Energy storage monitoring method and system based on spatio-temporal feature fusion and active safety early warning

Abstract

The embodiment of the invention provides an energy storage monitoring method and system based on spatio-temporal feature fusion and active safety early warning, and belongs to the technical field of energy storage monitoring. The energy storage monitoring method comprises the following steps: collecting multi-dimensional key data, including electric core level data, module level data and whole station level data, according to the running states of different levels of an energy storage system; performing data compression and synchronization on the obtained multi-dimensional key data; the transmitted multi-dimensional key data is preprocessed; feature extraction and weighting processing are carried out on the preprocessed multi-dimensional key data; an LSTM-CNN network model is constructed, and the weighted multi-dimensional key data and the corresponding fault result are used as training data to train the LSTM-CNN network model; and collecting the current multi-dimensional key data and carrying out preprocessing. According to the energy storage monitoring method, prediction and early warning can be carried out on each data of the energy storage power station based on the obtained multi-dimensional key data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage monitoring, in particular to an energy storage monitoring method and system based on spatiotemporal feature fusion and active safety early warning. BACKGROUND

[0002] Under the background of the deepening of the "double carbon" strategic goal, renewable energy (such as wind power and photovoltaic) continues to expand its grid-connected scale due to its clean and sustainable advantages. However, the output of renewable energy has obvious volatility and intermittency, which brings challenges to the stable operation of the power system. Electrochemical energy storage, with its millisecond-level flexible regulation capability, can effectively smooth out new energy output fluctuations and ensure power supply and demand balance, becoming a key supporting link in building a new power system.

[0003] In recent years, China's energy storage industry has shown explosive growth, with the cumulative installed capacity of new energy storage exceeding 34GW by the end of 2023. The number and scale of energy storage power stations continue to expand. However, at the same time, the problem of intrinsic safety and efficient operation and maintenance of energy storage systems has become increasingly prominent. On the one hand, during the operation of energy storage power stations, a large amount of time-series data is generated, including key parameters such as cell voltage, temperature, and charge and discharge current. Traditional monitoring methods are inefficient in the entire process of data acquisition, transmission, storage, and application, making it difficult to balance data compression rate and real-time requirements. On the other hand, existing monitoring technologies rely on a single parameter (such as temperature threshold) for fault judgment, lacking fusion analysis of spatiotemporal features, resulting in delayed early warning of micro-short circuit, thermal runaway, and other early faults. The sample distribution skew problem caused by the scarcity of fault samples further reduces the model's generalization ability and fault recognition accuracy. Therefore, there is a need for an energy storage monitoring method based on spatiotemporal feature fusion and active safety early warning to timely monitor and warn energy storage safety. SUMMARY

[0004] The purpose of the embodiments of the present application is to provide an energy storage monitoring method and system based on spatiotemporal feature fusion and active safety early warning, which can predict and warn various data of energy storage power stations based on the obtained multi-dimensional key data.

[0005] To achieve the above-mentioned purpose, the embodiments of the present application provide an energy storage monitoring method based on spatiotemporal feature fusion and active safety early warning, which comprises: For the running state of the energy storage system at different levels, multi-dimensional key data is collected, including cell-level data, module-level data, and whole-station-level data; The obtained multi-dimensional key data is compressed and synchronized; The transmitted multi-dimensional key data is preprocessed; The multi-dimensional key data after preprocessing is subjected to feature extraction and weighted processing; constructing an LSTM-CNN network model, and taking the weighted multi-dimensional key data and the corresponding fault result as training data to train the LSTM-CNN network model; collecting the current multi-dimensional key data and preprocessing it, so as to input the preprocessed multi-dimensional key data into the trained LSTM-CNN network model to obtain fault prediction data and fault prediction result; inputting the obtained fault prediction data and corresponding fault prediction result into the energy storage pre-digital twin model to intuitively display the fault prediction data and fault prediction result.

[0006] Optionally, the obtained multi-dimensional key data is compressed and synchronized, including: obtaining data with strong time sequence in the multi-dimensional key data as compressed data, including data of cell voltage and temperature; According to the obtained compressed data, the difference between adjacent data points is calculated to perform primary difference; After the primary difference is completed, the difference of the primary difference result is calculated again to obtain the secondary difference result, so as to compress the data with strong time sequence in the multi-dimensional key data; The compressed data and the uncompressed data are stored synchronously to facilitate subsequent calculation.

[0007] Optionally, the transmitted multi-dimensional key data is preprocessed, including: obtaining the multi-dimensional key data after data compression and synchronization; obtaining data of a category with a sample number less than a preset threshold as to-be-added data: randomly selecting any one data point in the to-be-added data; For the randomly selected data point, obtain the nearest K data points; select any one of the K data points and connect it with the randomly selected data point; Randomly select any one point on the connecting line as a new data point and add it to the to-be-added data to expand the data amount of the to-be-added data; Repeat the above steps until the sample data of the to-be-added data is greater than the preset threshold.

[0008] Optionally, the transmitted multi-dimensional key data is preprocessed, including: obtaining the multi-dimensional key data with increased sample number and not needing to increase sample number; filling the missing values in the multi-dimensional key data by linear interpolation method; by 3 Principle, the multi-dimensional key data after filling is checked for outliers to remove outliers in the multi-dimensional key data; After removing outliers, the multi-dimensional key data is standardized to eliminate dimensional differences in the multi-dimensional key data; The multi-dimensional key data after eliminating dimensional differences is normalized to complete the preprocessing of the multi-dimensional key data.

[0009] Optionally, feature extraction and weighting processing are performed on the preprocessed multi-dimensional key data, including: The preprocessed multi-dimensional key data is obtained, and the relevance of the data of each category in the multi-dimensional key data to the fault is calculated by the Gumbel Copula function to obtain the corresponding correlation coefficient; For the correlation coefficient corresponding to the multi-dimensional key data, the multi-dimensional key data corresponding to the correlation coefficient greater than the preset threshold is given a high weight; For the correlation coefficient corresponding to the multi-dimensional key data, the multi-dimensional key data corresponding to the correlation coefficient less than the preset threshold is given a low weight; The multi-dimensional key data with weights and the corresponding fault results are obtained to prepare for input into the LSTM-CNN network model.

[0010] Optionally, an LSTM-CNN network model is constructed, and the multi-dimensional key data after weighting processing and the corresponding fault results are used as training data to train the LSTM-CNN network model, including: An LSTM prediction network is constructed, and a Dropout layer is added after the LSTM layer to prevent overfitting and improve the generalization ability of the model; An attention mechanism is introduced based on the LSTM output, and the weighted multi-dimensional key data of the category is focused on, so that the network focuses on the time step that has the greatest impact on the fault result; The attention weighted LSTM is output to a one-dimensional convolutional neural network (CNN); The one-dimensional convolutional neural network extracts local time series patterns through a convolutional layer and performs feature compression in a pooling layer; The features after convolution are output through a fully connected layer to obtain the fault prediction data and the corresponding fault prediction result.

[0011] Optionally, an LSTM-CNN network model is constructed, and the multi-dimensional key data after weighting processing and the corresponding fault results are used as training data to train the LSTM-CNN network model, including: obtaining fault prediction data output by the LSTM-CNN network model in the training process and corresponding fault prediction results; constructing a loss function based on the fault prediction data; calculating the gradient in the loss function and determining whether the gradient is 0; in the case where the gradient is 0, outputting the fault prediction data and determining that the LSTM-CNN network model training is completed; in the case where the gradient is not 0, updating the LSTM-CNN network model, inputting the weighted multi-dimensional key data into the LSTM-CNN network model again, and returning to the step of obtaining the fault prediction data output by the LSTM-CNN network model in the training process and the corresponding fault prediction results until the gradient is 0.

[0012] Optionally, the cell-level data includes voltage, temperature, internal resistance and SOC of the cell; the module-level data includes cluster voltage consistency data and charge-discharge current stability data; and the station-level data includes PCS power output data, copper bar temperature data, environmental temperature and humidity data and fire extinguishing system state data.

[0013] In another aspect, the application further provides a storage energy monitoring system based on spatio-temporal feature fusion and active safety warning, which comprises: a data acquisition module, configured to collect multi-dimensional key data including cell-level data, module-level data and station-level data for different levels of operation states of the storage energy system; a storage energy monitoring module, configured to execute the storage energy monitoring method based on spatio-temporal feature fusion and active safety warning according to the data obtained by the data acquisition module, as described in any one of claims 1-8.

[0014] By the technical solution, the energy storage monitoring method and system based on spatio-temporal feature fusion and active safety warning can collect multi-dimensional key data according to the operation states of different levels of the energy storage system, which can include cell-level data, module-level data and whole-station-level data. After the multi-dimensional key data is obtained, the obtained multi-dimensional key data can be compressed and synchronized to facilitate transmission. Then the transmitted multi-dimensional key data can be preprocessed. After the preprocessing is completed, the multi-dimensional key data after preprocessing can be feature extracted and weighted to increase the weight of multi-dimensional key data with great influence. After the weighting, an LSTM-CNN network model can be constructed, and then the multi-dimensional key data after weighting and the corresponding fault result can be used as training data to train the LSTM-CNN network model. After the training is completed, the current multi-dimensional key data can be collected and preprocessed, and then the preprocessed multi-dimensional key data can be input into the trained LSTM-CNN network model, so that fault prediction data and fault prediction results can be obtained. After the fault prediction data and the corresponding fault prediction results are obtained, they can be input into the energy storage digital twin model, so that the fault prediction data and the fault prediction results can be intuitively displayed. The energy storage monitoring method can predict and warn each data of the energy storage power station based on the obtained multi-dimensional key data.

[0015] Other features and advantages of the embodiments of the present application will be described in detail in the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0016] The accompanying drawings are included to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the embodiments of the present application together with the following detailed description, but do not constitute a limitation of the embodiments of the present application. In the drawings: Figure 1 is a flowchart of an energy storage monitoring method based on spatio-temporal feature fusion and active safety warning according to an embodiment of the present application; Figure 2 is a flowchart of data compression and synchronization of an energy storage monitoring method based on spatio-temporal feature fusion and active safety warning according to an embodiment of the present application; Figure 3 is a first flowchart of data preprocessing of an energy storage monitoring method based on spatio-temporal feature fusion and active safety warning according to an embodiment of the present application; Figure 4 is a second flowchart of data preprocessing of an energy storage monitoring method based on spatio-temporal feature fusion and active safety warning according to an embodiment of the present application; Figure 5This is a flowchart of data weighting processing for an energy storage monitoring method based on spatiotemporal feature fusion and proactive safety early warning, according to an embodiment of the present invention. Figure 6 This is a first flowchart of a training model for an energy storage monitoring method based on spatiotemporal feature fusion and proactive safety early warning, according to an embodiment of the present invention. Figure 7 This is a second flowchart of a training model for an energy storage monitoring method based on spatiotemporal feature fusion and proactive safety early warning, according to an embodiment of the present invention. Detailed Implementation

[0017] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0018] In the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0019] Figure 1 This is a flowchart of an energy storage monitoring method based on spatiotemporal feature fusion and proactive safety early warning, according to an embodiment of the present invention. In this invention, the energy storage monitoring process may include: In step S1, multi-dimensional key data are collected for the different operating states of the energy storage system, including cell-level data, module-level data and whole-site-level data.

[0020] In step S2, the acquired multidimensional key data is compressed and synchronized.

[0021] In step S3, the transmitted multidimensional key data is preprocessed.

[0022] In step S4, feature extraction and weighting are performed on the preprocessed multidimensional key data.

[0023] In step S5, an LSTM-CNN network model is constructed, and the weighted multidimensional key data and the corresponding fault results are used as training data to train the LSTM-CNN network model.

[0024] In step S6, the current multidimensional key data is collected and preprocessed, so that the preprocessed multidimensional key data is fed into the trained LSTM-CNN network model to obtain fault prediction data and fault prediction results.

[0025] In step S7, the acquired fault prediction data and corresponding fault prediction results are input into the energy storage digital twin model to visually display the fault prediction data and fault prediction results.

[0026] In this invention, during energy storage monitoring, multi-dimensional key data can be collected for the operating status of different levels of the energy storage system. This multi-dimensional key data can include cell-level data, module-level data, and whole-site-level data. After acquiring the multi-dimensional key data, it can be compressed and synchronized for easy transmission. The transmitted multi-dimensional key data can then be preprocessed. After preprocessing, feature extraction and weighting can be performed on the preprocessed multi-dimensional key data to increase the weight of influential multi-dimensional key data. After weighting, an LSTM-CNN network model can be constructed, and the weighted multi-dimensional key data and corresponding fault results can be used as training data to train the LSTM-CNN network model. After training, current multi-dimensional key data can be collected and preprocessed, and then fed into the trained LSTM-CNN network model to obtain fault prediction data and fault prediction results. After acquiring fault prediction data and corresponding results, these can be input into the energy storage digital twin model, allowing for a direct and intuitive display of the fault prediction data and results. This energy storage monitoring method can predict and provide early warnings for various data points of the energy storage power station based on the acquired multi-dimensional key data.

[0027] In one embodiment of the present invention, such as Figure 2 As shown, the data compression and synchronization process may include: In step S8, data with strong time sequence from the multidimensional key data are acquired as data to be compressed, including data on cell voltage and temperature.

[0028] In step S9, the difference between adjacent data points is calculated based on the acquired data to be compressed in order to perform initial differencing.

[0029] In step S10, after the initial difference is completed, the difference is calculated again to obtain the result of the second difference, thereby compressing the time-series data in the multidimensional key data.

[0030] In step S11, the compressed data and the uncompressed data are stored synchronously to facilitate subsequent calculations.

[0031] In this invention, after acquiring multidimensional key data, the data that needs to be compressed can be compressed to reduce data redundancy in the time dimension. Data with strong time-series characteristics from the multidimensional key data can be used as the data to be compressed, which may include data on the voltage and temperature of the battery cell. After acquiring the data to be compressed, the difference between adjacent data points can be calculated to perform an initial difference. After the initial difference is completed, the difference can be calculated again to obtain the result of the secondary difference, thereby compressing the data with strong time-series characteristics from the multidimensional key data. After compression, the compressed data and the data that does not need compression can be stored synchronously for subsequent calculations.

[0032] In one embodiment of the present invention, such as Figure 3 As shown, the first step of data preprocessing may include: In step S12, multidimensional key data after data compression and synchronization is obtained.

[0033] In step S13, data of categories with a sample size less than a preset threshold in the multidimensional key data are obtained as data to be added.

[0034] In step S14, for the data to be added, any data point is randomly selected from the data to be added.

[0035] In step S15, for a randomly selected data point, the K nearest data points are obtained.

[0036] In step S16, select any one of the K data points and connect it with the randomly selected data point.

[0037] In step S17, any point on the connecting line is randomly selected as a new data point and added to the data to be added in order to expand the amount of data to be added. In step S18, the above steps are repeated until the sample data to be added is greater than the preset threshold.

[0038] In this invention, for data with a small sample size in the multidimensional data, data augmentation is required to increase the training weight of this type of data in the later calculation and prediction process. During augmentation, compressed and synchronized multidimensional key data can be acquired first. After acquiring the multidimensional key data, data from categories with a sample size less than a preset threshold can be designated as data to be added. For the data to be added, any data point can be randomly selected. For the randomly selected data point, its K nearest data points can be obtained, and then any one of these K data points can be connected to the previously randomly selected data point. After connecting the points, any point on the connection can be randomly selected as a new data point and added to the data to be added, thereby expanding the data volume. Through the above steps, the data volume of the data to be added can be increased. Repeating these steps continuously expands the data volume until the sample size of the data to be added exceeds a preset threshold, thus completing the augmentation of the data to be added.

[0039] In one embodiment of the present invention, such as Figure 4 As shown, the second step of data preprocessing may include: In step S19, multidimensional key data are obtained for both increased sample size and no increased sample size.

[0040] In step S20, missing values ​​in the multidimensional key data are filled using linear interpolation.

[0041] In step S21, through 3 The principle is to perform outlier checks on the filled multidimensional key data in order to remove outliers.

[0042] In step S22, after removing outliers, the multidimensional key data is standardized to eliminate dimensional differences in the multidimensional key data.

[0043] In step S23, the multidimensional key data after eliminating dimensional differences are normalized to complete the preprocessing of the multidimensional key data.

[0044] In this invention, when processing multidimensional key data, it is possible to obtain multidimensional key data with and without increased sample size, and then missing values ​​in the multidimensional key data can be filled using linear interpolation. (3) The principle is to perform outlier checks on the filled multidimensional key data to remove outliers. After outlier removal, the multidimensional key data can be standardized to eliminate dimensional differences. The multidimensional key data after dimensional differences are eliminated can then be normalized, completing the preprocessing of the multidimensional key data.

[0045] In one embodiment of the present invention, such as Figure 5 As shown, the data weighting process may include: In step S24, the preprocessed multidimensional key data is obtained, and the correlation between the data of each category in the multidimensional key data and the fault is calculated by the Gumbel Copula function to obtain the corresponding correlation coefficient.

[0046] In step S25, for the correlation coefficients corresponding to the multidimensional key data, the multidimensional key data with correlation coefficients greater than a preset threshold are assigned high weights.

[0047] In step S26, for the correlation coefficients corresponding to the multidimensional key data, the multidimensional key data with correlation coefficients less than a preset threshold are assigned low weights.

[0048] In step S27, multidimensional key data with assigned weights and corresponding fault results are obtained in preparation for being fed into the LSTM-CNN network model.

[0049] In this invention, when weighting multidimensional key data, preprocessed multidimensional key data can be obtained. The correlation between each category of data and the fault can be calculated using the Gumbel Copula function, thus obtaining the corresponding correlation coefficient. For the correlation coefficients corresponding to the multidimensional key data, multidimensional key data with correlation coefficients greater than a preset threshold are assigned high weights. For the correlation coefficients corresponding to the multidimensional key data, multidimensional key data with correlation coefficients less than a preset threshold are assigned low weights. Based on the obtained weighted multidimensional key data and corresponding fault results, preparation can be made for feeding into the LSTM-CNN network model.

[0050] In one embodiment of the present invention, such as Figure 6 As shown, the first step in training a model may include: In step S28, an LSTM prediction network is constructed, and a Dropout layer is added after the LSTM layer to prevent overfitting and improve the model's generalization ability.

[0051] In step S29, an attention mechanism is introduced on the basis of the LSTM output, and attention is paid to the multidimensional key data of the weighted processing category, so that the network focuses on the time step that has the greatest impact on the fault result.

[0052] In step S30, the attention-weighted LSTM output is fed into a one-dimensional convolutional neural network (CNN).

[0053] In step S31, the one-dimensional convolutional neural network extracts local time series patterns through convolutional layers and performs feature compression in pooling layers.

[0054] In step S32, the convolutional features are output as prediction results through a fully connected layer to obtain fault prediction data and corresponding fault prediction results.

[0055] In this invention, when training the LSTM-CNN network model, an LSTM prediction network can be constructed first, and a Dropout layer can be added after the LSTM layers to prevent overfitting and improve the model's generalization ability. An attention mechanism can be introduced based on the LSTM output, focusing on multi-dimensional key data of weighted categories, allowing the network to focus on the time step with the greatest impact on the fault prediction result. The attention-weighted LSTM output can be fed into a one-dimensional convolutional neural network. This one-dimensional convolutional neural network can extract local time-series patterns through convolutional layers and perform feature compression in pooling layers. The convolutional features can then be used to output prediction results through fully connected layers to obtain fault prediction data and corresponding fault prediction results.

[0056] In one embodiment of the present invention, such as Figure 7 As shown, the second step in training the model may include: In step S33, the fault prediction data and corresponding fault prediction results output by the LSTM-CNN network model during training are obtained.

[0057] In step S34, a loss function is constructed based on the fault prediction data.

[0058] In step S35, the gradient in the loss function is calculated, and it is determined whether the gradient is 0.

[0059] In step S36, with the gradient at 0, the fault prediction data is output, and the training of the LSTM-CNN network model is determined to be complete.

[0060] In step S37, if the gradient is not zero, the LSTM-CNN network model is updated, and the weighted multidimensional key data is input into the LSTM-CNN network model again. Then, the step of obtaining the fault prediction data and corresponding fault prediction results output by the LSTM-CNN network model during training is returned until the gradient is zero.

[0061] In this invention, during the training of the LSTM-CNN network model, the fault prediction data and corresponding fault prediction results output by the LSTM-CNN network model during training can be obtained, and then a loss function can be constructed based on this fault prediction data. The gradient in the loss function is calculated, and it can be determined whether the gradient is 0. If the gradient is 0, the fault prediction data is output, and the training of the LSTM-CNN network model is considered complete. If the gradient is not 0, the LSTM-CNN network model can be updated, and the weighted multidimensional key data can be input into the LSTM-CNN network model again. Then, the process can return to step S33 until the gradient is 0.

[0062] In this invention, the cell-level data may include: cell voltage, temperature, internal resistance, and state of charge (SOC). The module-level data may include: cluster voltage consistency data and charge / discharge current stability data. The station-level data may include: PCS power output data, copper busbar temperature data, ambient temperature and humidity data, and fire protection system status data.

[0063] On the other hand, the present invention can also provide an energy storage monitoring system based on spatiotemporal feature fusion and proactive safety early warning. The energy storage monitoring system includes a data acquisition module and an energy storage monitoring module.

[0064] The data acquisition module can be used to collect multi-dimensional key data for the different operating statuses of the energy storage system, including cell-level data, module-level data and whole-site-level data.

[0065] The energy storage monitoring module can be used to execute the energy storage monitoring method based on spatiotemporal feature fusion and proactive safety early warning, as described above, based on the data obtained by the data acquisition module.

[0066] Through the above technical solution, the energy storage monitoring method and system based on spatiotemporal feature fusion and proactive safety early warning provided by this invention can collect multi-dimensional key data for the operating status of different levels of the energy storage system. This multi-dimensional key data can include cell-level data, module-level data, and whole-site-level data. After acquiring the multi-dimensional key data, it can be compressed and synchronized for easy transmission. Then, the transmitted multi-dimensional key data can be preprocessed. After preprocessing, features can be extracted from the preprocessed multi-dimensional key data, and weighting can be performed to increase the weight of influential multi-dimensional key data. After weighting, an LSTM-CNN network model can be constructed, and the weighted multi-dimensional key data and corresponding fault results can be used as training data to train the LSTM-CNN network model. After training, the current multi-dimensional key data can be collected and preprocessed, and then fed into the trained LSTM-CNN network model to obtain fault prediction data and fault prediction results. After acquiring fault prediction data and corresponding results, these can be input into the energy storage digital twin model, allowing for a direct and intuitive display of the fault prediction data and results. This energy storage monitoring method can predict and provide early warnings for various data points of the energy storage power station based on the acquired multi-dimensional key data.

[0067] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0069] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0070] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0071] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0072] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0073] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0074] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0075] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for energy storage monitoring based on spatio-temporal feature fusion and active safety warning, characterized in that, The energy storage monitoring method comprises: For the running state of different levels of energy storage system, multi-dimensional key data is collected, including cell level data, module level data and whole station level data; The obtained multi-dimensional key data is compressed and synchronized; The transmitted multi-dimensional key data is preprocessed; Feature extraction is performed on the preprocessed multi-dimensional key data and weighted processing is performed; An LSTM-CNN network model is constructed, and the weighted multi-dimensional key data and the corresponding fault result are used as training data to train the LSTM-CNN network model; The current multi-dimensional key data is collected and preprocessed, and the preprocessed multi-dimensional key data is input into the trained LSTM-CNN network model to obtain fault prediction data and fault prediction result; The obtained fault prediction data and corresponding fault prediction result are input into the energy storage pre-digital twin model to intuitively display the fault prediction data and fault prediction result.

2. The energy storage monitoring method of claim 1, wherein, The obtained multi-dimensional key data is compressed and synchronized, comprising: The data with strong time sequence in the multi-dimensional key data is obtained as compressed data, including cell voltage and temperature data; According to the obtained compressed data, the difference between adjacent data points is calculated for primary difference; After the primary difference is completed, the difference of the primary difference result is calculated again to obtain the secondary difference result, so that the data with strong time sequence in the multi-dimensional key data is compressed; The compressed data and uncompressed data are stored synchronously to facilitate subsequent calculation.

3. The energy storage monitoring method of claim 1, wherein, The transmitted multi-dimensional key data is preprocessed, comprising: The multi-dimensional key data after data compression and synchronization is obtained; The data of the category with a sample number less than a preset threshold in the multi-dimensional key data is obtained as to-be-increased data: For the to-be-increased data, any one data point in the to-be-increased data is randomly selected; For the randomly selected data point, the nearest K data points are obtained; Any one of the K data points is selected and connected with the randomly selected data point; Randomly select any point on the connecting line as a new data point and add it to the to-be-increased data to expand the data amount of the to-be-increased data; Repeat the above steps until the sample data of the to-be-increased data is greater than the preset threshold.

4. The energy storage monitoring method of claim 3, wherein, The transmitted multi-dimensional key data is preprocessed, comprising: The multi-dimensional key data with increased sample number and without increased sample number is obtained; The missing values in the multi-dimensional key data are filled by linear interpolation method; By 3 principle, the multi-dimensional key data after filling is subjected to outlier check to remove outliers in the multi-dimensional key data; After removing the abnormal values, the multi-dimensional key data is standardized to eliminate the dimension difference in the multi-dimensional key data; The multi-dimensional key data after eliminating the dimension difference is normalized to complete the preprocessing of the multi-dimensional key data.

5. The energy storage monitoring method of claim 1, wherein, Feature extraction is performed on the preprocessed multi-dimensional key data and weighted processing is performed, comprising: Obtain the pre-processed multi-dimensional key data, and calculate the correlation between each category of data in the multi-dimensional key data and the fault through a Gumbel Copula function to obtain a corresponding correlation coefficient; For the correlation coefficient corresponding to the multi-dimensional key data, the multi-dimensional key data corresponding to the correlation coefficient greater than the preset threshold is given a high weight; For the correlation coefficient corresponding to the multi-dimensional key data, the multi-dimensional key data corresponding to the correlation coefficient less than the preset threshold is given a low weight; Obtain the multi-dimensional key data and the corresponding fault result after weighting to prepare for input into the LSTM-CNN network model.

6. The energy storage monitoring method of claim 1, wherein, An LSTM-CNN network model is constructed, and the multi-dimensional key data after weighting processing and the corresponding fault result are used as training data to train the LSTM-CNN network model, including: An LSTM prediction network is constructed, and a Dropout layer is added after the LSTM layer to prevent overfitting and improve the generalization ability of the model; An attention mechanism is introduced based on the LSTM output, and the weighted multi-dimensional key data of the category is focused on, so that the network focuses on the time step that has the greatest impact on the fault result; The attention-weighted LSTM output is input into a one-dimensional convolutional neural network (CNN); The one-dimensional convolutional neural network extracts local time sequence patterns through a convolutional layer and performs feature compression in a pooling layer; The features after convolution are output through a fully connected layer to obtain fault prediction data and corresponding fault prediction results.

7. The energy storage monitoring method of claim 6, wherein, An LSTM-CNN network model is constructed, and the multi-dimensional key data after weighting processing and the corresponding fault result are used as training data to train the LSTM-CNN network model, including: Obtain the fault prediction data and corresponding fault prediction result output by the LSTM-CNN network model during training; Based on the fault prediction data, a loss function is constructed; Calculate the gradient in the loss function and determine whether the gradient is 0; In the case where the gradient is 0, output the fault prediction data and determine that the LSTM-CNN network model training is complete; In the case where the gradient is not 0, update the LSTM-CNN network model, and again input the multi-dimensional key data after weighting processing into the LSTM-CNN network model, and return to the step of obtaining the fault prediction data and corresponding fault prediction result output by the LSTM-CNN network model during training until the gradient is 0.

8. The energy storage monitoring method of claim 1, wherein, The cell-level data includes the voltage, temperature, internal resistance, and SOC of the cell; the module-level data includes the intra-cluster voltage consistency data and the charge / discharge current stability data; and the station-level data includes the PCS power output data, copper bar temperature data, environmental temperature and humidity data, and fire extinguishing system state data.

9. An energy storage monitoring system based on spatio-temporal feature fusion and active safety warning, characterized in that, The energy storage monitoring system includes: A data acquisition module for collecting multi-dimensional key data including cell-level data, module-level data, and station-level data for different levels of operation of the energy storage system; The energy storage monitoring module is used for executing the energy storage monitoring method based on the spatio-temporal feature fusion and active safety early warning according to the data acquired by the data acquisition module.

Images