Battery power system digital twinning method and system applied to new energy heavy truck

Abstract

The invention discloses a battery power system digital twinning method and system applied to a new energy heavy truck. The method comprises the following steps: acquiring basic data; obtaining a multi-scale mapping model of the battery power system based on the basic data; adding a data driving model, performing information extraction on real-time monitoring data of the battery power system, supplementing the real-time monitoring data into the multi-scale mapping model, and constructing a digital twinborn model of the battery power system; the method comprises the following steps: abstracting sensor data of a battery power system into a network topological graph through a graph neural network-long and short-term memory neural network structure to carry out time-space relationship multi-scale modeling, and sensing operation state information of the battery power system by utilizing a constructed battery power system digital twinborn model; and constructing an incomplete multi-modal and multi-physical field deep semantic matching fusion model for the multi-modal data contained in the perceived operation state information by using a cross-modal deep semantic matching mechanism. The invention can be used.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a digital twin method and system for a battery power system applied to a new energy heavy-duty truck. Background Art

[0002] In the field of new energy heavy-duty truck battery power systems, existing technologies generally use rule-based energy management strategies and simplified battery models for system control. For example, traditional methods only build a single-level electrochemical model by monitoring apparent parameters such as battery voltage and temperature, and allocate power source power based on a fixed threshold. Although this solution implements basic energy management functions, it has significant technical defects: on the one hand, since it does not consider the dynamic impact of the coupling of multiple physical fields (such as electrochemical fields, temperature fields, and stress fields) on battery performance, the model cannot accurately reflect the actual state of the battery under complex working conditions; on the other hand, the lack of a real-time data-driven model update mechanism makes it difficult to adapt to the nonlinear characteristics of new energy heavy-duty trucks when operating in multiple working conditions. The above defects cause existing technologies to have problems such as low energy utilization efficiency and delayed system response when responding to dynamic environmental changes, and cannot meet the requirements of new energy heavy-duty trucks for high energy efficiency and high reliability. Summary of the Invention

[0003] The embodiments of the present invention provide a digital twin method and system for a battery power system applied to new energy heavy-duty trucks, which can.

[0004] An embodiment of the present invention provides a digital twin method for a battery power system of a new energy heavy-duty truck, comprising the following steps:

[0005] Obtain basic data for the digital twin of the battery power system used in new energy heavy-duty trucks. This data includes full-factor parameter information under the multi-physics coupling state of the electrochemical field, temperature field, and stress field of the battery power system of new energy heavy-duty trucks, as well as multi-scale data from battery cells to modules, complete packs, controllers, and drive motors.

[0006] Based on the basic data, an electrochemical-thermomechanical coupling model is constructed from the cell to module scale of the battery power system. A thermal-mechanical coupling model is constructed at the cell scale taking into account thermal stress and thermal expansion behavior. A geometric model is established and optimized at the module and whole package scales of the battery power system. At the same time, a dynamic model of the battery power system is established to achieve a comprehensive abstraction of the battery power system and obtain a multi-scale image model of the battery power system.

[0007] Adding a data-driven model to extract information from the real-time monitoring data of the battery power system, supplementing the extracted information into the multi-scale image model, and constructing a digital twin model of the battery power system;

[0008] Through the graph neural network-long short-term memory neural network structure, the sensor data of the battery power system is abstracted into a network topology graph for multi-scale modeling of spatiotemporal relationships, and the constructed digital twin model of the battery power system is used to perceive the operating status information of the battery power system;

[0009] Utilizing the cross-modal deep semantic matching mechanism, a shared feature subspace between modalities is constructed for the multimodal data contained in the perceived operating status information, and an invariant graph regularization factor is designed to establish a deep semantic matching fusion model for incomplete multimodal and multi-physical fields to reduce semantic deviation.

[0010] As an improvement to the above solution, based on the basic data, an electrochemical-thermomechanical coupling model is constructed from the cell to module scale of the battery power system. A thermal-mechanical coupling model is constructed at the single cell scale considering thermal stress and thermal expansion behavior. A geometric model is established and optimized from the module and whole package scales of the battery power system. At the same time, a dynamic model of the battery power system is established to achieve a comprehensive abstraction of the battery power system and obtain a multi-scale image model of the battery power system, including:

[0011] Cell-to-module scale modeling: At the cell scale, an electrochemical-thermal model is established based on the P2D model according to the energy conservation equation to obtain the cell-scale electrochemical-thermal model. At the cell scale, a thermal-mechanical coupling model of the power battery is constructed based on the thermal stress and thermal expansion behavior caused by temperature gradients during battery operation to obtain the cell-scale thermal-mechanical coupling model.

[0012] Module and full-pack scale modeling: Based on the cell-scale electrochemical-thermal model and the cell-scale thermal-mechanical coupling model, industrial modeling software is used to construct a 3D CAD model of the power battery. This accurately expresses its key structural parameters, the constraints and positioning relationships between cells and modules, and produces a 3D CAD model. Pixyz Studio is then used to perform geometric optimization to obtain the optimized geometric model.

[0013] Dynamic modeling: Using the optimized geometric model as input, the mechanical geometry, material parameters, force analysis, and velocity load information of the power battery pack, electric motor, inverter, and drive control unit are combined to establish a dynamic model that includes the mechanical geometry, material parameters, force analysis, and velocity load information.

[0014] As an improvement to the above solution, the data-driven model is added to extract information from the real-time monitoring data of the battery power system, and the extracted information is added to the multi-scale image model to construct a digital twin model of the battery power system, including:

[0015] Model system fusion: The multi-scale image model obtained in the multi-scale modeling step is integrated with the data-driven model to form a pre-fusion model system;

[0016] Real-time data processing: A data-driven approach is used to extract information from real-time monitoring data, obtaining the extracted information and supplementing it with the multi-scale image model in the pre-fusion model system to form a supplemented multi-scale image model. The real-time monitoring data includes sensor data from the physical layer and simulation data from the virtual layer.

[0017] Digital twin construction: Based on the supplemented multi-scale image model, a multi-scale digital twin model of the battery power system is constructed to achieve data-driven model fusion.

[0018] As an improvement to the above solution, the sensor data of the battery power system is abstracted into a network topology graph through the graph neural network-long short-term memory neural network structure to perform multi-scale modeling of spatiotemporal relationships, and the constructed digital twin model of the battery power system is used to perceive the operating status information of the battery power system, including:

[0019] Data abstraction processing: sensor data is abstracted into a network topology diagram of the sensor data time series accumulation process. The connections between nodes are defined as sequential events to form a network topology diagram of the sensor data.

[0020] Dynamic graph construction: Based on the network topology of the sensor data, a time-series dynamic graph structure of battery-motor-brake characteristic information fusion is constructed to obtain a time-series dynamic graph including battery-motor-brake characteristics;

[0021] Feature learning processing: A graph neural network-long short-term memory neural network structure is used to integrate the structural information and time information of the temporal dynamic graph, a hierarchical feature aggregation method is designed, and feature aggregators are learned in different depth neighborhoods to realize the flow of information from higher depths to nodes and obtain state perception results.

[0022] As an improvement to the above solution, the cross-modal deep semantic matching mechanism is used to construct a shared feature subspace between modalities for the multimodal data contained in the perceived operating status information, and an invariant graph regularization factor is designed to establish a deep semantic matching fusion model for incomplete multimodal and multi-physical fields, thereby reducing semantic deviation, including:

[0023] Cross-modal modeling: Using a cross-modal deep semantic matching mechanism and multi-layer nonlinear correlations between multi-level and multi-scale modal data, we establish an incomplete cross-modal deep semantic matching fusion model to obtain an initial fusion model.

[0024] Feature space construction: constructing a shared feature subspace between modalities based on the initial fusion model, learning the sharing of incomplete cross-modal data, and forming a shared feature subspace;

[0025] Regularization processing: designing an invariant graph regularization factor in the shared feature subspace to ensure the local similarity characteristics of each modal data and obtain a regularized shared subspace;

[0026] Objective function establishment: A new objective function is established based on the regularized shared subspace to form a complete multimodal data fusion model to reduce semantic bias.

[0027] Another embodiment of the present invention provides a digital twin system of a battery power system for a new energy heavy-duty truck, including:

[0028] An acquisition module is used to obtain basic data for the digital twin of the battery power system used in new energy heavy-duty trucks. The basic data includes: full-factor parameter information under the multi-physics coupling state of the electrochemical field, temperature field, and stress field of the battery power system of new energy heavy-duty trucks, as well as multi-scale data from battery cells to modules, complete packs, controllers, and drive motors;

[0029] A first construction module is configured to construct an electrochemical-thermomechanical coupling model of the battery power system from the cell to the module scale based on the basic data, construct a thermal-mechanical coupling model at the cell scale taking into account thermal stress and thermal expansion behavior, establish and optimize a geometric model at the module and entire package scales of the battery power system, and simultaneously establish a dynamic model of the battery power system, thereby achieving a comprehensive abstraction of the battery power system and obtaining a multi-scale image model of the battery power system;

[0030] A second construction module is configured to add a data-driven model, extract information from the real-time monitoring data of the battery power system, and add the extracted information to the multi-scale image model to construct a digital twin model of the battery power system;

[0031] A third construction module is used to abstract the sensor data of the battery power system into a network topology graph through a graph neural network-long short-term memory neural network structure to perform multi-scale modeling of spatiotemporal relationships, and use the constructed digital twin model of the battery power system to perceive the operating status information of the battery power system;

[0032] The fourth construction module is used to utilize the cross-modal deep semantic matching mechanism to construct a shared feature subspace between modalities for the multimodal data contained in the perceived operating status information, and to design an invariant graph regularization factor to establish a deep semantic matching fusion model for incomplete multimodal and multi-physical fields to reduce semantic deviation.

[0033] As an improvement to the above solution, the first building block is specifically configured to:

[0034] Cell-to-module scale modeling: At the cell scale, an electrochemical-thermal model is established based on the P2D model according to the energy conservation equation to obtain the cell-scale electrochemical-thermal model. At the cell scale, a thermal-mechanical coupling model of the power battery is constructed based on the thermal stress and thermal expansion behavior caused by temperature gradients during battery operation to obtain the cell-scale thermal-mechanical coupling model.

[0035] Module and full-pack scale modeling: Based on the cell-scale electrochemical-thermal model and the cell-scale thermal-mechanical coupling model, industrial modeling software is used to construct a 3D CAD model of the power battery. This accurately expresses its key structural parameters, the constraints and positioning relationships between cells and modules, and produces a 3D CAD model. Pixyz Studio is then used to perform geometric optimization to obtain the optimized geometric model.

[0036] Dynamic modeling: Using the optimized geometric model as input, the mechanical geometry, material parameters, force analysis, and velocity load information of the power battery pack, electric motor, inverter, and drive control unit are combined to establish a dynamic model that includes the mechanical geometry, material parameters, force analysis, and velocity load information.

[0037] As an improvement to the above solution, the second building block is specifically configured to:

[0038] Model system fusion: The multi-scale image model obtained in the multi-scale modeling step is integrated with the data-driven model to form a pre-fusion model system;

[0039] Real-time data processing: A data-driven approach is used to extract information from real-time monitoring data, obtaining the extracted information and supplementing it with the multi-scale image model in the pre-fusion model system to form a supplemented multi-scale image model. The real-time monitoring data includes sensor data from the physical layer and simulation data from the virtual layer.

[0040] Digital twin construction: Based on the supplemented multi-scale image model, a multi-scale digital twin model of the battery power system is constructed to achieve data-driven model fusion.

[0041] As an improvement to the above solution, the third building block is specifically configured to:

[0042] Data abstraction processing: sensor data is abstracted into a network topology diagram of the sensor data time series accumulation process. The connections between nodes are defined as sequential events to form a network topology diagram of the sensor data.

[0043] Dynamic graph construction: Based on the network topology of the sensor data, a time-series dynamic graph structure of battery-motor-brake characteristic information fusion is constructed to obtain a time-series dynamic graph including battery-motor-brake characteristics;

[0044] Feature learning processing: A graph neural network-long short-term memory neural network structure is used to integrate the structural information and time information of the temporal dynamic graph, a hierarchical feature aggregation method is designed, and feature aggregators are learned in different depth neighborhoods to realize the flow of information from higher depths to nodes and obtain state perception results.

[0045] As an improvement to the above solution, the fourth building block is specifically configured to:

[0046] Cross-modal modeling: Using a cross-modal deep semantic matching mechanism and multi-layer nonlinear correlations between multi-level and multi-scale modal data, we establish an incomplete cross-modal deep semantic matching fusion model to obtain an initial fusion model.

[0047] Feature space construction: constructing a shared feature subspace between modalities based on the initial fusion model, learning the sharing of incomplete cross-modal data, and forming a shared feature subspace;

[0048] Regularization processing: designing an invariant graph regularization factor in the shared feature subspace to ensure the local similarity characteristics of each modal data and obtain a regularized shared subspace;

[0049] Objective function establishment: A new objective function is established based on the regularized shared subspace to form a complete multimodal data fusion model to reduce semantic bias.

[0050] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0051] The embodiment of the present invention constructs a multi-scale digital twin model coupled with multiple physical fields, combines the data-driven method to realize dynamic model updates, and uses spatiotemporal feature fusion and cross-modal semantic matching technology to process multi-source heterogeneous data. Specifically, first, all the parameters and multi-scale data of the multi-physical fields (electrochemical field, temperature field, stress field) of the battery power system are collected to build a multi-level physical model from battery cells to the whole package; then, the real-time monitoring data is supplemented by the data-driven model to improve the accuracy of the model; then, the graph neural network-long short-term memory neural network structure is used to process the spatiotemporal relationship of the sensor data to achieve accurate perception of the operating status; finally, a shared feature subspace is established through a cross-modal semantic matching mechanism and a regularization factor is introduced to solve the problem of incomplete multimodal data. This technical concept breaks through the limitations of traditional models that rely only on apparent parameters and lack dynamic adaptability through hierarchical model construction and data fusion methods, and realizes all-round, high-precision simulation and real-time optimization of battery power systems. The embodiment of the present invention solves the technical defect that traditional methods cannot dynamically reflect the complex operating conditions of batteries through multi-physics field coupling modeling and real-time data-driven mechanism, realizes the accurate mapping and real-time optimization of the multi-scale operating status of the battery power system of new energy heavy-duty trucks, and significantly improves the system's energy utilization efficiency and dynamic response capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of a digital twin method for a battery power system applied to a new energy heavy-duty truck, provided by one embodiment of the present invention;

[0053] Figure 2 This is a structural schematic diagram of a digital twin system of a battery power system applied to new energy heavy-duty trucks provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0055] See also Figure 1 , is a flow chart of a digital twin method for a battery power system applied to a new energy heavy-duty truck, provided by one embodiment of the present invention. The digital twin method for a battery power system applied to a new energy heavy-duty truck comprises the following steps:

[0056] S10, obtaining basic data of a digital twin of a battery power system for a new energy heavy-duty truck, the basic data including: full-factor parameter information of the battery power system of the new energy heavy-duty truck under a multi-physics field coupling state of electrochemical field, temperature field, and stress field, as well as multi-scale data from battery cells to modules, entire packs, controllers, and drive motors;

[0057] S11, based on the basic data, constructing an electrochemical-thermomechanical coupling model from the cell to module scale of the battery power system, constructing a thermal-mechanical coupling model at the cell scale taking into account thermal stress and thermal expansion behavior, establishing and optimizing a geometric model at the module and entire package scales of the battery power system, and simultaneously establishing a dynamic model of the battery power system, thereby achieving a comprehensive abstraction of the battery power system and obtaining a multi-scale image model of the battery power system;

[0058] S12, adding a data-driven model to extract information from the real-time monitoring data of the battery power system, adding the extracted information to the multi-scale image model, and constructing a digital twin model of the battery power system;

[0059] S13, abstracting the sensor data of the battery power system into a network topology graph through a graph neural network-long short-term memory neural network structure to perform multi-scale modeling of spatiotemporal relationships, and using the constructed digital twin model of the battery power system to perceive the operating status information of the battery power system;

[0060] S14, using a cross-modal deep semantic matching mechanism, constructs a shared feature subspace between modalities for the multimodal data contained in the perceived operating status information, and designs an invariant graph regularization factor to establish a deep semantic matching fusion model for incomplete multimodal and multi-physical fields to reduce semantic deviation.

[0061] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0062] The embodiment of the present invention constructs a multi-scale digital twin model coupled with multiple physical fields, combines the data-driven method to realize dynamic model updates, and uses spatiotemporal feature fusion and cross-modal semantic matching technology to process multi-source heterogeneous data. Specifically, first, all the parameters and multi-scale data of the multi-physical fields (electrochemical field, temperature field, stress field) of the battery power system are collected to build a multi-level physical model from battery cells to the whole package; then, the real-time monitoring data is supplemented by the data-driven model to improve the accuracy of the model; then, the graph neural network-long short-term memory neural network structure is used to process the spatiotemporal relationship of the sensor data to achieve accurate perception of the operating status; finally, a shared feature subspace is established through a cross-modal semantic matching mechanism and a regularization factor is introduced to solve the problem of incomplete multimodal data. This technical concept breaks through the limitations of traditional models that rely only on apparent parameters and lack dynamic adaptability through hierarchical model construction and data fusion methods, and realizes all-round, high-precision simulation and real-time optimization of battery power systems. The embodiment of the present invention solves the technical defect that traditional methods cannot dynamically reflect the complex operating conditions of batteries through multi-physics field coupling modeling and real-time data-driven mechanism, realizes the accurate mapping and real-time optimization of the multi-scale operating status of the battery power system of new energy heavy-duty trucks, and significantly improves the system's energy utilization efficiency and dynamic response capabilities.

[0063] As one of the embodiments, in step S10, by comprehensively collecting multi-physical field coupling data and multi-scale structural data of the new energy heavy-duty truck battery power system, complete basic information is provided for subsequent digital twin modeling. Traditional technology relies only on battery apparent parameters (such as voltage and temperature) for modeling, which cannot reflect the dynamic coupling of internal multi-physical fields (electrochemical field, temperature field, stress field) and the interaction between multi-scale levels, resulting in insufficient model accuracy. This solution breaks through the single-dimensional limitation of traditional models by synchronously acquiring full-factor parameters (such as electrochemical characteristics, thermal distribution, mechanical stress) and hierarchical data from battery cells to entire packages (including controllers, drive motors, etc.), laying a data foundation for building a high-precision digital twin model, and ultimately achieving accurate mapping and optimization of the full life cycle status of the battery power system.

[0064] Specifically, for multi-physical field data acquisition: a distributed sensor network is used to monitor the electrochemical field (such as state of charge SOC, internal resistance), temperature field (such as cell / module temperature distribution) and stress field (such as mechanical deformation, thermal expansion) parameters of the battery power system in real time, covering the three operating conditions of dynamic charging and discharging, constant charging and static operation of the battery. Combined with simulation data (such as the thermal diffusion trend predicted by the virtual layer), a twin data set containing physical layer monitoring data and virtual layer simulation data is formed. For multi-scale structural data acquisition: microscopic electrochemical parameters (such as ion diffusion coefficient) are obtained at the cell level, series and parallel connection relationships and thermal management parameters are recorded at the module level, and overall energy flow and mechanical structure parameters are obtained at the whole package level. Industrial modeling software (such as CATIA) is used to build a three-dimensional CAD model of the power battery, accurately expressing key structural parameters (such as electrode material thickness and module spacing), and surface subdivision and topology correction are performed through Pixyz Studio to ensure that the geometric model is highly consistent with the physical entity. For sensor data time series accumulation: Sensor data is abstracted into a network topology graph, with nodes representing sensors (such as voltage sensors and temperature sensors) and edges representing data time series relationships, forming a dynamic spatiotemporal dataset. Methods such as non-negative matrix factorization (NMF) are used to reduce the dimensionality of multi-source heterogeneous data, extract core feature vectors, and reduce data redundancy.

[0065] As one embodiment, based on the basic data, an electrochemical-thermomechanical coupling model is constructed from the cell to module scale of the battery power system, a thermal-mechanical coupling model is constructed at the monomer scale considering thermal stress and thermal expansion behavior, a geometric model is established and optimized from the module and whole package scales of the battery power system, and a dynamic model of the battery power system is established at the same time, achieving a comprehensive abstraction of the battery power system and obtaining a multi-scale image model of the battery power system, including:

[0066] Cell-to-module scale modeling: At the cell scale, an electrochemical-thermal model is established based on the P2D model according to the energy conservation equation to obtain the cell-scale electrochemical-thermal model. At the cell scale, a thermal-mechanical coupling model of the power battery is constructed based on the thermal stress and thermal expansion behavior caused by temperature gradients during battery operation to obtain the cell-scale thermal-mechanical coupling model.

[0067] Module and full-pack scale modeling: Based on the cell-scale electrochemical-thermal model and the cell-scale thermal-mechanical coupling model, industrial modeling software is used to construct a 3D CAD model of the power battery. This accurately expresses its key structural parameters, the constraints and positioning relationships between cells and modules, and produces a 3D CAD model. Pixyz Studio is then used to perform geometric optimization to obtain the optimized geometric model.

[0068] Dynamic modeling: Using the optimized geometric model as input, the mechanical geometry, material parameters, force analysis, and velocity load information of the power battery pack, electric motor, inverter, and drive control unit are combined to establish a dynamic model that includes the mechanical geometry, material parameters, force analysis, and velocity load information.

[0069] In this embodiment, based on the basic data of the battery power system of the new energy heavy-duty truck obtained, models are constructed from different scales and levels, gradually achieving a comprehensive abstraction of the battery power system to obtain an accurate multi-scale image model. Traditional battery power system modeling often only focuses on a single scale or a single physical field, which makes it difficult to fully reflect the complex characteristics of the system. However, this solution comprehensively considers the coupling effects of multiple physical fields such as electrochemistry, heat, and force, as well as the dynamic characteristics of the system, from the battery cell to the module, monomer, module and whole package scales, and organically combines models of different scales. Through this hierarchical progressive and multi-dimensional modeling approach, the behavior and performance of the battery power system under different working conditions can be accurately captured, providing a solid foundation for the subsequent construction of the digital twin model, thereby significantly improving the prediction and analysis accuracy of the battery power system status, and helping to optimize the system design and operation strategy.

[0070] Specifically, the implementation process of this embodiment is as follows:

[0071] The process of cell-to-module scale modeling includes:

[0072] Cell-scale electrochemical-thermal model: Based on the energy conservation equation, an electrochemical-thermal model is established at the cell scale based on the P2D (pseudo-two-dimensional) model. The P2D model is a commonly used model to describe the electrochemical processes within lithium-ion batteries. The energy conservation equation can be expressed as:

[0073]

[0074] Where ρ is the density of the battery material, c p is the specific heat capacity, T is the temperature, k is the thermal conductivity, Q rxn is the heat generation rate of the chemical reaction, Q ohm is the ohmic heat generation rate. By solving this equation and combining the description of electrochemical processes such as electrode reaction and ion transport in the P2D model, the electrochemical-thermal model at the cell scale is obtained. Single-cell scale thermal-mechanical coupling model: At the single-cell scale, the thermal stress and thermal expansion behavior caused by the temperature gradient during battery operation are considered. Thermal stress can be calculated based on the theory of thermoelasticity, and thermal expansion behavior can be described by the thermal expansion coefficient. Assuming that the battery material is isotropic, the thermal stress σ th It can be expressed as:

[0075] σ th =EαΔT

[0076] Where E is the elastic modulus of the material, α is the thermal expansion coefficient, and ΔT is the temperature change. Based on this, we combine the mechanical equilibrium equation and the heat conduction equation to construct a thermal-mechanical coupling model for the power battery, resulting in a single-cell-scale thermal-mechanical coupling model.

[0077] Module and package scale modeling, the process includes:

[0078] 3D CAD Model Construction: Using a cell-scale electrochemical-thermal model and a cell-scale thermal-mechanical coupling model as a foundation, a 3D CAD model of the power battery is constructed using industrial modeling software (such as CATIA and SolidWorks). During the modeling process, key structural parameters (such as cell size and module layout) as well as the constraints and positioning relationships between cells and modules are accurately expressed. Accurate modeling of these parameters and relationships allows for a true reflection of the physical structure of the battery power system.

[0079] Geometry Optimization: Use Pixyz Studio to optimize the geometry of constructed 3D CAD models. The software can perform operations such as tessellation, topology correction, extraction, and repair. For example, tessellation can improve the surface accuracy of the model, while topology correction can resolve topological errors in the model, resulting in an optimized geometry model, improving model quality and analyzability.

[0080] Dynamic modeling, the process includes:

[0081] The optimized geometric model is used as input, and the mechanical geometry, material parameters, force analysis, and speed load information of the power battery pack, motor, inverter, and drive control unit are combined. Based on Newton's second law and the principle of dynamics, the dynamic equation is established. Assume that the total mass of the system is m and the total external force is F total , the acceleration is a, then the dynamic equation can be expressed as:

[0082] F total =m·a

[0083] At the same time, considering factors such as the torque of the motor and the power conversion of the inverter, the interaction between these components is described by establishing corresponding mathematical models. For example, the torque T of the motor can be expressed as:

[0084] T=k·I

[0085] Where k is the motor's torque constant and I is the current. By comprehensively considering these factors, a dynamic model is established that includes mechanical geometry, material parameters, force analysis, and speed load information.

[0086] As one embodiment, the data-driven model is added to extract information from the real-time monitoring data of the battery power system, and the extracted information is added to the multi-scale image model to construct a digital twin model of the battery power system, including:

[0087] Model system fusion: The multi-scale image model obtained in the multi-scale modeling step is integrated with the data-driven model to form a pre-fusion model system;

[0088] Real-time data processing: A data-driven approach is used to extract information from real-time monitoring data, obtaining the extracted information and supplementing it with the multi-scale image model in the pre-fusion model system to form a supplemented multi-scale image model. The real-time monitoring data includes sensor data from the physical layer and simulation data from the virtual layer.

[0089] Digital twin construction: Based on the supplemented multi-scale image model, a multi-scale digital twin model of the battery power system is constructed to achieve data-driven model fusion.

[0090] In this embodiment, a digital twin model that accurately reflects the actual operating status of the battery power system is constructed by integrating a multiscale image model with a data-driven model and updating and improving the model with real-time monitoring data. Traditional battery power system modeling often lacks real-time data updates and dynamic adjustments, making it difficult to adapt to complex and changing actual operating conditions. This solution innovatively introduces a data-driven model, using real-time monitoring data (including sensor data at the physical layer and simulation data at the virtual layer) to supplement and correct the multiscale image model. This approach enables the digital twin model to track the state changes of the battery power system in real time, improving the model's accuracy and reliability, and providing stronger support for optimized control, fault diagnosis, and predictive maintenance of new energy heavy-duty truck battery power systems. In summary, multiscale image models suffer from parameter deviation under complex operating conditions, while data-driven models can extract features from real-time data and supplement the physical model. Through this two-layer architecture of "multiscale image model + data-driven model," this solution achieves a dynamic improvement in model accuracy, enabling the digital twin model to accurately reflect the real-time status of the battery power system, ultimately improving the energy efficiency of new energy heavy-duty trucks.

[0091] Specifically, the implementation process of this embodiment is as follows:

[0092] The model system integration process includes:

[0093] The multi-scale image model M ms With data-driven model M dd Integration is achieved through the following methods: in represents the feature concatenation operation, For data-driven models (such as LSTM networks), D real-time For real-time monitoring data.

[0094] Real-time data processing, the process includes:

[0095] Feature extraction: real-time monitoring data D real-time (including sensor data D sensor and simulation data D sim ) applies the following non-negative matrix factorization (NMF):

[0096]

[0097] in, is the complete modal data, is incomplete modal data, U (υ) is the basis matrix, P C and is the shared feature matrix.

[0098] Information supplement: The extracted feature P C and Input multi-scale image model, update the parameters through the following data-driven model supplementation mechanism: Where α is the learning rate, ⊙ represents element-wise multiplication, and the ReLU activation function ensures non-negativity.

[0099] The digital twin construction process includes:

[0100] Based on the supplemented multi-scale image model M′ ms , build a digital twin model M DT :

[0101] in It is a GCN-LSTM network. The specific implementation is as follows:

[0102] Graph neural network layer: uses the GraphSAGE aggregation formula:

[0103] in is the set of k-th layer neighbor nodes of node υ.

[0104] Long Short-Term Memory Network layer: Apply the following LSTM units to the spatiotemporal features:

[0105] i t =σ(W xi x t +W hi h t-1 +b i ),

[0106] f t =σ(W xf x t +W hf h t-1 +b f ),

[0107] o t =σ(W xo x t +W ho h t-1 +b o ),

[0108]

[0109] h t =o t ☉tanh(c t ),

[0110] where x t is the current input, h t is the hidden state, c t For memory cells.

[0111] The above process explains: the output M of the model system fusion pre-fusion As the input of real-time data processing, feature P is extracted by NMF. C and The result of real-time data processing M′ ms As the initial state of digital twin construction, the spatiotemporal features are fused through the GCN-LSTM network and the final output is M DT .

[0112] As one of the embodiments, the sensor data of the battery power system is abstracted into a network topology graph through a graph neural network-long short-term memory neural network structure to perform multi-scale modeling of spatiotemporal relationships, and the constructed digital twin model of the battery power system is used to perceive the operating status information of the battery power system, including:

[0113] Data abstraction processing: sensor data is abstracted into a network topology diagram of the sensor data time series accumulation process. The connections between nodes are defined as sequential events to form a network topology diagram of the sensor data.

[0114] Dynamic graph construction: Based on the network topology of the sensor data, a time-series dynamic graph structure of battery-motor-brake characteristic information fusion is constructed to obtain a time-series dynamic graph including battery-motor-brake characteristics;

[0115] Feature learning processing: A graph neural network-long short-term memory neural network structure is used to integrate the structural information and time information of the temporal dynamic graph, a hierarchical feature aggregation method is designed, and feature aggregators are learned in different depth neighborhoods to realize the flow of information from higher depths to nodes and obtain state perception results.

[0116] In this embodiment, the problem that traditional sensor data processing methods cannot effectively capture the dynamic characteristics of multi-physical field coupling in battery power systems is solved through spatiotemporal feature fusion and dynamic graph modeling. Among them, sensor data has strong correlation and time series, but traditional methods only perform single time series analysis, resulting in feature loss. This solution abstracts sensor data into a network topology graph, combines the battery-motor-brake characteristics to construct a time series dynamic graph, and fuses spatial structure information with time series information through the GCN-LSTM structure. This solution breaks through the spatiotemporal decoupling limitations of traditional methods, realizes accurate perception of the operating status of the battery power system, and provides real-time and comprehensive status basis for subsequent control decisions.

[0117] Specifically, the implementation process of this embodiment is as follows:

[0118] Data abstraction processing, the process includes:

[0119] Network topology construction: sensor data is abstracted into a set G = (V, E, X, L), where: V = {υ1,υ2,…,υ n} is a set of sensor nodes (such as voltage, temperature, and current sensors); E = {(υ i ,υ j )|nodeυ i with υ j There is a temporal correlation}; is the node feature (such as sensor measurement value); L = {l1, l2, ..., l m} is the node label (such as battery status classification).

[0120] Modeling spatiotemporal relationships: using adjacency matrix Represents node connection, where A i，j =1 if and only if (υ i ,υ j )∈E.

[0121] The dynamic graph construction process includes:

[0122] Feature information fusion: Map battery characteristics (such as SOC, internal resistance), motor characteristics (such as torque, speed), and braking characteristics (such as braking force distribution) to network nodes to form a multimodal feature matrix

[0123] Time series dynamic graph generation: Use the GraphSAGE method to convert time series data into dynamic graphs: Among them G (t) It represents the dynamic graph at time t, and AGGR is the time series aggregation function.

[0124] Feature learning processing, the process includes:

[0125] GCN-LSTM structure:

[0126] Graph convolution layer: using aggregation formula: in is the k-th layer neighbor set of node υ, is the k-th layer embedding vector.

[0127] LSTM layer: Apply LSTM units to each layer of embedding vectors:

[0128] i t =σ(W xi x t +W hi h t-1 +b i ),

[0129] f t =σ(W xf x t +W hf h t-1 +b f ),

[0130] o t =σ(W xo x t +W ho h t-1 +b o ),

[0131]

[0132] h t =o t ☉tanh(c t ),

[0133] where x t is the current graph embedding, h t is the hidden state, c t For memory cells.

[0134] Hierarchical feature aggregation: Design a hierarchical feature aggregation method and learn feature aggregators for each k-layer neighborhood: Where K is the maximum neighborhood depth and CONCAT represents feature concatenation.

[0135] Supervised learning training: The model is trained using the categorical cross entropy loss function: where yυ，c is the true label of node υ belonging to category c, is the predicted probability.

[0136] The above process explains that the output G of data abstraction processing is used as the input of dynamic graph construction, and G is generated by fusion of feature information. (t) The result of dynamic graph construction G (t) Input feature learning processing, extract spatiotemporal features through GCN-LSTM, and finally output state perception results.

[0137] As one of the embodiments, the cross-modal deep semantic matching mechanism is used to construct a shared feature subspace between modalities for the multimodal data contained in the perceived operating status information, and an invariant graph regularization factor is designed to establish a deep semantic matching fusion model for incomplete multimodal and multi-physical fields to reduce semantic deviation, including:

[0138] Cross-modal modeling: Using a cross-modal deep semantic matching mechanism and multi-layer nonlinear correlations between multi-level and multi-scale modal data, we establish an incomplete cross-modal deep semantic matching fusion model to obtain an initial fusion model.

[0139] Feature space construction: constructing a shared feature subspace between modalities based on the initial fusion model, learning the sharing of incomplete cross-modal data, and forming a shared feature subspace;

[0140] Regularization processing: designing an invariant graph regularization factor in the shared feature subspace to ensure the local similarity characteristics of each modal data and obtain a regularized shared subspace;

[0141] Objective function establishment: A new objective function is established based on the regularized shared subspace to form a complete multimodal data fusion model to reduce semantic bias.

[0142] In this embodiment, for the multimodal data in the operating status information of the battery power system of new energy heavy trucks, a cross-modal deep semantic matching mechanism is used to solve the semantic deviation problem during the fusion of incomplete multimodal and multi-physical field data. Traditional data fusion methods have difficulty in handling the complex nonlinear relationships and data missing problems between multimodal data, resulting in inaccurate fusion results. This solution first establishes an initial fusion model through a cross-modal deep semantic matching mechanism to capture the multi-layer nonlinear correlations between multi-level and multi-scale modal data; then constructs a shared feature subspace between modalities to learn the sharing of incomplete cross-modal data; then designs an invariant graph regularization factor to ensure the local similarity characteristics of each modal data; finally, establishes a new objective function to form a complete multimodal data fusion model. In this way, semantic deviation can be effectively reduced, the accuracy of multimodal data fusion can be improved, and more reliable data support can be provided for accurate analysis and decision-making of battery power systems.

[0143] Specifically, the implementation process of this embodiment is as follows:

[0144] Cross-modal modeling: Using a cross-modal deep semantic matching mechanism, an incomplete cross-modal deep semantic matching fusion model is established. Assume that there are M modalities in the multimodal data, and the data of the mth modality is represented by X m , m=1,2,…,M. Through multi-layer nonlinear transformation function f m Processing of each modal data:

[0145] H m =f m (X m )

[0146] Among them H m is the feature representation of the mth mode after transformation. Then, considering the multi-layer nonlinear correlation between the modes, the initial fusion model F0 is established, which can be expressed as: F0 = g(H1, H2, ..., H M ),

[0147] Here g is a nonlinear function that integrates the features of each modality, for example, it can be a multi-layer perceptron (MLP) structure.

[0148] Feature space construction: Based on the initial fusion model F0, the shared feature subspace between the modalities is constructed. Using the non-negative matrix factorization (NMF) method, assuming that the basis matrix of the shared feature subspace is U and the coefficient matrix is P, the transformed feature H of each modality is m ,have:

[0149]

[0150] stU≥0, P m ≥0

[0151] By solving the above optimization problem, we can obtain the shared basis matrix U and the coefficient matrix P corresponding to each mode. m , thus forming a shared feature subspace.

[0152] Regularization: Design an invariant graph regularization factor in the shared feature subspace. Define the invariant graph regularization factor R to ensure the local similarity of each modal data. Let S be the local similarity matrix of each modal data, for example, which can be obtained by calculating the Euclidean distance or cosine similarity between data points. Then the invariant graph regularization factor R can be expressed as:

[0153]

[0154] Among them U i and U jare the i-th and j-th rows of the shared basis matrix U. Adding the regularization factor R to the original objective function, we get the regularized objective function:

[0155]

[0156] Where λ is the regularization coefficient, and the value of λ is adjusted to balance the constraints of data fitting and local similarity. Solve the above regularized objective function to obtain the regularized shared basis matrix U′ and coefficient matrix P′ m , thus obtaining the regularized shared subspace.

[0157] Objective function establishment: Establish a new objective function based on the regularized shared subspace. Let the regularized shared basis matrix be U′ and the coefficient matrix be P′ m , for prediction tasks (such as battery status classification), let the true label be Y and the predicted label be You can use the cross entropy loss function (for classification problems) or the mean square error loss function (for regression problems). Taking the cross entropy loss function as an example, the new objective function L can be expressed as:

[0158]

[0159] where Y i，c is the true label of the i-th sample belonging to the c-th class, is the prediction probability. By minimizing the objective function L and continuously adjusting the model parameters, a complete multimodal data fusion model is formed, thereby reducing semantic bias and improving the model's prediction accuracy for the battery power system's operating status.

[0160] As one example, after establishing a deep semantic matching fusion model for incomplete multimodal and multiphysical fields, this incomplete multimodal and multiphysical field deep semantic matching fusion model is the core enabling technology for the digital twin of the battery power system of new energy heavy-duty trucks. By eliminating the semantic gap between multi-source data, it achieves high-precision mapping between physical entities and virtual models, ultimately supporting real-time optimization control and full life cycle management of the battery power system. In addition, this model can be applied in the following ways:

[0161] Real-time condition monitoring and prediction: This model integrates multimodal data from the battery power system (such as electrochemical parameters, temperature distribution, and mechanical stress) to analyze the comprehensive performance of the battery under complex operating conditions in real time. For example, it combines the cell-level SOC (state of charge), the module-level temperature gradient, and the entire pack-level stress distribution to predict the battery pack's available capacity decay trend under different driving conditions.

[0162] Energy Management Optimization: Based on a fusion model's in-depth analysis of the battery, motor, and brake characteristics, regenerative braking energy recovery strategies can be optimized. For example, based on the real-time battery charging capacity constraints and motor power generation characteristics, the braking force distribution ratio can be dynamically adjusted to maximize energy recovery.

[0163] Fault diagnosis and lifespan assessment: Leveraging multi-physics coupling features extracted from a shared feature subspace, potential failure modes within the battery can be identified. For example, by analyzing the spatiotemporal correlation between temperature and stress fields, the development trend of microcracks within the battery cell can be located, enabling accurate prediction of battery life.

[0164] It can be understood that the fusion model is the "semantic engine" of the digital twin system, providing the virtual model with a deep understanding of multimodal data. It semantically aligns the physical layer sensor data with the virtual layer simulation data through a shared feature subspace to ensure that the digital twin model can reflect the true state of the battery power system in real time. The output of the fusion model (such as the state perception result) is used as a feedback signal to continuously calibrate the parameters of the digital twin model. For example, when it is detected that the temperature difference between battery cells exceeds the threshold, the thermal management strategy optimization of the digital twin model is automatically triggered. The semantic matching results of the model are directly input into the online optimal control mechanism driven by the digital twin. For example, in the multi-agent collaborative control framework, the battery health status assessment results provided by the fusion model can guide the energy management agent to dynamically adjust the charging and discharging strategy.

[0165] See also Figure 2 , is a schematic diagram of the structure of a digital twin system of a battery power system applied to a new energy heavy-duty truck provided by one embodiment of the present invention. The digital twin system of the battery power system applied to a new energy heavy-duty truck includes:

[0166] Acquisition module 10 is used to obtain basic data for the digital twin of the battery power system used in new energy heavy-duty trucks. The basic data includes: full-factor parameter information of the battery power system of the new energy heavy-duty truck in a multi-physics field coupling state of the electrochemical field, temperature field, and stress field, as well as multi-scale data from battery cells to modules, entire packs, controllers, and drive motors;

[0167] A first construction module 11 is configured to construct an electrochemical-thermomechanical coupling model of the battery power system from the cell to the module scale based on the basic data, construct a thermal-mechanical coupling model at the cell scale taking into account thermal stress and thermal expansion behavior, establish and optimize a geometric model at the module and entire package scales of the battery power system, and simultaneously establish a dynamic model of the battery power system, thereby achieving a comprehensive abstraction of the battery power system and obtaining a multi-scale image model of the battery power system;

[0168] A second construction module 12 is configured to add a data-driven model, extract information from the real-time monitoring data of the battery power system, and add the extracted information to the multi-scale image model to construct a digital twin model of the battery power system;

[0169] The third construction module 13 is used to abstract the sensor data of the battery power system into a network topology graph through a graph neural network-long short-term memory neural network structure to perform multi-scale modeling of spatiotemporal relationships, and use the constructed digital twin model of the battery power system to perceive the operating status information of the battery power system;

[0170] The fourth construction module 14 is used to utilize the cross-modal deep semantic matching mechanism to construct a shared feature subspace between modalities for the multimodal data contained in the perceived operating status information, and to design an invariant graph regularization factor to establish a deep semantic matching fusion model for incomplete multimodal and multi-physical fields to reduce semantic deviation.

[0171] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0172] The embodiment of the present invention constructs a multi-scale digital twin model coupled with multiple physical fields, combines the data-driven method to realize dynamic model updates, and uses spatiotemporal feature fusion and cross-modal semantic matching technology to process multi-source heterogeneous data. Specifically, first, all the parameters and multi-scale data of the multi-physical fields (electrochemical field, temperature field, stress field) of the battery power system are collected to build a multi-level physical model from battery cells to the whole package; then, the real-time monitoring data is supplemented by the data-driven model to improve the accuracy of the model; then, the graph neural network-long short-term memory neural network structure is used to process the spatiotemporal relationship of the sensor data to achieve accurate perception of the operating status; finally, a shared feature subspace is established through a cross-modal semantic matching mechanism and a regularization factor is introduced to solve the problem of incomplete multimodal data. This technical concept breaks through the limitations of traditional models that rely only on apparent parameters and lack dynamic adaptability through hierarchical model construction and data fusion methods, and realizes all-round, high-precision simulation and real-time optimization of battery power systems. The embodiment of the present invention solves the technical defect that traditional methods cannot dynamically reflect the complex operating conditions of batteries through multi-physics field coupling modeling and real-time data-driven mechanism, realizes the accurate mapping and real-time optimization of the multi-scale operating status of the battery power system of new energy heavy-duty trucks, and significantly improves the system's energy utilization efficiency and dynamic response capabilities.

[0173] As one embodiment, the first building block is specifically configured to:

[0174] Cell-to-module scale modeling: At the cell scale, an electrochemical-thermal model is established based on the P2D model according to the energy conservation equation to obtain the cell-scale electrochemical-thermal model. At the cell scale, a thermal-mechanical coupling model of the power battery is constructed based on the thermal stress and thermal expansion behavior caused by temperature gradients during battery operation to obtain the cell-scale thermal-mechanical coupling model.

[0175] Module and full-pack scale modeling: Based on the cell-scale electrochemical-thermal model and the cell-scale thermal-mechanical coupling model, industrial modeling software is used to construct a 3D CAD model of the power battery. This accurately expresses its key structural parameters, the constraints and positioning relationships between cells and modules, and produces a 3D CAD model. Pixyz Studio is then used to perform geometric optimization to obtain the optimized geometric model.

[0176] Dynamic modeling: Using the optimized geometric model as input, the mechanical geometry, material parameters, force analysis, and velocity load information of the power battery pack, electric motor, inverter, and drive control unit are combined to establish a dynamic model that includes the mechanical geometry, material parameters, force analysis, and velocity load information.

[0177] As one embodiment, the second building block is specifically configured to:

[0178] Model system fusion: The multi-scale image model obtained in the multi-scale modeling step is integrated with the data-driven model to form a pre-fusion model system;

[0179] Real-time data processing: A data-driven approach is used to extract information from real-time monitoring data, obtaining the extracted information and supplementing it with the multi-scale image model in the pre-fusion model system to form a supplemented multi-scale image model. The real-time monitoring data includes sensor data from the physical layer and simulation data from the virtual layer.

[0180] Digital twin construction: Based on the supplemented multi-scale image model, a multi-scale digital twin model of the battery power system is constructed to achieve data-driven model fusion.

[0181] As one embodiment, the third building block is specifically configured to:

[0182] Data abstraction processing: sensor data is abstracted into a network topology diagram of the sensor data time series accumulation process. The connections between nodes are defined as sequential events to form a network topology diagram of the sensor data.

[0183] Dynamic graph construction: Based on the network topology of the sensor data, a time-series dynamic graph structure of battery-motor-brake characteristic information fusion is constructed to obtain a time-series dynamic graph including battery-motor-brake characteristics;

[0184] Feature learning processing: A graph neural network-long short-term memory neural network structure is used to integrate the structural information and time information of the temporal dynamic graph, a hierarchical feature aggregation method is designed, and feature aggregators are learned in different depth neighborhoods to realize the flow of information from higher depths to nodes and obtain state perception results.

[0185] As one embodiment, the fourth building block is specifically configured to:

[0186] Cross-modal modeling: Using a cross-modal deep semantic matching mechanism and multi-layer nonlinear correlations between multi-level and multi-scale modal data, we establish an incomplete cross-modal deep semantic matching fusion model to obtain an initial fusion model.

[0187] Feature space construction: constructing a shared feature subspace between modalities based on the initial fusion model, learning the sharing of incomplete cross-modal data, and forming a shared feature subspace;

[0188] Regularization processing: designing an invariant graph regularization factor in the shared feature subspace to ensure the local similarity characteristics of each modal data and obtain a regularized shared subspace;

[0189] Objective function establishment: A new objective function is established based on the regularized shared subspace to form a complete multimodal data fusion model to reduce semantic bias.

[0190] It should be noted that the system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the system embodiment provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without inventive work.

[0191] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A digital twin method for a battery power system of a new energy heavy truck, characterized in that: The steps include: Obtain basic data for the digital twin of the battery power system used in new energy heavy-duty trucks. This data includes full-factor parameter information under the multi-physics coupling state of the electrochemical field, temperature field, and stress field of the battery power system of new energy heavy-duty trucks, as well as multi-scale data from battery cells to modules, complete packs, controllers, and drive motors. Based on the basic data, an electrochemical-thermomechanical coupling model is constructed from the cell to module scale of the battery power system. A thermal-mechanical coupling model is constructed at the cell scale taking into account thermal stress and thermal expansion behavior. A geometric model is established and optimized at the module and whole package scales of the battery power system. At the same time, a dynamic model of the battery power system is established to achieve a comprehensive abstraction of the battery power system and obtain a multi-scale image model of the battery power system. Adding a data-driven model to extract information from the real-time monitoring data of the battery power system, supplementing the extracted information into the multi-scale image model, and constructing a digital twin model of the battery power system; Through the graph neural network-long short-term memory neural network structure, the sensor data of the battery power system is abstracted into a network topology graph for multi-scale modeling of spatiotemporal relationships, and the constructed digital twin model of the battery power system is used to perceive the operating status information of the battery power system; Utilizing the cross-modal deep semantic matching mechanism, a shared feature subspace between modalities is constructed for the multimodal data contained in the perceived operating status information, and an invariant graph regularization factor is designed to establish a deep semantic matching fusion model for incomplete multimodal and multi-physical fields to reduce semantic deviation.

2. The digital twin method for the battery power system of a new energy heavy truck according to claim 1 is characterized in that: Based on the basic data, an electrochemical-thermomechanical coupling model is constructed from the cell to module scale of the battery power system. A thermal-mechanical coupling model is constructed at the single cell scale considering thermal stress and thermal expansion behavior. A geometric model is established and optimized from the module and whole package scales of the battery power system. At the same time, a dynamic model of the battery power system is established to achieve a comprehensive abstraction of the battery power system and obtain a multi-scale image model of the battery power system, including: Cell-to-module scale modeling: At the cell scale, an electrochemical-thermal model is established based on the P2D model according to the energy conservation equation to obtain the cell-scale electrochemical-thermal model. At the cell scale, a thermal-mechanical coupling model of the power battery is constructed based on the thermal stress and thermal expansion behavior caused by temperature gradients during battery operation to obtain the cell-scale thermal-mechanical coupling model. Module and full-pack scale modeling: Based on the cell-scale electrochemical-thermal model and the cell-scale thermal-mechanical coupling model, industrial modeling software is used to construct a 3D CAD model of the power battery. This accurately expresses its key structural parameters, the constraints and positioning relationships between cells and modules, and produces a 3D CAD model. Pixyz Studio is then used to perform geometric optimization to obtain the optimized geometric model. Dynamic modeling: Using the optimized geometric model as input, the mechanical geometry, material parameters, force analysis, and velocity load information of the power battery pack, electric motor, inverter, and drive control unit are combined to establish a dynamic model that includes the mechanical geometry, material parameters, force analysis, and velocity load information.

3. The digital twin method for battery power system applied to new energy heavy trucks according to claim 1 is characterized in that: The data-driven model is added to extract information from the real-time monitoring data of the battery power system, and the extracted information is added to the multi-scale image model to construct a digital twin model of the battery power system, including: Model system fusion: The multi-scale image model obtained in the multi-scale modeling step is integrated with the data-driven model to form a pre-fusion model system; Real-time data processing: A data-driven approach is used to extract information from real-time monitoring data, obtaining the extracted information and supplementing it with the multi-scale image model in the pre-fusion model system to form a supplemented multi-scale image model. The real-time monitoring data includes sensor data from the physical layer and simulation data from the virtual layer. Digital twin construction: Based on the supplemented multi-scale image model, a multi-scale digital twin model of the battery power system is constructed to achieve data-driven model fusion.

4. The digital twin method for battery power system applied to new energy heavy trucks according to claim 1 is characterized in that: The sensor data of the battery power system is abstracted into a network topology graph through the graph neural network-long short-term memory neural network structure to perform multi-scale modeling of spatiotemporal relationships, and the constructed digital twin model of the battery power system is used to perceive the operating status information of the battery power system, including: Data abstraction processing: sensor data is abstracted into a network topology diagram of the sensor data time series accumulation process. The connections between nodes are defined as sequential events to form a network topology diagram of the sensor data. Dynamic graph construction: Based on the network topology of the sensor data, a time-series dynamic graph structure of battery-motor-brake characteristic information fusion is constructed to obtain a time-series dynamic graph including battery-motor-brake characteristics; Feature learning processing: A graph neural network-long short-term memory neural network structure is used to integrate the structural information and time information of the temporal dynamic graph, a hierarchical feature aggregation method is designed, and feature aggregators are learned in different depth neighborhoods to realize the flow of information from higher depths to nodes and obtain state perception results.

5. The digital twin method of the battery power system applied to new energy heavy trucks according to claim 1 is characterized in that: The cross-modal deep semantic matching mechanism is used to construct a shared feature subspace between modalities for the multimodal data contained in the perceived operating status information, and an invariant graph regularization factor is designed to establish a deep semantic matching fusion model for incomplete multimodal and multi-physical fields to reduce semantic deviation, including: Cross-modal modeling: Using a cross-modal deep semantic matching mechanism and multi-layer nonlinear correlations between multi-level and multi-scale modal data, we establish an incomplete cross-modal deep semantic matching fusion model to obtain an initial fusion model. Feature space construction: constructing a shared feature subspace between modalities based on the initial fusion model, learning the sharing of incomplete cross-modal data, and forming a shared feature subspace; Regularization processing: designing an invariant graph regularization factor in the shared feature subspace to ensure the local similarity characteristics of each modal data and obtain a regularized shared subspace; Objective function establishment: A new objective function is established based on the regularized shared subspace to form a complete multimodal data fusion model to reduce semantic bias.

6. A digital twin system of a battery power system for new energy heavy trucks, characterized by: include: An acquisition module is used to obtain basic data for the digital twin of the battery power system used in new energy heavy-duty trucks. The basic data includes: full-factor parameter information under the multi-physics coupling state of the electrochemical field, temperature field, and stress field of the battery power system of new energy heavy-duty trucks, as well as multi-scale data from battery cells to modules, complete packs, controllers, and drive motors; A first construction module is configured to construct an electrochemical-thermomechanical coupling model of the battery power system from the cell to the module scale based on the basic data, construct a thermal-mechanical coupling model at the cell scale taking into account thermal stress and thermal expansion behavior, establish and optimize a geometric model at the module and entire package scales of the battery power system, and simultaneously establish a dynamic model of the battery power system, thereby achieving a comprehensive abstraction of the battery power system and obtaining a multi-scale image model of the battery power system; A second construction module is configured to add a data-driven model, extract information from the real-time monitoring data of the battery power system, and add the extracted information to the multi-scale image model to construct a digital twin model of the battery power system; A third construction module is used to abstract the sensor data of the battery power system into a network topology graph through a graph neural network-long short-term memory neural network structure to perform multi-scale modeling of spatiotemporal relationships, and use the constructed digital twin model of the battery power system to perceive the operating status information of the battery power system; The fourth construction module is used to utilize the cross-modal deep semantic matching mechanism to construct a shared feature subspace between modalities for the multimodal data contained in the perceived operating status information, and to design an invariant graph regularization factor to establish a deep semantic matching fusion model for incomplete multimodal and multi-physical fields to reduce semantic deviation.

7. The digital twin system of the battery power system applied to new energy heavy trucks as claimed in claim 6 is characterized in that: The first building block is specifically used for: Cell-to-module scale modeling: At the cell scale, an electrochemical-thermal model is established based on the P2D model according to the energy conservation equation to obtain the cell-scale electrochemical-thermal model. At the cell scale, a thermal-mechanical coupling model of the power battery is constructed based on the thermal stress and thermal expansion behavior caused by temperature gradients during battery operation to obtain the cell-scale thermal-mechanical coupling model. Module and full-pack scale modeling: Based on the cell-scale electrochemical-thermal model and the cell-scale thermal-mechanical coupling model, industrial modeling software is used to construct a 3D CAD model of the power battery. This accurately expresses its key structural parameters, the constraints and positioning relationships between cells and modules, and produces a 3D CAD model. Pixyz Studio is then used to perform geometric optimization to obtain the optimized geometric model. Dynamic modeling: Using the optimized geometric model as input, the mechanical geometry, material parameters, force analysis, and velocity load information of the power battery pack, electric motor, inverter, and drive control unit are combined to establish a dynamic model that includes the mechanical geometry, material parameters, force analysis, and velocity load information.

8. The digital twin system of the battery power system applied to new energy heavy trucks as claimed in claim 6 is characterized in that: The second building block is specifically configured to: Model system fusion: The multi-scale image model obtained in the multi-scale modeling step is integrated with the data-driven model to form a pre-fusion model system; Real-time data processing: A data-driven approach is used to extract information from real-time monitoring data, obtaining the extracted information and supplementing it with the multi-scale image model in the pre-fusion model system to form a supplemented multi-scale image model. The real-time monitoring data includes sensor data from the physical layer and simulation data from the virtual layer. Digital twin construction: Based on the supplemented multi-scale image model, a multi-scale digital twin model of the battery power system is constructed to achieve data-driven model fusion.

9. The digital twin system of the battery power system applied to new energy heavy trucks as claimed in claim 6, characterized in that: The third building block is specifically configured to: Data abstraction processing: sensor data is abstracted into a network topology diagram of the sensor data time series accumulation process. The connections between nodes are defined as sequential events to form a network topology diagram of the sensor data. Dynamic graph construction: Based on the network topology of the sensor data, a time-series dynamic graph structure of battery-motor-brake characteristic information fusion is constructed to obtain a time-series dynamic graph including battery-motor-brake characteristics; Feature learning processing: A graph neural network-long short-term memory neural network structure is used to integrate the structural information and time information of the temporal dynamic graph, a hierarchical feature aggregation method is designed, and feature aggregators are learned in different depth neighborhoods to realize the flow of information from higher depths to nodes and obtain state perception results.

10. The digital twin system of the battery power system applied to new energy heavy trucks according to claim 6, characterized in that: The fourth building block is specifically configured to: Cross-modal modeling: Using a cross-modal deep semantic matching mechanism and multi-layer nonlinear correlations between multi-level and multi-scale modal data, we establish an incomplete cross-modal deep semantic matching fusion model to obtain an initial fusion model. Feature space construction: constructing a shared feature subspace between modalities based on the initial fusion model, learning the sharing of incomplete cross-modal data, and forming a shared feature subspace; Regularization processing: designing an invariant graph regularization factor in the shared feature subspace to ensure the local similarity characteristics of each modal data and obtain a regularized shared subspace; Objective function establishment: A new objective function is established based on the regularized shared subspace to form a complete multimodal data fusion model to reduce semantic bias.

Images