Island detection method and device for energy storage system, and medium

Abstract

The invention discloses an energy storage system island detection method and device and a medium, and relates to the technical field of data processing. Comprising the following steps: extracting local key features of frequency domain features of a target energy storage system and a time sequence change rule of time domain features and dynamic features of the target energy storage system based on a hybrid model constructed by a convolutional neural network and a long-short term memory network; the local key features and the time sequence change rule are fused based on a full connection layer, a fused high-dimensional feature vector is obtained, and the high-dimensional feature vector is used for reflecting the operation state of the target energy storage system; the high-dimensional feature vectors are analyzed based on a classification model, a classification result of whether the operation state is an island state is obtained, and the hybrid model and the classification model are obtained through training based on a dynamic weight distribution mechanism. The island state under the condition that the matching degree of the load and the output power of the energy storage system is relatively high can be effectively identified through the hybrid model and the classification model trained by using the dynamic weight distribution mechanism, so that the accuracy of island detection is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, device and medium for detecting islanding in an energy storage system. Background Technology

[0002] With the rapid development of distributed energy resources (such as photovoltaic and wind power), energy storage systems are being used more and more widely in power systems. Energy storage systems can smooth out output fluctuations from distributed energy sources and improve energy utilization efficiency, but they also bring the safety hazard of islanding. Islanding detection, as a key technology to ensure the safe operation of energy storage systems, directly affects the stability of the power system and the safety of maintenance personnel.

[0003] Traditional islanding detection methods are mainly divided into two categories: passive and active. Passive methods detect islanding based on changes in electrical parameters such as voltage and frequency after a power outage. However, when the load and the energy storage system's output power are well-matched, the detection sensitivity is low, and false negatives are common. Active methods change electrical parameters by injecting disturbances into the system. While this can improve detection sensitivity, it can affect power quality, and the disturbance effect is unstable under complex load scenarios, resulting in a high false positive rate. Therefore, providing a method for detecting islanding in energy storage systems to improve the accuracy of islanding detection is a pressing issue that needs to be addressed. Summary of the Invention

[0004] This application provides a method, device, and medium for detecting islanding in an energy storage system, with the main purpose of improving the accuracy of islanding detection.

[0005] In a first aspect, embodiments of this application provide a method for detecting islanding in an energy storage system, the method comprising: A hybrid model based on convolutional neural networks and long short-term memory networks is used to extract local key features of the frequency domain characteristics of the target energy storage system, as well as the temporal variation patterns of the time domain and dynamic features of the target energy storage system. Based on the fully connected layer, the local key features and the temporal variation pattern are fused to obtain a fused high-dimensional feature vector, which is used to reflect the operating status of the target energy storage system. The high-dimensional feature vector is analyzed based on the classification model to obtain the classification result of whether the running state is an island state. The hybrid model and the classification model are trained based on a dynamic weight allocation mechanism.

[0006] In one implementation of this application, after obtaining the classification result of whether the running state is an island state, the method includes: If the classification result shows that the operating state is an islanded state, then the connection between the target energy storage system and the load is cut off based on the circuit breaker.

[0007] In one implementation of this application, before extracting the local key features of the frequency domain characteristics of the target energy storage system and the temporal variation law of the time domain characteristics of the target energy storage system using a hybrid model constructed based on convolutional neural networks and long short-term memory networks, the method includes: The target energy storage system's frequency domain characteristics, time domain characteristics, and dynamic characteristics are collected based on a combination of voltage and current sensors installed at the grid connection point of the target energy storage system. The time domain characteristics of the target energy storage system include voltage, current, and frequency within the target energy storage system. The frequency domain characteristics of the target energy storage system include harmonic components of voltage and current within the target energy storage system, with a sampling frequency of 10kHz. The dynamic characteristics include power change rate and load impedance change within the target energy storage system, with a sampling frequency of 10kHz.

[0008] In one implementation of this application, after acquiring the frequency domain characteristics, time domain characteristics, and dynamic characteristics of the target energy storage system based on a combination of voltage and current sensors installed at the grid connection point of the target energy storage system, the method includes: Outlier cleaning is performed on the frequency domain characteristics, time domain characteristics, and dynamic characteristics of the target energy storage system. The frequency domain characteristics, time domain characteristics, and dynamic characteristics of the target energy storage system after cleaning are standardized. The frequency domain features, time domain features, and dynamic features of the standardized target energy storage system are reduced in dimensionality to extract the local key features and the time-series variation patterns.

[0009] In one implementation of this application, the hybrid model constructed based on convolutional neural networks and long short-term memory networks extracts local key features of the frequency domain characteristics of the target energy storage system and the temporal variation law of the time domain characteristics of the target energy storage system, including: The local key information is extracted through the convolutional and pooling layers of the convolutional neural network; The temporal variation pattern is extracted through the hidden layers of the Long Short-Term Memory network.

[0010] In one implementation of this application, before extracting the local key features of the frequency domain characteristics of the target energy storage system and the temporal variation law of the time domain characteristics of the target energy storage system using a hybrid model constructed based on convolutional neural networks and long short-term memory networks, the method includes: An initial hybrid model was constructed based on convolutional neural networks and long short-term memory networks; An initial classification model was constructed based on the random forest algorithm; An initial weight is assigned to each training sample based on the classification difficulty corresponding to each sample in the training sample set. The training samples are the historical time-domain features, historical frequency-domain features, and historical dynamic features of the target energy storage system when different loads are connected to the target energy storage system. The classification difficulty is determined based on the load type and operating conditions corresponding to the training sample. The initial hybrid model and the initial classification model are trained based on the training samples after the initial weight allocation to obtain the classification result corresponding to each training sample. The weight of each training sample is adjusted according to the classification result, and the initial hybrid model and the classification model that have completed one round of training are trained again using the training samples with adjusted weights until the condition for stopping training is met, so as to obtain the hybrid model and the classification model.

[0011] In one implementation of this application, the historical time-domain features, the historical frequency-domain features, and the historical dynamic features are all obtained based on voltage and current sensors at the grid connection point of the target energy storage system. The historical time-domain features include historical voltage, historical current, and historical frequency within the target energy storage system. The frequency-domain features of the target energy storage system include historical harmonic components of historical voltage and historical current within the target energy storage system, with a sampling frequency of 10kHz. The historical dynamic features include historical power change rate and historical load impedance change within the target energy storage system, with a sampling frequency of 10kHz.

[0012] In one implementation of this application, before initially assigning weights to each training sample based on the classification difficulty corresponding to each sample in the training sample set, the method includes: Outlier cleaning is performed on the historical frequency domain features, the historical time domain features, and the historical dynamic features; The historical frequency domain features, historical time domain features, and historical dynamic features after cleaning are standardized. The standardized historical frequency domain features, standardized historical time domain features, and standardized historical dynamic features are dimensionality reduced so that the dimensionality-reduced historical frequency domain features, dimensionality-reduced historical time domain features, and dimensionality-reduced historical dynamic features can be used to construct the training sample set.

[0013] Secondly, embodiments of this application also provide an energy storage system islanding detection device, the device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: extract local key features of the frequency domain features of the target energy storage system and the temporal variation patterns of the time domain features and dynamic features of the target energy storage system based on a hybrid model constructed based on a convolutional neural network and a long short-term memory network; fuse the local key features and the temporal variation patterns based on a fully connected layer to obtain a fused high-dimensional feature vector, the high-dimensional feature vector being used to reflect the operating state of the target energy storage system; and analyze the high-dimensional feature vector based on a classification model to obtain a classification result indicating whether the operating state is an islanding state, wherein the hybrid model and the classification model are trained based on a dynamic weight allocation mechanism.

[0014] This application provides a method, device, and medium for detecting islanding in an energy storage system. The method includes: extracting local key features of the frequency domain characteristics of a target energy storage system and the temporal variation patterns of its time domain and dynamic features using a hybrid model constructed based on a convolutional neural network and a long short-term memory network; fusing the local key features and the temporal variation patterns using a fully connected layer to obtain a fused high-dimensional feature vector, which reflects the operating state of the target energy storage system; and analyzing the high-dimensional feature vector using a classification model to obtain a classification result indicating whether the operating state is islanded. The hybrid model and the classification model are trained based on a dynamic weight allocation mechanism. This method offers the following advantages: by using the hybrid model and the classification model trained with the dynamic weight allocation mechanism to check the islanding state of the target energy storage system, it can effectively identify islanding states where the load and the energy storage system's output power have a high degree of matching. Furthermore, islanding state detection based on the hybrid model and the classification model does not require introducing disturbances into the target system, thereby improving the accuracy of islanding detection. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart of an energy storage system islanding detection method provided in this application embodiment; Figure 2 This is a schematic diagram of the internal structure of an energy storage system islanding detection device provided in an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] To facilitate understanding of the embodiments of this application, the following terms are explained: Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Principal Component Analysis (PCA), and the island effect, which refers to the phenomenon that an energy storage system continues to supply power to surrounding loads even after being disconnected from the power grid. This application provides an energy storage system islanding detection method, device, and medium to achieve the following effect: providing an energy storage system islanding detection method to improve the accuracy of islanding detection.

[0018] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart illustrating an islanding detection method for an energy storage system provided in an embodiment of this application. Figure 1 As shown in the figure, the islanding detection method for an energy storage system provided in this application embodiment specifically includes the following steps: Step 101: Extract local key features of the frequency domain features of the target energy storage system and the temporal variation patterns of the time domain features and dynamic features of the target energy storage system using a hybrid model constructed based on convolutional neural networks and long short-term memory networks; In order to accurately output the classification results of the operating status of the target energy storage system, the local key features and the time-series change rules are extracted based on the hybrid model to obtain multimodal features. Then, the two multimodal features mentioned above are used as the input for classification, which improves the accuracy of the classification results.

[0020] Step 102: Based on the fully connected layer, the local key features and the temporal variation pattern are fused to obtain a fused high-dimensional feature vector, which is used to reflect the operating status of the target energy storage system; To input the extracted multimodal features into the classification model, the multimodal features are fused to obtain the fused high-dimensional feature vector. By collecting and fusing multimodal features, the operating status of the energy storage system can be comprehensively reflected, overcoming the limitations of single features in complex scenarios, improving detection accuracy, and solving the problem of low detection accuracy in existing methods.

[0021] Step 103: Analyze the high-dimensional feature vector based on the classification model to obtain the classification result of whether the running state is an island state. The hybrid model and the classification model are trained based on a dynamic weight allocation mechanism.

[0022] By introducing a dynamic weight allocation mechanism, the classification model pays more attention to samples that are difficult to classify during training, which improves the model's adaptability to complex load types and operating conditions, enhances the model's generalization ability, and solves the problem of poor generalization ability in existing methods.

[0023] The classification model trained by the dynamic weight allocation mechanism, combined with the high-dimensional feature vector fused with multimodal features, can effectively identify islanding states when the load and the output power of the energy storage system are well matched, thereby improving the accuracy of islanding state identification of the target energy storage system.

[0024] This application provides a method for islanding detection in energy storage systems, comprising: extracting local key features of the frequency domain characteristics of the target energy storage system and the temporal variation patterns of the time domain characteristics and dynamic features of the target energy storage system using a hybrid model constructed based on a convolutional neural network and a long short-term memory network; fusing the local key features and the temporal variation patterns based on a fully connected layer to obtain a fused high-dimensional feature vector, which reflects the operating state of the target energy storage system; and analyzing the high-dimensional feature vector based on a classification model to obtain a classification result indicating whether the operating state is an islanded state. The hybrid model and the classification model are trained based on a dynamic weight allocation mechanism. This method has the following advantages: by using the hybrid model and the classification model trained by the dynamic weight allocation mechanism to check the islanding state of the target energy storage system, it can effectively identify islanding states where the load and the output power of the energy storage system have a high degree of matching. Furthermore, islanding state detection based on the hybrid model and the classification model does not require introducing disturbances into the target system, thereby improving the accuracy of islanding detection.

[0025] As a refinement of the embodiments of this application, after obtaining the classification result of whether the operating state is an islanded state, the method may also adopt, but is not limited to, the following implementation methods, for example: if the classification result shows that the operating state is an islanded state, then the connection between the target energy storage system and the load is cut off based on the circuit breaker.

[0026] In some embodiments, when the islanding state prediction result received from the classification model indicates that islanding has occurred, the circuit breaker of the energy storage system is opened to disconnect the connection with the load. Specifically, the trained model is deployed to the control unit of the energy storage system to receive and process sensor data in real time. When islanding is detected, the control unit issues a command to open the circuit breaker of the energy storage system within 0.1 seconds, disconnecting the connection with the load.

[0027] As a refinement of the above embodiments, before extracting the local key features of the frequency domain features and the temporal variation law of the time domain features of the target energy storage system using a hybrid model constructed based on convolutional neural networks and long short-term memory networks, the method may also adopt, but is not limited to, the following implementation methods, for example: based on the combination of voltage and current sensors installed at the grid connection point of the target energy storage system, the frequency domain features, the time domain features, and the dynamic features of the target energy storage system are collected. The time domain features of the target energy storage system include the voltage, current, and frequency within the target energy storage system. The frequency domain features of the target energy storage system include the harmonic components of the voltage and current within the target energy storage system, with a sampling frequency of 10kHz. The dynamic features include the power change rate and load impedance change within the target energy storage system, with a sampling frequency of 10kHz.

[0028] To facilitate understanding of the data acquisition process described above, an energy storage system in an industrial and commercial park is selected as an example. This system integrates distributed energy sources such as photovoltaics and wind power, and its loads include various types of appliances such as motors, lighting equipment, and office appliances. Sensors installed at the grid connection point collect real-time data on time-domain characteristics such as voltage, current, and frequency; frequency-domain characteristics such as harmonic components of voltage and current; and dynamic characteristics such as power change rate and load impedance changes. The sampling frequency is 10kHz. Simultaneously, islanding status data under different load types and operating conditions is obtained by simulating grid disconnection.

[0029] As a refinement of the above embodiments, after collecting the frequency domain features, time domain features, and dynamic features of the target energy storage system based on the voltage and current sensors installed at the grid connection point of the target energy storage system, the following implementation methods can also be adopted, but are not limited to: performing outlier cleaning on the frequency domain features, time domain features, and dynamic features of the target energy storage system; standardizing the cleaned frequency domain features, time domain features, and dynamic features; and reducing the dimensionality of the standardized frequency domain features, time domain features, and dynamic features to extract the local key features and the time-series variation patterns using the reduced dimensionality frequency domain features, time domain features, and dynamic features.

[0030] In some embodiments, the preprocessing module preprocesses the collected feature data, including data cleaning (removing outliers and noise), standardization (converting features of different magnitudes to the same scale so that the mean of each feature is 0 and the standard deviation is 1), and feature dimensionality reduction (using principal component analysis (PCA) to reduce feature dimensionality, reduce computational complexity, and protect principal components with a cumulative contribution rate of 95%).

[0031] As a refinement of the above embodiments, when extracting the local key features of the frequency domain features and the temporal variation law of the time domain features of the target energy storage system using the hybrid model constructed based on the convolutional neural network and the long short-term memory network in step 101, the following implementation methods can also be adopted, but are not limited to: extracting the local key information through the convolutional layer and pooling layer of the convolutional neural network; and extracting the temporal variation law through the hidden layer of the long short-term memory network.

[0032] As a refinement of the above embodiments, before extracting the local key features of the frequency domain features and the temporal variation law of the time domain features of the target energy storage system from the hybrid model constructed based on the convolutional neural network and the long short-term memory network, the method may also adopt, but is not limited to, the following implementation methods, for example: constructing an initial hybrid model based on the convolutional neural network and the long short-term memory network; constructing an initial classification model based on the random forest algorithm; assigning initial weights to each training sample based on the classification difficulty corresponding to each sample in the training sample set, wherein the training samples are the historical time domain features, historical frequency domain features, and historical dynamic features of the target energy storage system when the target energy storage system is connected to different loads, and the classification difficulty is determined based on the load type and operating conditions corresponding to the training sample; training the initial hybrid model and the initial classification model based on the training samples after the initial weight assignment to obtain the classification result corresponding to each training sample; adjusting the weight of each training sample according to the classification result; and using the weight-adjusted training samples to train the initial hybrid model and the classification model after completing one round of training again until the condition for stopping training is met, thereby obtaining the hybrid model and the classification model.

[0033] In some embodiments, a CNN-LSTM hybrid model is constructed, where the CNN part contains two convolutional layers and two pooling layers, and the LSTM part contains two hidden layers. A random forest classification model is constructed, with 100 decision trees and a maximum depth of 10. Training and test sets are divided, with the training set accounting for 70% of the total data and the test set accounting for 30%. During training, the weights are dynamically adjusted based on the classification difficulty of the samples; for misclassified samples, the weight is increased by 10% in the next training round. After multiple rounds of iterative training, the model achieves optimal accuracy, recall, and precision on the test set.

[0034] In one possible implementation of this application embodiment, the historical time-domain features, the historical frequency-domain features, and the historical dynamic features are all obtained based on voltage and current sensors at the grid connection point of the target energy storage system. The historical time-domain features include historical voltage, historical current, and historical frequency within the target energy storage system. The historical frequency-domain features of the target energy storage system include historical harmonic components of historical voltage and historical current within the target energy storage system, with a sampling frequency of 10kHz. The historical dynamic features include historical power change rate and historical load impedance change within the target energy storage system, with a sampling frequency of 10kHz.

[0035] As a refinement of the above embodiments, before initially assigning weights to each training sample based on the classification difficulty corresponding to each sample in the training sample set, the method may also adopt, but is not limited to, the following implementation methods, for example: performing outlier cleaning on the historical frequency domain features, the historical time domain features, and the historical dynamic features; standardizing the cleaned historical frequency domain features, the cleaned historical time domain features, and the cleaned historical dynamic features; and reducing the dimensionality of the standardized historical frequency domain features, the standardized historical time domain features, and the standardized historical dynamic features, so as to construct the training sample set using the dimensionality-reduced historical frequency domain features, the dimensionality-reduced historical time domain features, and the dimensionality-reduced historical dynamic features.

[0036] The collection and preprocessing process of the training samples described above is the same as the data collection and preprocessing process in the application of this application. For clarity and brevity, the collection and preprocessing process of the training samples will not be described again in this embodiment. Please refer to the above content.

[0037] It is important to understand that this application, in addition to its application in islanding detection for energy storage systems, can also be applied to the field of islanding detection in microgrids. Islanding issues also exist in microgrids. When applying this application to microgrids, it is only necessary to adjust the range of feature acquisition and the sample data for model training according to the load characteristics and operating parameters of the microgrid. For example, by adding the output characteristics of distributed power sources in the microgrid, accurate detection of the islanding state of the microgrid can be achieved, thereby improving the safety and stability of microgrid operation.

[0038] In summary, the embodiments of this application can achieve the following effects, including: By using the hybrid model trained with a dynamic weight allocation mechanism and the classification model to perform islanding checks on the target energy storage system, islanding can be effectively identified when the load and the energy storage system's output power are well matched. Furthermore, islanding detection based on the hybrid and classification models does not require introducing disturbances into the target system, thereby improving the accuracy of islanding detection. The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide an energy storage system islanding detection device, the structure of which is as follows: Figure 2 As shown.

[0039] Figure 2 This is a schematic diagram of the internal structure of an energy storage system islanding detection device provided in an embodiment of this application. Figure 2 As shown, the device includes: At least one processor 201; And a memory 202 that is communicatively connected to at least one processor; The memory 202 stores instructions executable by at least one processor. These instructions are executed by at least one processor 201 to enable the processor 201 to: extract local key features of the frequency domain characteristics of the target energy storage system and the temporal variation patterns of the time domain and dynamic features of the target energy storage system using a hybrid model constructed based on a convolutional neural network and a long short-term memory network; fuse the local key features and the temporal variation patterns based on a fully connected layer to obtain a fused high-dimensional feature vector, which reflects the operating state of the target energy storage system; and analyze the high-dimensional feature vector based on a classification model to obtain a classification result indicating whether the operating state is an isolated state. The hybrid model and the classification model are trained based on a dynamic weight allocation mechanism.

[0040] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium stores computer-executable instructions, which are configured as follows: A hybrid model based on convolutional neural networks and long short-term memory networks extracts local key features of the frequency domain characteristics of a target energy storage system, as well as the temporal variation patterns of the time domain and dynamic features of the target energy storage system; the local key features and the temporal variation patterns are fused based on a fully connected layer to obtain a fused high-dimensional feature vector, which reflects the operating state of the target energy storage system; and a classification model is used to analyze the high-dimensional feature vector to obtain a classification result indicating whether the operating state is an isolated state. The hybrid model and the classification model are trained based on a dynamic weight allocation mechanism.

[0041] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0042] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0043] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0044] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0045] 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.

[0046] 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.

[0047] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0048] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0049] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0050] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0051] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for detecting islanding in an energy storage system, characterized in that, The method includes: A hybrid model based on convolutional neural networks and long short-term memory networks is used to extract local key features of the frequency domain characteristics of the target energy storage system, as well as the temporal variation patterns of the time domain and dynamic features of the target energy storage system. Based on the fully connected layer, the local key features and the temporal variation pattern are fused to obtain a fused high-dimensional feature vector, which is used to reflect the operating status of the target energy storage system. The high-dimensional feature vector is analyzed based on the classification model to obtain the classification result of whether the running state is an island state. The hybrid model and the classification model are trained based on a dynamic weight allocation mechanism.

2. The method for detecting islanding in an energy storage system according to claim 1, characterized in that, After obtaining the classification result of whether the running state is an island state, the method includes: If the classification result shows that the operating state is an islanded state, then the connection between the target energy storage system and the load is cut off based on the circuit breaker.

3. The method for detecting islanding in an energy storage system according to claim 1, characterized in that, Before extracting the local key features of the frequency domain characteristics of the target energy storage system and the temporal variation law of the time domain characteristics of the target energy storage system using a hybrid model constructed based on convolutional neural networks and long short-term memory networks, the method includes: The target energy storage system's frequency domain characteristics, time domain characteristics, and dynamic characteristics are collected based on a combination of voltage and current sensors installed at the grid connection point of the target energy storage system. The time domain characteristics of the target energy storage system include voltage, current, and frequency within the target energy storage system. The frequency domain characteristics of the target energy storage system include harmonic components of voltage and current within the target energy storage system, with a sampling frequency of 10kHz. The dynamic characteristics include power change rate and load impedance change within the target energy storage system, with a sampling frequency of 10kHz.

4. The method for detecting islanding in an energy storage system according to claim 3, characterized in that, After acquiring the frequency domain characteristics, time domain characteristics, and dynamic characteristics of the target energy storage system based on a combination of voltage and current sensors installed at the grid connection point of the target energy storage system, the method includes: Outlier cleaning is performed on the frequency domain characteristics, time domain characteristics, and dynamic characteristics of the target energy storage system. The frequency domain characteristics, time domain characteristics, and dynamic characteristics of the target energy storage system after cleaning are standardized. The frequency domain features, time domain features, and dynamic features of the standardized target energy storage system are reduced in dimensionality to extract the local key features and the time-series variation patterns.

5. The method for detecting islanding in an energy storage system according to claim 4, characterized in that, A hybrid model based on convolutional neural networks and long short-term memory networks extracts local key features of the frequency domain characteristics of the target energy storage system and the temporal variation patterns of the time domain characteristics of the target energy storage system, including: The local key information is extracted through the convolutional and pooling layers of the convolutional neural network; The temporal variation pattern is extracted through the hidden layers of the Long Short-Term Memory network.

6. A method for detecting islanding in an energy storage system according to any one of claims 1-5, characterized in that, Before extracting the local key features of the frequency domain characteristics of the target energy storage system and the temporal variation law of the time domain characteristics of the target energy storage system using a hybrid model constructed based on convolutional neural networks and long short-term memory networks, the method includes: An initial hybrid model was constructed based on convolutional neural networks and long short-term memory networks; An initial classification model was constructed based on the random forest algorithm; An initial weight is assigned to each training sample based on the classification difficulty corresponding to each sample in the training sample set. The training samples are the historical time-domain features, historical frequency-domain features, and historical dynamic features of the target energy storage system when different loads are connected to the target energy storage system. The classification difficulty is determined based on the load type and operating conditions corresponding to the training sample. The initial hybrid model and the initial classification model are trained based on the training samples after the initial weight allocation to obtain the classification result corresponding to each training sample. The weight of each training sample is adjusted according to the classification result, and the initial hybrid model and the classification model that have completed one round of training are trained again using the training samples with adjusted weights until the condition for stopping training is met, so as to obtain the hybrid model and the classification model.

7. The method for detecting islanding in an energy storage system according to claim 1, characterized in that, The historical time-domain features, historical frequency-domain features, and historical dynamic features are all obtained based on voltage and current sensors at the grid connection point of the target energy storage system. The historical time-domain features include historical voltage, historical current, and historical frequency within the target energy storage system. The frequency-domain features of the target energy storage system include historical harmonic components of historical voltage and historical current within the target energy storage system, with a sampling frequency of 10kHz. The historical dynamic features include historical power change rate and historical load impedance change within the target energy storage system, with a sampling frequency of 10kHz.

8. The method for detecting islanding in an energy storage system according to claim 7, characterized in that, Before assigning initial weights to each training sample based on the classification difficulty corresponding to each sample in the training sample set, the method includes: Outlier cleaning is performed on the historical frequency domain features, the historical time domain features, and the historical dynamic features; The historical frequency domain features, historical time domain features, and historical dynamic features after cleaning are standardized. The standardized historical frequency domain features, standardized historical time domain features, and standardized historical dynamic features are dimensionality reduced so that the dimensionality-reduced historical frequency domain features, dimensionality-reduced historical time domain features, and dimensionality-reduced historical dynamic features can be used to construct the training sample set.

9. An islanding detection device for an energy storage system, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform an energy storage system islanding detection method as described in any one of claims 1-8.

10. A computer storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed, they implement the energy storage system islanding detection method as described in any one of claims 1-8.

Images