A battery pack intelligent monitoring system and method based on deep learning

Abstract

The present invention relates to the field of deep learning technology, and invents a deep learning-based intelligent monitoring system and method for a battery pack, comprising performing data standardization and outlier denoising operations on the historical electrical data set group to obtain an electrical data set sequence group; performing electrical feature extraction and correlation feature fusion on the electrical data set sequence group to obtain a fused electrical feature sequence group; performing graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence; performing self-attention time series convolution on the electrical graph feature sequence to obtain electrical time series features; performing variational anomaly monitoring and anomaly recognition on the electrical time series features to obtain a battery pack monitoring result. The present invention can improve the accuracy of battery pack monitoring.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a battery pack intelligent monitoring system and method based on deep learning. Background Art

[0002] A battery pack is a whole composed of multiple single batteries connected in a certain way. It is commonly used in electric vehicles, energy storage systems, uninterruptible power supplies and other occasions to provide stable power output. In order to ensure the safety of customers' electricity use, it is necessary to carry out intelligent monitoring of the battery pack.

[0003] Most of the existing battery pack monitoring methods are based on physical sensor signals, that is, they rely on various physical sensor systems to monitor the electrical data of the battery pack, and determine the corresponding abnormality when the electrical data is unstable or exceeds a preset numerical threshold. However, in actual application, the monitoring method based on physical sensor signals can only detect specific physical phenomena and cannot handle more complex causes of abnormal phenomena. In addition, the existing battery pack monitoring method often monitors the battery pack as a whole, and it is difficult to monitor the abnormality of the battery pack caused by current disorder between batteries, which may lead to low accuracy in battery pack monitoring. Summary of the invention

[0004] The present invention provides a battery pack intelligent monitoring system and method based on deep learning, the main purpose of which is to solve the problem of low accuracy when monitoring the battery pack.

[0005] To achieve the above object, the present invention provides a battery pack intelligent monitoring system based on deep learning, the system includes a data cleaning module, a feature extraction module, a graph reconstruction module, a time series convolution module and an abnormality monitoring module, wherein:

[0006] The data cleaning module is used to obtain a historical electrical data set group of the battery group, perform data standardization and outlier denoising operations on the historical electrical data set group, and obtain an electrical data set sequence group;

[0007] The feature extraction module is used to perform electrical feature extraction and correlation feature fusion on the electrical data set sequence group to obtain a fused electrical feature sequence group. When the feature extraction module performs electrical feature extraction and correlation feature fusion on the electrical data set sequence group to obtain a fused electrical feature sequence group, the module includes: performing electrical feature extraction on the electrical data set sequence group to obtain an electrical feature set sequence group; regrouping the electrical feature set sequence group according to a preset data type to obtain a single-type electrical feature set; performing feature standardization on the single-type electrical feature set to obtain a standard electrical feature set; and calculating the electrical correlation between each standard electrical feature class in the standard electrical feature set using the following electrical correlation algorithm:

[0008] in, Refers to the standard electrical characteristics class set of standard electrical characteristics class and standard electrical characteristics electrical correlation, i and j are the serial numbers of the standard electrical characteristic classes in the standard electrical characteristic class set, The standard electrical characteristics are The total number of features, The standard electrical characteristics are The total number of features, k, is the feature number, is the covariance operator symbol, is the standard deviation operator, Refers to the standard electrical characteristics class The kth feature in Refers to the standard electrical characteristics class The features; performing correlation feature fusion on the electrical feature set sequence group according to all electrical correlations of the standard electrical feature set to obtain a fused electrical feature sequence group;

[0009] The graph reconstruction module is used to obtain the battery pack structure of the battery pack, and perform graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence;

[0010] The timing convolution module is used to perform self-attention timing convolution on the electrical diagram feature sequence to obtain electrical timing features;

[0011] The abnormality monitoring module is used to perform variational abnormality monitoring and abnormality identification on the electrical timing characteristics to obtain battery pack monitoring results.

[0012] Optionally, when the data cleaning module performs data standardization and outlier denoising operations on the historical electrical data set group to obtain the electrical data set sequence group, the data cleaning module includes:

[0013] Performing a data format unification operation on the historical electrical data set group to obtain a standard electrical data set group;

[0014] Deduplication of data is performed on the standard electrical data set group to obtain a deduplication electrical data set group;

[0015] Performing outlier screening on the deduplicated electrical data set group to obtain a de-noised electrical data set group;

[0016] Sorting the data in the denoised electrical data set group in a time sequence order to obtain a denoised electrical data set sequence group;

[0017] Gaussian interpolation is performed on the denoised electrical data set sequence group to obtain an electrical data set sequence group.

[0018] Optionally, when the data cleaning module performs outlier screening on the deduplicated electrical data set group to obtain a denoised electrical data set group, the data cleaning module includes:

[0019] Regrouping the deduplicated electrical data set group according to preset data types to obtain a single-type electrical data set;

[0020] Selecting one by one the single-type electrical data groups in the single-type electrical data group set as target single-type electrical data groups, and selecting one by one the data in the target single-type electrical data groups as target electrical data;

[0021] The outlier contour distance of the target electrical data in the target single-type electrical data group is calculated using the following outlier contour distance algorithm:

[0022] ;

[0023] in, is the outlier contour distance, p is the target electrical data, N is the total number of data in the target single-type electrical data group, n is the data index, Refers to the nth data in the target single type electrical data group, is the transpose symbol, is the symbol of the covariance function;

[0024] Filtering outlier electrical data sets from the single-type electrical data set according to the outlier contour distance of each target electrical data in the single-type electrical data set;

[0025] The outlier electrical data set is used to perform an outlier data screening operation on the deduplicated electrical data set group to obtain a denoised electrical data set group.

[0026] Optionally, when the feature extraction module performs correlation feature fusion on the electrical feature set sequence group according to all electrical correlations of the standard electrical feature set to obtain a fused electrical feature sequence group, the feature extraction module includes:

[0027] generating an electrical correlation matrix according to all electrical correlations of the standard electrical feature set;

[0028] Calculating a weight matrix of the standard electrical feature set according to the electrical correlation matrix;

[0029] Performing a matrix row sum operation on the weight matrix to obtain a row feature weight group;

[0030] Normalizing each row feature weight in the row feature weight group to obtain a feature weight group;

[0031] A feature weighted fusion operation is performed on each electrical feature set in the electrical feature set sequence group according to the feature weight group to obtain a fused electrical feature sequence group.

[0032] Optionally, when the graph reconstruction module performs graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence, the graph reconstruction module includes:

[0033] extracting a battery node set from the battery pack structure;

[0034] Performing edge mapping on the battery node set using the battery group structure to obtain a battery group graph;

[0035] Selecting electrical feature sequences in the electrical feature sequence group one by one as target electrical feature sequences, and taking battery nodes corresponding to the target electrical feature sequences in the battery node set as target battery nodes;

[0036] Performing node feature mapping on the target battery node in the battery group diagram using the target electrical feature sequence to obtain a primary diagram feature sequence, until the target electrical feature sequence is the last electrical feature sequence in the electrical feature sequence group, and using the primary diagram feature sequence as a reconstructed diagram feature sequence;

[0037] A graph convolution operation is performed on each reconstructed graph feature in the reconstructed graph feature sequence to obtain an electrical graph feature sequence.

[0038] Optionally, when the timing convolution module performs self-attention timing convolution on the electrical diagram feature sequence to obtain the electrical timing feature, it includes:

[0039] Performing short-term recursive convolution on the electrical diagram feature sequence to obtain short-term electrical features;

[0040] Performing long-term jump convolution on the electrical graph feature sequence to obtain long-term electrical features;

[0041] Attention encoding is performed on the electrical graph feature sequence according to a self-attention mechanism to obtain an attention electrical feature;

[0042] The short-term electrical feature, the long-term electrical feature and the attention electrical feature are fused using the following attention connection algorithm to obtain an electrical timing feature: ;

[0043] in, is the characteristic value of the electrical timing characteristic at time step t, is the characteristic weight of the short-term electrical characteristic, is the characteristic value of the short-term electrical feature at the tth time step, r, t is the time step index, y is the number of hidden units skipped by the long-term electrical feature when performing long-term skip convolution, is the characteristic weight of the long-term electrical characteristic at the rth time step, The long-term electrical characteristics are described in The eigenvalues ​​of the time steps, is the characteristic value of the attention electrical feature at the tth time step, and b is the preset bias coefficient.

[0044] Optionally, when the abnormality monitoring module performs variational abnormality monitoring and decision tree abnormality monitoring on the electrical timing characteristics to obtain the battery pack monitoring result, it includes:

[0045] Performing variational anomaly monitoring on the electrical timing characteristics to obtain primary monitoring results;

[0046] Performing decision tree abnormality monitoring on the electrical timing characteristics to obtain secondary monitoring results;

[0047] A battery pack monitoring result is generated according to the primary monitoring result and the secondary monitoring result.

[0048] Optionally, when the abnormality monitoring module performs variational abnormality monitoring on the electrical timing feature to obtain a primary monitoring result, the abnormality monitoring module includes:

[0049] Performing latent space mapping on the electrical timing characteristics to obtain electrical latent distribution;

[0050] Reparameterizing the electrical potential distribution to obtain potential electrical variables;

[0051] Performing latent space inverse mapping on the latent electrical variables to obtain electrical reconstruction time series features;

[0052] Performing reconstruction error calculation on the electrical reconstruction timing characteristics and the electrical timing characteristics to obtain an electrical error;

[0053] Performing time series convolution on the electrical error to obtain a time series feature of the error;

[0054] Anomaly mapping is performed on the error time series characteristics to obtain primary monitoring results.

[0055] Optionally, when the abnormality monitoring module performs decision tree abnormality monitoring on the electrical timing feature to obtain a secondary monitoring result, it includes:

[0056] Performing feature space splitting on the electrical timing features to obtain an electrical timing space feature group;

[0057] Generate an abnormal decision tree group according to the electrical time series spatial feature group;

[0058] Using the abnormal decision tree group to perform leaf node matching on the electrical timing characteristics to obtain an abnormal leaf node group;

[0059] Performing outlier matching on the abnormal leaf node group to obtain an outlier group;

[0060] An abnormal decision is made on the abnormal value group to obtain a secondary monitoring result.

[0061] In order to solve the above problems, the present invention also provides a battery pack intelligent monitoring method based on deep learning, the method comprising:

[0062] Acquire a historical electrical data set group of the battery group, perform data standardization and outlier denoising operations on the historical electrical data set group, and obtain an electrical data set sequence group;

[0063] The electrical feature extraction and correlation feature fusion are performed on the electrical data set sequence group to obtain a fused electrical feature sequence group, wherein the electrical feature extraction and correlation feature fusion are performed on the electrical data set sequence group to obtain a fused electrical feature sequence group, including: extracting electrical features from the electrical data set sequence group to obtain an electrical feature set sequence group; regrouping the electrical feature set sequence group according to a preset data type to obtain a single-type electrical feature set; performing feature standardization operation on the single-type electrical feature set to obtain a standard electrical feature set; and calculating the electrical correlation between each standard electrical feature class in the standard electrical feature set using the following electrical correlation algorithm:

[0064] ;

[0065] in, Refers to the standard electrical characteristics class set of standard electrical characteristics class and standard electrical characteristics electrical correlation, i and j are the serial numbers of the standard electrical characteristic classes in the standard electrical characteristic class set, The standard electrical characteristics are The total number of features, The standard electrical characteristics are The total number of features, k, is the feature number, is the covariance operator symbol, is the standard deviation operator, Refers to the standard electrical characteristics class The kth feature in Refers to the standard electrical characteristics class The features; performing correlation feature fusion on the electrical feature set sequence group according to all electrical correlations of the standard electrical feature set to obtain a fused electrical feature sequence group;

[0066] Acquire the battery pack structure of the battery pack, and perform graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence;

[0067] Performing self-attention time series convolution on the electrical graph feature sequence to obtain electrical time series features;

[0068] Variational abnormality monitoring and abnormality identification are performed on the electrical timing characteristics to obtain a battery pack monitoring result.

[0069] The embodiment of the present invention obtains an electrical data set sequence group by performing data standardization and outlier denoising operations, which can effectively remove noise data in the historical electrical data set group, thereby improving the representation of subsequent electrical time series features and the accuracy of electrical feature extraction. By performing electrical feature extraction and correlation feature fusion, the electrical data of different modes in a single battery can be fused into an overall fused electrical feature, thereby improving the comprehensiveness of battery monitoring and facilitating subsequent intelligent monitoring of the relationship between batteries. By performing graph reconstruction and graph feature extraction, the graph neural network can be used to extract the correlation features of the electrical data between each battery in the battery group, thereby realizing block-based monitoring of the battery group, thereby improving the monitoring resolution and further improving the accuracy of intelligent monitoring.

[0070] By performing self-attention temporal convolution, it is possible to combine graph neural networks and temporal neural networks to simultaneously capture the spatial dependencies in the graph structure and the dynamic changes in the time series, thereby improving the comprehensiveness of battery pack monitoring and the accuracy of anomaly analysis. By performing variational anomaly monitoring and decision tree anomaly monitoring, it is possible to combine unsupervised models and supervised models for anomaly analysis, thereby improving the comprehensiveness of anomaly analysis and the accuracy of intelligent monitoring, thereby improving the accuracy of battery pack monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 A functional module diagram of a battery pack intelligent monitoring system based on deep learning provided by one embodiment of the present invention;

[0072] Figure 2 A flowchart of generating an electrical data set sequence group provided by an embodiment of the present invention;

[0073] Figure 3 A flowchart of extracting a feature sequence of graph events provided by an embodiment of the present invention;

[0074] Figure 4 A schematic diagram of a flow chart of a battery pack intelligent monitoring method based on deep learning provided by an embodiment of the present invention;

[0075] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0076] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0077] The embodiment of the present application provides a battery pack intelligent monitoring system based on deep learning. The execution subject of the battery pack intelligent monitoring system based on deep learning includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the system provided by the embodiment of the present application. In other words, the battery pack intelligent monitoring system based on deep learning can be executed by software or hardware installed on a terminal device or a server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms.

[0078] like Figure 1, is a functional module diagram of a battery pack intelligent monitoring system based on deep learning provided in one embodiment of the present invention.

[0079] The battery pack intelligent monitoring system 100 based on deep learning of the present invention can be installed in an electronic device. According to the functions implemented, the battery pack intelligent monitoring system 100 based on deep learning can include a data cleaning module 101, a feature extraction module 102, a graph reconstruction module 103, a time series convolution module 104 and an abnormality monitoring module 105. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0080] In this embodiment, the functions of each module / unit are as follows:

[0081] The data cleaning module is used to obtain a historical electrical data set group of the battery group, perform data standardization and outlier denoising operations on the historical electrical data set group, and obtain an electrical data set sequence group.

[0082] In detail, the battery pack is a power system composed of multiple single batteries connected in series, parallel or hybrid. The battery pack stores energy through electrochemical reactions and releases electrical energy when needed. It is commonly found in electric vehicles, backup power supplies, uninterruptible power supply systems and renewable energy storage systems.

[0083] Specifically, each historical electrical data set in the historical electrical data set group corresponds to an electrical data set of a single battery in the battery pack. The electrical data set refers to a data set of various electrical parameter information related to the operating status of the electrical system or electrical equipment. The electrical data set includes data such as voltage, current, power, electric energy, resistance, temperature and humidity.

[0084] In detail, the battery pack may be monitored using a variety of electrical sensors to obtain a historical electrical data set. The electrical sensors may be voltage sensors, current sensors, temperature sensors, humidity sensors and other sensors.

[0085] In the embodiment of the present invention, refer to Figure 2 As shown, when the data cleaning module performs data standardization and outlier denoising operations on the historical electrical data set group to obtain the electrical data set sequence group, it includes:

[0086] S21, performing a data format unification operation on the historical electrical data set to obtain a standard electrical data set;

[0087] S22, performing data deduplication on the standard electrical data set group to obtain a deduplicated electrical data set group;

[0088] S23, screening outliers from the deduplicated electrical data set to obtain a denoised electrical data set;

[0089] S24, sorting the data in the denoised electrical data set group in chronological order to obtain a denoised electrical data set sequence group;

[0090] S25 . Perform Gaussian interpolation on the denoised electrical data set sequence group to obtain an electrical data set sequence group.

[0091] In detail, the data format unification refers to unifying the format of data with the same data type in the historical electrical data set, for example, unifying the unit of data with the data type of voltage into volts, and unifying the unit of data with the data type of temperature into degrees Celsius.

[0092] Specifically, data deduplication can be performed using hash deduplication, double pointer deduplication or Bloom filter, and each deduplication electrical data set sequence in the deduplication electrical data set sequence group represents an electrical data set of a battery in the battery group sorted in chronological order.

[0093] In detail, when the data cleaning module performs outlier screening on the deduplicated electrical data set group to obtain a denoised electrical data set group, it includes:

[0094] Regrouping the deduplicated electrical data set group according to preset data types to obtain a single-type electrical data set;

[0095] Selecting one by one the single-type electrical data groups in the single-type electrical data group set as target single-type electrical data groups, and selecting one by one the data in the target single-type electrical data groups as target electrical data;

[0096] Calculating the outlier contour distance of the target electrical data in the target single-type electrical data group;

[0097] Filtering outlier electrical data sets from the single-type electrical data set according to the outlier contour distance of each target electrical data in the single-type electrical data set;

[0098] The outlier electrical data set is used to perform an outlier data screening operation on the deduplicated electrical data set group to obtain a denoised electrical data set group.

[0099] In detail, the data type refers to the type of electrical data, such as voltage, current, and temperature, etc. Each single-type electrical data group in the single-type electrical data group set corresponds to all data of a certain data type in the deduplicated electrical data set set.

[0100] Specifically, the outlier contour distance is calculated using the following outlier contour distance algorithm:

[0101] ;

[0102] in, is the outlier contour distance, p is the target electrical data, N is the total number of data in the target single-type electrical data group, n is the data index, Refers to the nth data in the target single type electrical data group, is the transpose symbol, is the symbol of the covariance function.

[0103] In detail, the outlier profile distance algorithm can combine the covariance relationship between data to perform distance calculation, and is applicable to multidimensional data with non-uniform scales or different correlations, thereby facilitating the discovery of outlier data in each target single-type electrical data group.

[0104] Specifically, the outlier electrical data group refers to target electrical data in the single-type electrical data group set whose outlier profile distance is greater than a preset data threshold as outlier electrical data collected into the outlier electrical data group.

[0105] In detail, the outlier data screening refers to screening out the data corresponding to the outlier electrical data set from the deduplicated electrical data set group, the Gaussian interpolation refers to using a Gaussian kernel function to smoothly insert missing data for smoothing processing, and the missing data refers to the data corresponding to the outlier electrical data set screened out by the outlier data screening.

[0106] In the embodiment of the present invention, by performing data standardization and outlier denoising operations to obtain an electrical data set sequence group, the noise data in the historical electrical data set group can be effectively removed, thereby improving the representation of subsequent electrical timing characteristics and the accuracy of electrical feature extraction.

[0107] The feature extraction module is used to perform electrical feature extraction and correlation feature fusion on the electrical data set sequence group to obtain a fused electrical feature sequence group.

[0108] In detail, each fused electrical feature sequence in the fused electrical feature sequence group corresponds to the electrical features of all electrical data of a storage battery in the battery group in a past time period.

[0109] In the embodiment of the present invention, refer to Figure 3 As shown, when the feature extraction module performs electrical feature extraction and correlation feature fusion on the electrical data set sequence group to obtain a fused electrical feature sequence group, it includes:

[0110] S31, extracting electrical features from the electrical data set sequence group to obtain an electrical feature set sequence group;

[0111] S32, regrouping the electrical feature set sequence group according to preset data types to obtain a single-type electrical feature set;

[0112] S33, performing a feature standardization operation on the single-type electrical feature set to obtain a standard electrical feature set;

[0113] S34, calculating the electrical correlation between each standard electrical characteristic class in the standard electrical characteristic class set;

[0114] S35 , performing correlation feature fusion on the electrical feature set sequence group according to all electrical correlations of the standard electrical feature set to obtain a fused electrical feature sequence group.

[0115] In detail, the electrical feature extraction refers to vectorizing each electrical data in the electrical data set sequence group into electrical features, each single-type electrical feature class in the single-type electrical feature class set represents a class group of all electrical features of a data type in the electrical feature set sequence group, and the feature standardization refers to normalizing each electrical feature in the single-type electrical feature class set to a data form with a mean of 0 and a mean square error of 1.

[0116] Specifically, the electrical correlation is calculated using the following electrical correlation algorithm:

[0117] ;

[0118] in, Refers to the standard electrical characteristics class set of standard electrical characteristics class and standard electrical characteristics electrical correlation, i and j are the serial numbers of the standard electrical characteristic classes in the standard electrical characteristic class set, The standard electrical characteristics are The total number of features, The standard electrical characteristics are The total number of features, k, is the feature number, is the covariance operator symbol, is the standard deviation operator, Refers to the standard electrical characteristics class The kth feature in Refers to the standard electrical characteristics class The Features.

[0119] In detail, the electrical correlation algorithm can handle the complex linear relationship between multi-dimensional feature groups, evenly consider the impact of each feature pair, and eliminate the difference in dimensions through normalization, and is suitable for comprehensive correlation analysis of electrical feature groups.

[0120] Specifically, when the feature extraction module performs correlation feature fusion on the electrical feature set sequence group according to all electrical correlations of the standard electrical feature set to obtain a fused electrical feature sequence group, it includes:

[0121] generating an electrical correlation matrix according to all electrical correlations of the standard electrical feature set;

[0122] Calculating a weight matrix of the standard electrical feature set according to the electrical correlation matrix;

[0123] Performing a matrix row sum operation on the weight matrix to obtain a row feature weight group;

[0124] Normalizing each row feature weight in the row feature weight group to obtain a feature weight group;

[0125] A feature weighted fusion operation is performed on each electrical feature set in the electrical feature set sequence group according to the feature weight group to obtain a fused electrical feature sequence group.

[0126] In detail, the electrical correlation matrix refers to a matrix that arranges the electrical correlations of each standard electrical feature class in the standard electrical feature class set, and the weight matrix is ​​a matrix composed of each matrix element in the electrical correlation matrix.

[0127] Specifically, the row summation operation refers to calculating the sum of the weights of each row of the weight matrix as the row feature weight, and normalization operation can be performed using normalization functions such as softmax or sigmoid. The feature weighted fusion operation refers to multiplying the electrical features of each corresponding data type in the electrical feature set by the feature weight of the corresponding data type in the feature weight group, and taking the sum of all products as the fused electrical feature.

[0128] In the embodiment of the present invention, by performing electrical feature extraction and correlation feature fusion, the electrical data of different modes in a single battery can be fused into an overall fused electrical feature, thereby improving the comprehensiveness of battery monitoring and facilitating subsequent intelligent monitoring of the relationship between batteries.

[0129] The graph reconstruction module is used to obtain the battery pack structure of the battery pack, and perform graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence.

[0130] In detail, the battery pack structure refers to a structural representation that describes the connection relationship and positional relationship between each battery in the battery pack. Each electrical diagram feature in the electrical diagram feature sequence corresponds to a certain moment in a past time period. The electrical characteristic relationship between each battery in the battery pack can be obtained by obtaining the circuit diagram of the battery pack.

[0131] In the embodiment of the present invention, when the graph reconstruction module performs graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence, it includes:

[0132] extracting a battery node set from the battery pack structure;

[0133] Performing edge mapping on the battery node set using the battery group structure to obtain a battery group graph;

[0134] Selecting electrical feature sequences in the electrical feature sequence group one by one as target electrical feature sequences, and taking battery nodes corresponding to the target electrical feature sequences in the battery node set as target battery nodes;

[0135] Performing node feature mapping on the target battery node in the battery group diagram using the target electrical feature sequence to obtain a primary diagram feature sequence, until the target electrical feature sequence is the last electrical feature sequence in the electrical feature sequence group, and using the primary diagram feature sequence as a reconstructed diagram feature sequence;

[0136] A graph convolution operation is performed on each reconstructed graph feature in the reconstructed graph feature sequence to obtain an electrical graph feature sequence.

[0137] In detail, each battery node in the battery node set is a node corresponding to a single storage battery in the battery pack structure, and the edge mapping refers to extracting the connection relationship between the battery nodes corresponding to the battery node set from the battery pack structure, and generating a node edge between two battery nodes with a connection relationship, wherein the connection relationship can be extracted by means of an adjacency matrix.

[0138] Specifically, the graph convolution (GCN) is a neural network model specially used to process graph structure data, which is used to extract the features of nodes in the graph and consider the connection relationship between nodes.

[0139] In an embodiment of the present invention, by performing graph reconstruction and graph feature extraction, a graph neural network can be used to extract the correlation characteristics of electrical data between each battery in a battery pack, thereby realizing block-based monitoring of the battery pack, thereby improving the monitoring resolution and further improving the accuracy of intelligent monitoring.

[0140] The timing convolution module is used to perform self-attention timing convolution on the electrical diagram feature sequence to obtain electrical timing features.

[0141] In detail, the electrical timing characteristics are timing characteristics corresponding to the electrical diagram characteristic sequence, which can reflect the change relationship of the electrical data of the battery pack over time in the past time period, and can effectively improve the monitoring efficiency of the battery pack.

[0142] In the embodiment of the present invention, when the time series convolution module performs self-attention time series convolution on the electrical diagram feature sequence to obtain the electrical time series feature, it includes:

[0143] Performing short-term recursive convolution on the electrical diagram feature sequence to obtain short-term electrical features;

[0144] Performing long-term jump convolution on the electrical graph feature sequence to obtain long-term electrical features;

[0145] Attention encoding is performed on the electrical graph feature sequence according to a self-attention mechanism to obtain an attention electrical feature;

[0146] Feature fusion is performed on the short-term electrical feature, the long-term electrical feature, and the attention electrical feature.

[0147] Specifically, the temporal recursive convolution and the temporal jump convolution can be performed using the gate structure of a pre-trained long short term memory (LSTM) neural network, wherein the gate structure includes a forget gate, an input gate, and an output gate. The long short term memory neural network is a special recursive neural network specially designed to process and predict long dependencies in time series data.

[0148] In detail, the electrical timing features are calculated using the following attention connection algorithm: ;

[0149] in, is the characteristic value of the electrical timing characteristic at time step t, is the characteristic weight of the short-term electrical characteristic, is the characteristic value of the short-term electrical feature at the tth time step, r, t is the time step index, y is the number of hidden units skipped by the long-term electrical feature when performing long-term skip convolution, is the characteristic weight of the long-term electrical characteristic at the rth time step, The long-term electrical characteristics are described in The eigenvalues ​​of the time steps, is the characteristic value of the attention electrical feature at the tth time step, and b is the preset bias coefficient.

[0150] Specifically, the attention connection algorithm can simultaneously capture short-term fluctuations in electrical data, long-term trends in electrical data, and global attention correlations, and can dynamically adjust the contribution of different features in the final result according to the characteristics of the data, thereby improving the accuracy of anomaly detection.

[0151] In the embodiment of the present invention, by performing self-attention temporal convolution, the graph neural network and the temporal neural network can be combined to simultaneously capture the spatial dependencies in the graph structure and the dynamic changes in the time series, thereby improving the comprehensiveness of battery pack monitoring and the accuracy of abnormality analysis.

[0152] The abnormality monitoring module is used to perform variational abnormality monitoring and decision tree abnormality monitoring on the electrical timing characteristics to obtain battery pack monitoring results.

[0153] In detail, the battery pack monitoring result refers to the result of abnormal monitoring of the battery pack in the past time period. The battery pack monitoring result may be charging and discharging abnormalities, capacity attenuation abnormalities, or an unbalanced state inside the battery pack of a battery in the battery pack.

[0154] In the embodiment of the present invention, when the abnormality monitoring module performs variational abnormality monitoring and decision tree abnormality monitoring on the electrical timing characteristics to obtain the battery pack monitoring result, it includes:

[0155] Performing variational anomaly monitoring on the electrical timing characteristics to obtain primary monitoring results;

[0156] Performing decision tree abnormality monitoring on the electrical timing characteristics to obtain secondary monitoring results;

[0157] A battery pack monitoring result is generated according to the primary monitoring result and the secondary monitoring result.

[0158] In detail, the primary monitoring result and the secondary monitoring result are monitoring results obtained by performing an abnormality analysis on the electrical timing characteristics, including the abnormalities of the battery pack corresponding to the electrical timing characteristics at various times in the past time period.

[0159] Specifically, when the abnormality monitoring module performs variational abnormality monitoring on the electrical timing characteristics to obtain a primary monitoring result, it includes:

[0160] Performing latent space mapping on the electrical timing characteristics to obtain electrical latent distribution;

[0161] Reparameterizing the electrical potential distribution to obtain potential electrical variables;

[0162] Performing latent space inverse mapping on the latent electrical variables to obtain electrical reconstruction time series features;

[0163] Performing reconstruction error calculation on the electrical reconstruction timing characteristics and the electrical timing characteristics to obtain an electrical error;

[0164] Performing time series convolution on the electrical error to obtain a time series feature of the error;

[0165] Anomaly mapping is performed on the error time series characteristics to obtain primary monitoring results.

[0166] In detail, the latent space mapping refers to compressing the electrical timing features into a low-dimensional latent space using an encoder composed of multiple convolutional layers, recurrent neural networks, or long-short time series neural networks. The electrical latent distribution is the normal distribution of the electrical timing features in the latent space. Since the electrical latent distribution is not a definite value, this will make the distribution sampling process random, and thus the gradient cannot be directly back-propagated from the loss function back to the encoder. The reparameterization operation converts the sampling process into a differentiable form, thereby enabling back-propagation.

[0167] Specifically, the latent space inverse mapping method is the inverse step of the latent space mapping method, and the reconstruction error can be calculated using the cross entropy loss function or the Euclidean distance algorithm, that is, the distance between the electrical reconstruction timing feature and the electrical timing feature at each time index is calculated.

[0168] In detail, a long-short time series neural network or a recurrent neural network can be used for time series convolution, and the time series convolution can extract the time series characteristics of the electrical error, so as to analyze the abnormality of the battery pack in terms of trend. The method of abnormal mapping can be to perform abnormal activation based on the activation layer of the trained long-short time series neural network or recurrent neural network, or to perform abnormal analysis based on the numerical size of the error time series characteristics combined with a threshold judgment method.

[0169] In detail, when the abnormality monitoring module performs decision tree abnormality monitoring on the electrical timing characteristics to obtain a secondary monitoring result, it includes:

[0170] Performing feature space splitting on the electrical timing features to obtain an electrical timing space feature group;

[0171] Generate an abnormal decision tree group according to the electrical time series spatial feature group;

[0172] Using the abnormal decision tree group to perform leaf node matching on the electrical timing characteristics to obtain an abnormal leaf node group;

[0173] Performing outlier matching on the abnormal leaf node group to obtain an outlier group;

[0174] An abnormal decision is made on the abnormal value group to obtain a secondary monitoring result.

[0175] Specifically, the feature space splitting refers to dividing the feature space into multiple feature space regions according to the size of the eigenvalue of the electrical timing feature, and making the electrical splitting point set corresponding to each feature space region have the smallest mean square error, wherein the gradient iteration method based on the gradient algorithm can be used to select the optimal features and segmentation points for feature space splitting.

[0176] In detail, the boosting method can be used to generate an abnormal decision tree group according to the electrical timing space feature group. The leaf node matching refers to using each abnormal decision tree in the abnormal decision tree group to make an abnormal decision on the electrical timing feature, and taking the leaf node of the corresponding path as the abnormal leaf node.

[0177] Specifically, the outlier matching refers to aggregating the predicted values ​​corresponding to each abnormal leaf node in the abnormal leaf node group as outliers into an outlier group. The predicted values ​​can be predicted values ​​calculated using a pre-trained XGBoost model. The abnormal decision refers to weighted summing of the results of the outlier group to obtain a standard outlier, and performing an abnormal decision of the decision tree based on the standard outlier to obtain a secondary monitoring result.

[0178] In detail, generating the battery pack monitoring result according to the primary monitoring result and the secondary monitoring result means taking the fusion result of the primary monitoring result and the secondary monitoring result as the battery pack monitoring result.

[0179] In the embodiment of the present invention, by performing variational anomaly monitoring and decision tree anomaly monitoring, anomaly analysis can be performed in combination with an unsupervised model and a supervised model, thereby improving the comprehensiveness of anomaly analysis and the accuracy of intelligent monitoring.

[0180] Reference Figure 4 FIG. 1 is a flow chart of a method for intelligently monitoring a battery pack based on deep learning according to an embodiment of the present invention. In this embodiment, the method for intelligently monitoring a battery pack based on deep learning includes:

[0181] S1. Obtain a historical electrical data set group of a battery group, perform data standardization and outlier denoising operations on the historical electrical data set group, and obtain an electrical data set sequence group.

[0182] S2, extracting electrical features and fusing correlation features of the electrical data set sequence group to obtain a fused electrical feature sequence group, wherein the extracting electrical features and fusing correlation features of the electrical data set sequence group to obtain a fused electrical feature sequence group includes: extracting electrical features of the electrical data set sequence group to obtain an electrical feature set sequence group; regrouping the electrical feature set sequence group according to a preset data type to obtain a single-type electrical feature set; performing feature standardization on the single-type electrical feature set to obtain a standard electrical feature set; and calculating the electrical correlation between each standard electrical feature class in the standard electrical feature set using the following electrical correlation algorithm: ;

[0183] in, Refers to the standard electrical characteristics class set of standard electrical characteristics class and standard electrical characteristics electrical correlation, i and j are the serial numbers of the standard electrical characteristic classes in the standard electrical characteristic class set, The standard electrical characteristics are The total number of features, The standard electrical characteristics are The total number of features, k, is the feature number, is the covariance operator symbol, is the standard deviation operator, Refers to the standard electrical characteristics class The kth feature in Refers to the standard electrical characteristics class The features; and performing correlation feature fusion on the electrical feature set sequence group according to all electrical correlations of the standard electrical feature set to obtain a fused electrical feature sequence group.

[0184] S3. Obtain the battery pack structure of the battery pack, and perform graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence.

[0185] S4. Perform self-attention temporal convolution on the electrical graph feature sequence to obtain electrical temporal features.

[0186] S5. Perform variational abnormality monitoring and abnormality identification on the electrical timing characteristics to obtain a battery pack monitoring result.

[0187] The embodiment of the present invention obtains an electrical data set sequence group by performing data standardization and outlier denoising operations, which can effectively remove noise data in the historical electrical data set group, thereby improving the representation of subsequent electrical time series features and the accuracy of electrical feature extraction. By performing electrical feature extraction and correlation feature fusion, the electrical data of different modes in a single battery can be fused into an overall fused electrical feature, thereby improving the comprehensiveness of battery monitoring and facilitating subsequent intelligent monitoring of the relationship between batteries. By performing graph reconstruction and graph feature extraction, the graph neural network can be used to extract the correlation features of the electrical data between each battery in the battery group, thereby realizing block-based monitoring of the battery group, thereby improving the monitoring resolution and further improving the accuracy of intelligent monitoring.

[0188] By performing self-attention temporal convolution, it is possible to combine graph neural networks and temporal neural networks to simultaneously capture the spatial dependencies in the graph structure and the dynamic changes in the time series, thereby improving the comprehensiveness of battery pack monitoring and the accuracy of anomaly analysis. By performing variational anomaly monitoring and decision tree anomaly monitoring, it is possible to combine unsupervised models and supervised models for anomaly analysis, thereby improving the comprehensiveness of anomaly analysis and the accuracy of intelligent monitoring, thereby improving the accuracy of battery pack monitoring.

[0189] In detail, the battery pack intelligent monitoring method based on deep learning in the embodiment of the present invention is used in the same manner as above. Figure 1 The same technical means are used as the deep learning-based battery pack intelligent monitoring system described in the text, and can produce the same technical effects, so I will not go into details here.

[0190] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0191] Acquire a historical electrical data set group of the battery group, perform data standardization and outlier denoising operations on the historical electrical data set group, and obtain an electrical data set sequence group;

[0192] The electrical feature extraction and correlation feature fusion are performed on the electrical data set sequence group to obtain a fused electrical feature sequence group, wherein the electrical feature extraction and correlation feature fusion are performed on the electrical data set sequence group to obtain a fused electrical feature sequence group, including: extracting electrical features from the electrical data set sequence group to obtain an electrical feature set sequence group; regrouping the electrical feature set sequence group according to a preset data type to obtain a single-type electrical feature set; performing feature standardization operation on the single-type electrical feature set to obtain a standard electrical feature set; and calculating the electrical correlation between each standard electrical feature class in the standard electrical feature set using the following electrical correlation algorithm:

[0193] ;

[0194] in, Refers to the standard electrical characteristics class set of standard electrical characteristics class and standard electrical characteristics electrical correlation, i and j are the serial numbers of the standard electrical characteristic classes in the standard electrical characteristic class set, The standard electrical characteristics are The total number of features, The standard electrical characteristics are The total number of features, k, is the feature number, is the covariance operator symbol, is the standard deviation operator, Refers to the standard electrical characteristics class The kth feature in Refers to the standard electrical characteristics class The features; performing correlation feature fusion on the electrical feature set sequence group according to all electrical correlations of the standard electrical feature set to obtain a fused electrical feature sequence group;

[0195] Acquire the battery pack structure of the battery pack, and perform graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence;

[0196] Performing self-attention time series convolution on the electrical graph feature sequence to obtain electrical time series features;

[0197] Variational abnormality monitoring and abnormality identification are performed on the electrical timing characteristics to obtain a battery pack monitoring result.

[0198] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and systems can be implemented in other ways. For example, the system embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0199] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0200] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0201] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0202] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0203] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is a theory, system, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0204] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or systems stated in the system embodiment can also be implemented by one unit or system through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any specific order.

[0205] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A battery pack intelligent monitoring system based on deep learning, characterized in that: The system includes a data cleaning module, a feature extraction module, a graph reconstruction module, a time series convolution module and an anomaly monitoring module, wherein: The data cleaning module is used to obtain a historical electrical data set group of the battery pack, perform data standardization and outlier denoising operations on the historical electrical data set group, and obtain an electrical data set sequence group. When the data cleaning module performs data standardization and outlier denoising operations on the historical electrical data set group to obtain an electrical data set sequence group, it includes: performing a data format unification operation on the historical electrical data set group to obtain a standard electrical data set group; performing data deduplication on the standard electrical data set group to obtain a deduplicated electrical data set group; performing outlier screening on the deduplicated electrical data set group to obtain a denoised electrical data set group; sorting the data in the denoised electrical data set group in chronological order to obtain A denoised electrical data set sequence group; performing Gaussian interpolation on the denoised electrical data set sequence group to obtain an electrical data set sequence group, wherein the data cleaning module, when performing outlier screening on the deduplicated electrical data set group to obtain the denoised electrical data set group, comprises: regrouping the deduplicated electrical data set group according to a preset data type to obtain a single-type electrical data set set; selecting the single-type electrical data sets in the single-type electrical data set set one by one as target single-type electrical data sets, and selecting the data in the target single-type electrical data set one by one as target electrical data; and calculating the outlier contour distance of the target electrical data in the target single-type electrical data set using the following outlier contour distance algorithm: ; Wherein, L refers to the outlier contour distance, p refers to the target electrical data, N is the total number of data in the target single-type electrical data group, n is the data index, refers to the nth data in the target single-type electrical data group, T is the transposition symbol, is the sign of the covariance function; filtering outlier electrical data sets from the single-type electrical data set according to the outlier contour distance of each target electrical data in the single-type electrical data set; performing an outlier data screening operation on the deduplicated electrical data set group using the outlier electrical data set to obtain a denoised electrical data set group; The feature extraction module is used to perform electrical feature extraction and correlation feature fusion on the electrical data set sequence group to obtain a fused electrical feature sequence group. When the feature extraction module performs electrical feature extraction and correlation feature fusion on the electrical data set sequence group to obtain a fused electrical feature sequence group, the module includes: performing electrical feature extraction on the electrical data set sequence group to obtain an electrical feature set sequence group; regrouping the electrical feature set sequence group according to a preset data type to obtain a single-type electrical feature set; performing feature standardization on the single-type electrical feature set to obtain a standard electrical feature set; and calculating the electrical correlation between each standard electrical feature class in the standard electrical feature set using the following electrical correlation algorithm: ; in, Refers to the standard electrical characteristics class set of standard electrical characteristics class and standard electrical characteristics electrical correlation, i and j are the serial numbers of the standard electrical characteristic classes in the standard electrical characteristic class set, The standard electrical characteristics are The total number of features, The standard electrical characteristics are The total number of features, k and l are the feature numbers, is the covariance operator symbol, is the standard deviation operator, Refers to the standard electrical characteristics class The kth feature in Refers to the standard electrical characteristics class The first feature in the standard electrical feature set; performing correlation feature fusion on the electrical feature set sequence group according to all electrical correlations of the standard electrical feature set to obtain a fused electrical feature sequence group; The graph reconstruction module is used to obtain the battery pack structure of the battery pack, and perform graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence; The timing convolution module is used to perform self-attention timing convolution on the electrical diagram feature sequence to obtain electrical timing features; The abnormality monitoring module is used to perform variational abnormality monitoring and abnormality identification on the electrical timing characteristics to obtain battery pack monitoring results.

2. The deep learning-based intelligent monitoring system for battery packs according to claim 1, characterized in that: When the feature extraction module performs correlation feature fusion on the electrical feature set sequence group according to all electrical correlations of the standard electrical feature set to obtain a fused electrical feature sequence group, it includes: generating an electrical correlation matrix according to all electrical correlations of the standard electrical feature set; Calculating a weight matrix of the standard electrical feature set according to the electrical correlation matrix; Performing a matrix row sum operation on the weight matrix to obtain a row feature weight group; Normalizing each row feature weight in the row feature weight group to obtain a feature weight group; A feature weighted fusion operation is performed on each electrical feature set in the electrical feature set sequence group according to the feature weight group to obtain a fused electrical feature sequence group.

3. The deep learning-based intelligent monitoring system for battery packs according to claim 1, characterized in that: When the graph reconstruction module performs graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence, the graph reconstruction module includes: extracting a battery node set from the battery pack structure; Performing edge mapping on the battery node set using the battery group structure to obtain a battery group graph; Selecting electrical feature sequences in the electrical feature sequence group one by one as target electrical feature sequences, and taking battery nodes corresponding to the target electrical feature sequences in the battery node set as target battery nodes; Performing node feature mapping on the target battery node in the battery group diagram using the target electrical feature sequence to obtain a primary diagram feature sequence, until the target electrical feature sequence is the last electrical feature sequence in the electrical feature sequence group, and using the primary diagram feature sequence as a reconstructed diagram feature sequence; A graph convolution operation is performed on each reconstructed graph feature in the reconstructed graph feature sequence to obtain an electrical graph feature sequence.

4. The deep learning-based intelligent monitoring system for battery packs according to claim 1, characterized in that: When the timing convolution module performs self-attention timing convolution on the electrical diagram feature sequence to obtain electrical timing features, it includes: Performing short-term recursive convolution on the electrical diagram feature sequence to obtain short-term electrical features; Performing long-term jump convolution on the electrical graph feature sequence to obtain long-term electrical features; Attention encoding is performed on the electrical graph feature sequence according to a self-attention mechanism to obtain an attention electrical feature; The short-term electrical feature, the long-term electrical feature and the attention electrical feature are fused using the following attention connection algorithm to obtain an electrical timing feature: ; in, is the characteristic value of the electrical timing characteristic at time step t, is the characteristic weight of the short-term electrical characteristic, is the characteristic value of the short-term electrical feature at the tth time step, r, t is the time step index, y is the number of hidden units skipped by the long-term electrical feature when performing long-term skip convolution, is the characteristic weight of the long-term electrical characteristic at the rth time step, The long-term electrical characteristics are described in The eigenvalues ​​of the time steps, is the characteristic value of the attention electrical feature at the tth time step, and b is the preset bias coefficient.

5. The deep learning-based intelligent monitoring system for battery packs according to claim 1, characterized in that: When the abnormality monitoring module performs variational abnormality monitoring and decision tree abnormality monitoring on the electrical timing characteristics to obtain the battery pack monitoring result, it includes: Performing variational anomaly monitoring on the electrical timing characteristics to obtain primary monitoring results; Performing decision tree abnormality monitoring on the electrical timing characteristics to obtain secondary monitoring results; A battery pack monitoring result is generated according to the primary monitoring result and the secondary monitoring result.

6. The deep learning-based intelligent monitoring system for battery packs according to claim 5, characterized in that: When the abnormality monitoring module performs variational abnormality monitoring on the electrical timing characteristics to obtain a primary monitoring result, it includes: Performing latent space mapping on the electrical timing characteristics to obtain electrical latent distribution; Reparameterizing the electrical potential distribution to obtain potential electrical variables; Performing latent space inverse mapping on the latent electrical variables to obtain electrical reconstruction time series features; Performing reconstruction error calculation on the electrical reconstruction timing characteristics and the electrical timing characteristics to obtain an electrical error; Performing time series convolution on the electrical error to obtain a time series feature of the error; Anomaly mapping is performed on the error time series characteristics to obtain primary monitoring results.

7. The deep learning-based intelligent monitoring system for battery packs according to claim 5, characterized in that: When the abnormality monitoring module performs decision tree abnormality monitoring on the electrical timing characteristics to obtain a secondary monitoring result, it includes: Performing feature space splitting on the electrical timing features to obtain an electrical timing space feature group; Generate an abnormal decision tree group according to the electrical time series spatial feature group; Using the abnormal decision tree group to perform leaf node matching on the electrical timing characteristics to obtain an abnormal leaf node group; Performing outlier matching on the abnormal leaf node group to obtain an outlier group; An abnormal decision is made on the abnormal value group to obtain a secondary monitoring result.

8. A battery pack intelligent monitoring method based on deep learning, characterized in that: The method comprises: Acquire a historical electrical data set group of a storage battery group, perform data standardization and outlier denoising operations on the historical electrical data set group, and obtain an electrical data set sequence group, wherein the data standardization and outlier denoising operations on the historical electrical data set group to obtain an electrical data set sequence group include: performing a data format unification operation on the historical electrical data set group to obtain a standard electrical data set group; performing data deduplication on the standard electrical data set group to obtain a deduplicated electrical data set group; performing outlier screening on the deduplicated electrical data set group to obtain a de-noised electrical data set group; sorting the data in the de-noised electrical data set group in chronological order to obtain a de-noised electrical data set group. The method comprises the following steps: performing Gaussian interpolation on the denoised electrical data set sequence group to obtain an electrical data set sequence group, wherein the outlier screening on the deduplicated electrical data set group to obtain a denoised electrical data set group comprises: regrouping the deduplicated electrical data set group according to a preset data type to obtain a single-type electrical data set set; selecting the single-type electrical data sets in the single-type electrical data set set one by one as target single-type electrical data sets, and selecting the data in the target single-type electrical data set one by one as target electrical data; and calculating the outlier contour distance of the target electrical data in the target single-type electrical data set using the following outlier contour distance algorithm: ; Wherein, L refers to the outlier contour distance, p refers to the target electrical data, N is the total number of data in the target single-type electrical data group, n is the data index, refers to the nth data in the target single-type electrical data group, T is the transposition symbol, is the sign of the covariance function; filtering outlier electrical data sets from the single-type electrical data set according to the outlier contour distance of each target electrical data in the single-type electrical data set; performing an outlier data screening operation on the deduplicated electrical data set group using the outlier electrical data set to obtain a denoised electrical data set group; The electrical feature extraction and correlation feature fusion are performed on the electrical data set sequence group to obtain a fused electrical feature sequence group, wherein the electrical feature extraction and correlation feature fusion are performed on the electrical data set sequence group to obtain a fused electrical feature sequence group, including: extracting electrical features from the electrical data set sequence group to obtain an electrical feature set sequence group; regrouping the electrical feature set sequence group according to a preset data type to obtain a single-type electrical feature set; performing feature standardization operation on the single-type electrical feature set to obtain a standard electrical feature set; and calculating the electrical correlation between each standard electrical feature class in the standard electrical feature set using the following electrical correlation algorithm: ; in, Refers to the standard electrical characteristics class set of standard electrical characteristics class and standard electrical characteristics electrical correlation, i and j are the serial numbers of the standard electrical characteristic classes in the standard electrical characteristic class set, The standard electrical characteristics are The total number of features, The standard electrical characteristics are The total number of features, k and l are the feature numbers, is the covariance operator symbol, is the standard deviation operator, Refers to the standard electrical characteristics class The kth feature in Refers to the standard electrical characteristics class The first feature in the standard electrical feature set; performing correlation feature fusion on the electrical feature set sequence group according to all electrical correlations of the standard electrical feature set to obtain a fused electrical feature sequence group; Acquire the battery pack structure of the battery pack, and perform graph reconstruction and graph feature extraction on the fused electrical feature sequence group according to the battery pack structure to obtain an electrical graph feature sequence; Performing self-attention time series convolution on the electrical graph feature sequence to obtain electrical time series features; Variational abnormality monitoring and abnormality identification are performed on the electrical timing characteristics to obtain a battery pack monitoring result.

Images