Lithium battery soh dynamic correction method fusing capacity increment analysis and timing characteristics

Abstract

The application relates to the technical field of battery management, in particular to a lithium battery SOH dynamic correction method fusing capacity increment analysis and time sequence characteristics, which is suitable for battery operation state evaluation of electric vehicles. The application constructs a battery health state tree, generates candidate nodes by using complete or fragment charging and discharging data, combines a main path to simulate an evolution process, calculates a change degree caused by disturbance in an embedding space, judges a hanging mode of the candidate nodes, and realizes dynamic correction and update of an SOH value. The application has good adaptability and real-time performance, and can improve the accuracy and stability of SOH evaluation under complex working conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery management, in particular to a lithium battery SOH dynamic correction method fusing capacity increment analysis and time sequence characteristics. BACKGROUND

[0002] The existing state of health (SOH) estimation method often relies on complete charge and discharge cycle data acquisition in actual operation process, and in typical use scenarios such as electric vehicles, the load changes frequently, resulting in the existence of fragmented battery data. In the face of such incomplete data, traditional algorithms are difficult to effectively utilize, resulting in insufficient accuracy of SOH estimation results. In addition, the existing technology generally uses linear interpolation, simple regression or alignment based on historical templates, which is difficult to reflect the small disturbance changes in the evolution process of the battery state, resulting in insensitive response to the trend of health state change and poor dynamic correction ability. Especially under the condition of superposition of multiple complex working conditions, the existing method lacks effective modeling means for disturbance influence, and finally it is difficult to realize accurate, continuous and robust evaluation of battery SOH. In order to solve the above problems, the present application designs a lithium battery SOH dynamic correction method fusing capacity increment analysis and time sequence characteristics. SUMMARY

[0003] The technical problem to be solved by the present application is to provide a lithium battery SOH dynamic correction method fusing capacity increment analysis and time sequence characteristics, which is suitable for battery operation state evaluation of electric vehicles. By constructing a battery health state tree, and using complete or fragmented charge and discharge data to generate candidate nodes, combining the trunk path to simulate the evolution process, calculating the degree of change caused by disturbance in the embedding space, judging the hanging mode of the candidate nodes, realizing the dynamic correction and update of SOH value. The present application has good adaptability and real-time performance, and can improve the accuracy and stability of SOH evaluation under complex working conditions.

[0004] To achieve the above purpose, the present application provides the following technical scheme:

[0005] The lithium battery SOH dynamic correction method fusing capacity increment analysis and time sequence characteristics is applied to a battery management system, the battery management system comprising at least one battery monitoring module and an SOH maintenance module configured in a cloud server, the battery monitoring module being in communication connection with the SOH maintenance module, a battery health state tree being configured in the SOH maintenance module, the method comprising:

[0006] Obtaining running data, and constructing candidate nodes according to the running data, wherein the running data comprises capacity increment characteristics and time sequence characteristics of complete or fragmented charge and discharge cycles;

[0007] According to the main path of the battery state of health tree, a first evolution path and a second evolution path are simulated in combination with the candidate node;

[0008] The first evolution path and the second evolution path are respectively subjected to embedding space disturbance evaluation, and a first change degree and a second change degree are calculated, wherein the first change degree and the second change degree are calculated based on a hash collision density of the candidate node and a high-dimensional embedding vector, the high-dimensional embedding vector is generated by a capacity increment feature, a time sequence feature and a node weight coding of a node in the main path, and the hash collision density is used to calculate a disturbance degree of the candidate node on an evolution trend of the main path;

[0009] If the first change degree is less than the second change degree, the candidate node is connected to an end of the main path of the battery state of health tree, and a corrected SOH value is output according to the updated main path.

[0010] The battery state of health tree is constructed, including:

[0011] A main path of the battery state of health tree is initialized, and operation data with complete charging and discharging cycles in a historical charging and discharging cycle are selected as initial main path nodes;

[0012] For each initial main path node, a capacity increment feature and a time sequence feature are extracted, and a corresponding SOH value is calculated;

[0013] According to a descending order of the SOH values, the initial main path nodes are sequentially connected, as a main path of the battery state of health tree;

[0014] On the basis of the main path, a branch path is expanded, and segment data without complete charging and discharging cycles in the historical charging and discharging cycle are selected to be connected to the corresponding branch path in a half-node form, wherein the half-node form is a weight decay node, and a node corresponding to the main path to which the branch path is connected is subjected to weight assignment according to a partial capacity increment feature and a truncated time sequence feature of the half-node.

[0015] The first evolution path and the second evolution path are simulated in combination with the candidate node, including:

[0016] The candidate node is connected to a main path end node closest to the operation time in the battery state of health tree, to generate a first evolution path, and the first evolution path represents a generated main path after the candidate node is incorporated;

[0017] According to the capacity increment trend and the timing evolution curve of the plurality of end nodes in the preset time window in the main path of the battery health state tree, a second evolution path is generated by interpolating and connecting the corresponding virtual nodes, and the second evolution path represents the main path generated at the same time without the candidate node.

[0018] The first evolution path and the second evolution path are respectively subjected to embedding space disturbance evaluation, and the first change degree and the second change degree are calculated, including:

[0019] The nodes of the main path in the battery health state tree are embedded into a high-dimensional topological vector space based on their weights, capacity increment characteristics and timing characteristics, and an embedding reference space is constructed.

[0020] The candidate nodes are constructed in vector form and projected into the embedding reference space to generate candidate embedding nodes.

[0021] The signature representation of the candidate embedding node is generated by random disturbance mapping, and the hash collision density of the signature representation and the nodes in the embedding reference space is calculated, and the change degree is calculated according to the hash collision density.

[0022] The embedding reference space is constructed, including:

[0023] The weights, capacity increment characteristics and timing characteristics of the nodes of the main path in the battery health state tree are encoded into embedding vectors.

[0024] The embedding vectors are disturbed by a preset K hash function and mapped into a plurality of hash buckets.

[0025] A plurality of hash buckets are organized into a hash index atlas to generate an embedding reference space.

[0026] The signature representation of the candidate embedding node is generated by random disturbance mapping, and the hash collision density of the signature representation and the nodes in the embedding reference space is calculated, including:

[0027] The candidate embedding node is disturbed by a preset K hash function to generate K hash signatures.

[0028] The K hash signatures are used as signature representations to query a plurality of hash buckets respectively, and the hit hash bucket is determined.

[0029] The hash collision density is calculated according to the number of nodes in the hit hash bucket and the embedding vector.

[0030] If the first change degree is less than the second change degree, the candidate node is connected to the end of the main path of the battery health state tree, including:

[0031] If the operation data corresponding to the candidate node is complete charge-discharge cycle data, the corresponding candidate node is connected as a backbone path node to the end of the backbone path;

[0032] If the operation data corresponding to the candidate node is fragment charge-discharge cycle data, the missing interval charge-discharge cycle data is predicted according to the capacity increment characteristics and time sequence characteristics of the corresponding candidate node, and the capacity evolution trend of the parent node in the backbone path, the candidate node is completed, and the completed candidate node is connected as a backbone path node to the end of the backbone path.

[0033] If the first change degree is greater than or equal to the second change degree, the method further comprises:

[0034] The candidate node is connected in the form of a half node to a branch path of a current end node in the backbone path of the battery health state tree, wherein the candidate node is a leaf node of the branch path.

[0035] The weight of the candidate node to the current end node is corrected, wherein the correction amount is calculated according to the hash collision density of the candidate node in the high-dimensional embedding space.

[0036] The corrected SOH value is output according to the updated backbone path, including:

[0037] When the backbone path is updated, the current SOH value of the battery health state tree is corrected according to the capacity increment characteristics and time sequence characteristics of the end node.

[0038] When the branch path is updated, the current SOH value of the battery health state tree is corrected according to the weight of the end node.

[0039] The current SOH value is updated and replaced according to the correction result.

[0040] The method further comprises:

[0041] According to the newly connected candidate node or the half node in the branch path, the embedding reference space is updated by local adjustment.

[0042] Compared with the prior art, the beneficial effects of the present application are:

[0043] The lithium battery SOH dynamic correction method provided in this application, which integrates capacity increment analysis and time-series characteristics, can effectively utilize data containing incomplete charge-discharge cycles for health state estimation under actual operating conditions where charge-discharge data is fragmented. By constructing a battery health state tree and introducing a perturbation evaluation mechanism in a high-dimensional embedding space, dynamic attachment judgment of candidate nodes is achieved, effectively reducing estimation bias caused by missing data. Furthermore, the perturbation degree is measured by hash collision density and path selection is assisted, improving the stability and sensitivity of SOH value updates. At the same time, this application maintains the long-term validity of the embedding space through local updates, giving the entire estimation framework good scalability and real-time response performance, making it suitable for dynamic battery health state management under various actual operating environments of electric vehicles. Attached Figure Description

[0044] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0045] Figure 1 This is a schematic diagram illustrating an exemplary application scenario of an embodiment of this application;

[0046] Figure 2 This is a schematic diagram of the cloud server modules in an embodiment of this application;

[0047] Figure 3 This is a flowchart illustrating the lithium battery SOH dynamic correction method that integrates capacity increment analysis and timing characteristics according to an embodiment of this application.

[0048] Figure 4 This is a schematic diagram of the battery health status tree in an embodiment of this application;

[0049] Figure 5 This is a schematic diagram illustrating the process of calculating the degree of change in the embodiments of this application;

[0050] Figure 6 This is a schematic diagram illustrating the principle of battery health status tree update in an embodiment of this application. Detailed Implementation

[0051] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0052] Reference herein to an "embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive of one another. Those skilled in the art will appreciate that embodiments described herein can be combined with other embodiments.

[0053] The application is suitable for lithium-ion battery systems with periodic charging and discharging behavior and energy state degradation characteristics in electric drive platforms represented by electric tricycles, electric motorcycles, electric light vehicles, etc., and is particularly suitable for edge environments with highly fragmented running data distribution and non-standardized charging scenarios.

[0054] It can be understood that such battery systems generally lack high-precision sensor support, and the running data has data incompleteness problems caused by factors such as charging interruption, shallow cycle behavior, and environmental drift, resulting in insufficient accuracy or even failure of traditional SOH estimation methods.

[0055] The application scenarios of the application include but are not limited to:

[0056] There are a large number of incomplete charging and discharging cycles in typical charging behavior, and the data upload frequency is unstable;

[0057] The battery operates in a multi-user, multi-terrain environment, and the timing information is discontinuous or abnormal interference is frequent;

[0058] The SOH evaluation algorithm needs to run under low-frequency sampling, high-drop characteristic signals, and irregular data windows;

[0059] The battery state evolution is significantly nonlinear, and traditional empirical models and mean filtering methods are difficult to fit the boundary state.

[0060] The selection of application scenarios is based on the common characteristics of battery systems, and the application scenarios can be applied to the method of the application if they have one of the following characteristics:

[0061] There are a large number of semi-node hanging behaviors in the battery health state tree, and the main path is not easy to extend;

[0062] The candidate data segment forms a disturbance or pseudo change in the state trend, and the traditional pruning path is difficult to model;

[0063] The capacity attenuation trend is disturbed by multiple fragments, and the SOH correction needs to integrate structural priori and non-structural disturbance evaluation;

[0064] The path cost of the inversion simulation is too high, i.e., the charging times are irregular and relatively frequent.

[0065] It should be noted that the lithium battery SOH dynamic correction method proposed in the present application does not rely on complete SOC trajectory or continuous temperature and current joint sampling as a prerequisite, but is aimed at SOH estimation problem in high noise data disturbance scene.

[0066] Referring to Figure 1 , the figure is a schematic diagram of an exemplary application scenario provided by an embodiment of the present application.

[0067] As Figure 1 shown, the present application is applied to a battery management system, which includes a lithium battery body, a battery monitoring module, a communication interface and a cloud server, wherein:

[0068] The lithium battery body is used to provide energy output for terminal devices such as electric tricycles or electric bicycles. The lithium battery body can be a single body or a battery pack composed of multiple series and parallel combinations, has the electrochemical characteristics of periodic charging and discharging, and its state continuously degrades with the use cycle, and needs to be regularly evaluated for health status (State of Health, SOH) to ensure safety and reliability.

[0069] The battery monitoring module is connected with the lithium battery body, and is used to collect key parameter information of the battery in the running process, which includes but is not limited to: voltage, current, temperature, remaining capacity, state of charge, charging and discharging rate and charging and discharging time period. The battery monitoring module also performs preliminary processing, data compression and feature extraction operations on the collected key parameter information, so as to be transmitted and processed subsequently.

[0070] The communication interface is used to realize the bidirectional data transmission between the battery monitoring module and the cloud server. The communication interface can be a cellular communication module, a LoRa low-power module, a Bluetooth or Wi-Fi communication module, etc., and the specific selection can be determined according to the networking mode and operation environment of the target terminal. Through the communication interface 103, the battery running data can be uploaded to the cloud.

[0071] The cloud server is deployed with the SOH dynamic correction method proposed in the present application, and has high-dimensional vector modeling capability and health status tree management function. The cloud server can perform structured processing on the received battery running data, and calculate the disturbance degree of the candidate data segment based on the embedding space disturbance evaluation mechanism, and then realize the dynamic update of the main path and the correction of SOH. The cloud server can further provide the terminal device with the corrected SOH estimation result, support operation decision and life prediction management.

[0072] Referring to Figure 2 , the figure is a module schematic diagram of the cloud server provided by an embodiment of the present application.

[0073] The SOH maintenance module is configured in the cloud server, wherein the SOH maintenance module is a core calculation and storage structure of the SOH dynamic correction method, is specially used for modeling, matching and correction processing of the battery running state, and specifically includes:

[0074] A battery health state tree is configured in the SOH maintenance module, and is used for storing and maintaining a state evolution structure constructed by historical running data of the battery. The battery health state tree includes a plurality of main path nodes and a plurality of branch path nodes, and is used for describing an SOH evolution track of the battery in use. The main path nodes correspond to complete charging and discharging cycle data, the branch path nodes express auxiliary effects of fragment data on the main path in the form of half nodes, and a health evolution topology is established through a connection relationship between the nodes.

[0075] An embedding space processor is configured to perform high-dimensional embedding coding and perturbation collision analysis on the candidate nodes and the historical nodes in the battery health state tree. The embedding space processor can generate a high-dimensional vector representation based on the capacity increment feature, the time sequence feature and the node weight of the node, and construct a hash signature combined with a local sensitive hash algorithm, so as to evaluate the perturbation degree of the candidate node after being included in the path, and to determine whether the candidate node can be connected to the main path.

[0076] An updating unit is configured to dynamically update the structure of the battery health state tree and the SOH value after the embedding space perturbation evaluation. If the evaluation result of the candidate node meets the access condition of the main path, the end of the main path is updated and the SOH value is re-estimated; if the evaluation result does not meet the condition, the candidate node is connected to the branch path in the form of a half node, and the SOH value at the end of the main path is indirectly corrected through a node weight correction mechanism.

[0077] Next, the lithium battery SOH dynamic correction method provided by the embodiment of the application is introduced in combination with the drawings, Figure 3 The method is applied to a battery management system, the battery management system includes at least one battery monitoring module and an SOH maintenance module configured in a cloud server, the battery monitoring module and the SOH maintenance module are in communication connection, a battery health state tree is configured in the SOH maintenance module, and the method includes:

[0078] S1: obtaining running data, and constructing a candidate node according to the running data;

[0079] In the embodiment, the running data is collected in real time by the battery monitoring module, including voltage, current, temperature and time sequence of charging and discharging original signals, and capacity increment features and time sequence features are extracted after preprocessing, and are encapsulated as candidate nodes in the form of nodes. The candidate node can be derived from a complete charging and discharging cycle, or can only include partial fragment data.

[0080] It can be understood that the specific collection method and type of operation data can be selected according to different product configurations, for example, the sampling frequency and sensor arrangement method can be flexibly set, as long as the basic analysis function of supporting capacity increment and time sequence behavior can be supported, and the application does not limit this.

[0081] S2: According to the main path of the battery health state tree, a first evolution path and a second evolution path are simulated in combination with the candidate node;

[0082] In the embodiment, based on the existing battery health state tree main path, two time consistency paths are respectively constructed:

[0083] The first evolution path is a new path formed by directly connecting the candidate node to the end of the main path;

[0084] The second evolution path is a hypothetical path without the candidate node, a virtual node is generated by the evolution trend of multiple nodes in the historical window of the main path, and an interpolation path is formed;

[0085] It should be noted that in the path construction process, whether the candidate node is a fragment data does not affect the path generation capability; when it is a fragment data, its incomplete structure will be retained to facilitate subsequent evaluation of its weight contribution as a half node.

[0086] S3: The first evolution path and the second evolution path are respectively evaluated by embedding space disturbance, and the first change degree and the second change degree are calculated;

[0087] In the embodiment, each node in the evolution path is embedded into a high-dimensional topological vector space, and a disturbed signature representation is generated by a plurality of hash functions, so as to be mapped to an embedding reference space. The candidate node is also subjected to hash disturbance, and the signature of the node in the embedding reference space is compared, and the hash collision density is calculated as a disturbance index. The first change degree and the second change degree correspond to the disturbance intensity between the first evolution path and the second evolution path and the embedding reference space respectively, and are used to reflect the interference degree of the new node to the original health evolution trend.

[0088] S4: If the first change degree is less than the second change degree, the candidate node is connected to the end of the main path of the battery health state tree, and a corrected SOH value is output according to the updated main path;

[0089] In this embodiment, when the disturbance caused by the candidate node accessing the backbone path is less than the natural fluctuation of the original evolution trend, it is considered to conform to the evolution direction of the healthy state, and it is directly connected as a new node at the end of the backbone path. If it is a segment node, it is first predicted and completed according to the capacity change trend of the parent node before being connected. Then, the SOH value is recalculated according to the capacity increment and time sequence information of the new end node of the backbone path, and the corrected state is output.

[0090] Before expanding the specific technical content of the embodiments of the present application, it is necessary to clarify the problems faced in actual use scenarios.

[0091] Taking the lithium battery widely used in tricycles or electric vehicles as an example, due to frequent load fluctuations, its charging and discharging behavior often presents non-periodic and non-continuous characteristics. For example, under the conditions of short-distance operation, frequent start-stop, downhill energy feedback or short-time charging, the traditional complete charging and discharging cycle is difficult to form; a large amount of operation data presents a fragmented sequence of discontinuous or local charging segments / discharge segments.

[0092] As can be appreciated by those skilled in the art, existing SOH estimation methods mostly use complete charging and discharging cycle data as input, lack effective modeling means for fragmented data, resulting in problems such as evaluation window, error accumulation and poor robustness of the estimation model.

[0093] The core idea of the processing logic of the present application is not to directly improve the SOH regression accuracy, but to start from the perspective of evolution process modeling, to dynamically maintain the battery health state as a health state tree based on time series, and to form a structured decision logic with evolution explanation ability by combining local disturbance evaluation and high-dimensional topology mapping. Under this logic, even if the input is fragmented data, the health state can be continuously maintained and dynamically corrected through node-level evolution trend embedding and disturbance judgment.

[0094] Next, the part of the method of the present application about operation data is further expanded.

[0095] Specifically, the capacity increment feature described in the present application refers to the related information of the capacity change in the battery charging and discharging process under unit voltage interval or equal interval time, which is used to represent the local change trend of the degree of participation of the active material of the battery in the reaction; and the time sequence feature refers to the dynamic mode of the evolution of variables such as voltage, current and temperature of the battery over time in the charging and discharging process, including but not limited to inflection point position, change slope, stable interval and its duration, etc. Time sequence structure information. The features are derived from actual data collected by the battery monitoring module during operation, such as voltage, current, temperature and time stamp, etc. original monitoring values.

[0096] It can be understood that the capacity increment feature and the timing feature can be extracted through a series of calculation models, including differential curve calculation, moving window integration, and feature point detection based on time series, and the present application does not repeat here.

[0097] It should be noted that the present application does not rely on complete charge and discharge cycle data for feature extraction. For fragmented data, even if part of the voltage segment is missing or a complete cycle is not completed, local capacity increment features and truncated timing information can be extracted according to the available interval, and candidate nodes can be constructed.

[0098] In the present embodiment, the form of the candidate node is a feature vector representation generated based on the running data.

[0099] Next, the part of the method of the present application related to the construction of the battery health state tree is further expanded.

[0100] For example, the battery health state tree can be understood by referring to Figure 4 Figure 4 is a structural diagram of the battery health state tree of the embodiment of the present application.

[0101] Figure 4 The battery health state tree is shown, including a main path and branch paths, wherein the branch paths are extended according to the main path nodes in the main path.

[0102] Figure 4 The basic structure form of the battery health state tree is further shown to assist in understanding the state modeling in the present application. The basic structure form includes a main path generated from complete charge and discharge cycle data, and a plurality of branch paths extended from the main path nodes. The white nodes represent the main path nodes, and the gray nodes are branch path nodes constructed from fragmented data, which are connected to the branches at different positions of the main path, and are used to supplement the state information of the main path under non-ideal operating conditions.

[0103] It can be understood that the nodes in the main path are arranged in descending order of SOH value, which logically reflects the degradation trajectory of the battery over time. Each main path node is generated from complete charge and discharge cycle data, has a clear capacity increment curve and complete timing distribution, and constitutes the core reference of state evolution. The branch path is constructed by connecting half nodes at the main path nodes, and each half node corresponds to fragmented charge and discharge data, and generates a weight decay value based on the feature similarity with the main path node, so that its influence on the overall state tree is controllable.

[0104] It should be noted that Figure 4 ​The node layout and connection relationship shown in the figure are only exemplary and are intended to assist in understanding the state modeling method of the present application, and do not limit the number of nodes, branch structure hierarchy or specific feature generation method applicable to the present application. In actual deployment, the node construction and path expansion strategy can be flexibly adjusted according to the data integrity, collection frequency and battery system design requirements.

[0105] In one example, the construction of the battery health state tree comprises:

[0106] Initializing the main path of the battery health state tree, and selecting the operation data with complete charge and discharge cycles in the historical charge and discharge cycles as initial main path nodes;

[0107] For each initial main path node, extracting its capacity increment feature and time sequence feature and calculating the corresponding SOH value;

[0108] According to the descending order of the SOH value, the initial main path nodes are sequentially connected as the main path of the battery health state tree;

[0109] On the basis of the main path, the branch path is expanded, and the segment data without complete charge and discharge cycles in the historical charge and discharge cycles is selected to be connected to the corresponding branch path in the form of a half node, wherein the half node form is a weight decay node, and the corresponding node of the main path connected by the branch path is weighted according to the partial capacity increment feature and the truncated time sequence feature of the half node.

[0110] Specifically, the construction process of the battery health state tree is based on the structured organization and feature mapping of historical operation data, aiming to form a graph basis that can be used for subsequent state evolution simulation and path disturbance evaluation without introducing artificial prior structure division. The construction of the main path node is based on the data segment with complete charge and discharge cycles, and the SOH value is calculated by extracting the key change interval of the capacity increment curve and the corresponding time sequence distribution. By sequentially connecting these nodes in descending order of the SOH value, the natural evolution trend of the gradual degradation of the battery state can be reflected in a physical sense.

[0111] Further, in order to improve the compatibility of the model to incomplete or non-standard working condition data, the feature derived from the segment data is mapped to an auxiliary node with weight decay through the half node logic, and is connected to the branch position of the main path, and indirectly participates in the construction of the health state evolution graph. The contribution degree of such node is weighted and evaluated through the similarity of the feature of the main path node and the overlapping relationship on the time axis, so as to give the main path node a dynamically adjusted structure confidence index.

[0112] Next, the part of the method of the present application about the evolution path is further expanded.

[0113] In one example, the specific steps of S2 are as follows:

[0114] S2.1: Connect the candidate node with the end node of the main path in the battery state of health tree closest to its running time, generate a first evolution path, which represents the generated main path after the candidate node is incorporated;

[0115] It is easy to understand that the SOH evolution process has strong time correlation, and the capacity evolution trend and aging state between adjacent time periods are more continuous. This step can effectively ensure that the connection of the candidate node will not cause abrupt discontinuity to the historical evolution logic of the main path. After generating the connection relationship, the capacity increment features of the candidate node are fused with the existing features of the end node of the main path to construct a new state vector, and the main path sequence is extended accordingly to form a first evolution path.

[0116] S2.2: According to the capacity increment trend and time evolution curve of a plurality of end nodes in the main path of the battery state of health tree within a preset time window, calculate and connect the corresponding virtual nodes through an interpolation algorithm to generate a second evolution path, which represents the main path generated without incorporating the candidate node at the same time;

[0117] In this embodiment, the preset time window is generally set to N periods before the current time, and the setting of the preset time window is based on the slow change characteristic of the battery SOH, which ensures that the extracted node sequence can cover the recent evolution trend. In the specific operation process, first, the capacity increment sequence and its time distribution vector of each node in the window are extracted, and a fitting function is constructed according to the time interval between nodes.

[0118] Further, the state vector of the virtual node is calculated at the current node time point by combining the fitting function through the interpolation algorithm, so that the second evolution path logically continues the natural degradation trend of the main path without introducing the influence of the candidate node.

[0119] Further, to enhance the comparability of the second evolution path and the first evolution path, in addition to capacity value fitting, the time features are also aligned in the interpolation process, so that the generated virtual node has the same time anchor point dimension as the candidate node when the vector is embedded.

[0120] It can be understood that through the construction of the first evolution path and the second evolution path, a pair of contrast path sequences with high matching of structure and time dimension are formed. On the basis of maintaining the logical continuity of the main path, the disturbance degree between the candidate node and the natural evolution trend is taken as the decision basis for state update. The problem of relying on single-point similarity matching or single-feature deviation threshold judgment in the traditional method is avoided, and more attention is paid to the influence change of the whole evolution structure after the node is included. It has higher dynamic adaptability and anti-data incompleteness interference ability, and is especially suitable for running scenes with a large number of fragment data.

[0121] Next, the part of the method of the application related to the change degree calculation is further expanded.

[0122] Referring to Figure 5 The figure is a change degree calculation process schematic diagram provided by the embodiment of the application.

[0123] As can be appreciated by those skilled in the art, in the scenario of long-term accumulation of running data, the number of nodes in the battery health state tree grows exponentially, especially in the actual working conditions of frequent start-stop and unstructured charging and discharging cycles of electric vehicles, the expansion speed of the main path and the branch path is much higher than the capacity that can be carried by the conventional linear modeling framework. With the continuous increase of historical nodes, if the traditional Euclidean distance or similarity measurement method based on time sequence sliding window is still used, there will be a great calculation pressure and matching error accumulation problem.

[0124] Specifically, if the path disturbance evaluation is directly performed in the original feature space, the evaluation operation of any candidate node needs to traverse all existing path nodes, and the time complexity is at least O(n), and when the branch path structure is complex, it even rises to O(nxd), where n is the total number of nodes and d is the feature dimension of each node. This will cause a serious performance bottleneck in the whole SOH dynamic correction process after the expansion of the data volume, and cannot guarantee the real-time performance.

[0125] In the embodiment, to solve the problem of node explosion, the path nodes are embedded into a unified high-dimensional topological vector space. The original multi-source heterogeneous features are uniformly expressed by means of space mapping, and then the fast approximate similarity judgment between the candidate node and the embedded reference space is realized by the disturbance mechanism based on local sensitive hashing. The core of the method is to avoid the exponential calculation complexity caused by the accurate distance calculation in the high-dimensional space. The similar vectors are mapped into the same hash bucket in a probabilistic way by using a hash function, so as to convert the node similarity calculation into a hash bucket hit problem. In the actual execution process, the average query complexity of hash collision detection can be reduced to O(1) in the best case, and the total complexity is O(K), which is much lower than the O(nxd) or O(nxlogn) level of the traditional brute-force matching method, where K is the number of preset hash functions.

[0126] It can be understood that, on the one hand, the application can spatially align the path node features in a structural compression manner, so that the evolution trend of the main path is spatially recognizable; on the other hand, by projecting the disturbance of the candidate node and the hash collision statistics, the influence degree of the candidate node on the original structure can be quickly judged, the discrimination conflict of path selection when there are too many branch paths can be effectively alleviated, and the stability and response speed of the overall judgment can be improved. Especially when dealing with non-standard nodes such as fragment charging and discharging cycles or signal abnormal points, stable judgments can be made based on the spatial disturbance trend without explicit function modeling, which is suitable for health state updating requirements under large-scale dynamic data.

[0127] In one example, the specific steps of S3 are as follows:

[0128] S3.1: Embedding the nodes of the main path in the battery health state tree into a high-dimensional topological vector space based on their weights, capacity increment features and time sequence features, and constructing an embedding reference space;

[0129] It can be understood that this step can obtain the embedding reference space in the offline state through pre-processing, which can be directly used as a reference for subsequent steps.

[0130] Specifically, this step aims to establish a standard reference structure for evaluating the disturbance ability of candidate nodes. By uniformly encoding the running state of each node in the main path, the node has quantifiable and comparable distribution characteristics in the high-dimensional space, forming an embedding reference space for path change trend evaluation. The embedding reference space can reflect the overall evolution trend of the main path and the distribution rule of the node features, providing a spatial geometric reference framework for subsequent disturbance analysis.

[0131] In this embodiment, the capacity increment feature and the time sequence feature of each node in the main path are first extracted, and the node weight is calculated in combination with the contribution degree in the historical evolution process. The three types of features are uniformly mapped to embedding vectors of fixed dimensions. In order to ensure the topological continuity of the mapping result, normalization processing and feature standardization technology are used for preprocessing. Subsequently, K local sensitive hash functions are used to disturb the mapping of the embedding vectors, and the embedding vectors are stored in multiple hash buckets. Each hash bucket constitutes a hash index atlas, which constitutes an embedding reference space.

[0132] In one example, the specific steps of S3.1 are as follows:

[0133] S3.1.1: Encoding the weight, capacity increment feature and time sequence feature of the node of the main path in the battery health state tree into an embedding vector;

[0134] S3.1.2: Disturbing the embedding vector by a preset K hash function, and mapping it into multiple hash buckets;

[0135] S3.1.3: Organize the plurality of hash buckets into a hash index atlas to generate an embedding reference space.

[0136] In this embodiment, for each backbone path node, first extract its three types of core structure information.

[0137] The first type of core structure information is the weight value formed by the cumulative evolution of the node in the path.

[0138] The second type of core structure information is the capacity increment feature extracted in a typical charge-discharge cycle.

[0139] The third type of core structure information is the timing feature associated with the capacity change.

[0140] Further, the aforementioned three types of core structure information will be normalized according to the preset vector dimension standard and mapped to the corresponding field slot in the vector space, thereby generating an embedding vector with uniform scale and sequential arrangement. The embedding vector not only encodes the evolution performance of the node in the battery life cycle, but also implicitly retains the evolution trend of its upstream and downstream path structure.

[0141] Further, in order to reduce the high-dimensional operation overhead brought by accurate distance measurement of all embedding vectors in subsequent perturbation calculation, this embodiment maps each embedding vector through a preset K hash function. The perturbation method can include but is not limited to a linear hash function based on random projection, so that each embedding vector is mapped to a set of hash signatures while maintaining a certain approximate similarity expression. Each set of hash signatures corresponds to a logical hash bucket, and all embedding vectors are respectively projected into their corresponding hash buckets to form a plurality of hash bucket structures.

[0142] It should be noted that the hash function needs to meet the following two aspects:

[0143] The first aspect is local sensitivity, that is, embedding vectors that are close to each other have a higher probability of being mapped to the same hash bucket after perturbation;

[0144] The second aspect is the stability of the perturbation range, that is, the distribution after perturbation can reflect the micro variation trend of the candidate node, while avoiding excessive false collisions.

[0145] Further, after generating the hash bucket structure, in order to achieve the requirements of fast retrieval, update and local fine-tuning of space index management, all hash buckets are organized and constructed into a structured hash index atlas. The construction logic of the hash index atlas can be based on the number of signature overlaps between nodes, the density distribution trend in the bucket, and the evolution window to which the node belongs, thereby forming a lightweight graph structure oriented to the embedding space.

[0146] S3.2: construct the candidate node into a vector form and project it into the embedding reference space to generate a candidate embedding node;

[0147] In particular, the purpose of this step is to enable the candidate node to participate in the spatial comparison with consistent structure, and to reflect its relationship with each node of the backbone path through position mapping, so as to provide a unified calculation basis for disturbance measurement and path decision. The feature encoding mode of the candidate node should be consistent with that of the backbone path node to ensure that its semantic position in the space is interpretable.

[0148] In this embodiment, the capacity increment feature and the timing feature of the candidate node are first extracted in the same dimension as the backbone path node, and an initial weight value is assigned according to its context operation stage. The initial weight value can be calculated by referring to the adjacent cycle SOH rate. The constructed candidate embedding vector is then input into the same set of hash functions as the backbone path to obtain the corresponding K hash bucket indexes, realizing the projection of the vector in the embedding reference space. The position of the candidate embedding node in the space will serve as the starting point for disturbance evaluation, and its spatial adjacency relationship can be used to infer its degree of fit to the path trend.

[0149] S3.3: generate a signature representation of the candidate embedding node through random disturbance mapping, and calculate the hash collision density of the signature representation with the nodes in the embedding reference space, and calculate the degree of change according to the hash collision density;

[0150] In particular, the core of this step is to quantify the possible path changes caused by the introduction of the candidate node based on the stability of the spatial structure, and to determine whether the candidate node is suitable for inclusion in the backbone path based on the spatial proximity between the path structures before and after the disturbance. Hash collision density, as an approximate similarity measurement method in high-dimensional space, can efficiently identify the structural relationship between nodes in large-scale embedding structures.

[0151] In this embodiment, the candidate embedding node generates a hash signature through K hash functions, and each hash signature represents its spatial projection position under a certain disturbance condition. The corresponding K hash buckets in the embedding reference space are queried respectively, the number of backbone path nodes that collide with the candidate node (i.e. mapped to the same hash bucket) in each hash bucket is counted, and the hash collision density of the candidate node in the embedding reference space is calculated by combining the weight and timing span of these hit nodes. Hash collision density reflects the spatial compatibility between the candidate node and the evolution trend of the backbone path. When the hash collision density is in the high value interval, it means that the features of the candidate node are consistent with the current path trend, and the disturbance is small; otherwise, it means that its introduction may cause a twist or deviation in the path direction.

[0152] In one example, the specific steps of S3.3 are as follows:

[0153] S3.3.1: perturb the candidate embedding node by a preset K hash functions to generate K hash signatures;

[0154] Specifically, in order to quickly evaluate the perturbation effect of the candidate embedding node relative to the existing node distribution structure in the high-dimensional embedding space, the embedding vector of the candidate node needs to be perturbed and mapped in this step. The purpose of the perturbation operation is to introduce a certain spatial perturbation redundancy, so that the slight difference between the vectors can be reflected through the difference of the hash signatures, thereby facilitating the subsequent dynamic perturbation measurement based on local collision statistics.

[0155] It can be understood that the hash function here is consistent with the hash function described above.

[0156] In this embodiment, the perturbation operation is based on a set of preset K hash functions. In the implementation process, a perturbation factor can be injected for each dimension of the candidate embedding vector. The perturbed vector can generate a K-dimensional hash signature after being processed by the hash function, which is used as the basis for subsequent bucket lookup.

[0157] S3.3.2: query a plurality of hash buckets respectively based on the K hash signatures as signature representations, and determine the hit hash bucket;

[0158] Specifically, based on the generated K hash signatures, this step takes them as the unique structural identifier of the candidate node, and quickly matches the hash bucket structure stored in the embedding reference space through the hash index atlas.

[0159] In this embodiment, in order to balance the efficiency and accuracy of the query, the index structure of the hash bucket has been hierarchically encoded according to the signature fragment during initial construction, and a signature redundancy tolerance mechanism is attached, so that partial signature bit offset or perturbation will not cause the target bucket query to fail. Each signature will retrieve the main index bit and several auxiliary index bits during the query process, and finally the hit hash buckets will be aggregated into a candidate matching set in order.

[0160] S3.3.3: calculate the hash collision density according to the number of nodes in the hit hash bucket and the embedding vector;

[0161] Specifically, the hash collision density is an important indicator for measuring whether the candidate node causes local perturbation in the current embedding space. Its essence is to jointly count the number distribution and embedding similarity of the existing nodes in the hit hash bucket. This statistical result can not only reflect whether the candidate node is located in a dense or sparse area in the space, but also reveal the evolution deviation degree of its main path trend.

[0162] In this embodiment, the calculation of the hash collision density adopts a double index mechanism:

[0163] The first heavy index mechanism is a quantitative level, which counts the number of effective nodes in all hit buckets and normalizes and compares it with the average bucket capacity of the current embedding space to determine whether the candidate nodes are significantly clustered or deviated from the evolution direction of the main path;

[0164] The second heavy index mechanism is a feature level, which further calculates the vector angle and feature coincidence degree between the hit nodes and the candidate embedding nodes to construct gradient information of the disturbance change direction.

[0165] Further, the two heavy indexes are quantified and then weighted and averaged to obtain the hash collision density.

[0166] In an optional embodiment, the change degree can be obtained according to the inverse ratio of the hash collision density, that is, the greater the hash collision density, the smaller the change degree.

[0167] Taking the battery state of health tree update as an example, reference can be made to Figure 6 for understanding, Figure 6 which is a schematic diagram of the battery state of health tree update principle of the embodiments of the present application.

[0168] In the present embodiment, the first change degree and the second change degree corresponding to the two evolution paths can be obtained according to the foregoing steps, and the comparison of the change degrees includes two cases:

[0169] In the first case, if the first change degree is less than the second change degree, that is, the disturbance to the overall embedding space structure after the candidate node is included in the main path is less than the structural deviation caused by the evolution prediction when the candidate node is not included, it indicates that the capacity increment feature and the time sequence feature of the candidate node have strong consistency or continuity with the current evolution trend of the main path, and the candidate node can be reasonably absorbed and the stability of the evolution trajectory can be enhanced.

[0170] It can be understood that, at this time, the candidate node is connected to the end of the main path of the battery state of health tree, which not only does not destroy the structural smoothness of the current path, but also helps to enhance the adaptability of the main path to new working conditions or segment working condition data, and avoids the trend invalidation caused by time window or state fluctuation. In actual deployment, this operation also represents that the running data uploaded this time will be officially adopted and affect the update logic of the subsequent SOH correction curve. Therefore, under the premise of preserving the prediction accuracy of the main path, this judgment criterion maximizes the absorption of reliable incremental information, ensures the data utilization efficiency and the online evolution ability of the model.

[0171] Further, in order to cope with the situation that the candidate node provides charging and discharging period data for the segment, in the preferred embodiment of the present application, a process of predicting and completing the capacity increment feature and the time sequence trajectory of the missing part of the candidate node is also included. The completion process is modeled based on the evolution trend, capacity change slope and typical segment trajectory in the historical data of the adjacent node at the end of the main path to obtain a relatively complete fitting charging and discharging process, so as to be converted into complete node data that can be connected to the main path, and finally the main path connection is completed in the form of the main path node. This way not only improves the engineering value of the segment data, but also strengthens the fault tolerance and robustness of the main path to data input in complex running scenarios.

[0172] In the second case, if the first change degree is greater than or equal to the second change degree, it means that the structural disturbance caused by the candidate node after being included in the main path is greater than the natural change trend of the main path in the evolution process when the candidate node is not included. At this time, it can be inferred that the capacity increment feature and the time sequence feature of the candidate node are inconsistent with the healthy evolution trajectory of the current main path, and if the main path is forcibly connected, it may cause the trend of the main path to break, abnormally bend, or cause the subsequent SOH calculation accuracy to decrease. Therefore, in order to maintain the continuity and prediction stability of the main path, such candidate nodes should be avoided to be directly included in the main path.

[0173] Specifically, in this case, the strategy of branch path connection is adopted, and the candidate node is connected to the branch path led by the current end node of the main path in the form of a half node. The half node is a node type with limited weight, and its capacity increment feature and time sequence feature will not directly participate in the SOH evolution reasoning of the main path, but will affect the weight update and state propagation of the end node of the main path in the form of a derived path.

[0174] In this embodiment, the newly added candidate half node in the branch path does not have the absolute weight of the main path, but calculates the influence degree of its state on the end node of the main path through the hash collision density between it and the adjacent node in the embedded reference space, and updates the node weight of the end node accordingly. The adjustment of the weight can be realized by the disturbance back-propagation method: when the disturbance density of the half node is concentrated near the end of the main path, it means that the node is not completely consistent, but still within a reasonable deviation range, and its trend can be reflected by fine-tuning the weight of the main node; on the contrary, if the disturbance distribution is divergent or discrete, its weight contribution will be automatically compressed to near zero to avoid misleading feedback.

[0175] In one example, the corrected SOH value is obtained according to the updated main path output, including:

[0176] When the main path is updated, the current SOH value of the battery health state tree is corrected according to the capacity increment feature and the timing feature of the end node;

[0177] When the branch path is updated, the current SOH value of the battery health state tree is corrected according to the weight of the end node;

[0178] The current SOH value is updated and replaced according to the correction result;

[0179] According to the newly accessed candidate node or the half node in the branch path, the embedding reference space is updated through local fine-tuning.

[0180] In the embodiment, the core is to establish a real-time mapping relationship between the structural node state and the SOH evaluation value. Specifically, each node in the battery health state tree contains capacity increment features, timing features and node weight information, which collectively represent the position of the corresponding running state in the multi-dimensional health evaluation space. To achieve the goal of dynamically correcting the SOH value, after the path structure is updated, the health features of the end node need to be calculated in real time and converted into the current SOH estimate value.

[0181] Specifically, when the main path is updated, that is, a new candidate node is connected to the end of the main path, the capacity increment feature and the timing feature carried by the current end node will be input vectors, which are processed through the SOH value mapping function to output a new set of SOH values. In the embodiment, the mapping function can be modeled based on the historical path evolution process, and specifically, the SOH output of the current node can be calculated based on trend fitting, time window aggregation or feature matching interpolation. For example, nonlinear fitting is performed between the SOH values corresponding to multiple historical end nodes in a fixed window, and the new node is projected into the fitting curve to obtain the corrected SOH value of the node.

[0182] Further, in the case where the main path does not change but the branch path is updated, for example, a new half node is connected to the end of the main path branch path, the structural disturbance influence of the new node in the branch path on the end node of the main path needs to be evaluated. In the embodiment, the weight of the end node of the main path is updated by the hash collision density corresponding to the disturbance degree of the half node in the embedding space. The new weight value will affect the trustworthiness or representativeness of the end node in the battery health state tree, and finally be reflected as a weighted correction coefficient in SOH evaluation. Based on the weight change, the system reweights and fuses the health index value of the corresponding node and the historical trend to output the corrected SOH value.

[0183] Further, in order to ensure the adaptability of the embedding space structure to the new state, a local fine-tuning mechanism of the embedding space is also introduced in the embodiment. That is, after the new node is connected, its capacity increment feature, timing feature and perturbation weight are converted into a new high-dimensional vector, which is matched with the adjacent node vectors in the current embedding reference space, and the index structure or density distribution in the corresponding bucket is adjusted through the hash bucket structure. The fine-tuning process does not affect the stability of the global embedding system, and only updates the distribution of the node cluster in a local range, so as to improve the identification accuracy of future similar candidate nodes and the consistency of hash mapping.

[0184] Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A lithium battery SOH dynamic correction method fusing capacity increment analysis and timing characteristics, applied to a battery management system, characterized in that, The battery management system comprises at least one battery monitoring module and an SOH maintenance module configured in a cloud server, the battery monitoring module is in communication connection with the SOH maintenance module, a battery health state tree is configured in the SOH maintenance module, and the method comprises: obtaining running data, and constructing a candidate node according to the running data, wherein the running data comprises capacity increment characteristics and timing characteristics of complete or fragment charging and discharging cycles; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; 2. The lithium battery SOH dynamic correction method of claim 1, wherein combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; 3.The lithium battery SOH dynamic correction method of fusing capacity increment analysis and timing characteristics according to claim 1, characterized in that, combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at the end of a main path in the battery health state tree closest to the running time of the candidate node, generating a first evolution path, and the first evolution path represents a main path generated after the candidate node is incorporated; combining the candidate node with a node at constructing the candidate node into a vector form and projecting the vector into the embedding reference space to generate a candidate embedding node; generating a signature representation of the candidate embedding node through random perturbation mapping, and calculating a hash collision density of the signature representation with nodes in the embedding reference space, and calculating a variation degree according to the hash collision density.

4. The lithium battery SOH dynamic correction method of claim 3, wherein, The embedding reference space is constructed, including: encoding the weight, capacity increment feature and timing feature of the trunk path node in the battery state of health tree into an embedding vector; perturbing the embedding vector through a preset K hash function and mapping it into a plurality of hash buckets; organizing a plurality of hash buckets into a hash index atlas to generate an embedding reference space.

5. The lithium battery SOH dynamic correction method of claim 4, wherein, The method for generating a signature representation of the candidate embedding node through random perturbation mapping, and calculating a hash collision density of the signature representation with nodes in the embedding reference space, includes: perturbing the candidate embedding node through a preset K hash function to generate K hash signatures; querying a plurality of hash buckets with the K hash signatures as signature representations respectively to determine a hit hash bucket; calculating a hash collision density according to the number of nodes in the hit hash bucket and the embedding vector. 6.The lithium battery SOH dynamic correction method of claim 1, wherein, If the first variation degree is less than the second variation degree, the candidate node is connected to the end of the trunk path of the battery state of health tree, including: If the running data corresponding to the candidate node is complete charge and discharge cycle data, the corresponding candidate node is connected to the end of the trunk path as a trunk path node. If the running data corresponding to the candidate node is fragment charge and discharge cycle data, the missing interval of the charge and discharge cycle data is predicted according to the capacity increment feature and timing feature of the corresponding candidate node, and the candidate node is completed, and the completed candidate node is connected to the end of the trunk path as a trunk path node.

7. The lithium battery SOH dynamic correction method of claim 3, wherein, If the first variation degree is greater than or equal to the second variation degree, the method further includes: connecting the candidate node in the form of a half node to a branch path of a current end node in the trunk path of the battery state of health tree, wherein the candidate node is a leaf node of the branch path; correcting the weight of the current end node according to the candidate node, wherein the correction amount is calculated according to the hash collision density of the candidate node in the high-dimensional embedding space.

8. The lithium battery SOH dynamic correction method of claim 7, wherein, Outputting a corrected SOH value according to the updated trunk path, including: When the trunk path is updated, correcting the current SOH value of the battery state of health tree according to the capacity increment feature and timing feature of the end node; When the branch path is updated, correcting the current SOH value of the battery state of health tree according to the weight of the end node; updating and replacing the current SOH value according to the correction result. 9.The lithium battery SOH dynamic correction method of claim 7, wherein, The method further includes: updating the embedding reference space through local adjustment according to the newly accessed candidate node or the half node in the branch path. The method further includes: updating the embedding reference space through local adjustment according to the newly accessed candidate node or the half node in the branch path.

Images