Storage battery fault detection method and system based on multi-modal characteristics

Abstract

The invention discloses a transformer substation storage battery pack abnormal state identification method based on multi-modal characteristics. The method comprises the following steps of obtaining current, temperature, humidity and vibration frequency of a storage battery in historical data; formalizing a fault prediction task and constructing a fault detection model, and importing preprocessed data into a feature extraction module to obtain local features; importing the local features into a time sequence feature extraction module to obtain time sequence features; time sequence features are imported into a global feature extraction module; global features are obtained; importing the local features, the time sequence features and the global features to a feature fusion module to obtain multi-modal features; and importing the multi-modal features to a prediction module to obtain a predicted fault type. According to the method, the local features, the time sequence features and the global features are extracted in sequence by using a plurality of detection indexes of the transformer substation storage battery pack, and the features are subjected to multi-modal data fusion, so that the capture capability of the model on data exception is improved, and the recognition accuracy of the abnormal state of the transformer substation storage battery pack is improved.

Description

Technical Field

[0001] The invention belongs to the field of storage batteries, and in particular relates to a storage battery fault detection method based on multi-modal features. Background Art

[0002] The DC operating power supply system is one of the indispensable secondary equipment in power plants and substations. As the core equipment of the DC operating power supply system, the battery provides protection power in time when the station uses AC failure to maintain the reliable operation of the protection equipment of the power station and the substation. The high reliability, long cycle life and durability of the battery have become the focus of research and development. Based on the high reliability of the battery, this project develops a small, light, portable battery with simple wiring and convenient installation. In addition, the battery is in a floating charge state during long-term operation, and it is difficult to judge the performance of the battery, such as reduced capacity, increased internal resistance, and virtual connection of the battery pole. At present, the only way to monitor the performance of the battery is to perform a qualitative discharge test on the battery. Generally, it is carried out once every 1 to 2 years according to the requirements of relevant regulations and standards. This method is not real-time enough and cannot detect potential faults in time. On-site maintenance is also required, with low overall efficiency, high cost and large workload. In the prior art, there is an online monitoring system embedded in a portable power supply, which can monitor the voltage, resistance and other parameters of the battery, control the charging and discharging of the battery pack and the battery status estimation, so as to achieve high-precision monitoring and control of the high-voltage battery pack.

[0003] The Chinese patent with patent publication number CN107150602B provides a BMS battery management system for detecting the battery system, but there are technical problems such as failure to fully utilize the detection data and to effectively predict the battery failure in real time. Summary of the invention

[0004] In view of the technical problems existing in the tree trunk whitening machine in the prior art, a method for evaluating the abnormal state of a substation battery pack based on multimodal features is provided, and the method comprises the following steps:

[0005] Step S1: Obtain historical operating data of the voltage, current, temperature, humidity and vibration frequency of the battery pack through the battery management system, and intercept and quantify the data according to the actual abnormal state type and its severity;

[0006] Step S2: pre-process the data by using abnormal filtering, format correction, data splicing and data standardization to obtain pre-processed data in array format, enhance the data with abnormal state by downsampling, flipping and twisting, and construct a training sample data set;

[0007] Step S3: Formalize the abnormal state assessment task and construct an intelligent assessment model based on a deep learning network structure, wherein the network structure includes a local feature extraction module, a temporal feature extraction module, a global feature extraction module, a feature fusion module, and a recognition output module; import the preprocessed data into the feature extraction module, the temporal feature extraction module, and the global feature extraction module in sequence, obtain local features, temporal features, and global features respectively, and perform multi-scale fusion on the extracted feature data to obtain multi-modal features, perform intelligent recognition of the multi-modal features through the recognition output module, and obtain evaluation indicators for each abnormal state type;

[0008] Step S4: Use quantile loss to construct a model loss function, randomly select 30 groups of training sample set data in batches and import them into the intelligent evaluation model, and use the Adam optimization algorithm to train the intelligent evaluation model described in S3 with the goal of minimizing the loss function to obtain the intelligent evaluation model after parameter optimization;

[0009] Step S5: Acquire the voltage, current, temperature, humidity and vibration frequency data of the on-site battery pack, import the optimized intelligent evaluation model after the preprocessing described in S2, and output the prediction result of the battery pack state type.

[0010] Furthermore, the abnormal conditions are battery leakage, internal short circuit, overcharge, undercharge, installation failure, abnormal ambient temperature and abnormal ambient humidity, and include combinations of different abnormal conditions, and each abnormal condition indicator is divided into three levels: mild, severe and critical.

[0011] Furthermore, it is characterized in that the formalized prediction task in S3 is specifically:

[0012] Given a set of preprocessed data X = {x1, x2, x3, x4, x5}, where x1, x2, x3, x4, x5 represent the voltage, current, temperature, humidity and vibration frequency of the battery pack, the corresponding abnormal state type and severity are diagnosed by the intelligent evaluation model, which is expressed as:

[0013] (y1,y2)=f(X);

[0014] Wherein, y1 represents the first state with the highest severity, y2 represents the second state with the second lowest severity, and f(·) represents the abnormal state assessment model.

[0015] Furthermore, the local feature extraction module in S3 is a combination of a ResNeSt block and a Feature refinement block.

[0016] Furthermore, the temporal feature extraction module in S3 first performs a maximum pooling operation on the output F1 of the local feature extraction module to obtain x t, and then extract the temporal dependency features through the long short-term memory network, specifically:

[0017] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0018] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0019]

[0020] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0021] h t =o t *tanh(C t );

[0022] in, represents the candidate cell state, C t Indicates updating the cell state, h t represents the hidden state, W f , W i , W C and W o Both represent the preset weight matrix, b f , b i , b C and b o All represent the preset weight parameters, σ(·) represents the SigMoid activation function, tanh(·) represents the tanh activation function, and h t represents the hidden state, that is, the time series feature F2, h t-1 Represents the hidden state at the previous moment.

[0023] Furthermore, the global feature extraction module in S3 processes the temporal features through the multi-head self-attention mechanism as follows:

[0024] First, the input time series feature F2 is transformed into the query matrix Q, key matrix K and value matrix V through three different linear transformations; then, the scaled dot product attention is used to calculate the attention weight of each head; the above process is processed in parallel to obtain the global feature F3:

[0025] Q=WQ F2,K=W K F2,V=W V F2;

[0026]

[0027] MultiHead(Q,K,V)=Concat(head1,head2,...,head h )W O ;

[0028] F3=MultiHead(Q,K,V);

[0029] Among them, W Q ,W K ,W V is a learnable weight matrix; d k is the dimension of the key vector; and W O is a learnable weight matrix.

[0030] Furthermore, the multimodal feature F is obtained in S3 y The specific operations are:

[0031] F y =Concat(F1,F2,F3);

[0032] Furthermore, the recognition output module described in S3 is a fully connected layer that uses a probability density function to calculate the probability of each abnormal state type, sorts the corresponding abnormal state types according to their sizes, and uses the first two types and their probabilities as outputs.

[0033] The positive and progressive effects of the present invention are:

[0034] 1) The present invention improves the ability to capture detection data and the generalization ability of the model by extracting multiple detection indicators of the battery module and extracting local features, time series features and global features in sequence, and can perform fault type analysis on multiple scales.

[0035] 2) The present invention makes full use of historical detection data and can timely obtain two fault types predicted by probability, thereby simplifying the procedures for battery maintenance and inspection and improving efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The present invention is a flowchart of the steps of a method for identifying abnormal status of a substation battery pack based on multimodal features. DETAILED DESCRIPTION

[0037] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.

[0038] Reference Figure 1 , a method for evaluating abnormal state of a substation battery pack based on multimodal features, in one example, the method comprises the following steps:

[0039] Step S1: Obtain historical operating data of the voltage, current, temperature, humidity and vibration frequency of the battery pack through the battery management system, and intercept and quantify the data according to the actual abnormal state type and its severity;

[0040] Step S2: pre-process the data by using abnormal filtering, format correction, data splicing and data standardization to obtain pre-processed data in array format, enhance the data with abnormal state by downsampling, flipping and twisting, and construct a training sample data set;

[0041] Step S3: Formalize the abnormal state assessment task and construct an intelligent assessment model based on a deep learning network structure, wherein the network structure includes a local feature extraction module, a temporal feature extraction module, a global feature extraction module, a feature fusion module, and a recognition output module; import the preprocessed data into the feature extraction module, the temporal feature extraction module, and the global feature extraction module in sequence, obtain local features, temporal features, and global features respectively, and perform multi-scale fusion on the extracted feature data to obtain multi-modal features, perform intelligent recognition of the multi-modal features through the recognition output module, and obtain evaluation indicators for each abnormal state type;

[0042] Step S4: Use quantile loss to construct a model loss function, randomly select 30 groups of training sample set data in batches and import them into the intelligent evaluation model, and use the Adam optimization algorithm to train the intelligent evaluation model described in S3 with the goal of minimizing the loss function to obtain the intelligent evaluation model after parameter optimization;

[0043] Step S5: Acquire the voltage, current, temperature, humidity and vibration frequency data of the on-site battery pack, import the optimized intelligent evaluation model after the preprocessing described in S2, and output the prediction result of the battery pack state type.

[0044] Furthermore, in one example, the abnormal state is battery leakage, internal short circuit, overcharge, undercharge, installation failure, abnormal ambient temperature and abnormal ambient humidity, and includes combinations of different abnormal states, and each abnormal state indicator is divided into three levels: mild, severe and critical.

[0045] Furthermore, in one example, it is characterized in that the formalized prediction task in S3 is specifically:

[0046] Given a set of preprocessed data X = {x1, x2, x3, x4, x5}, where x1, x2, x3, x4, x5 represent the voltage, current, temperature, humidity and vibration frequency of the battery pack, the corresponding abnormal state type and severity are diagnosed by the intelligent evaluation model, which is expressed as:

[0047] (y1,y2)=f(X);

[0048] Wherein, y1 represents the first state with the highest severity, y2 represents the second state with the second lowest severity, and f(·) represents the abnormal state assessment model.

[0049] Furthermore, in one example, the local feature extraction module in S3 is a combination of a ResNeSt block and a Featurerefinement block.

[0050] Further, in one example, the temporal feature extraction module in S3 first performs a maximum pooling operation on the output F1 of the local feature extraction module to obtain x t , and then extract the temporal dependency features through the long short-term memory network, specifically:

[0051] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0052] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0053]

[0054] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0055] h t =o t *tanh(C t );

[0056] in, represents the candidate cell state, C tIndicates updating the cell state, h t represents the hidden state, W f , W i , W C and W o Both represent the preset weight matrix, b f 、b i 、b C and b o All represent the preset weight parameters, σ(·) represents the SigMoid activation function, tanh(·) represents the tanh activation function, and h t represents the hidden state, that is, the time series feature F2, h t-1 Indicates the hidden state at the previous moment.

[0057] Furthermore, in one example, the global feature extraction module in S3 processes the temporal features through the multi-head self-attention mechanism as follows:

[0058] First, the input time series feature F2 is transformed into the query matrix Q, key matrix K and value matrix V through three different linear transformations; then, the scaled dot product attention is used to calculate the attention weight of each head; the above process is processed in parallel to obtain the global feature F3:

[0059] Q=W Q F2,K=W K F2,V=W V F2;

[0060]

[0061] MultiHead(Q,K,V)=Concat(head1,head2,·..,head h )W O ;

[0062] F3=MultiHead(Q,K,V);

[0063] Among them, W Q ,W K ,W V is a learnable weight matrix; d k is the dimension of the key vector; and W O is a learnable weight matrix.

[0064] Furthermore, in one example, the multimodal feature F is obtained in S3 y The specific operations are:

[0065] F y =Concat(F1,F2,F3);

[0066] Furthermore, in one example, the recognition output module in S3 is a fully connected layer that uses a probability density function to calculate the probability of each abnormal state type, sorts the corresponding abnormal state types according to their sizes, and uses the first two types and their probabilities as outputs.

[0067] The present invention is described in detail above in conjunction with the embodiments of the accompanying drawings. A person skilled in the art can make various variations of the present invention according to the above description. Therefore, some details in the embodiments should not be construed as limiting the present invention, and the present invention shall be protected by the scope defined by the attached claims.

Claims

1. A method for evaluating abnormal state of a substation battery pack based on multimodal features, characterized in that: The method comprises the following steps: Step S1: Obtain historical operating data of the voltage, current, temperature, humidity and vibration frequency of the battery pack through the battery management system, and intercept and quantify the data according to the actual abnormal state type and its severity; Step S2: pre-process the data by using abnormal filtering, format correction, data splicing and data standardization to obtain pre-processed data in array format, enhance the data with abnormal state by downsampling, flipping and twisting, and construct a training sample data set; Step S3: Formalize the abnormal state assessment task and construct an intelligent assessment model based on a deep learning network structure, wherein the network structure includes a local feature extraction module, a temporal feature extraction module, a global feature extraction module, a feature fusion module, and a recognition output module; import the preprocessed data into the feature extraction module, the temporal feature extraction module, and the global feature extraction module in sequence, obtain local features, temporal features, and global features respectively, and perform multi-scale fusion on the extracted feature data to obtain multi-modal features, perform intelligent recognition of the multi-modal features through the recognition output module, and obtain evaluation indicators for each abnormal state type; Step S4: Use quantile loss to construct a model loss function, randomly select 30 groups of training sample set data in batches and import them into the intelligent evaluation model, and use the Adam optimization algorithm to train the intelligent evaluation model described in S3 with the goal of minimizing the loss function to obtain the intelligent evaluation model after parameter optimization; Step S5: Acquire the voltage, current, temperature, humidity and vibration frequency data of the on-site battery pack, import the optimized intelligent evaluation model after the preprocessing described in S2, and output the prediction result of the battery pack state type.

2. According to claim 1, a method for evaluating abnormal state of a substation battery pack based on multimodal features is characterized in that: The abnormal conditions are battery leakage, internal short circuit, overcharge, undercharge, installation failure, abnormal ambient temperature and abnormal ambient humidity, and include combinations of different abnormal conditions. Each abnormal condition indicator is divided into three levels: mild, severe and critical.

3. According to claim 1, a method for evaluating abnormal state of a substation battery pack based on multimodal features is characterized in that: The formal prediction tasks in S3 are as follows: Given a set of preprocessed data ,in, They represent the voltage, current, temperature, humidity and vibration frequency of the battery pack respectively. The corresponding abnormal state type and severity are diagnosed through the intelligent evaluation model, which are expressed as: ； in, Indicates the first state with the highest severity, Indicates the second, less severe state. Represents the abnormal state assessment model.

4. According to claim 1, a method for evaluating abnormal state of a substation battery pack based on multimodal features is characterized in that: The local feature extraction module described in S3 is a combination of ResNeSt block and Feature refinement block.

5. The method for evaluating abnormal state of a substation battery pack based on multimodal features according to claim 1 is characterized in that: The temporal feature extraction module in S3 first performs a maximum pooling operation on the output F1 of the local feature extraction module to obtain , and then extract the temporal dependency features through the long short-term memory network, specifically: ； ； ； ； ； ； in, represents the candidate cell state, Indicates updating the cell status. Indicates the hidden state, , , and Both represent the preset weight matrix, , , and Both represent preset weight parameters. represents the SigMoid activation function, represents the tanh activation function, Represents the hidden state, that is, the time series feature , Represents the hidden state at the previous moment.

6. The method for evaluating abnormal state of a substation battery pack based on multimodal features according to claim 1, characterized in that: The global feature extraction module in S3 processes the temporal features through the multi-head self-attention mechanism as follows: First, the input time series features The query matrix Q, key matrix K and value matrix V are obtained through three different linear transformations; then, the attention weight of each head is calculated using scaled dot product attention; the above process is processed in parallel to obtain the global feature : ； ； ； MultiHead Concat ; ； in, is a learnable weight matrix; is the dimension of the key vector; and is a learnable weight matrix.

7. The method for evaluating abnormal state of a substation battery pack based on multimodal features according to claim 1, characterized in that: Obtaining multimodal features in S3 The specific operations are: 。 8. The method for evaluating abnormal state of a substation battery pack based on multimodal features according to claim 1, characterized in that: The recognition output module described in S3 is a fully connected layer that uses a probability density function to calculate the probability of each abnormal state type, sorts the corresponding abnormal state types according to their sizes, and outputs the first two types and their probabilities.

Images