the battery material, expressed in kg / m³. For lithium iron phosphate batteries, a typical value is used. ; Specific heat capacity, sourced from a preset material library, physically refers to the specific heat capacity of the battery material, and is measured in units of... Take typical values ; Thermal conductivity, sourced from a preset material library, physically represents the material's thermal conductivity, and is measured in units of... Considering the anisotropy of the battery's internal structure, the in-plane thermal conductivity is set. Normal thermal conductivity ; The internal heat generation rate is calculated according to Joule's law and spatial mapping rules, and its formula is as follows: in, The current in the pure state data; The internal resistance function of the battery cell varies with temperature. To address the computational bottleneck caused by missing variable mapping, this embodiment defines it as a specific Arrhenius form fitting formula: Among them, the reference internal resistance reference temperature Activation energy coefficient ; The contact resistance is calculated based on the voltage difference, and its calculation follows Ohm's law, as shown in the formula: in, and These are the real-time readings of the voltage sensors deployed at the busbar and the terminal in Example 2; To prevent numerically stable terms with a denominator of zero, the value is taken as... When detected At that time, the system retains the previous moment's... The value remains unchanged; Let be the geometric volume of the battery cell, and take the value. , For the polar volume, This is a spatial position indication function, and its specific mathematical expression adopts the definition of the generalized Heaviside step function: in For a predefined set of polar geometric regions, This is an indicator function; it takes the value 1 when the condition is met, and 0 otherwise; based on three-dimensional spatial coordinates. The system determines the value of a coordinate as 1 if it falls within the preset polarity geometry, and 0 otherwise. This formula clarifies the mapping logic from the contact resistance heat source to the volume heat source in Example 2. The unit is W / m³. By utilizing the electrochemical reaction kinetic equations and mass transport partial differential equations embedded in the model, the activity level of side reactions inside the battery is calculated, generating chemical reaction rate field data. To solve the problem of extrapolating a three-dimensional concentration field from one-dimensional SOC data. To address this challenge, this embodiment introduces a mass transport equation into the model as a physical constraint, and its calculation formula is as follows: Based on this transport equation, and in response to the requirement in Example 2 to correct the intensity of side reactions using gas concentration, this example constructs a reaction rate formula that includes a gas correction factor: in, This is chemical reaction rate field data, sourced from model output, with the physical meaning being the intensity of side reactions, and the unit is... / (m³·s); This is a pre-exponential factor, derived from laboratory calibration, and its physical meaning is a reaction frequency factor. For the SEI film thickening side reaction, its typical value range is... ; Activation energy, derived from laboratory calibration, is the minimum energy required for a reaction, and its unit is 1. The typical range of values ​​is ; Let be the ideal gas constant, and take the value of . ; The diffusion coefficient is derived from a pre-defined material library or based on the Arrhenius relation. The calculation, in physical terms, is the solid-phase diffusion rate of lithium ions in the electrode material, with units of m² / s; The reactant concentration is derived from the joint solution of the model based on the mass transport equation and the global SOC integral constraint. Its physical meaning is the local concentration of substances participating in the side reactions, and the unit is mol / m³. The reaction order is derived from laboratory calibration and its physical meaning is the sensitivity of the reaction rate to concentration. It is dimensionless and typically takes the value of 1. The gas concentration originates from the environmental sensing unit and is physically represented by a characteristic gas, such as CO, with concentration in ppm, and is measured over time. known functions Substitute into the equation; This is the gas correction factor, derived from experimental calibration. Its physical meaning is the gain weight of gas concentration on reaction rate, used to dynamically correct the model output when gas evolution is detected. This embodiment utilizes the strong constraints of physical equations, enabling the model to inversely calculate the temperature and reaction rate that conform to physical laws based on known boundary conditions in areas without physical sensors. This method breaks through the spatial limitations of physical sensor deployment and realizes holographic perception of the internal state of the energy storage cabinet, especially the ability to capture the precursors of thermal runaway in deep cells. To enable those skilled in the art to implement the aforementioned physical information neural network model, this embodiment discloses specific network structure parameters and the method for constructing the training loss function. Regarding the network structure: the model adopts a fully connected feedforward neural network architecture with an input layer dimension of 6, corresponding to spatiotemporal coordinates. and the real-time current as an external excitation and gas concentration In this embodiment, the time-varying external current excitation and ambient gas concentration are explicitly used as input variables for the neural network. Regarding the construction of the loss function: total loss function Defined as the weighted sum of data-driven loss, global physical constraint loss, and partial differential equation residual loss, the formula is as follows: in, The mean squared error (MSE) is used to constrain the consistency between the model output and the sensor's measured data at the pure state points. To avoid forcibly smoothing the reaction rate field, only point-to-point constraints on temperature and concentration are retained here, and their calculation formulas are as follows: in, Voltage based on pure state data With current Using preset curves and Estimated electrode surface concentration obtained from equation inversion; The specific inversion calculation logic is as follows: Based on the equivalent circuit model, the ohmic voltage drop is stripped, and the estimated open-circuit voltage is calculated using the following formula: in, The pre-calibrated DC internal resistance; using Inverse function of mapping relationship Obtain surface SOC estimates To address the code blocking issue caused by the lack of mapping relationships preventing label generation, and to correct the error in the original fitting formula that violates physical principles at boundary points (where the original formula yields a negative value at the full-fill point), this embodiment discloses a corrected inverse function polynomial fitting formula. For the lithium iron phosphate system, the calculation formula is as follows: Considering the strong nonlinear characteristics of lithium iron phosphate batteries—a flat voltage plateau with steep ends—simple low-order polynomials cannot meet the accuracy requirements. Therefore, this embodiment employs a 9th-order polynomial for fitting to ensure inversion accuracy; coefficients By standard The curve is pre-acquired through least-squares fitting and stored in the controller's non-volatile memory; in, The normalized voltage is calculated using the following formula: This embodiment focuses on the lithium iron phosphate battery system and sets... , Combined with maximum embedding concentration Calculated ;in For the lithium iron phosphate system in this embodiment, the value is... Based on the conversion between theoretical material density and compacted density; this step effectively eliminates polarization voltage interference during dynamic operation, providing... The deduction provides data references with clear physical meaning; This is a global current conservation constraint used to constrain the microscopic reaction rate field in an integral sense, replacing the original point-to-point MSE constraint to avoid contradictions in principle. Its calculation formula is as follows: in, For the first The infinitesimal volume or integral weight coefficient corresponding to each configuration point The effective volume of the active material in the battery cell is set as follows: ; The electron transfer number is set to 1 for lithium-ion intercalation reactions; here, it is used. The number is marked to distinguish it from the reaction order in the reaction kinetic equation. Eliminate parameter ambiguity; The value is Faraday's constant, taken as 96485 C / mol; The current global state of charge of the battery cluster is derived from observations calculated in real time by the BMS using the ampere-hour integration method, and serves as a global constraint term; this formula constrains the microscopic velocity field output by the model. The volume component needs to match the macroscopic current. This allows the model to adaptively generate non-uniform reaction rate distributions; The residual of the heat conduction equation is used to constrain the non-observed region to conform to the heat diffusion law. Its calculation formula is: The residuals of the reaction kinetic equations are: The residuals of the mass transport equation and the SOC integral constraint are used to ensure the consistency between the local concentration field and the global SOC. The calculation formula is as follows: in, This represents the current global state of charge of the battery cluster, derived from observations calculated in real-time by the BMS using the ampere-hour integration method. The number of sensor sampling points. The number of configuration points is usually taken as... Weighting coefficients Dynamic adjustment is performed using a linear annealing strategy, and its update formula is as follows: in, For the current training round, As a pre-set warm-up round, this strategy ensures that the model focuses primarily on data fitting in the early stages of training, and gradually forces the model to meet physical constraints in the later stages. The initial value is set to 0.1, and gradually increases to 1.0 as the training rounds increase, in order to ensure the dominance of physical laws.

[0024] Example 5: Through multidimensional coupling analysis, the electrochemical-thermal coupling hysteresis entropy characterizing the degree of internal damage in the system is calculated, including: performing time integration calculation on virtual temperature field data to obtain the total amount of heat accumulation; calculating the difference between the peak time of the chemical reaction rate field data and the peak time of the voltage change in the pure state data to obtain the electrothermal response hysteresis time; performing dimensionless normalization on the total amount of heat accumulation and the electrothermal response hysteresis time respectively; and using preset weighting coefficients, performing weighted summation on the normalized total amount of heat accumulation and the normalized electrothermal response hysteresis time to obtain the electrochemical-thermal coupling hysteresis entropy. This embodiment is a further specification of the entropy calculation step in Embodiment 1; this step introduces a composite index to characterize the degree of internal damage in the battery system; the system uses virtual temperature field data Perform time integration calculations to obtain the total amount of heat accumulation. The formula is as follows: in, The total amount of heat accumulation is derived from integral calculations and its physical meaning is the cumulative thermal effect over a time period, with units of K·s. The reference temperature is sourced from the environmental sensing unit and its physical meaning is the ambient reference temperature, with the unit being K. The integration start time corresponds to the triggering time of the response to the high-rate charge / discharge state, and the unit is seconds (s). The integration termination time corresponds to the current sampling time and is expressed in seconds. System calculates chemical reaction rate field data peak time Peak voltage change times in pure state data The difference; to address the issue of unknown future moments in real-time data stream processing, this embodiment employs a sliding time window local extremum detection algorithm to obtain the difference. and Specifically, the system maintains a length of For example, in a 30-second first-in-first-out data cache queue, when the value of the queue center point is detected to be greater than the values ​​of the adjacent points before and after it, that moment is locked as the local peak moment. Meanwhile, to prevent misjudgments due to data fluctuations during non-active reaction periods, a noise threshold determination mechanism is introduced: a noise threshold is only applied when the chemical reaction rate field data... The instantaneous value exceeds the preset floor noise threshold, for example... At that time, the above-mentioned lag time Only the calculation result is considered a valid value; otherwise, a mandatory command is enforced. ; If the current phase is a monotonically increasing phase without a peak, the current moment is used as a temporary peak moment to calculate the minimum lag time under the dynamic trend; the electrothermal response lag time is obtained. The formula is as follows: in, The hysteresis time of the electrothermal response is derived from differential calculation. Its physical meaning is the time misalignment of the thermal response relative to the electrical excitation, measured in seconds (s). It is used for... and Dimensionless normalization is then performed; the specific normalization formula is as follows: in, To ensure that the denominator is clearly defined during code calculation, the maximum allowable heat accumulation threshold is set as a constant in this embodiment. This value is calculated based on the total integral of the battery module under adiabatic conditions from the operating temperature to the critical temperature of thermal runaway, such as 130°C. This embodiment sets the maximum allowable lag time threshold as a fixed value. This value is based on the thermal diffusion physical time constant calculated from the cell's thermal conductivity and geometric radius; it utilizes preset weighting coefficients. and For the normalized total heat accumulation With normalized electrothermal response hysteresis time By performing a weighted summation, the electrochemical-thermal coupling hysteresis entropy is obtained. The formula is as follows: in, The electrochemical-thermal coupling hysteresis entropy is derived from weighted calculations and its physical meaning represents the degree of thermal runaway risk of the system; it is dimensionless. The weighting coefficient is a preset value, and its physical meaning is the relative importance of heat accumulation and lag time. This embodiment constructs an innovative index called electrochemical-thermal coupling hysteresis entropy, which keenly captures the response hysteresis phenomenon caused by the thermal island effect inside the battery. Even when the current temperature value has not exceeded the limit, the increase of this index can reveal the potential risk of the thermoelectric balance inside the system being broken, thus providing a quantitative basis for latent period early warning. Regarding the weighting coefficients mentioned in the embodiments and The specific method for obtaining the parameters is disclosed in this embodiment, which uses the parameter setting logic based on the Analytic Hierarchy Process (AHP). Considering that the thermal accumulation effect is more directly harmful than the time lag effect in the early stage of thermal runaway, a judgment matrix is ​​constructed to set the total amount of thermal accumulation. Relative to the hysteresis time of the electrothermal response The importance scale is 3, meaning it is slightly important; the weight vector is calculated using the following formula: Based on the differences in thermal stability of different battery chemical systems The value range is set to , The value range is set to And satisfy For ternary lithium batteries, it is recommended to take... For lithium iron phosphate batteries, it is recommended to take... The aforementioned weighting coefficients have been verified through an adiabatic accelerated calorimeter (ARC) experiment: In the simulation experiment, the weighting coefficients were... When the weight is set to 0.75, the system has the smallest prediction error for the thermal runaway trigger time, RMSE < 5s, which proves that the weight allocation is consistent with the failure characteristics of lithium iron phosphate batteries, which are dominated by thermal accumulation and have a delayed temperature rise.

[0025] Example 6: Based on the comparison between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold, an adaptive hierarchical protection strategy is generated, including: a preset first threshold, a second threshold, and a third threshold, wherein the first threshold is less than the second threshold, and the second threshold is less than the third threshold. If the electrochemical-thermal coupling hysteresis entropy is greater than or equal to the first threshold and less than the second threshold, it is determined to be a latent risk and a first-level thermal suppression strategy is generated. The first-level thermal suppression strategy includes: maintaining the main loop closed and outputting instructions to adjust the opening of the liquid cooling flow channel distribution valve and bypass specific high-risk modules. If the electrochemical-thermal coupling hysteresis entropy is greater than or equal to the second threshold and less than the third threshold, it is determined to be a critical period risk, and a secondary thermal neutralization strategy is generated. The secondary thermal neutralization strategy includes: cutting off the local circuit and outputting instructions for the targeted release of trace amounts of inhibitor. If the electrochemical-thermal coupling hysteresis entropy is greater than or equal to the third threshold, it is determined to be a risk of outbreak, and a three-level blocking strategy is generated. The three-level blocking strategy includes: cutting off the main circuit of the system and outputting instructions for the release of the total flooding extinguishing medium and the activation of the explosion-proof pressure relief device. This embodiment is a further specification of the decision generation and execution steps in Embodiment 1; the strategy is based on electrochemical-thermal coupled hysteresis entropy. The system dynamically matches non-interruptible or interruptible protection actions within a given value range; the system presets three incremental thresholds. ; In response to The system determines that there is a latent risk. At this time, the system executes a first-level thermal suppression strategy. Under the premise of maintaining the main circuit closed and not cutting off the external power supply, the system outputs a command to adjust the opening of the liquid cooling flow channel distribution valve to increase the cooling flow in the high entropy area, and controls the electrical topology switch to bypass specific high-risk modules, so that they temporarily exit the cycle for cryotherapy. The specific electrical topology for bypassing a specific module while maintaining the main circuit closure is as follows: Each battery module in the battery cluster is equipped with a bypass switch unit, which includes a normally open bypass contactor or bidirectional thyristor connected in parallel with the module, and a normally closed isolation contactor connected in series with the module. When the bypass command is executed, the controller follows the turn-on-then-off timing logic, closing the bypass contactor to establish a shunt path and disconnecting the series isolation contactor, thereby electrically isolating the high-risk module without interrupting the main circuit current path of the battery cluster. In response to The system determines that the risk is in the critical period. At this time, the system executes a two-stage thermal neutralization strategy, cuts off the local DC / DC connection of the risk cluster, and outputs instructions to the fire extinguishing medium pipeline valve to release trace amounts of inhibitors such as perfluorohexanone gas in a directional manner to neutralize free radicals and block the chain reaction. In response to The system determines that there is a risk of an outbreak. At this time, the system executes a three-level blocking strategy, immediately disconnects the main contactor of the high-voltage box to cut off the main circuit of the system, and simultaneously outputs a command to start the total flooding fire extinguishing medium release and explosion-proof pressure relief device to prevent physical explosion and fire spread. This embodiment breaks the binary opposition of traditional protection systems that require either on or off through a hierarchical adaptive strategy. In the first and second risk stages, the system can maintain some or all of its external power output while ensuring safety, maximizing the availability of energy storage devices. This refined control method is particularly suitable for power grid frequency regulation scenarios with extremely high continuity requirements, achieving a dynamic balance between safety and efficiency. Regarding the grading threshold mentioned in the embodiments To support the specific acquisition methods and the actual effect verification of the system, this embodiment provides the following specific numerical examples and experimental data. Threshold setting logic: Based on the electrochemical-thermal coupling hysteresis entropy distribution characteristics of 1000 thermal runaway events in the historical fault database, the threshold is determined using the statistical quantile method. The latency threshold is set as the 95th quantile of the entropy distribution, with the specific value taken as... ; The critical period threshold is set as the 98th quantile of the entropy distribution, with the specific value taken as... ; The threshold for the outbreak period is set as the 99.9th quantile of the entropy distribution, with the specific value taken as... ; Simulation verification data: A simulation environment for a 50Ah lithium iron phosphate battery cluster was constructed to simulate a micro short circuit fault in the central cell of the module; the experimental group used the PINN drive protection system of this embodiment, while the control group used a traditional BMS based on surface temperature monitoring; Experimental results show that the system's blind zone recognition capability is as follows: 50 seconds after the fault occurs, the virtual temperature field reconstructed by this system... The temperature rise of the central hot spot was detected, but the surface sensor temperature rise was only 0.5℃, which did not trigger a traditional BMS alarm. Quantitative indicator response: In At that time, the calculated electrochemical-thermal coupling hysteresis entropy Reaching 0.48, exceeding The system immediately triggers the first-level thermal suppression strategy; Final result comparison: This system, through early intervention, controlled the maximum temperature below 45℃; while the control group... Protection is triggered only when the surface temperature exceeds a preset threshold. At this point, the internal core temperature has already exceeded 120°C, leading to thermal runaway. The above data strongly demonstrates that this system can accurately identify blind zone risks and effectively quantify the thermal island effect. Example 7: Please see Figure 2A battery energy storage cabinet protection system includes: a data acquisition module for acquiring multi-source heterogeneous data through a multi-source sensor array; a model building module for establishing a physical information neural network model; and a core processing module including: a noise cleaning unit for removing dark current disturbance noise and obtaining pure state data based on a non-uniform aging feature extraction algorithm. The field reconstruction unit is used to input pure state data into the physical information neural network model to generate virtual temperature field data and chemical reaction rate field data; the entropy calculation unit is used to combine virtual temperature field data, chemical reaction rate field data and pure state data to calculate electrochemical-thermal coupling hysteresis entropy. The decision generation unit is used to generate an adaptive hierarchical protection strategy based on the comparison results between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold; the execution control unit is used to output control signals of the energy storage cabinet actuator according to the adaptive hierarchical protection strategy. This embodiment discloses a hardware and software entity architecture for executing the above method; the system is composed of multiple functional modules working together; the data acquisition module is connected to a multi-source sensor array via a CAN bus to acquire multi-source heterogeneous data such as voltage, current, temperature and gas concentration in real time; the model building module is responsible for building and storing a PINN model with an embedded PDE structure; The core processing module serves as the computational hub of the system. Its internal noise cleaning unit runs a non-uniform aging feature extraction algorithm and uses frequency domain analysis to remove dark current disturbance noise. The field reconstruction unit maps the cleaned data to the PINN model and generates virtual field data in the blind zone. The entropy calculation unit performs integration and difference operations to solve the electrochemical-thermal coupling hysteresis entropy. The decision generation unit compares the entropy value with a preset threshold and dynamically selects the protection strategy. The execution control unit sends specific control level signals to the liquid cooling pump, solenoid valve, DC / DC converter and fire-fighting equipment of the energy storage cabinet via hard-wired connection; This embodiment achieves closed-loop control from perception and computation to execution through modular hardware and software design; in particular, the core processing module integrates AI inference and physical modeling, enabling the device to be installed as an independent intelligent safety controller in existing energy storage systems, which greatly improves the safety and intelligence level of existing equipment.

[0026] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for protecting a battery energy storage cabinet, characterized in that, include: A multi-source sensor array is deployed in the battery clusters and cooling circuit inside the battery storage cabinet. The multi-source sensor array includes an electrical sensing unit and an environmental sensing unit. A physical information neural network model was constructed and initialized, which embedded electrochemical reaction kinetic equations and heat conduction partial differential equations. In response to the battery energy storage cabinet being in a high-rate charge / discharge operation state, a multi-physics field reconstruction and graded protection process is triggered, including: Step 1: Based on the multi-source heterogeneous data collected by the electrical sensing unit and the environmental sensing unit, the dark current disturbance noise caused by the difference in cell aging is removed by the non-uniform aging feature extraction algorithm to obtain pure state data. Step 2: Input the pure state data into the physical information neural network model, and through dynamic thermo-electrochemical field reconstruction processing, generate virtual temperature field data and chemical reaction rate field data within the sensor blind zone; Step 3: Combining virtual temperature field data, chemical reaction rate field data, and pure state data, the electrochemical-thermal coupling hysteresis entropy characterizing the degree of internal damage of the system is calculated through multidimensional coupling analysis. Step 4: Based on the comparison results between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold, an adaptive hierarchical protection strategy is generated. Step 5: Based on the adaptive hierarchical protection strategy, control signals for the energy storage cabinet actuators are output through hierarchical adjustment of electrical topology control and dielectric release control.

2. The battery energy storage cabinet protection method according to claim 1, characterized in that, Multi-source sensor array, including: Voltage and current sensors are installed at the battery module terminals and busbars; Temperature sensors are installed on the surface of the battery pack and at the inlet of the cooling channel; Combustible gas concentration sensor installed on the top of the cabinet.

3. The battery energy storage cabinet protection method according to claim 2, characterized in that, By employing a non-uniform aging feature extraction algorithm, dark current disturbance noise caused by differences in cell aging is removed, including: Obtain the historical charge-discharge curves and current voltage response data of each individual cell within the battery cluster; A benchmark consistency model is constructed based on historical charge and discharge curves, and the difference sequence between the current voltage response data and the output of the benchmark consistency model is calculated. Frequency domain analysis is performed on the difference sequence to identify fluctuation components with frequencies higher than the preset cutoff frequency as dark current disturbance noise, which are then filtered out from multi-source heterogeneous data.

4. The battery energy storage cabinet protection method according to claim 3, characterized in that, Through dynamic thermo-electrochemical field reconstruction processing, virtual temperature field data and chemical reaction rate field data within the sensor blind zone are generated, including: Pure state data is used as boundary condition constraints and input into the physical information neural network model; By using the partial differential equation of heat conduction in the physical information neural network model, the heat diffusion distribution in the area not covered by the sensor is calculated, and virtual temperature field data is generated. By utilizing the electrochemical reaction kinetic equations in the physical information neural network model, the activity level of side reactions inside the battery is calculated, generating chemical reaction rate field data.

5. A battery energy storage cabinet protection method according to claim 4, characterized in that, Through multidimensional coupling analysis, the electrochemical-thermal coupling hysteresis entropy characterizing the degree of internal damage in the system is calculated, including: The total amount of heat accumulation is obtained by performing time integration calculation on the virtual temperature field data; The difference between the peak time of the chemical reaction rate field data and the peak time of the voltage change in the pure state data is calculated to obtain the electrothermal response hysteresis time. Dimensionless normalization was performed on the total heat accumulation and the electrothermal response hysteresis time, respectively. Using preset weighting coefficients, the normalized total heat accumulation and the normalized electrothermal response lag time are weighted and summed to obtain the electrochemical-thermal coupling lag entropy.

6. A battery energy storage cabinet protection method according to claim 5, characterized in that, Based on the comparison between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold, an adaptive hierarchical protection strategy is generated, including: A first threshold, a second threshold, and a third threshold are preset, wherein the first threshold is less than the second threshold, and the second threshold is less than the third threshold; If the electrochemical-thermal coupling hysteresis entropy is greater than or equal to the first threshold and less than the second threshold, it is determined to be a latent risk and a first-level thermal suppression strategy is generated. The first-level thermal suppression strategy includes: maintaining the main loop closed and outputting instructions to adjust the opening of the liquid cooling flow channel distribution valve and bypass specific high-risk modules. If the electrochemical-thermal coupling hysteresis entropy is greater than or equal to the second threshold and less than the third threshold, it is determined to be a critical period risk, and a secondary thermal neutralization strategy is generated. The secondary thermal neutralization strategy includes: cutting off the local circuit and outputting instructions for the targeted release of trace amounts of inhibitor.

7. A battery energy storage cabinet protection method according to claim 6, characterized in that, Based on the comparison between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold, an adaptive hierarchical protection strategy is generated, which also includes: If the electrochemical-thermal coupling hysteresis entropy is greater than or equal to the third threshold, it is judged as a risk of outbreak, and a three-level blocking strategy is generated. The three-level blocking strategy includes: cutting off the main circuit of the system and outputting instructions for the release of the total flooding extinguishing medium and the activation of the explosion-proof pressure relief device.

8. A battery energy storage cabinet protection system, applied to the battery energy storage cabinet protection method according to any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire multi-source heterogeneous data through a multi-source sensor array; The model building module is used to build physical information neural network models; The core processing module includes: The noise cleaning unit is used to remove dark current disturbance noise based on the non-uniform aging feature extraction algorithm to obtain clean state data. The field reconstruction unit is used to input pure state data into the physical information neural network model to deduce and generate virtual temperature field data and chemical reaction rate field data. The entropy calculation unit is used to combine virtual temperature field data, chemical reaction rate field data and pure state data to calculate the electrochemical-thermal coupling hysteresis entropy. The decision generation unit is used to generate an adaptive hierarchical protection strategy based on the comparison results between the electrochemical-thermal coupling hysteresis entropy and the preset risk level threshold. The execution control unit is used to output control signals for the energy storage cabinet actuators according to the adaptive hierarchical protection strategy.

Images