Battery operating status monitoring and evaluation method, system, electronic device and storage medium

Abstract

The present application provides a battery operating status monitoring and evaluation method, system, electronic device and storage medium, which relates to the field of battery monitoring technology. The method includes: collecting multimodal monitoring data of the battery during the charging and discharging process; preprocessing the multimodal monitoring data based on a multi-view encoder of deep learning to obtain a multimodal fusion feature vector; performing causal chain reasoning on the multimodal fusion feature vector based on a preset fused lead-acid battery knowledge graph to obtain a battery status causal chain, and performing status evaluation on the lead-acid battery based on the battery status causal chain to obtain a real-time status evaluation result; inputting the multimodal fusion feature vector into a preset battery performance degradation trend estimation model to obtain a battery performance degradation trend; and comprehensively evaluating the status of the lead-acid battery based on the real-time status evaluation result and the battery performance degradation trend to obtain a comprehensive evaluation result. The above method improves the accuracy of the lead-acid battery status evaluation result.

Description

Technical Field

[0001] The present application relates to the field of battery monitoring technology, and in particular to a battery operating status monitoring and evaluation method, system, electronic device and storage medium. Background Art

[0002] Lead-acid batteries are widely used in energy storage power stations, electric vehicles, and other fields due to their low cost, high safety, and long cycle life. To ensure the reliable operation and extend the service life of lead-acid batteries, real-time monitoring and evaluation of their operating status is necessary.

[0003] In existing technologies, battery health monitoring and assessment primarily involves collecting monitoring data from a single modality, such as voltage, current, and temperature, and evaluating the battery status based on preset thresholds or empirical models. However, the degradation process of lead-acid batteries is affected by multiple factors, such as charge and discharge rates, temperature, and corrosion. Relying solely on monitoring data from a single modality makes it difficult to fully characterize the battery's health status. Furthermore, existing assessment methods are mostly based on simplified empirical models and lack in-depth consideration of battery degradation mechanisms, resulting in inaccurate assessment results. Summary of the Invention

[0004] The present application provides a battery operating status monitoring and evaluation method, system, electronic device and storage medium to improve the accuracy of lead-acid battery status evaluation results.

[0005] In a first aspect, the present application provides a method for monitoring and evaluating a battery operating status, the method comprising:

[0006] Collect multimodal monitoring data of lead-acid batteries during the charging and discharging process;

[0007] A multi-view encoder based on deep learning pre-processes multimodal monitoring data to obtain a multimodal fusion feature vector;

[0008] Based on the preset fusion lead-acid battery knowledge graph, causal chain reasoning is performed on the multimodal fusion feature vector to obtain a series of battery status causal chains. Based on the battery status causal chains, the lead-acid battery status is evaluated to obtain real-time status evaluation results.

[0009] Input the multimodal fusion feature vector into a preset battery performance degradation trend estimation model to obtain the battery performance degradation trend;

[0010] Based on the real-time status evaluation results and the battery performance degradation trend, the status of the lead-acid battery is comprehensively evaluated to obtain a comprehensive evaluation result.

[0011] In this technical solution, multimodal monitoring data from lead-acid batteries during the charge and discharge process is collected and preprocessed using a deep learning-based multi-view encoder to generate a multimodal fusion feature vector. This approach fully utilizes the complementary information between different modal data and effectively fuses this data through deep learning technology, resulting in a feature representation that comprehensively characterizes the battery's health status.

[0012] Based on this, this method combines multimodal fusion feature vectors with a pre-defined fused lead-acid battery knowledge graph for causal chain reasoning, generating a series of battery status causal chains. The introduction of the knowledge graph not only considers the complex causal relationships within the battery degradation process but also incorporates rich battery domain knowledge, making the status assessment process more consistent with the physical mechanisms of battery degradation. By assessing the status of lead-acid batteries based on this battery status causal chain, the resulting real-time status assessment results are more accurate and interpretable.

[0013] This method also inputs the multimodal fusion feature vector into a pre-defined battery performance degradation trend estimation model to derive the battery performance degradation trend. This model not only considers historical degradation data but also incorporates field knowledge such as degradation mechanisms and the causal chain of battery states, enabling a more accurate characterization of the dynamic characteristics and state transition patterns of the degradation process. The resulting battery performance degradation trend exhibits enhanced long-term stability and extrapolation capabilities.

[0014] Finally, this method comprehensively evaluates the status of lead-acid batteries by combining real-time status assessment results with battery performance degradation trends. This multi-dimensional, multi-scale assessment method can fully reflect the battery's health status and provide a more reliable basis for battery health management and maintenance decisions.

[0015] In a second aspect of the present application, a battery operating status monitoring and evaluation system is provided, the system comprising:

[0016] Data acquisition module, used to collect multimodal monitoring data of lead-acid batteries during the charging and discharging process;

[0017] Feature extraction module, which is used for multi-view encoder based on deep learning to pre-process multimodal monitoring data and obtain multimodal fusion feature vectors;

[0018] The real-time status assessment module is used to perform causal chain reasoning on the multimodal fusion feature vector based on the preset fusion lead-acid battery knowledge graph to obtain a series of battery status causal chains, and then perform status assessment on the lead-acid battery based on the battery status causal chain to obtain real-time status assessment results;

[0019] A degradation trend prediction module is used to input the multimodal fusion feature vector into a preset battery performance degradation trend estimation model to obtain the battery performance degradation trend;

[0020] The comprehensive evaluation module is used to comprehensively evaluate the status of the lead-acid battery based on the real-time status evaluation result and the battery performance degradation trend to obtain a comprehensive evaluation result.

[0021] In a third aspect of the present application, a computer storage medium is provided. The computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the above method steps.

[0022] In the fourth aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs the above method.

[0023] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0024] 1. This application collects multimodal monitoring data from lead-acid batteries during the charging and discharging process and preprocesses the multimodal monitoring data using a deep learning-based multi-view encoder to obtain a multimodal fusion feature vector. This method fully utilizes the complementary information between different modal data and achieves effective data fusion through deep learning technology, thereby forming a feature representation that comprehensively depicts the battery health status.

[0025] 2. This application performs causal chain reasoning on the multimodal fusion feature vector and the preset fusion lead-acid battery knowledge graph to obtain a series of battery status causal chains. The introduction of the knowledge graph not only takes into account the complex causal relationship of the battery degradation process, but also integrates rich battery domain knowledge, making the status assessment process more consistent with the physical mechanism of battery degradation. By performing status assessment on the lead-acid battery based on the battery status causal chain, the real-time status assessment results obtained have higher accuracy and interpretability.

[0026] 3. This application comprehensively evaluates the health of lead-acid batteries by considering both real-time status assessment results and battery performance degradation trends. This multi-dimensional, multi-scale assessment method can fully reflect the health status of batteries and provide a more reliable basis for battery health management and maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flowchart of a battery operating status monitoring and evaluation method provided in an embodiment of the present application;

[0028] Figure 2 An architectural diagram of a battery operating status monitoring and evaluation system provided in an embodiment of the present application;

[0029] Figure 3 This is a schematic diagram of the structure of an electronic device provided in this application. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0031] In the description of the embodiments of this application, words such as "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "for example" or "for instance" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "for example" or "for instance" is intended to present the relevant concepts in a concrete manner.

[0032] In the description of the embodiments of the present application, the term "multiple" means two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0033] In order to facilitate understanding of the method and system provided by the embodiments of the present application, before introducing the embodiments of the present application, the background of the embodiments of the present application is first introduced.

[0034] Lead-acid batteries, with their low cost, excellent safety, and long cycle life, are widely used in energy storage power stations, electric vehicles, and other fields. To ensure the reliable operation and extend the service life of lead-acid batteries, real-time monitoring and evaluation of their operating status is essential.

[0035] Currently, battery health monitoring and assessment technologies primarily collect monitoring data from a single modality, such as voltage, current, and temperature, and assess the battery's condition based on pre-set thresholds or empirical models. However, the degradation of lead-acid batteries is a complex physical and chemical process, influenced by multiple factors such as charge and discharge rates, temperature, and corrosion. Relying solely on monitoring data from a single modality makes it difficult to comprehensively and accurately characterize the battery's health status. Furthermore, existing assessment methods are mostly based on simplified empirical models and lack in-depth consideration and modeling of battery degradation mechanisms, resulting in a need to improve the accuracy and reliability of the assessment results.

[0036] After the background introduction of the above content, those skilled in the art can understand the problems existing in the prior art. The technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0037] On the basis of the above background technology, further, please refer to Figure 1 , Figure 1 This is a flow chart of a method for monitoring and evaluating the operating status of a battery provided in an embodiment of the present application. The system can be implemented by a computer program or can be run as an independent tool application. Specifically, in the embodiment of the present application, the method can be applied on a server, but can also be applied to electronic devices such as servers. A method for monitoring and evaluating the operating status of a battery includes the following steps:

[0038] S101, collecting multimodal monitoring data of the lead-acid battery during the charging and discharging process;

[0039] Specifically, during the charge and discharge process of a lead-acid battery, multiple sensors deployed on the battery synchronously collect various physical quantities such as the battery's voltage, current, temperature, and internal resistance, generating time-synchronized multimodal monitoring data. The voltage sensor measures the battery's terminal voltage, the current sensor measures the battery's charge and discharge current, the temperature sensor measures the battery's surface temperature, and the internal resistance sensor measures the battery's internal resistance. These sensors collect data in real time at a specific sampling frequency and transmit the data to a data acquisition unit for synchronization and storage.

[0040] By collecting multimodal monitoring data, the status characteristics of lead-acid batteries can be reflected from different perspectives. For example, voltage data reflects the battery's state of charge and polarization, current data reflects the battery's charge and discharge rate and power characteristics, temperature data reflects the battery's thermal stability and heat dissipation performance, and internal resistance data reflects the battery's conductivity and health. These different modal data are correlated and complementary to each other, and fusion analysis can provide more accurate and comprehensive battery status information.

[0041] S102, a multi-view encoder based on deep learning pre-processes the multimodal monitoring data to obtain a multimodal fusion feature vector;

[0042] Specifically, the multimodal monitoring data is first standardized to unify the data of different modalities to the same scale and range, eliminating the impact of dimensional differences. Then, the standardized multimodal monitoring data is input into a pre-trained deep learning multi-view encoder. The encoder consists of multiple sub-encoders, each corresponding to a data modality, such as a voltage sub-encoder, a current sub-encoder, a temperature sub-encoder, and an internal resistance sub-encoder. Each sub-encoder uses a deep learning model such as a convolutional neural network (CNN) or a long short-term memory network (LSTM). Through multi-layer nonlinear transformation and feature extraction, the original time series data is mapped to a low-dimensional feature space to obtain the feature vector of the modality.

[0043] After obtaining the feature vectors for each modality, the multi-view encoder performs a weighted fusion of these feature vectors through an attention mechanism to produce a multimodal fused feature vector. This attention mechanism adaptively adjusts the weights of features from different modalities, highlighting features with strong expressiveness and high information content while suppressing redundant and noisy features, resulting in a more refined and effective feature representation.

[0044] By preprocessing multimodal monitoring data using a deep learning-based multi-view encoder, we can fully exploit the potential correlations and complementary information between different modal data, learning more abstract and high-level feature representations. This multimodal fusion feature vector is not only lower in dimensionality and more information-focused, but also has stronger semantic expression and noise robustness, helping to improve the accuracy and efficiency of subsequent state assessments.

[0045] Based on the above embodiment, as an optional embodiment, a multi-view encoder based on deep learning pre-processes the multimodal monitoring data to obtain a multimodal fusion feature vector, including:

[0046] S201, unifying the multimodal monitoring data to the same scale and distribution to obtain preprocessed multimodal monitoring data;

[0047] Specifically, for each modality's monitoring data, the mean and standard deviation are first calculated. Then, the data is scaled using a normalization (Z-score) method. The mean of the modality is subtracted from each data point, and then divided by the standard deviation to obtain normalized data with a mean of 0 and a standard deviation of 1. This method unifies the numerical ranges of data from different modalities to the same scale, eliminating the effects of dimension.

[0048] Next, the standardized data undergoes a distribution transformation. The distribution characteristics of each modality, such as skewness and kurtosis, are observed to determine whether they conform to a normal distribution. For data that deviates significantly from a normal distribution, methods such as the Box-Cox transformation are used to correct the distribution, bringing it as close to a normal distribution as possible. This distribution transformation improves the statistical properties of the data and facilitates the learning and generalization of deep learning models.

[0049] After completing the scale and distribution transformations, preprocessed multimodal monitoring data is obtained. This data achieves consistency in numerical range and distribution characteristics, significantly reducing heterogeneity between different modalities. This unified data representation not only facilitates subsequent feature extraction and fusion, but also accelerates the training convergence of deep learning models, improving the learning effect and robustness of the models.

[0050] S202 , performing multi-view feature extraction on the pre-processed multimodal monitoring data to obtain a multimodal fusion feature vector.

[0051] Specifically, for the preprocessed multimodal monitoring data, three parallel feature extractors are designed to extract features from the time domain, frequency domain, and spatial domain perspectives. The time domain feature extractor uses a long short-term memory network (LSTM) to recursively learn time series data to extract the dynamic change characteristics of the data and characterize the evolution of the battery status over time. The frequency domain feature extractor uses a convolutional neural network (CNN) to extract the frequency domain features of the data through convolution and pooling operations on the spectrum graph, capturing abnormal patterns of the battery status on the spectrum. The spatial domain feature extractor uses a graph convolutional network (GCN) to extract the spatial topological features of the data through convolution operations on the battery structure graph, and explore the correlation patterns between the components within the battery.

[0052] During feature extraction, different feature extraction strategies are employed for monitoring data of different modalities. For one-dimensional time series data such as voltage and current, a time-domain feature extractor is primarily used. For frequency-domain data such as impedance spectra and EIS, a frequency-domain feature extractor is primarily used. For structured data such as battery structural parameters and material properties, a spatial-domain feature extractor is primarily used. Through targeted feature extraction, the characteristics of different modal data can be fully utilized to extract more effective and comprehensive state features.

[0053] After completing multi-view feature extraction, feature vectors from the three perspectives are obtained. These feature vectors are combined using a specific splicing or fusion method to form a multimodal fusion feature vector. This feature vector integrates feature information from the time, frequency, and spatial domains, comprehensively characterizing the battery state from multiple dimensions and providing stronger representation and differentiation capabilities. This multi-view fusion feature can effectively improve the accuracy of subsequent state assessment and prediction, enhancing the model's generalization and robustness.

[0054] S103: Perform causal chain reasoning on the multimodal fusion feature vector based on the preset fused lead-acid battery knowledge graph to obtain a series of battery status causal chains, and perform status assessment on the lead-acid battery based on the battery status causal chains to obtain real-time status assessment results;

[0055] Specifically, a fused lead-acid battery knowledge graph is first constructed. This knowledge graph formally represents the key components, degradation mechanisms, and health status of lead-acid batteries in the form of an ontology. Causal reasoning rules are then constructed through semantic associations between ontologies. For example, the knowledge graph includes causal reasoning rules such as "lead sulfate crystallization -> active material shedding -> capacity decay." The multimodal fusion feature vector obtained in step S102 is then mapped into the knowledge graph. Through feature matching and semantic mapping, the nodes and relationships most relevant to the current feature vector are identified.

[0056] Next, using the matched node as a starting point, the algorithm performs forward and backward reasoning using pre-set causal reasoning rules, resulting in a series of causal chains for the battery's status. For example, based on the causal chain of "high temperature -> electrolyte evaporation -> increased resistance," it can be inferred that the battery is likely exposed to high temperatures, causing electrolyte evaporation and resulting in increased internal resistance. These causal chains reflect the battery's health and degradation mechanisms from different perspectives, providing important diagnostic and predictive value.

[0057] Finally, based on the resulting battery health causal chain, the system comprehensively analyzes various battery health indicators to produce a real-time health assessment. Specifically, based on the key nodes and paths in the causal chain, the system determines the battery's current health state (normal, mildly degraded, moderately degraded, or severely degraded) and assigns a corresponding health metric score. Furthermore, based on the degradation mechanisms and trends reflected in the causal chain, the system predicts the battery's future degradation trends and estimates its remaining service life.

[0058] Based on the above embodiment, as an optional embodiment, causal chain reasoning is performed on the multimodal fusion feature vector based on a preset fused lead-acid battery knowledge graph to obtain a series of battery status causal chains, including:

[0059] S301, linking and matching the multimodal fusion feature vector with entities and attributes in a preset fusion lead-acid battery knowledge graph to obtain a mapping relationship network between the multimodal fusion feature vector and the preset fusion lead-acid battery knowledge graph;

[0060] Specifically, we first constructed a fused knowledge graph for lead-acid batteries, encompassing concepts, entities, attributes, and relationships related to battery health assessment. This knowledge graph organizes domain knowledge in an ontology format, encompassing knowledge entities and relationships across multiple areas, such as battery structure, materials, performance parameters, degradation mechanisms, and failure modes. By extracting and integrating knowledge from domain literature, expert knowledge, and experimental data, we constructed a comprehensive, accurate, and consistent knowledge representation and reasoning framework.

[0061] Each feature in the multimodal fusion feature vector is then semantically linked to entities and attributes in the knowledge graph. By calculating the similarity between features, entities, and attributes, the most relevant knowledge elements for each feature are identified, and a mapping relationship between features and knowledge is established. For example, the capacity feature can be linked to entities such as "rated capacity" and "capacity decay" in the knowledge graph, while the internal resistance feature can be linked to entities such as "ohmic internal resistance" and "polarization internal resistance." This semantic linking connects data-driven feature representation with knowledge-driven concept representation, achieving the fusion of feature space and knowledge space.

[0062] After linking and matching features with knowledge, a mapping network is generated, representing the multimodal fusion feature vectors and the fused lead-acid battery knowledge graph. This mapping network uses the knowledge graph as its framework and feature vectors as its nodes, organically integrating the two through semantic links. Each feature node establishes a semantic association with related entities and attributes in the knowledge graph through links, reflecting the physical meaning and causal role of the feature in the battery health status. This knowledge graph-based feature linking provides important prior knowledge and constraints for subsequent causal reasoning, helping to improve the accuracy and interpretability of reasoning.

[0063] S302: searching for causal chain paths based on the mapping relationship network to obtain a set of candidate causal chains, and calculating the causal strength of the causal chains in the set of candidate causal chains to obtain the causal strength of each causal chain in the set of candidate causal chains;

[0064] Specifically, a causal chain path search is first performed based on the mapping relationship network. Starting from the feature node in the multimodal fusion feature vector, graph search algorithms such as breadth-first search are used to search for all possible paths from the starting node to the target state node in the knowledge graph. These paths represent the causal chains that the feature node may influence through the entities and relationships in the knowledge graph. By setting constraints such as path length and direction, the depth and breadth of the search are controlled, balancing the integrity and complexity of the causal chain. During the search process, the paths are scored using the relevance weights of the semantic links, prioritizing paths with high relevance. Ultimately, a set of candidate causal chains is obtained, where each causal chain represents a possible causal influence path.

[0065] Then, for each causal chain in the set of candidate causal chains, its causal strength is calculated. Causal strength measures the degree and likelihood of a causal chain's impact on state changes and requires comprehensive consideration of multiple factors, including the causal chain's semantic relevance, knowledge reliability, and data support. First, the semantic relevance score of the entire causal chain is calculated using the weights of the semantic links in the causal chain. Second, the reliability of the domain knowledge on which the causal chain relies is assessed by combining the confidence levels of entities and relationships in the knowledge graph. Third, the causal chain is instantiated using multimodal monitoring data, and the data support for the causal chain is calculated. Finally, by combining these factors and using methods such as weighted summation, a causal strength value is obtained for each causal chain. A higher causal strength indicates a greater impact of the causal chain on state changes and a higher likelihood of occurrence.

[0066] After searching for causal chains and calculating causal strength, a set of candidate causal chains with strength attributes is generated. This set reveals the possible causes and impact paths of state changes from multiple perspectives, providing a rich causal basis for state assessment. By analyzing the strength rankings of different causal chains, the key influencing factors most likely to have led to the current state can be identified, guiding subsequent decision-making and intervention. Furthermore, the semantic interpretability of causal chains makes assessment results more transparent and trustworthy, facilitating expert analysis and verification of model outputs.

[0067] S303 , sorting the causal chains based on causal strength, and selecting the top N causal chains with the largest causal strength as a series of battery status causal chains.

[0068] Specifically, all causal chains in the candidate causal chain set are first sorted in descending order by causal strength. Causal strength reflects the degree and likelihood of each causal chain's impact on the state change. A higher causal strength indicates a more likely key cause of the current state. Descending sorting prioritizes causal chains with higher causal strengths, placing them first and lower ones later. This sorting approach helps quickly identify the most important and relevant causal chains, facilitating subsequent top-N selection. Then, from the sorted causal chain set, the top N causal chains with the highest causal strengths are selected as the battery state causal chains. The value of N can be set based on the specific problem and requirements; a smaller value is typically chosen to limit the number of selected causal chains. These N causal chains represent the causal paths that have the greatest impact on the current battery state and are most likely to cause the state change, reflecting the primary influencing factors and key processes of the state change. By focusing on these key causal chains, the causal mechanisms underlying the state change can be highly summarized and condensed, improving the efficiency and accuracy of subsequent state assessment and decision-making.

[0069] After causal chain sorting and top-N selection, a series of battery status causal chains are generated. These chains are prioritized by importance and relevance, using causal strength as weight, and represent the key influencing paths of battery status changes. These causal chains can be presented to experts or decision makers through visualization or natural language description, illustrating the primary causes and influencing factors of the current battery status. This extraction and interpretation of causal chains helps enhance the comprehensibility and credibility of status assessment results, providing clear guidance and basis for subsequent maintenance decisions and interventions.

[0070] Based on the above embodiment, as an optional embodiment, before performing causal chain reasoning on the multimodal fusion feature vector based on the preset fused lead-acid battery knowledge graph to obtain a series of battery status causal chains, the method further includes:

[0071] Obtaining lead-acid battery knowledge text related to lead-acid batteries, and building a lead-acid battery ontology knowledge base based on the lead-acid battery knowledge text;

[0072] Obtain sample data of lead-acid battery monitoring materials, and map the sample data of lead-acid battery monitoring materials to the ontology framework based on the lead-acid battery ontology knowledge base to form a semantic network containing concept nodes, attribute edges, and relationship edges, and obtain the preset fused lead-acid battery knowledge graph.

[0073] Specifically, knowledge texts related to lead-acid batteries were obtained, including textbooks, papers, patents, standards, reports, and other document formats. Text mining and natural language processing were performed on the obtained knowledge texts to extract key concepts, attributes, relationships, and other elements from the text. Using ontology construction methods such as Ontology Development 101 and IDEF5, the extracted knowledge elements were organized into a hierarchical, structured conceptual system. Based on this conceptual system, attribute associations and semantic relationships between concepts were defined to form a complete ontology graph. The ontology graph was formalized using standard ontology languages ​​such as OWL and RDF and stored in an ontology database. Through the above process, an ontology knowledge base covering core concepts in the lead-acid battery field was constructed, characterizing knowledge on various aspects of lead-acid batteries, including their composition, structure, material properties, operating principles, and failure mechanisms, forming a conceptual paradigm and semantic specification for describing lead-acid batteries.

[0074] Next, various monitoring materials and sample data from lead-acid battery monitoring sites are collected, including multi-source heterogeneous data such as production logs, process parameters, online monitoring data, and energy spectrum images. These monitoring materials and sample data are semantically annotated and linked, mapping them to the conceptual framework of the ontology knowledge base. Structured data such as parameters and logs can be directly mapped to the corresponding data type attributes in the ontology. For unstructured data such as text and images, named entity recognition and relationship extraction techniques are used to identify key concepts and relationships within the text and images, linking them to corresponding concept nodes and relationship edges in the ontology knowledge base. Through semantic mapping and linking, prior knowledge is integrated with data instances, forming a semantic network consisting of concept nodes, attribute edges, and relationship edges. Concept nodes represent core concepts involved in battery health management, such as physical quantities, components, materials, and processes; attribute edges characterize the specific values ​​assigned to concept attributes; and relationship edges characterize the semantic associations between concepts. This integrated semantic network becomes the pre-defined integrated lead-acid battery knowledge graph.

[0075] Based on the above embodiment, as an optional embodiment, a state assessment of a lead-acid battery is performed based on a battery state causal chain to obtain a real-time state assessment result, including:

[0076] S401, matching the multimodal monitoring data with the battery status causal chain to obtain key indicator values ​​in the multimodal monitoring data, and comparing and calculating the key indicator values ​​with corresponding state nodes in the battery status causal chain to obtain the degree of deviation between the key indicator values ​​and a preset state threshold;

[0077] Specifically, the multimodal monitoring data is first matched with the battery status causal chain. Each causal chain is traversed to identify the associated monitoring data types and features. For example, the capacity decay causal chain may correspond to monitoring data such as voltage, current, and temperature, while the internal resistance increase causal chain may correspond to monitoring data such as voltage, temperature, and EIS. Through semantic mapping and data type matching, a correspondence is established between the state nodes in the causal chain and the features in the monitoring data. This matching process links the abstract causal chain with the specific monitoring data, laying the foundation for subsequent indicator extraction and comparative calculation.

[0078] Then, the key indicator values ​​related to the causal chain are extracted from the matching monitoring data. For each state node, the key monitoring indicators that best reflect the state are identified, such as capacity, internal resistance, voltage, etc. Statistical analysis, signal processing and other methods are used to calculate the real-time values ​​or statistical values ​​of these key indicators from the original monitoring data. For example, the capacity value is extracted from the charge and discharge curve, the internal resistance value is extracted from the EIS spectrum, and the voltage mean and fluctuation range are extracted from the voltage time series. The extracted key indicator values ​​represent the actual performance of the battery at the corresponding state node and quantify the key characteristics of the battery state.

[0079] Next, the extracted key indicator values ​​are compared and calculated with the preset thresholds of the corresponding state nodes in the battery state causal chain. Each state node has a normal operating range or threshold, which reflects the health level of the state. By comparing the size relationship between the key indicator value and the preset threshold, the degree to which the current battery state deviates from the normal level is judged. For example, if the capacity indicator value is lower than the preset lower capacity threshold, it indicates that the battery capacity has severely decayed; if the internal resistance indicator value is higher than the preset upper internal resistance threshold, it indicates that the battery internal resistance has increased significantly. By calculating the difference or ratio between the key indicator value and the threshold, the indicator deviation, that is, the degree of abnormality of the battery at the corresponding state node, can be quantitatively characterized.

[0080] Based on the above embodiment, as an optional embodiment, the sample data of lead-acid battery inspection materials is mapped to the ontology framework based on the lead-acid battery ontology knowledge base to form a semantic network containing concept nodes, attribute edges, and relationship edges, thereby obtaining a preset fused lead-acid battery knowledge graph, including:

[0081] Preprocess the sample data of lead-acid battery monitoring information to obtain lead-acid battery monitoring data;

[0082] Specifically, data cleansing was performed on sample lead-acid battery inspection data. The data was examined for outliers, missing values, and other noisy data. Outliers were identified and addressed using algorithms such as anomaly detection and isolation forests. Missing values ​​were estimated and filled using methods such as mean filling and regression interpolation. Low-quality and inconsistent data was filtered and corrected to improve data validity and reliability. Furthermore, data consistency was checked to identify conflicts and contradictions between data from different sources and formats. Data fusion and knowledge reasoning techniques were used to eliminate data conflicts and ensure data consistency. The cleaned inspection data sample was then transformed and standardized. Unstructured data such as text and images was converted into structured attribute values. For example, keywords in text were extracted as attribute labels, and regional features in images were extracted as attribute vectors. Numerical attributes were normalized and dimensionless to eliminate the impact of differences in dimension and numerical range. Categorical attributes were encoded, converting qualitative descriptions into quantitative representations. Through attribute valorization and quantification, a structured representation of the original inspection data sample was achieved.

[0083] After data cleaning and conversion, the structured inspection data is semantically annotated. Based on the concepts, attributes, and relationships in the lead-acid battery ontology knowledge base, the attributes of the inspection data are semantically annotated. Attribute names are mapped to ontology concepts, and attribute values ​​are mapped to concept instances, forming semantically annotated knowledge. Using techniques such as ontology reasoning and semantic similarity calculation, the semantic consistency of the annotated knowledge with the ontology knowledge base is checked, and ambiguous or conflicting annotation results are disambiguated and corrected. This annotation imparts clear semantics to the inspection data, paving the way for subsequent knowledge integration.

[0084] Finally, the semantically annotated inspection data is semantically linked and fused according to the ontology, forming a semantic network consisting of concept nodes, attribute edges, and relationship edges. The annotated attribute items are linked to the ontology concept nodes, with the attribute names serving as attribute edges and the attribute values ​​serving as attribute edge weights. The annotated concept instances are linked to the corresponding concept nodes and, based on the relationship edges defined in the ontology, semantic associations are established with other concept instances. Through these links, the dispersed inspection data is connected into a tightly connected semantic network. This fused semantic network becomes the pre-defined fused lead-acid battery knowledge graph.

[0085] Based on the lead-acid battery ontology knowledge base, data mapping is performed on the lead-acid battery monitoring data. The entities in the lead-acid battery monitoring data are mapped to the corresponding ontology concept nodes in the ontology knowledge base, and the attributes in the structured data are mapped to the attribute edges of the ontology concept nodes.

[0086] Specifically, preprocessing is performed on sample data from lead-acid battery monitoring documentation. Sample data typically comes from various stages of battery production, use, and maintenance, and includes structured parameters and log data, as well as unstructured text and image data. This heterogeneous data undergoes unified cleaning, filtering, and conversion. For structured data, data quality is improved through methods such as data completion, outlier correction, and consistency verification. For unstructured data, key information elements are extracted using techniques such as text parsing and image segmentation. Following data cleansing, data type conversion and normalization are performed on the monitoring data, based on the data graph definitions in the ontology knowledge base, to ensure that it meets the data format requirements of the ontology mapping. Furthermore, security measures such as data desensitization and access control are implemented to ensure the privacy and confidentiality of the sample data during processing. Through a series of preprocessing operations, the dispersed and heterogeneous sample data is converted into unified and standardized lead-acid battery monitoring data, providing a high-quality data foundation for ontology mapping. Data mapping is then performed on the lead-acid battery monitoring data based on the ontology knowledge base. Each data item in the audit data is traversed to identify its entity type, such as battery, positive electrode, electrolyte, charge / discharge process, etc. The identified entities are mapped to ontology concept nodes of the same semantic type in the ontology knowledge base. For example, "Lead-acid Battery 12#" in the audit data is mapped to an instance of the "Lead-acid Battery" concept node in the ontology knowledge base. Based on entity mapping, attribute information of the entities in the audit data is extracted and mapped to attribute edges of the corresponding concept nodes. For example, attributes such as the production date and rated capacity of "Lead-acid Battery 12#" are mapped to data type attributes of the "Lead-acid Battery" concept. Through entity and attribute mapping, key information elements in the audit data are semantically associated with the concept system in the ontology knowledge base and integrated into the semantic network of the knowledge base. Furthermore, using the relationship types defined in the ontology knowledge base, semantic relationships between concept nodes in the audit data are identified and relationship edges are formed. For example, the "include" relationship between "Lead-acid Battery 12#" and "Positive Plate A" and the "participate" relationship between "Lead-acid Battery 12#" and "Charge / Discharge Record X" are identified. By linking entities in the monitoring data through relationship edges, a tightly connected and semantically rich knowledge network is formed. This completes the semantic mapping of lead-acid battery monitoring data to the ontology knowledge base, forming a fused semantic network consisting of concept nodes, attribute edges, and relationship edges—the resulting fused lead-acid battery knowledge graph.

[0087] Based on the lead-acid battery ontology knowledge base, relationship extraction is performed on the lead-acid battery monitoring data, the semantic relationship between the named entities in the lead-acid battery monitoring data is extracted, and the semantic relationship is mapped into the relationship edge between the ontology concept nodes;

[0088] Specifically, based on the lead-acid battery ontology knowledge base, relationship extraction is performed on the lead-acid battery monitoring data. Each data item in the monitoring data is scanned to identify the named entities that appear therein, such as "lead-acid battery 12#", "positive plate A", "charge and discharge record X", etc. Then, the contextual information between the named entities is analyzed, such as entity co-occurrence, word order relationship, etc., and natural language processing technologies such as syntactic analysis and dependency analysis are used to identify the semantic relationship between the entities and form relationship knowledge in the form of triples, such as ("lead-acid battery 12#", "includes", "positive plate A"), ("lead-acid battery 12#", "participates", "charge and discharge record X"), etc. In order to improve the accuracy and coverage of relationship extraction, this embodiment adopts a multi-strategy relationship extraction method based on rules, pattern matching, and machine learning. The rule-based method customizes a batch of high-confidence relationship extraction rules for quickly and accurately extracting common explicit relationships. A pattern matching approach was used to construct a pattern library for relation extraction. Pattern matching was used to discover relational patterns within sentences and extract implicit and complex semantic relationships. A machine learning approach employed a neural network model with an attention mechanism to extract relations through end-to-end relation classification, improving the generalization and robustness of relation extraction. Through multi-strategy relation extraction, rich relational knowledge was obtained from lead-acid battery inspection data.

[0089] The extracted semantic relationships are then mapped as relationship edges between ontology concept nodes. Each relationship triple is traversed, and its subject and object entities are mapped to the constructed ontology concept nodes. The relationship type between the two concept nodes is determined based on the predicates in the triple. For example, for the relationship triple ("Lead-acid Battery 12#", "Contains", "Positive Plate A"), querying the graph definition of the ontology knowledge base reveals that there is a "Contains" relationship between "Lead-acid Battery" and "Positive Plate". Therefore, a "Contains" type relationship edge is added between the "Lead-acid Battery 12#" node and the "Positive Plate A" node. For another example, for the relationship triple ("Lead-acid Battery 12#", "Participate", "Charge and Discharge Record X"), querying the graph reveals that there is a "Participate" relationship between "Lead-acid Battery" and "Charge and Discharge Record", so a "Participate" type relationship edge is added. All extracted semantic relations are mapped and converted into relationship edges between ontology concept nodes, which are then integrated into the existing semantic network containing concept nodes and attribute edges, ultimately forming a fused semantic network with closer associations and richer semantics, namely the preset fused lead-acid battery knowledge graph.

[0090] Based on ontology concept nodes, attribute edges and relationship edges, a preset fusion lead-acid battery knowledge graph is constructed.

[0091] Specifically, the data model of the graph specifies the types of components of the graph and their logical relationships, and is the basis for constructing the graph. This embodiment uses a property graph (PropertyGraph) as the data model of the graph. The property graph consists of nodes (Node) and edges (Edge), and both nodes and edges can contain several properties (Property). Each ontology concept node corresponds to a node in the property graph, and the attributes of the node are the attribute edges of the concept. Each relationship edge corresponds to a directed edge between nodes in the property graph, and the attributes of the edge are the metadata of the relationship, such as relationship type, confidence, etc. The property graph is intuitive and flexible, and can express concepts, attributes, relationships and other elements at the same time. It is suitable for modeling complex semantic networks in the field of lead-acid batteries.

[0092] After defining the graph data model, we begin constructing the lead-acid battery knowledge graph step by step. First, we add ontology concept nodes to the graph. We traverse each ontology concept node and create a corresponding node in the graph. The node's label is the name of the concept. For different instances of the same concept, such as different instances of "lead-acid battery," we create multiple nodes, distinguishing them by unique identifiers (such as URIs). Next, we add attribute edges to the graph. We traverse the attribute edges of each ontology concept node and add them as attributes to the corresponding node. The attribute key is the attribute name, and the attribute value is the attribute's numeric value. Next, we add relationship edges to the graph. We traverse each relationship triple and create a directed edge between its subject and object nodes. The edge label is the relationship type. If necessary, we add attributes to the relationship edges, such as metadata such as the relationship's confidence and source. Through these steps, we integrate all ontology concept nodes, attribute edges, and relationship edges into a property graph, initially constructing the pre-defined integrated lead-acid battery knowledge graph.

[0093] To optimize the quality and performance of the knowledge graph, the initially constructed graph also requires necessary processing. On the one hand, noise and redundancy in the graph need to be eliminated, such as by removing duplicate nodes, deleting invalid edges, and correcting incorrect attribute values, thereby improving the graph's accuracy and simplicity. On the other hand, the graph's semantic information needs to be enriched, such as by adding synonyms and abbreviations to concepts, exploring the hierarchical relationships between concepts, and linking to external knowledge bases, thereby enhancing the graph's semantic expressiveness. Furthermore, the graph's storage and retrieval performance needs to be optimized, employing technologies such as graph databases and indexing mechanisms to improve query efficiency and scalability. After a series of optimization processes, a high-quality, high-performance knowledge graph for lead-acid batteries was ultimately obtained.

[0094] S402, determining the state node of the current lead-acid battery in the battery state causal chain according to the degree of deviation;

[0095] Specifically, the battery's abnormality level at each state node is determined based on the deviation of key indicators at each state node. The deviation of each state node is compared with a preset abnormality level threshold to determine the abnormality level of that node. For example, multiple levels can be set, such as normal, mildly abnormal, moderately abnormal, and severely abnormal, with corresponding deviation thresholds set for different key indicators and state nodes. If the deviation of a state node exceeds the corresponding abnormality level threshold, the battery is considered to have experienced an abnormal state of the corresponding level at that node. By grading the deviation of all state nodes, the distribution of battery abnormalities across the entire causal chain can be comprehensively assessed. Then, the abnormality level of each state node is comprehensively analyzed to determine the battery's current overall state stage. Given that battery state evolution is a gradual process, the state nodes in the causal chain have a sequential order and causal dependencies. Typically, a battery starts in a normal state and gradually progresses through early, mid, and late stages of decline, ultimately reaching a failure state. Based on the abnormality level distribution of each state node, combined with the topological structure and logical relationships of the causal chain, the most likely current state stage of the battery can be inferred. For example, if an early-stage degradation node is abnormal but a mid- to late-stage node is normal, the battery may be in the early-stage degradation phase. If all degradation nodes are abnormal, the battery may have entered the late-stage degradation phase. This integrated reasoning establishes temporal and causal connections between status nodes, forming a dynamic assessment of the battery's overall state.

[0096] S403, calculating the state transition probability of each state node in the battery state causal chain;

[0097] Specifically, the method first extracts feature data related to state transitions from a battery operation history database. A representative sample of batteries is selected, and various monitoring data over their entire lifecycle, such as voltage, current, temperature, capacity, and internal resistance, are collected. For each battery sample, the main state nodes it experiences are identified, such as normal, early-stage decay, mid-stage decay, and late-stage decay, and the start and end times and duration of each state node are recorded. Key characteristic indicators at each state node, such as capacity decay rate, internal resistance growth rate, and mean temperature, are then extracted to form a state node feature vector. By summarizing and analyzing the state node features of multiple battery samples, a state transition feature dataset is generated. This state transition feature dataset is then used to train a state transition probability model. Common modeling methods include Markov chains, Bayesian networks, and survival analysis. Taking the Markov chain as an example, the battery state causal chain is abstracted as a discrete-time Markov chain, with state nodes corresponding to states of the chain and state transitions corresponding to state transitions. Using methods such as maximum likelihood estimation, the parameters of the state transition probability matrix are learned from state transition feature data, resulting in a Markov chain model that describes the state transition patterns. Similarly, Bayesian networks construct conditional dependencies between state nodes, and survival analysis estimates the distribution function of state durations to establish corresponding state transition probability models. Through model training and validation, the model with the best predictive performance is selected as the final state transition probability model.

[0098] After the state transition probability model is built, it can be used to calculate the state transition probability of each state node. Given the state node the battery is currently in, the trained state transition probability model is used to infer the probability of the battery transitioning from its current state to another state node based on the state transition feature vector.

[0099] S404 , performing real-time state evaluation based on the state node and state transition probability of the current lead-acid battery state in the battery state causal chain to obtain a real-time state evaluation result.

[0100] Specifically, the current battery state node and the transition probabilities for each state node are first obtained from steps S402 and S403. The current battery's real-time monitoring data is input into the state node determination model to identify the battery's position in the causal chain and determine its state node, such as normal state or early degradation state. Simultaneously, the feature vector of the current state node is input into the state transition probability model to calculate the probability distribution of transitions from the current state node to other state nodes. These two types of information describe the battery's health status from static and dynamic perspectives, respectively. Then, based on the state nodes and transition probabilities, a real-time state assessment indicator system is constructed. A series of state assessment indicators are designed for various key battery properties, such as capacity, lifespan, and efficiency, to quantitatively describe the battery's performance in these areas. For example, the capacity state indicator can be measured as the ratio of current capacity to rated capacity, while the lifespan indicator can be measured as the ratio of cycles used to rated cycle life. Each evaluation indicator is associated with a specific state node and reflects the battery's performance level in that state. By mapping the current battery's state node to each evaluation indicator, a set of indicator values ​​reflecting the current battery's real-time state is obtained.

[0101] Next, the real-time state assessment indicators are dynamically weighted using state transition probabilities. Starting from the current state node, the probability distribution of each state node that the battery may reach in the future is calculated based on the state transition probabilities. The probabilities of these state nodes are used as weights and weighted averaged with the assessment indicator values ​​for the corresponding state to obtain a set of expected state assessment indicator values ​​that account for state transition risks. These expected values ​​reflect the evolution of the battery state over a period of time and depict the expected path of changes in the battery's health level.

[0102] Based on the above embodiment, as an optional embodiment, before inputting the multimodal fusion feature vector into a preset battery performance degradation trend estimation model to obtain the battery performance degradation trend, the method further includes:

[0103] S501, collecting lead-acid battery degradation history data, preprocessing the lead-acid battery degradation history data, and obtaining preprocessed lead-acid battery degradation history data;

[0104] Specifically, first determine the collection indicators and data sources for historical degradation data based on the needs of estimating battery performance degradation trends. Historical battery degradation data generally includes performance parameters such as capacity, internal resistance, self-discharge rate, charge and discharge efficiency, as well as operating parameters such as voltage, current, and temperature. The data source can be the battery management system's monitoring database, operation and maintenance record archives, or third-party test reports. Based on the different characteristics of the data source, design corresponding data collection interfaces and parsing rules, such as database query statements and text parsing scripts, to extract structured historical degradation data from heterogeneous data sources. Data from different battery samples and different time periods require data cleaning and alignment to ensure data integrity and consistency.

[0105] Then, the collected degradation history data is preprocessed to improve data quality and highlight degradation trend characteristics. Common preprocessing operations include outlier removal, noise filtering, data normalization, feature construction, etc. Outlier removal identifies and removes abnormal data points that deviate significantly from the normal range by setting thresholds or statistical criteria. Noise filtering uses signal processing methods such as moving average and Kalman filtering to smooth short-term fluctuations and extract long-term trends. Data normalization maps parameters of different dimensions to the same scale to eliminate the impact of differences in value ranges. Feature construction is based on the original parameters and derives some composite features that reflect degradation rate and acceleration factor, such as capacity attenuation rate, internal resistance growth rate, etc. After a series of preprocessing operations, a high-quality degradation history data set is obtained.

[0106] After preprocessing the degradation history data, the dataset needs to be further partitioned to construct training and test sets for the degradation trend estimation model. The preprocessed degradation history data is divided into training and test sets according to a specific ratio, such as 8:2. The training set is used for parameter learning and optimization of the degradation trend estimation model, while the test set is used to evaluate the model's generalization performance and prediction results. To enhance sample representativeness, data partitioning should ensure a balanced distribution of samples with different degradation modes and operating conditions in the training and test sets. Furthermore, to verify the stability of the model, methods such as cross-validation can be used to randomly partition the dataset multiple times to evaluate the distribution of model performance.

[0107] S502, performing feature extraction on the pre-processed lead-acid battery degradation history data to obtain degradation feature data;

[0108] Specifically, based on the degradation mechanism and failure mode of lead-acid batteries, the degradation feature extraction method and indicator system are first determined. Common degradation feature extraction methods include time domain analysis, frequency domain analysis, and wavelet analysis. Time domain analysis characterizes the overall trend and fluctuation level of the degradation process by calculating statistics of time series data, such as mean, variance, and peak value. Frequency domain analysis uses techniques such as Fourier transform to convert time series data into the frequency domain to extract the periodicity and frequency characteristics of the degradation process. Wavelet analysis extracts degradation features at different time scales by performing multi-scale decomposition of the time series, capturing local details and mutation points of the degradation process. For typical degradation modes such as capacity decay, internal resistance growth, and increased self-discharge, a series of characteristic indicators are designed to quantitatively evaluate key attributes of the degradation process, such as rate, acceleration, and step change.

[0109] Then, using the defined feature extraction method and indicator system, the preprocessed degradation history data is characterized. Taking time-domain analysis as an example, statistical characteristics such as the initial degradation value, final degradation value, degradation mean, and degradation standard deviation are calculated for time series data of key performance parameters such as capacity and internal resistance. This characterizes the starting point, end point, central trend, and degree of dispersion of the entire degradation process. For derived parameters such as degradation rate, their maximum, minimum, and average values ​​are calculated to reflect the extreme and overall levels of degradation speed. Similarly, frequency-domain analysis and wavelet analysis can extract the spectral characteristics, energy distribution characteristics, and multi-scale decomposition coefficients of the degradation process, characterizing the dynamic characteristics of the degradation trend from different perspectives.

[0110] S503, performing degradation pattern recognition on the pre-processed lead-acid battery degradation history data to obtain a degradation pattern;

[0111] Specifically, based on domain knowledge of lead-acid batteries, we first summarize common degradation pattern types and their identification rules. Common degradation patterns include linear, exponentially accelerating, and multi-stage degradation patterns. A linear degradation pattern describes a battery performance parameter that exhibits a nearly linear decline, with a roughly constant degradation rate. An exponentially accelerating degradation pattern describes an exponentially increasing degradation rate, accelerating the degradation process and significantly shortening the time to end-of-life. The multi-stage degradation pattern divides the entire degradation process into several stages, with distinct degradation rates in different stages, exhibiting piecewise linear or exponential characteristics. Based on the typical characteristics of different degradation patterns, we design a series of identification rules, such as linear fit, exponential goodness of fit, and segmented inflection point locations, for qualitative identification and quantitative evaluation of degradation patterns. Then, using these degradation pattern identification rules, we perform degradation pattern recognition on preprocessed degradation history data. For each degradation history data series, we fit a linear and exponential degradation trend line, and calculate corresponding goodness-of-fit metrics, such as the coefficient of determination and root mean square error. Comparing the goodness-of-fit index with a preset threshold determines whether the data series belongs to a linear or exponential degradation model. For data series with a lower goodness-of-fit, a multi-stage degradation model is considered for segmented fitting. Optimization methods such as dynamic programming and greedy algorithms are used to search for optimal segmentation points along the timeline, dividing the degradation process into several linear or exponential degradation sub-stages. Degradation pattern recognition identifies the degradation pattern type and key parameters for each historical degradation data entry, such as degradation rate, acceleration factor, and segmentation nodes.

[0112] S504, using battery domain knowledge, constructing a degradation mechanism battery state causal chain reflecting the degradation mechanism;

[0113] Specifically, the key physical quantities involved in the degradation process of lead-acid batteries and their internal connections are sorted out. The battery degradation process is usually manifested as capacity decay and increased internal resistance, and these macroscopic performance degradation phenomena are an integrated reflection of changes in multiple microscopic states. At the positive electrode, the shedding of active substances, sulfate crystallization, plate corrosion, etc. will lead to a decrease in active substances and a decrease in active specific surface area. At the negative electrode, the shedding of active substances, sulfate crystallization, plate passivation, etc. will lead to a decrease in active specific surface area and deterioration of conductive performance. In the electrolyte, water evaporation and changes in acid composition will lead to a decrease in conductivity. Combining the relationship between these microscopic state quantities and macroscopic performance energies, a multi-layer state causal chain covering electrodes, electrolytes, and macroscopic performance is preliminarily constructed, reflecting the correspondence between performance degradation phenomena and internal state changes.

[0114] S505: Construct a battery performance degradation trend estimation model based on the degradation characteristic data, the degradation mechanism battery status causal chain, and the degradation pattern, and construct a battery performance degradation trend estimation model based on the lead-acid battery degradation history data. When the prediction accuracy of the battery performance degradation trend estimation model is greater than a preset accuracy threshold, a preset battery performance degradation trend estimation model is obtained.

[0115] Specifically, the basic framework of a battery performance degradation trend estimation model is established based on the causal chain of degradation mechanism states. Key state variables in the causal chain serve as the model's state variables. These state variables are causally transferred and iteratively updated through the quantitative relationships within the causal chain, ultimately outputting battery degradation performance indicators. Building on this causal chain, data-driven degradation pattern information is introduced. The quantitative relationships between state variables are defined in segments, tailored to the characteristics of different degradation stages, enabling a phased characterization of degradation trends. Operating conditions are used as model input variables, and an operating condition modulation function is introduced to dynamically adjust the influence weights between state variables, enabling the model to adapt to changing operating conditions. The battery performance degradation model, constructed based on the mechanism framework, degradation patterns, and operating condition modulation, combines mechanistic explanatory power with data flexibility. Degradation characteristic data is then used to identify parameters and optimize the model's structure. Based on historical degradation data samples, parameter identification methods such as least squares and maximum likelihood estimation are used to fit the model's quantitative relationship parameters, such as the influence coefficients between state variables and the weight coefficients of the operating condition modulation function. Based on parameter identification, model structure optimization algorithms, such as regularization and feature selection, are used to optimize the model's structural complexity, reduce redundant state variables and causal relationships, and improve the model's generalization performance. Through parameter identification and structural optimization, the degradation trend model can best fit the historical degradation data while avoiding the risk of overfitting.

[0116] After model construction and optimization, cross-validation and other methods are used to evaluate the model's predictive accuracy. The degradation history data is divided into a training set and a test set. The model is trained using the training set data and evaluated using the test set data to obtain model prediction accuracy metrics such as root mean square error (RMS) and mean absolute percentage error (MAPE). Multiple cross-validations are performed to comprehensively evaluate the model's predictive performance and stability. When the model's prediction accuracy exceeds the preset accuracy threshold, the model is considered to have met the expected predictive performance requirements, resulting in the preset battery performance degradation trend estimation model.

[0117] S104, inputting the multimodal fusion feature vector into a preset battery performance degradation trend estimation model to obtain the battery performance degradation trend;

[0118] Specifically, a battery performance degradation trend estimation model is first constructed based on a large amount of historical operating data. This model, based on a long short-term memory (LSTM) network, characterizes the dynamic patterns of battery performance indicators over time by sequentially learning feature vectors across multiple time steps. During model training, the degradation curves of key performance indicators such as capacity and internal resistance are used as targets, and end-to-end supervised learning is performed using historical data, enabling the model to adaptively fit the degradation patterns of different batteries.

[0119] After obtaining a trained battery performance degradation trend estimation model, the multimodal fusion feature vector obtained in step S102 is used as the model input. A forward propagation calculation is performed to determine the battery's performance degradation trend over the next period of time. Specifically, based on the input feature vector, the model gradually predicts the performance indicators, such as capacity and internal resistance, for each future time step and connects these predicted values ​​to form a complete degradation curve. This degradation curve reflects the long-term degradation trend of the battery in its current state and is important for predicting the battery's remaining useful life and formulating maintenance strategies.

[0120] By using a pre-defined battery performance degradation trend estimation model for long-term prediction, we can fully exploit the degradation patterns and trends contained in historical operating data and learn the general characteristics of the battery degradation process. This data-driven approach not only overcomes the limitations of physical models in long-term prediction, but also adaptively fits degradation patterns under different operating conditions and environments, providing greater robustness and extrapolation capabilities.

[0121] S105 , performing a comprehensive evaluation on the state of the lead-acid battery based on the real-time state evaluation result and the battery performance degradation trend to obtain a comprehensive evaluation result.

[0122] Specifically, the real-time status assessment results and battery performance degradation trends are first fused at the feature level. The key nodes in the battery status causal chain obtained in step S103 are used as real-time status features, and the key time points in the battery performance degradation trend obtained in step S104 are used as long-term trend features to form a comprehensive evaluation feature set. These features reflect the battery's health status from different perspectives, encompassing static and dynamic, local, and global multi-granularity information. Then, based on the comprehensive evaluation feature set, a multi-level evaluation index system is constructed. This index system, centered on battery health status, designs evaluation indicators based on multiple dimensions, such as capacity, internal resistance, self-discharge rate, and cycle life. The indicators are divided into different levels based on their dependencies and importance. For example, the top-level indicator is battery health, the middle-level indicators include capacity retention rate and internal resistance growth rate, and the bottom-level indicators include active material shedding rate and electrolyte concentration. This hierarchical and graded indicator design allows for a comprehensive assessment of battery health from the macro to the micro level, and from the overall to the local level. Next, the analytic hierarchy process (AHP) is used to quantitatively evaluate and weightedly fuse the indicators at each level. First, battery experts were invited to compare the importance of each layer of indicators, generating a judgment matrix. The eigenvalue method was then used to calculate the maximum eigenvalue and corresponding eigenvector of the judgment matrix. The eigenvectors were normalized and used as weight vectors for the indicators. Finally, a weighted summation method was used to combine the evaluation results of each layer of indicators from bottom to top, layer by layer, to obtain a comprehensive assessment score for the battery's health status.

[0123] On the other hand, the present application also provides a battery operation status monitoring and evaluation system, such as Figure 2 , the system comprises:

[0124] Data acquisition module 1, used to collect multimodal monitoring data of the lead-acid battery during the charging and discharging process;

[0125] Feature extraction module 2, which is used for preprocessing multimodal monitoring data using a multi-view encoder based on deep learning to obtain a multimodal fusion feature vector;

[0126] The real-time status assessment module 3 is used to perform causal chain reasoning on the multimodal fusion feature vector based on the preset fusion lead-acid battery knowledge graph to obtain a series of battery status causal chains, and perform status assessment on the lead-acid battery based on the battery status causal chains to obtain real-time status assessment results;

[0127] Degradation trend prediction module 4, used to input the multimodal fusion feature vector into a preset battery performance degradation trend estimation model to obtain the battery performance degradation trend;

[0128] The comprehensive evaluation module 5 is used to comprehensively evaluate the state of the lead-acid battery based on the real-time state evaluation result and the battery performance degradation trend to obtain a comprehensive evaluation result.

[0129] Please refer to Figure 3 The present application also discloses an electronic device. Figure 3 The electronic device 300 may include: at least one processor 301 , at least one network interface 304 , a user interface 303 , a memory 305 , and at least one communication bus 302 .

[0130] The communication bus 302 is used to implement the connection and communication between these components.

[0131] The user interface 303 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0132] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0133] The processor 301 may include one or more processing cores. Using various interfaces and circuits, the processor 301 connects to various components within the server. It executes instructions, programs, code sets, or instruction sets stored in the memory 305, as well as accesses data stored in the memory 305, to perform various server functions and process data. Optionally, the processor 301 may be implemented using at least one of the following hardware forms: a digital signal processing (DSP), a field-programmable gate array (FPGA), or a programmable logic array (PLA). The processor 301 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU primarily processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing content displayed on the display; and the modem handles wireless communications. It is understood that the modem may also be implemented as a separate chip, rather than integrated into the processor 301.

[0134] Among them, the memory 305 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read~Only Memory). Optionally, the memory 305 includes a non-transitory computer-readable medium. The memory 305 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 305 may also optionally be at least one storage system located away from the aforementioned processor 301. Reference Figure 3 , the memory 305 as a computer storage medium may include an operating system, a network communication module, a user interface module and an application program of a battery operation status monitoring and evaluation method.

[0135] exist Figure 3 In the electronic device 300 shown, the user interface 303 is mainly used to provide an input interface for the user and obtain the data input by the user; and the processor 301 can be used to call the application program storing the road assessment method in the memory 305. When executed by one or more processors 301, the electronic device 300 executes one or more methods in the above-mentioned embodiments. It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should know that this application is not limited to the described order of actions, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for this application. In the above-mentioned embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0136] In the several embodiments provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interface, and the indirect coupling or communication connection of the system or unit can be electrical or other forms.

[0137] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0138] The present application also provides a computer storage medium that can store multiple instructions, which are suitable for being loaded and executed by a processor as described above. Figure 1 The road assessment method of the embodiment shown, the specific execution process can be found in Figure 1 The detailed description of the illustrated embodiment will not be repeated here.

[0139] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0140] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of this application, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of this application. The aforementioned memory includes various media that can store program code, such as USB flash drives, mobile hard drives, magnetic disks, or optical disks.

[0141] The above are merely exemplary embodiments of the present disclosure and are not intended to limit the scope of the present disclosure. In other words, any equivalent variations and modifications made in accordance with the teachings of the present disclosure are still within the scope of the present disclosure. Those skilled in the art will readily conceive of other embodiments of the present disclosure after considering the disclosure and the practical implications thereof.

[0142] This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not described herein. The description and examples are to be considered as exemplary only, and the scope and spirit of the present disclosure are to be defined by the claims.

Claims

1. A method for monitoring and evaluating battery operating status, characterized in that: The method comprises: Collect multimodal monitoring data of lead-acid batteries during the charging and discharging process; A multi-view encoder based on deep learning pre-processes the multimodal monitoring data to obtain a multimodal fusion feature vector; Performing causal chain reasoning on the multimodal fusion feature vector based on a preset fused lead-acid battery knowledge graph to obtain a series of battery status causal chains, and performing a status assessment on the lead-acid battery based on the battery status causal chains to obtain a real-time status assessment result, wherein the fused lead-acid battery knowledge graph is obtained by performing causal reasoning on the key components, degradation mechanisms, and health status of the lead-acid battery; Inputting the multimodal fusion feature vector into a preset battery performance degradation trend estimation model to obtain a battery performance degradation trend; Performing a comprehensive assessment of the state of the lead-acid battery based on the real-time state assessment result and the battery performance degradation trend to obtain a comprehensive assessment result; The multi-view encoder based on deep learning pre-processes the multi-modal monitoring data to obtain a multi-modal fusion feature vector, including: Calculating the mean and standard deviation of the multimodal monitoring data, performing a scale transformation on the multimodal monitoring data based on the mean and standard deviation using a preset Z-score method to obtain standardized data, and performing distribution correction on the standardized data to obtain preprocessed multimodal monitoring data; Performing feature extraction on the preprocessed multimodal monitoring data from a time domain perspective, a frequency domain perspective, and a spatial domain perspective to obtain a time domain feature vector corresponding to the time domain perspective, a frequency domain feature vector corresponding to the frequency domain perspective, and a spatial domain feature vector corresponding to the spatial domain perspective; The time domain feature vector, the frequency domain feature vector and the spatial domain feature vector are feature-concatenated to obtain the multimodal fusion feature vector.

2. The method according to claim 1, characterized in that The multimodal fusion feature vector is subjected to causal chain reasoning based on the preset fusion lead-acid battery knowledge graph to obtain a series of battery status causal chains, including: Linking and matching the multimodal fusion feature vector with entities and attributes in the preset fusion lead-acid battery knowledge graph to obtain a mapping relationship network between the multimodal fusion feature vector and the preset fusion lead-acid battery knowledge graph; Performing a causal chain path search based on the mapping relationship network to obtain a set of candidate causal chains, and performing causal strength calculation on the causal chains in the set of candidate causal chains to obtain the causal strength of each causal chain in the set of candidate causal chains; The causal chains are sorted based on the causal strengths, and the top N causal chains with the largest causal strengths are selected as a series of battery status causal chains.

3. The method according to claim 1, characterized in that The state evaluation of the lead-acid battery based on the battery state causal chain to obtain a real-time state evaluation result includes: Matching the multimodal monitoring data with the battery status causal chain to obtain key indicator values ​​in the multimodal monitoring data, and comparing and calculating the key indicator values ​​with corresponding status nodes in the battery status causal chain to obtain a degree of deviation between the key indicator values ​​and a preset status threshold; Determining, based on the degree of deviation, a state node where the lead-acid battery is currently located in the battery state causal chain; Calculating the state transition probability of each state node in the battery state causal chain; A real-time state evaluation is performed based on the current state of the lead-acid battery being a state node in the battery state causal chain and the state transition probability to obtain the real-time state evaluation result.

4. The method according to claim 1, wherein Before inputting the multimodal fusion feature vector into a preset battery performance degradation trend estimation model to obtain the battery performance degradation trend, the method further includes: Collecting lead-acid battery degradation history data, and preprocessing the lead-acid battery degradation history data to obtain preprocessed lead-acid battery degradation history data; Performing feature extraction on the preprocessed lead-acid battery degradation history data to obtain degradation feature data; performing degradation pattern recognition on the preprocessed lead-acid battery degradation history data to obtain a degradation pattern; Leveraging battery domain knowledge, we construct a battery status causal chain that reflects degradation mechanisms. A battery performance degradation trend estimation model is constructed based on the degradation characteristic data, the degradation mechanism battery state causal chain, and the degradation pattern, and a battery performance degradation trend estimation model is constructed based on the lead-acid battery degradation history data. When the prediction accuracy of the battery performance degradation trend estimation model is greater than a preset accuracy threshold, the preset battery performance degradation trend estimation model is obtained.

5. The method according to claim 4, characterized in that Before performing causal chain reasoning on the multimodal fusion feature vector based on the preset fused lead-acid battery knowledge graph to obtain a series of battery state causal chains, the method further includes: Obtaining lead-acid battery knowledge text related to lead-acid batteries, and constructing a lead-acid battery ontology knowledge base based on the lead-acid battery knowledge text; Obtain sample data of lead-acid battery monitoring materials, and map the sample data of lead-acid battery monitoring materials into an ontology framework based on the lead-acid battery ontology knowledge base to form a semantic network including concept nodes, attribute edges, and relationship edges, thereby obtaining the preset fused lead-acid battery knowledge graph.

6. The method according to claim 5, characterized in that The lead-acid battery ontology knowledge base is based on mapping the lead-acid battery inspection material sample data into the ontology framework to form a semantic network including concept nodes, attribute edges, and relationship edges, thereby obtaining the preset fused lead-acid battery knowledge graph, including: Preprocessing the lead-acid battery monitoring data sample data to obtain lead-acid battery monitoring data; Performing data mapping on the lead-acid battery monitoring data based on the lead-acid battery ontology knowledge base, mapping entities in the lead-acid battery monitoring data to corresponding ontology concept nodes in the ontology knowledge base, and mapping attributes in the structured data to attribute edges of the ontology concept nodes; Performing relationship extraction on the lead-acid battery monitoring data based on the lead-acid battery ontology knowledge base, extracting semantic relationships between named entities in the lead-acid battery monitoring data, and mapping the semantic relationships into relationship edges between the ontology concept nodes; Based on the ontology concept nodes, the attribute edges and the relationship edges, the preset fused lead-acid battery knowledge graph is constructed.

7. A battery operation status monitoring and evaluation system, characterized in that: The system comprises: Data acquisition module, used to collect multimodal monitoring data of lead-acid batteries during the charging and discharging process; A feature extraction module, configured to pre-process the multimodal monitoring data using a deep learning-based multi-view encoder to obtain a multimodal fusion feature vector; A real-time status assessment module is configured to perform causal chain reasoning on the multimodal fusion feature vector based on a preset fused lead-acid battery knowledge graph to obtain a series of battery status causal chains, and perform status assessment on the lead-acid battery based on the battery status causal chains to obtain a real-time status assessment result; a degradation trend prediction module, configured to input the multimodal fusion feature vector into a preset battery performance degradation trend estimation model to obtain a battery performance degradation trend, wherein the fused lead-acid battery knowledge graph is obtained by performing causal reasoning on the key components, degradation mechanism, and health status of the lead-acid battery; A comprehensive evaluation module, configured to comprehensively evaluate the state of the lead-acid battery based on the real-time state evaluation result and the battery performance degradation trend, to obtain a comprehensive evaluation result; The feature extraction module is also used to calculate the mean and standard deviation of the multimodal monitoring data, and based on the mean and standard deviation, use a preset Z-score method to scale the multimodal monitoring data to obtain standardized data, and perform distribution correction on the standardized data to obtain preprocessed multimodal monitoring data; perform feature extraction on the preprocessed multimodal monitoring data from the time domain perspective, frequency domain perspective, and spatial domain perspective to obtain a time domain feature vector corresponding to the time domain perspective, a frequency domain feature vector corresponding to the frequency domain perspective, and a spatial domain feature vector corresponding to the spatial domain perspective; and perform feature splicing on the time domain feature vector, the frequency domain feature vector, and the spatial domain feature vector to obtain the multimodal fusion feature vector.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executed by a method according to any one of claims 1 to 6.

9. An electronic device, characterized in that: The electronic device comprises a processor, a memory and a transceiver, wherein the memory is used to store instructions, the transceiver is used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 6.

Images