Multi-element time sequence real-time anomaly detection method for lithium battery formation and grading process, electronic device and storage medium

Abstract

The present application relates to a kind of lithium battery formation and the multi-element time series real-time anomaly detection method of partial capacity process, electronic equipment, storage medium.Formation process and partial capacity process are indispensable step in lithium battery production process, if not in time to find abnormal data in production process will lead to battery bulging even fire safety accident.The present application carries out abnormal detection to the data of lithium battery charge-discharge by the method of real-time anomaly detection, finds abnormal data in time to lithium battery in production process, reduces the probability of safety accident, protects lithium battery and equipment and plays a key role.

Description

TECHNICAL FIELD

[0001] The present application relates to real-time exception detection of battery formation and capacity process, and relates to a time series real-time, unsupervised clustering exception detection method for data when the battery in the lithium battery formation and capacity process is charged and discharged. BACKGROUND

[0002] There are few outliers in time series flow data, and the limit time series anomaly detection of the exception standard has very wide application in industry, such as finding malicious buyers in transaction data of online retail industry, finding intruders in network security traffic data, finding diseases or mutations in biological genes, etc. Under the goal of carbon neutralization, the state builds lithium battery factories to produce batteries, and a large number of batteries are produced. In the battery formation and capacity stage, there are few abnormal data in the time flow data, and it is difficult to set up artificial exception judgment and difficult to tag. Therefore, online real-time detection of abnormal batteries, substandard batteries and abnormal lithium battery detection equipment in the charging and discharging process has become an important research direction.

[0003] The existing detection means are whether the extreme value, maximum value, minimum value, etc. of the voltage, internal resistance, capacity, probe voltage, port voltage, temperature, negative pressure, etc. of the battery within a unit time is within the artificially preset range. There may be exceptions in normal batteries, and the effectiveness is not high, which may lead to the failure to find abnormal batteries in time, and may cause fire and bulging in the charging and discharging process. The present application jointly performs abnormal detection and judgment through RRCF (robust random cut forest anomaly detection based on flow) and autonomously set exception judgment mode, and achieves very ideal effect. SUMMARY

[0004] The main function of the present application is to improve the protection mechanism of the existing battery production process, promote enterprises to improve the yield of lithium battery production and reduce the probability of production accidents. According to the time flow data of each battery in the formation and capacity process, and each battery only according to its own historical time flow data to perform online real-time exception detection on the time data of the future input battery.

[0005] To achieve the purpose, the technical scheme of the present application is:

[0006] According to one aspect of the present application, a multi-element time series real-time exception detection method for lithium battery formation and capacity process is provided, comprising:

[0007] S1, collecting time series data of normal and abnormal formation and capacity process in the lithium battery charging and discharging process based on parameter protection;

[0008] S2, pre-processing the time series data to achieve feature selection and change;

[0009] S3. Sample normal and abnormal time series data. The samples are used to train the model to generate anomaly thresholds for lithium batteries. The anomaly thresholds are score thresholds used for anomaly determination.

[0010] S4. For each step of the decomposition and capacity formation process, the first M lines of data stream for each step are ignored, where M is a set value.

[0011] S5. After preprocessing the lithium battery sample data stream to be tested, input it into the model, and combine it with the abnormal threshold and the abnormality ignoring of the first M rows of each step to jointly determine whether the currently input data sample point is abnormal.

[0012] As an improved approach, step S2's filtering of time-series data 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 of the same lithium battery charging and discharging step, so that some or all of the numerical features of that step are expanded or reduced in magnitude but remain at the same order of magnitude. Further, the transformation performed on the time-series data in step S2 further includes: if the length of the input 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 used for anomaly detection. In this invention, the data collection for the formation and capacity testing process in step S1 further includes converting the binary data stream transmitted from the formation and capacity testing cabinet into a dataset R = {S1, S2, S3, ..., S...}. x The dataset R contains S lithium battery data samples, and each S has N = {D1, D2, D3, ..., D}. x The data consists of} D-dimensional feature samples, S, which accumulates over time. The dimension N is configured to partially or fully include lithium battery voltage, process number, process type, current, temperature, capacity, energy, port voltage, probe voltage, contact impedance, electrode failure voltage difference, and date / time. Step S2 further includes filtering the time series data for each sampling point N. i Remove categorical features, date / time features, and irrelevant numerical features, and retain numerical features.

[0013] As an improved approach, step S3 samples normal and abnormal time-series data to train the model and generate anomaly thresholds for lithium batteries, further including:

[0014] S31. Initialize the number of binary trees, the number of leaves in each tree, the time window length, and construct the binary trees for the Robust Random Cut Forest (RRCF) algorithm.

[0015] S32, sample the normal sample data and the abnormal sample data and input into the RRCF algorithm for abnormality detection and obtain an abnormal score, determine an abnormal threshold of each working step according to the abnormal score, and then input a continuous fixed-length data stream into the forest in a time windowing batch mode, each tree obtains an abnormal score, and the average of the abnormal scores of all trees is taken as the abnormal score of the data point.

[0016] Further, step S4 ignores the abnormality of the first M rows of data stream of each working step, and further comprises:

[0017] S41, according to the change characteristics of the data of each working step of the lithium battery, define a sample index of a current time series structured by the RRCF, and perform self-increment 1 each time a sample point is detected, if an abnormality occurs, compare the working step number of the current abnormal sample with the working step number of the previous sample point, if they are not equal, it is determined that they belong to different working steps, and it is determined that the current detected sample is an initial collection sample of the working step, and then the first M rows of the working step are ignored.

[0018] As another improvement scheme, step S5 further comprises:

[0019] S51, combine the abnormal threshold of each working step and the condition of ignoring the first M rows of each working step as an online real-time abnormality determination condition of the current time series data;

[0020] S52, when the condition of S51 is met, determine that it is an abnormal point, and return the name of the lithium battery and the abnormal data point of the lithium battery.

[0021] According to another aspect of the present application, an electronic device is provided, wherein the electronic device comprises:

[0022] a processor; and

[0023] a memory arranged to store computer executable instructions, the executable instructions, when executed, causing the processor to perform the method.

[0024] According to another aspect of the present application, a computer readable storage medium is provided, wherein the computer readable storage medium stores one or more programs, the one or more programs, when executed by a processor, implement the method.

[0025] The above description is only a summary of the technical solutions of the present application, in order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the description, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0026] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments with reference made to the accompanying drawings. The drawings are for purposes of illustration only and are not intended to be limiting in

[0027] In the drawings:

[0028] Figure 1 A modeling flowchart of the anomaly detection method of the present application is shown;

[0029] Figure 2 Test data for the anomaly detection method of the present method is shown;

[0030] Figure 3 An example diagram of a binary tree for RRCF model construction is shown;

[0031] Figure 4 Sample data and anomaly score graphs for negative pressure anomaly detection are shown;

[0032] Figure 5 A structural diagram of the electronic device of the present application is shown;

[0033] Figure 6 A structural diagram of the computer readable storage medium of the present application is shown. DETAILED DESCRIPTION

[0034] Exemplary embodiments of the present disclosure will be described more fully hereinafter with reference to the accompanying drawings, in which exemplary embodiments of the present disclosure are shown. It should be understood that the present disclosure can be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art.

[0035] Figure 1 A flowchart of the online real-time anomaly detection method for chemical composition and capacity provided by the present embodiment is shown. The present embodiment is based on an electronic device, such as a computer device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, which can be applied on a server. The processor can implement a multi-element time series real-time anomaly detection method for lithium battery formation and capacity process when executing the program, wherein the method includes the following steps:

[0036] S1. Collecting time series data of normal and abnormal formation and capacity processes during the lithium battery charging and discharging process based on parameter protection;

[0037] S2. Preprocessing the time series data to achieve feature selection and change;

[0038] S3, sampling normal and abnormal time series data, the sample is used to train a model to generate an abnormal threshold of a lithium battery, the abnormal threshold is a score threshold for abnormality determination;

[0039] S4, for each process step of formation and dispensing, the first M rows of data flow of each process step are ignored for abnormality, wherein M is a set value;

[0040] S5, the sample data flow of the lithium battery to be detected is input into the model after preprocessing, and the abnormal threshold and the abnormality of the first M rows of each process step are combined to determine whether the current input data sample point is abnormal.

[0041] The evaluation index of the abnormality detection effect is accuracy (Accuracy), precision (Precision), recall (Recall) and F-Score.

[0042] TP (True Positive): Correct positive example, an instance is a positive class and is also determined as a positive class.

[0043] FN (False Negative): False negative, which is a positive class but is determined as a false class.

[0044] FP (False Positive): False positive, which is a false class but is determined as a positive class.

[0045] TN (True Negative): Correct negative example, an instance is a false class and is also determined as a false class.

[0046]

[0047]

[0048]

[0049]

[0050] The abnormal sample and normal sample of different types are fused for abnormality detection result as shown in Figure 2 .

[0051] With the continuous improvement of equipment and process, the abnormal lithium battery becomes very few in the production process, and from the results Figure 2 , the detection accuracy of the negative pressure value abnormality, master-slave communication abnormality and static voltage lower limit abnormality generated in the formation and dispensing stage is high, reaching 99%. Many abnormal batteries are found in many normal batteries, such as abnormal data found in advance in the trigger parameter protection in the negative pressure value abnormality.

[0052] It should be noted that the present application is applicable to online real-time anomaly detection of the formation and capacity grading stage of any lithium battery.

[0053] The present application has the following advantages over the existing safety parameter protection and traditional supervised algorithm in the formation and capacity grading stage:

[0054] 1. Successfully detects the master-slave communication anomaly, step negative pressure overrun, and step voltage lower limit anomaly in abnormal batteries in the formation and capacity grading stage.

[0055] 2. The existing abnormal protection value is predicted in advance, and the occurrence of lithium battery fire and other safety accidents is avoided in advance, and equipment problems can be detected in time.

[0056] 3. Detects the abnormal value in normal batteries and the value triggers the traditional protection mechanism.

[0057] 4. Because the data in the first few steps often changes greatly during the charging process of the battery, it is easy to trigger the traditional protection mechanism, but it is a normal situation. The present application can solve this situation and has high flexibility.

[0058] 5. In the actual production process, the abnormal data changes constantly, and it is difficult for traditional supervised algorithms to obtain labels, new abnormalities cannot be detected, and the number of parameters protected by human subjective setting and subjective judgment is not accurate, which leads to the failure to discover new abnormalities. The present application can timely, efficiently, accurately, and comprehensively apply to this application scenario.

[0059] As an embodiment of the above scheme, in the present application, the data collection of step S1 can be specifically configured to include:

[0060] S11, the formation and capacity grading system converts the binary data stream transmitted by the formation and capacity grading cabinet into a data set R={S1, S2, S3, …, S x}, R has S lithium battery data samples, each S has N={D1, D2, D3, …, D x} D-dimensional feature samples, S is continuously accumulated with time sampling points, and the dimension of N includes lithium battery voltage, step number, step type, current, temperature, capacity, energy, port voltage, probe voltage, contact impedance, tab bad pressure difference, time and date, etc.

[0061] On this basis, step S2 is further configured to include:

[0062] S21, for each sampling point N i, remove the category feature, time date feature, irrelevant (distinguished by software personnel) numerical value feature, and retain the numerical value feature.According to the change trend of each feature of the lithium battery, the data magnitude of the same lithium battery charging and discharging step is processed, and part or all of the numerical value features of the step are enlarged or reduced in magnitude and then kept in the same order of magnitude.

[0063] S22, when the length of the input time series data is within the preset threshold, directly input to the model for anomaly detection.If greater than the preset threshold, take the latest set length equal to the threshold for detection.

[0064] In the above, for the training phase, the binary data stream generated by the formation and distribution cabinet is converted into a numerical data stream by the formation and distribution software platform, and the normal and abnormal lithium battery time series data is removed and retained only the required features, and then the specific feature magnitude change is performed according to the specific step, which 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, lithium battery data is removed and retained only the required features, and then the specific feature magnitude change is performed according to the specific step, when the length of the input data sample is less than the set value, directly input into the model for anomaly detection, if greater than the set value, take the latest set size of the input data stream, can control the timeliness of anomaly detection.

[0065] As an improved scheme, the step S3 of the present application is further configured to include:

[0066] S31, initialize the number of binary trees of Robust Random Cut Forest (RRCF) algorithm, the number of leaves of each tree, the length of time series window, and build binary tree;

[0067] S32, sample and input the normal sample data and abnormal sample data into the RRCF algorithm for anomaly detection and obtain an abnormal score, determine the abnormal threshold of each step according to the abnormal score, and then input the data stream of continuous fixed length into the forest in time series windowing batch, each tree obtains an abnormal score, and the average value of the abnormal scores of all trees is taken as the abnormal score of the data point.

[0068] In the above, by initializing the model framework of RRCF, the processed lithium battery data is input into the model, and the abnormal score of each sample point is obtained, and the abnormal threshold of each step can be determined.

[0069] The RRCF algorithm mentioned in the present application is a very efficient, accurate and stable algorithm in real-time anomaly detection, and the key principles of the algorithm are described in detail as follows.

[0070] (1) The principle of constructing a tree in S31:

[0071] Input: dataset S, dimension d of data samples.

[0072] Output: a binary tree consisting of several branches and several leaves.

[0073] 1. Calculate the range of each dimension:

[0074] 2. Dimension selection: randomly select a dimension q according to the probability The larger the range span of a dimension, the greater the probability of being selected as the dimension for splitting.

[0075] 3. Split point selection: randomly select a split point p of the dimension q According to the principle of uniform probability distribution, divide the sample set with values less than p as the left child of the current node, and the sample set with values greater than p as the right child of the current node.

[0076] 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.

[0077] 5. If the sample set has only one point, mark the sub-node as a leaf node of the tree.

[0078] 6. If the number of sample points in the sample set is greater than 1, recursively perform steps 1-5. Until all points become leaf nodes (leaf), the tree T is constructed, as shown in Figure 3 .

[0079] (2) The principle of inserting a sample point v of S set in S32:

[0080] 1. If the S set is empty, return the RRCF tree T consisting of only the v point.

[0081] 2. If the S set is not empty, insert the sample point v into the binary tree T constructed by RRCF (set S), and generate the tree T' from RRCF (S U v). Randomly generate a new random cut value p' in the random dimension, and check if 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 the sub-tree T (S) as the other child node as the left and right sub-trees. 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).

[0082] (3) The codisp anomaly score calculation in S32:

[0083] The RRCF tree T can be described by a binary vector of each leaf node, where 0 means that the node is the left child of its parent node, and 1 means the right child, and the vector length is the depth of the node in the tree. Therefore, the structure of the tree can be represented by the binary vector of all leaf nodes, for example Figure 3 Leaf 1 (0), 2 (1, 0, 1), 3 (1, 0, 0), 4 (1, 1, 0, 0), 5 (1, 1, 0, 1), 0 (1, 1, 1). The complexity of the tree is the sum of the depths of all child nodes, that is, the total amount of bits, Figure 3 The complexity of the tree is 18.

[0084] 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 represents the degree of abnormality of a point, and 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, then deleting the abnormal point will not cause a significant change in the tree structure, so the displacement of the point will not be very large, and it is easy to misjudge it as a non-abnormal point. When calculating co-disp, the point with high similarity to the target point is deleted, and then the amount of change in the complexity of the tree is calculated.

[0085] In the detection process, the point to be detected is inserted into each tree in the forest, and the anomaly score (co-disp) of all trees for the point is calculated, and the average value is taken as the anomaly score of the point.

[0086] (4) The anomaly detection process of the RRCF algorithm:

[0087] 1. First, initialize tree_num empty trees, and the tree_size of the tree.

[0088] 2. Insert a new point to determine the anomaly of the sample (insert the new point into each tree, 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, determine the anomaly, and delete the new point from each tree).

[0089] 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.

[0090] On this basis, preferably, step S4 ignores the anomaly of the first M rows of data flow of each working step, further comprising:

[0091] S41, according to the change characteristics of the data of each step of the lithium battery, define a sample index of a current time sequence structured by the RRCF, and perform self-increment by 1 each time a sample point is detected, and if an abnormality occurs, compare the step number of the current abnormal sample with the step number of the previous sample point, if they are not equal, it is determined that they belong to different steps, and it is determined that the currently detected sample is the initial collection sample of the step, and then the first M rows of the step are ignored.

[0092] Through the above setting, the false triggering phenomenon of the traditional protection mechanism in normal cases can be avoided, and the detection method has very high flexibility.

[0093] Figure 4 The sample data and abnormal score graph of the negative pressure anomaly detection are shown in the figure, and the negative pressure characteristic data abnormal point and the abnormal score of the point are boxed. The points with higher abnormal scores are the switching between steps, and the abnormal scores of the points are ignored during detection.

[0094] As another improvement scheme, step S5 further comprises:

[0095] S51, the abnormal threshold value of each step and the condition of ignoring the first M rows of each step are combined as the online real-time anomaly determination condition of the current time sequence data;

[0096] S52, when the condition of S51 is met, it is determined to be an abnormal point, and the name of the lithium battery and the abnormal data point of the lithium battery are returned.

[0097] The abnormal threshold value obtained by the model and the abnormal score of the current sample data point and the abnormality of the first few rows of each step are used to determine whether the newly input sample point is abnormal, as shown in the following formula: Figure 2 As shown, as the lithium battery data sample increases from 500 to 2000, good robustness and high accuracy are shown.

[0098] It should be noted that:

[0099] The method used in this embodiment can be converted into program steps and devices that can be stored in a computer storage medium, and is implemented by being called and executed by a controller, wherein the device should be understood as a functional module implemented by a computer program.

[0100] The algorithms and displays presented herein are not inherently related to any particular computer, virtual apparatus, or other apparatus. Various general purpose systems can be used with programs in accordance with the teachings herein, or it can prove convenient to construct more specialized apparatus to perform the required method steps. The required structure for a variety of these systems will be apparent from the description above. In addition, the present application is not intended to be limited to any particular programming language. It will be appreciated that there are many programming languages that can be used to implement the teachings herein, and any specific language can be chosen for use in this application.

[0101] In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application can be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been described in detail in order to not obscure the understanding of this description.

[0102] Similarly, it is to be understood that the above description is intended to be illustrative and not restrictive. Many embodiments of the application will be apparent to those of skill in the art upon reviewing the above description, and it is therefore contemplated that the claims should be construed in light of the full scope of the disclosure and the claims, and that the scope of the claims should not be limited to the particular examples disclosed. In addition, the described embodiments are to be considered merely exemplary and are not intended to limit the scope of the application to the precise details of these examples. Various modifications and changes can be made thereto by those of ordinary skill in the art, which modifications and changes are to be considered as falling within the scope of the application as set forth in the claims.

[0103] Those skilled in the art will appreciate that the modules in the apparatuses in the embodiments can be adapted and placed in one or more apparatuses other than the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and further can be divided into more sub-modules or sub-units or sub-components. Any combination of all the features disclosed in the specification (including the accompanying claims, abstract and drawings), and any method or apparatus otherwise disclosed in the specification, can be made, except that at least some of such features and / or processes or units are mutually exclusive. Unless explicitly stated otherwise, each feature disclosed in the specification (including the accompanying claims, abstract and drawings) can be replaced by alternative features that serve the same, equivalent or similar purpose.

[0104] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features but not others included in other embodiments, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.

[0105] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing some or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium, or can take the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0106] For example, Figure 5 A schematic diagram of an electronic device according to an embodiment of the present invention is shown. The electronic device conventionally includes a processor 51 and a memory 52 arranged to store computer-executable instructions (program code). The memory 52 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. The memory 52 has storage space 53 for storing program code 54 for performing any method steps in the embodiments. For example, the storage space 53 for program code may include various program codes 54 respectively for implementing the various steps in the above methods. 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, CDs, memory cards, or floppy disks. Such computer program products are typically, for example... Figure 6 The aforementioned computer-readable storage medium. This computer-readable storage medium may have the same characteristics as... Figure 5 The memory 52 in the electronic device is arranged in a similar manner as storage segments, storage spaces, etc. The program code can be compressed, for example, in a suitable form. Typically, the storage unit stores program code 61 for performing the method steps according to the invention, i.e., program code that can be read by a processor such as 51, which, when run by the electronic device, causes the electronic device to perform the various steps of the method described above.

[0107] It should be noted that the described embodiments illustrate rather than limit the application, and that those skilled in the art will be able to design many alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The application can be implemented by means of both hardware and software, and any combination thereof. In a unitary claim, several devices or means can be listed, comprising means which can be implemented by one and the same item of hardware. The use of the words "first", "second", and "third", etc. do not imply any order. These words are used to name the elements.

Claims

1. A multi-dimensional time-series real-time anomaly detection method for lithium battery formation and capacity testing processes, characterized in that... The multivariate time series real-time anomaly detection method includes the following steps: S1. During the charging and discharging process of a lithium battery based on parameter protection, collect time series data of normal and abnormal formation and capacity testing processes. S2. Preprocess time series data to achieve feature selection and change analysis; S3. Sample normal and abnormal time series data. The samples are used to train the model to generate anomaly thresholds for lithium batteries. The anomaly thresholds are score thresholds used for anomaly determination. S4. For each step of the decomposition and capacity formation process, the first M lines of data stream for each step are ignored, where M is a set value. S5. After preprocessing the lithium battery sample data stream to be tested, input it into the model, and combine it with the abnormal threshold and the abnormality ignoring of the first M rows of each step to 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 data magnitude processing on the same lithium battery charging and discharging step, so that some or all of the numerical features of the step are expanded or reduced by the same order of magnitude and then kept at the same order of magnitude.

3. The multivariate time series real-time anomaly detection method according to claim 2, characterized in that, Step S2, the transformation performed on the time series data, further includes: If the length of the input time series data is within the preset threshold, it is directly input into the model for anomaly detection; otherwise, the length is taken equal to the preset threshold for anomaly detection.

4. The multivariate time series real-time anomaly detection method according to claim 3, characterized in that: The data collection for the formation and capacity testing process in step S1 further includes converting the binary data stream transmitted from the formation and capacity testing cabinet into a dataset R = {S1, S2, S3, ..., S}. x The dataset R contains S lithium battery data samples, and each S has N = {D1, D2, D3, ..., D}. x The feature samples are D-dimensional, and the sampling points S accumulate over time. The dimension N is configured to include, partially or completely, lithium battery voltage, step number, step type, current, temperature, capacity, energy, port voltage, probe voltage, contact impedance, electrode failure voltage difference, and time / date. Step S2, which involves filtering the time series data, further includes filtering each sampling point N. i Remove categorical features, date / time 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 involves sampling normal and abnormal time-series data to train the model and generate anomaly thresholds for lithium batteries, further including: S31. Initialize the number of binary trees, the number of leaves in each tree, the time window length, and construct the binary trees for the Robust Random Cut Forest (RRCF) algorithm. S32. Sample normal and abnormal sample data and input them into the RRCF algorithm for anomaly detection and obtain anomaly scores. Determine the anomaly threshold for each step based on the anomaly scores. Then, batch input the continuous fixed-length data stream into the forest in a time-series window. Each tree obtains an anomaly score, and the average of the anomaly scores of all trees is taken as the anomaly score of that data point.

6. The multivariate time series real-time anomaly detection method according to claim 5, characterized in that, Step S4 involves ignoring the first M rows of data stream for each step, and further includes: S41. Based on the changing characteristics of the data in each step of the lithium battery process, define a sample index of the time series currently structured by RRCF. Each time a sample point is detected, increment the index by 1. If an anomaly occurs, compare the step number of the current abnormal sample with the step number of the previous sample point. If they are not equal, it is determined that they belong to different steps. The sample currently detected can be determined as the initial collection sample of the step, and the first M rows of the step can be ignored.

7. The multivariate time series real-time anomaly detection method according to claim 1, characterized in that, Step S5 further includes: S51. Combine the abnormal threshold determined in each step with the condition of ignoring the first M rows in each step as the online real-time abnormal judgment condition for the current time series data. S52. When the conditions in S51 are met, it is determined to be an anomaly point, and the name of the lithium battery and the abnormal data point of the lithium battery are returned.

8. A storage medium, wherein, The storage medium stores one or more programs, which, when executed by a processor, implement the method of any one of claims 1 to 7.

9. An electronic device, characterized in that, include: processor; as well as, A memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the method according to any one of claims 1 to 7.

Images