Method for establishing battery anomaly detection model and method for battery anomaly detection

Abstract

The present invention relates to a method for establishing a battery anomaly detection model, a method for battery anomaly detection, a computer storage medium, and a computer device for implementing the above method. According to one aspect of the present invention, a method for establishing a battery anomaly detection model is proposed, which includes the following steps: processing a battery feature sequence via a first network structure to generate semantic information of the battery feature sequence; processing a cell feature sequence via a second network structure to generate semantic information of the cell feature sequence; merging the semantic information of the battery feature sequence and the semantic information of the cell feature sequence; and inputting the merged semantic information of the battery feature sequence and the semantic information of the cell feature sequence via a fully connected layer to construct a classifier and train the classifier, so as to establish the battery anomaly detection model based on the trained classifier.

Description

Technical Field

[0001] The present invention relates to the field of batteries, and more particularly to a method for establishing a battery anomaly detection model, a method for battery anomaly detection, a computer storage medium and a computer device for implementing the above method. Background Art

[0002] With the increasing popularity of electric vehicles, onboard battery safety monitoring has received widespread attention in the industry, becoming a key cornerstone for the development of the electric vehicle industry. Accurately detecting battery anomalies is crucial for maintaining battery safety and improving the user experience of electric vehicles.

[0003] The causes of battery abnormalities are diverse and complex, such as micro-short circuits inside the battery caused by impurities such as dust and metal powder introduced into the electrode, long-term humidity in the external environment, incomplete insulation of external circuits, external electronic circuits caused by poor isolation of the battery casing, and the shedding and peeling of active substances in the electrode material, which leads to a decrease in capacitance.

[0004] Currently, battery anomaly detection relies primarily on a series of preset rules. For example, the battery anomaly is determined by checking whether indicators such as the battery's voltage, current, and temperature exceed a set reasonable range. However, this method requires extensive charge and discharge experiments in a battery laboratory to calibrate a large number of parameter thresholds for electric vehicles under multiple operating conditions. Furthermore, these parameter thresholds cannot fully reflect the coupling relationship between battery parameter characteristics. Due to the variable operating conditions of on-board batteries and the diverse and complex causes of anomalies, the above-mentioned method is prone to omissions and false alarms when detecting battery anomalies, making it impossible to accurately detect battery anomalies. Summary of the Invention

[0005] In order to solve or at least alleviate one or more of the above problems, the following technical solutions are provided.

[0006] According to a first aspect of the present invention, a method for establishing a battery anomaly detection model is provided, the method comprising the following steps: processing a battery feature sequence via a first network structure to generate semantic information of the battery feature sequence; processing a cell feature sequence via a second network structure to generate semantic information of the cell feature sequence; merging the semantic information of the battery feature sequence and the semantic information of the cell feature sequence; and inputting the merged semantic information of the battery feature sequence and the semantic information of the cell feature sequence via a fully connected layer to construct a classifier and train the classifier, so as to establish the battery anomaly detection model based on the trained classifier.

[0007] According to a method for establishing a battery anomaly detection model according to an embodiment of the present invention, the first network structure is implemented based on a concept machine recursive neural network and a convolutional neural network encoder, and the second network structure is implemented based on a bidirectional gated recurrent unit and an attention mechanism network.

[0008] According to the method for establishing a battery anomaly detection model according to one embodiment of the present invention or any of the above embodiments, processing a battery feature sequence via a first network structure to generate semantic information of the battery feature sequence includes: using a concept machine recursive neural network to represent the battery feature sequence as a pseudo image feature map; and using a convolutional neural network encoder to encode the pseudo image feature map to generate semantic information of the battery feature sequence.

[0009] According to the method for establishing a battery anomaly detection model described in one embodiment of the present invention or any one of the above embodiments, processing a cell feature sequence via a second network structure to generate semantic information of the cell feature sequence includes: feature encoding the cell feature sequence to obtain shallow sequence features of the cell feature sequence; using a bidirectional gated recurrent unit to obtain the correlation between sequences of shallow sequence features of the cell feature sequence, and determining the deep sequence features of the cell feature sequence based on the correlation; and using an attention mechanism network to obtain the weighting of the sequence dimension of the deep sequence features of the cell feature sequence to generate the semantic information of the cell feature sequence based on the deep sequence features of the cell feature sequence and the weighting of the sequence dimension of the deep sequence features of the cell feature sequence.

[0010] According to the method for establishing a battery anomaly detection model described in one embodiment or any of the above embodiments of the present invention, the battery feature sequence and the battery cell feature sequence are obtained in the following manner: acquiring and preprocessing battery data to generate multiple battery data sequence fragments, wherein the battery data includes normal battery data and abnormal battery data; and performing feature extraction on the multiple battery data sequence fragments to generate the battery feature sequence and the battery cell feature sequence, respectively.

[0011] According to the method for establishing a battery anomaly detection model according to one embodiment of the present invention or any of the above embodiments, preprocessing battery data to generate multiple battery data sequence segments includes: dividing the battery data into battery data during a period when the battery power changes and battery data during a period when the battery power does not change; splitting the battery data during the period when the battery power changes with a predetermined SOC interval as a window length; and splitting the battery data during the period when the battery power does not change with a predetermined time interval as a window length.

[0012] According to the method for establishing a battery anomaly detection model according to one embodiment of the present invention or any one of the above embodiments, feature extraction of the multiple battery data sequence fragments to generate the battery feature sequence includes: determining statistical features corresponding to one or more of the battery cell voltage difference, SOC and ambient temperature in each battery data sequence fragment of the multiple battery data sequence fragments, wherein the statistical features include category features and continuous value features; and encoding the category features in a preset manner and normalizing the continuous value features to generate the battery feature sequence.

[0013] According to the method for establishing a battery anomaly detection model according to one embodiment of the present invention or any of the above embodiments, the categorical features include operating conditions, and the continuous value features include one or more of the mean, maximum, minimum, quantile, skewness, and kurtosis.

[0014] According to the method for establishing a battery abnormality detection model according to one embodiment of the present invention or any one of the above embodiments, feature extraction of the multiple battery data sequence fragments to generate the battery cell feature sequence includes: for each battery data sequence fragment in the multiple battery data sequence fragments, determining a set of voltage means of each battery cell in the multiple battery cells of the battery; normalizing the set of voltage means of each battery cell in the multiple battery cells of the battery to generate voltage sequence data; and processing the voltage sequence data using a feature sampling method to generate the battery cell feature sequence including a first part of the battery cell feature sequence and a second part of the battery cell feature sequence.

[0015] According to the method for establishing a battery anomaly detection model described in one embodiment of the present invention or any one of the above embodiments, processing the voltage sequence data using a feature sampling method to generate the first part of the battery cell feature sequence includes: traversing one or more battery data sequence fragments among the multiple battery data sequence fragments, determining a first plurality of battery cells associated with the maximum value, minimum value and median value in the voltage series data corresponding to the one or more battery data sequence fragments, and a second plurality of battery cells adjacent to the first plurality of battery cells in the battery cell topology structure; and sampling the normalized voltage mean corresponding to the first plurality of battery cells and the second plurality of battery cells in the voltage series data to generate the first part of the battery cell feature sequence.

[0016] According to the method for establishing a battery anomaly detection model described in one embodiment of the present invention or any one of the above embodiments, processing the voltage sequence data using a feature sampling method to generate the second part of the battery cell feature sequence includes: traversing each battery data sequence fragment in the multiple battery data sequence fragments, determining a third plurality of battery cells associated with the maximum value, minimum value and median value in the voltage sequence data and a fourth plurality of battery cells adjacent to the third plurality of battery cells in the battery cell topology structure; and sampling the normalized voltage mean corresponding to the third plurality of battery cells and the fourth plurality of battery cells in the voltage sequence data to generate the second part of the battery cell feature sequence.

[0017] According to the method for establishing a battery anomaly detection model according to one embodiment of the present invention or any of the above embodiments, one or more battery data sequence fragments among the multiple battery data sequence fragments include the first battery data sequence fragment and the last battery data sequence fragment among the multiple battery data sequence fragments.

[0018] According to the method for establishing a battery anomaly detection model according to one embodiment of the present invention or any of the above embodiments, training the classifier to establish the battery anomaly detection model based on the trained classifier includes: dynamically training the classifier until convergence using training data including a training set and a validation set; and selecting the trained classifier with the smallest average validation loss of the validation set as the battery anomaly detection model.

[0019] According to a second aspect of the present invention, a method for battery anomaly detection is provided, the method comprising the following steps: acquiring battery data from the cloud and preprocessing the battery data to generate a plurality of battery data sequence fragments; performing feature extraction on the plurality of battery data sequence fragments to respectively generate a battery feature sequence and a cell feature sequence; and processing the battery feature sequence and the cell feature sequence using a battery anomaly detection model established by the method for establishing a battery anomaly detection model according to the first aspect of the present invention to generate a battery anomaly detection result.

[0020] According to a method for battery anomaly detection described in one embodiment of the present invention, preprocessing battery data to generate multiple battery data sequence segments includes: dividing the battery data into battery data during a period when the battery power changes and battery data during a period when the battery power does not change; splitting the battery data during the period when the battery power changes with a predetermined SOC interval as a window length; and splitting the battery data during the period when the battery power does not change with a predetermined time interval as a window length.

[0021] According to the method for battery anomaly detection described in one embodiment of the present invention or any one of the above embodiments, feature extraction of the multiple battery data sequence fragments to generate the battery feature sequence includes: determining statistical features corresponding to one or more of the battery cell voltage difference, SOC and ambient temperature in each battery data sequence fragment of the multiple battery data sequence fragments, wherein the statistical features include category features and continuous value features; and encoding the category features in a preset manner and normalizing the continuous value features to generate the battery feature sequence.

[0022] According to the method for battery abnormality detection described in one embodiment or any one of the above embodiments of the present invention, the category feature includes an operating condition, and the continuous value feature includes one or more of an average value, a maximum value, a minimum value, a quantile, a skewness, and a kurtosis.

[0023] According to the method for battery abnormality detection described in one embodiment of the present invention or any one of the above embodiments, feature extraction of the multiple battery data sequence fragments to generate the battery cell feature sequence includes: for each battery data sequence fragment in the multiple battery data sequence fragments, determining a set of voltage means of each battery cell in the multiple battery cells of the battery; normalizing the set of voltage means of each battery cell in the multiple battery cells of the battery to generate voltage sequence data; and processing the voltage sequence data using a feature sampling method to generate the battery cell feature sequence including a first part of the battery cell feature sequence and a second part of the battery cell feature sequence.

[0024] According to the method for battery abnormality detection described in one embodiment of the present invention or any one of the above embodiments, processing the voltage sequence data using a feature sampling method to generate the first part of the battery cell feature sequence includes: traversing one or more battery data sequence fragments among the multiple battery data sequence fragments, determining a first plurality of battery cells associated with the maximum value, minimum value and median value in the voltage series data corresponding to the one or more battery data sequence fragments, and a second plurality of battery cells adjacent to the first plurality of battery cells in the battery cell topology structure; and sampling the normalized voltage mean corresponding to the first plurality of battery cells and the second plurality of battery cells in the voltage series data to generate the first part of the battery cell feature sequence.

[0025] According to the method for battery abnormality detection described in one embodiment of the present invention or any one of the above embodiments, processing the voltage sequence data using a feature sampling method to generate the second part of the cell feature sequence includes: traversing each battery data sequence fragment in the multiple battery data sequence fragments, determining a third plurality of battery cells associated with the maximum value, minimum value and median value in the voltage sequence data and a fourth plurality of battery cells adjacent to the third plurality of battery cells in the battery cell topology structure; and sampling the normalized voltage mean corresponding to the third plurality of battery cells and the fourth plurality of battery cells in the voltage sequence data to generate the second part of the cell feature sequence.

[0026] According to the method for battery abnormality detection according to one embodiment of the present invention or any of the above embodiments, one or more battery data sequence fragments among the multiple battery data sequence fragments include the first battery data sequence fragment and the last battery data sequence fragment among the multiple battery data sequence fragments.

[0027] According to a third aspect of the present invention, a computer storage medium is provided, wherein the computer storage medium includes instructions, and when the instructions are run, the steps of the method for establishing a battery abnormality detection model according to the first aspect of the present invention are executed.

[0028] According to a fourth aspect of the present invention, a computer storage medium is provided, wherein the computer storage medium comprises instructions, and when the instructions are executed, the steps of the method for battery abnormality detection according to the second aspect of the present invention are executed.

[0029] According to a fifth aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of the method for establishing a battery abnormality detection model according to the first aspect of the present invention are implemented.

[0030] According to a sixth aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of the method for battery abnormality detection according to the second aspect of the present invention are implemented.

[0031] The method for establishing a battery anomaly detection model according to one or more embodiments of the present invention can mine battery data from both the overall and local perspectives to generate semantic information for battery feature sequences and semantic information for cell feature sequences, and establish a battery anomaly detection model based on the combined semantic information for the battery feature sequences and the semantic information for the cell feature sequences. This improves the robustness of the battery anomaly detection model and the accuracy of battery anomaly detection using the battery anomaly detection model, thereby improving the availability and safety of the battery. Furthermore, the method for establishing a battery anomaly detection model according to one or more embodiments of the present invention eliminates the need for extensive parameter calibration, saving manpower and material resources spent on model establishment. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above and / or other aspects and advantages of the present invention will become clearer and easier to understand through the following description of various aspects in conjunction with the accompanying drawings, in which the same or similar elements are represented by the same reference numerals. In the accompanying drawings:

[0033] Figure 1 FIG. 1 is a flowchart of a method for establishing a battery abnormality detection model according to one or more embodiments of the present invention.

[0034] Figure 2 FIG. 1 is a flow chart of a method for battery abnormality detection according to one or more embodiments of the present invention.

[0035] Figure 3 FIG. 1 is a block diagram of a battery abnormality detection system according to one or more embodiments of the present invention.

[0036] Figure 4 A block diagram of a computer device according to one or more embodiments of the invention. DETAILED DESCRIPTION

[0037] The description of the following specific embodiments is merely exemplary in nature and is not intended to limit the disclosed technology or the application and use of the disclosed technology. In addition, there is no intention to be bound by any express or implied theory presented in the foregoing technical field, background technology or the following specific embodiments.

[0038] In the following detailed description of the embodiments, numerous specific details are set forth to provide a more thorough understanding of the disclosed technology. However, it will be apparent to one of ordinary skill in the art that the disclosed technology can be practiced without these specific details. In other instances, well-known features are not described in detail to avoid unnecessarily complicating the description.

[0039] Terms such as "comprising" and "including" indicate that, in addition to the units and steps directly and explicitly stated in the specification, the technical solution of the present invention does not exclude the presence of other units and steps not directly or explicitly stated. Terms such as "first" and "second" do not indicate the order of units in terms of time, space, size, etc., but are merely used to distinguish between units.

[0040] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

[0041] Figure 1 FIG. 1 is a flowchart of a method for establishing a battery abnormality detection model according to one or more embodiments of the present invention.

[0042] like Figure 1 As shown in , in step S101 , a battery feature sequence is processed via a first network structure to generate semantic information of the battery feature sequence.

[0043] Optionally, the first network structure can be implemented based on a conceptor recurrent neural network (CNN) and a convolutional neural network (CNN) encoder, where the CNN encoder can include an input layer, a convolution layer, a flattening layer, and a fully connected layer. For example, the convolution layer in the CNN encoder can be implemented as various forms of convolution layers, including but not limited to grouped convolution, separable convolution, dilated convolution, depthwise convolution, and the like.

[0044] Optionally, in step S101, a concept machine recurrent neural network can be used to represent the battery feature sequence as a pseudo-image feature map, and a convolutional neural network encoder can be used to encode the pseudo-image feature map to generate semantic information of the battery feature sequence. By using the concept machine recurrent neural network and the convolutional neural network encoder to obtain the semantic information of the battery feature sequence, the semantic information of the battery feature sequence can be mined holistically to improve the accuracy of the obtained semantic information of the battery feature sequence, thereby improving the robustness of the battery anomaly detection model.

[0045] In one or more embodiments, the battery feature sequence may be obtained by preprocessing collected battery data to generate multiple battery data sequence segments and performing feature extraction on the multiple battery data sequence segments, wherein the collected battery data may include normal battery data and abnormal battery data.

[0046] Optionally, preprocessing the collected battery data to generate multiple battery data sequence segments may include: dividing the collected battery data into battery data during a period when the battery charge changes and battery data during a period when the battery charge does not change; segmenting the battery data during a period when the battery charge changes using a predetermined SOC interval as a window length; and segmenting the battery data during a period when the battery charge does not change using a predetermined time interval as a window length. Optionally, when the length of the last battery data sequence segment is less than the window length, the length of the battery data sequence segment is compared with 1 / 2 of the window length. If the length of the battery data sequence segment is less than or equal to 1 / 2 of the window length, the battery data sequence segment and the previous window are merged into one battery data sequence segment; if the length of the battery data sequence segment is greater than 1 / 2 of the window length, the battery data sequence segment is separated into a new battery data sequence segment.

[0047] In one embodiment, the battery data during a period when the battery charge changes may include battery data during charging and battery data during discharging (e.g., vehicle driving), and the battery data during a period when the battery charge does not change may include battery data during a period when the vehicle is stopped and not charging. For example, the battery data during a period when the battery charge changes may be segmented using a 1% SOC interval as a window length to generate 100 battery data sequence segments. For example, the battery data during a period when the battery charge does not change during a day may be segmented using a 1-hour time interval as a window length to generate 24 battery data sequence segments.

[0048] Optionally, extracting features from multiple battery data sequence segments to generate a battery feature sequence may include: determining statistical features corresponding to one or more of cell voltage differential, SOC, and ambient temperature in each of the multiple battery data sequence segments, where the statistical features include categorical features and continuous-valued features; encoding the categorical features in a predetermined manner and normalizing the continuous-valued features to generate the battery feature sequence. Optionally, the length of the generated battery feature sequence for each battery may be maintained uniform; shorter battery feature sequences may be padded with leading zeros to increase their length. Optionally, categorical features may include operating conditions (e.g., charging, discharging, or vehicle stopped and not charging), and continuous-valued features may include mean, maximum, minimum, quantile, skewness, and kurtosis. In one embodiment, categorical features may be encoded using a "one-hot" approach, which can be understood as a binary encoding method where the encoded result contains only one binary "1." For example, when using a "one-hot" approach to encode the charging process, discharging process, and vehicle stopped and not charging process included in the operating state, the charging process can be encoded as "000", the discharging process can be encoded as "010", and the vehicle stopped and not charging process can be encoded as "100". In one embodiment, the maximum and minimum value method can be used to normalize the continuous value features.

[0049] In step S103 , the cell feature sequence is processed via a second network structure to generate semantic information of the cell feature sequence.

[0050] Optionally, the second network structure can be implemented based on a bidirectional gated recurrent unit (Bi-GRU) and an attention mechanism network. Alternatively, the Bi-GRU can be replaced with LSTM, GRU, RNN, Bi-LSTM, Bi-RNN, and other network units.

[0051] Optionally, in step S103, the cell feature sequence can be feature encoded to obtain shallow sequence features of the cell feature sequence, and then a bidirectional gated recurrent unit is used to obtain the correlation between the sequences of the shallow sequence features of the cell feature sequence, and based on the correlation, the deep sequence features of the cell feature sequence are determined, and then an attention mechanism network is used to obtain the weighted sequence dimension of the deep sequence features of the cell feature sequence, so as to generate semantic information of the cell feature sequence based on the deep sequence features of the cell feature sequence and the weighted sequence dimension of the deep sequence features of the cell feature sequence. By using a bidirectional gated recurrent unit and an attention mechanism network to obtain the semantic information of the cell feature sequence, the semantic information of the cell feature sequence can be locally mined to improve the accuracy of the semantic information of the obtained cell feature sequence, thereby improving the robustness of the battery anomaly detection model.

[0052] In one or more embodiments, the cell feature sequence can be obtained by preprocessing collected battery data to generate multiple battery data sequence segments and performing feature extraction on the multiple battery data sequence segments, wherein the collected battery data may include normal battery data and abnormal battery data. The process of preprocessing the collected battery data to generate multiple battery data sequence segments is similar to the preprocessing process described in step S101 and will not be further described here.

[0053] Optionally, performing feature extraction on multiple battery data sequence segments to generate a cell feature sequence may include: determining, for each of the multiple battery data sequence segments, a set of voltage mean values ​​for each of the multiple cells of the battery; normalizing the set of voltage mean values ​​for each of the multiple cells of the battery to generate voltage sequence data; and processing the voltage sequence data using a feature sampling method to generate a cell feature sequence comprising a first portion of the cell feature sequence and a second portion of the cell feature sequence. Optionally, the first portion of the cell feature sequence and the second portion of the cell feature sequence may be concatenated in a certain order along a feature dimension to generate the cell feature sequence. For example, assuming a battery includes 100 cells, the set of voltage mean values ​​for each of the multiple cells of the battery may be represented as a set of voltage mean values ​​for the 100 cells. For example, the set of voltage mean values ​​for the 100 cells may be normalized using a maximum-minimum method to generate the voltage sequence data for each of the battery cells.

[0054] Optionally, processing the voltage sequence data using a feature sampling method to generate the first portion of the cell feature sequence may include: traversing one or more battery data sequence segments from a plurality of battery data sequence segments, determining a first plurality of cells associated with maximum, minimum, and median values ​​in the voltage sequence data corresponding to the one or more battery data sequence segments, and a second plurality of cells adjacent to the first plurality of cells in a cell topology; and sampling the normalized voltage mean values ​​corresponding to the first plurality of cells and the second plurality of cells in the voltage sequence data to generate the first portion of the cell feature sequence. Optionally, the one or more battery data sequence segments from the plurality of battery data sequence segments may include the first battery data sequence segment and the last battery data sequence segment from the plurality of battery data sequence segments. In one embodiment, the first battery data sequence segment may be traversed to determine the cells corresponding to the top three maximum, minimum, and median values ​​in the voltage sequence data of the first battery data sequence segment, and two cells adjacent to the first plurality of cells in a cell topology; and sampling the normalized voltage mean values ​​corresponding to the first plurality of cells and the two cells adjacent to the first plurality of cells in the voltage sequence data to generate the first portion of the cell feature sequence. In one embodiment, the last battery data sequence segment can be traversed to determine the battery cells corresponding to the top three maximum, minimum, and median values ​​in the voltage sequence data of the last battery data sequence segment, as well as the two battery cells adjacent thereto in the battery cell topology structure, and the normalized voltage mean corresponding to the battery cell and the two battery cells adjacent thereto in the battery cell topology structure can be sampled in the voltage sequence data to generate the first part of the battery cell feature sequence. In another embodiment, the above-mentioned processing can also be performed on the first battery data sequence segment and the last battery data sequence segment in multiple battery data sequence segments to generate the first part of the battery cell feature sequence.

[0055] Optionally, processing the voltage sequence data using a feature sampling method to generate the second part of the cell feature sequence may include: traversing each battery data sequence segment in a plurality of battery data sequence segments, determining a third plurality of battery cells associated with the maximum value, minimum value, and median value in the voltage sequence data, and a fourth plurality of battery cells adjacent to the third plurality of battery cells in the cell topology; and sampling the normalized voltage mean corresponding to the third plurality of battery cells and the fourth plurality of battery cells in the voltage sequence data to generate the second part of the cell feature sequence. In one embodiment, each battery data sequence segment in a plurality of battery data sequence segments may be traversed to determine the battery cells corresponding to the top three of the maximum value, minimum value, and median value in the voltage sequence data, and two battery cells adjacent thereto in the cell topology, and sampling the normalized voltage mean corresponding to the battery cells and the two battery cells adjacent thereto in the cell topology in the voltage sequence data to generate the second part of the cell feature sequence.

[0056] The above feature sampling method is suitable for feature extraction of battery cells with different topological structures and can accurately obtain typical battery cell features of different battery types. Therefore, it has high engineering application value and generalization.

[0057] Optionally, the length of the cell feature sequence generated for each battery can be kept consistent and can be kept consistent with the length of the battery feature sequence generated in step S101. The cell feature sequence with a shorter length can be supplemented with 0 in front of its sequence to increase its length.

[0058] In the above steps S101 and S103, by using a bidirectional gated recurrent unit and an attention mechanism network to mine typical cell feature sequences of batteries with different topologies, and by using a concept machine recurrent neural network and a convolutional neural network encoder to mine the attribute statistical characteristics of the battery feature sequence, the semantic information of the battery data is mined locally and globally, respectively, thereby improving the robustness of the established battery anomaly detection model. In addition, it should be noted that, without departing from the spirit and scope of the present invention, the above steps S101 and S103 can be performed in parallel, or step S103 can be performed before step S101.

[0059] In step S105 , the semantic information of the battery feature sequence generated in step S101 and the semantic information of the cell feature sequence generated in step S103 are merged.

[0060] In step S107, the semantic information of the combined battery feature sequence and the semantic information of the cell feature sequence are input through a fully connected layer to construct a classifier and train the classifier to establish a battery anomaly detection model based on the trained classifier. Alternatively, the semantic information of the combined battery feature sequence and the semantic information of the cell feature sequence can be input into a fully connected layer with an output node of 1 and a sigmoid function as the activation function to construct a binary classifier.

[0061] Optionally, in step S107, the classifier can be dynamically trained to convergence using training data including a training set and a validation set, and the trained classifier with the smallest average validation loss of the validation set is selected as the battery anomaly detection model. In one embodiment, a dynamic K-fold training method can be used to train the classifier, wherein the training data can be divided into a training set and a validation set at a ratio of k:1, and there is no overlap in the batteries in the training set and the validation set during the division. The normal battery data sequence is set to 0 and the abnormal battery data sequence is set to 1, and a binary cross entropy function can be used as the loss function during training. When the average validation loss of the validation set is the smallest, the trained classifier can be determined to be a model with the optimal weight and a battery anomaly detection model can be established based on this.

[0062] The method for establishing a battery anomaly detection model according to one or more embodiments of the present invention can mine battery data from both the overall and local perspectives to generate semantic information for battery feature sequences and semantic information for cell feature sequences, and establish a battery anomaly detection model based on the combined semantic information for the battery feature sequences and the semantic information for the cell feature sequences. This improves the robustness of the battery anomaly detection model and the accuracy of battery anomaly detection using the battery anomaly detection model, thereby improving the availability and safety of the battery. Furthermore, the method for establishing a battery anomaly detection model according to one or more embodiments of the present invention eliminates the need for extensive parameter calibration, saving manpower and material resources spent on model establishment.

[0063] Figure 2 FIG. 1 is a flow chart of a method for battery abnormality detection according to one or more embodiments of the present invention.

[0064] like Figure 2 As shown in , in step S201 , battery data is obtained from the cloud and pre-processed to generate a plurality of battery data sequence segments.

[0065] Optionally, in step S201, preprocessing the battery data to generate multiple battery data sequence segments may include: dividing the battery data into battery data during a period when the battery charge changes and battery data during a period when the battery charge does not change; segmenting the battery data during a period when the battery charge changes using a predetermined SOC interval as a window length; and segmenting the battery data during a period when the battery charge does not change using a predetermined time interval as a window length. Optionally, when the length of the last battery data sequence segment is less than the window length, the length of the battery data sequence segment is compared with 1 / 2 of the window length. If the length of the battery data sequence segment is less than or equal to 1 / 2 of the window length, the battery data sequence segment and the previous window are merged into one battery data sequence segment; if the length of the battery data sequence segment is greater than 1 / 2 of the window length, the battery data sequence segment is separated into a new battery data sequence segment.

[0066] In step S203 , feature extraction is performed on the plurality of battery data sequence segments to generate a battery feature sequence and a cell feature sequence respectively.

[0067] Optionally, extracting features from multiple battery data sequence segments to generate a battery feature sequence may include: determining statistical features corresponding to one or more of cell voltage differential, SOC, and ambient temperature in each of the multiple battery data sequence segments, where the statistical features include categorical features and continuous-valued features; encoding the categorical features in a predetermined manner and normalizing the continuous-valued features to generate the battery feature sequence. Optionally, the length of the generated battery feature sequence for each battery may be maintained uniform; shorter battery feature sequences may be padded with leading zeros to increase their length. Optionally, categorical features may include operating conditions (e.g., charging, discharging, or vehicle stopped and not charging), and continuous-valued features may include mean, maximum, minimum, quantile, skewness, and kurtosis. In one embodiment, categorical features may be encoded using a "one-hot" approach, which can be understood as a binary encoding method where the encoded result contains only one binary "1." For example, when using a "one-hot" approach to encode the charging process, discharging process, and vehicle stopped and not charging process included in the operating state, the charging process can be encoded as "000", the discharging process can be encoded as "010", and the vehicle stopped and not charging process can be encoded as "100". In one embodiment, the maximum and minimum value method can be used to normalize the continuous value features.

[0068] Optionally, performing feature extraction on multiple battery data sequence fragments to generate a cell feature sequence may include: determining, for each battery data sequence fragment in the multiple battery data sequence fragments, a set of voltage means of each cell in the multiple cells of the battery; normalizing the set of voltage means of each cell in the multiple cells of the battery to generate voltage sequence data; and processing the voltage sequence data using a feature sampling method to generate a cell feature sequence including a first part of the cell feature sequence and a second part of the cell feature sequence. Optionally, the first part of the cell feature sequence and the second part of the cell feature sequence may be connected together in a certain order in the feature dimension to obtain the cell feature sequence. It should be noted that the process of the feature sampling method is the same as Figure 1 The process of the feature sampling method described in step S103 is similar and will not be further described here.

[0069] In step S205, using the combination Figure 1The battery anomaly detection model established by the method described processes the battery feature sequence and the cell feature sequence to generate a battery anomaly detection result. Optionally, the battery anomaly detection result can be represented by an anomaly score ranging from 0 to 1, and the anomaly score can be uploaded to the battery management cloud center. It is understandable that an anomaly score of 0 indicates that the battery is working normally, and an anomaly score of 1 indicates that the battery is working abnormally. In addition, according to the battery usage scenario and actual needs, a value between 0 and 1 can be selected as the threshold. When the anomaly score is less than the threshold, it can be judged that the battery is working normally, and when the anomaly score is greater than or equal to the threshold, it can be judged that the battery is working abnormally.

[0070] The method for battery anomaly detection according to one or more embodiments of the present invention can improve the accuracy of battery anomaly detection and has a higher recall rate and precision rate than other battery anomaly detection methods, thereby improving the availability and safety of the battery.

[0071] Figure 3 FIG. 1 is a block diagram of a battery abnormality detection system according to one or more embodiments of the present invention.

[0072] like Figure 3 As shown in , the battery abnormality detection system 300 includes a data acquisition module 310 , a cloud processing module 320 , a feature extraction module 330 and an abnormality detection module 340 .

[0073] The data acquisition module 310 may be configured to collect battery data such as battery temperature, voltage, SOC, etc., and may be configured to upload the collected battery data to the cloud processing module 320 at a predetermined frequency.

[0074] The cloud processing module 320 may be configured to store the collected battery data in a cloud database.

[0075] The feature extraction module 330 can be configured to obtain the collected battery data from the cloud database and determine the collected battery data corresponding to each battery according to the corresponding battery ID. The feature extraction module 330 can be configured to pre-process the collected battery data to generate multiple battery data sequence segments, and perform feature extraction on the multiple battery data sequence segments to generate a battery feature sequence and a cell feature sequence respectively. The above pre-processing process and feature extraction process can be referred to Figure 1 The preprocessing process and feature extraction process described in step S101 and step S103 in FIG will not be further described here.

[0076] The abnormality detection module 340 can be configured to process the battery feature sequence and the cell feature sequence to generate a battery abnormality detection result. Figure 1Optionally, the anomaly detection module 340 can upload the battery anomaly detection results to the battery management cloud center.

[0077] Figure 4 FIG. 1 is a block diagram of a computer device according to an embodiment of the present invention. Figure 4 As shown in FIG, a computer device 400 includes a memory 410, a processor 420, and a computer program 430 stored in the memory 410 and executable on the processor 420. When the processor 420 executes the computer program 430, the processor 420 implements the various steps of the method for establishing a battery abnormality detection model according to one or more embodiments of the present invention, or implements the various steps of the method for battery abnormality detection according to one or more embodiments of the present invention.

[0078] In addition, as described above, the present invention can also be implemented as a computer storage medium, in which a program is stored for causing a computer to execute a method for establishing a battery abnormality detection model according to one or more embodiments of the present invention, or a program is stored for causing a computer to execute a method for battery abnormality detection according to one or more embodiments of the present invention.

[0079] Here, as computer storage media, various types of computer storage media can be used, such as disks (for example, magnetic disks, optical disks, etc.), cards (for example, memory cards, optical cards, etc.), semiconductor memories (for example, read-only memories, non-volatile memories, etc.), and tapes (for example, magnetic tapes, cassette tapes, etc.).

[0080] In the case of applicable, hardware, software or a combination of hardware and software can be used to realize the various embodiments provided by the present invention. Moreover, in the case of applicable, without departing from the scope of the present invention, the various hardware components and / or software components set forth herein can be combined into composite components comprising software, hardware and / or both. In the case of applicable, without departing from the scope of the present invention, the various hardware components and / or software components set forth herein can be divided into subcomponents comprising software, hardware or both. In addition, in the case of applicable, it is contemplated that software components can be implemented as hardware components, and vice versa.

[0081] Software according to the present invention (such as program code and / or data) can be stored on one or more computer storage media. It is also contemplated that the software identified herein can be implemented using one or more general-purpose or special-purpose computers and / or computer systems, networked and / or otherwise. Where applicable, the order of the various steps described herein can be changed, combined into composite steps, and / or divided into sub-steps to provide the features described herein.

[0082] The embodiments and examples set forth herein are provided to best illustrate embodiments according to the present invention and its specific applications, and thereby enable those skilled in the art to make and use the invention. However, those skilled in the art will appreciate that the above description and examples are provided for ease of illustration and example only. The descriptions set forth are not intended to encompass all aspects of the invention or to limit the invention to the precise forms disclosed.

Claims

1. A method for establishing a battery anomaly detection model, characterized in that: The method comprises the following steps: Processing a battery feature sequence via a first network structure to generate semantic information of the battery feature sequence; Processing the cell feature sequence via a second network structure to generate semantic information of the cell feature sequence; Merging the semantic information of the battery feature sequence and the semantic information of the cell feature sequence; as well as Inputting the combined semantic information of the battery feature sequence and the semantic information of the cell feature sequence through a fully connected layer to construct a classifier and train the classifier, so as to establish the battery anomaly detection model based on the trained classifier, Processing the battery feature sequence via the first network structure to generate semantic information of the battery feature sequence includes: Characterizing the battery feature sequence as a pseudo image feature map using a concept machine recursive neural network; and Encoding the pseudo image feature map using a convolutional neural network encoder to generate semantic information of the battery feature sequence, Processing the cell feature sequence via the second network structure to generate semantic information of the cell feature sequence includes: Performing feature encoding on the cell feature sequence to obtain shallow sequence features of the cell feature sequence; Obtaining correlations between sequences of shallow sequence features of the cell feature sequence using a bidirectional gated recurrent unit, and determining deep sequence features of the cell feature sequence based on the correlations; and An attention mechanism network is used to obtain the weighted sequence dimension of the deep sequence features of the battery cell feature sequence, so as to generate semantic information of the battery cell feature sequence based on the deep sequence features of the battery cell feature sequence and the weighted sequence dimension of the deep sequence features of the battery cell feature sequence.

2. The method according to claim 1, wherein the battery characteristic sequence and the cell characteristic sequence are obtained by: Acquiring and preprocessing battery data to generate a plurality of battery data sequence segments, wherein the battery data includes normal battery data and abnormal battery data; and Feature extraction is performed on the multiple battery data sequence segments to generate the battery feature sequence and the cell feature sequence respectively.

3. The method according to claim 2, wherein pre-processing the battery data to generate a plurality of battery data sequence segments comprises: Dividing the battery data into battery data during a period when the battery power level changes and battery data during a period when the battery power level does not change; Segmenting the battery data during a period when the battery power level changes using a predetermined SOC interval as a window length; as well as The battery data during a period in which the battery power level does not change is divided using a predetermined time interval as a window length.

4. The method according to claim 2, wherein performing feature extraction on the plurality of battery data sequence segments to generate the battery feature sequence comprises: Determining statistical features corresponding to one or more of a cell voltage difference, a SOC, and an ambient temperature in each of the plurality of battery data sequence segments, wherein the statistical features include categorical features and continuous value features; as well as The category features are encoded in a preset manner and the continuous value features are normalized to generate the battery feature sequence. 5 . The method of claim 4 , wherein the categorical feature comprises an operating state, and the continuous-valued feature comprises one or more of a mean, a maximum, a minimum, a quantile, a skewness, and a kurtosis.

6. The method according to claim 2, wherein extracting features from the plurality of battery data sequence segments to generate the battery cell feature sequence comprises: determining, for each battery data sequence segment of the plurality of battery data sequence segments, a set of voltage mean values ​​of each battery cell in a plurality of battery cells of the battery; Normalizing a set of voltage mean values ​​of each of the plurality of battery cells comprising the battery to generate voltage series data; as well as The voltage sequence data is processed by using a feature sampling method to generate the cell feature sequence including the first part of the cell feature sequence and the second part of the cell feature sequence.

7. The method according to claim 6, wherein processing the voltage sequence data using a feature sampling method to generate the first part of the cell feature sequence comprises: traversing one or more battery data sequence segments of the plurality of battery data sequence segments, and determining a first plurality of battery cells associated with maximum values, minimum values, and median values ​​in the voltage sequence data corresponding to the one or more battery data sequence segments, and a second plurality of battery cells adjacent to the first plurality of battery cells in a battery cell topology; as well as Normalized voltage averages corresponding to the first plurality of battery cells and the second plurality of battery cells are sampled from the voltage sequence data to generate a first part of the battery cell feature sequence.

8. The method according to claim 6, wherein processing the voltage sequence data using a feature sampling method to generate the second part of the cell feature sequence comprises: traversing each battery data sequence segment of the plurality of battery data sequence segments, and determining a third plurality of battery cells associated with a maximum value, a minimum value, and a median value in the voltage sequence data, and a fourth plurality of battery cells adjacent to the third plurality of battery cells in a battery cell topology structure; as well as Normalized voltage averages corresponding to the third plurality of battery cells and the fourth plurality of battery cells are sampled from the voltage sequence data to generate a second part of the battery cell feature sequence. 9 . The method according to claim 7 , wherein the one or more battery data sequence segments of the plurality of battery data sequence segments include a first battery data sequence segment and a last battery data sequence segment of the plurality of battery data sequence segments.

10. The method according to claim 1, wherein training the classifier to establish the battery anomaly detection model based on the trained classifier comprises: Dynamically training the classifier to convergence using training data including a training set and a validation set; as well as The classifier trained when the average validation loss of the validation set is the smallest is selected as the battery anomaly detection model.

11. A method for detecting battery abnormality, characterized in that: The method comprises the following steps: acquiring battery data from a cloud and preprocessing the battery data to generate a plurality of battery data sequence segments; Performing feature extraction on the multiple battery data sequence fragments to generate a battery feature sequence and a cell feature sequence respectively; as well as The battery abnormality detection model established according to any one of claims 1 to 10 is used to process the battery feature sequence and the cell feature sequence to generate a battery abnormality detection result.

12. The method according to claim 11, wherein pre-processing the battery data to generate a plurality of battery data sequence segments comprises: Dividing the battery data into battery data during a period when the battery power level changes and battery data during a period when the battery power level does not change; Segmenting the battery data during a period when the battery power level changes using a predetermined SOC interval as a window length; as well as The battery data during a period in which the battery power level does not change is divided using a predetermined time interval as a window length.

13. The method according to claim 11, wherein performing feature extraction on the plurality of battery data sequence segments to generate the battery feature sequence comprises: Determining statistical features corresponding to one or more of a cell voltage difference, a SOC, and an ambient temperature in each of the plurality of battery data sequence segments, wherein the statistical features include categorical features and continuous value features; as well as The category features are encoded in a preset manner and the continuous value features are normalized to generate the battery feature sequence. 14 . The method of claim 13 , wherein the categorical feature comprises an operating state, and the continuous-valued feature comprises one or more of a mean, a maximum, a minimum, a quantile, a skewness, and a kurtosis.

15. The method according to claim 11, wherein performing feature extraction on the plurality of battery data sequence segments to generate the battery cell feature sequence comprises: determining, for each battery data sequence segment of the plurality of battery data sequence segments, a set of voltage mean values ​​of each battery cell in a plurality of battery cells of the battery; Normalizing a set of voltage mean values ​​of each of the plurality of battery cells comprising the battery to generate voltage series data; as well as The voltage sequence data is processed by using a feature sampling method to generate the cell feature sequence including the first part of the cell feature sequence and the second part of the cell feature sequence.

16. The method according to claim 15, wherein processing the voltage sequence data using a feature sampling method to generate the first part of the cell feature sequence comprises: traversing one or more battery data sequence segments of the plurality of battery data sequence segments, and determining a first plurality of battery cells associated with maximum values, minimum values, and median values ​​in the voltage sequence data corresponding to the one or more battery data sequence segments, and a second plurality of battery cells adjacent to the first plurality of battery cells in a battery cell topology; as well as Normalized voltage averages corresponding to the first plurality of battery cells and the second plurality of battery cells are sampled from the voltage sequence data to generate a first part of the battery cell feature sequence.

17. The method according to claim 15, wherein processing the voltage sequence data using a feature sampling method to generate the second part of the cell feature sequence comprises: traversing each battery data sequence segment of the plurality of battery data sequence segments, and determining a third plurality of battery cells associated with a maximum value, a minimum value, and a median value in the voltage sequence data, and a fourth plurality of battery cells adjacent to the third plurality of battery cells in a battery cell topology structure; as well as Normalized voltage averages corresponding to the third plurality of battery cells and the fourth plurality of battery cells are sampled from the voltage sequence data to generate a second part of the battery cell feature sequence. 18 . The method of claim 16 , wherein the one or more battery data sequence segments of the plurality of battery data sequence segments include a first battery data sequence segment and a last battery data sequence segment of the plurality of battery data sequence segments.

19. A computer storage medium, characterized in that The computer storage medium comprises instructions that, when executed, perform the method according to any one of claims 1 to 10.

20. A computer storage medium, characterized in that The computer storage medium comprises instructions that, when executed, perform the method according to any one of claims 11 to 18.

21. A computer device, characterized in that: The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 10 when executing the program.

22. A computer device, characterized in that: The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 11 to 18 when executing the program.

23. A computer program product, characterized in that The computer program product comprises instructions which, when executed by a processor, implement the method according to any one of claims 1 to 10.

24. A computer program product, characterized in that The computer program product comprises instructions which, when executed by a processor, implement the method according to any one of claims 11 to 18.

Images