A method and system for monitoring the operating state of a lithium-ion battery electrolyte system

Abstract

This invention discloses a method and system for monitoring the operating status of a lithium-ion battery electrolyte system, comprising: firstly, acquiring a basic information set of system operating conditions, which includes battery operating data of the target battery and property data of the target electrolyte; then, extracting features from the basic information set to obtain at least two operating characterization vectors; for each operating characterization vector, extracting a coupled characterization vector formed by non-target operating characterization vectors based on a coupled dynamic evolution mechanism; subsequently, performing feature correlation modeling on the operating characterization vectors and coupled characterization vectors to obtain a corresponding first associated coupled characterization vector; finally, constructing an operating status correlation index based on each first associated coupled characterization vector, which accurately indicates the operating coupling state relationship between the target battery and the target electrolyte. This method improves the accuracy and reliability of monitoring through dynamic coupling analysis.

Description

Technical Field

[0001] This invention relates to the field of battery technology, and more specifically, to a method and system for monitoring the operating status of a lithium-ion battery electrolyte system. Background Technology

[0002] In the field of lithium-ion battery technology, the stable operating state of the electrolyte system is crucial to battery performance and safety. Existing monitoring methods typically analyze battery operating parameters or electrolyte property parameters independently, lacking an effective assessment of the dynamic coupling relationship between the battery and the electrolyte. This isolated analysis leads to inaccurate state monitoring, thereby affecting battery life and safety. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for monitoring the operating status of a lithium-ion battery electrolyte system.

[0004] In a first aspect, embodiments of the present invention provide a method for monitoring the operating status of a lithium-ion battery electrolyte system, the method comprising: Acquire a system operating condition basic information set, which includes battery operation data of the target battery and target electrolyte property data of the target electrolyte. The battery operation data includes lithium-ion battery operation parameters reflecting at least one of lithium-ion insertion and extraction kinetics, lithium metal deposition tendency, solid electrolyte interface phase state evolution characteristics, and high-voltage operation safety margin. The target electrolyte property data includes lithium-ion battery electrolyte parameters characterizing at least one of lithium-ion solvation structure, lithium-ion migration efficiency, lithium salt concentration, and decomposition by-product concentration. Feature extraction is performed on the system operating condition basic information set to obtain at least two operating characteristic vectors; For each of the aforementioned operational representation vectors, a coupling representation vector corresponding to the operational representation vector is extracted, and feature association modeling is performed on the operational representation vector and the coupling representation vector to obtain a first associated coupling representation vector corresponding to the operational representation vector. The coupling representation vector is a feature vector obtained by feature extraction based on a coupling dynamic evolution mechanism from non-target operational representation vectors other than the operational representation vector among the at least two operational representation vectors. Based on the first associated coupling representation vector associated with each of the aforementioned operational representation vectors, an operational state association index is obtained, which indicates the operational coupling state relationship between the target battery and the target electrolyte.

[0005] In a second aspect, embodiments of the present invention provide a server system, including a server, the server being used to execute the method described in the first aspect.

[0006] Compared to existing technologies, the beneficial effects provided by this invention include: The invention discloses a method and system for monitoring the operating status of a lithium-ion battery electrolyte system, comprising: firstly, acquiring a basic information set of system operating conditions, which includes battery operating data of the target battery and property data of the target electrolyte; then, extracting features from the basic information set to obtain at least two operating characterization vectors; for each operating characterization vector, extracting a coupled characterization vector formed by non-target operating characterization vectors based on a coupled dynamic evolution mechanism; subsequently, performing feature correlation modeling on the operating characterization vectors and coupled characterization vectors to obtain a corresponding first associated coupled characterization vector; finally, constructing an operating status correlation index based on each first associated coupled characterization vector, which accurately indicates the operating coupling state relationship between the target battery and the target electrolyte. This method improves the accuracy and reliability of monitoring through dynamic coupling analysis. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating the steps of a lithium-ion battery electrolyte system operation status monitoring method provided in an embodiment of the present invention. Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0010] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0011] In order to solve the technical problems mentioned in the background art Figure 1 This is a flowchart illustrating the method for monitoring the operating status of a lithium-ion battery electrolyte system provided in this embodiment. The following is a detailed description of this method for monitoring the operating status of a lithium-ion battery electrolyte system.

[0012] Step S201: Obtain the system operating condition basic information set. The system operating condition basic information set includes the battery operation data of the target battery and the target electrolyte property data of the target electrolyte. The battery operation data includes lithium-ion battery operation parameters reflecting at least one of lithium-ion insertion and extraction kinetics, lithium metal deposition tendency, solid electrolyte interface phase state evolution characteristics, and high-voltage operation safety margin. The target electrolyte property data includes lithium-ion battery electrolyte parameters characterizing at least one of lithium-ion solvation structure, lithium-ion migration efficiency, lithium salt concentration, and decomposition by-product concentration. Step S202: Extract features from the system operating condition basic information set to obtain at least two operating characteristic vectors; Step S203: For each of the running representation vectors, extract the coupling representation vector corresponding to the running representation vector, and perform feature association modeling on the running representation vector and the coupling representation vector to obtain the first associated coupling representation vector corresponding to the running representation vector. The coupling representation vector is a feature vector obtained by feature extraction based on the coupling dynamic evolution mechanism of non-target running representation vectors other than the running representation vector among the at least two running representation vectors. Step S204: Based on the first associated coupling representation vector associated with each of the operation representation vectors, an operation state association index is obtained, which indicates the operation coupling state relationship between the target battery and the target electrolyte.

[0013] In this embodiment of the invention, the core of the lithium-ion battery electrolyte system operation status monitoring method disclosed herein lies in achieving accurate assessment of the coupling state of battery and electrolyte operation through multi-dimensional data acquisition and deep coupling modeling. The method is executed by a server, and the specific execution process is as follows: First, the server acquires a basic system operating condition information set, which serves as the foundational data source for subsequent analysis. This set comprises two main categories: battery operating data for the target battery and target electrolyte property data for the target electrolyte. The battery operating data includes parameters reflecting the kinetics of lithium-ion insertion and extraction, such as the potential difference between oxidation and reduction peaks and the peak current density ratio in cyclic voltammetry curves acquired using an in-situ electrochemical workstation, as well as the charge transfer resistance and Warburg impedance coefficient extracted from electrochemical impedance spectroscopy. These parameters directly correlate with the migration rate of lithium ions in the lattice of the positive and negative electrode materials and the interfacial reactivity. The data also includes data reflecting the tendency of lithium metal deposition, such as the time of lithium dendrite formation and growth rate observed by in-situ optical microscopy under low-temperature charging conditions, combined with the negative... The data also includes the overpotential of the electrode surface, with higher overpotentials indicating a greater risk of lithium metal deposition; data on the evolution characteristics of the solid electrolyte interphase (SEI) phase are also included, such as the intensity ratio of characteristic peaks of LiF, CO, and C=O bonds in the SEI film obtained by X-ray photoelectron spectroscopy analysis, and the exothermic peak temperature and heat release of SEI decomposition measured by differential scanning calorimetry, to assess the chemical and thermal stability of the SEI film; in addition, high-voltage operation safety margin data are also indispensable, such as the thermal runaway trigger temperature, maximum temperature rise rate, and capacity retention rate after 500 cycles of the battery at a full charge of 4.5V. The target electrolyte property data includes parameters characterizing the lithium-ion solvation structure, such as the area ratio of the characteristic peaks of the solvation sheath to those of free ions in Raman spectroscopy, and the chemical shift of solvent molecules in 1H NMR spectroscopy; the product of ionic conductivity and lithium-ion transference number, characterizing lithium-ion migration efficiency; the electrolyte density (calculated from a preset density-concentration calibration curve) and the lithium salt anion concentration directly measured by ion chromatography, characterizing lithium salt concentration; and data characterizing the concentration of decomposition byproducts, such as the concentrations of byproducts like HF and CO2 detected by gas chromatography-mass spectrometry, and the water content determined by high-performance liquid chromatography. The server collects these raw data in real-time or periodically through sensor arrays (such as temperature, voltage, and current sensors) deployed within the battery pack, external electrochemical workstations, spectrometers, and laboratory analysis data interfaces. After preprocessing (such as baseline correction, noise filtering, and data standardization), the data is integrated into a structured set of basic system operating information.

[0014] Next, the server extracts features from the system's basic operating condition information set to obtain at least two operational characterization vectors. The purpose of this step is to transform the high-dimensional, heterogeneous raw data into low-dimensional, interpretable feature vectors for subsequent coupled modeling. Specifically, the server employs a feature extraction strategy that combines domain knowledge guidance with data-driven approaches. For battery operating data, key features with clear physical meaning are selected, such as peak potential difference and charge transfer resistance in lithium-ion insertion / deintercalation kinetics, dendrite growth rate and overpotential in lithium deposition tendency, LiF characteristic peak intensity ratio and decomposition heat of release in the SEI state, and thermal runaway temperature and capacity retention in high-voltage safety margin. For electrolyte property data, core parameters such as solvation structure peak area ratio, ion migration efficiency, lithium salt concentration, and HF concentration are selected. Subsequently, principal component analysis (PCA) is used to reduce the dimensionality of the selected high-dimensional features, retaining principal components with a cumulative contribution rate exceeding 95%, and nonlinear feature fusion is performed using a multilayer perceptron (MLP) to ultimately generate multiple operational characterization vectors. For example, in a certain implementation scenario, the server generates four operational characterization vectors: a kinetic characterization vector (including features such as peak potential difference, charge transfer resistance, and ion migration efficiency), a deposition tendency characterization vector (including features such as dendrite growth rate, overpotential, and lithium salt concentration), an SEI state characterization vector (including features such as LiF peak intensity ratio, decomposition heat of release, and HF concentration), and a high-voltage safety characterization vector (including features such as thermal runaway temperature, capacity retention, and CO2 concentration). Each operational characterization vector is Z-score normalized to ensure that the dimensions of each feature are consistent, facilitating subsequent coupling calculations between vectors.

[0015] Then, for each operational representation vector, the server extracts its corresponding coupling representation vector and performs feature correlation modeling on the operational representation vector and the coupling representation vector to obtain the first associated coupling representation vector corresponding to that operational representation vector. Here, the coupling representation vector refers to the feature vector obtained by performing feature extraction based on a coupling dynamic evolution mechanism on at least two operational representation vectors other than the target operational representation vector. It reflects the comprehensive coupling influence of other operational dimensions on the target dimension in the system. Specifically, for each target operational representation vector (e.g., kinetic representation vector V1), the server first integrates the features of all other non-target operational representation vectors (e.g., deposition tendency representation vector V2, SEI state representation vector V3, and high-voltage safety representation vector V4). For example, it merges V2, V3, and V4 into a basic operational coupling representation using a weighted average. The weight coefficients are set according to the physical correlation between each non-target vector and the target vector (e.g., the weight of deposition tendency on kinetics is higher than that on high-voltage safety, so V2 has a larger weight). Next, feature extraction based on a coupling dynamic evolution mechanism is performed on this basic operational coupling representation to generate the coupling representation vector. This dynamic evolution mechanism for coupling typically includes nonlinear transformation, attention weight generation, and weighted fusion steps: First, a first nonlinear transformation (e.g., through a neural network layer with ReLU activation) is performed on the basic representation of the operational coupling to generate an attention weight vector that evaluates the contribution of each feature dimension to the overall coupling state. Then, the attention weight vector is fused with the basic representation of the operational coupling dimension through a weighted fusion step-by-step to highlight key coupling features. Finally, a second nonlinear transformation and feature dimension reduction (e.g., through a fully connected layer with tanh activation) are used to extract deep coupling features, resulting in a coupling representation vector (e.g., C1). After obtaining the target operational representation vector (V1) and its corresponding coupling representation vector (C1), the server performs feature association modeling on the two, specifically using a parameter-by-parameter coupling calculation method. For example, the corresponding dimensional features of V1 and C1 are multiplied, and their weighted sum (with weights set by a preset coupling strength coefficient) is superimposed to obtain the first associated coupling representation vector (A1). This vector not only retains the feature information of the target operational representation vector itself but also integrates the coupling influence transmitted by other operational dimensions through the dynamic evolution mechanism, achieving a preliminary association between a single dimension and the overall system state. This process will be executed one by one for each running characterization vector. For example, for the sedimentation tendency characterization vector V2, the server will take V1, V3, and V4 as non-target running characterization vectors, repeat the above feature integration, coupling dynamic evolution to extract the coupling characterization vector C2, and couple it with V2 to calculate the first associated coupling characterization vector A2. This process will continue in the same way until the first associated coupling characterization vector set corresponding to all running characterization vectors is finally obtained.

[0016] Finally, based on the first correlation coupling representation vector associated with each operational representation vector, the server obtains an operational state correlation index, which is used to quantitatively indicate the operational coupling relationship between the target battery and the target electrolyte. To comprehensively capture the complex coupling effects in the system, the server further integrates multi-source correlation information based on the first correlation coupling representation vector. Specifically, this includes: First, generating a second associated coupling representation vector. For each operational representation vector, it is modeled as a pairwise direct association with other operational representation vectors (e.g., V1 and V2, V1 and V3, etc.). This is achieved through simple coupling methods such as element-wise multiplication, reflecting the direct interaction between dimensions in the system. Second, generating a deep associated coupling representation vector. The server integrates the features of all operational representation vectors (e.g., concatenating them column-wise into a high-dimensional integrated representation vector). Then, this integrated representation vector is processed by a deep coupling dynamic evolution mechanism. This mechanism is also based on a multi-level evolution weighted by dynamic feature importance. A global attention weight vector is generated through a first nonlinear transformation. After weighted fusion of the integrated representation vector, a second nonlinear transformation and dimensional reduction are performed to extract global, deep coupling features across all operational representation vectors. Subsequently, the server concatenates all the first associated coupling representation vectors, the second associated coupling representation vectors, and the deep associated coupling representation vectors to obtain the integrated coupling representation vector. This result contains multi-scale coupling information of the system from local to global and from shallow to deep layers. Finally, the server performs coupling modeling based on the coupling representation vector integration results. This coupling modeling is a multi-level coupling modeling used to reflect the coupling adaptability of the target battery to the target electrolyte, typically implemented using a multilayer perceptron (MLP). The input layer of the MLP is the coupling representation vector integration results, and feature depth mining is performed through nonlinear transformations in the hidden layers (such as the ReLU activation function). The output layer outputs a normalized operating state correlation index (typically ranging from [0,1]) through the Sigmoid activation function. The closer the index is to 1, the better the operating coupling state of the battery and electrolyte, the better the coupling adaptability, and the lower the probability of performance degradation or safety risks. Conversely, a lower index indicates poor coupling state, requiring electrolyte parameter adjustment or battery maintenance. For example, in the monitoring scenario of a new energy vehicle battery system, the server calculates the operating status correlation index as 0.85 through the above steps. Combined with the preset operating threshold judgment conditions (such as the threshold being set to 0.7), it determines that the current battery and electrolyte coupling state is good. If the index drops to 0.65 in subsequent cycles, a system warning is triggered, indicating that the electrolyte needs to be optimized or the internal state of the battery needs to be checked.

[0017] Through the above steps, the server achieves comprehensive, dynamic, and in-depth monitoring of the operating status of the lithium-ion battery electrolyte system, providing key quantitative basis for battery system safety management, life prediction, and electrolyte formulation optimization.

[0018] In this embodiment of the invention, the extraction of the coupling representation vector corresponding to the running representation vector can be performed through the following example.

[0019] Feature integration is performed on the non-target running representation vectors (excluding the running representation vectors) from the at least two running representation vectors to obtain the running coupling basic representation corresponding to the running representation vectors; Feature extraction based on a coupling dynamic evolution mechanism is performed on the basic representation of the operation coupling to obtain the coupling representation vector corresponding to the operation representation vector; The step of performing feature association modeling on the running representation vector and the coupled representation vector to obtain the first associated coupled representation vector corresponding to the running representation vector includes: The parameter-by-parameter coupling calculation is performed on the running representation vector and the corresponding coupling representation vector to obtain the first associated coupling representation vector corresponding to the running representation vector.

[0020] In an embodiment of the present invention, exemplarily, Taking the server monitoring process of a new energy vehicle lithium-ion battery system (NCM811 / graphite system, electrolyte is 1mol / LLiPF6-EC / DEC (volume ratio 3:7)) as an example, the server has generated 4 operation characterization vectors: kinetic characterization vector V1=[0.82,-0.35,0.12,0.68,0.75] (5-dimensional, reflecting insertion / deintercalation kinetics), deposition tendency characterization vector V2=[1.21,-0.53,-0.28,0.41,-0.15] (reflecting lithium dendrite risk), SEI state characterization vector V3=[0.93,0.72,-0.45,-0.31,0.58] (reflecting SEI stability), and high voltage safety characterization vector V4=[0.65,0.88,-0.62,0.59,-0.23] (reflecting high voltage safety margin). Now, using V1 as the target representation vector, we extract its coupled representation vector C1 and generate the first associated coupled representation vector A1. The specific process is as follows: The server first performs feature integration on the non-target operational representation vectors (V2, V3, V4) to obtain the basic representation of operational coupling. Based on the physical correlation between each vector and V1 (deposition tendency has the greatest impact on dynamics, followed by SEI state, and high voltage safety is lower), weights α2=0.4, α3=0.3, and α4=0.3 are set, and the vectors are integrated by weighted average: M_avg=0.4×V2+0.3×V3+0.3×V4. Substituting the specific values, we get: M_avg first dimension = 0.4×1.21+0.3×0.93+0.3×0.65=0.484+0.279+0.195=0.958, second dimension = 0.4×(-0.53)+0.3×0.72+0.3×0.88=-0.212+0.216+0.264=0.268, and subsequent dimensions are calculated in sequence, finally obtaining the basic representation of runtime coupling M_avg=[0.958,0.268,-0.389,0.460,0.120].

[0021] Next, the server performs feature extraction on M_avg based on a coupled dynamic evolution mechanism to generate a coupled representation vector C1. This mechanism involves a multi-level evolution with dynamic feature importance weighting: first, M_avg undergoes a first nonlinear transformation through a ReLU-activated neural network layer to generate an attention weight vector W=[0.22,0.35,0.10,0.25,0.08] (the weights of each dimension sum to 1, with the second dimension contributing the most due to its corresponding dendrite growth rate); then, W is fused with M_avg dimension by dimension with weights: the first dimension of the weighted M_avg = 0.22×0. 958≈0.211, the second dimension = 0.35×0.268≈0.094, and so on, to obtain the weighted vector = [0.211,0.094,-0.039,0.115,0.010]; finally, the second nonlinear transformation and dimension reduction are performed through the fully connected layer activated by tanh to extract the deep coupling features, and the coupling representation vector C1 = [0.192,0.087,-0.035,0.106,0.009].

[0022] Finally, the server performs parameter-by-parameter coupling calculations on V1 and C1 to generate the first associated coupling representation vector A1. The coupling formula of "item-by-item product + weighted sum" is adopted: A1[i]=V1[i]×C1[i]+λ×(V1[i]+C1[i]) (λ=0.1 is the coupling strength coefficient). Taking the first dimension as an example: A1[1]=0.82×0.192+0.1×(0.82+0.192)=0.157+0.1×1.012=0.157+0.101=0.258; the second dimension: A1[2]=(-0.35)×0.087+0.1×(-0.35+0.087)=-0.030+0.1×(-0.263)=-0.030-0.026=-0.056; the remaining dimensions are calculated in sequence, and finally the first correlation coupling representation vector A1=[0.258,-0.056,0.008,0.156,0.098] is obtained. This vector integrates the dynamic coupling influence of dynamic characteristics and other dimensions of the system, and provides the core basis for the subsequent calculation of the correlation index of the operating state.

[0023] In this embodiment of the invention, the process of obtaining the operation status correlation index based on the first correlation coupling representation vector associated with each of the operation representation vectors can be implemented through the following example.

[0024] For each of the aforementioned operational representation vectors, feature association modeling is performed on each operational representation vector and each associated operational representation vector to obtain a second associated coupling representation vector corresponding to the operational representation vector. The associated operational representation vector is any operational representation vector among the at least two operational representation vectors excluding the operational representation vector. The first associated coupled representation vectors associated with each of the aforementioned operational representation vectors and the second associated coupled representation vectors associated with each of the aforementioned operational representation vectors are integrated to obtain the coupled representation vector integration result. Based on the integration result of the coupling characterization vector, the operational coupling state relationship between the target battery and the target electrolyte is determined, and the operational state correlation index is obtained.

[0025] In an embodiment of the present invention, taking the server monitoring of a lithium-ion battery system (NCM811 / graphite system) for a new energy vehicle as an example, the server has obtained the following runtime characterization vectors: V1=[0.82,-0.35,0.12,0.68,0.75] (kinetics), V2=[1.21,-0.53,-0.28,0.41,-0.15] (lithium deposition tendency), V3=[0.93,0.72,-0.45,-0.31,0.58] (SEI state), and V4=[0.65,0.88, -0.62, 0.59, -0.23] (high voltage safety), and generate the corresponding first associated coupling representation vectors A1=[0.258, -0.056, 0.008, 0.156, 0.098], A2=[0.185, 0.221, -0.072, 0.103, 0.065], A3=[0.210, 0.153, 0.198, -0.084, 0.122], A4=[0.167, 0.205, -0.110, 0.141, 0.089] (5 dimensions each). The server generates the running status associated indicators according to the following steps: Generate the second correlation coupling representation vector: For each runtime representation vector, the server performs feature association modeling on it with any other associated runtime representation vector (excluding itself as a single vector), generating a second associated coupling representation vector (using term-by-term product to highlight direct pairwise coupling). Taking V1 as an example, its associated runtime representation vectors are V2, V3, and V4, generating: -B 12 (V1 and V2 are coupled): Term-by-term product yields [0.82×1.21,(-0.35)×(-0.53),0.12×(-0.28),0.68×0.41,0.75×(-0.15)]≈[0.992,0.186,-0.034,0.279,-0.112]; -B 13 (V1 and V3 coupling): [0.82×0.93,(-0.35)×0.72,0.12×(-0.45),0.68×(-0.31),0.75×0.58]≈[0.763,-0.252,-0.054,-0.211,0.435]; -B 14 (V1 and V4 coupling): [0.82×0.65,(-0.35)×0.88,0.12×(-0.62),0.68×0.59,0.75×(-0.23)]≈[0.533,-0.308,-0.074,0.401,-0.173].

[0026] Similarly, V2 generates B 21(V2 and V1), B 23 (V2 and V3), B 24 (V2 and V4); V3 generates B 31 B 32 B 34 V4 generates B 41 B 42 B 43 There are a total of 12 second-related coupling representation vectors (each with 5 dimensions).

[0027] Coupling representation vector integration: The server concatenates the first associated coupling representation vector (A1-A4, a total of 4 × 5 dimensions = 20 dimensions) and the second associated coupling representation vector (12 × 5 dimensions = 60 dimensions) column by column to obtain the coupling representation vector integration result Z (80-dimensional vector), which covers all features of the dynamic coupling between a single dimension and the system as a whole (A series) and the direct interaction between dimensions (B series).

[0028] Determine the relevant indicators for operational status: The server inputs Z into a pre-trained multilayer perceptron (MLP) model (80-dimensional input, 32 / 16-dimensional hidden layers, and sigmoid activation in the output layer). The model learns the mapping relationship between coupling features in the sample data and measured coupling states (such as cycle life and thermal runaway risk), and outputs an operational state correlation index S. Substituting Z into the current Z, we get S=0.89 (∈[0,1]), indicating that the target battery and electrolyte have a good operational coupling state, and the dynamics, lithium deposition, SEI, and high-voltage safety dimensions are well-coordinated, resulting in stable system operation.

[0029] In this embodiment of the invention, the coupling representation vector integration of the first associated coupling representation vector and the second associated coupling representation vector associated with each of the running representation vectors to obtain the coupling representation vector integration result can be implemented through the following example.

[0030] The aforementioned operational representation vectors are integrated to obtain an integrated representation vector; A deep coupling dynamic evolution mechanism is applied to the integrated representation vector to obtain a deep correlation coupling representation vector; The first associated coupling representation vector, the second associated coupling representation vector, and the deep associated coupling representation vector are integrated to obtain the coupling representation vector integration result.

[0031] In an embodiment of the present invention, taking the server monitoring of a lithium-ion battery system (NCM811 / graphite system) for a new energy vehicle as an example, the server has acquired the following operational characterization vectors: V1=[0.82,-0.35,0.12,0.68,0.75] (kinetics, 5-dimensional), V2=[1.21,-0.53,-0.28,0.41,-0.15] (lithium deposition tendency, 5-dimensional), V3=[0.93,0.72,-0.45,-0.31,0.58] (SEI state, 5-dimensional), V4=[0.65,0.88,-0.62,0.59,-0.23] (high voltage safety, 5-dimensional), the first associated coupling characterization vectors A1-A4 (each 5-dimensional), and the second associated coupling characterization vectors B series (e.g., B...). 12 =[0.992,0.186,-0.034,0.279,-0.112], B 13 =[0.763,-0.252,-0.054,-0.211,0.435] etc., a total of 12, each with 5 dimensions. The server integrates the coupled representation vectors according to the following steps: Generate integrated representation vectors: The server performs feature ensemble on the four runtime representation vectors (V1~V4), concatenating them column-wise to form the ensemble representation vector. Since each of V1~V4 contains 5-dimensional features, the concatenated vector yields a 20-dimensional ensemble representation vector V_all. V_all=[0.82,-0.35,0.12,0.68,0.75,1.21,-0.53,-0.28,0.41,-0.15,0.93,0.72,-0.45,-0.31,0.58,0.65,0.88,-0.62,0.59,-0.23] Generate deep association and coupling representation vectors: The server executes a deep-coupled dynamic evolution mechanism on V_all (a multi-level evolution based on dynamic feature importance weighting): 1. First nonlinear transformation to generate attention weights: V_all is transformed by a ReLU activated neural network layer to generate a 20-dimensional attention weight vector W (the weights of each dimension sum to 1). Among them, the dimensions related to lithium deposition rate (second dimension of V2), LiF ratio in SEI (first dimension of V3), and thermal runaway temperature (second dimension of V4) have higher weights, W=[0.05,0.08,0.03,0.06,0.04,0.07,0.12,0.02,0.05,0.03,0.10,0.06,0.04,0.03,0.05,0.04,0.09,0.03,0.05,0.03].

[0032] 2. Weighted fusion: W and V_all are weighted dimension by dimension to obtain the weighted integrated representation vector (20 dimensions). For example, the second dimension (V1 second dimension, -0.35): 0.08×(-0.35)=-0.028; the seventh dimension (V2 second dimension, -0.53): 0.12×(-0.53)=-0.064. The overall weighted vector highlights the key coupling features.

[0033] 3. Second nonlinear transformation and dimensionality reduction: The 20-dimensional weighted vector is reduced to a 5-dimensional deep correlation coupling representation vector D=[0.22,0.18,-0.09,0.15,0.11] through a fully connected layer activated by tanh.

[0034] The fusion yields the integrated result of the coupled representation vectors: The server concatenates A1-A4 (4×5=20 dimensions), B series (12×5=60 dimensions), and D (5 dimensions) column by column to obtain the coupling representation vector integration result Z (20+60+5=85 dimensions), which covers the dynamic coupling (A) between a single dimension and the system as a whole, the direct coupling between pairs (B), and the global deep coupling (D) features, providing full input for subsequent calculation of operational status correlation indicators.

[0035] In this embodiment of the invention, the fusion of the first associated coupling representation vector, the second associated coupling representation vector, and the deep associated coupling representation vector associated with each of the running representation vectors to obtain the coupling representation vector integration result can be implemented through the following example.

[0036] The first associated coupling representation vector, the second associated coupling representation vector, and the deep associated coupling representation vector associated with each of the aforementioned operational representation vectors are concatenated to obtain the integrated result of the coupling representation vectors.

[0037] In this embodiment of the invention, taking the server monitoring of a new energy vehicle lithium-ion battery system (NCM811 / graphite system) as an example, the server has obtained: The first associated coupling representation vectors are: A1=[0.258,-0.056,0.008,0.156,0.098], A2=[0.185,0.221,-0.072,0.103,0.065], A3=[0.210,0.153,0.198,-0.084,0.122], and A4=[0.167,0.205,-0.110,0.141,0.089] (4 in total, each with 5 dimensions, for a total of 20 dimensions). Second correlation coupling representation vector: B 12=[0.992,0.186,-0.034,0.279,-0.112] (V1 and V2 are coupled), B 13 =[0.763,-0.252,-0.054,-0.211,0.435] (V1 and V3), B 14 =[0.533,-0.308,-0.074,0.401,-0.173] (V1 and V4), B 21 =[0.208,0.195,-0.042,0.121,0.087] (V2 and V1), B 23 =[1.125,-0.382,-0.126,0.127,0.093] (V2 and V3), B 24 =[0.787,-0.466,-0.174,0.242,0.035] (V2 and V4), B 31 =[0.239,-0.113,0.011,0.083,0.089] (V3 and V1), B 32 =[1.125,-0.382,-0.126,0.127,0.093] (V3 and V2), B 34 =[0.605,0.634,-0.279,-0.183,0.133] (V3 and V4), B 41 =[0.137,-0.062,0.007,0.116,0.068] (V4 and V1), B 42 =[0.202,0.242,-0.080,0.115,0.047] (V4 and V2), B 43 =[0.205,0.137,0.178,-0.075,0.099] (V4 and V3) (12 in total, 5 dimensions each, for a total of 60 dimensions); Deeply correlated coupling representation vector: D=[0.22,0.18,-0.09,0.15,0.11] (5-dimensional).

[0038] The server concatenates vectors in the order of "first associated coupling representation vector → second associated coupling representation vector → deep associated coupling representation vector": 1. Concatenate the A series: Concatenate A1, A2, A3, and A4 sequentially to form a 20-dimensional vector segment: [0.258, -0.056, 0.008, 0.156, 0.098, 0.185, 0.221, -0.072, 0.103, 0.065, 0.210, 0.153, 0.198, -0.084, 0.122, 0.167, 0.205, -0.110, 0.141, 0.089]; 2. Assembling Series B: Assembling Series B sequentially. 12 To B 43 This forms a 60-dimensional vector segment, for example, the first segment is [0.992,0.186,-0.034,0.279,-0.112,0.763,-0.252,-0.054,-0.211,0.435,...] (which contains 12 5-dimensional vectors in total). 3. Concatenate D: Finally, concatenate D to supplement the 5-dimensional vector segment [0.22, 0.18, -0.09, 0.15, 0.11].

[0039] Ultimately, the server obtains the integrated result Z of the 85-dimensional coupling representation vector. This vector integrates all the features of the system's local dynamic coupling (A series), pairwise direct coupling (B series), and global deep coupling (D), providing complete input for subsequent calculation of operational status correlation indicators.

[0040] In this embodiment of the invention, the determination of the operational coupling state relationship between the target battery and the target electrolyte based on the integration result of the coupling characterization vector, and the obtaining of the operational state correlation index, can be implemented through the following example.

[0041] Based on the integration result of the coupling representation vector, coupling modeling is performed to obtain the operating state correlation index. The coupling modeling is a multi-level coupling modeling used to reflect the coupling adaptability of the target battery to the target electrolyte. The method further includes: when the operating status correlation index meets the operating threshold determination condition, outputting an operating medium matching command corresponding to the target electrolyte to the target battery.

[0042] In an embodiment of the present invention, for example, taking the server monitoring of a new energy vehicle lithium-ion battery system (NCM811 / graphite system, electrolyte is 1mol / LLiPF6-EC / DEC (3:7 volume ratio)) as an example, the server has obtained the coupling characterization vector integration result Z (85 dimensions, including the splicing result of the first associated coupling characterization vector A1-A4, the second associated coupling characterization vector B series and the deep associated coupling characterization vector D).

[0043] Multi-level coupling modeling based on the integrated results of coupling representation vectors: The server employs multi-level coupling modeling (Multilayer Perceptron, MLP) to process Z to reflect the coupling compatibility between the battery and the electrolyte. This MLP model is a pre-trained model (training data includes 1000 sets of Z under different coupling states and corresponding measured coupling compatibility scores), with the following structure: an 85-dimensional input layer (corresponding to the dimension of Z), two hidden layers (32-dimensional and 16-dimensional, with ReLU activation function), and a 1-dimensional output layer (operating state correlation index S, with Sigmoid activation function, output range [0,1]). The server inputs the 85-dimensional vector Z into the MLP, and the model extracts cross-dimensional coupling features through deep nonlinear transformations, calculating the operating state correlation index S=0.89 (the closer S is to 1, the better the coupling compatibility).

[0044] Operating threshold determination and output of operating medium matching instructions: The server's preset operating threshold condition is "S≥0.7" (indicating good coupling compatibility, no adjustment of electrolyte parameters is needed). Currently, S=0.89>0.7, meeting the condition. The server then generates an operating medium matching command, which includes: confirming good operating coupling between the target battery and the current electrolyte (lithium salt concentration 1mol / L, solvent EC / DEC volume ratio 3:7, moisture content <20ppm), stable lithium-ion solvation structure (Raman peak area ratio 0.82), migration efficiency 0.75, SEI film thermal stability meeting standards (decomposition deheating peak temperature 128℃), and low lithium deposition risk (dendritic growth rate <0.1μm / h). This command is sent to the Battery Management System (BMS) via the CAN bus and simultaneously displayed on the vehicle terminal, prompting "Electrolyte and battery operating status matched; it is recommended to maintain the current charge / discharge strategy (1C rate, voltage range 3.0-4.2V)."

[0045] Through the above process, the server achieves quantitative assessment and dynamic command output of battery-electrolyte coupling compatibility, ensuring that the system operates in the optimal coupling state.

[0046] In this embodiment of the invention, the method is implemented through an electrolyte coupling monitoring model, which includes a running feature normalization unit and a first multidimensional running correlation modeling unit. The first multidimensional running correlation modeling unit is connected to the running feature normalization unit, and the first multidimensional running correlation modeling unit includes a first feature integration network, a first multi-level running evolution unit, and a first parameter coupling response unit, which are sequentially cascaded. The process of extracting features from the system's basic operating condition information set to obtain at least two operational representation vectors can be implemented through the following example.

[0047] The system operating condition basic information set is input into the operation feature normalization unit to obtain the at least two operation characterization vectors; The extraction of the coupling representation vector corresponding to each of the running representation vectors can be performed through the following example.

[0048] For each of the aforementioned operational representation vectors, the non-target operational representation vectors other than the operational representation vectors from the at least two operational representation vectors are input into the first feature integration network to obtain the operational coupling basic representation corresponding to the operational representation vector; the operational coupling basic representation corresponding to the operational representation vector is input into the first multi-level operational evolution unit to obtain the coupling representation vector corresponding to the operational representation vector. The step of performing feature association modeling on the running representation vector and the coupling representation vector to obtain the first associated coupling representation vector corresponding to the running representation vector includes: inputting each of the running representation vectors and the coupling representation vector corresponding to the running representation vector into the first parameter coupling response unit to obtain the first associated coupling representation vector corresponding to the running representation vector.

[0049] In an embodiment of the present invention, for example, taking the server monitoring process of a new energy vehicle lithium-ion battery system (NCM811 / graphite cathode, electrolyte is 1mol / LLiPF6-EC / DEC (3:7 volume ratio)) as an example, the electrolyte coupling monitoring model deployed on the server includes an operation feature normalization unit and a first multi-dimensional operation association modeling unit (the latter is composed of a first feature integration network, a first multi-level operation evolution unit, and a first parameter coupling response unit in sequential cascade).

[0050] The system operating condition basic information set is input into the operation feature normalization unit to generate the operation representation vector: The server collects basic system operating information through the Battery Management System (BMS) and external testing equipment, including: battery operating data (lithium-ion insertion / extraction kinetics: cyclic voltammetry peak potential difference 0.12V, charge transfer resistance 85Ω; lithium metal deposition tendency: lithium dendrite appearance time at -10℃ charging 420s, growth rate 0.08μm / h; SEI state: LiF characteristic peak intensity ratio in XPS 0.68, decomposition deheating peak temperature 128℃; high voltage safety margin: 4.5V full-charge thermal runaway temperature 215℃, capacity retention after 500 cycles 89%); electrolyte property data (lithium-ion solvation structure: Raman spectroscopy solvation sheath peak area ratio 0.75, ¹H-NMR chemical shift shift 0.32ppm; migration efficiency: ionic conductivity 1.2×10⁻⁶). -²S / cm, migration number 0.42, migration efficiency index 0.00504; lithium salt concentration: density 1.12 g / cm³ corresponding to 1.02 mol / L; decomposition byproduct concentration: HF concentration 8 ppm, CO2 concentration 12 ppm). The server inputs the above raw data (a total of 12 parameters) into the running feature normalization unit. This unit generates four 5-dimensional running characterization vectors through Z-score normalization (according to the preset "parameter-mean-standard deviation" calibration curve, such as lithium salt concentration mean 1.0 mol / L, standard deviation 0.05 mol / L, normalized concentration = (1.02-1.0) / 0.05 = 0.4) and principal component analysis (retaining the four principal components with a cumulative contribution rate of 96%): kinetic characterization vector V1=[0.82, -0.35 [0.12, 0.68, 0.75] (reflecting insertion / deintercalation kinetics), deposition tendency characterization vector V2=[1.21, -0.53, -0.28, 0.41, -0.15] (reflecting lithium dendrite risk), SEI state characterization vector V3=[0.93, 0.72, -0.45, -0.31, 0.58] (reflecting SEI stability), and high voltage safety characterization vector V4=[0.65, 0.88, -0.62, 0.59, -0.23] (reflecting high voltage safety margin).

[0051] The non-target runtime representation vectors are used to generate coupled representation vectors through the first multi-dimensional runtime association modeling unit: The server uses V1 as the target runtime representation vector and inputs the non-target runtime representation vectors (V2, V3, V4) into the first multi-dimensional runtime association modeling unit to generate their corresponding coupled representation vector C1. 1. The first feature integration network generates the basic characterization of operation coupling: The first feature integration network receives V2, V3, and V4, and performs weighted integration based on physical correlation (weight of lithium deposition tendency on kinetic influence 0.4, SEI state 0.3, high voltage safety 0.3): the basic characterization of operation coupling M_avg=0.4×V2+0.3×V3+0.3×V4. Substituting the specific values, the first dimension of M_avg = 0.4 × 1.21 + 0.3 × 0.93 + 0.3 × 0.65 = 0.484 + 0.279 + 0.195 = 0.958; the second dimension = 0.4 × (-0.53) + 0.3 × 0.72 + 0.3 × 0.88 = -0.212 + 0.216 + 0.264 = 0.268; subsequent dimensions are calculated sequentially, resulting in the basic representation of runtime coupling, M_avg = [0.958, 0.268, -0.389, 0.460, 0.120] (5 dimensions).

[0052] 2. The first multi-level evolutionary unit generates the coupled representation vector: The first multi-level evolutionary unit performs a coupled dynamic evolution mechanism on M_avg based on dynamic feature importance weighting: first through a ReLU activated neural network layer (weight matrix) The pre-training parameters are [[0.12,0.08,0.05,0.09,0.06],[0.07,0.15,0.04,0.08,0.06],...], and the bias is... The first nonlinear transformation is applied to M_avg to generate an attention weight vector W = [0.22, 0.35, 0.10, 0.25, 0.08] (the sum of the weights of each dimension is 1, and the second dimension has the highest contribution due to its corresponding lithium dendrite growth rate). Then, W and M_avg are weighted and fused dimension by dimension: after weighting, the first dimension of M_avg = 0.22 × 0.958 ≈ 0.211, and the second dimension = 0.35 × 0.268 ≈ 0.094, resulting in a weighted vector = [0.211, 0.094, -0.039, 0.115, 0.010]. Finally, a fully connected layer activated by tanh (weight matrix) is applied. bias Perform a second nonlinear transformation and feature dimension reduction to output a coupled representation vector C1=[0.192,0.087,-0.035,0.106,0.009] (5-dimensional).

[0053] The first associated coupled representation vector is generated by the coupled response unit based on the input first parameter of the representation vector and the coupled representation vector. The server inputs the target running representation vector V1=[0.82,-0.35,0.12,0.68,0.75] and the coupling representation vector C1=[0.192,0.087,-0.035,0.106,0.009] into the first parameter coupling response unit. This unit performs a preset parameter-by-parameter coupling calculation: A1[i]=V1[i]×C1[i]+λ×(V1[i]+C1[i]) (λ=0.1 is the coupling strength coefficient, which is preset in the unit parameter library). Calculation process: Dimension 1 A1[1] = 0.82 × 0.192 + 0.1 × (0.82 + 0.192) = 0.157 + 0.1 × 1.012 = 0.258; Dimension 2 A1[2] = (-0.35) × 0.087 + 0.1 × (-0.35 + 0.087) = -0.030 + 0.1 × (-0.263) = -0.056; Subsequent dimensions are calculated sequentially to obtain the first associated coupling representation vector A1 = [0.258, -0.056, 0.008, 0.156, 0.098] (5 dimensions). The server repeats the above process for V2, V3, and V4 to generate the corresponding first associated coupling representation vectors A2, A3, and A4, respectively, completing the preliminary association modeling of the single running dimension and the overall coupling state in the system.

[0054] Through the above process, the server utilizes the functional units of the electrolyte coupling monitoring model to transform the original operating condition data into the first associated coupling representation vector, laying the foundation for the generation of subsequent operating status associated indicators.

[0055] In this embodiment of the invention, the electrolyte coupling monitoring model further includes a second multidimensional operational correlation modeling unit, a deep coupling evolution calculation unit, an operational feature integration unit, and a coupling modeling unit. Both the second multidimensional operational correlation modeling unit and the deep coupled evolution calculation unit are connected to the operational feature normalization unit. The second multidimensional operational correlation modeling unit is used to perform second-level coupled modeling on different operational representation vectors. The second multidimensional operational correlation modeling unit includes a second parameter coupled response unit. The deep coupling evolutionary computation unit is used to splice the various running representation vectors and implement a deep coupling dynamic evolution mechanism. The deep coupling evolutionary computation unit includes a sequentially cascaded second feature integration network and a second multi-level running evolution unit. The first multidimensional operation association modeling unit, the second multidimensional operation association modeling unit, and the deep coupling evolution calculation unit are all connected to the operation feature integration unit, and the operation feature integration unit is connected to the coupling modeling unit. The operational feature integration unit is used to perform vector concatenation on the outputs of the first multidimensional operational correlation modeling unit, the second multidimensional operational correlation modeling unit, and the deep coupling evolution calculation unit, and input the vector concatenation result into the coupling modeling unit. The coupling modeling unit is used to output the operational state correlation index.

[0056] In an embodiment of the invention, taking the server monitoring of a new energy vehicle lithium-ion battery system (NCM811 / graphite cathode, electrolyte is 1mol / L LiPF6-EC / DEC (3:7 volume ratio)) as an example, the electrolyte coupling monitoring model deployed on the server, based on the aforementioned units, adds a second multi-dimensional operation correlation modeling unit (including a second parameter coupling response unit), a deep coupling evolution calculation unit (including a second feature integration network and a second multi-level operation evolution unit), an operation feature integration unit, and a coupling modeling unit. The connection relationship between each unit is as follows: the operation feature normalization unit is connected to the first multi-dimensional operation correlation modeling unit, the second multi-dimensional operation correlation modeling unit, and the deep coupling evolution calculation unit, respectively; the outputs of the above three units are connected to the operation feature integration unit, and then output from the integration unit to the coupling modeling unit. The server coordinates with each unit to generate operation status correlation indicators according to the following process: The second multidimensional operational correlation modeling unit generates a secondary coupling representation vector. The second multidimensional operational correlation modeling unit receives four operational representation vectors (V1=[0.82,-0.35,0.12,0.68,0.75] (kinetics), V2=[1.21,-0.53,-0.28,0.41,-0.15] (lithium deposition), V3=[0.93,0.72,-0.45,-0.31,0.58] (SEI), V4=[0.65,0.88,-0.62,0.59,-0.23] (high voltage safety)) output by the operational feature normalization unit. It then performs "secondary coupling modeling" (direct coupling between pairs of operational representation vectors, highlighting interdimensional interactions) through the second parameter coupling response unit. Taking V1 as an example, it needs to be coupled separately to each of the other individual operational representation vectors (V2, V3, V4). V1 and V2 coupling: The second parameter coupling response unit performs term-by-term multiplication (preset coupling rule) to generate the second associated coupling representation vector B. 12 =[0.82×1.21,(-0.35)×(-0.53),0.12×(-0.28),0.68×0.41,0.75×(-0.15)]=[0.992,0.186,-0.034,0.279,-0.112](5-dimensional) V1 and V3 are coupled: B is generated similarly. 13 =[0.82×0.93,(-0.35)×0.72,0.12×(-0.45),0.68×(-0.31),0.75×0.58]=[0.763,-0.252,-0.054,-0.211,0.435](5-dimensional) V1 and V4 are coupled: generating B 14 =[0.82×0.65,(-0.35)×0.88,0.12×(-0.62),0.68×0.59,0.75×(-0.23)]=[0.533,-0.308,-0.074,0.401,-0.173](5-dimensional).

[0057] Following this logic, the server performs coupling calculations with other vectors on V2, V3, and V4 respectively, generating a total of 12 second-association coupling representation vectors (B). 12 -B 43 (5 dimensions each, totaling 60 dimensions), stored in the output cache of the second multidimensional running association modeling unit.

[0058] The deep-coupling evolutionary computational unit generates a deep-coupling representation vector: The deep coupled evolutionary computation unit performs global coupled feature extraction on the runtime representation vector through the second feature integration network and the second multi-level runtime evolution unit: 1. Second feature integration network splicing and running representation vector: Receive V1-V4, splice them column by column to form an integrated representation vector V_all (4 5-dimensional vectors spliced ​​into 20 dimensions): V_all=[0.82,-0.35,0.12,0.68,0.75,1.21,-0.53,-0.28,0.41,-0.15,0.93,0.72,-0.45,-0.31,0.58,0.65,0.88,-0.62,0.59,-0.23].

[0059] 2. The second multi-level evolutionary unit performs deep-coupled dynamic evolution: Multi-level evolution based on dynamic feature importance weighting is performed on V_all. Attention weights are generated by transforming V_all through a ReLU-activated neural network layer (weight matrix W3∈20, pre-trained parameters) to generate a 20-dimensional attention weight vector W=[0.05,0.08,0.03,0.06,0.04,0.07,0.12,0.02,0.05,0.03,0.10,0.06,0.04,0.03,0.05,0.04,0.09,0.03,0.05,0.03] (the weights sum to 1, and the 7th dimension has the highest contribution due to the lithium dendrite growth rate corresponding to V2). Weighted fusion and dimensional reduction: W and V_all are weighted dimension by dimension (e.g., the 7th dimension: 0.12×(-0.53)=-0.064) to obtain a 20-dimensional weighted vector; then, feature reduction is performed through a fully connected layer activated by tanh (output dimension 5) to extract global coupling features across all running representation vectors, resulting in a deep association coupling representation vector D=[0.22,0.18,-0.09,0.15,0.11] (5 dimensions), which is stored in the output cache of the deep coupling evolution computation unit.

[0060] The feature integration unit concatenates the multi-source coupled representation vector: The runtime feature integration unit receives the first correlation coupling representation vector (A1=[0.258,-0.056,0.008,0.156,0.098], A2=[0.185,0.221,-0.072,0.103,0.065], A3=[0.210,0.153,0.198,-0.084,0.122], A4=[) output by the first multi-dimensional runtime correlation modeling unit. [0.167, 0.205, -0.110, 0.141, 0.089], totaling 4×5=20 dimensions), 12 second association coupling representation vectors (60 dimensions) output by the second multidimensional running association modeling unit, and D (5 dimensions) output by the deep coupling evolution calculation unit are concatenated in the order of "first association → second association → deep association" to generate the coupling representation vector integration result Z (20+60+5=85 dimensions).

[0061] The coupling modeling unit outputs operational status correlation indicators: The coupling modeling unit uses a pre-trained multilayer perceptron (MLP) to perform multi-level coupling modeling of Z: an 85-dimensional input layer (corresponding to the dimension of Z), two hidden layers (32-dimensional and 16-dimensional, ReLU activation), and a 1-dimensional output layer (Sigmoid activation, outputting the running state correlation index S∈[0,1]). The server inputs the 85-dimensional Z into the MLP, and the model performs deep nonlinear transformations (such as the first hidden layer output: ReLU(W4·Z+b4), W4∈ 32×85 b4∈ 32 The cross-dimensional coupling pattern was extracted, and the final calculation yielded S=0.89, which quantitatively indicates the operational coupling relationship between the target battery and the electrolyte (the closer S is to 1, the better the coupling adaptability).

[0062] Through the collaboration of the aforementioned multiple units, the server has achieved full-process modeling from raw operating condition data to operational status correlation indicators, providing a quantitative basis for battery-electrolyte coupling status assessment.

[0063] In this embodiment of the invention, the deep coupling dynamic evolution mechanism is a multi-level evolution mechanism based on dynamic feature importance weighting; The deep coupling dynamic evolution mechanism for the integrated representation vector, which yields a deep-coupling representation vector, can be implemented through the following example.

[0064] The integrated representation vector is subjected to a first nonlinear transformation to generate an attention weight vector for evaluating the contribution of each feature dimension in the overall coupling state. The attention weight vector and the integrated representation vector are weighted and fused dimension by dimension to obtain the weighted integrated representation vector. The weighted integrated representation vector is subjected to a second nonlinear transformation and feature dimension reduction to extract global, deep coupling features across multiple running representation vectors, thereby obtaining the deep correlation coupling representation vector.

[0065] In an embodiment of the present invention, exemplarily, taking the server monitoring of a new energy vehicle lithium-ion battery system (NCM811 / graphite cathode, electrolyte is 1 mol / L LiPF6-EC / DEC (3:7 volume ratio)) as an example, the server has, through the second feature integration network of the deep coupled evolutionary computing unit, integrated the running characterization vectors V1=[0.82,-0.35,0.12,0.68,0.75] (kinetics, 5-dimensional), V2=[1.21,-0.53,-0.28,0.41,-0.15] (lithium deposition tendency, 5-dimensional), and V3=[0.9 V1 = [0.82, -0.72, -0.45, -0.31, 0.58] (SEI state, 5-dimensional) and V4 = [0.65, 0.88, -0.62, 0.59, -0.23] (high voltage safety, 5-dimensional) are concatenated into a 20-dimensional integrated representation vector V_all = [0.82, -0.35, 0.12, 0.68, 0.75, 1.21, -0.53, -0.28, 0.41, -0.15, 0.93, 0.72, -0.45, -0.31, 0.58, 0.65, 0.88, -0.62, 0.59, -0.23]. The server executes a multi-level evolution mechanism based on dynamic feature importance weighting to generate a deep-linked coupling representation vector D: The first nonlinear transformation generates the attention weight vector: The server performs the first nonlinear transformation on V_all (using a ReLU-activated neural network layer with weight matrix W3∈) through the second multi-level running evolution unit of the deeply coupled evolutionary computation unit. 20×20 The pre-trained parameters are learned through sample data, and the bias b3∈ 20This generates an attention weight vector W (20 dimensions, with the weights of each dimension summing to 1) to evaluate the contribution of each feature dimension to the overall coupling state. Specifically, V_all is input into this neural network layer, Z1 = W3·V_all + b3 is calculated, and then W is obtained through the ReLU activation function (ReLU(x) = max(0,x)). For example, for the second dimension of V_all (the second dimension of V1, -0.35, corresponding to charge transfer resistance), the corresponding row vector in W3 is multiplied by V_all and then added to b3, resulting in a ReLU output weight of 0.08; the seventh dimension (the second dimension of V2, -0.53, corresponding to lithium dendrite growth rate) has the greatest impact on the overall coupling state, resulting in an output weight of 0.12. The final attention weight vector is generated as W=[0.05,0.08,0.03,0.06,0.04,0.07,0.12,0.02,0.05,0.03,0.10,0.06,0.04,0.03,0.05,0.04,0.09,0.03,0.05,0.03], where the weights of each dimension reflect the contribution of the corresponding feature to the global coupling.

[0066] Attention weight vector and ensemble representation vector are fused dimension-wise: The server performs a dimension-wise weighted fusion of the attention weight vector W and the ensemble representation vector V_all, calculated as follows: the i-th dimension of the weighted ensemble representation vector = W[i] × V_all[i] (i = 1~20). Taking the key dimension as an example: Second dimension (V1 second dimension, charge transfer resistance): 0.08×(-0.35)=-0.028 (weight 0.08, eigenvalue -0.35); 7th dimension (V2 2nd dimension, lithium dendrite growth rate): 0.12×(-0.53)=-0.064 (weight 0.12, eigenvalue -0.53, highest contribution); 11th dimension (V3 1st dimension, LiF proportion in SEI): 0.10 × 0.93 = 0.093 (weight 0.10, eigenvalue 0.93, reflecting contribution to SEI stability). 17th dimension (V4 2nd dimension, thermal runaway temperature): 0.09 × 0.88 = 0.079 (weight 0.09, eigenvalue 0.88, reflecting the contribution to high voltage safety).

[0067] After weighting, a 20-dimensional integrated characterization vector is obtained: [-0.041,-0.028,0.004,0.041,0.030,0.085,-0.064,-0.006,0.021,-0.004,0.093,0.043,-0.018,-0.010,0.029,0.026,0.079,-0.019,0.030,-0.007]. This vector highlights the contributions of key coupling features such as lithium dendrite growth rate, LiF ratio, and thermal runaway temperature.

[0068] The second nonlinear transformation and feature dimension reduction generate a deeply correlated and coupled representation vector: The server performs a second nonlinear transformation and feature dimension reduction on the weighted 20-dimensional ensemble representation vector: through a fully connected layer activated by tanh (weight matrix W4∈ 5×20 Pre-trained parameters; bias b4∈ 5 The 20-dimensional vector is reduced to 5-dimensional, and global, deep coupling features across V1 to V4 are extracted. The specific process is as follows: The weighted vector is input into the fully connected layer, and Z2 = W4·weighted vector + b4 is calculated, and then activated by tanh (output range [-1,1]). For example, the first dimension of Z2 = W4[1,:]·weighted vector + b4[1]≈0.22, and the output after tanh is 0.22; the second dimension ≈0.18, and the output is 0.18; and so on, finally obtaining the 5-dimensional deep correlation coupling representation vector D=[0.22,0.18,-0.09,0.15,0.11]. This vector comprehensively reflects the global deep coupling relationship between the battery operating state and electrolyte characteristics, and provides the core input for the subsequent calculation of operating state correlation indicators.

[0069] In this embodiment of the invention, the following implementation methods are also provided.

[0070] When the operation status correlation index does not meet the operation threshold judgment condition, the reverse optimization analysis process is initiated. In the reverse optimization analysis process, the part representing the battery operating state in the integrated result of the coupled characterization vector is fixed, and an optimization problem with electrolyte property parameters as variables is constructed. The solution is to find the electrolyte property parameter adjustment amount that optimizes the operating state correlation index in the direction of satisfying the operating threshold judgment condition. Based on the adjusted electrolyte property parameters obtained from the solution, adjustment suggestions are generated, including specific adjustment directions and ranges. The adjustment suggestions are then output to the user interface or battery management system.

[0071] In this embodiment of the invention, taking the server monitoring of a new energy vehicle lithium-ion battery system (NCM811 / graphite cathode, electrolyte is 1mol / L LiPF6-EC / DEC (3:7 volume ratio)) as an example, the server presets the operating threshold judgment condition as "operating status correlation index S≥0.7" (indicating good coupling compatibility between the battery and the electrolyte). In a certain monitoring, the coupling modeling unit output S=0.65 (<0.7), which does not meet the judgment condition. The server then starts the reverse optimization analysis process, and the specific steps are as follows: Initiating the reverse optimization analysis process and fixing the battery operating state: The server first locates the portion representing the battery operating state in the integrated result Z (85 dimensions) of the coupled characterization vectors. Z is composed of the first associated coupled characterization vector (A1-A4), the second associated coupled characterization vector (B series), and the deep associated coupled characterization vector (D). The battery operating state portion corresponds to the features generated from the original battery operating data (such as the lithium-ion insertion / extraction peak potential difference, lithium dendrite growth rate, and the proportion of LiF in the SEI, which have been fixed as the first 40 dimensions of Z). The server fixes the feature values ​​of this portion through masking operations, and only sets the features related to electrolyte properties in the last 45 dimensions (corresponding to the lithium-ion solvation structure Raman peak area ratio of 0.75, migration efficiency of 0.00504, lithium salt concentration of 1 mol / L, HF concentration of 8 ppm, etc.) as optimization variables.

[0072] Construct an optimization problem with electrolyte property parameters as variables and solve for the adjustment amount.

[0073] The server constructs an optimization problem with the objective of "making S ≥ 0.7": minimize |S_target - S| (S_target = 0.7), where the variables are the electrolyte property parameter vector X = [lithium salt concentration c, solvent EC / DEC volume ratio r, HF concentration h] (units are mol / L, -, ppm, respectively). Based on a pre-trained coupled modeling unit (MLP), the Jacobian matrix is ​​solved using gradient descent to obtain the parameter adjustment amount to optimize S towards 0.7. Lithium salt concentration c: currently 1 mol / L, needs to be increased by 0.1 mol / L (adjustment amount Δc = +0.1) to enhance the stability of the ion-solvated sheath (Raman peak area ratio is expected to increase from 0.75 to 0.82). Solvent volume ratio r: currently 3:7, needs to be adjusted to 4:6 (Δr = +10% EC percentage) to improve the electrolyte dielectric constant (ion migration efficiency is expected to increase from 0.00504 to 0.0058). HF concentration h: currently 8 ppm, needs to be reduced by 3 ppm (Δh=-3) to reduce SEI film erosion (LiF characteristic peak intensity is expected to increase from 0.68 to 0.75).

[0074] Generate adjustment suggestions and output them to the user interface: Based on the adjustments obtained from the solution, the server generates electrolyte parameter adjustment suggestions: "The target battery currently has insufficient coupling compatibility with the electrolyte (S=0.65<0.7). The following adjustments are recommended: ① Increase the lithium salt concentration from 1.0mol / L to 1.1mol / L (+0.1mol / L); ② Adjust the solvent EC / DEC volume ratio from 3:7 to 4:6 (EC percentage +10%); ③ Reduce the HF concentration from 8ppm to 5ppm (-3ppm) through a dehydration process. After the adjustments, the lithium-ion migration efficiency is expected to improve by 15%, SEI stability by 10%, and the operating status correlation index can be optimized to S=0.72 (≥0.7)." This suggestion is simultaneously output to the vehicle user interface and battery management system (BMS) to guide electrolyte maintenance operations.

[0075] In this embodiment of the invention, the electrolyte coupling monitoring model is trained in the following manner: Obtain a sample dataset, which includes multiple sets of basic information on the operating conditions of sample systems and their corresponding labeled operating status correlation indicators generated by high-precision simulation. The sample system operating condition basic information set is input into the electrolyte coupling monitoring model to be trained to obtain the predicted operating status correlation index. The difference between the predicted operational state correlation index and the labeled operational state correlation index is calculated. With the goal of minimizing the difference, the network parameters of the operational feature normalization unit, the first multidimensional operational correlation modeling unit, the second multidimensional operational correlation modeling unit, the deep coupling evolution calculation unit, the operational feature integration unit, and the coupling modeling unit are updated simultaneously using the error backpropagation algorithm. In each parameter update process, the first multi-level operation evolution unit in the first multi-dimensional operation association modeling unit and the second multi-level operation evolution unit in the second multi-dimensional operation association modeling unit use the same parameter matrix for operation to achieve parameter sharing.

[0076] In this embodiment of the invention, taking the training of an electrolyte coupling monitoring model for a lithium-ion battery system (NCM811 / graphite cathode) of a new energy vehicle as an example, the server needs to optimize the parameters of each unit of the model through sample data to achieve accurate evaluation of the battery-electrolyte coupling adaptability. In the training scenario, the target electrolyte is a LiPF6-carbonate system with different ratios (lithium salt concentration 0.8-1.2 mol / L, solvent EC / DEC volume ratio 2:8 to 5:5), and the server follows the training process as follows: Obtain the sample dataset: The server collects sample datasets from laboratory databases and cloud-based battery testing platforms, containing 1000 sets of basic information on sample system operating conditions and their corresponding labeled operational status indicators (S-labels). Each set of basic information on sample system operating conditions includes: Battery operating data: Lithium-ion insertion / extraction peak potential difference (0.08-0.15V), lithium dendrite growth rate (0.05-0.12μm / h), LiF characteristic peak intensity ratio in SEI (0.55-0.75), thermal runaway temperature (190-230℃) and other data were obtained through charge and discharge tests. Electrolyte property data: The lithium-ion solvation Raman peak area ratio (0.68-0.85), ion mobility efficiency (0.004-0.006 S / cm), lithium salt concentration (0.8-1.2 mol / L), and HF concentration (5-15 ppm) were obtained through spectral / chromatographic analysis.

[0077] The operating status correlation index (S-label) is derived from the experimentally determined "comprehensive score of battery-electrolyte coupling compatibility": it is obtained by comprehensively weighting the cycle life test (capacity retention rate of 60%-90% after 1000 cycles), the low temperature performance test (capacity retention rate of 50%-80% after discharge at -20℃), and the thermal safety test (thermal runaway trigger time of 30-100s). S-label ∈ [0,1] (e.g., for a sample with a cycle life retention rate of 85%, a low temperature capacity retention rate of 75%, and a thermal runaway time of 80s, the corresponding S-label = 0.82).

[0078] Input samples into the model to be trained to generate predictive runtime status related metrics: The server divides the 1000 sets of basic system operating information samples into a training set (700 sets) and a validation set (300 sets) in a 7:3 ratio. Taking a sample from the training set as an example (lithium salt concentration 1 mol / L, EC / DEC volume ratio 3:7, HF concentration 8 ppm): 1. Run the feature normalization unit: Perform Z-score normalization on the sample data to generate 4 running representation vectors (e.g., V1=[0.82,-0.35,0.12,0.68,0.75], as in the previous example); 2. Multi-unit collaborative computation: The first multi-dimensional operational correlation modeling unit generates the first correlation coupling representation vector (A1-A4), the second multi-dimensional operational correlation modeling unit generates the second correlation coupling representation vector (B series), and the deep coupling evolution computation unit generates the deep correlation coupling representation vector (D). 3. Coupling representation vector integration and prediction: The running feature integration unit concatenates A, B, and D to obtain an 85-dimensional Z, which is input into the coupling modeling unit (MLP to be trained) and outputs the predicted running state correlation index S_pred (initially during training, due to random parameters, S_pred=0.58, which differs from the sample S label=0.82).

[0079] Error backpropagation and parameter update (including parameter sharing): The server performs the following steps with the goal of minimizing the difference between the predicted S_pred and the labeled S_label: 1. Calculate the loss: Use the mean squared error (MSE) loss function L=1 / NΣ(S_pred,iSlabel,i)² (N=700 is the number of training set samples). The current sample L=(0.58-0.82)²=0.0576, and the initial L=0.12 for the entire training set.

[0080] 2. Backpropagation parameter update: The gradient of L with respect to the network parameters of each unit is calculated using the backpropagation algorithm (BP algorithm), and the parameters are updated synchronously. Run the feature normalization unit: update the normalization mean / standard deviation parameters (e.g., fine-tune the lithium salt concentration mean from 1.0 mol / L to 0.98 mol / L to improve normalization accuracy); First / Second Multidimensional Operational Association Modeling Unit: Update feature integration weights (e.g., adjust lithium deposition tendency weight from 0.4 to 0.42), and couple strength coefficient λ of parameter coupling response unit (fine-tuned from 0.1 to 0.11). Deeply coupled evolutionary computational unit: Update the attention weight generation matrix of the second multi-level running evolutionary unit (enhance the focus on the lithium dendrite growth rate dimension); Coupled modeling unit: Update the weights of the hidden layer of the MLP (e.g., the weight matrix W of the first layer). 32×85 Adjusted from random values ​​to [[0.02,0.05,...],...]).

[0081] 3. Parameter Sharing Implementation: The first multi-level operational evolution unit in the first multi-dimensional operational correlation modeling unit and the second multi-level operational evolution unit in the second multi-dimensional operational correlation modeling unit use the same parameter matrix W (∈ 5×5 The initial values ​​for pre-training are [[0.12,0.08,...],[0.07,0.15,...],...]). During backpropagation, the gradients of both are synchronously accumulated to W to ensure that a consistent coupled dynamic evolution mechanism is learned (such as the unified weighting rule for the lithium dendrite growth rate characteristics).

[0082] Training convergence and model validation: After 500 rounds of iterative training on the server, the training set loss L decreased to below 0.01, and the validation set L=0.015 (prediction accuracy > 95%), indicating model convergence. At this point, the shared parameter matrix W of the first / second multi-level evolutionary units stabilized as [[0.14,0.09,0.06,0.10,0.07],[0.08,0.16,0.05,0.09,0.07],...]. The parameters of each unit synergistically reflect the battery-electrolyte coupling law and can be used for subsequent online monitoring.

[0083] In this embodiment of the invention, the system operating condition basic information set also includes environmental operating condition data, which includes at least environmental temperature data and environmental humidity data; The acquisition of the system operating condition basic information set can be implemented through the following example.

[0084] Simultaneously collect battery operation data of the target battery, property data of the target electrolyte, and environmental operating condition data; The feature extraction of the system operating condition basic information set yields at least two operational representation vectors, including: The battery operation data, the target electrolyte property data, and the environmental condition data are subjected to joint feature extraction and normalization processing to form at least two operation characterization vectors that reflect the internal state of the battery, the characteristics of the electrolyte, and its interaction with the external environment.

[0085] In an embodiment of the present invention, taking the server monitoring of a lithium-ion battery system for a new energy vehicle (NCM811 / graphite cathode, electrolyte is 1mol / L LiPF6-EC / DEC (3:7 volume ratio)) as an example, the system operating condition basic information set adds environmental operating condition data (ambient temperature, ambient humidity) to the original battery operating data and electrolyte property data. The server synchronously collects data from multiple sources and performs joint feature extraction to generate an operating characterization vector reflecting the interaction between the battery, electrolyte, and external environment. The specific process is as follows: Simultaneous acquisition of target battery, electrolyte, and environmental condition data: The server, through an NTC temperature sensor (sampling frequency 1Hz, accuracy ±0.5℃) deployed within the battery pack, an onboard ambient humidity sensor (sampling frequency 1Hz, accuracy ±5%RH), and external detection equipment, synchronously collects timestamps of battery operating data, electrolyte property data, and environmental condition data (all aligned with GPS timestamps to ensure data corresponds to the system state at the same moment). In a winter low-temperature driving scenario, the server continuously collected three sets of typical operating condition data (timestamps t=10:00:00, t=10:00:01, t=10:00:02), with the data at t=10:00:00 as follows: Environmental operating data: Ambient temperature T_env = -5℃ (from the temperature sensor on the outside of the battery pack, reflecting the low external temperature environment), ambient humidity H_env = 60%RH (from the vehicle environment sensor, reflecting the high humidity environment); Battery operating data: Lithium-ion insertion / extraction kinetics (cyclic voltammetry peak potential difference 0.14V, 0.03V higher than at room temperature (25℃), due to low temperature inhibiting ion migration); Lithium metal deposition tendency (lithium dendrite appearance time 320s, 100s shorter than at room temperature, low temperature accelerates dendrite growth); SEI state (LiF characteristic peak intensity ratio 0.62, 0.06 lower than at room temperature, SEI stability decreases at low temperature); High voltage safety margin (thermal runaway temperature 205℃, 10℃ lower than at room temperature, increased risk of thermal runaway at low temperature); Electrolyte properties: Lithium-ion solvation structure (Raman peak area ratio 0.70, down 0.05 from room temperature, indicating a loose solvation sheath at low temperatures); Ion mobility efficiency 0.0042 S / cm (down 0.0008 S / cm from room temperature, indicating a decrease in ionic conductivity at low temperatures); Lithium salt concentration 1.0 mol / L (unchanged, but actual degree of dissociation decreased at low temperatures); HF concentration 9 ppm (up 1 ppm from room temperature, indicating that the high humidity environment caused the electrolyte to absorb a small amount of water, promoting the hydrolysis of LiPF6 to generate HF).

[0086] Joint feature extraction and normalization processes generate runtime representation vectors: The server performs joint feature extraction on synchronously collected battery operation data, electrolyte property data, and environmental condition data, focusing on exploring the interactive influence of environmental factors on the battery-electrolyte coupling state. Then, it generates four operational representation vectors (each with six dimensions, plus a new one-dimensional environmental interaction feature) through Z-score normalization. The specific process is as follows: 1. Dynamic characterization vector V1 (adding the dimension of the influence of ambient temperature on migration efficiency) Original characteristic dimensions (5 dimensions): Cyclic voltammetric peak potential difference 0.14V (normalized to 0.78), charge transfer resistance 92Ω (normalized to -0.42), Warburg impedance coefficient 0.15 (normalized to 0.18), electrolyte migration efficiency 0.0042S / cm (normalized to 0.55), solvation Raman peak area ratio 0.70 (normalized to 0.68); Added environmental interaction feature dimension (6th dimension): Environmental temperature influence coefficient = (normal temperature migration efficiency - current migration efficiency) / (normal temperature - current temperature) = (0.0050 - 0.0042) / (25 - (-5)) = 0.0008 / 30 ≈ 0.000027 (normalized to -0.32, negative value indicates that low temperature inhibits migration); Normalized V1: [0.78, -0.42, 0.18, 0.55, 0.68, -0.32].

[0087] 2. Deposition tendency characterization vector V2 (adding a dimension on the influence of ambient temperature on dendrite growth)

[0088] Original characteristic dimensions (5 dimensions): lithium dendrite appearance time 320s (normalized to 1.05), growth rate 0.12μm / h (normalized to 0.85), negative electrode overpotential 0.18V (normalized to 0.72), lithium salt concentration 1.0mol / L (normalized to 0.40), electrolyte moisture content 22ppm (normalized to 0.35, high humidity environment leads to increased moisture); Added environmental interaction feature dimension (6th dimension): Acceleration factor of ambient temperature on dendrite growth = current growth rate / room temperature growth rate = 0.12 / 0.08 = 1.5 (normalized to 0.65, a positive value indicates that low temperature accelerates dendrite growth). Normalized V2: [1.05, 0.85, 0.72, 0.40, 0.35, 0.65].

[0089] 3. SEI State Characterization Vector V3 (Added dimension: influence of ambient humidity on HF concentration)

[0090] Original characteristic dimensions (5 dimensions): LiF characteristic peak intensity ratio 0.62 (0.60 after normalization), SEI decomposition heat release peak temperature 122℃ (0.58 after normalization), heat release during decomposition 45J / g (-0.42 after normalization), HF concentration 9ppm (0.45 after normalization), SEI thickness 12nm (0.38 after normalization); Added environmental interaction feature dimension (6th dimension): Environmental humidity influence coefficient = (current HF concentration - HF concentration in a dry environment at room temperature) / (current humidity - dry humidity) = (9-7) / (60-30) = 2 / 30≈0.067 (0.52 after normalization, a positive value indicates that high humidity promotes HF generation); Normalized V3: [0.60, 0.58, -0.42, 0.45, 0.38, 0.52].

[0091] 4. High-voltage safety characterization vector V4 (adds a dimension on the influence of ambient temperature on thermal runaway temperature)

[0092] Original characteristic dimensions (5 dimensions): thermal runaway temperature 205℃ (normalized to 0.70), maximum temperature rise rate 15℃ / s (normalized to -0.55), 4.5V capacity retention 85% (normalized to 0.65), CO2 byproduct concentration 14ppm (normalized to 0.40), electrolyte flash point 65℃ (normalized to 0.50); Added environmental interaction feature dimension (6th dimension): Sensitivity of ambient temperature to thermal runaway temperature = (Ambient thermal runaway temperature - Current thermal runaway temperature) / (Ambient temperature - Current temperature) = (215 - 205) / (25 - (-5)) = 10 / 30 ≈ 0.333 (normalized to -0.48, negative value indicates that low temperature reduces thermal runaway temperature); Normalized V4: [0.70, -0.55, 0.65, 0.40, 0.50, -0.48].

[0093] Through the joint feature extraction and normalization described above, the four operational representation vectors (V1~V4, each with 6 dimensions) generated by the server not only retain the internal state of the battery and the characteristics of the electrolyte, but also quantify the influence of environmental temperature and humidity on the coupling state of the two by adding an environmental interaction feature dimension (such as low temperature accelerating dendrite growth and high humidity promoting HF generation), providing a more comprehensive input for subsequent deep coupling modeling and ensuring that the operational state correlation indicators can reflect the system adaptability under real environment.

[0094] This invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned method for monitoring the operating status of a lithium-ion battery electrolyte system. Figure 2 As shown, Figure 2 This is a structural block diagram of a computer device 100 provided in an embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112, and a communication unit 113. To enable data transmission or interaction, the memory 111, processor 112, and communication unit 113 are electrically connected to each other directly or indirectly. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0095] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in accordance with the foregoing teachings. These embodiments were chosen and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the disclosure and to employ various embodiments with different modifications to suit a particular intended application.

Claims

1. A method for monitoring the operating status of a lithium-ion battery electrolyte system, characterized in that, The method includes: Acquire a system operating condition basic information set, which includes battery operation data of the target battery and target electrolyte property data of the target electrolyte. The battery operation data includes lithium-ion battery operation parameters of at least one of lithium-ion intercalation and deintercalation kinetics, lithium metal deposition tendency, solid electrolyte interface phase state evolution characteristics, and high voltage operation safety margin. The target electrolyte property data includes lithium-ion battery electrolyte parameters of at least one of lithium-ion solvation structure, lithium-ion migration efficiency, lithium salt concentration, and decomposition by-product concentration. Feature extraction is performed on the system operating condition basic information set to obtain at least two operating characteristic vectors; For each of the aforementioned operational representation vectors, a coupling representation vector corresponding to the operational representation vector is extracted, and feature association modeling is performed on the operational representation vector and the coupling representation vector to obtain a first associated coupling representation vector corresponding to the operational representation vector. The coupling representation vector is a feature vector obtained by feature extraction based on a coupling dynamic evolution mechanism from non-target operational representation vectors other than the operational representation vector among the at least two operational representation vectors. Based on the first associated coupling representation vector associated with each of the aforementioned operational representation vectors, an operational state association index is obtained, which indicates the operational coupling state relationship between the target battery and the target electrolyte.

2. The method according to claim 1, characterized in that, The extraction of the coupled representation vector corresponding to the running representation vector includes: Feature integration is performed on the non-target running representation vectors (excluding the running representation vectors) from the at least two running representation vectors to obtain the running coupling basic representation corresponding to the running representation vectors; Feature extraction based on a coupling dynamic evolution mechanism is performed on the basic representation of the operation coupling to obtain the coupling representation vector corresponding to the operation representation vector; The step of performing feature association modeling on the running representation vector and the coupled representation vector to obtain the first associated coupled representation vector corresponding to the running representation vector includes: The parameter-by-parameter coupling calculation is performed on the running representation vector and the corresponding coupling representation vector to obtain the first associated coupling representation vector corresponding to the running representation vector.

3. The method according to claim 1 or 2, characterized in that, The first associated coupling representation vector, which is associated with each of the aforementioned operational representation vectors, is used to obtain the operational status association index, including: For each of the aforementioned operational representation vectors, feature association modeling is performed on each operational representation vector and each associated operational representation vector to obtain a second associated coupling representation vector corresponding to the operational representation vector. The associated operational representation vector is any operational representation vector among the at least two operational representation vectors excluding the operational representation vector. The aforementioned operational representation vectors are integrated to obtain an integrated representation vector; A deep coupling dynamic evolution mechanism is applied to the integrated representation vector to obtain a deep correlation coupling representation vector; The first associated coupling representation vector, the second associated coupling representation vector, and the deep associated coupling representation vector associated with each of the aforementioned operational representation vectors are concatenated to obtain the integrated result of the coupling representation vectors. Based on the integration results of the coupling representation vector, coupling modeling is performed to obtain the operating state correlation index. The coupling modeling is a multi-level coupling modeling used to reflect the coupling adaptability of the target battery to the target electrolyte.

4. The method according to claim 1, characterized in that, The method is implemented through an electrolyte coupling monitoring model, which includes an operating feature normalization unit and a first multidimensional operating correlation modeling unit. The first multidimensional operating correlation modeling unit is connected to the operating feature normalization unit. The first multidimensional operating correlation modeling unit includes a first feature integration network, a first multi-level operating evolution unit, and a first parameter coupling response unit, which are sequentially cascaded. The feature extraction of the system operating condition basic information set yields at least two operational representation vectors, including: The system operating condition basic information set is input into the operation feature normalization unit to obtain the at least two operation characterization vectors; The step of extracting the coupling representation vector corresponding to each of the running representation vectors includes: for each of the running representation vectors, inputting the non-target running representation vectors other than the running representation vectors from the at least two running representation vectors into the first feature integration network to obtain the running coupling basic representation corresponding to the running representation vector; inputting the running coupling basic representation corresponding to the running representation vector into the first multi-level running evolution unit to obtain the coupling representation vector corresponding to the running representation vector. The step of performing feature association modeling on the running representation vector and the coupled representation vector to obtain the first associated coupled representation vector corresponding to the running representation vector includes: Each of the operational representation vectors and the corresponding coupling representation vectors are input into the first parameter coupling response unit to obtain the first associated coupling representation vector corresponding to the operational representation vector.

5. The method according to claim 4, characterized in that, The electrolyte coupling monitoring model further includes a second multidimensional operational correlation modeling unit, a deep coupling evolution calculation unit, an operational feature integration unit, and a coupling modeling unit. Both the second multidimensional operational correlation modeling unit and the deep coupled evolution calculation unit are connected to the operational feature normalization unit. The second multidimensional operational correlation modeling unit is used to perform second-level coupled modeling on different operational representation vectors. The second multidimensional operational correlation modeling unit includes a second parameter coupled response unit. The deep coupling evolutionary computation unit is used to splice the various running representation vectors and implement a deep coupling dynamic evolution mechanism. The deep coupling evolutionary computation unit includes a sequentially cascaded second feature integration network and a second multi-level running evolution unit. The first multidimensional operation association modeling unit, the second multidimensional operation association modeling unit, and the deep coupling evolution calculation unit are all connected to the operation feature integration unit, and the operation feature integration unit is connected to the coupling modeling unit. The operational feature integration unit is used to perform vector concatenation on the outputs of the first multidimensional operational correlation modeling unit, the second multidimensional operational correlation modeling unit, and the deep coupling evolution calculation unit, and input the vector concatenation result into the coupling modeling unit. The coupling modeling unit is used to output the operational state correlation index.

6. The method according to claim 3, characterized in that, The deep coupling dynamic evolution mechanism is a multi-level evolution mechanism based on the importance weighting of dynamic features; The deep coupling dynamic evolution mechanism for the integrated representation vector, to obtain a deep-coupling representation vector, includes: The integrated representation vector is subjected to a first nonlinear transformation to generate an attention weight vector for evaluating the contribution of each feature dimension in the overall coupling state. The attention weight vector and the integrated representation vector are weighted and fused dimension by dimension to obtain the weighted integrated representation vector. The weighted integrated representation vector is subjected to a second nonlinear transformation and feature dimension reduction to extract global, deep coupling features across multiple running representation vectors, thereby obtaining the deep correlation coupling representation vector.

7. The method according to claim 3, characterized in that, The method further includes: When the associated indicators of the operating status do not meet the operating threshold judgment conditions, the reverse optimization analysis process is initiated. In the reverse optimization analysis process, the part representing the battery operating state in the integrated result of the coupled characterization vector is fixed, and an optimization problem with electrolyte property parameters as variables is constructed. The solution is to find the electrolyte property parameter adjustment amount that optimizes the operating state correlation index in the direction of satisfying the operating threshold judgment condition. Based on the adjusted electrolyte property parameters obtained from the solution, adjustment suggestions are generated, including specific adjustment directions and ranges. The adjustment suggestions are then output to the user interface or battery management system.

8. The method according to claim 5, characterized in that, The electrolyte coupling monitoring model was trained in the following manner: Obtain a sample dataset, which includes multiple sets of basic information on the operating conditions of sample systems and their corresponding labeled operating status correlation indicators generated by high-precision simulation. The sample system operating condition basic information set is input into the electrolyte coupling monitoring model to be trained to obtain the predicted operating status correlation index. The difference between the predicted operational state correlation index and the labeled operational state correlation index is calculated. With the goal of minimizing the difference, the network parameters of the operational feature normalization unit, the first multidimensional operational correlation modeling unit, the second multidimensional operational correlation modeling unit, the deep coupling evolution calculation unit, the operational feature integration unit, and the coupling modeling unit are updated simultaneously using the error backpropagation algorithm. In each parameter update process, the first multi-level operation evolution unit in the first multi-dimensional operation association modeling unit and the second multi-level operation evolution unit in the second multi-dimensional operation association modeling unit use the same parameter matrix for operation to achieve parameter sharing.

9. The method according to claim 1, characterized in that, The system operating condition basic information set also includes environmental operating condition data, which includes at least environmental temperature data and environmental humidity data. The acquisition of the system operating condition basic information set includes: Simultaneously collect battery operation data of the target battery, property data of the target electrolyte, and environmental operating condition data; The feature extraction of the system operating condition basic information set yields at least two operational representation vectors, including: The battery operation data, the target electrolyte property data, and the environmental condition data are subjected to joint feature extraction and normalization processing to form at least two operation characterization vectors that reflect the internal state of the battery, the characteristics of the electrolyte, and its interaction with the external environment.

10. A server system, characterized in that, Includes a server, the server being used to perform the method according to any one of claims 1-9.

Images