A battery energy storage system sensor fault diagnosis method and system
By combining hierarchical fault location and inspection with a time-frequency domain dual-stream network, the problem of the diversity and complexity of sensor fault signals in battery energy storage systems is solved, and high-precision and efficient sensor fault diagnosis is achieved in low signal-to-noise ratio environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XI AN JIAOTONG UNIV
- Filing Date
- 2026-02-13
- Publication Date
- 2026-05-29
AI Technical Summary
In low signal-to-noise ratio environments, sensor fault signals in battery energy storage systems exhibit diversity and complexity. Existing diagnostic methods struggle to effectively separate abrupt and gradual faults, resulting in insufficient generalization ability of diagnostic models under complex operating conditions. Furthermore, massive data input leads to computational redundancy and a decrease in diagnostic accuracy.
A hierarchical fault location and inspection strategy is adopted, which combines a time-frequency domain dual-stream network. By segmenting data through a sliding window and performing hierarchical fault location and inspection, a dual-stream heterogeneous neural network is constructed to extract mutation-type and gradual-type features respectively, and then perform feature fusion. Finally, the fault type is output through a classifier.
It significantly improves diagnostic robustness and accuracy in low signal-to-noise ratio environments, reduces computational redundancy, and achieves accurate separation and efficient diagnosis of multimodal fault characteristics, making it suitable for online monitoring of large-scale battery energy storage systems.
Smart Images
Figure CN122108225A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of battery energy storage system safety monitoring technology, specifically relating to a method and system for diagnosing sensor faults in battery energy storage systems. Background Technology
[0002] In low signal-to-noise ratio operating environments, sensor fault signals in battery energy storage systems exhibit significant diversity and complexity. On the one hand, abrupt faults such as bias faults, jamming faults, noise interference faults, and periodic interference faults typically manifest as high-frequency energy spikes or textural changes within a local timescale. On the other hand, gradual faults such as drift faults and slow inaccuracies caused by aging exhibit trend shifts over long timescales, and their amplitude changes are often masked by system noise or operating condition fluctuations.
[0003] Existing diagnostic methods based on a single feature domain or a single network structure typically require a trade-off between "mutation sensitivity" and "trend stability": convolutional neural networks, which focus on high-frequency features, are prone to missing gradual-type faults; while recurrent neural networks, which focus on time-dependent modeling, struggle to accurately identify abrupt anomalies in noisy environments. This contradiction is particularly pronounced in battery energy storage systems with numerous and diverse sensors (voltage, current, temperature, electrochemical), directly resulting in insufficient generalization ability of diagnostic models under complex operating conditions.
[0004] Furthermore, in large-scale sensor cluster scenarios, directly inputting all sensor data into a deep learning model not only results in computational redundancy but also weakens the model's ability to distinguish fault characteristics due to the introduction of a large amount of normal data, thereby affecting diagnostic accuracy and real-time performance. Therefore, how to ensure diagnostic accuracy while effectively separating abrupt and gradual fault characteristics and reducing the input complexity of the deep learning model constitutes a pressing technical challenge in this field. Summary of the Invention
[0005] The technical problem to be solved by the present invention is to provide a sensor fault diagnosis method and system for battery energy storage systems, which addresses the shortcomings of the prior art. The method and system first locates the fault and then diagnoses it through a hierarchical strategy and combines a time-frequency domain dual-stream network to solve the technical problems of difficulty in screening massive amounts of data and conflict in the extraction of multimodal fault features. It has the advantages of accurate positioning and comprehensive classification, significantly improves the diagnostic robustness in low signal-to-noise ratio environments, and meets the needs of online diagnosis.
[0006] The present invention adopts the following technical solution: A method for diagnosing sensor faults in a battery energy storage system includes the following steps: S1. Obtain time-series data of the sensor cluster in the battery energy storage system, segment the time-series data using a sliding window, and construct a sensor cluster data matrix. S2. Perform hierarchical fault location and inspection (HFLC) on the sensor cluster data matrix. The hierarchical fault location and inspection includes a global detection stage and a local identification stage. In the global detection stage, a set of suspected sensors is selected. In the local identification stage, the final faulty sensor is locked and the faulty data fragment is extracted. S3. Construct a dual-stream heterogeneous neural network, which includes a first branch for parallel time-frequency feature extraction and a second branch for trend feature extraction, and input the fault data fragments into the first branch and the second branch respectively; S4. In the first branch, the fault data segment is subjected to continuous wavelet transform to generate a scale spectrum map, and the high-frequency spatial texture features in the scale spectrum map are extracted using a two-dimensional convolutional neural network to obtain a first feature vector. S5. In the second branch, the trend component is extracted and reconstructed by discrete wavelet transform of the fault data segment, and the long-term time-dependent features in the trend component are extracted by bidirectional long short-term memory network to obtain the second feature vector. S6. The first feature vector and the second feature vector are fused, and the fused feature vector is input into the classifier. The classifier outputs the sensor fault type to complete the sensor fault diagnosis.
[0007] Preferably, in step S1, the sensor cluster includes a voltage sensor and a temperature sensor, the voltage sensor has a sampling frequency of 20Hz, and the temperature sensor has a sampling frequency of 1Hz; the sliding window has a length of 1024 sampling points, and the overlap rate of the sliding window is 50%.
[0008] Preferably, in step S2, the specific process of the global detection phase is as follows: Calculate the energy of the difference sequence between the sensor and its neighboring sensors, and calculate the energy ratio based on the energy of the difference sequence. If the energy ratio exceeds a preset threshold, the corresponding sensor is recorded in the suspect sensor set. The sensor cluster data matrix is Z-score normalized, the covariance matrix and the average correlation coefficient between the sensor and other sensors are calculated, the correlation ratio is calculated based on the average correlation coefficient, and if the correlation ratio exceeds a preset threshold, the corresponding sensor is recorded in the suspect sensor set. Generate reference signals for the sensors in the suspected sensor set.
[0009] Preferably, the preset threshold corresponding to the energy ratio is 2, and the preset threshold corresponding to the correlation ratio is 0.3.
[0010] Preferably, in step S2, the specific process of the local identification stage is as follows: Using the median of neighboring sensor readings as a reference signal, relative difference and cumulative energy difference are calculated based on the reference signal. If the relative difference exceeds a preset threshold or the cumulative energy difference exceeds a preset threshold, the corresponding sensor is identified as a faulty sensor, and the time series data of the faulty sensor is extracted as the fault data segment.
[0011] Preferably, the preset threshold corresponding to the relative difference is 0.04, and the preset threshold corresponding to the cumulative energy difference is 5.0.
[0012] Preferably, in step S4, the continuous wavelet transform uses the generalized Morse wavelet as the mother wavelet, and the two-dimensional convolutional neural network is a three-layer cascaded convolution-batch normalization-ReLU-pooling structure, used to capture the abrupt fault characteristics of the sensor.
[0013] Preferably, in step S5, the discrete wavelet transform uses biorthogonal wavelets to perform multi-layer decomposition and reconstruction of the fault data segment, with a decomposition layer of 10; the bidirectional long short-term memory network is a three-layer stacked structure with 128 hidden units in each layer, used to capture the gradual fault characteristics of the sensor.
[0014] Preferably, in step S6, the fusion process is feature vector concatenation; the classifier's processing involves inputting the concatenated feature vector into a fully connected layer for nonlinear mapping, then calculating the fault probability and outputting the fault type through a Softmax layer. The fully connected layer has a 2-layer structure and the dropout rate of the Dropout layer is 0.4. The fault types include bias faults, jamming faults, noise faults, periodic faults, drift faults, and combined faults.
[0015] Secondly, embodiments of the present invention provide a sensor fault diagnosis system for a battery energy storage system, comprising: The data module is used to acquire time-series data of the sensor cluster in the battery energy storage system, and to segment the time-series data using a sliding window to construct a sensor cluster data matrix. The positioning module is signal-connected to the data module and is used to receive the sensor cluster data matrix, perform hierarchical fault location and inspection HFLC, filter the set of suspected sensors through the global detection stage, and then lock the faulty sensor and extract the faulty data fragment through the local identification stage. An extraction module, signal-connected to the positioning module, is used to receive the fault data segment. The module includes a parallel time-frequency feature extraction first branch and a trend feature extraction second branch. The first branch performs continuous wavelet transform on the fault data segment to generate a scale spectrum and extracts high-frequency spatial texture features through a two-dimensional convolutional neural network to obtain a first feature vector. The second branch performs discrete wavelet transform on the fault data segment to extract and reconstruct trend components and extracts long-term time-dependent features through a bidirectional long short-term memory network to obtain a second feature vector. The fusion module is signal-connected to the extraction module and is used to receive the first feature vector and the second feature vector and perform fusion processing to obtain a fused feature vector. The classification module, which is signal-connected to the fusion module, is used to receive the fused feature vector, input it into the classifier, and output the sensor fault type to complete the sensor fault diagnosis.
[0016] Thirdly, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described battery energy storage system sensor fault diagnosis method.
[0017] Fourthly, embodiments of the present invention provide a computer-readable storage medium including a computer program, which, when executed by a processor, implements the steps of the above-described battery energy storage system sensor fault diagnosis method.
[0018] Fifthly, a chip includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described battery energy storage system sensor fault diagnosis method.
[0019] In a sixth aspect, embodiments of the present invention provide an electronic device including a computer program, which, when executed by the electronic device, implements the steps of the above-described battery energy storage system sensor fault diagnosis method.
[0020] Compared with the prior art, the present invention has at least the following beneficial effects: A method for diagnosing sensor faults in battery energy storage systems is proposed. First, a sliding window segmentation of the raw sensor cluster data provides a standardized data matrix input for subsequent processing. Second, a hierarchical fault location and inspection strategy is introduced. This step is not only data preprocessing but also the logical starting point of the entire invention, transforming the analysis of the entire cluster data into the analysis of suspected fault segments through location. Finally, a dual-stream heterogeneous neural network is constructed, which for the first time in the field of battery energy storage diagnosis physically decouples abrupt feature extraction and gradual feature extraction, modeling them separately and then fusing them for decision-making. This overcomes the inherent trade-off between abrupt sensitivity and trend stability in traditional single-network models. Through division of labor at the architectural level, CNN focuses on local high-frequency anomalies, while RNN focuses on global low-frequency drift, fundamentally solving the mutual inhibition problem of multimodal fault feature extraction. Simultaneously, the pre-localization mechanism significantly reduces the amount of input data for subsequent networks, solving the computational redundancy problem caused by massive sensor data and enabling online real-time diagnosis.
[0021] Furthermore, differentiated sampling frequencies are set for the different monitoring characteristics of voltage and temperature sensors. 20Hz voltage sampling captures high-frequency electrical signal changes, while 1Hz temperature sampling adapts to slow heating characteristics, avoiding insufficient sampling or data redundancy. A sliding window with 1024 sampling points and a 50% overlap rate ensures both the continuity and feature integrity of the time-series data, effectively segmenting the data to form a standardized matrix. This provides standardized input for subsequent batch calculations of the HFLC algorithm. Simultaneously, the overlap rate design avoids missed detections caused by fault segments being segmented by the window, significantly improving the effectiveness of data preprocessing and the accuracy of subsequent fault location.
[0022] Furthermore, by employing a dual-dimensional approach of energy ratio and correlation ratio, abnormal sensors are identified from two perspectives: signal energy variation and spatial correlation characteristics. Compared to a single-dimensional approach, this significantly reduces the probability of missing faulty sensors. First, the energy ratio of the difference sequence is calculated to capture signal amplitude anomalies. Then, Z-score normalization is used to eliminate the influence of dimensions, and the correlation ratio is calculated to analyze the cooperative anomalies between sensors. After these two steps, a reference signal is generated, providing a foundation for subsequent local identification. This achieves rapid, coarse-grained fault screening across the entire system, significantly reducing the computational load of subsequent local identification and improving the overall efficiency of fault location.
[0023] Furthermore, the energy ratio threshold of 2 and the correlation ratio threshold of 0.3 are optimal values calibrated based on the actual operating data and fault characteristics of the battery energy storage system sensors. This avoids the possibility of missed detection of faulty sensors due to excessively high thresholds, while also preventing the introduction of too many invalid suspect sensors due to excessively low thresholds, effectively controlling the size of the suspect sensor set. This makes the judgment process in the global detection phase reproducible and verifiable, eliminating the drawbacks of empirical judgment and ensuring the consistency and stability of global detection under different operating conditions.
[0024] Furthermore, using the median of neighboring sensor readings as a reference signal provides stronger resistance to extreme value interference compared to the mean reference signal, and better reflects the actual operating conditions of battery energy storage systems. A refined judgment is made using both relative difference and cumulative energy difference criteria. Relative difference captures real-time shift anomalies in the signal, while cumulative energy difference captures long-term cumulative anomalies. The combination of these two criteria enables precise identification of faulty sensors. Simultaneously, fault time-series data is extracted as fault samples, ensuring that the subsequent input to the dual-stream heterogeneous neural network consists only of fault-related data. This significantly reduces model input complexity, enhances the saliency of fault features, and improves the efficiency and accuracy of the diagnostic model.
[0025] Furthermore, the relative difference threshold of 0.04 and the cumulative energy difference threshold of 5.0 are optimal thresholds calibrated for low signal-to-noise ratio environments of battery energy storage systems. These thresholds adapt to the characteristic differences of various fault types, such as bias and drift, effectively distinguishing normal operating condition fluctuations from genuine sensor faults and avoiding misjudgments caused by environmental noise and changes in operating conditions. This quantification standardizes the judgment process in the local identification stage, ensuring the accuracy of fault sensor location. Simultaneously, the precisely extracted fault data fragments make subsequent feature extraction more targeted, effectively improving the ability of the dual-stream heterogeneous neural network to capture fault features and providing a high-quality data foundation for refined fault diagnosis.
[0026] Furthermore, the generalized Morse wavelet is used as the mother wavelet for continuous wavelet transform. Its time-frequency resolution is adjustable, enabling it to accurately capture the non-stationary transient features of abrupt faults. This transforms one-dimensional time-series data into a two-dimensional scale spectrum, achieving visualization of time-frequency domain features. The 2D-CNN with a three-layer cascaded convolution-batch normalization-ReLU-pooling structure can extract high-frequency spatial texture features from the scale spectrum layer by layer. It effectively captures high-frequency energy surges and texture structure changes in abrupt faults such as bias, jamming, and noise. The batch normalization layer can also improve the training efficiency and robustness of the network, ensuring accurate extraction of abrupt fault features even in low signal-to-noise ratio environments.
[0027] Furthermore, using biorthogonal wavelets for discrete wavelet transform with a decomposition layer of 10 effectively separates low-frequency trend components, eliminates high-frequency noise interference, and fully preserves the long-term trend characteristics of gradual faults such as drift and aging inaccuracies. Biorthogonal wavelets also avoid waveform phase distortion, ensuring the authenticity of trend features. A three-layer stacked BiLSTM network with 128 hidden units per layer can bidirectionally capture the time-dependent features of trend components, fully exploring the long-term trend shift patterns of gradual faults. Compared to unidirectional LSTM, it provides more comprehensive feature extraction for gradual faults, significantly improving the ability to identify low-amplitude, slowly varying gradual faults and filling the gap in single-network diagnosis of gradual faults.
[0028] Furthermore, a feature vector concatenation fusion method is adopted to fully preserve high-frequency features in the time-frequency domain and trend features in the time domain, avoiding information loss during feature fusion and ensuring that the fused features simultaneously contain key information about both abrupt and gradual faults. Two fully connected layers combined with a Dropout layer with a 0.4 dropout rate achieve non-linear feature mapping while effectively preventing network overfitting. The Softmax layer implements probabilistic determination of multiple fault types. Six fault types are explicitly supported, covering the main fault forms of sensors in battery energy storage systems, enabling refined fault type identification, making diagnostic results more instructive, and providing accurate basis for the operation and maintenance of battery energy storage systems.
[0029] It is understood that the beneficial effects of the second to sixth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.
[0030] In summary, this invention achieves data dimensionality reduction and fault enhancement through hierarchical localization, physical decoupling and parallel extraction of abrupt / gradual features through a dual-stream heterogeneous network, and accurate decision-making through feature fusion and classification. The end-to-end collaborative design fundamentally solves the difficulties of massive data screening and the conflicts in multimodal feature extraction. In environments with a low signal-to-noise ratio of 40-60dB, the localization accuracy is >98%, and the classification accuracy is 99.5%-99.9%, significantly outperforming existing single-network architectures, and possessing the engineering advantages of high precision, high robustness, and low latency.
[0031] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0032] Figure 1 This is a flowchart illustrating the entire process of the method of the present invention; Figure 2 A flowchart illustrating the HFLC hierarchical fault location algorithm; Figure 3 This is a schematic diagram of a two-stream heterogeneous neural network architecture and feature fusion. Figure 4 This is a comparison chart of the diagnostic accuracy of different methods in a noisy environment in Embodiment 1 of the present invention, illustrating the performance advantages of the method of the present invention under different signal-to-noise ratios; Figure 5 The comparison of the classification confusion matrix of the 2D-CNN network and BiLSTM network in Embodiment 1 of the present invention with that of the dual-stream heterogeneous neural network of the present invention shows that the method of the present invention has higher accuracy. Among them, (a) is the 2D-CNN network, (b) is the BiLSTM network, and (c) is the network of the present invention.
[0033] Figure 6A schematic diagram of a computer device provided in an embodiment of the present invention; Figure 7 This is a block diagram of a chip provided according to an embodiment of the present invention.
[0034] Among them, 60. Computer equipment; 61. Processor; 62. Memory; 63. Computer program; 600. Electronic device; 610. Processing unit; 620. Storage unit; 6201. Random access memory unit; 6202. Cache memory unit; 6203. Read-only memory unit; 6204. Program / utility; 6205. Program module; 630. Bus; 640. Display unit; 650. Input / output interface; 660. Network adapter; 700. External device. Detailed Implementation
[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0036] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.
[0037] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.
[0038] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.
[0039] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.
[0040] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."
[0041] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.
[0042] This invention provides a sensor fault diagnosis method for battery energy storage systems. First, time-series data from multiple types of sensors in the battery energy storage system are acquired. A sensor cluster data matrix is constructed using a sliding window. Then, a hierarchical fault location and inspection strategy is employed to perform dimensionality reduction filtering and fault segment extraction on the sensor data, thereby reducing the input scale of the subsequent diagnostic model and enhancing the saliency of fault features. Based on this, a dual-stream heterogeneous neural network coupled with the hierarchical fault location results is constructed. The located fault data is input into a time-frequency feature extraction branch and a trend feature extraction branch, respectively. The time-frequency branch extracts abrupt abnormal features from the sensor signals based on continuous wavelet transform and a two-dimensional convolutional neural network, while the trend branch extracts gradual abnormal features from the sensor signals based on discrete wavelet transform and a bidirectional long short-term memory neural network. Finally, a feature fusion and classification module outputs the sensor fault type. This invention achieves precise localization and refined diagnosis of various sensor faults in low signal-to-noise ratio environments through a collaborative design of hierarchical localization and heterogeneous modeling. This effectively reduces computational redundancy and improves diagnostic accuracy and real-time performance, making it suitable for online monitoring and maintenance of large-scale battery energy storage systems. It effectively utilizes the spatial correlation of sensor clusters in battery energy storage systems, performing dimensionality reduction and fault segment extraction on sensor data through hierarchical fault localization and inspection. This reduces the input size of subsequent deep learning models and enhances the saliency of fault features. Based on this, a dual-stream heterogeneous neural network coupled with the hierarchical fault localization results is constructed to perform differentiated modeling of abrupt and gradual features in the localized fault data, thereby achieving refined diagnosis of various sensor faults.
[0043] Please see Figure 1 The present invention provides a method for diagnosing sensor faults in a battery energy storage system, comprising the following steps: S1. Obtain time-series data of the sensor cluster in the battery energy storage system, and construct the sensor cluster data matrix to be analyzed using a sliding window. ; Real-time output data from voltage sensors (sampling frequency 20Hz) and temperature sensors (sampling frequency 1Hz) in the battery energy storage system are collected to establish a sensor cluster data matrix. The data acquisition module can be configured to acquire data in real time through the Battery Management System (BMS) interface; a sliding window is used to segment the data, with a window length of 1024 sampling points and an overlap rate of 50%. Assuming the total number of sensors in the cluster is... The length of the collected data is Using sliding windows The original data is segmented to obtain the data matrix at the current time. .
[0044] S2. Perform Hierarchical Fault Localization and Check (HFLC), which includes a global detection phase and a local identification phase. In the global detection phase, the energy ratio and correlation ratio of each sensor are calculated in parallel to filter out a set of suspected sensors. In the local identification phase, relative differences and cumulative energy differences are calculated based on neighborhood reference signals to pinpoint the final faulty sensor and extract fault data segments. The hierarchical fault localization and check step is used not only to determine the index position of the faulty sensor, but also to extract time series segments with statistically abnormal characteristics, so that the input data of the dual-stream heterogeneous neural network is distinguishable in terms of time-frequency characteristics and trend characteristics, thereby improving the ability of different branches to distinguish the corresponding fault types.
[0045] Please see Figure 2 The global detection phase aims to quickly filter out normal nodes by leveraging the spatial redundancy of the sensor cluster. The specific calculations are as follows: S301, Computational Sensor Energy of the difference sequence between its data and that of neighboring sensors (spatially adjacent sensors). And calculate the energy ratio. :
[0046] like Exceeding the threshold (In this embodiment, the threshold is 2), and it is added to the suspect set. .
[0047] S302. Perform Z-score standardization on the data and calculate the covariance matrix. and average correlation coefficient ; Computational Sensors Average correlation coefficient with other sensors And calculate the correlation ratio. :
[0048] like or Exceeding the threshold (In this embodiment, the threshold is 0.3), and it is added to the suspect set. .
[0049] S303, regarding the suspected group The sensor in the middle generates a reference signal. .
[0050] The local discrimination stage aims to address the false alarm problem in the global screening process; firstly, a reference signal is generated. The median of the neighboring sensor readings is used, and then the relative difference criterion and the cumulative energy difference criterion are calculated:
[0051]
[0052] If the relative difference criterion is met (The threshold in this embodiment is 0.04) or cumulative energy difference criterion (In this embodiment, the threshold is 5.0), then it is identified as a faulty sensor, and the time series data of the sensor is extracted as a fault sample. .
[0053] S3. Construct a two-stream heterogeneous neural network and input the extracted fault data segments into two parallel branches of the network respectively; the first branch is the time-frequency feature extraction branch, and the second branch is the trend feature extraction branch; S4. In the first branch, the fault data is subjected to continuous wavelet transform (CWT) to generate a scale spectrum, and high-frequency spatial texture features are extracted using a two-dimensional convolutional neural network (2D-CNN). The continuous wavelet transform uses the generalized Morse wavelet as the mother wavelet to convert one-dimensional time series data into a two-dimensional scale spectrum to capture the non-stationary transient features of the signal; the 2D-CNN network contains multiple cascaded convolutional layers, batch normalization layers and max pooling layers, with the kernel size decreasing layer by layer.
[0054] S5. In the second branch, the fault data is subjected to Discrete Wavelet Transform (DWT) to extract the reconstructed trend components, and long-term time-dependent features are extracted using a Bidirectional Long Short-Term Memory (BiLSTM) network. Discrete wavelet transform uses bioorthogonal wavelets to perform multi-level decomposition and reconstruction of the signal, separating the low-frequency trend component. To eliminate high-frequency noise interference and preserve gradual fault characteristics; the BiLSTM network is used to capture bidirectional time dependencies in the trend component.
[0055] Please see Figure 3 The modulus squared wavelet coefficient matrix is obtained by performing continuous wavelet transform on the fault data and multi-scale decomposition of the original time series data.
[0056] The wavelet coefficients are defined as follows:
[0057] in, This embodiment uses the generalized Morse wavelet (symmetric parameter) to represent the mother wavelet function. The preferred value is 3, time-frequency lumped parameter Preferably 20). Indicates complex conjugation. Parameter To indicate translation, this embodiment sets the sampling points for a single processing step. Regarding the sampling time of the voltage sensor The temperature sensor's sampling time is 1 second. Parameters Represented as a scale. Indicates the center frequency; Indicates the target frequency. The spectrum selection follows a logarithmic distribution. , This indicates the height of the final scale spectrum. In this invention, the final scale spectrum size is 44*1024. Furthermore, to avoid noise interference at the Nyquist level for the target frequency, it must satisfy… Thus, we obtained This ensures the fine resolution of the scaled spectrum in the frequency domain. The modulus-squared wavelet coefficient matrix can then be obtained:
[0058] The scale spectral map is input into a three-layer cascaded 2D-CNN network (containing a 3-layer convolution-batch normalization-ReLU-pooling structure), with the convolution kernels being as follows: , , Extracting high-frequency features It is used to capture sudden faults such as noise and periodicity; The discrete wavelet decomposition of the fault samples is performed using bioorthogonal wavelets. The specific decomposition process is as follows:
[0059] in, Indicates the number of decomposition layers. Indicates the first The approximation coefficients of the layer represent the low-frequency trend after signal decomposition; Indicates the first The detailed parameters of the layers mainly represent the high-frequency information of the signal. To capture the long-term variation information of the sensor's gradual faults and to prevent phase distortion of the waveform after discrete wavelet transform decomposition and reconstruction, this embodiment uses a decomposition layer of 10. Thus, the reconstructed time-domain trend information can be obtained:
[0060] The trend signal is input into a three-layer stacked BiLSTM network (each layer has 128 hidden units in this embodiment) to extract trend features. It is used to detect gradual faults such as drift. Concatenate the feature vectors: .
[0061] S6. The feature vectors extracted from the first and second branches are fused together, and the specific sensor fault type is output by the classifier to realize the fault diagnosis of the battery energy storage sensor.
[0062] The features are mapped to a size of [size missing] using a fully connected layer (two layers in this embodiment, with a dropout rate of 0.4 for the Dropout layer). The vector corresponds to the number of classification categories. This produces the raw logits used for probability calculation. .
[0063] Feature fusion refers to the process of combining feature vectors extracted by 2D-CNN. and the feature vector extracted by BiLSTM After concatenation and mapping through a fully connected layer, a Softmax classifier is used to output the fault probability distribution. ; in, This represents the original score for the i-th category. It is the natural base.
[0064] Supported fault types include bias fault (constant offset added to the output signal), jam fault (output stuck at a fixed value), noise fault (increased noise variance), periodic fault (periodic interference), drift fault (gradual offset), and combined fault (multiple superpositions).
[0065] The model training in this invention employs the cross-entropy loss function, with Adam being the preferred optimizer, a learning rate preferably between 0.0001 and 0.01, and a batch size of 128. The model training process is as follows: The cross-entropy loss is calculated on the training set. The network parameters are updated through backpropagation until the model converges; the data to be diagnosed is input into the trained model, and the fault type label is output.
[0066] In another embodiment of the present invention, a battery energy storage system sensor fault diagnosis system is provided. This system can be used to implement the above-mentioned battery energy storage system sensor fault diagnosis method. Specifically, the battery energy storage system sensor fault diagnosis system includes a data module, a positioning module, an extraction module, a fusion module, and a classification module.
[0067] The data module is used to acquire time-series data of the sensor cluster in the battery energy storage system, and to segment the time-series data using a sliding window to construct a sensor cluster data matrix. The positioning module is signal-connected to the data module and is used to receive the sensor cluster data matrix, perform hierarchical fault location and inspection HFLC, filter the set of suspected sensors through the global detection stage, and then lock the faulty sensor and extract the faulty data fragment through the local identification stage. An extraction module, signal-connected to the positioning module, is used to receive the fault data segment. The module includes a parallel time-frequency feature extraction first branch and a trend feature extraction second branch. The first branch performs continuous wavelet transform on the fault data segment to generate a scale spectrum and extracts high-frequency spatial texture features through a two-dimensional convolutional neural network to obtain a first feature vector. The second branch performs discrete wavelet transform on the fault data segment to extract and reconstruct trend components and extracts long-term time-dependent features through a bidirectional long short-term memory network to obtain a second feature vector. The fusion module is signal-connected to the extraction module and is used to receive the first feature vector and the second feature vector and perform fusion processing to obtain a fused feature vector. The classification module, which is signal-connected to the fusion module, is used to receive the fused feature vector, input it into the classifier, and output the sensor fault type to complete the sensor fault diagnosis.
[0068] This invention provides a terminal device comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment can be used in the operation of a sensor fault diagnosis method for a battery energy storage system, including: A time-series data set of sensor clusters in a battery energy storage system is acquired. This time-series data is segmented using a sliding window to construct a sensor cluster data matrix. Hierarchical fault location and inspection (HFLC) is performed on the sensor cluster data matrix. HFLC includes a global detection phase and a local identification phase. In the global detection phase, a set of suspected sensors is selected. In the local identification phase, the final faulty sensor is identified, and fault data fragments are extracted. A dual-stream heterogeneous neural network is constructed, comprising a first branch for parallel time-frequency feature extraction and a second branch for trend feature extraction. The fault data fragments are input into the first and second branches respectively. In the first branch, continuous wavelet transform is performed on the fault data fragments to generate a scale spectrum. A two-dimensional convolutional neural network is used to extract high-frequency spatial texture features from the scale spectrum to obtain a first feature vector. In the second branch, discrete wavelet transform is performed on the fault data fragments to extract and reconstruct trend components. A bidirectional long short-term memory network is used to extract long-term time-dependent features from the trend components to obtain a second feature vector. The first and second feature vectors are fused. The fused feature vector is input into a classifier, which outputs the sensor fault type, completing the sensor fault diagnosis.
[0069] Please see Figure 6 The terminal device is a computer device. In this embodiment, the computer device 60 includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When executed by the processor 61, the computer program 63 implements the battery energy storage system sensor fault diagnosis method of this embodiment. To avoid repetition, details are omitted here. Alternatively, when executed by the processor 61, the computer program 63 implements the functions of each model / unit in the battery energy storage system sensor fault diagnosis system of this embodiment. To avoid repetition, details are omitted here.
[0070] Computer device 60 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. Computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will understand that... Figure 6 This is merely an example of computer device 60 and does not constitute a limitation on computer device 60. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device may also include input / output devices, network access devices, buses, etc.
[0071] The processor 61 may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.
[0072] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the computer device 60.
[0073] Furthermore, the memory 62 may include both internal storage units of the computer device 60 and external storage devices. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.
[0074] Please see Figure 7 The terminal device is an electronic device 600, which is manifested in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including storage unit 620 and processing unit 610), a display unit 640, etc.
[0075] The storage unit stores program code, which can be executed by the processing unit 610 to perform the steps described in the method section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 610 can perform actions such as... Figure 1 The steps are shown in the figure.
[0076] Storage unit 620 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 6201 and / or cache memory 6202, and may further include a read-only memory (ROM) 6203.
[0077] Storage unit 620 may also include a program / utility 6204 having a set (at least one) program module 6205, such program module 6205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.
[0078] Bus 630 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the multiple bus structures.
[0079] Electronic device 600 can also communicate with one or more external devices 700 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 600, and / or with any device that enables electronic device 600 to communicate with one or more other computing devices (e.g., router, modem). This communication can be performed via input / output interface 650. Furthermore, electronic device 600 can also communicate with one or more networks (e.g., local area network, wide area network, and / or public network, such as the Internet) via network adapter 660. Network adapter 660 can communicate with other modules of electronic device 600 via bus 630. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.
[0080] Example 2 This invention also provides a storage medium, specifically a computer-readable storage medium, which is a memory device in a terminal device for storing programs and data. It is understood that the computer-readable storage medium here can include both built-in storage media in the terminal device and extended storage media supported by the terminal device; it can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). More specific examples of the computer-readable storage medium include: an electrical connection with one or more wires, a portable disk, a hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical fiber, portable compact disk read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.
[0081] Computer-readable storage media also include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium can also be any readable medium other than a readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, radio frequency, etc., or any suitable combination thereof.
[0082] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages such as Java and C++, and conventional procedural programming languages such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).
[0083] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the battery energy storage system sensor fault diagnosis method in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor to perform the following steps: A time-series data set of sensor clusters in a battery energy storage system is acquired. This time-series data is segmented using a sliding window to construct a sensor cluster data matrix. Hierarchical fault location and inspection (HFLC) is performed on the sensor cluster data matrix. HFLC includes a global detection phase and a local identification phase. In the global detection phase, a set of suspected sensors is selected. In the local identification phase, the final faulty sensor is identified, and fault data fragments are extracted. A dual-stream heterogeneous neural network is constructed, comprising a first branch for parallel time-frequency feature extraction and a second branch for trend feature extraction. The fault data fragments are input into the first and second branches respectively. In the first branch, continuous wavelet transform is performed on the fault data fragments to generate a scale spectrum. A two-dimensional convolutional neural network is used to extract high-frequency spatial texture features from the scale spectrum to obtain a first feature vector. In the second branch, discrete wavelet transform is performed on the fault data fragments to extract and reconstruct trend components. A bidirectional long short-term memory network is used to extract long-term time-dependent features from the trend components to obtain a second feature vector. The first and second feature vectors are fused. The fused feature vector is input into a classifier, which outputs the sensor fault type, completing the sensor fault diagnosis.
[0084] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0085] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0086] By employing a collaborative design of hierarchical fault localization and dual-stream heterogeneous diagnosis, this invention significantly reduces the computational burden on deep learning models while maintaining diagnostic accuracy. Compared to methods that directly perform end-to-end modeling of all sensor data, this invention reduces the scale of model input data through pre-layered localization, making the diagnostic process more suitable for online operating environments.
[0087] Please see Figure 4 and Figure 5 This invention is significantly superior to single CNN or BiLSTM architectures, demonstrating good engineering applicability.
[0088] In the actual testing of sensor fault diagnosis in battery energy storage systems, this invention conducts comparative tests on six types of faults—bias, jamming, noise, periodicity, drift, and combined faults—as well as normal conditions, under different signal-to-noise ratios ranging from 30dB to 70dB. Simultaneously, it benchmarks the performance against traditional diagnostic methods such as single 2D-CNN and single BiLSTM. The test samples cover full-condition operating data of voltage and temperature sensors, with a sample size exceeding 100,000.
[0089] In terms of positioning performance, the HFLC hierarchical fault location algorithm of this invention achieves a fault sensor positioning accuracy of over 98% under various signal-to-noise ratios. Compared with traditional methods without prior positioning, the positioning miss rate is reduced by 87% and the false judgment rate is reduced by 92%. Even in a low signal-to-noise ratio environment of 30dB, it can still effectively distinguish between operating condition fluctuations and real faults, accurately extract fault data segments, reduce the input data scale of subsequent models by more than 80%, and significantly reduce computational redundancy. The positioning time on a general industrial computer is only 0.2s / time, which meets the real-time requirements of online diagnosis.
[0090] In terms of diagnostic accuracy, this invention achieves a classification accuracy of 99.5%-99.9% within the common signal-to-noise ratio (SNR) range of 40-60dB in practical engineering applications, with a 99.9% accuracy rate for abrupt fault diagnosis and a 99.5% accuracy rate for gradual fault diagnosis. Even in an extremely low SNR environment of 30dB, the accuracy remains above 95%, while the single 2D-CNN method achieves only 78% accuracy for gradual fault diagnosis, and the single BiLSTM method achieves only 82% accuracy for abrupt fault diagnosis. Regarding the classification confusion matrix, this invention achieves a cross-misclassification rate of less than 0.5% for all six fault types, while the cross-misclassification rate of traditional single networks reaches as high as 15%. This fully demonstrates that the dual-stream heterogeneous neural network of this invention can effectively separate abrupt and gradual fault features, achieving refined diagnosis.
[0091] From an engineering applicability perspective, under the same hardware conditions, this invention reduces the overall diagnostic time by 65% compared to traditional end-to-end modeling methods, improves model training convergence speed by 40%, and has been verified through actual engineering deployments. This method can adapt to the operating condition fluctuations of battery energy storage systems and changes in the size of sensor clusters. Even in large-scale cluster scenarios with 100+ sensors, it can still maintain stable positioning and diagnostic performance, providing reliable technical support for the safe and stable operation of battery energy storage systems. Compared to traditional methods, it can effectively reduce the system operation risks caused by sensor failures and improve operation and maintenance efficiency.
[0092] In summary, this invention provides a method and system for diagnosing sensor faults in battery energy storage systems. Addressing industry pain points in sensor fault diagnosis for battery energy storage systems, it proposes a collaborative technical solution combining hierarchical localization and dual-stream heterogeneous diagnosis. This effectively solves the technical problems of difficulty in screening massive amounts of data, conflicts in multimodal fault feature extraction, and insufficient diagnostic robustness in low signal-to-noise ratio environments. The HFLC algorithm achieves accurate localization of faulty sensors and data segments, significantly reducing model input complexity. The dual-stream heterogeneous neural network performs differentiated feature extraction for abrupt and gradual faults, breaking the performance trade-off dilemma of traditional diagnosis. This invention achieves a localization accuracy of over 98% and a classification accuracy of 99.5%-99.9% at a signal-to-noise ratio of 40-60dB, significantly outperforming single CNN or BiLSTM architectures. Simultaneously, it reduces computational redundancy, meets online diagnostic requirements, and is suitable for the safety monitoring and maintenance of large-scale battery energy storage systems.
[0093] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0094] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0095] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.
[0096] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.
[0097] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0098] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0099] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random-access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.
[0100] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, 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.
[0101] 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.
[0102] 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.
[0103] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.
Claims
1. A method for diagnosing sensor faults in a battery energy storage system, characterized in that, Includes the following steps: S1. Obtain time-series data of the sensor cluster in the battery energy storage system, segment the time-series data using a sliding window, and construct a sensor cluster data matrix. S2. Perform hierarchical fault location and inspection (HFLC) on the sensor cluster data matrix. The hierarchical fault location and inspection includes a global detection stage and a local identification stage. In the global detection stage, a set of suspected sensors is selected. In the local identification stage, the final faulty sensor is locked and the faulty data fragment is extracted. S3. Construct a dual-stream heterogeneous neural network, which includes a first branch for parallel time-frequency feature extraction and a second branch for trend feature extraction, and input the fault data fragments into the first branch and the second branch respectively; S4. In the first branch, the fault data segment is subjected to continuous wavelet transform to generate a scale spectrum map, and the high-frequency spatial texture features in the scale spectrum map are extracted using a two-dimensional convolutional neural network to obtain a first feature vector. S5. In the second branch, the trend component is extracted and reconstructed by discrete wavelet transform of the fault data segment, and the long-term time-dependent features in the trend component are extracted by bidirectional long short-term memory network to obtain the second feature vector. S6. The first feature vector and the second feature vector are fused, and the fused feature vector is input into the classifier. The classifier outputs the sensor fault type to complete the sensor fault diagnosis.
2. The battery energy storage system sensor fault diagnosis method according to claim 1, characterized in that, In step S1, the sensor cluster includes a voltage sensor and a temperature sensor. The sampling frequency of the voltage sensor is 20Hz, and the sampling frequency of the temperature sensor is 1Hz. The length of the sliding window is 1024 sampling points, and the overlap rate of the sliding window is 50%.
3. The battery energy storage system sensor fault diagnosis method according to claim 1, characterized in that, In step S2, the specific process of the global detection phase is as follows: Calculate the energy of the difference sequence between the sensor and its neighboring sensors, and calculate the energy ratio based on the energy of the difference sequence. If the energy ratio exceeds a preset threshold, the corresponding sensor is recorded in the suspect sensor set. The sensor cluster data matrix is Z-score normalized, the covariance matrix and the average correlation coefficient between the sensor and other sensors are calculated, the correlation ratio is calculated based on the average correlation coefficient, and if the correlation ratio exceeds a preset threshold, the corresponding sensor is recorded in the suspect sensor set. Generate reference signals for the sensors in the suspected sensor set.
4. The battery energy storage system sensor fault diagnosis method according to claim 3, characterized in that, The preset threshold for the energy ratio is 2, and the preset threshold for the correlation ratio is 0.
3.
5. The battery energy storage system sensor fault diagnosis method according to claim 1, characterized in that, In step S2, the specific process of the local identification stage is as follows: Using the median of neighboring sensor readings as a reference signal, relative difference and cumulative energy difference are calculated based on the reference signal. If the relative difference exceeds a preset threshold or the cumulative energy difference exceeds a preset threshold, the corresponding sensor is identified as a faulty sensor, and the time series data of the faulty sensor is extracted as the fault data segment.
6. The battery energy storage system sensor fault diagnosis method according to claim 5, characterized in that, The preset threshold corresponding to the relative difference is 0.04, and the preset threshold corresponding to the cumulative energy difference is 5.
0.
7. The battery energy storage system sensor fault diagnosis method according to claim 1, characterized in that, In step S4, the continuous wavelet transform uses the generalized Morse wavelet as the mother wavelet, and the two-dimensional convolutional neural network is a three-layer cascaded convolution-batch normalization-ReLU-pooling structure, used to capture the abrupt fault characteristics of the sensor.
8. The battery energy storage system sensor fault diagnosis method according to claim 1, characterized in that, In step S5, the discrete wavelet transform uses biorthogonal wavelets to perform multi-layer decomposition and reconstruction of fault data segments, with a decomposition layer of 10; the bidirectional long short-term memory network is a three-layer stacked structure with 128 hidden units in each layer, used to capture the gradual fault characteristics of the sensor.
9. The battery energy storage system sensor fault diagnosis method according to claim 1, characterized in that, In step S6, the fusion process is feature vector concatenation; the classifier process is to input the concatenated feature vector into a fully connected layer for nonlinear mapping, and then calculate the fault probability and output the fault type through a Softmax layer. The fully connected layer has a 2-layer structure and the dropout rate of the Dropout layer is 0.
4. The fault types include bias fault, jamming fault, noise fault, periodic fault, drift fault and comprehensive fault.
10. A sensor fault diagnosis system for a battery energy storage system, characterized in that, include: The data module is used to acquire time-series data of the sensor cluster in the battery energy storage system, and to segment the time-series data using a sliding window to construct a sensor cluster data matrix. The positioning module is signal-connected to the data module and is used to receive the sensor cluster data matrix, perform hierarchical fault location and inspection HFLC, filter the set of suspected sensors through the global detection stage, and then lock the faulty sensor and extract the faulty data fragment through the local identification stage. An extraction module, signal-connected to the positioning module, is used to receive the fault data segment. The module includes a parallel time-frequency feature extraction first branch and a trend feature extraction second branch. The first branch performs continuous wavelet transform on the fault data segment to generate a scale spectrum and extracts high-frequency spatial texture features through a two-dimensional convolutional neural network to obtain a first feature vector. The second branch performs discrete wavelet transform on the fault data segment to extract and reconstruct trend components and extracts long-term time-dependent features through a bidirectional long short-term memory network to obtain a second feature vector. The fusion module is signal-connected to the extraction module and is used to receive the first feature vector and the second feature vector and perform fusion processing to obtain a fused feature vector. The classification module, which is signal-connected to the fusion module, is used to receive the fused feature vector, input it into the classifier, and output the sensor fault type to complete the sensor fault diagnosis.