Energy storage container fault early warning method and device based on real-time monitoring

Abstract

The invention provides an energy storage container fault early warning method and device based on real-time monitoring, and relates to the field of intelligent models. The method comprises the following steps: constructing a standardized multi-dimensional time series data set based on multi-modal operation state data; constructing a state estimation digital twin based on the multi-dimensional time sequence data set and the energy storage container topology model; defining a system association graph structure based on the energy storage container topology model, and constructing a state reference representation based on the state estimation digital twin; constructing a space-time diagram convolutional neural network model by taking the system association diagram structure and the state reference representation as constraint conditions; outputting a system health state evaluation vector through the space-time diagram convolutional neural network model, and calculating a system fault probability index value; and outputting maintenance early warning information corresponding to the target energy storage container based on the system fault probability index value. The problem that the fault diagnosis accuracy of the energy storage container is greatly reduced due to the fact that the incidence relation between multi-source weak anomalies cannot be recognized in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of intelligent models, and in particular to a method and device for early warning of energy storage container failures based on real-time monitoring. Background Technology

[0002] With the large-scale application of energy storage technology in scenarios such as renewable energy consumption, grid peak and frequency regulation, and emergency backup power, energy storage containers, as integrated and modular energy storage units, have become a key infrastructure of modern energy systems. They typically integrate battery packs, thermal management systems, power conversion devices, and structural protection units, undertaking multiple functions of energy storage, conversion, and safety isolation. However, energy storage systems face multiple risks during long-term operation, including cell aging, thermal runaway, insulation failure, and loose connections. There is an urgent need for high-precision, high-timeliness state perception and risk prediction mechanisms to ensure the safe and stable operation of the system throughout its entire lifecycle. Therefore, fault diagnosis of energy storage containers has become an important research direction.

[0003] In existing technologies, although some energy storage containers are equipped with basic temperature and voltage monitoring modules, they generally lack the ability to collaboratively perceive the coupling characteristics of multiple physical fields. When complex operating conditions overlap, the operating status changes rapidly, or abnormal signs are in the early stages of evolution, they are prone to misjudging complex anomalies caused by the combined effects of electrical, thermodynamic, chemical, and mechanical factors as normal fluctuations of a single physical quantity, or failing to identify the correlation between multiple weak anomalies, thus significantly reducing the accuracy of fault diagnosis for energy storage containers.

[0004] Therefore, there is an urgent need for a method and device for early warning of energy storage container failures based on real-time monitoring. Summary of the Invention

[0005] This application provides a method and device for early warning of energy storage container faults based on real-time monitoring, which solves the problem that the existing technology cannot identify the correlation between multiple weak anomalies, thus causing a significant reduction in the accuracy of fault diagnosis of energy storage containers.

[0006] The first aspect of this application provides a method for early warning of energy storage container faults based on real-time monitoring. The method includes: preprocessing multimodal operating state data corresponding to the target energy storage container to construct a standardized multidimensional time-series dataset; the multimodal operating state data includes electrical state data, thermodynamic state data, chemical environment data, and mechanical vibration data; constructing a multi-physics coupled energy storage container topology model, and constructing a state estimation digital twin based on the multidimensional time-series dataset and the energy storage container topology model; defining a system correlation graph structure based on the multi-physics coupled energy storage container topology model, and constructing a state reference representation based on the state estimation digital twin; constructing a spatiotemporal graph convolutional neural network model using the system correlation graph structure as a spatial constraint for spatiotemporal feature propagation and the state reference representation as a health state mapping constraint; outputting a system health state evaluation vector through the spatiotemporal graph convolutional neural network model; calculating a system fault probability index value based on the health state evaluation vector, and outputting maintenance early warning information corresponding to the target energy storage container based on the system fault probability index value.

[0007] Optionally, preprocessing operations are performed on the multimodal operating status data corresponding to the target energy storage container to construct a standardized multidimensional time-series dataset. Specifically, this includes: acquiring multimodal operating status data based on a multimodal sensor system, which includes a voltage sensor, a current sensor, a distributed fiber optic temperature sensor, an infrared thermal imager, an electrochemical state sensing unit, and a microelectromechanical system accelerometer; performing preprocessing operations on the multimodal operating status data and constructing a standardized multidimensional time-series dataset; the preprocessing operations include timestamp alignment, data denoising, normalization, and missing value imputation.

[0008] Optionally, a multi-physics coupled energy storage container topology model is constructed, specifically including: based on electrical state data and chemical environment data, a single-particle model is used to describe the diffusion and embedding process of lithium ions in positive and negative electrode particles to construct an electrochemical model; based on thermodynamic state data, the Navier-Stokes equation and energy conservation equation are used to simulate the flow and heat transfer of coolant in the pipeline to construct a thermal management model; based on mechanical vibration data, the Fourier heat conduction equation is used to describe the heat exchange process between the external environment and internal equipment to construct a shell heat conduction model; a multi-physics coupled energy storage container topology model is constructed based on the electrochemical model, thermal management model, and shell heat conduction model; an extended Kalman filter algorithm is used, and a standardized multi-dimensional time-series dataset is used as the observation set to iteratively update the state vectors in the digital twin that are not directly measured; the state vectors include the state of charge, health state, remaining effective lifetime, internal temperature, lithium plating degree, and coolant flow rate of each battery cell.

[0009] Optionally, the extended Kalman filter algorithm includes a prediction phase and an update phase, iteratively updating the state vector in the digital twin that is not directly measured. Specifically, in the prediction phase, the predicted observations and the covariance matrix corresponding to the state vector are predicted using the energy storage container topology model; in the update phase, the predicted observations are compared with the actual observations in the multidimensional time series dataset to calculate the Kalman increment; and the state vector and the covariance matrix corresponding to the state vector are iteratively corrected based on the Kalman increment.

[0010] Optionally, a system correlation graph structure is defined based on the multi-physics coupling energy storage container topology model, and a state reference representation is constructed based on the state estimation digital twin. Specifically, this includes: defining a system correlation graph structure based on the multi-physics coupling energy storage container topology model; the system correlation graph structure includes nodes and edges, where nodes represent battery cells, battery modules, and key components, and edges represent physical connections, electrical coupling paths, and thermal conduction paths between nodes, with edge weights quantified based on the Euclidean distance, electrical impedance, or thermal resistance between components; and constructing a state reference representation based on the state estimation digital twin, which represents the state reference characteristics of the target energy storage container under reference operating conditions. The state reference characteristics include a baseline health state, which is determined by the steady-state distribution established by the state estimation digital twin operating under fault-free historical data.

[0011] Optionally, the system failure probability index value is calculated based on the health status assessment vector, and maintenance warning information corresponding to the target energy storage container is output based on the system failure probability index value. Specifically, this includes: performing deviation analysis between the system health status assessment vector and the baseline health status using Mahalanobis distance metric, and calculating the system failure probability index value; if it is confirmed that the system failure probability index value is greater than a first preset threshold and less than or equal to a second preset threshold, then the maintenance warning information is output as a Level 1 maintenance warning information; the Level 1 maintenance warning information is used to prompt maintenance personnel to check relevant components through a human-machine interface and record abnormal event logs; if it is confirmed that the system failure probability index value is greater than a second preset threshold and less than or equal to a third preset threshold, then the maintenance warning information is output as a Level 2 maintenance warning information; the Level 2 maintenance warning information is used to send a power derating command to the energy conversion system of the target energy storage container and a forced cooling command to the thermal management system of the target energy storage container; if it is confirmed that the system failure probability index value is greater than a third preset threshold, then the maintenance warning information is output as a Level 3 maintenance warning information; and is used to electrically isolate the faulty unit in the target energy storage container and initiate a directional inert gas fire extinguishing procedure.

[0012] Optionally, after performing deviation analysis between the system health status assessment vector and the baseline health status using Mahalanobis distance metric and calculating the system failure probability index value, the method further includes: based on the contribution values ​​of each feature dimension in the health status assessment vector, outputting a fault diagnosis report corresponding to the target energy storage container, including the fault cause, fault propagation path, and fault repair measures.

[0013] A second aspect of this application provides a fault early warning device for energy storage containers based on real-time monitoring. The device includes an acquisition module and a processing module, wherein... The acquisition module is used to preprocess the multimodal operating status data corresponding to the target energy storage container to construct a standardized multidimensional time-series dataset; the multimodal operating status data includes electrical status data, thermodynamic status data, chemical environment data, and mechanical vibration data.

[0014] The processing module is used to construct a multi-physics coupled energy storage container topology model, and build a state estimation digital twin based on a multi-dimensional time-series dataset and the energy storage container topology model; define the system correlation graph structure based on the multi-physics coupled energy storage container topology model, and construct a state reference representation based on the state estimation digital twin; construct a spatiotemporal graph convolutional neural network model using the system correlation graph structure as the spatial constraint for spatiotemporal feature propagation and the state reference representation as the health state mapping constraint; output the system health state evaluation vector through the spatiotemporal graph convolutional neural network model; calculate the system failure probability index value based on the health state evaluation vector, and output the maintenance warning information corresponding to the target energy storage container based on the system failure probability index value.

[0015] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform any of the methods described above.

[0016] A fourth aspect of this application provides a non-transitory computer-readable storage medium storing a computer program, the computer program being executed by a processor using any of the methods described above.

[0017] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. Construct a standardized multi-dimensional time-series dataset based on multimodal operational status data; construct a state estimation digital twin based on the multi-dimensional time-series dataset and the energy storage container topology model; define the system correlation graph structure based on the energy storage container topology model, and construct a state reference representation based on the state estimation digital twin; construct a spatiotemporal graph convolutional neural network model with the system correlation graph structure and the state reference representation as constraints; output the system health status assessment vector and calculate the system failure probability index value through the spatiotemporal graph convolutional neural network model; output the maintenance early warning information corresponding to the target energy storage container based on the system failure probability index value, thereby realizing multi-physics collaborative perception and dynamic consistency assessment of the target energy storage container's operational status, transforming fault identification from a passive response based on a single threshold to an active early warning based on state evolution trends and probability indicators, revealing potential fault risks in advance and providing quantitative basis for operation and maintenance decisions, reducing the probability of sudden failures and improving the safety and reliability of the energy storage container's entire life cycle operation.

[0018] 2. In the prediction phase, the predicted observations and the covariance matrix corresponding to the state vectors in the multidimensional time-series dataset are predicted using the energy storage container topology model. In the update phase, the predicted observations are compared with the actual observations in the multidimensional time-series dataset to calculate the Kalman increment. Based on the Kalman increment, the state vectors and the covariance matrix corresponding to the state vectors are iteratively corrected. Thus, under the constraints of the multiphysics coupling mechanism and the combined effect of multimodal observation data, a dynamic consistency estimate of the unmeasured operating state of the target energy storage container is achieved. This allows the state vectors and their uncertainties to gradually converge to a range that matches the actual operating state over time, providing a stable and reliable state basis for subsequent health status assessment, fault probability calculation, and fault diagnosis.

[0019] 3. Based on the contribution values ​​of each feature dimension in the health status assessment vector, output the fault diagnosis report corresponding to the target energy storage container, including the fault cause, fault propagation path and fault repair measures. This transforms the system-level health assessment results into interpretable fault location and handling basis, enabling the fault diagnosis process to clearly identify the source of the anomaly, the propagation mode and the corresponding handling direction, improve the accuracy and pertinence of fault analysis, and provide direct support for operation and maintenance personnel to carry out targeted maintenance and risk control. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a method for early warning of energy storage container failures based on real-time monitoring, provided in an embodiment of this application. Figure 2 This is a schematic diagram of the core principle framework of a spatiotemporal graph convolutional neural network model provided in an embodiment of this application; Figure 3This is a schematic diagram of a module of an energy storage container fault early warning device based on real-time monitoring, provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0021] Explanation of reference numerals in the attached figures: 31, acquisition module; 32, processing module; 401, processor; 402, communication bus; 403, user interface; 404, network interface; 405, memory. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0023] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0024] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0025] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0026] Please refer to Figure 1 The diagram illustrates a flowchart of a real-time monitoring-based early warning method for energy storage containers provided in this application embodiment. The flowchart mainly includes the following steps: S101 to S106.

[0027] Step S101: Preprocess the multimodal operating status data corresponding to the target energy storage container to construct a standardized multidimensional time series dataset.

[0028] Specifically, the method of the technical solution in this application first deploys a multimodal sensor network inside the target energy storage container and on its key components to collect multimodal operating status data of the energy storage container in real time and synchronously. This multimodal operating status data includes electrical status data, thermodynamic status data, chemical environment data, and mechanical vibration data. The target energy storage container refers to the specific energy storage container entity selected as the object of operating status monitoring, status estimation, health assessment, and fault diagnosis during the execution of the method in this application. It serves as an encapsulation carrier for the integrated deployment of battery cells, battery modules, and related key components, used to realize the storage, conversion, and safe operation of electrical energy.

[0029] After collecting multimodal operational status data, data preprocessing steps are performed. Data preprocessing includes timestamp alignment, data denoising, normalization, and imputation of missing values, ultimately generating a standardized multidimensional time-series dataset.

[0030] In one possible implementation, step S101 further includes: acquiring multimodal operating state data based on a multimodal sensor system, the multimodal sensor system including a voltage sensor, a current sensor, a distributed fiber optic temperature sensor, an infrared thermal imager, an electrochemical state sensing unit, and a microelectromechanical system accelerometer; performing preprocessing operations on the multimodal operating state data and constructing a standardized multidimensional time-series dataset; the preprocessing operations include timestamp alignment, data denoising, normalization, and missing value imputation.

[0031] Specifically, electrical state data is acquired through voltage sensors, current sensors, and an online electrochemical impedance spectroscopy (EIS) monitor deployed on each battery cell or module. Voltage sensors acquire cell voltage values ​​at a sampling frequency of at least one hundred times per second. Current sensors monitor the current intensity in the charge / discharge circuit in real time using the Hall effect principle. The EIS monitor applies a sinusoidal perturbation signal with an amplitude of ten millivolts and a frequency range of 0.1 Hz to 1 kHz when the system is in a quiescent or low-power operating state, measuring the battery's impedance response at different frequencies to obtain frequency domain characteristics reflecting the electrode interface state and electrolyte conductivity. Thermodynamic state data is acquired jointly by a distributed fiber Bragg grating temperature sensor array laid along the battery cluster rack and cooling pipes, and an infrared thermal imager installed on the top of the container. The distributed fiber optic temperature sensor array has a spatial resolution of 5 cm and a temperature measurement accuracy of ±0.5 degrees Celsius, enabling continuous acquisition of the longitudinal and lateral temperature distribution within the battery stack. An infrared thermal imager scans the upper surface area inside the container at a rate of 30 frames per second, generating a two-dimensional thermal image with a resolution of at least 640 x 480 pixels. The data from both arrays are fused to form a high-density three-dimensional temperature field distribution map. Chemical environmental data is collected through an array of hydrogen, carbon monoxide, and photoionization gas sensors placed in the gaps between the battery stack and at the exhaust vents. The hydrogen sensor utilizes the catalytic combustion principle, with a detection range of 0 to 4% volume concentration and a response time of less than 10 seconds. The carbon monoxide sensor is based on electrochemical principles, with a range of 0 to 500 ppm. The photoionization gas sensor detects volatile organic compounds, with a detection limit of up to 10 ppb. These three sensors work together to monitor the concentrations of characteristic gases released during electrolyte leakage, the initial decomposition products of thermal runaway, and the aging of insulating materials. Mechanical vibration data is collected by microelectromechanical system accelerometers fixed on rotating and switching components such as converters, fans and main contactors. The sampling frequency is set to 5 kHz to capture abnormal vibration modes during equipment operation, such as mechanical fault characteristics like bearing wear, contactor bounce or fan imbalance.

[0032] All sensor nodes are connected via industrial Ethernet and global clock synchronization is performed using a precise time protocol to ensure consistent time bases for all data streams, with time synchronization errors controlled within microseconds. Timestamp alignment is achieved by resampling all raw data according to a unified time grid, with the time grid interval set to 20 milliseconds. Data denoising employs a wavelet thresholding algorithm, using the Daubechies fourth-order wavelet basis function to perform a five-level decomposition of the original signal. Soft thresholding is applied to high-frequency detail coefficients, with the threshold adaptively determined based on noise estimation to preserve abrupt changes in the signal while suppressing high-frequency noise. Normalization maps data of different dimensions to a zero-to-one interval. Electrical state data is normalized using maximum and minimum values, thermodynamic state data is linearly scaled based on historical operating extreme values, and chemical environment data and mechanical vibration data are normalized using Z-score before interval mapping. Missing value imputation employs a collaborative filtering algorithm based on matrix factorization, treating the multidimensional time series dataset as a user-item rating matrix, where "user" corresponds to the time step and "item" corresponds to the sensor channel. Low-rank latent factors are extracted through singular value decomposition to reconstruct missing data points, and the imputation error is controlled within five percent of the original signal standard deviation.

[0033] The calculation methods for the preprocessing steps of various types of data are as follows: Electrical condition data: in, It is voltage characteristic standardization. It's voltage. It is the nominal voltage. This refers to the rated voltage range. Thermodynamic state data:

[0034] in, It is temperature characteristic standardization. It's temperature. It is the ambient reference temperature. This is the upper limit of the permissible temperature rise. Chemical environmental data (hydrogen) carbon monoxide Volatile organic compounds )

[0035] in, It's the hydrogen concentration. It is the carbon monoxide concentration. This refers to the concentration of volatile organic compounds. Mechanical vibration data:

[0036] in, It is the amplitude of the acoustic signal. It is the sound pressure level (SPL), dB (decibels).

[0037] in, It is the sound power level (SWL). It is a directional factor. This is the distance from the sound source to the center of gravity of the container. Learnable weight matrix:

[0038] in, For learnable weight matrix, , , , This is the learning rate.

[0039] Step S102: Construct a multi-physics coupled energy storage container topology model, and build a state estimation digital twin based on the multi-dimensional time series dataset and the energy storage container topology model.

[0040] Specifically, a dynamically updated digital twin for state estimation is constructed based on a standardized multidimensional time-series dataset and a pre-defined physical topology model of an energy storage container. The physical topology model is a multiphysics coupled model, which includes the electrochemical model of the battery system, the fluid dynamics model of the thermal management system, and the heat conduction model of the container shell.

[0041] In one possible implementation, step S102 further includes: constructing an electrochemical model by using a single-particle model to describe the diffusion and embedding process of lithium ions in the positive and negative electrode particles based on electrical state data and chemical environment data; constructing a thermal management model by using the Navier-Stokes equation and energy conservation equation to simulate the flow and heat transfer of coolant in the pipeline based on thermodynamic state data; constructing a shell thermal conduction model by using the Fourier heat conduction equation to describe the heat exchange process between the external environment and the internal equipment based on mechanical vibration data; constructing a multi-physics coupled energy storage container topology model based on the electrochemical model, thermal management model, and shell thermal conduction model; and employing an extended Kalman filter algorithm, and Using a standardized multidimensional time-series dataset as the observation set, the state vectors in the digital twin that are not directly measured are iteratively updated. The extended Kalman filter algorithm includes a prediction phase and an update phase. In the prediction phase, the predicted observations corresponding to the multidimensional time-series dataset and the covariance matrix corresponding to the state vectors are predicted using the energy storage container topology model. In the update phase, the predicted observations are compared with the actual observations in the multidimensional time-series dataset to calculate the Kalman increment. The state vectors and the covariance matrix corresponding to the state vectors are iteratively corrected based on the Kalman increment. The state vectors include the state of charge, state of health, remaining effective lifetime, internal temperature, degree of lithium plating, and coolant flow rate of each battery cell.

[0042] Specifically, regarding the construction of the electrochemical model, the single-particle model abstracts the positive and negative electrode particles of each battery cell as equivalent spherical solid-phase diffusers, respectively. The radial diffusion of solid-phase lithium concentration is used to characterize the insertion and deintercalation kinetics, enabling a recursively consistent relationship between the external observables reflected by electrical state data and chemical environment data and the internal reaction state. Under the condition of a uniform time grid interval of twenty milliseconds after discretization in the preprocessing step, the voltage characteristics of each time step are standardized. Standardized quantity of current characteristics and gas characteristic standardized quantities , , The organization provides electrochemical correlation features on the observation side, thus providing observations that are correlated one-to-one with the model state for subsequent extended Kalman filtering; the state evolution of solid-phase diffusion in the single-particle model can be characterized by the following equation: in, Indicates at time Radial position of particles Solid-phase lithium concentration at the location, The value ranges from the particle center to the particle surface. The solid-phase diffusion coefficient is represented, and its value can be obtained through calibration experiments or online identification. This formula describes the diffusion homogenization process of the concentration inside the particle over time through the physical mechanism of concentration gradient-driven diffusion flux. The diffusion flux on the particle surface is related to the reaction current density, and the boundary conditions can be expressed as follows: in, Indicates particle radius, This represents the surface reaction current density. This represents the Faraday constant; this boundary condition reflects that the lithium flux consumed or generated by the electrochemical reaction is equal to the flux diffused from the solid phase to the surface, enabling current information to drive the update of the internal concentration state. It can be estimated from the terminal current and standardized with the current characteristics. Establish a consistent scale mapping, for example, by preprocessing the scaled data. The equivalent terminal current is reduced by inverse normalization, and then the current term for model driving is obtained by allocating it according to the parallel branches of individual units; state of charge It can be obtained from the average concentration of the solid phase: in, This represents the average solid concentration over the particle volume. and These represent the minimum and maximum achievable solid phase concentration boundaries, respectively, and their values ​​can be determined by battery material parameters or calibration. This formula reflects the state of charge corresponding to the "occupancy ratio of lithium intercalation capacity within the concentration range." The electrochemical composition of the terminal voltage can be expressed by the superposition of the equivalent open-circuit voltage and overpotential, and the pre-processed voltage characteristics are also considered. Establish observational relationships with model voltages: in, This indicates the model's predicted terminal voltage. This represents the functional relationship between open-circuit voltage and state of charge. This represents the equivalent overpotential caused by charge transfer and diffusion, Indicates the internal resistance of the ohm. This represents the terminal current; this formula reflects the mechanism that the terminal voltage is calculated by subtracting irreversible losses from the reversible potential; on the observation side, it can be... Through the same nominal voltage With rated voltage range Standardize it to make it compatible with preprocessing. Comparisons are made on the same scale to ensure consistency in residual calculations during the Kalman update phase.

[0043] For the construction of the thermal management model, the thermodynamic state data is denoised and normalized to form standardized temperature characteristic quantities. Furthermore, it couples with the flow and heat transfer process of the cooling circuit, enabling the digital twin to synchronously estimate the relationship between "coolant flow rate, cooling capacity, heat source release intensity, and temperature rise accumulation" at a unified time step of twenty milliseconds; the flow of coolant in the pipeline can be described by the incompressible Navier-Stokes equations: in, Indicates the density of the coolant. This represents the velocity vector, whose components take values ​​in the pipeline coordinate system. Represents a pressure field. Indicates dynamic viscosity. This represents the body force term or equivalent drag term; this equation reflects that "inertia, pressure, viscosity, and external forces jointly determine the flow velocity evolution," thus providing a dynamic constraint on the coolant flow velocity in the state vector; heat transfer and temperature field can be described by the energy conservation equation: in, Indicates specific heat capacity. Represents the temperature field in a fluid or solid region. Indicates thermal conductivity. The term represents the heat source; this equation reflects how convection and heat conduction together determine temperature evolution, with the release of heat source added as a superposition. This can be correlated with the Joule heat and heat of reaction on the electrochemical side, for example, let It changes with the terminal current and internal resistance, thus affecting the pre-processed... Its inverse normalization is used for heat source modeling; in order to integrate with preprocessed temperature features Consistency can be achieved by adjusting the model temperature. Using the same ambient reference temperature With the upper limit of allowable temperature rise Scale the data so that the predicted and actual observations of the digital twin are calculated in the same standardized domain.

[0044] For the construction of the shell heat conduction model, the heat exchange between the shell and internal equipment can be described by the Fourier heat conduction equation and coupled with the thermal management model at the boundary conditions. This model is used to constrain the consistency of "shell temperature distribution, external heat dissipation capacity, and internal temperature rise accumulation" under external environmental disturbances, so that thermal imaging or distributed temperature observations can drive the update of shell-related states. Its basic expression is as follows: in, This indicates the temperature field of the shell or structural component. Indicates thermal diffusivity, This represents the equivalent heat source term within the structure. and Let represent the density and specific heat capacity of the structural material, respectively; this formula reflects that "temperature propagates within the structure via diffusion and is driven by the heat source"; the heat exchange between the shell and the environment and the internal cavity can be expressed using convective boundary conditions. in, Indicates the thermal conductivity of the shell. This represents the temperature gradient along the outward normal direction. Indicates the convective heat transfer coefficient. This formula represents the ambient temperature; it reflects the principle that "the thermal flux through conduction on the shell surface is equal to the convective heat transfer flux," thus establishing the ambient reference temperature. Boundary constraints were incorporated in accordance with environmental observations; simultaneously, the sound power level obtained from the preprocessed mechanical vibration data was... It can be used as a proxy for the operating load of key components such as fans and converters to correct the convective heat transfer coefficient. or equivalent heat source item This reflects the "changes in heat exchange and heat source release under enhanced vibration and noise conditions," thereby enabling cross-domain correlation of preprocessing parameters in a multi-physics model.

[0045] For the state estimation part of the extended Kalman filter, the electrochemical model, thermal management model, and shell thermal conduction model are combined to form a multi-physics coupled energy storage container topology model. State vectors and observation vectors are defined on a unified time grid, and model prediction and observation correction are used to achieve recursive consistency of the digital twin. The state vector can cover non-directly measured quantities such as the state of charge, state of health, remaining effective life, internal temperature, degree of lithium plating, and coolant flow rate of each battery cell. Among these, the state of health, remaining effective life, and degree of lithium plating can be represented by parameterized degradation sub-models or slow variables related to impedance characteristics, and corrected through observation residuals in long-term recursion. In the prediction stage, state recursion and covariance propagation can be written as: in, Indicates the first The predicted state vector at each time step. This represents the updated state vector from the previous time step. This represents a selectable known control variable or operating condition variable, such as a charge / discharge command, fan speed setting, or pump speed setting; (function) The discretized multiphysics coupling model described above is used to deduce the state evolution through a combination mechanism of electrochemical diffusion, flow heat transfer, and structural heat transfer; the covariance propagation is as follows: in, Represents the predicted state covariance matrix. This represents the covariance matrix after the previous time step update. express The Jacobian matrix of the state is used to characterize the propagation of small perturbations in the state during prediction. The equation represents the process noise covariance, used to characterize model incompleteness and unmodeled disturbances; it reflects the propagation of uncertainty with the linearization of dynamics and the superposition of process disturbances; the observation equation maps the preprocessed multidimensional time series dataset into observation vectors. And associate it with the predicted observations corresponding to the predicted state: in, Indicates the first The actual observation vector at each time step can be derived from... , , , , , and current normalization quantity Composed of, etc. This represents the mapping relationship from state to observation, for example, ... Mapping internal resistance to a standardized voltage, mapping internal temperature and heat transfer state to a standardized temperature, and mapping slow degradation variables to gas or impedance-related features; to fuse observations from different modes using a unified strategy, the learnable weight matrix you provided can be used to weight the observation sub-vectors, making them consistent with the node features on the graph convolution side. The weight organization can be written as: in, Indicates the first layer or first The fusion weight structure of the stages, arrive This represents the learnable fusion coefficients for each modality, and their values ​​are typically limited to a range of values. arrive The parameters satisfy a normalization constraint, which reflects that the relative contributions of different modalities to the current health representation can be adaptively adjusted during training. The terms in parentheses are preprocessed standardized feature combinations; the normalization constraint is: in, arrive The value can be obtained through end-to-end training or online adaptive update to avoid unbounded weight growth and ensure consistency in the fusion interpretation. During the update phase, the innovation residual is first calculated and a Kalman increment is formed to correct the state. The innovation residual is: in, Representing the observation residuals, reflecting the deviation between actual observations and model predictions; the residual covariance is: in, express The Jacobian matrix of the state is used to characterize the sensitivity of state perturbations to observations. The observation noise covariance is used to characterize sensor noise and preprocessing residual error; the Kalman increment is: in, The Kalman increment matrix is ​​used to form an adaptive trade-off between trust model predictions and trust observation information; the state correction is as follows: in, This represents the updated state vector; the covariance is corrected as follows: in, Let represent the updated covariance matrix. The identity matrix is ​​represented by the above update mechanism, which achieves iterative convergence of the unmeasured state through residual-driven iteration, and enables the corrected state estimate to serve as the basis for subsequent health mapping constraints and baseline health state construction.

[0046] By controlling the time synchronization error to the microsecond level and using a 20-millisecond resampling time grid, the filter step size can be determined together. And observation alignment strategy, thereby Organized on the same time reference; the five-level decomposition and soft thresholding of wavelet thresholding reduce the impact of high-frequency noise. The contribution of this allows the observation noise covariance to be set according to the "residual noise level after denoising"; maximum and minimum normalization, linear scaling, and Z-score standardization followed by interval mapping ensure... , , and It has comparable scales during fusion, making arrive The learning process is not dominated by magnitude differences; matrix factorization interpolation reconstructs missing points through low-rank latent factors, ensuring the continuity of the observation vector within the time window, thereby avoiding residual distortion caused by missing measurements during the Kalman update stage. Simultaneously, the upper limit constraint on interpolation error can be used to... Provide a basis for setting prior upper bounds; if the denoising process needs to be explicitly written into a reproducible computational expression, the soft threshold operator can be written as: in, These represent the high-frequency detail coefficients obtained from wavelet decomposition. This represents the detail coefficients after thresholding. The threshold value can be adaptively determined by noise estimation, reflecting the processing mechanism of "preserving abrupt changes while suppressing small noise"; the low-rank reconstruction of the singular value decomposition used for missing value imputation can be written as: in, This represents a matrix representation of a multidimensional time-series dataset organized according to "time step - sensor channel". and Represents a left and right singular vector matrix. Representing a singular value diagonal matrix, a low-rank approximation is achieved by preserving the principal singular values, thereby reconstructing the missing elements; the normalized features of the preprocessed output and the continuous observations after interpolation together determine... The composition of the components also determines the fusion weight. arrive The learnable space and the stability of the filter update enable consistent coupling of electrochemistry, thermal management, shell heat transfer and state estimation under the same data scale, the same time step and the same noise assumption.

[0047] Step S103: Define the system association graph structure based on the energy storage container topology model of multi-physics coupling, and construct a state reference representation based on the state estimation digital twin.

[0048] Specifically, by abstracting the topology model of the energy storage container with multi-physics coupling, the physical connection, electrical coupling, and thermal conduction relationships between individual battery cells, battery modules, and key components within the energy storage container are uniformly mapped into an association graph structure to form spatial association constraints that can characterize the coupling characteristics of multi-physics fields. At the same time, based on the state estimation results output by the state estimation digital twin during continuous operation, a state reference representation is constructed to characterize the state distribution characteristics of the target energy storage container under reference operating conditions, so that the subsequent spatiotemporal feature modeling and health status assessment processes can be carried out under the combined effect of physical structural constraints and state benchmark constraints.

[0049] In one possible implementation, step S103 further includes: defining a system correlation graph structure based on a multi-physics coupling energy storage container topology model; the system correlation graph structure includes nodes and edges, where nodes represent battery cells, battery modules, and key components, and edges represent physical connections, electrical coupling paths, and thermal conduction paths between nodes, with the weights of the edges quantified based on the Euclidean distance, electrical impedance, or thermal resistance between components; and constructing a state reference representation based on a state estimation digital twin, where the state reference representation is used to represent the state reference characteristics of the target energy storage container under reference operating conditions, and the state reference characteristics include a baseline health state, which is determined by the steady-state distribution established by the state estimation digital twin operating under fault-free historical data.

[0050] Specifically, a battery cell refers to the smallest electrochemical energy unit in the target energy storage container capable of independently completing electrochemical energy storage and discharge. During operation, it can generate cell voltage, branch current distribution, cell temperature, and characteristic quantities related to the electrochemical state. A battery module refers to a battery functional integration unit formed by combining multiple battery cells in series, parallel, or series-parallel configurations and through connectors, busbars, and housing structures. During operation, it exhibits group characteristics such as module-level voltage consistency, thermal distribution consistency, and electrical connection integrity. Key components refer to equipment units in the target energy storage container that, in addition to battery cells and battery modules, directly affect energy conversion, thermal management, safety isolation, and operational reliability. These include at least converters, DC contactors, fuses / circuit breakers, coolant pumps, fans, cooling pipe heat exchange components, and actuators related to gas monitoring and fire-fighting linkage. These units are represented by nodes in the system association diagram structure to ensure that subsequent spatiotemporal feature propagation can revolve around the actual physical composition and coupling relationships of the target energy storage container.

[0051] Physical connections refer to the geometric adjacency or assembly relationships formed by battery cells, battery modules, and key components in terms of structural installation, wiring harness / pipeline layout, and the constraints of mounting brackets and housings. They are used to characterize the spatial proximity conditions for cross-domain links such as "mechanical loosening—poor contact—localized heating—increased vibration" in fault propagation. Electrical coupling paths refer to the equivalent electrical connection paths between the series and parallel topologies of battery cells and battery modules, the connections between busbars and wiring harnesses, the conduction links between contactors and protection devices, and the DC-side busbar of the converter to the battery-side circuit. These paths are used to characterize current distribution and voltage drop. The impact of accumulated and local internal resistance anomalies on the voltage and current characteristics of adjacent nodes is transmitted; the heat conduction path refers to the heat transfer channel formed by the contact heat exchange between battery cells and battery modules, the conduction and convection heat exchange between modules and cooling plates / cooling pipes, the conduction and heat dissipation between key components and the shell, and the convection heat transfer boundary between the shell and the external environment. It is used to characterize the propagation relationship of thermal risks caused by temperature rise diffusion, hot spot migration and cooling failure on the topology. The above three types of relationships together constitute the semantic type of the edges in the system association graph structure, so that the edges not only express whether they are connected, but also express the mechanism by which they influence each other.

[0052] In the system's relational graph structure, edge weights are used to quantify the strength of the association represented by the edges. Edge weights can be calculated separately for different relation types and then uniformly scaled to reflect the propagation mechanism where stronger coupling neighboring nodes contribute more significantly during graph convolutional aggregation. When quantization is based on Euclidean distance between components, Euclidean distance reflects the degree of physical proximity. The formula for calculating Euclidean distance is: in, Represents a node With nodes Euclidean distance, Represents a node Three-dimensional position coordinates in the internal coordinate system of the target energy storage container Represents a node The three-dimensional position coordinates in the same coordinate system; the Euclidean distance is obtained through structural design coordinates, installation positioning coordinates, or three-dimensional modeling parameters; when forming edge weights for physical connections based on Euclidean distance, the mechanism of stronger coupling with closer distances can be solidified into a weight mapping through a distance decay function. The formula for calculating the distance weight is: in, This represents the edge weight components obtained from the Euclidean distance. The distance attenuation coefficient is used to adjust the attenuation rate. Its value can be set according to the statistical distribution of the spacing between typical devices inside the target energy storage container, so that the weight of adjacent installed units remains high while the weight of distant units decreases rapidly. This mapping reflects the propagation mechanism in which nearby units are more likely to form the superposition of structural disturbance transmission and local thermal effects.

[0053] When quantified based on electrical impedance, electrical impedance reflects the conductivity and energy loss of an electrical coupling path. Electrical impedance can be characterized using DC equivalent impedance or AC equivalent impedance. Under DC or quasi-steady-state conditions, the equivalent impedance can be formed by the ratio of voltage drop to current. The formula for calculating electrical impedance is: in, Represents a node To the node The equivalent impedance along the electrical coupling path between them. This represents the equivalent voltage drop along the path. This represents the equivalent current along the path; It can be obtained from voltage observations or model estimation of the corresponding circuit. The impedance can be obtained by observing current sensors and combining it with the loop topology; the larger the impedance, the weaker the conduction and the higher the risk of heat generation. Therefore, the impedance can be mapped to a weighted component with a larger weight for smaller impedances. The formula for calculating the impedance weight is: in, This represents the edge weight component obtained from the electrical impedance. This represents the impedance scaling factor, used to map different circuit impedance levels to a comparable range. Its value can be determined based on historical operating data or the statistical mean of impedance under fault-free conditions. This mapping reflects the more synchronized coupling mechanism of nodes with tighter electrical connections in terms of voltage and current disturbances, while suppressing the contribution of high-resistance paths to highlight the changes in the coupling structure caused by potential contact defects, bus aging, and other anomalies.

[0054] When quantified based on thermal resistance, thermal resistance reflects the heat transfer capability of the heat conduction path. A lower thermal resistance indicates stronger heat transfer, and hot spots are more easily diffused to adjacent nodes. Thermal resistance can be determined based on the equivalent thickness of the heat-conducting layer, the heat transfer area, and the material's thermal conductivity. The formula for calculating thermal resistance is: in, Represents a node With nodes The equivalent thermal resistance along the heat conduction path between them. This represents the equivalent length or equivalent thickness of the effective heat transfer path between two nodes, and its value is determined by the structural assembly and the configuration of the dielectric layer. This represents the effective thermal conductivity, and its value is determined by the parameters of the contact interface material, thermal pad, or dielectric material. This represents the effective heat transfer area, the value of which is determined by the contact area, the cooling plate coverage area, or the effective area of ​​the heat exchanger. When the edge weight of the heat conduction path is formed based on thermal resistance, an inverse proportional mapping isomorphic to the impedance can be used. The formula for calculating the thermal resistance weight is: in, This represents the edge weight component obtained from thermal resistance. This represents the thermal resistance scaling factor, used to map the thermal resistance levels of different heat transfer channels to a comparable range. Its value can be determined by steady-state fitting of the temperature field under fault-free operating conditions or by calibration of design parameters. This mapping reflects the mechanism that "the smoother the heat transfer channel, the stronger the propagation of temperature disturbances between nodes," thus enabling the graphical structure to express the impact of cooling failure and thermal interface degradation on the heat propagation channel.

[0055] To unify edge weights obtained from different quantification criteria within the same system's association graph structure, one can... , and Normalization and fusion are performed to make the edge weights comparable in the numerical domain and stable for use in graph convolution. The formula for calculating the fused edge weights is as follows: in, Represents nodes in the system association graph structure With nodes The final edge weight between them , , These represent the fusion coefficients for the weights of the edges in the physical connection relationship, the electrical coupling path, and the heat conduction path, respectively, and their values ​​range from... arrive The normalization constraint can be satisfied between them to avoid imbalance of fusion weights; the fusion coefficient can be determined by design experience, historical data fitting or learnable parameters in the training phase, so as to adaptively highlight the main propagation channel under different working conditions; the fusion mechanism reflects that the coupling strength of the same pair of nodes is jointly determined by multiple physical channels, and enables the edge weight to simultaneously carry three types of coupling factors: spatial proximity, electrical synchronization and thermal diffusion.

[0056] Step S104: Using the system correlation graph structure as the spatial constraint for spatiotemporal feature propagation and the state reference representation as the health state mapping constraint, a spatiotemporal graph convolutional neural network model is constructed.

[0057] For details, please refer to Figure 2The document presents a schematic diagram illustrating the core principle framework of a spatiotemporal graph convolutional neural network model provided in this application embodiment. The spatiotemporal graph convolutional neural network model is composed of multiple stacked spatiotemporal convolutional layers, each containing a graph convolutional network unit and a gated recurrent unit. The graph convolutional network unit aggregates the spatial feature information of neighboring nodes at each time step, mathematically expressed as:

[0058] in, Given an adjacency matrix with self-loops, Its degree matrix, For the first Layer in time step The node feature matrix, For learnable weight matrix, for Activation function. The gated recurrent unit receives the output of the graph convolutional unit and captures the evolution of features over time. Its update mechanism is as follows:

[0059] in, For graph convolution output, In hidden state, and These are the update door and the reset door, respectively.

[0060] Step S105: Output the system health status assessment vector through the spatiotemporal graph convolutional neural network model.

[0061] Specifically, the spatiotemporal graph convolutional neural network model ultimately aggregates the temporal features of all nodes into a fixed-dimensional system health status assessment vector through a global average pooling layer. Each dimension of this vector corresponds to a potential fault mode or health indicator, such as thermal runaway risk, lithium plating tendency, and contactor aging degree. The training process of the spatiotemporal graph convolutional neural network model includes: acquiring historical multimodal operating status data of the energy storage container under normal operating conditions, various typical fault injection conditions (such as single-cell overcharging, cooling failure, insulation failure, and mechanical loosening), and the entire life cycle aging process, and constructing a training sample set. Each sample contains standardized multidimensional time-series data for ten consecutive minutes and its corresponding real system status label. The labels are divided into four categories: normal, first-level abnormality, second-level alarm, and third-level emergency, determined by experts based on post-fault analysis and accelerated aging experiments. Supervised learning is adopted, and optimization is performed by minimizing the cross-entropy loss function between the system health status assessment vector output by the model and the real label. The loss function is defined as:

[0062] in, For the number of categories, One-hot encoding for the real label. To predict probabilities for the model, the network weights were adjusted using the backpropagation algorithm and the Adam optimizer. The initial learning rate was set to 0.001, and the batch size was 32. Training continued until the validation set loss no longer decreased for ten consecutive epochs.

[0063] Step S106: Calculate the system failure probability index value based on the health status assessment vector, and output the maintenance warning information corresponding to the target energy storage container based on the system failure probability index value.

[0064] Specifically, after obtaining the system health status assessment vector, a deviation analysis is performed between it and the baseline health status of the state estimation digital twin to calculate the system failure probability index value. The baseline health status is determined by the steady-state distribution established by the digital twin operating under fault-free historical data, including the mean vector and covariance matrix of each state parameter.

[0065] In one possible implementation, step S106 further includes: performing deviation analysis on the system health status assessment vector and the baseline health status using Mahalanobis distance metric, and calculating the system failure probability index value; if it is confirmed that the system failure probability index value is greater than a first preset threshold and less than or equal to a second preset threshold, then outputting maintenance warning information as a level one maintenance warning information; the level one maintenance warning information is used to prompt maintenance personnel to check relevant components and record abnormal event logs through a human-machine interface; if it is confirmed that the system failure probability index value is greater than the second preset threshold and less than or equal to a third preset threshold, then outputting maintenance warning information as a level two maintenance warning information; the level two maintenance warning information is used to send a power derating command to the energy conversion system of the target energy storage container and a forced cooling command to the thermal management system of the target energy storage container; if it is confirmed that the system failure probability index value is greater than the third preset threshold, then outputting maintenance warning information as a level three maintenance warning information; and is used to electrically isolate the faulty unit in the target energy storage container and initiate a directional inert gas fire extinguishing procedure.

[0066] Specifically, the deviation analysis uses the Mahalanobis distance metric: in, This is the system health status assessment vector. and These are the baseline mean vector and covariance matrix, respectively. System failure probability index value. Obtained by mapping Mahalanobis distance to probability space:

[0067] in, The scaling parameter is determined by fitting historical fault data to ensure... The failure rate is less than 5% under normal operating conditions and greater than 90% under Level 3 emergency conditions. Based on the system failure probability index value, tiered early warning and proactive safety control responses are executed. When the system failure probability index value exceeds the first preset threshold (which can be set to 30%), a Level 1 maintenance early warning message is generated, prompting maintenance personnel to check relevant components via the human-machine interface and recording abnormal event logs. When the system failure probability index value exceeds the second preset threshold (which can be set to 60%), a Level 2 alarm message is generated, and a power derating command is sent to the energy conversion system of the energy storage container, limiting the maximum charging and discharging power to 70% of the rated value; simultaneously, a forced cooling command is sent to the thermal management system, starting the backup fan and increasing the coolant pump speed to 120% of its maximum value. When the system failure probability index value exceeds the third preset threshold (which can be set to 85%), a Level 3 emergency alarm message is generated, and electrical isolation of the faulty unit and initiation of a directional inert gas fire suppression procedure are performed. Electrical isolation of faulty units is achieved through a fault location algorithm: analyzing the abnormal contribution of each node dimension in the system health status assessment vector, and combining it with the state estimation of the digital twin, the smallest electrical unit that has experienced or is about to experience a fault is accurately located; then, a disconnect command is sent to the DC contactor of the circuit containing that unit via the controller area network bus to cut off its electrical connection with the main circuit. The directional inert gas fire suppression procedure, based on the hotspot areas located by the infrared thermal imager, activates the nitrogen or perfluorohexanone nozzles in the corresponding zones to implement local fire suppression, avoiding resource waste and secondary damage caused by the release of gas throughout the entire compartment.

[0068] In one possible implementation, step S106 further includes: based on the contribution values ​​of each feature dimension in the health status assessment vector, outputting a fault diagnosis report corresponding to the target energy storage container, including the fault cause, fault propagation path, and fault repair measures.

[0069] Specifically, the technical solution of this application also employs a Bayesian network model for causal reasoning. Upon receiving the fault probability index output by the fault diagnosis and probability prediction module, the Bayesian network model can infer the root cause chain leading to the fault based on the contribution of each feature dimension in the system health status assessment vector. The nodes of the Bayesian network include potential causes such as sensor anomalies, cooling failures, overcharging, internal short circuits, and mechanical loosening; the edges represent causal dependencies; and the conditional probability table is obtained through learning from historical fault cases. Finally, the Bayesian network model generates a structured fault diagnosis report, including fault causes, fault propagation paths, and fault repair measures. The fault causes indicate the fundamental anomalies that contribute the most to the fault probability index under causal dependency constraints. The fault propagation paths describe the causal evolution chain of the fundamental anomalies propagating step by step along electrical coupling paths, thermal conduction paths, or physical connections under multi-physics coupling conditions, leading to the current fault state. The fault repair measures, based on the fault causes and fault propagation paths, provide targeted handling suggestions for the least affected unit in the target energy storage container. This transforms fault handling from experience-based judgment based on appearances to precise intervention based on causal reasoning results, thereby improving the certainty of fault location, the pertinence of repair decisions, and the safety and reliability of the overall operation and maintenance process.

[0070] Please refer to Figure 3 The diagram illustrates a module schematic of a real-time monitoring-based energy storage container fault early warning device according to an embodiment of this application. The device includes an acquisition module 31 and a processing module 32, wherein... The acquisition module 31 is used to preprocess the multimodal operating status data corresponding to the target energy storage container to construct a standardized multidimensional time series dataset; the multimodal operating status data includes electrical status data, thermodynamic status data, chemical environment data, and mechanical vibration data.

[0071] Processing module 32 is used to construct a multi-physics coupled energy storage container topology model, and to construct a state estimation digital twin based on a multi-dimensional time-series dataset and the energy storage container topology model; to define a system correlation graph structure based on the multi-physics coupled energy storage container topology model, and to construct a state reference representation based on the state estimation digital twin; to construct a spatiotemporal graph convolutional neural network model using the system correlation graph structure as a spatial constraint for spatiotemporal feature propagation, and the state reference representation as a health state mapping constraint; to output a system health state evaluation vector through the spatiotemporal graph convolutional neural network model; to calculate the system failure probability index value based on the health state evaluation vector, and to output the maintenance warning information corresponding to the target energy storage container based on the system failure probability index value.

[0072] In one possible implementation, the acquisition module 31 is used to preprocess the multimodal operating status data corresponding to the target energy storage container to construct a standardized multidimensional time-series dataset. Specifically, this includes: acquiring multimodal operating status data based on a multimodal sensor system, which includes a voltage sensor, a current sensor, a distributed fiber optic temperature sensor, an infrared thermal imager, an electrochemical state sensing unit, and a microelectromechanical system accelerometer; performing preprocessing operations on the multimodal operating status data and constructing a standardized multidimensional time-series dataset; the preprocessing operations include timestamp alignment, data denoising, normalization, and missing value imputation.

[0073] In one possible implementation, the processing module 32 is used to construct a multi-physics coupled energy storage container topology model, specifically including: constructing an electrochemical model by using a single-particle model to describe the diffusion and embedding process of lithium ions in positive and negative electrode particles based on electrical state data and chemical environment data; constructing a thermal management model by using the Navier-Stokes equation and energy conservation equation to simulate the flow and heat transfer of coolant in the pipeline based on thermodynamic state data; constructing a shell heat conduction model by using the Fourier heat conduction equation to describe the heat exchange process between the external environment and internal equipment based on mechanical vibration data; constructing a multi-physics coupled energy storage container topology model based on the electrochemical model, thermal management model, and shell heat conduction model; and iteratively updating the state vectors in the digital twin that are not directly measured by using an extended Kalman filter algorithm and a standardized multi-dimensional time-series dataset as the observation set; the state vectors include the state of charge, health state, remaining effective lifetime, internal temperature, degree of lithium plating, and coolant flow rate of each battery cell.

[0074] In one possible implementation, the processing module 32 is used to extend the Kalman filter algorithm to include a prediction phase and an update phase, and to iteratively update the state vector in the digital twin that is not directly measured. Specifically, in the prediction phase, the predicted observations and the covariance matrix corresponding to the state vector are predicted using the energy storage container topology model; in the update phase, the predicted observations and the actual observations in the multidimensional time series dataset are compared to calculate the Kalman increment; and the state vector and the covariance matrix corresponding to the state vector are iteratively corrected based on the Kalman increment.

[0075] In one possible implementation, the processing module 32 is used to define a system correlation graph structure based on a multi-physics coupled energy storage container topology model, and to construct a state reference representation based on a state estimation digital twin. Specifically, this includes: defining a system correlation graph structure based on a multi-physics coupled energy storage container topology model; the system correlation graph structure includes nodes and edges, where nodes represent battery cells, battery modules, and key components, and edges represent physical connections, electrical coupling paths, and thermal conduction paths between nodes, with the weights of the edges quantified based on the Euclidean distance, electrical impedance, or thermal resistance between components; and constructing a state reference representation based on a state estimation digital twin, where the state reference representation is used to represent the state reference characteristics of the target energy storage container under reference operating conditions, and the state reference characteristics include a baseline health state, which is determined by the steady-state distribution established by the state estimation digital twin operating under fault-free historical data.

[0076] In one possible implementation, the processing module 32 is used to calculate the system failure probability index value based on the health status assessment vector, and output maintenance warning information corresponding to the target energy storage container based on the system failure probability index value. Specifically, this includes: performing deviation analysis between the system health status assessment vector and the baseline health status using Mahalanobis distance metric, and calculating the system failure probability index value; if it is confirmed that the system failure probability index value is greater than a first preset threshold and less than or equal to a second preset threshold, then outputting maintenance warning information as a level one maintenance warning information; the level one maintenance warning information is used to prompt maintenance personnel to check relevant components through a human-machine interface and record abnormal event logs; if it is confirmed that the system failure probability index value is greater than the second preset threshold and less than or equal to a third preset threshold, then outputting maintenance warning information as a level two maintenance warning information; the level two maintenance warning information is used to send a power derating command to the energy conversion system of the target energy storage container and a forced cooling command to the thermal management system of the target energy storage container; if it is confirmed that the system failure probability index value is greater than the third preset threshold, then outputting maintenance warning information as a level three maintenance warning information; and electrically isolating the faulty unit in the target energy storage container and initiating a directional inert gas fire extinguishing procedure.

[0077] In one possible implementation, after the processing module 32 performs deviation analysis between the system health status assessment vector and the baseline health status using Mahalanobis distance metric and calculates the system failure probability index value, the method further includes: based on the contribution value of each feature dimension in the health status assessment vector, outputting a fault diagnosis report corresponding to the target energy storage container, including the fault cause, fault propagation path and fault repair measures.

[0078] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0079] This application also provides an electronic device. (See reference...) Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 401, at least one communication bus 402, a user interface 403, at least one network interface 404, and a memory 405.

[0080] The communication bus 402 is used to enable communication between these components.

[0081] The user interface 403 may include a display screen and a camera. Optionally, the user interface 403 may also include a standard wired interface and a wireless interface.

[0082] The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0083] The processor 401 may include one or more processing cores. The processor 401 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 405, and by calling data stored in memory 405. Optionally, the processor 401 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 401 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 401.

[0084] The memory 405 may include random access memory (RAM) or read-only memory. Optionally, the memory 405 may include a non-transitory computer-readable storage medium. The memory 405 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 405 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 405 may also be at least one storage device located remotely from the aforementioned processor 401. (Refer to...) Figure 3 The memory 405, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a fault early warning application for energy storage containers based on real-time monitoring.

[0085] exist Figure 3In the illustrated electronic device, the user interface 403 is primarily used to provide an input interface for the user and acquire user input data; while the processor 401 can be used to call the energy storage container fault early warning application stored in the memory 405 based on real-time monitoring. When executed by one or more processors 401, the electronic device performs one or more of the methods described in the above embodiments. It should be noted that, for the foregoing method embodiments, for the sake of simplicity, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0086] This application also provides a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.

[0087] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0088] In the various embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0089] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0090] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0091] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0092] The above description is merely an exemplary embodiment disclosed in this application and should not be construed as limiting the scope of this application. Any equivalent changes and modifications made in accordance with the teachings of this application shall still fall within the scope of this application.

[0093] This application is intended to cover any variations, uses, or adaptations disclosed herein that follow the general principles disclosed herein and include common knowledge or customary technical means in the art that are not described in this application.

Claims

1. A method for early warning of energy storage container failures based on real-time monitoring, characterized in that, The method includes: The multimodal operating status data corresponding to the target energy storage container is preprocessed to construct a standardized multidimensional time-series dataset; the multimodal operating status data includes electrical status data, thermodynamic status data, chemical environment data, and mechanical vibration data; A multi-physics coupled energy storage container topology model is constructed, and a state estimation digital twin is built based on the multi-dimensional time series dataset and the energy storage container topology model. The system correlation graph structure is defined based on the multi-physics coupling energy storage container topology model, and a state reference representation is constructed based on the state estimation digital twin. A spatiotemporal graph convolutional neural network model is constructed using the system association graph structure as the spatial constraint for spatiotemporal feature propagation and the state reference representation as the health state mapping constraint. The spatiotemporal graph convolutional neural network model outputs a system health status assessment vector. The system failure probability index value is calculated based on the health status assessment vector, and the maintenance warning information corresponding to the target energy storage container is output based on the system failure probability index value.

2. The method according to claim 1, characterized in that, The preprocessing of the multimodal operating status data corresponding to the target energy storage container to construct a standardized multidimensional time-series dataset specifically includes: The multimodal operating status data is acquired based on a multimodal sensor system, which includes a voltage sensor, a current sensor, a distributed fiber optic temperature sensor, an infrared thermal imager, an electrochemical state sensing unit, and a microelectromechanical system accelerometer. The multimodal operational status data is preprocessed to construct a standardized multidimensional time-series dataset. The preprocessing operations include timestamp alignment, data denoising, normalization, and missing value imputation.

3. The method according to claim 2, characterized in that, The construction of the multi-physics coupled energy storage container topology model specifically includes: Based on the electrical state data and the chemical environment data, a single-particle model is used to describe the diffusion and embedding process of lithium ions in positive and negative electrode particles to construct an electrochemical model. Based on the aforementioned thermodynamic state data, the Navier-Stokes equation and energy conservation equation are used to simulate the flow and heat transfer of coolant in the pipeline to construct a thermal management model. Based on the mechanical vibration data, the Fourier heat conduction equation is used to describe the heat exchange process between the external environment and the internal equipment in order to construct a shell heat conduction model; A multi-physics coupled topology model of the energy storage container is constructed based on the electrochemical model, the thermal management model, and the shell thermal conduction model. An extended Kalman filter algorithm is used, and the standardized multidimensional time-series dataset is used as the observation set to iteratively update the state vectors in the digital twin that are not directly measured. The state vectors include the state of charge, state of health, remaining effective lifetime, internal temperature, degree of lithium plating, and coolant flow rate of each battery cell.

4. The method according to claim 3, characterized in that, The extended Kalman filter algorithm includes a prediction phase and an update phase. The iterative update of the state vector in the digital twin that is not directly measured specifically includes: In the prediction phase, the predicted observations corresponding to the multidimensional time series dataset and the covariance matrix corresponding to the state vector are predicted using the energy storage container topology model. In the update phase, the predicted observations are compared with the actual observations in the multidimensional time-series dataset to calculate the Kalman increment; The state vector and the covariance matrix corresponding to the state vector are iteratively corrected based on the Kalman increment.

5. The method according to claim 1, characterized in that, The energy storage container topology model based on the multi-physics coupling defines the system correlation graph structure and constructs a state reference representation based on the state estimation digital twin, specifically including: The system correlation graph structure is defined based on the multi-physics coupling energy storage container topology model. The system correlation graph structure includes nodes and edges. The nodes represent battery cells, battery modules and key components. The edges represent the physical connection relationship, electrical coupling path and heat conduction path between the nodes. The weight of the edges is quantified according to the Euclidean distance, electrical impedance or thermal resistance between the components. A state reference representation is constructed based on the state estimation digital twin. The state reference representation is used to represent the state reference characteristics of the target energy storage container under reference operating conditions. The state reference characteristics include a baseline health state, which is determined by the steady-state distribution established by the state estimation digital twin operating under fault-free historical data.

6. The method according to claim 5, characterized in that, The step of calculating the system failure probability index value based on the health status assessment vector and outputting the maintenance early warning information corresponding to the target energy storage container based on the system failure probability index value specifically includes: The deviation between the system health status assessment vector and the baseline health status is analyzed using Mahalanobis distance metric, and the system failure probability index value is calculated. If the system failure probability index value is confirmed to be greater than the first preset threshold and less than or equal to the second preset threshold, the maintenance warning information is output as a level 1 maintenance warning information; the level 1 maintenance warning information is used to prompt maintenance personnel to check relevant components through the human-machine interface and record abnormal event logs. If it is confirmed that the system failure probability index value is greater than the second preset threshold and less than or equal to the third preset threshold, then the maintenance warning information is output as a level 2 maintenance warning information; the level 2 maintenance warning information is used to send a power derating instruction to the energy conversion system of the target energy storage container and a forced heat dissipation instruction to the thermal management system of the target energy storage container. If the system failure probability index value is confirmed to be greater than the third preset threshold, the maintenance warning information is output as a level 3 maintenance warning information; the level 3 maintenance warning information is used to electrically isolate the faulty unit in the target energy storage container and to initiate a directional inert gas fire extinguishing procedure.

7. The method according to claim 6, characterized in that, After performing deviation analysis between the system health status assessment vector and the baseline health status using Mahalanobis distance metric, and calculating the system failure probability index value, the method further includes: Based on the contribution values ​​of each feature dimension in the health status assessment vector, a fault diagnosis report corresponding to the target energy storage container is output. The fault diagnosis report includes the cause of the fault, the fault propagation path, and the fault repair measures.

8. A fault early warning device for energy storage containers based on real-time monitoring, characterized in that, The device includes an acquisition module and a processing module, wherein, The acquisition module is used to preprocess the multimodal operating status data corresponding to the target energy storage container to construct a standardized multidimensional time-series dataset; the multimodal operating status data includes electrical status data, thermodynamic status data, chemical environment data, and mechanical vibration data; The processing module is used to construct a multi-physics coupled energy storage container topology model, and to construct a state estimation digital twin based on the multi-dimensional time-series dataset and the energy storage container topology model; to define a system correlation graph structure based on the multi-physics coupled energy storage container topology model, and to construct a state reference representation based on the state estimation digital twin; to construct a spatiotemporal graph convolutional neural network model using the system correlation graph structure as a spatial constraint for spatiotemporal feature propagation and the state reference representation as a health state mapping constraint; to output a system health state evaluation vector through the spatiotemporal graph convolutional neural network model; to calculate a system failure probability index value based on the health state evaluation vector, and to output maintenance warning information corresponding to the target energy storage container based on the system failure probability index value.

9. An electronic device, characterized in that, The device includes a processor, a communication bus, a user interface, a network interface, and a memory. The memory is used to store instructions. The user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.

Images