A method and system for fault diagnosis of an energy storage device

Abstract

This invention relates to a fault diagnosis method and system for energy storage devices, belonging to the field of energy storage system monitoring technology. The method includes the following steps: inputting real-time features extracted from real-time energy storage business object information data into a trained deep learning model based on a self-attention mechanism to obtain fault diagnosis results and expert suggestions for the energy storage device; the dataset used during training of the deep learning model includes temporally ordered features extracted from an expert knowledge base; the expert knowledge base is constructed based on historical energy storage business object information data, including fault types and handling strategies for energy storage devices; the historical energy storage business object information data includes historical fault reports, fault handling records, and expert diagnoses of the energy storage device. This invention utilizes the parallel computing capabilities and self-attention mechanism of the deep learning model based on a self-attention mechanism to improve the model's performance and efficiency, effectively improving the fault diagnosis efficiency of energy storage devices.

Description

Technical Field

[0001] This invention relates to a fault diagnosis method and system for energy storage devices, belonging to the field of energy storage system monitoring technology. Background Technology

[0002] With the rapid development of renewable energy, energy storage systems are playing an increasingly important role in the power system. Especially with a high proportion of renewable energy integrated into the grid, the importance of energy storage systems as regulation and buffer devices is becoming increasingly prominent.

[0003] Currently, battery energy storage systems, as one of the most widely used energy storage technologies, mainly consist of battery cells, battery management systems, inverters, temperature control systems, and other auxiliary equipment. The combined performance of these components directly determines the overall efficiency, safety, and lifespan of the energy storage system. However, while traditional monitoring and maintenance technologies for energy storage systems provide basic operational guarantees for energy storage power stations, they still have many shortcomings.

[0004] On the one hand, the diversity and complexity of energy storage system monitoring data pose challenges to existing monitoring systems. The monitored data is not only distributed at different levels such as individual battery cells, battery clusters, and battery boxes, but also needs to take into account the impact of environmental factors and load changes. However, most current monitoring systems only provide data display and analysis from a single dimension or a limited number of dimensions, making it difficult to comprehensively and intuitively present the operating status of the energy storage system and accurately obtain the status assessment results of the energy storage devices in the system.

[0005] On the other hand, traditional fault diagnosis methods for energy storage devices in energy storage systems rely on human experience and expert judgment. Although they can effectively solve some fault problems, as the scale of equipment expands and the operating time extends, the types and complexity of faults continue to increase, limiting the efficiency and accuracy of traditional methods. This results in slow response speed and low efficiency of remote monitoring and control systems when faced with sudden faults. Summary of the Invention

[0006] The purpose of this invention is to provide a fault diagnosis method and system for energy storage devices to solve the problem of low efficiency in existing fault diagnosis methods.

[0007] To achieve the above objectives, the present invention includes: A fault diagnosis method for energy storage devices according to the present invention includes the following steps: The real-time features extracted from the real-time energy storage business object information data are input into a trained deep learning model based on the self-attention mechanism to obtain the fault diagnosis results and expert suggestions for the energy storage equipment. The dataset used for training the deep learning model includes time-series features extracted from an expert knowledge base; the expert knowledge base is built based on historical energy storage business object information data, and includes the fault types and handling strategies of energy storage equipment. Historical energy storage business object information data includes historical fault reports of energy storage equipment, historical fault handling records, and historical expert diagnoses.

[0008] Furthermore, the real-time energy storage business object information data is the real-time energy storage business object information data preprocessed through filtering algorithms and interpolation methods.

[0009] Furthermore, the historical energy storage business object information data is the historical energy storage business object information data preprocessed through filtering algorithms and interpolation methods.

[0010] Furthermore, the fault assessment results in historical fault reports are obtained in the following way: After cleaning and standardizing the raw battery data across the entire lifecycle of the energy storage power station (total station, battery stack, battery cluster, and individual battery cells), principal component analysis is used to extract key feature sets reflecting the battery's health status from four dimensions: safety assessment, performance assessment, lifespan assessment, and consistency analysis. A sliding window method is then used to extract dynamic feature sets with time-series characteristics from the raw battery data at different levels and dimensions. Finally, fault assessment results reflecting the battery's operating status are synthesized based on the key feature sets and the dynamic feature sets.

[0011] Furthermore, the cleaning process employs filtering algorithms and interpolation methods.

[0012] Furthermore, the normalization process is implemented using an encoding definition module, an instantiation module, and a data mapping module.

[0013] Furthermore, temporal features with time order are obtained by positional encoding of the temporal features input to the deep learning model.

[0014] Furthermore, the deep learning model employs a Transformer model based on a temporal attention mechanism.

[0015] Furthermore, the extraction of real-time features is achieved using the window partitioning method; the extraction of time-series features is also achieved using the window partitioning method.

[0016] The present invention provides a fault diagnosis system for energy storage devices, comprising a processor for executing a computer program to implement the steps of the fault diagnosis method for energy storage devices as described above.

[0017] The beneficial effects of this invention are: This invention is pioneering, providing a fault diagnosis method for energy storage devices. It utilizes the parallel computing capabilities and self-attention mechanism of a deep learning model based on this mechanism to improve the model's performance and efficiency, effectively enhancing the fault diagnosis efficiency of energy storage devices in energy storage systems. The generated fault diagnosis results and expert suggestions provide effective maintenance decision support for energy storage system operators. The method inputs real-time features extracted from real-time energy storage business object information data into a trained deep learning model based on self-attention, obtaining fault diagnosis results and expert suggestions for the energy storage devices. The dataset used during training of the self-attention deep learning model includes temporally ordered features extracted from an expert knowledge base. The expert knowledge base is constructed based on historical energy storage business object information data, including fault types and handling strategies for energy storage devices. This historical energy storage business object information data includes historical fault reports, historical fault handling records, and historical expert diagnoses of the energy storage devices. Attached Figure Description

[0018] Figure 1 This is a flowchart of the usage of a deep learning model based on the self-attention mechanism; Figure 2 This is a flowchart of a fault diagnosis method for energy storage equipment; Figure 3 It is a functional model diagram of a multi-level and multi-dimensional energy storage power station; Figure 4 This is the operation and maintenance architecture diagram of the energy storage business of the energy storage system. Detailed Implementation

[0019] To address the problems in the background art, this invention provides a fault diagnosis method for energy storage devices. It uses accurate historical data to train a deep learning model based on a self-attention mechanism, and leverages the advantages of the self-attention mechanism to achieve efficient fault diagnosis of energy storage devices in the energy storage system.

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0021] An embodiment of a fault diagnosis method for energy storage devices: A fault diagnosis method for energy storage devices, such as Figure 1 As shown, it includes the following steps: Real-time features are extracted from real-time energy storage business object information data. The extracted real-time features are then input into a trained deep learning model based on a self-attention mechanism. The trained deep learning model based on a self-attention mechanism outputs fault diagnosis results and expert suggestions for energy storage equipment. The obtained fault diagnosis results and expert suggestions for energy storage equipment can provide effective maintenance decision support for operators.

[0022] The deep learning model is trained using a dataset that includes time-series features extracted from an expert knowledge base. This knowledge base is built upon historical energy storage business object information data, including fault types and handling strategies for energy storage equipment. The historical energy storage business object information data includes historical fault reports, historical fault handling records, and historical expert diagnoses. The real-time energy storage business object information data includes real-time status information of the energy storage equipment.

[0023] To ensure data reliability, the real-time energy storage business object information data is prepared by noise filtering using a filtering algorithm and missing value imputation using interpolation. Alternatively, other existing methods can also be used to preprocess the real-time energy storage business object information data.

[0024] To ensure data reliability, the historical energy storage business object information data is prepared by noise filtering using a filtering algorithm and missing value imputation using interpolation. Alternatively, other existing methods can be used to preprocess the historical energy storage business object information data.

[0025] Specifically, temporal features with time sequence are obtained by positional encoding of the temporal features input to the deep learning model.

[0026] Specifically, the deep learning model uses the Transformer model based on the temporal attention mechanism, but other existing deep learning models based on the self-attention mechanism can also be used.

[0027] Specifically, real-time feature extraction is achieved using a window partitioning method.

[0028] Specifically, the extraction of temporal features is achieved using the window partitioning method.

[0029] Specifically, data augmentation methods can be used to generate more training samples for the training model when there are few training samples, thereby enhancing the model's generalization ability.

[0030] Historical fault reports, historical fault handling records, and historical expert diagnoses of energy storage devices can all be obtained using existing methods.

[0031] The fault assessment results in the historical fault reports were obtained in the following ways: After cleaning and standardizing the raw battery data across the entire lifecycle of the energy storage power station (total station, battery stack, battery cluster, and individual battery cells), Principal Component Analysis (PCA) is used to extract key feature sets reflecting the battery's health status from four dimensions: safety assessment, performance assessment, lifespan assessment, and consistency analysis. Furthermore, a sliding window method is employed to extract dynamic feature sets with time-series characteristics from the raw battery data at different levels and dimensions. Finally, a fault assessment result reflecting the battery's operating status is synthesized based on the key feature sets and the dynamic feature sets.

[0032] The cleaning process can be implemented using other existing methods, or by employing filtering algorithms and interpolation methods to ensure the accuracy and continuity of the data. Filtering methods can include Kalman filtering or wavelet transform; interpolation methods can include linear interpolation or spline interpolation, etc.

[0033] The normalization process is implemented using existing coding definition modules, instantiation modules, and data mapping modules to facilitate subsequent analysis operations.

[0034] This invention achieves automated fault diagnosis and expert suggestion generation by extracting features from energy storage equipment fault diagnosis and expert suggestions, thereby providing effective maintenance decision support for operators.

[0035] Among them, such as Figure 2 As shown, automated fault diagnosis and expert suggestion generation are achieved through the following steps: 1. Data acquisition and preprocessing: Collect and analyze historical data of energy storage business objects (historical energy storage business object information), which includes historical fault reports, historical fault handling records, historical expert diagnoses and other documents of energy storage equipment.

[0036] Data reliability is ensured by using noise filtering and missing value imputation preprocessing methods. An expert knowledge base is built based on historical energy storage business object information data to form a database of storage fault types and handling strategies.

[0037] 2. Extraction of time-series data features (time series features) and training of the model: Taking the Transformer model based on the temporal attention mechanism as an example, firstly, time series data processing is performed to transform the fault data in the expert knowledge base into a format suitable for time series analysis. This includes serializing the equipment operating status, sensor data, etc. through window partitioning to generate a time step sequence suitable for the input of the Transformer model, that is, extracting the temporal features.

[0038] Next, positional encoding is added to the input data to generate temporal features with time order, ensuring that the Transformer model can understand the time order.

[0039] In addition, data augmentation methods can be used to generate more training samples, thereby enhancing the model's generalization ability.

[0040] Based on the standard Transformer architecture, a deep learning model adapted to fault diagnosis of energy storage equipment is constructed. This model includes multiple self-attention layers and feedforward neural network layers, which can capture complex temporal relationships and fault modes in the input data.

[0041] Finally, the dataset used by this deep learning model includes time-series features extracted from an expert knowledge base for training. The trained Transformer model is then used to process real-time data from energy storage devices to identify potential failure modes.

[0042] This model can classify different types of faults, such as battery faults, control system faults, and connectivity problems. While identifying fault types, the model can also assess the severity of faults based on data characteristics, providing a basis for subsequent fault handling.

[0043] Feature extraction from time-series data using Transformer models based on temporal attention mechanisms has become an important direction in time-series data processing in recent years. While Transformer initially achieved great success in Natural Language Processing (NLP), its powerful self-attention mechanism effectively captures dependencies between long time steps, avoiding the training difficulties and information loss problems of traditional convolutional neural networks in long sequences. The self-attention mechanism allows the model to dynamically pay attention to relevant information from other time steps while extracting features from the input at each time step, thus obtaining richer contextual representations. It has also demonstrated excellent performance in time-series data analysis, especially in capturing long-term dependencies and complex patterns.

[0044] The fault assessment results in the historical fault reports of energy storage equipment are obtained through the following methods: 1. Data Acquisition and Processing: Collect raw battery data for the entire lifecycle of the energy storage power station across four levels: the entire energy storage power station, the battery stack, the battery cluster, and the individual battery cell.

[0045] Among them, the whole-station energy storage power station, that is, the whole-station level: collect raw battery data from the overall level of the energy storage power station, which includes the overall monitoring of parameters such as temperature, voltage, and internal resistance of all battery packs.

[0046] Battery stack, or stack hierarchy: Data is collected from each stack to obtain the overall performance of the batteries within the stack.

[0047] Battery clusters, or cluster hierarchies: Each cluster contains multiple battery cells, allowing for more refined data collection and analysis to capture the characteristics of each cluster.

[0048] Battery cell level, or cell level: the most granular level, collects data for each individual battery cell to ensure that the status of each battery can be evaluated and monitored individually.

[0049] Because external interference (such as electrical noise and temperature fluctuations) may occur during battery data acquisition, filtering algorithms (such as Kalman filtering and wavelet transform) are needed to remove noise to ensure data accuracy. Missing values ​​may appear in battery data over long-term operation; these can be filled using interpolation methods (such as linear interpolation and spline interpolation) to ensure data continuity.

[0050] Furthermore, the data cleaned using the above methods is standardized. Through existing coding definition modules, instantiation modules, and data correspondence modules, multi-dimensional business-oriented battery data is standardized to facilitate subsequent analysis.

[0051] 2. Obtain the characteristic fingerprint: "Feature fingerprinting" refers to the extraction of a set of key features that reflect the current health status of a battery by conducting multi-dimensional analysis of its operational data at different time scales, levels, and dimensions. This feature data can not only help identify the current state of the battery, but also reveal potential health risks, performance degradation trends, and possible failure modes.

[0052] Specifically, the principal component analysis (PCA) method is used to screen out a set of key features closely related to the battery health status from a large amount of data. For raw battery data of different levels and dimensions, the sliding window method is used to extract a set of dynamic features with time-series characteristics.

[0053] For the high-dimensional features in battery data, dimensionality reduction techniques such as PCA are used to extract several principal components that can explain the main variability of the data, which helps to simplify the model and improve evaluation efficiency.

[0054] After the above steps, multiple feature sets can be generated from data at different time scales, levels, and dimensions. These feature sets are then combined to form the battery's "fingerprint." These fingerprints contain information about the battery's health status at a specific moment or time period, reflecting the battery's overall health condition.

[0055] 3. Results: Through the above steps, the final output is a battery operating status assessment, providing accurate analysis results and in-depth lifespan predictions. It achieves "feature fingerprint" data analysis of batteries with different characteristics in energy storage power stations, including temperature, voltage, internal resistance, and faults, providing decision support for the operation and management of energy storage power stations, including recommendations for battery repair, replacement, and long-term operation optimization.

[0056] This invention proposes a battery operating status assessment method based on multi-level, multi-dimensional, and multi-timescale battery sample feature data. By analyzing raw battery data across four levels—"whole station," "stack," "cluster," and "single cell"—the method standardizes multi-dimensional operational battery data. Through threshold and data analysis at different time scales, it obtains "feature fingerprints" of batteries with different characteristics to assess battery status. Raw data from energy storage stations is obtained at four levels: "whole station," "stack," "cluster," and "single cell." After data cleaning techniques, existing coding definition modules, instantiation modules, and data correspondence modules are used to standardize the multi-dimensional operational battery data, facilitating subsequent analysis. This enables the analysis of "feature fingerprint" data such as temperature, voltage, internal resistance, and faults of batteries with different characteristics in energy storage power stations. It comprehensively perceives the operating status, operational faults, and energy consumption of core equipment in the power station, proactively analyzes multi-level physical entities such as "single cell," "cluster," and "stack," and, combined with correlated data, deeply explores the causes of early warnings. The functional model of the multi-level, multi-dimensional energy storage power station is as follows: Figure 3 As shown.

[0057] The "multi-dimensional" aspect refers to the innovative use of four dimensions—safety assessment, performance assessment, lifespan assessment, and consistency analysis—to evaluate the battery's operating status. The "multi-timescale" aspect refers to analyzing the battery data across different time thresholds, with short timescales set between 5 and 15 minutes and long timescales ranging from days to months to years. This battery operating status evaluation method meets the long-term safe and stable operation requirements of energy storage power stations.

[0058] This invention evaluates the safety performance of energy storage power stations through multi-layer, multi-dimensional, and multi-time battery operating status assessment, providing accurate analysis results (fault assessment results), in-depth lifespan prediction, and fault diagnosis functions. Furthermore, by extracting features from fault diagnosis and expert recommendations for energy storage equipment, it achieves automated fault diagnosis and expert recommendation generation, providing effective maintenance decision support for operators and further improving the efficiency and quality of operation and maintenance work. This comprehensively enhances the monitoring, analysis, and management level of energy storage power stations, providing strong support for the development of energy storage technology. It addresses the long-term safe and stable operation requirements of energy storage power stations by researching multi-level and multi-dimensional early warning technologies, enabling online assessment and proactive early warning of energy storage stations. This provides comprehensive and accurate monitoring and operation and maintenance support, improves the intelligence level of energy storage stations, significantly enhances the transparency and controllability of system operation, and achieves refined and automated operation and maintenance of energy storage power stations.

[0059] The operation and maintenance architecture of energy storage systems, such as... Figure 4 As shown, the system consists of a data acquisition layer, a data analysis layer, and an advanced application layer. The data acquisition layer uses a parallel and scalable architecture to collect, process, and store battery data. It employs a unified modeling approach to transform the four remote sensing models (remote monitoring, remote sensing, remote telemetry, and remote sensing) into energy storage business data, forming a "unified business model" that accurately describes the energy storage system from multiple dimensions, including battery consistency, energy consumption, equipment performance, safety, lifespan, and individual battery characteristics. The data analysis layer uses a distributed database and distributed services to ensure efficient data computation and processing across multiple time scales. The advanced application layer, supported by data analysis, assists in the fusion of mechanistic models and AI models to conduct equipment early warning and intelligent operation and maintenance.

[0060] An embodiment of a fault diagnosis system for energy storage devices: A fault diagnosis system for energy storage devices includes a processor for executing a computer program to implement the steps of a fault diagnosis method for energy storage devices. The fault diagnosis method for energy storage devices has been described in detail in an embodiment of such a method and will not be repeated here.

Claims

1. A failure diagnosis method for an energy storage device, characterized by, Includes the following steps: The real-time features extracted from the real-time energy storage business object information data are input into a trained deep learning model based on the self-attention mechanism to obtain the fault diagnosis results and expert suggestions for the energy storage equipment. The dataset used for training the deep learning model includes temporal features extracted from an expert knowledge base; the expert knowledge base is constructed based on historical energy storage business object information data, and includes the fault types and handling strategies of energy storage equipment. The historical energy storage business object information data includes historical fault reports, historical fault handling records, and historical expert diagnoses of energy storage equipment.

2. The failure diagnosis method for an energy storage device according to claim 1, characterized by, The real-time energy storage business object information data is the real-time energy storage business object information data preprocessed by filtering algorithms and interpolation methods.

3. The failure diagnosis method for an energy storage device according to claim 1 or 2, characterized by, The historical energy storage business object information data is the historical energy storage business object information data preprocessed through filtering algorithms and interpolation methods.

4. The failure diagnosis method for an energy storage device according to claim 1, wherein The fault assessment results in the historical fault reports are obtained in the following manner: After cleaning and standardizing the raw battery data across the entire lifecycle of the energy storage power station (total station, battery stack, battery cluster, and individual battery cells), principal component analysis is used to extract key feature sets reflecting the battery's health status from four dimensions: safety assessment, performance assessment, lifespan assessment, and consistency analysis. A sliding window method is then used to extract dynamic feature sets with time-series characteristics from the raw battery data at different levels and dimensions. Finally, a fault assessment result reflecting the battery's operating status is synthesized based on the key feature sets and the dynamic feature sets.

5. The failure diagnosis method for an energy storage device according to claim 4, wherein The cleaning process is implemented using filtering algorithms and interpolation methods.

6. The fault diagnosis method for energy storage equipment according to claim 4, characterized in that, The normalization process is implemented using an encoding definition module, an instantiation module, and a data mapping module.

7. The fault diagnosis method for energy storage equipment according to claim 1, characterized in that, The temporal features with time sequence are obtained by positional encoding of the temporal features input to the deep learning model.

8. The fault diagnosis method for energy storage equipment according to claim 1, characterized in that, The deep learning model adopts the Transformer model based on the temporal attention mechanism.

9. The fault diagnosis method for energy storage equipment according to claim 1, characterized in that, The extraction of real-time features is achieved using a window partitioning method; the extraction of temporal features is achieved using a window partitioning method.

10. A fault diagnosis system for energy storage devices, comprising a processor, characterized in that, The processor is used to execute a computer program to implement the steps of the fault diagnosis method for energy storage devices as described in any one of claims 1 to 9.

Images