Island detection method, device and equipment and storage medium

Abstract

The invention discloses an island detection method and device, equipment and a storage medium. The method comprises the following steps: acquiring basic information data in a power distribution network accessed to a photovoltaic power generation field; preprocessing the basic information data to obtain to-be-detected data; inputting the to-be-detected data into a pre-constructed island detection model based on an improved space-time diagram to obtain an island detection result of the power distribution network; wherein the island detection model comprises a graph convolutional network GCN model and a time feature extraction model; the GCN model is used for extracting spatial features of the to-be-detected data, and the time feature extraction model is used for extracting time features of the to-be-detected data. According to the method, the to-be-detected data in the power distribution network is input into the island detection model based on the improved space-time diagram to obtain the island detection result of the power distribution network, and the detection speed and precision can be improved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the technical field of island detection, and in particular to an island detection method, device, equipment and storage medium. Background Art

[0002] With the rapid integration of distributed generation (DG), distributed generation systems have attracted more and more attention. In traditional power systems, power generation is centralized and transmitted to users through transmission and distribution networks. The introduction of distributed generation, although it has brought many advantages and benefits, has also caused new challenges and problems in the power system, such as unplanned islanding caused by accidental disconnection of the circuit breaker (CB) connecting the main grid to the DG system. When unplanned islanding occurs, it may endanger the lives of grid line maintenance personnel and users, interfere with the normal closing of the grid, and damage distribution equipment and user equipment.

[0003] At present, research on island detection technology is ongoing, and these technologies are mainly divided into three categories: local technology, remote technology, and intelligent technology. Local technology has power quality problems. The detection speed of passive island detection methods is low, the detection time of hybrid methods is long, and the detection time of intelligent island detection technology is not enough, which has a great impact on power quality. It can be seen that the existing island detection technology still has shortcomings in terms of detection speed, accuracy, and impact on power quality. Summary of the invention

[0004] The present invention provides an island detection method, device, equipment and storage medium to solve the problem that the island detection technology in the prior art still has deficiencies in detection speed, accuracy and impact on power quality.

[0005] According to one aspect of the present invention, a method for detecting an island is provided, the method comprising:

[0006] Obtain basic information data in the distribution network connected to the photovoltaic power plant;

[0007] Preprocessing the basic information data to obtain data to be detected;

[0008] Inputting the data to be detected into a pre-built island detection model based on an improved space-time graph to obtain an island detection result of the distribution network;

[0009] Among them, the island detection model includes a graph convolutional network GCN model and a temporal feature extraction model; the graph convolutional network GCN model is used to extract the spatial features of the data to be detected, and the temporal feature extraction model is used to extract the temporal features of the data to be detected.

[0010] According to another aspect of the present invention, there is provided an island detection device, the device comprising:

[0011] An acquisition module is used to acquire basic information data in the distribution network connected to the photovoltaic power plant;

[0012] A processing module, used for preprocessing the basic information data to obtain data to be detected;

[0013] A detection module, used for inputting the data to be detected into a pre-built island detection model based on an improved spatiotemporal graph to obtain an island detection result of the distribution network;

[0014] Among them, the island detection model includes a graph convolutional network GCN model and a temporal feature extraction model; the graph convolutional network GCN model is used to extract the spatial features of the data to be detected, and the temporal feature extraction model is used to extract the temporal features of the data to be detected.

[0015] According to another aspect of the present invention, there is provided an electronic device, the electronic device comprising: at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the island detection method described in any embodiment of the present invention.

[0018] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the island detection method described in any embodiment of the present invention when executed.

[0019] An island detection method, device, equipment and storage medium of an embodiment of the present invention, the method includes: obtaining basic information data in a distribution network connected to a photovoltaic power plant; preprocessing the basic information data to obtain data to be detected; inputting the data to be detected into a pre-built island detection model based on an improved spatiotemporal graph to obtain an island detection result of the distribution network; wherein the island detection model includes a graph convolutional network GCN model and a time feature extraction model; the graph convolutional network GCN model is used to extract the spatial features of the data to be detected, and the time feature extraction model is used to extract the time features of the data to be detected. The method obtains the island detection result of the distribution network by inputting the data to be detected in the distribution network into the island detection model based on the improved spatiotemporal graph, which can improve the detection speed and accuracy; and the island phenomenon is detected by the model-based intelligent detection technology, and there is no need to actively inject a disturbance signal, so that the power quality will not be affected when the island phenomenon is detected, which solves the problem that the island detection technology in the prior art is still insufficient in terms of detection speed, accuracy and impact on power quality.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 A schematic diagram of a flow chart of an island detection method provided in Embodiment 1 of the present invention;

[0023] Figure 2 A schematic diagram of a flow chart of an island detection method provided in Embodiment 2 of the present invention;

[0024] Figure 3 A schematic diagram of the structure of an improved GCN module provided in an embodiment of the present invention;

[0025] Figure 4 A schematic diagram of the structure of an island detection model based on an improved space-time graph provided by an embodiment of the present invention;

[0026] Figure 5 A schematic diagram of the structure of an island detection device provided in Embodiment 3 of the present invention;

[0027] Figure 6 Schematic diagram of the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only an embodiment of a part of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should belong to the scope of protection of the present invention. It should be understood that the various steps recorded in the method implementation of the present invention can be performed in different orders and / or in parallel. In addition, the method implementation may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0029] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, any variations of the terms "including" and "having" are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] It should be noted that the modifications of "one" and "plurality" mentioned in the present invention are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0032] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only used for illustrative purposes, and are not used to limit the scope of these messages or information.

[0033] At present, there are still many shortcomings in local technology, remote technology and intelligent technology for island detection technology. Local technology includes active, passive and hybrid island detection technology. Active island detection methods such as Slip-Mode Frequency Shift (SMFS), Sliding Mode Voltage Shift (SVS) and Sandia Frequency Shift (SFS) have noise and power quality problems. Passive island detection methods such as undervoltage / overvoltage and frequency, phase jump and voltage harmonic measurement have large non-detection zones (NDZ) and low detection speed. Hybrid methods combine active and passive technologies. Although they can reduce the non-detection zone, they usually have a long detection time and are complex to execute. Remote technology requires a communication link and is more reliable for large-scale systems, but it is difficult to implement for small-scale and microgrid systems due to its complexity and high cost. Intelligent island detection technology is faster and more reliable than local and remote methods. The main advantages of smart technology are fast detection, low NDZ, high accuracy and less impact on the power quality of the DG power system.

[0034] Existing intelligent island detection technologies include hybrid island detection technology based on decision tree and SFS, methods based on active and reactive power, detection methods using naive Bayes classifiers, and detection methods for multiple DGs and single-phase inverter DGs based on artificial neural networks (ANN) and support vector machines (SVM). In addition, there are island detection technologies using intelligent algorithms such as fuzzy logic (FL), fuzzy positive feedback (PF), adaptive network fuzzy inference systems (ANFIS), and convolutional neural networks (CNN).

[0035] However, the current detection technology still has many problems in practical application. Existing local and remote island detection technologies often have shortcomings in detection speed, accuracy, non-detection area and impact on power quality. In order to solve these problems, the present invention proposes an island detection method, which is as follows:

[0036] Embodiment 1

[0037] Figure 1A flow chart of an island detection method provided in Embodiment 1 of the present invention is applicable to the detection of island phenomena occurring in a distribution network with a high photovoltaic penetration rate. The method can be performed by an island detection device, wherein the device can be implemented by software and / or hardware and is generally integrated on an electronic device. In this embodiment, the electronic device includes but is not limited to: computers and other devices.

[0038] like Figure 1 As shown, an island detection method provided in Embodiment 1 of the present invention includes the following steps:

[0039] S110: Acquire basic information data in the distribution network connected to the photovoltaic power plant.

[0040] Among them, the basic information data can be information in the distribution network.

[0041] In this embodiment, the distribution network can be a distribution network with a high photovoltaic penetration rate, that is, the photovoltaic power plants connected to the distribution network are greater than a certain proportion, and the specific proportion of photovoltaic power plants is not limited in this embodiment. Before performing island detection in this embodiment, it is necessary to first obtain basic information data of the distribution network connected to the photovoltaic power plant.

[0042] In one embodiment, each piece of data in the basic information data includes at least: voltage data, current data, power data, frequency data, operating status, location and number of each node in the distribution network, operating status of the branch, line connection information and timestamp data corresponding to each piece of data; wherein the line connection information includes a topological structure diagram of the distribution network, and the topological structure diagram is used to describe the connection status between each node, the impedance and admittance of the wires.

[0043] Among them, the voltage data may include phase voltage and line voltage. In addition to the voltage data of ordinary nodes, the voltage data may also include the voltage data of the point of common coupling (PCC), which is the connection point of more than one user load in the power system. The current data may be the current information of each node, such as the current flow of each branch. The power data may be the active power and reactive power of each node, such as the output data of photovoltaic power generation. The frequency data may be the frequency change of the distribution network, which is used to detect abnormal fluctuations in the distribution network. The operating status of the node and the operating status of the branch may include the closing or opening of the switch. The location and number of the node may be the location and number of all nodes in the distribution network. The line connection information may be the connection between each node, and the line connection information may be the topological structure diagram of the distribution network. The topological structure diagram may be used to describe the connection between each node, the impedance and admittance of the wire and other physical parameters. The timestamp data may be the timestamp marked for each collected data point, and the timestamp data may be used to ensure the timing of the data and ensure that the data is collected at multiple time points to construct a time series data set.

[0044] In this embodiment, when performing island detection on a photovoltaic high-penetration distribution network, the basic information data collected may include: voltage data, current data, power data, frequency data of each node in the distribution network, the operating status, location and number of the node, the operating status of the branch, branch connection information and the timestamp data corresponding to each data.

[0045] S120: pre-process the basic information data to obtain data to be detected.

[0046] The data to be detected may be a processed data set.

[0047] In this embodiment, the basic information data may be preprocessed, and the preprocessing may include denoising, missing data filling, data standardization and other processing to obtain the data to be detected. This embodiment does not limit the specific method of preprocessing.

[0048] S130, inputting the data to be detected into a pre-built island detection model based on an improved space-time graph to obtain an island detection result of the distribution network.

[0049] Among them, the island detection model includes a graph convolutional network GCN model and a temporal feature extraction model; the graph convolutional network GCN model is used to extract the spatial features of the data to be detected, and the temporal feature extraction model is used to extract the temporal features of the data to be detected.

[0050] Among them, the island detection model based on the improved spatiotemporal graph can be a model that can detect the island phenomenon, and the island detection model can include a graph convolutional network model (Graph Convolutional Networks, GCN) and a time feature extraction model. The GCN model can be used to extract the spatial features of the data to be detected, and the time feature extraction model can be used to extract the time features of the data to be detected. The island detection result can be used to determine the location and corresponding time of the island phenomenon in the distribution network.

[0051] In this embodiment, the data to be detected can be input into a pre-built island detection model to obtain an island detection result of the distribution network. The location and time of the island phenomenon in the distribution network can be determined through the island detection result.

[0052] This embodiment can pre-train the island detection model. Before training, it is necessary to collect relevant basic information data for building and training the model to ensure that it can accurately detect the island phenomenon. Exemplarily, when collecting data, a device with high precision and fast response characteristics can be used to monitor the basic information data in the distribution network in real time, such as smart meters and data recorders. After collecting the basic information data, the data can be transmitted to the central data storage system through a reliable communication network, and real-time storage and regular backup can be performed. When pre-processing the basic information data, it can include denoising, missing data filling, and data standardization. For example, signal processing technology is used to remove noise in the data and improve data quality; interpolation, regression analysis and other methods are used to fill in missing values ​​in the data to ensure the integrity of the data set; the data is standardized to eliminate the influence of different dimensions and units on model training.

[0053] After preprocessing the data, data annotation and data set division are required, including:

[0054] Labeling of island events in the data set: Analyze historical data to identify and label the time when island events occur. These labels can be used for supervised learning of the model.

[0055] Classify normal and abnormal states: classify data into normal operating states and abnormal states (including island phenomena);

[0056] Divide the training set: used for model training, including data in normal and abnormal states;

[0057] Divide the test set: reserve a portion of the data for testing and validating the model to ensure the generalization ability of the model.

[0058] An island detection method provided in the first embodiment of the present invention includes: obtaining basic information data in a distribution network connected to a photovoltaic power plant; preprocessing the basic information data to obtain data to be detected; inputting the data to be detected into a pre-built island detection model based on an improved spatiotemporal graph to obtain an island detection result of the distribution network; wherein the island detection model includes a graph convolutional network GCN model and a time feature extraction model; the graph convolutional network GCN model is used to extract the spatial features of the data to be detected, and the time feature extraction model is used to extract the time features of the data to be detected. The method obtains the island detection result of the distribution network by inputting the data to be detected in the distribution network into the island detection model based on the improved spatiotemporal graph, which can improve the detection speed and accuracy; and the island phenomenon is detected by the model-based intelligent detection technology, and there is no need to actively inject a disturbance signal, so that the power quality will not be affected when the island phenomenon is detected, which solves the problem that the island detection technology in the prior art is still insufficient in terms of detection speed, accuracy and impact on power quality.

[0059] Embodiment 2

[0060] Figure 2 This is a flow chart of an island detection method provided by Embodiment 2 of the present invention. Embodiment 2 is optimized on the basis of the above embodiments. For details not yet fully described in this embodiment, please refer to Embodiment 1.

[0061] like Figure 2 As shown, a method for detecting an island provided by Embodiment 2 of the present invention includes the following steps:

[0062] S210: Acquire basic information data in the distribution network connected to the photovoltaic power plant.

[0063] S220: Preprocess the basic information data to obtain data to be detected.

[0064] S230 , extracting spatial features of the data to be detected by using a GCN model to obtain a first output feature quantity.

[0065] The first output feature quantity may be a hidden feature quantity output by the GCN model.

[0066] In this embodiment, the spatial features of the data to be detected can be extracted through the GCN model to obtain the first output feature quantity.

[0067] In one embodiment, the GCN model includes an improved GCN module. Accordingly, the spatial feature extraction of the data to be detected through the GCN model includes: combining the distance correction matrix, step correction matrix and spatial attention matrix in the improved GCN module to extract spatial features of the data to be detected; wherein the distance correction matrix refers to a matrix for correcting the distance between each node in the distribution network, the step correction matrix refers to a matrix for correcting the step size between each node in the distribution network, and the spatial attention matrix includes the mutual influence between each node in the distribution network.

[0068] Among them, the improved GCN module can be a module in the GCN model. The improved GCN module can include a distance correction matrix, a step correction matrix and a spatial attention matrix. The distance correction matrix can be a matrix for correcting the distance between each node in the distribution network. The step correction matrix can be a matrix for correcting the step size between each node in the distribution network. The spatial attention matrix can include the mutual influence between each node in the distribution network.

[0069] In this embodiment, the distance correction matrix, step correction matrix and spatial attention matrix in the improved GCN module can be used to extract spatial features of the data to be detected. By introducing the distance correction matrix, step correction matrix and spatial attention mechanism, this embodiment can enhance the understanding of the relationship between nodes, improve the accuracy and robustness of island detection, effectively reduce the non-detection zone (NDZ), and ensure reliable detection under various operating conditions.

[0070] For example, Figure 3 A schematic diagram of the structure of an improved GCN module provided by an embodiment of the present invention is shown in FIG. Figure 3 As shown, the improvements of the improved GCN module include a distance correction matrix D, a step correction matrix P, and a spatial attention matrix S'. The distance correction matrix D, the step correction matrix P, and the spatial attention matrix S' are fused to obtain the final feature representation, which can comprehensively consider the influence of various relationships between nodes. Since the distance between different nodes in the distribution network (such as substations, distributed energy access points) plays an important role in their mutual influence, the distance correction matrix of this embodiment can adjust its influence according to the shortest path distance between the nodes, ensuring that the smaller impact of distant nodes on system stability is taken into account when performing island detection. This is particularly important for accurately detecting possible islanding phenomena in distribution networks, because islanding phenomena may occur at distant distributed energy access points.

[0071] The distance correction matrix of this embodiment can be used to adjust and normalize the distance influence between nodes, so that nodes with closer distances have a greater weight influence on each other; the step correction matrix can be used to consider the complexity of the connection path between nodes, and more accurately describe the connection relationship between nodes by adjusting the step size; the spatial attention matrix can dynamically adjust the mutual influence weights between nodes by learning the relationship between node features, so as to better capture the subtle changes and complex relationships between nodes.

[0072] S240: Input the first output feature quantity into a time feature extraction model to extract the time feature, and obtain an island detection result of the distribution network.

[0073] In this embodiment, the first output feature quantity output by the GCN model can be input into the time feature extraction model to extract the time feature, so as to obtain the island detection result of the distribution network.

[0074] The second embodiment of the present invention provides an island detection method, which specifically inputs the data to be detected into a pre-built island detection model based on an improved space-time graph to obtain the island detection result of the distribution network, including: extracting spatial features of the data to be detected through a GCN model to obtain a first output feature quantity; inputting the first output feature quantity into a time feature extraction model to extract time features to obtain the island detection result of the distribution network. This method uses different models to extract spatial features and time features of the data to be detected, respectively, and can accurately capture the complex spatial relationships and time series characteristics between different nodes in the distribution network.

[0075] Based on the above embodiment, a variant embodiment of the above embodiment is proposed. It should be noted that in order to make the description concise, only the differences from the above embodiment are described in the variant embodiment.

[0076] In one embodiment, the element D(i,j) of the distance correction matrix D is:

[0077]

[0078] Among them, d(v i ,v j ) represents the node v in the distribution network i and v j The shortest path distance between max It represents the maximum value of the shortest path distance between nodes in the distribution network;

[0079] The step correction matrix is ​​used to reflect the connection complexity between each node. The element P(i, j) of the step correction matrix P is:

[0080]

[0081] Among them, P(v i ,v j ) represents node v i and v j The minimum step length between max It represents the maximum value of the minimum step length between nodes in the distribution network;

[0082] The attention matrix S is calculated by the following formula:

[0083] S=W1·σ((C t W2)W3(C t W4) T +b)

[0084] Among them, σ represents the activation function, C t is the hidden feature matrix obtained by transforming the feature set of each node in the distribution network at time t, W1, W2, W3, W4 are parameter matrices, b is the bias vector, and T represents transposition;

[0085] The normalized spatial attention matrix S' is calculated by the following formula:

[0086]

[0087] Among them, S' is an N×N matrix, which represents the influence of each node on other nodes; S is the attention matrix, and e is a natural constant.

[0088] The feature set of each node at time t may be obtained by processing the data to be detected.

[0089] In this embodiment, the improvement of GCN includes defining a distance correction matrix and a step correction matrix. The step correction matrix can reflect the complexity of the connection between nodes and take into account the complexity of the path of power transmission. This is particularly important for evaluating the impact of distributed energy access points on the entire distribution network. By accurately modeling these complex paths, the model can better predict the power distribution and stability in the island state. Improvements to GCN also include improvements to the dynamic weight allocation and feature extraction in the spatial attention mechanism. The spatial attention mechanism can dynamically allocate the influence weights between different nodes and automatically learn the interactions between nodes in a data-driven manner. The spatial attention mechanism can capture the subtle changes and complex relationships between each node, and improve the accuracy of island detection, especially in the face of data noise or anomalies. In the improved GCN model, the spatial attention mechanism is used to study the influence between road network nodes in the distribution network. By observing the input data, the influence between all nodes is calculated.

[0090] In one embodiment, the distance correction matrix, the step correction matrix and the spatial attention matrix in the improved GCN module are combined to extract spatial features of the data to be detected, including:

[0091] The data to be detected is converted into the input of the hidden layer of the GCN model, and then the spatial features of the data to be detected are extracted by the following formula:

[0092] H l =σ(((A k ⊙D⊙P⊙S′)H l-1 W l-1 ))

[0093] Among them, σ represents the activation function, A is the adjacency matrix, D is the distance correction matrix, P is the step correction matrix, S' is the spatial attention matrix, W is the learnable parameter matrix, H represents the output of the lth hidden layer, and k represents the number of hidden layers of the GCN model.

[0094] In this embodiment, the data to be detected can be converted into a feature matrix X that can be input into the GCN model, and the feature matrix is ​​converted as the input H of the hidden layer of the GCN model, and the spatial feature of H is extracted by the above formula to finally obtain the first output feature quantity Y. This formula can represent the calculation of the graph convolution layer l, which contains multiple layers of convolution operations to extract spatial features. Among them, the feature matrix X is an N×F feature matrix, which contains the feature information of all nodes in the distribution network, and Y is an N×P output matrix, which represents P output features.

[0095] In the GCN model, Fourier transform and convolution can also be performed to extract spatial features in the graph using the Fourier transform of the graph Laplace matrix L, which is defined as:

[0096]

[0097] Where D is the diagonal matrix, A is the adjacency matrix, I is the identity matrix, and the convolution kernel g is approximated using Chebyshev polynomials. θ :

[0098]

[0099] Among them, θ k is the Chebyshev coefficient vector, Λ is the Laplace eigenvalue matrix, T k is the Chebyshev polynomial, k is the number of hidden layers of the GCN model. This method is more computationally efficient and can capture the spatial features in the graph.

[0100] In this embodiment, the input of GCN includes the feature matrix and adjacency matrix of the nodes in the distribution network. The feature matrix represents the attributes of each node (such as voltage, current, power, etc.), and the adjacency matrix represents the connection relationship between nodes. Through the propagation of multiple layers of GCN, the high-order features of the nodes can be gradually extracted. GCN extracts features by weighted summation of adjacent node features, and the weight is determined by the connection relationship between nodes. Each layer of GCN contains an activation function to enhance nonlinear characteristics.

[0101] In one embodiment, the inputting the first output feature quantity into a time feature extraction model for time feature extraction to obtain the island detection result of the distribution network includes: adding a position coding vector to the first output feature quantity through position coding in the time feature extraction model; performing time feature extraction on the first output feature quantity after adding the position coding vector through an encoder and a decoder in the time feature extraction model to obtain the island detection result of the distribution network.

[0102] In this embodiment, the position coding vector can be added to the first output feature quantity through the position coding in the time feature extraction model, and then the time feature extraction is performed on the first output feature quantity after adding the position coding vector through the encoder and decoder to obtain the island detection result of the distribution network. This embodiment is aimed at the special needs of the photovoltaic high penetration distribution network, and constructs an island detection model with powerful spatiotemporal feature extraction capabilities. Through the combination of position coding, encoder and decoder, it can process complex time series data and accurately reflect the instability and volatility of photovoltaic power generation. In the case of large fluctuations in the output of the photovoltaic system, it can still ensure efficient and accurate detection of island phenomena, thereby improving the safety and stability of the distribution network.

[0103] For example, Figure 4 A schematic diagram of the structure of an island detection model based on an improved spatiotemporal graph provided by an embodiment of the present invention is shown in FIG. Figure 4 As shown, the GCN model may include two identical improved GCN modules, and the distribution network spatiotemporal data may be the data to be detected. In this embodiment, the temporal feature extraction model may be a Transformer model, and the Transformer model includes position encoding, an encoder (Encoder) and a decoder (Decoder). In the Transformer model, since there is no loop or convolution structure, the model cannot directly perceive the timing of the data, so position encoding is introduced to maintain the time series information of the input data, and the position encoding vector is calculated by the following formula:

[0104]

[0105] Among them, pos is the position index, i is the dimension index, and d is the vector dimension. By combining the position encoding vector with the first output feature value, the position information of the time series can be retained.

[0106] The encoder can be composed of N identical sub-layers, each of which includes two sub-layers: Multi-Head Attention Mechanism and Feed Forward Network. Before the data is input into the encoder, positional encoding is required to clarify the relative position relationship of each node.

[0107] Multi-head self-attention mechanism: The multi-head self-attention mechanism is used to capture the long-range dependencies of time series data, and calculates the correlation of different subspaces in parallel through multiple heads. The first output feature after adding the position encoding vector can be divided into three matrices, the query matrix Q, the key matrix K, and the value matrix V. These three matrices are used as the input of the multi-head self-attention mechanism, and the output of different heads is obtained through linear transformation:

[0108] head i =Attention(QW i Q ,KW i K ,VW i V )

[0109] Where Q, K, and V represent the query matrix, key matrix, and value matrix, respectively. i Q , W i K , W i V are the transformation matrices of the query, key, and value of the i-th head respectively.

[0110] After the outputs of all heads are connected, the final output is obtained through linear transformation:

[0111] MultiHead(Q,K,V)=Concat(head1,...,head h )W O

[0112] Among them, W O is the final output transformation matrix, and Concat represents the concatenation operation.

[0113] Feedforward neural network: Feedforward neural network is used for nonlinear transformation, including two linear mappings and an activation function ReLU:

[0114] I hidden =Linear(ReLU(Linear(I attention )))

[0115] In this embodiment, each sublayer is followed by layer normalization and residual connection to accelerate training and stabilize the network.

[0116] The decoder can be composed of N layers, each of which includes three sublayers: masked multi-head self-attention mechanism, multi-head encoder-decoder attention mechanism, and feed-forward neural network. Like the encoder part, the decoder part also has layer normalization and residual connections. Masked multi-head self-attention mechanism In the decoder, in order to ensure the causal relationship of the generated sequence, the masking operation is used to block out the future moment information. Multi-head encoder-decoder attention mechanism The decoder uses the information output by the encoder as context when generating each step to help generate the correct output.

[0117] In this embodiment, the high-order node features output by the GCN module are used as the input of the Transformer, and the temporal features and global dependencies are further extracted through the Transformer. The final output of the Transformer model is the feature representation at each moment, which can be input into the classification layer (such as the fully connected layer) to classify the island state.

[0118] The embodiments of the present invention provide several specific implementation methods based on the technical solutions of the above embodiments.

[0119] As a specific implementation method of this embodiment, the following steps are included:

[0120] Step 1: Determine the data type and range required for island detection in PV high-penetration distribution networks, collect data and perform data preprocessing, label the data and divide the data sets;

[0121] Step 2: Build an improved graph convolutional network (GCN) model, in which the GCN module in the GCN model combines the distance correction matrix, the step correction matrix and the spatial attention mechanism to enhance the ability to capture the complex relationships between different nodes in the distribution network;

[0122] Step 3: Propose an island detection and identification model based on the improved spatiotemporal graph Transformer. On the basis of the improved GCN proposed in step 2, the Transformer model is introduced to process the temporal characteristics of the distribution network data. The model consists of a position encoder, an encoder, and a decoder.

[0123] Step 4: Offline training is performed on the constructed island detection and identification model based on the improved spatiotemporal graph Transformer, aiming to optimize the model parameters to ensure that the model can accurately identify the island state of the distribution network, and to evaluate the performance of the model.

[0124] Use the training set prepared in step 1, including data of normal state and abnormal state (such as island events). The training data should include rich scenarios and states to improve the generalization ability of the model. Select the cross-entropy loss function for the classification task to measure the difference between the model output and the true label. The expression of the cross-entropy loss function is as follows:

[0125]

[0126] Where C is the total number of categories of island events, y i is the indicator function of the i-th class of the true label, p i is the probability of the i-th class predicted by the model.

[0127] In this embodiment, the Adam optimizer can be used to update the model parameters to minimize the loss function, set the learning rate to 0.0002, and the batch size to 512 to ensure stable convergence of the training process. The training process is: input training data, the model calculates the output through forward propagation, evaluates the difference between the output and the true label through the loss function, and then updates the model parameters through back propagation. The training loss and validation loss should be monitored during the training process to prevent overfitting. Finally, the reserved test set is used to evaluate the model performance to ensure that the model has good predictive ability on unseen data.

[0128] The island detection model of this embodiment can detect the island state in real time and quickly, reducing the impact that the traditional detection method may have on the normal operation of the power grid. Since there is no need to actively inject disturbance signals during the detection process, it will not cause power quality problems. At the same time, the model's efficient detection capability can also respond to island events in a timely manner, avoiding safety hazards caused by delayed detection, and improving the operational reliability of the power system.

[0129] Embodiment 3

[0130] Figure 5 This is a structural schematic diagram of an island detection device provided in Example 3 of the present invention. The device can be used to detect islanding phenomena occurring in a distribution network with a high photovoltaic penetration rate. The device can be implemented by software and / or hardware and is generally integrated on an electronic device.

[0131] like Figure 5 As shown, the device comprises:

[0132] An acquisition module 310 is used to acquire basic information data in a distribution network connected to a photovoltaic power plant;

[0133] The processing module 320 is used to pre-process the basic information data to obtain data to be detected;

[0134] A detection module 330 is used to input the data to be detected into a pre-built island detection model based on an improved spatiotemporal graph to obtain an island detection result of the distribution network;

[0135] Among them, the island detection model includes a graph convolutional network GCN model and a temporal feature extraction model; the graph convolutional network GCN model is used to extract the spatial features of the data to be detected, and the temporal feature extraction model is used to extract the temporal features of the data to be detected.

[0136] This embodiment provides an island detection device, including: an acquisition module, used to acquire basic information data in a distribution network connected to a photovoltaic power plant; a processing module, used to pre-process the basic information data to obtain data to be detected; a detection module, used to input the data to be detected into a pre-built island detection model based on an improved spatiotemporal graph to obtain an island detection result of the distribution network; wherein the island detection model includes a graph convolutional network GCN model and a time feature extraction model; the graph convolutional network GCN model is used to extract the spatial features of the data to be detected, and the time feature extraction model is used to extract the time features of the data to be detected. By inputting the data to be detected in the distribution network into the island detection model based on the improved spatiotemporal graph to obtain the island detection result of the distribution network, the detection speed and accuracy can be improved; and by detecting the island phenomenon through the model-based intelligent detection technology, there is no need to actively inject a disturbance signal, so that the power quality will not be affected when the island phenomenon is detected, which solves the problem that the island detection technology in the prior art is still insufficient in terms of detection speed, accuracy and impact on power quality.

[0137] Furthermore, the detection module 330 is specifically configured to:

[0138] Performing spatial feature extraction on the data to be detected by using a GCN model to obtain a first output feature quantity;

[0139] The first output feature quantity is input into a time feature extraction model to perform time feature extraction to obtain an island detection result of the distribution network.

[0140] Further, the GCN model includes an improved GCN module. Accordingly, the spatial feature extraction of the data to be detected by using the GCN model includes:

[0141] Combine the distance correction matrix, the step correction matrix and the spatial attention matrix in the improved GCN module to extract spatial features of the data to be detected;

[0142] Among them, the distance correction matrix refers to the matrix that corrects the distance between each node in the distribution network, the step correction matrix refers to the matrix that corrects the step size between each node in the distribution network, and the spatial attention matrix includes the mutual influence between each node in the distribution network.

[0143] Furthermore, the element D(i,j) of the distance correction matrix D is:

[0144]

[0145] Among them, d(v i ,v j ) represents the node v in the distribution network i and v j The shortest path distance between max It represents the maximum value of the shortest path distance between nodes in the distribution network;

[0146] The step correction matrix is ​​used to reflect the connection complexity between each node. The element P(i, j) of the step correction matrix P is:

[0147]

[0148] Among them, P(v i ,v j ) represents node v i and v j The minimum step length between max It represents the maximum value of the minimum step length between nodes in the distribution network;

[0149] The attention matrix S is calculated by the following formula:

[0150] S=W1·σ((C t W2)W3(C t W4) T +b)

[0151] Among them, σ represents the activation function, C t is the hidden feature matrix obtained by transforming the feature set of each node in the distribution network at time t, W1, W2, W3, W4 are parameter matrices, b is the bias vector, and T represents transposition;

[0152] The normalized spatial attention matrix S' is calculated by the following formula:

[0153]

[0154] Among them, S' is an N×N matrix, which represents the influence of each node on other nodes; S is the attention matrix, and e is a natural constant.

[0155] Furthermore, the distance correction matrix, step correction matrix and spatial attention matrix in the improved GCN module are combined to extract spatial features of the data to be detected, including:

[0156] The data to be detected is converted into the input of the hidden layer of the GCN model, and then the spatial features of the data to be detected are extracted by the following formula:

[0157] H l =σ(((A k ⊙D⊙P⊙S′)H l-1 W l-1 ))

[0158] Among them, σ represents the activation function, A is the adjacency matrix, D is the distance correction matrix, P is the step correction matrix, S' is the spatial attention matrix, W is the learnable parameter matrix, H represents the output of the lth hidden layer, and k represents the number of hidden layers of the GCN model.

[0159] Furthermore, the step of inputting the first output feature quantity into a time feature extraction model to extract the time feature to obtain the island detection result of the distribution network includes:

[0160] Adding a position coding vector to the first output feature quantity by using the position coding in the time feature extraction model;

[0161] The encoder and decoder in the time feature extraction model are used to extract the time features of the first output feature quantity after adding the position encoding vector to obtain the island detection result of the distribution network.

[0162] Furthermore, each piece of data in the basic information data includes at least: voltage data, current data, power data, frequency data, operation status, location and number of each node in the distribution network, operation status of the branch, branch connection information and timestamp data corresponding to each piece of data;

[0163] The network connection information includes a topological structure diagram of the distribution network, and the topological structure diagram is used to describe the connection status between nodes, impedance and admittance of wires.

[0164] The above-mentioned island detection device can execute the island detection method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0165] Embodiment 4

[0166] Figure 6A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0167] like Figure 6 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0168] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0169] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as an island detection method.

[0170] In some embodiments, the island detection method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the island detection method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the island detection method in any other appropriate manner (e.g., by means of firmware).

[0171] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0172] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0173] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0174] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0175] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0176] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0177] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0178] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An island detection method, characterized in that: The method comprises: Obtain basic information data in the distribution network connected to the photovoltaic power plant; Preprocessing the basic information data to obtain data to be detected; Inputting the data to be detected into a pre-built island detection model based on an improved space-time graph to obtain an island detection result of the distribution network; Among them, the island detection model includes a graph convolutional network GCN model and a temporal feature extraction model; the graph convolutional network GCN model is used to extract the spatial features of the data to be detected, and the temporal feature extraction model is used to extract the temporal features of the data to be detected.

2. The method according to claim 1, characterized in that The step of inputting the data to be detected into a pre-built island detection model based on an improved space-time graph to obtain the island detection result of the distribution network includes: Performing spatial feature extraction on the data to be detected by using a GCN model to obtain a first output feature quantity; The first output feature quantity is input into a time feature extraction model to perform time feature extraction to obtain an island detection result of the distribution network.

3. The method according to claim 2, characterized in that The GCN model includes an improved GCN module. Accordingly, the spatial feature extraction of the to-be-detected data by using the GCN model includes: Combine the distance correction matrix, the step correction matrix and the spatial attention matrix in the improved GCN module to extract spatial features of the data to be detected; Among them, the distance correction matrix refers to the matrix that corrects the distance between each node in the distribution network, the step correction matrix refers to the matrix that corrects the step size between each node in the distribution network, and the spatial attention matrix includes the mutual influence between each node in the distribution network.

4. The method according to claim 3, characterized in that: The element D(i,j) of the distance correction matrix D is: Among them, d(v i ,v j ) represents the node v in the distribution network i and v j The shortest path distance between max It represents the maximum value of the shortest path distance between nodes in the distribution network; The step correction matrix is ​​used to reflect the connection complexity between each node. The element P(i, j) of the step correction matrix P is: Among them, P(v i ,v j ) represents node v i and v j The minimum step length between max It represents the maximum value of the minimum step length between nodes in the distribution network; The attention matrix S is calculated by the following formula: S=W1·σ((C t W2)W3(C t W4) T +b) Among them, σ represents the activation function, C t is the hidden feature matrix obtained by transforming the feature set of each node in the distribution network at time t, W1, W2, W3, W4 are parameter matrices, b is the bias vector, and T represents transposition; The normalized spatial attention matrix S' is calculated by the following formula: Among them, S' is an N×N matrix, which represents the influence of each node on other nodes; S is the attention matrix, and e is a natural constant.

5. The method according to claim 3, characterized in that: The step of combining the distance correction matrix, the step correction matrix and the spatial attention matrix in the improved GCN module to extract spatial features of the data to be detected includes: The data to be detected is converted into the input of the hidden layer of the GCN model, and then the spatial features of the data to be detected are extracted by the following formula: H l =σ(((A k ☉D⊙P☉S')H l-1 W l-1 )) Among them, σ represents the activation function, A is the adjacency matrix, D is the distance correction matrix, P is the step correction matrix, S' is the spatial attention matrix, W is the learnable parameter matrix, H represents the output of the lth hidden layer, and k represents the number of hidden layers of the GCN model.

6. The method according to claim 2, characterized in that The step of inputting the first output feature quantity into a time feature extraction model to extract the time feature to obtain the island detection result of the distribution network includes: Adding a position coding vector to the first output feature quantity by using the position coding in the time feature extraction model; The encoder and decoder in the time feature extraction model are used to extract the time features of the first output feature quantity after adding the position encoding vector to obtain the island detection result of the distribution network.

7. The method according to claim 1, characterized in that Each piece of data in the basic information data includes at least: voltage data, current data, power data, frequency data, operation status, location and number of each node in the distribution network, operation status of the branch, branch connection information and timestamp data corresponding to each piece of data; The network connection information includes a topological structure diagram of the distribution network, and the topological structure diagram is used to describe the connection status between nodes, impedance and admittance of wires.

8. An island detection device, characterized in that: The device comprises: An acquisition module is used to acquire basic information data in the distribution network connected to the photovoltaic power plant; A processing module, used for preprocessing the basic information data to obtain data to be detected; A detection module, used for inputting the data to be detected into a pre-built island detection model based on an improved spatiotemporal graph to obtain an island detection result of the distribution network; Among them, the island detection model includes a graph convolutional network GCN model and a temporal feature extraction model; the graph convolutional network GCN model is used to extract the spatial features of the data to be detected, and the temporal feature extraction model is used to extract the temporal features of the data to be detected.

9. An electronic device, characterized in that: The device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the island detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the island detection method according to any one of claims 1 to 7 when executed.

Images