Multivariate time series-based real-time anomaly detection method for lithium battery formation and capacity-grading processes, and electronic device and storage medium
By using the RRCF algorithm and a self-defined anomaly detection method, time-series data in lithium battery formation and capacity testing processes are preprocessed and anomaly thresholds are trained, solving the real-time and accuracy problems of anomaly detection in lithium battery production and improving production safety and yield.
Patent Information
- Application Number
- PCT/CN2024/094036
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-17
- Filing Date
- 2024-05-17
- Publication Date
- 2024-11-21
AI Technical Summary
In existing lithium battery formation and capacity testing processes, it is difficult to identify abnormal batteries in real time and accurately, leading to safety hazards and production accidents.
A robust random logging forest anomaly detection algorithm based on RRCF is adopted, combined with a self-defined anomaly judgment method. The time series data during the charging and discharging process of lithium batteries are preprocessed and anomaly thresholds are trained. Real-time anomaly detection is performed by ignoring the anomaly threshold and the first M rows of data in the process step.
It improved the yield rate of lithium battery production, reduced the probability of production accidents, and enabled timely detection and safety protection of abnormal batteries.
Smart Images

Figure CN2024094036_21112024_PF_FP_ABST
Abstract
Description
Multivariate time series real-time anomaly detection method, electronic device, and storage medium for lithium battery formation and capacity separation processes
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS
[0002] This application claims priority to Chinese patent application number 2023105626735 filed with the China Patent Office on May 17, 2023, entitled “Multivariate time series real-time anomaly detection method, electronic device, and storage medium for lithium battery formation and capacity fractionation processes,” the entire contents of which are incorporated herein by reference. Technical Field
[0003] The present invention relates to real-time anomaly detection in battery formation and capacity grading processes, and to an anomaly detection method for performing real-time, unsupervised clustering of time series data when a lithium battery is charged and discharged in the battery formation and capacity grading processes. Background Art
[0004] Time series stream data has very few outliers, and the anomaly standard is well within the bounds. Time series anomaly detection has a wide range of industrial applications, such as identifying malicious buyers in online retail transaction data, finding intruders in network security traffic data, and identifying lesions or mutations in biological genes. To achieve carbon neutrality, China is vigorously developing lithium battery factories for battery production. This results in a large number of batteries. During the battery grading and capacity stage, the time series data generated by these factories has very few anomalies, and manual anomaly assessment is extremely difficult to set and label. Therefore, online and real-time detection of abnormal batteries, defective batteries, and anomalies in lithium battery testing equipment during the charging and discharging process has become a key research direction.
[0005] Existing detection methods include checking whether the extreme values, maximum values, and minimum values of battery characteristics such as voltage, internal resistance, capacity, probe voltage, port voltage, temperature, and negative pressure within a unit time are within a preset range. This may result in abnormalities in normal batteries, and the effectiveness is low, which can easily lead to abnormal batteries not being discovered in time, causing fires and bulging during the charging and discharging process. The present invention uses RRCF (robust random felling forest anomaly detection based on flow) and an independently set anomaly determination method to jointly perform anomaly detection and determination, achieving very ideal results.
[0006] Summary of the Invention
[0007] The main purpose of this invention is to improve the protection mechanism of the existing battery production process, promote enterprises to improve the yield rate of lithium battery production and reduce the probability of production accidents. Based on the time flow data of each battery cell during the composition and capacity process, each battery cell only uses its own historical time flow data to perform online real-time anomaly detection on the time data of future battery cells.
[0008] To achieve the above-mentioned purpose, the technical solution of the present invention is as follows:
[0009] According to one aspect of the present invention, a multivariate time series real-time anomaly detection method for lithium battery formation and capacity separation processes is provided, comprising:
[0010] S1. During the charging and discharging process of lithium batteries based on parameter protection, collect time series data of normal and abnormal chemical composition and capacity processes;
[0011] S2, preprocessing time series data to achieve feature screening and changes;
[0012] S3. Sampling normal and abnormal time series data. The samples are used to train the model to generate an abnormality threshold for lithium batteries. The abnormality threshold is a score threshold for abnormality determination.
[0013] S4. For each step of the content conversion, ignore the first M rows of data streams of each step, where M is a set value.
[0014] S5. The lithium battery sample data stream to be tested is pre-processed and then input into the model. The abnormal threshold and the abnormality ignoring of the first M rows of each step are combined to jointly determine whether the current input data sample point is abnormal.
[0015] As an improved solution, the screening performed on the time series data in step S2 is configured to include retaining numerical features; the transformation performed on the time series data in step S2 is configured to include processing the data magnitude for the same lithium battery charging and discharging process, so that part or all of the numerical features of the process step are expanded or reduced in magnitude and then maintained at the same magnitude. Furthermore, the transformation performed on the time series data in step S2 further includes: if the input length of the time series data is within a preset threshold, it is directly input into the model for anomaly detection, otherwise a length equal to the preset threshold is taken for anomaly detection. In the present invention, the data collection of the chemical fractionation process in step S1 further includes converting the binary data stream transmitted by the chemical fractionation cabinet into a data set R = {S1, S2, S3, ..., S x}, there are S lithium battery data samples in the data set R, each S has N = {D1, D2, D3, ..., D x} D-dimensional feature samples, S is accumulated over time sampling points, and the dimension of N is configured to partially or completely include lithium battery voltage, step number, step type, current, temperature, capacity, energy, port voltage, probe voltage, contact impedance, tab bad pressure difference, time and date; step S2 performs filtering on the time series data, further including filtering each sampling point N i , remove categorical features, time and date features, and irrelevant numerical features, and retain numerical features.
[0016] As another improved solution, step S3 samples normal and abnormal time series data for training the model to generate abnormal thresholds for lithium batteries, further comprising:
[0017] S31. Initialize the number of binary trees, the number of leaves in each tree, the time window length, and construct the binary tree for the Robust Random Cut Forest (RRCF) algorithm.
[0018] S32. Sample normal sample data and abnormal sample data and input them into the RRCF algorithm for anomaly detection and obtain anomaly scores. Determine the anomaly threshold of each work step based on the anomaly score. Then, input the continuous fixed-length data stream into the forest in batches in a time-series windowed manner. Each tree obtains an anomaly score, and the average of the anomaly scores of all trees is taken as the anomaly score of the data point.
[0019] Furthermore, step S4 ignores the abnormality of the first M rows of data streams of each step, further comprising:
[0020] S41. Based on the changing characteristics of the data of each process step of the lithium battery, a sample index of the time series currently structured by RRCF is defined. The index is incremented by 1 each time a sample point is detected. If an abnormality occurs, the process step number of the current abnormal sample is compared with the process step number of the previous sample point. If they are not equal, it is determined that they belong to different process steps. The currently detected sample can be determined as the initial collection sample of the process step, and the first M rows of the process step are ignored.
[0021] As another improved solution, step S5 further includes:
[0022] S51, combining the abnormality threshold determined for each work step and the condition of ignoring the first M rows for each work step as an online real-time abnormality determination condition for the current time series data;
[0023] S52: When the condition of S51 is met, it is determined to be an abnormal point, and the lithium battery name and the abnormal data point of the lithium battery are returned.
[0024] According to another aspect of the present invention, an electronic device is provided, wherein the electronic device includes:
[0025] processor; and,
[0026] A memory is arranged to store computer executable instructions which, when executed, cause a processor to perform the method described.
[0027] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores one or more programs, and when the one or more programs are executed by a processor, the method described above is implemented.
[0028] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be construed as limiting the present invention. The same reference numerals are used throughout the accompanying drawings to denote the same components.
[0030] In the attached figure:
[0031] FIG1 shows a modeling flow chart of the anomaly detection method of the present invention;
[0032] Figure 2 shows the test data of the anomaly detection method of the present method;
[0033] FIG3 shows an example diagram of a binary tree constructed by the RRCF model;
[0034] Figure 4 shows sample data and anomaly score graph for negative pressure anomaly detection;
[0035] FIG5 is a schematic structural diagram of an electronic device of the present invention;
[0036] FIG6 is a schematic diagram of the structure of a computer-readable storage medium of the present invention. DETAILED DESCRIPTION
[0037] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.
[0038] Figure 1 is a flow chart of the online real-time anomaly detection method for formation and capacity grading provided by this embodiment. This embodiment is implemented based on an electronic device, such as a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program can be applied on a server. When the processor executes the program, a multivariate time series real-time anomaly detection method for lithium battery formation and capacity grading processes can be implemented, wherein the method includes the following steps:
[0039] S1. During the charging and discharging process of lithium batteries based on parameter protection, collect time series data of normal and abnormal chemical composition and capacity processes;
[0040] S2, preprocessing time series data to achieve feature screening and changes;
[0041] S3. Sampling normal and abnormal time series data. The samples are used to train the model to generate an abnormality threshold for lithium batteries. The abnormality threshold is a score threshold for abnormality determination.
[0042] S4. For each step of the content conversion, ignore the first M rows of data streams of each step, where M is a set value.
[0043] S5. The lithium battery sample data stream to be tested is pre-processed and then input into the model. The abnormal threshold and the abnormality ignoring of the first M rows of each step are combined to jointly determine whether the current input data sample point is abnormal.
[0044] The evaluation indicators of anomaly detection effect are accuracy, precision, recall, and F-Score.
[0045] TP (True Positive): A correct positive example, an instance is a positive class and is also judged to be a positive class.
[0046] FN (False Negative): A false negative example, which belongs to the positive class but is judged as a false class.
[0047] FP (False Positive): False positive example, which belongs to the false class but is judged as the positive class.
[0048] TN (True Negative): A correct counterexample, an instance is a false class and is also judged to be a false class.
[0049] Figure 2 shows the anomaly detection results of a manufacturer's lithium battery data by fusing different types of abnormal samples with normal samples.
[0050] With the continuous improvement of equipment and processes, abnormal lithium batteries are becoming extremely rare during production. As shown in Figure 2, the present invention has a 99% accuracy rate for detecting abnormalities such as negative pressure values generated during the fractionation phase, abnormal master-slave communication, and the lower limit of static voltage. Many abnormal cells can be found among many normal batteries, for example, by pre-detecting abnormal data in the case of negative pressure values, using trigger parameter protection.
[0051] It should be noted that the present invention is applicable to online real-time abnormality detection in any lithium battery's formation and capacity formation stage.
[0052] The present invention has the following advantages over existing safety parameter protection in the formation and fractionation stages and traditional supervised algorithms:
[0053] 1. Successfully detected all abnormal master-slave communication abnormalities, step negative pressure exceeding the limit, and step static voltage lower limit abnormalities in the abnormal batteries in the capacity formation stage.
[0054] 2. Predict the existing values that may cause abnormal protection in advance, avoid safety accidents such as lithium battery fires, and detect equipment problems in a timely manner.
[0055] 3. Detecting abnormal values in normal batteries and triggering traditional protection mechanisms.
[0056] 4. During the battery charging process, the first few data items often change significantly, which can easily trigger the traditional protection mechanism. However, this is normal. The present invention can solve this problem and has extremely high flexibility.
[0057] 5. In the actual production process, abnormal data are ever-changing. Traditional supervised algorithms are difficult to obtain labels and cannot detect new anomalies. In addition, the number of parameter protections and subjective judgments set by humans are inaccurate, resulting in the failure to discover new anomalies. The present invention can timely, efficiently, accurately and comprehensively solve this application scenario.
[0058] As an implementation example of the above solution, in the present invention, the data collection in step S1 can be specifically configured to include:
[0059] S11, the storage system converts the binary data stream transmitted by the storage cabinet into a data set R = {S1, S2, S3, ..., S x}, there are S lithium battery data samples in R, each S has N={D1,D2,D3,…,D x} D-dimensional feature samples, S is continuously accumulated with the sampling points over time, and the dimensions of N include lithium battery voltage, process step number, process step type, current, temperature, capacity, energy, port voltage, probe voltage, contact impedance, tab bad pressure difference, time and date, etc.
[0060] On this basis, step S2 is further configured to include:
[0061] S21, for each sampling point N i , remove categorical features, time and date features, and irrelevant numerical features (differentiated and set by software developers), retaining numerical features. Based on the changing trends of various lithium battery features, the data level of the same lithium battery charging and discharging process step is processed to expand or reduce the level of some or all numerical features of the process step and then maintain the same level.
[0062] S22: When the length of the input time series data is within the preset threshold, it is directly input into the model for anomaly detection. If it is greater than the preset threshold, the nearest set length equal to the threshold is used for detection.
[0063] In the above, during the training phase, the binary data stream generated by the capacity storage cabinet is converted into a numerical data stream through the capacity storage software platform. The normal and abnormal lithium battery time series data is feature-removed to retain the required features. The magnitude of the specific features is then varied according to specific process steps. This can improve the anomaly detection accuracy of the trained model. For the actual anomaly detection phase after the model is trained, the input data stream and lithium battery data are feature-removed to retain the required features. The magnitude of the specific features is then varied according to specific process steps. When the length of the input data sample is less than the set value, it is directly input into the model for anomaly detection. If it is greater than the set value, the latest set size of the input data stream is used to control the timeliness of anomaly detection.
[0064] As an improved solution, step S3 of the present invention is further configured to include:
[0065] S31. Initialize the number of binary trees, the number of leaves in each tree, the time window length, and construct the binary tree for the Robust Random Cut Forest (RRCF) algorithm.
[0066] S32. Sample normal sample data and abnormal sample data and input them into the RRCF algorithm for anomaly detection and obtain anomaly scores. Determine the anomaly threshold of each work step based on the anomaly score. Then, input the continuous fixed-length data stream into the forest in batches in a time-series windowed manner. Each tree obtains an anomaly score, and the average of the anomaly scores of all trees is taken as the anomaly score of the data point.
[0067] In the above, by initializing the RRCF model framework, the processed lithium battery data is input into the model, and the anomaly score of each sample point is obtained, and the anomaly threshold of each process step can be determined.
[0068] The RRCF algorithm mentioned in this invention is very efficient and ranks among the best in terms of accuracy and stability among real-time anomaly detection algorithms. The key principles of the algorithm are described in detail below.
[0069] (1) The construction principle of the tree in S31:
[0070] Input: Dataset S, dimension d of data samples.
[0071] Output: A binary tree consisting of several branches and several leaves.
[0072] 1. Count the range in each dimension:
[0073] 2. Dimension selection: according to The probability of randomly selecting a dimension q is , if the range span of a dimension is larger, the probability of the dimension being selected as the segmentation dimension will be greater.
[0074] 3. Cut point selection: arbitrarily select a cut point of this dimension According to the principle of uniform probability distribution, the sample set is divided into samples with a value less than p as the left child of the current node, and samples with a value greater than p as the right child of the current node.
[0075] 4. The dimension q and the corresponding split point p form a branch, which divides the dataset S into two sub-branch nodes: S1 = {x|x∈S,x q ≤p},S2=S-S1.
[0076] 5. If there is only one point in the sample set, mark the child node as the leaf node of the tree.
[0077] 6. If the number of sample points in the sample set is greater than 1, recurse steps 1 to 5 until all points become leaf nodes. The tree T is constructed, as shown in Figure 3.
[0078] (2) The principle of inserting the sample point v of set S into S32:
[0079] 1. If the set S is empty, return the RRCF tree T consisting only of v points.
[0080] 2. If the set S is not empty, insert the sample point v into the binary tree T constructed by RRCF (set S) and generate a tree T' from RRCF (SU v). Generate a new random cut value p' in a random dimension and check whether the cut separates S and v. If the cut value p' separates S and v, create a new parent node with v as one child node and subtree T(S) as the other child node as the left and right subtrees. If the cut value does not separate S' and v, follow the existing cut value p in the tree (for example, if v is less than the existing cut value, go to the left child node, otherwise go to the right child node).
[0081] (3) Calculation of codisp anomaly score in S32:
[0082] An RRCF tree T can be described using a binary vector for each leaf node, where 0 indicates the node is the left child of its parent, and 1 indicates the node is the right child. The length of the vector is the node's depth in the tree. Therefore, the tree structure can be represented by the binary vectors of all leaf nodes. For example, in Figure 3, leaves 1(0), 2(1,0,1), 3(1,0,0), 4(1,1,0,0), 5(1,1,0,1), and 0(1,1,1). The complexity of the tree is the sum of the depths of all child nodes, or the total number of bits. The complexity of the tree in Figure 3 is 18.
[0083] Deleting a leaf node will change the structure and complexity of the tree. When a point is deleted, the complexity of the tree will change, and the amount of change is called the displacement of the point. The displacement indicates the degree of abnormality of a point. The larger the displacement, the more likely the point is an abnormal point. However, if there are points in the tree that are very close to the deleted point, deleting the abnormal point will not cause a significant change in the tree structure, so the displacement of the point will not be large, and it is easy to misjudge it as a non-abnormal point. When calculating co-disp, it is necessary to consider deleting points with a high similarity to the target point and then calculate the change in the complexity of the tree.
[0084] During the detection process, the point to be detected is inserted into each tree in the forest, and the anomaly scores (co-disp) of all trees for the point are calculated, and their average value is taken as the anomaly score of the point.
[0085] (4) RRCF algorithm anomaly detection process:
[0086] 1. First initialize tree_num empty trees and tree_size of the tree.
[0087] 2. Insert a new point and perform an anomaly judgment on the sample (insert the new point into each tree and calculate the anomaly score co-disp of each tree for the point. Calculate the average of the anomaly scores of all trees as the anomaly score of the point. Compare the anomaly score with the anomaly threshold, perform an anomaly judgment, and delete the new point from each tree).
[0088] 3. Randomly select a tree in the forest and insert the new point into the tree. If the size of the tree has reached tree_size, delete the oldest point in the tree and insert it into the position of the deleted point.
[0089] On this basis, preferably, step S4 further includes: ignoring the abnormality of the first M rows of data streams in each step;
[0090] S41. Based on the changing characteristics of the data of each process step of the lithium battery, a sample index of the time series currently structured by RRCF is defined. The index is incremented by 1 each time a sample point is detected. If an abnormality occurs, the process step number of the current abnormal sample is compared with the process step number of the previous sample point. If they are not equal, it is determined that they belong to different process steps. The currently detected sample can be determined as the initial collection sample of the process step, and the first M rows of the process step are ignored.
[0091] Through the above settings, the false triggering phenomenon that would normally trigger the traditional protection mechanism can be avoided, making the detection method extremely flexible.
[0092] Figure 4 shows sample data and anomaly scores for negative pressure anomaly detection. The circled points in the figure indicate anomalies in the negative pressure feature data and their anomaly scores. Other points with high anomaly scores are transitions between process steps and are ignored during detection.
[0093] As another improved solution, step S5 further includes:
[0094] S51, combining the abnormality threshold determined for each work step and the condition of ignoring the first M rows for each work step as an online real-time abnormality determination condition for the current time series data;
[0095] S52: When the condition of S51 is met, it is determined to be an abnormal point, and the lithium battery name and the abnormal data point of the lithium battery are returned.
[0096] The anomaly threshold obtained by the model, the anomaly score of the current sample data point, and the ignoring of anomalies in the first few rows of each step are used to determine whether the latest input sample point is abnormal. As shown in Figure 2, as the number of lithium battery data samples increases from 500 to 2000, it shows good robustness and high accuracy.
[0097] It should be noted that:
[0098] The method used in this embodiment can be converted into program steps and devices that can be stored in a computer storage medium and implemented by being called and executed by a controller, wherein the device should be understood as a functional module implemented by a computer program.
[0099] The algorithm and display provided herein are not inherently related to any particular computer, virtual device or other equipment. Various general-purpose devices can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing this type of device. In addition, the present invention is not directed to any specific programming language. It should be understood that various programming languages can be utilized to implement the content of the present invention described herein, and the above description of specific languages is for the purpose of disclosing the best mode of the present invention.
[0100] In the description provided herein, numerous specific details are described. However, it is understood that embodiments of the present invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.
[0101] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Accordingly, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of the invention.
[0102] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition may be divided into multiple submodules or subunits or subcomponents. All features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed herein may be combined in any combination, except that at least some of such features and / or processes or units are mutually exclusive. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.
[0103] Furthermore, those skilled in the art will appreciate that although some embodiments described herein include certain features included in other embodiments but not other features, combinations of features from different embodiments are intended to be within the scope of the present invention and to form different embodiments. For example, in the claims below, any of the claimed embodiments may be used in any combination.
[0104] The various component embodiments of the present invention can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and a computer program product) for performing a portion or all of the methods described herein. Such a program implementing the present invention can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.
[0105] For example, FIG5 shows a schematic diagram of the structure of an electronic device according to one embodiment of the present invention. The electronic device conventionally includes a processor 51 and a memory 52 arranged to store computer-executable instructions (program code). The memory 52 can be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. The memory 52 has a storage space 53 for storing program code 54 for executing any method steps in the embodiments. For example, the storage space 53 for program code can include individual program codes 54 for implementing various steps in the above method. These program codes can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards, or floppy disks. Such computer program products are typically computer-readable storage media, such as those shown in FIG6 . The computer-readable storage medium can have storage segments, storage space, etc. arranged similarly to the memory 52 in the electronic device of FIG5 . The program code can be compressed, for example, in a suitable form. Typically, the storage unit stores program codes 61 for executing the method steps according to the present invention, that is, program codes that can be read by a processor such as 51. When these program codes are run by an electronic device, the electronic device executes the various steps in the method described above.
[0106] It should be noted that the embodiments described illustrate rather than limit the invention, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between brackets should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising several different elements and by means of suitably programmed computers. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third etc. does not indicate any order. These words may be interpreted as names.
Claims
1. A multivariate time series real-time anomaly detection method for lithium battery formation and capacity separation process, characterized in that ,The multivariate time series real-time anomaly detection method includes the following steps: S1. During the charging and discharging process of lithium batteries based on parameter protection, collect time series data of normal and abnormal chemical composition processes; S2, preprocessing the time series data to achieve feature screening and changes; S3. Sampling normal and abnormal time series data, the samples are used to train the model to generate an abnormal threshold value of lithium batteries, and the abnormal threshold value is a score threshold value for abnormality determination; S4. For each step of the content conversion, the first M rows of data streams of each step are abnormally ignored, where M is a set value; S5. Preprocess the lithium battery sample data stream to be tested and input it into the model. Combined with the abnormal threshold and the abnormal neglect of the first M rows of each process step, jointly determine whether the currently input data sample point is abnormal.
2. The multivariate time series real-time anomaly detection method according to claim 1, characterized in that: The filtering performed on the time series data in step S2 is configured to include retaining numerical features; The transformation performed on the time series data in step S2 is configured to include processing the data magnitude for the same lithium battery charging and discharging process step, so that part or all of the numerical features of the process step are enlarged or reduced in magnitude and then maintained at the same magnitude.
3. The multivariate time series real-time anomaly detection method according to claim 2, characterized in that: The transformation performed on the time series data in step S2 further includes: If the input length of the time series data is within the preset threshold, it is directly input into the model for anomaly detection, otherwise the length equal to the preset threshold is taken for anomaly detection.
4. The multivariate time series real-time anomaly detection method according to claim 3, characterized in that: The data collection of the chemical separation process in step S1 further includes converting the binary data stream transmitted by the chemical separation cabinet into a data set R = {S1, S2, S3, ..., S x }, there are S lithium battery data samples in the data set R, each of which has N = {D1, D2, D3, ..., D x } D-dimensional feature samples, S is accumulated over time, and the dimension of N is configured to partially or completely include lithium battery voltage, process step number, process step type, current, temperature, capacity, energy, port voltage, probe voltage, contact impedance, tab bad pressure difference, time and date; Step S2 performs the screening of the time series data, further comprising: i , remove categorical features, time and date features, and irrelevant numerical features, and retain numerical features.
5. The multivariate time series real-time anomaly detection method according to claim 1, characterized in that: Step S3 samples normal and abnormal time series data for training the model to generate abnormal thresholds for lithium batteries, further comprising: S31, initialize the number of binary trees, the number of leaves in each tree, the time window length of the Robust Random Cut Forest (RRCF) algorithm, and construct the binary tree; S32. Sample normal sample data and abnormal sample data and input them into the RRCF algorithm for anomaly detection and obtain anomaly scores. Determine the abnormal threshold of each process step based on the anomaly score. Then, input the continuous fixed-length data stream into the forest in batches in a time-series windowed manner. Each tree obtains an anomaly score, and the average of the anomaly scores of all trees is taken as the anomaly score of the data point.
6. The multivariate time series real-time anomaly detection method according to claim 5, characterized in that: Step S4 ignores the abnormality of the first M rows of data streams in each step, and further includes: S41. According to the changing characteristics of the data of each process step of the lithium battery, a sample index of the time series currently structured by RRCF is defined. The index is incremented by 1 each time a sample point is detected. If an abnormality occurs, the process step number of the current abnormal sample is compared with the process step number of the previous sample point. If they are not equal, it is determined that they belong to different process steps. The currently detected sample can be determined as the initial collection sample of the process step, and the first M rows of the process step can be ignored.
7. The multivariate time series real-time anomaly detection method according to claim 1, characterized in that: Step S5 further comprises: S51, combining the abnormal threshold determined by each process step and the condition of ignoring the first M rows of each process step as the online real-time abnormality determination condition of the current time series data; S52: When the condition of S51 is met, it is determined to be an abnormal point, and the lithium battery name and the abnormal data point of the lithium battery are returned.
8. A storage medium, wherein: The storage medium stores one or more programs, and when the one or more programs are executed by a processor, the method according to any one of claims 1 to 7 is implemented.
9. An electronic device, characterized in that: include: processor; as well as, A memory arranged to store computer executable instructions which, when executed, cause the processor to perform a method according to any one of claims 1 to 7.
Citation Information
Patent Citations
Time sequence exception processing method and device, electronic equipment and storage medium
CN112905671A
Method and device for detecting abnormal data
CN113779103A
Time sequence signal point anomaly detection method, system, equipment and medium
CN114265882A
Lithium battery detection data preprocessing method based on multi-feature fusion
CN115963409A
Multivariate time sequence real-time anomaly detection method for lithium battery formation and capacity grading process, electronic equipment and storage medium
CN116561545A
Cited By
Lithium battery abnormal sample detection model training method and related device
CN119646658A
Active inhibition method and system for thermal runaway behavior of ultrathin battery and medium
CN120974769A
Abnormal impedance analysis method and system for lithium battery storage location probe
CN122592308A
