Active distribution network island detection method based on multi-criterion fusion

Abstract

The invention discloses an active distribution network island detection method based on multi-criterion fusion, and relates to the technical field of electric power automation. The method aims at solving the problems that electric power feature selection and extraction are not intelligent enough, model training cannot fully utilize the advantages of multi-source data, and a traditional island detection method is prone to being affected by the environment or possibly introduces additional interference. According to the method, power features are extracted from historical operation data, and a multi-criterion fusion model is constructed by using a convolutional neural network; real-time operation data are collected and preprocessed, real-time electric power features are extracted, and the island state probability is calculated through a multi-criterion fusion model; analyzing an island occurrence mode based on a judgment result, and predicting an island occurrence probability by using an adaptive filter; according to the invention, through automatic feature engineering and a multi-criterion fusion model, the accuracy and reliability of island detection are improved, an early warning capability is provided, equipment damage and safety accidents are effectively prevented, the maintenance cost is reduced, and the overall stability and intelligent management level of a power grid are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power automation, and in particular, to an active distribution network island detection method based on multi-criterion fusion. Background Art

[0002] With the wide access of distributed energy resources (DERs), such as renewable energy sources like solar energy and wind energy, the complexity of modern power grids, especially active distribution networks (AND), and the dynamic nature of the operating environment have increased significantly. The islanding phenomenon, that is, an independent power supply area formed after the distributed power source is disconnected from the main power grid, has become a key factor affecting the safe and stable operation of the power grid. The islanding phenomenon may not only endanger the safety of power grid equipment, but also cause power supply interruption to electricity users, and even lead to safety accidents.

[0003] In the field of island detection technology, traditional detection methods are mainly divided into two categories: active detection methods and passive detection methods. Active detection methods identify the islanding state by actively injecting specific disturbance signals into the power grid, such as the frequency offset method, voltage offset method, etc. Although these methods can improve the sensitivity of island detection to a certain extent, they may also have an adverse impact on the normal operation of the power grid, such as introducing additional interference and affecting the power quality. Passive detection methods rely on detecting changes in electrical characteristics after the power grid loses power, such as changes in voltage and frequency, to judge the islanding state. However, passive detection methods are easily affected by environmental factors, such as load changes and fluctuations in distributed power source output, resulting in low accuracy and reliability of detection results.

[0004] In recent years, with the rapid development of machine learning and artificial intelligence technologies, using big data and intelligent algorithms to improve the accuracy of island detection has become a research hotspot. Especially in active distribution networks, due to the randomness and intermittency of distributed power sources, traditional detection methods are difficult to effectively cope with complex dynamic environments. Researchers have begun to explore data-driven methods, by collecting and analyzing historical operation data of the power grid, using intelligent algorithms to extract power operation characteristics, and building models to predict and identify the islanding state. This method has improved the sensitivity and reliability of island detection to a certain extent, but there are still some problems.

[0005] At present, most of the existing data-driven methods focus on the use of a single criterion and lack effective integration of multi-source heterogeneous data. In active distribution networks, power operation data is diverse and complex, and a single criterion often fails to comprehensively reflect the operating state of the power grid. In addition, the existing methods lack intelligent means in the feature selection and model construction processes and are difficult to adapt to changes in different power grid environments and operating conditions. This limits the applicability and generalization ability of the detection methods in complex scenarios.

[0006] For example, CN115580019A discloses a new method for active distribution network topology identification and island detection based on 5G communication; this method mainly aims at the active distribution network after the access of high-penetration distributed power sources. By collecting the voltage and current information of each line segment of the distributed new energy distribution network in real time and combining the characteristics of high speed and low delay of 5G communication technology, island detection is realized. It has certain innovation and practicability in island detection, but there are still problems such as data transmission and processing delay, complex algorithm and large calculation amount, high system dependence on 5G communication technology, poor adaptability to complex and changeable power grid environments, and lack of multi-criterion fusion technology defects.

[0007] Another example is that CN118150944B discloses a method and system for identifying fault sections in active distribution networks using the energy of characteristic signals; this method aims to inject characteristic signals into the active distribution network and use the energy of these signals for fault section identification and island detection. However, this solution still has limitations in characteristic signal injection, low accuracy of signal detection in complex power grid environments, complex fault identification, high requirements for the computing power of the system, island detection relying on the drop of the amplitude of the characteristic current signal at the outlet of IIDG, which is not accurate or reliable enough in some cases, highly dependent on specific hardware devices and control strategies, and the scalability and compatibility of the system are limited, and there are also technical defects in lacking multi-criterion fusion.

[0008] To solve the above problems, the present invention proposes an active distribution network island detection method based on multi-criterion fusion. Through intelligent feature selection and model construction, this method makes full use of multi-source heterogeneous data to achieve effective fusion of multiple criteria, thereby improving the accuracy and reliability of island detection and ensuring the safe and stable operation of the power grid. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to provide an active distribution network island detection method based on multi-criterion fusion to solve the problems existing in the field of power automation technology, especially island detection in active distribution networks. Specifically, with the wide application of distributed energy, the complexity of the power grid has increased significantly. Traditional island detection methods have obvious limitations: passive detection methods rely on detecting changes in electrical characteristics after the power grid loses power, such as changes in voltage and frequency, but are easily affected by environmental factors; while active detection methods identify the island state by injecting specific disturbance signals, but may introduce additional interference and affect the power quality; in addition, most existing data-driven methods focus on the use of a single criterion and lack effective integration of multi-source heterogeneous data, limiting the applicability of the detection method in complex scenarios.

[0010] To solve the above technical problems, the technical solution adopted by the present invention is: an active distribution network island detection method based on multi-criterion fusion, including the following steps: Step1: Extract historical power operation characteristics from historical operation data; Step2: Input the historical power operation characteristics into a convolutional neural network model for training to construct a multi-criterion fusion model; Step3: Collect real-time operation data and preprocess the real-time operation data; Step4: Extract real-time power operation characteristics from the preprocessed real-time operation data; Step5: Calculate the comprehensive score of the island state probability based on the multi-criterion fusion model to obtain the island state judgment result; Step6: Analyze the island occurrence mode based on the island state judgment result and combined with historical operation data; Step7: Based on the island occurrence mode, use an adaptive filter to predict the probability of island appearance.

[0011] In a preferred solution, the specific steps of Step1 include: Step1.1: Obtain historical operation data from the database; Step1.2: Use statistical methods to extract comprehensive power characteristics; Step1.3: Apply time series analysis methods to extract dynamic power characteristics; Step1.4: Adopt dimensionality reduction technology to reduce the dimensions of the comprehensive power characteristics and dynamic power characteristics to obtain historical power operation characteristics.

[0012] In a preferred solution, the comprehensive power characteristics in Step1.2 include mean, variance, standard deviation, maximum value, minimum value, and range; the dynamic power characteristics in Step1.3 include linear trend, periodic seasonality, and autocorrelation coefficient.

[0013] In a preferred embodiment, the specific steps of the Step2 include: Step2.1: Use a convolutional neural network model as the basic model; Step2.2: Convert the historical power operation characteristics into a time series through timestamps and input them into the input layer; Step2.3: Pass through multiple convolutional layers, each layer containing multiple convolutional kernels, for extracting local features; Step2.4: Add a pooling layer after the convolutional layer; Step2.5: After flattening the features after convolution and pooling, further process them through a fully connected layer; Step2.6: Use the softmax activation function as the output layer to output a comprehensive score representing the probability of the islanding state; Step2.7: Use the cross-entropy loss function and the gradient descent method to train and optimize the performance of the multi-criterion fusion model.

[0014] In a preferred embodiment, the real-time operation data in the Step3 includes real-time voltage, real-time current, real-time frequency, real-time power, and real-time energy; the preprocessing includes data cleaning, data normalization, and outlier identification and replacement.

[0015] In a preferred embodiment, the specific expression for calculating the comprehensive score of the islanding state probability based on the multi-criterion fusion model in the Step5 is: (1) In the formula, is the comprehensive score of the islanding state probability, is the activation function, is the weight vector from the hidden layer to the output layer, is the weight matrix from the input layer to the hidden layer, is the standardized data matrix, is the real-time power operation feature matrix corresponding to the first principal components, is the number of principal components, is the bias vector of the hidden layer,

[0016] In a preferred embodiment, the specific steps for obtaining the islanding state judgment result in the Step5 include: Step5.1: Set a threshold based on historical data, and compare the comprehensive score of the islanding state probability with the threshold ; Step5.2: When When it is, it means that the active distribution network is in the island state; Step5.3: When it is, it means that the active distribution network is in the non-island state.

[0017] In a preferred solution, the specific steps of the said Step6 include: Step6.1: When the active distribution network is in the non-island state, analyze the historical operation data through the rolling window analysis method to identify the typical characteristics before the occurrence of the island event; Step6.2: Use statistical techniques to identify outliers and periodic changes in the historical operation data; Step6.3: Based on the identified outliers and periodic changes, determine the island occurrence mode.

[0018] In a preferred solution, the specific expression for predicting the island appearance probability by using the adaptive filter based on the island occurrence mode in the said Step7 is: (2) In the formula, is the island appearance probability, is the constant vector, is the number of real-time power operation characteristics, is the index vector, is the th importance weight of the real-time power operation characteristics, is the th standardized value of the real-time power operation characteristics, is the coefficient vector of the adaptive filter, is the vector transpose, is the real-time power operation characteristic sequence.

[0019] In a preferred solution, the specific steps for predicting the island appearance probability in the said Step7 include: Step7.1: Set the threshold through the historical data, and compare the island appearance probability with the threshold ; Step7.2: When it is, it means that the probability of the occurrence of the island is very high, and immediately take warning notification to start the protection mechanism; Step7.3: When it is, it means that the probability of the occurrence of the island is relatively low.

[0020] In a preferred solution, the said detection method further includes: when it is predicted that the island appearance probability exceeds the preset threshold, immediately take warning notification and start the protection mechanism.

[0021] A computer device is a device used to implement the steps of an active distribution network islanding detection method based on multi-criterion fusion described in any one of the above. It includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the islanding detection method.

[0022] In a preferred embodiment, the computer device further includes a communication interface for data exchange with external devices.

[0023] In a preferred embodiment, the memory is provided with a readable storage medium on which a computer program is stored, and the processor executes the computer program to implement the steps of the islanding detection method.

[0024] In a preferred embodiment, the storage medium is a solid-state drive, a flash memory, or an optical disc.

[0025] An active distribution network islanding detection method based on multi-criterion fusion provided by the present invention has the following beneficial effects: 1. The present invention proposes an innovative active distribution network islanding detection method based on multi-criterion fusion. This method significantly improves the accuracy and reliability of islanding detection in multiple aspects, providing a strong guarantee for the safe and stable operation of the power grid.

[0026] 2. The present invention effectively solves the problems of insufficient intelligence in power feature selection and extraction, and the failure to fully utilize the advantages of multi-source data during the model training process. Traditional islanding detection methods are vulnerable to environmental influences or may introduce additional interference, while existing data-driven methods mostly focus on single criteria, limiting their applicability in complex scenarios. The present invention overcomes these technical limitations by introducing a multi-criterion fusion model and comprehensively considering the change trends of multiple power parameters.

[0027] 3. The present invention extracts historical power operation characteristics from historical operation data. These characteristics are obtained through statistical methods, time series analysis methods, and dimensionality reduction techniques, ensuring the representativeness and effectiveness of the characteristics. Through an automated feature engineering process, automatic selection and construction of features are achieved, improving the intelligence level of feature selection and providing a high-quality data basis for model training.

[0028] 4. The present invention inputs the extracted historical power operation characteristics into a convolutional neural network model for training to construct a multi-criterion fusion model. This model comprehensively considers the change trends of multiple power parameters such as voltage, current, frequency, power, and energy, realizes the effective fusion of multiple criteria, not only enhances the robustness and generalization ability of the model, but also significantly improves the accuracy and reliability of islanding detection.

[0029] 5. The present invention introduces an adaptive filter to predict the probability of islanding occurrence, achieving early warning, enabling grid managers to timely detect potential islanding events and take corresponding preventive and handling measures, thereby further enhancing the security and foresight of grid management.

[0030] 6. Through the method of the present invention, the power grid can reduce the risk of equipment damage and other safety accidents and reduce maintenance costs; at the same time, the present invention improves the overall stability and intelligent management level of the power grid, providing strong support for the long-term stable operation of the power grid.

[0031] 7. The method of the present invention is not only applicable to single-inverter systems, but also optimized for multi-inverter grid-connected systems; by solving problems such as dilution effect, the present invention improves the applicability in complex grid environments and provides an effective solution for islanding detection in multi-inverter grid-connected systems.

[0032] 8. The present invention has made significant progress and has practicality in the field of islanding detection. By introducing technical means such as multi-criterion fusion models, automated feature engineering processes, and adaptive filters, the accuracy and reliability of islanding detection are significantly improved, providing strong guarantee for the safe and stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention will be further described below in conjunction with the drawings and embodiments: Figure 1 is the flow chart of the detection method of the present invention; Figure 2 is the flow chart of judging the islanding state of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] The technical solutions in the present invention will be further described below in conjunction with the drawings and embodiments: Embodiment 1 As Figure 1 shown, this embodiment provides an active distribution network islanding detection method based on multi-criterion fusion, and the specific steps are as follows: Step S1: Extract historical power operation characteristics from historical operation data S1.1: Obtain historical operation data: Obtain the historical operation data of the active distribution network in the past period (such as one year) from the database, and these data include historical voltage, historical current, historical frequency, historical power, and historical energy, etc.

[0035] S1.2: Feature extraction: S1.2.1: Comprehensive power feature extraction: Use statistical methods to calculate statistics such as the mean, variance, standard deviation, maximum value, minimum value, and range of historical operation data, so as to reflect the average state over a long period, the degree of data dispersion, the maximum and minimum values in extreme cases, and the maximum fluctuation range of the data.

[0036] S1.2.2: Dynamic power feature extraction: Apply time series analysis methods to calculate the linear trend, periodic seasonality, and autocorrelation coefficient of historical operation data, etc., so as to reflect the long-term change trend, periodic change, and correlation at different time intervals of the data.

[0037] S1.3: Dimensionality reduction processing: Adopt dimensionality reduction techniques such as principal component analysis (PCA, Principal Component Analysis) to reduce the dimensions of comprehensive power features and dynamic power features, retain the main information, and obtain historical power operation features.

[0038] Step S2: Build a multi-criterion fusion model S2.1: Model selection: Use the convolutional neural network (CNN) model as the basic model. This model has strong local feature extraction ability and multi-scale feature capture ability, and is suitable for processing complex patterns in time series data.

[0039] S2.2: Model construction: S2.2.1: Data conversion: Convert historical power operation features into time series through timestamps and input them into the input layer of the CNN (Convolutional Neural Network) model.

[0040] S2.2.2: Feature extraction: Extract local features through multiple convolutional layers and pooling layers. The convolutional layer contains multiple convolutional kernels, which are used to slide on the input data and calculate the weighted sum of local regions; the pooling layer is used to downsample the feature map output by the convolutional layer, reduce the computational amount, and extract the main features.

[0041] S2.2.3: Processing of the fully connected layer: After flattening the features after convolution and pooling, further processing is performed through the fully connected layer to integrate local features into global features.

[0042] S2.2.4: Output layer: Use the softmax activation function as the output layer to output a comprehensive score representing the probability of the islanding state.

[0043] S2.2.5: Model training: Use the cross-entropy loss function and the gradient descent method to train and optimize the performance of the multi-criterion fusion model. By continuously iterating and adjusting the model parameters, make the comprehensive score of the islanding state probability output by the model as close as possible to the true islanding state.

[0044] Step S3: Real-time operating data processing and feature extraction S3.1: Data collection: Collect real-time operating data through sensors installed in the power grid (such as current transformers, voltage transformers, and temperature sensors), including real-time voltage, real-time current, real-time frequency, real-time power, and real-time energy, etc.

[0045] S3.2: Data preprocessing: Preprocess the collected real-time operating data, including data cleaning (removing obvious incorrect data points), data normalization (scaling the data to the same scale), and outlier identification and replacement (using the Z-Score method to identify outliers and replacing them with upper and lower limits).

[0046] S3.3: Feature extraction: Extract real-time power operation features from the preprocessed real-time operating data. The extraction method is the same as that in step S1.2.

[0047] Step S4: Islanding state judgment S4.1: Comprehensive score calculation: Based on the real-time power operation features, calculate the comprehensive score of the islanding state probability through the constructed multi-criterion fusion model.

[0048] S4.2: State judgment: Set a threshold based on historical data and compare the calculated comprehensive score of the islanding state probability with the threshold. When the comprehensive score is greater than the threshold, it indicates that the active distribution network is in the islanding state; otherwise, it indicates that the active distribution network is in the non-islanding state.

[0049] Step S5: Islanding occurrence mode analysis When the active distribution network is in the non-islanding state, analyze the historical operation data through the rolling window analysis method to identify the typical features before the occurrence of the islanding event. Use statistical techniques to identify outliers and periodic changes in the historical operation data, and combine the real-time operation data and historical data to analyze the islanding occurrence mode, providing a basis for subsequent islanding prediction.

[0050] Step S6: Islanding appearance probability prediction Based on the islanding occurrence mode and real-time power operation features, use an adaptive filter to predict the probability of the islanding appearance. By comprehensively considering various factors (such as the change trend of real-time power operation features, the periodicity of historical islanding occurrence modes, etc.), use the adaptive filter to continuously adjust the prediction model parameters to improve the prediction accuracy. Set a prediction probability threshold. When the prediction probability exceeds the threshold, immediately issue a warning notice and activate the protection mechanism to ensure the safe and stable operation of the power grid.

[0051] Embodiment 2 In another preferred embodiment, on the basis of the above Embodiment 1, this embodiment is further refined on the basis of Embodiment 1.

[0052] This embodiment provides an active distribution network islanding detection method based on multi-criterion fusion, including the following steps: S1. Extract historical power operation characteristics from historical operation data.

[0053] S1.1. Obtain historical operation data from the database; S1.2. For the historical operation data, use statistical methods to extract comprehensive power characteristics, apply time series analysis methods to extract dynamic power characteristics, and adopt dimensionality reduction techniques to reduce the dimensions of comprehensive power characteristics and dynamic power characteristics, so as to obtain historical power operation characteristics.

[0054] It should be noted that the historical operation data includes historical voltage, historical current, historical frequency, historical power, and historical energy.

[0055] Extract comprehensive power characteristics using statistical methods; Calculate the mean values of historical voltage, historical current, historical frequency, historical power, and historical energy. The expression is: (3) In the formula, is the mean value of historical operation data, is the observed value at time point , is the total number of time points; Calculate the variances of historical voltage, historical current, historical frequency, historical power, and historical energy. The expression is: (4) In the formula, is the variance of historical operation data; Calculate the standard deviations of historical voltage, historical current, historical frequency, historical power, and historical energy. The expression is: (5) In the formula, is the standard deviation of historical operation data; Calculate the maximum values of historical voltage, historical current, historical frequency, historical power, and historical energy. The expression is: (6) In the formula, is the maximum value of historical operation data, is the standardized data matrix; Calculate the minimum values of historical voltage, historical current, historical frequency, historical power, and historical energy. The expression is: (7) In the formula, is the maximum value of historical operation data; Calculate the range of historical voltage, historical current, historical frequency, historical power, and historical energy, with the expression: (8) In the formula, is the range of historical operating data; By calculating the mean, variance, standard deviation, maximum value, minimum value, and range of historical voltage, historical current, historical frequency, historical power, and historical energy, the following results are obtained: Mean: Reflects the average state over a long period; Variance: Reflects the degree of data dispersion; Standard deviation: The square root of the variance, which also reflects the degree of data dispersion; Maximum value: Reflects the maximum value in extreme cases; Minimum value: Reflects the minimum value in extreme cases; Range: Reflects the maximum fluctuation range of the data; Obtain the comprehensive power characteristics from these results; Apply time series analysis methods to extract dynamic power characteristics; Calculate the linear trend of historical voltage, historical current, historical frequency, historical power, and historical energy, with the expression: (9) In the formula, is the trend value at time point , is the intercept obtained through linear regression, is the slope obtained through linear regression; Calculate the periodic seasonality of historical voltage, historical current, historical frequency, historical power, and historical energy, with the expression: (10) In the formula, is the seasonal component at time point , is the original data value at time point ; Calculate the autocorrelation coefficient of historical voltage, historical current, historical frequency, historical power, and historical energy, with the expression: (11) In the formula, is the autocorrelation coefficient at lag, is the order of lag, is the mean of the observed values; By calculating the linear trends, periodic seasonality, and autocorrelation coefficients of historical voltage, historical current, historical frequency, historical power, and historical energy, the following results are obtained: Linear trend: Reflects the long-term change trend of the data; Periodic seasonality: Reflects the periodic changes of the data; Autocorrelation coefficient: Reflects the correlation of the data at different time intervals; Dynamic power characteristics are obtained from these results; Dimensionality reduction techniques are used to reduce the dimensions of the comprehensive power characteristics and dynamic power characteristics, obtaining historical power operation characteristics; Principal component analysis (PCA) is used to reduce the dimensions of the comprehensive power characteristics and dynamic power characteristics. The expression is: (12) In the formula, is the covariance matrix of the data, is the data matrix after standardization, is the matrix The transpose matrix of; Perform eigenvalue decomposition on the covariance matrix. The expression is: (13) In the formula, is the eigenvector matrix, is the eigenvalue diagonal matrix, is the matrix The transpose matrix of; Select the eigenvectors corresponding to the first largest eigenvalues as the principal components. The expression is: (14) In the formula, is the data matrix after dimensionality reduction, is the matrix composed of the first eigenvectors.

[0056] S2. Input the historical power operation characteristics into the convolutional neural network model for training to construct a multi-criterion fusion model.

[0057] S2.1. Take the convolutional neural network model as the basic model.

[0058] S2.1.1. Take the convolutional neural network model as the basic model; S2.1.2. Convert the historical power operation characteristics into a time series through timestamps and input them into the input layer; S2.1.3. Through multiple convolutional layers, each layer contains multiple convolutional kernels, which are used to extract local features; S2.1.4. Add a pooling layer after the convolutional layer; S2.1.5. After flattening the features after convolution and pooling, further process them through a fully connected layer; S2.1.6. Use the softmax activation function as the output layer to output a comprehensive score representing the probability of the island state; S2.1.7. Use the cross-entropy loss function and the gradient descent method to train and optimize the performance of the multi-criterion fusion model.

[0059] It should be noted that using the convolutional neural network model as the basic model to construct the multi-criterion fusion model is because the convolutional neural network model (CNN) has strong local feature extraction ability and multi-scale feature capture ability, can effectively process complex patterns in time series data, improve the accuracy and generalization ability of the model, and thus achieve more accurate and reliable island detection.

[0060] S3. The real-time operating data includes real-time voltage, real-time current, real-time frequency, real-time power, and real-time energy.

[0061] It should be noted that the real-time operating data is collected through sensors installed in the power grid; The sensors include current transformers, voltage transformers, and temperature sensors.

[0062] S3.1. The preprocessing includes data cleaning, data normalization, and outlier identification and replacement.

[0063] Specifically, remove the obviously incorrect data points, such as outliers outside the normal range; check the integrity of the data to ensure that all necessary fields have values; mark the missing values, identify and record the missing parts in the data for further processing.

[0064] Normalize the cleaned data. The expression is: (15) In the formula, is the normalized data value, is the original data value, is the average value of the real-time operating data, is the standard deviation of the data.

[0065] Use Z-Score to identify outliers in the normalized data and replace the outliers with upper and lower limit values. The expression is: (16) Regard the observation value as an outlier; Replace the outlier with the upper and lower limit values of the mean plus or minus 3 times the standard deviation.

[0066] S4. Extract real-time power operation features from the preprocessed real-time operation data, and calculate the comprehensive score of the islanding state probability based on the multi-criterion fusion model.

[0067] S4.1. Extract real-time power operation features from the preprocessed real-time operation data through statistical methods, time series analysis methods, and dimensionality reduction techniques.

[0068] S4.2. Calculate the comprehensive score of the islanding state probability based on the real-time power operation features through the multi-criterion fusion model. The expression is: (1) It should be noted that the activation function is defined by the Sigmoid function, and the expression is: (17) Equation represents the output value of the activation function for the input ; is a linear combination, representing the weighted sum of the standardized data matrix and the weight matrix plus the bias term; is the weight matrix from the input layer to the hidden layer, and the expression is: (18) In the formula, are the respective weight values in the weight matrix, is the number of neurons in the hidden layer, is the number of features after dimensionality reduction; It should be noted that the standardized data matrix is obtained by preprocessing the original power data (including data cleaning and standardization) and feature extraction (including features such as voltage fluctuations, current changes, frequency fluctuations, and power changes); is the standardized data matrix, and the expression is: (19) In the formula, represents each element in the standardized data matrix, that is, the original data value, represents the number of samples, represents the number of features.

[0069] It should also be noted that the real-time power operation feature matrix is a matrix obtained by extracting features from the preprocessed power data and retaining the main information using dimensionality reduction techniques such as principal component analysis (PCA); Real-time power operation feature matrix , the expression is: (20) Among them, represents the element in the real-time power operation feature matrix, is the number of original features, is the number of principal components.

[0070] S5. Obtain the island state judgment result.

[0071] Step5.1: Set a threshold based on historical data , and compare the comprehensive score of the island state probability with the threshold ; Step5.2: When , it means that the active distribution network is in the island state; Step5.3: When , it means that the active distribution network is in the non-island state.

[0072] It should be noted that the historical data includes voltage level and fluctuations, frequency fluctuations, power changes (active power and reactive power), current level and changes, and load data; By analyzing the characteristic correlation distribution in the historical data, a suitable threshold that can distinguish highly correlated and lowly correlated features is selected.

[0073] S6. Analyze the island occurrence mode based on the island state judgment result and combined with historical operation data.

[0074] S6.1. When the active distribution network is in the non-island state, analyze the historical operation data through the rolling window analysis method, identify the typical features before the occurrence of the island event, use statistical techniques to identify the outliers and periodic changes in the historical operation data, and identify the island occurrence mode.

[0075] It should be noted that the rolling window analysis method (RWA, Rolling Window Analysis) is a commonly used technique in time series analysis for processing and analyzing data that changes over time; The specific process includes: selecting a fixed window size, moving the window from the starting position of the time series data, calculating the statistics (such as mean, variance, etc.) within each window, extracting features, and identifying the outliers and periodic changes in the data through statistical techniques, and finally determining the typical features before the occurrence of the island event; It also needs to be noted that by using statistical techniques to identify the outliers and periodic changes in the historical operation data, based on the identified typical features before the occurrence of the island event, the island occurrence mode is further identified.

[0076] S7. Based on the islanding occurrence mode, an adaptive filter is used to predict the probability of islanding occurrence.

[0077] S7.1. Based on the islanding occurrence mode and real-time power operation characteristics, an adaptive filter is used to predict the probability of islanding occurrence. The expression is: (2) It should be noted that the real-time power operation characteristic sequence is a sequence formed in chronological order after extracting features (including dynamic features such as voltage fluctuation, current change, frequency fluctuation, power change, etc.) from the preprocessed power data.

[0078] S7.2. Set a threshold through historical data , and compare the probability of islanding occurrence with the threshold ; When , it indicates that the probability of islanding occurrence is very high, and an early warning notice is immediately taken to activate the protection mechanism; When , it indicates that the probability of islanding occurrence is relatively low.

[0079] Embodiment 3 In another preferred embodiment, on the basis of the above Embodiment 2, this embodiment further provides a computer device applicable to the case of an active distribution network islanding detection method based on multi-criterion fusion, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the active distribution network islanding detection method based on multi-criterion fusion as proposed in the above embodiment.

[0080] The computer device can be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be implemented through WIFI (Wireless Fidelity), a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0081] Embodiment 4 In another preferred embodiment, based on the above Embodiment 3, this embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the method for detecting an island in an active distribution network based on multi-criterion fusion proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0082] In summary, the present invention extracts historical power features through an automated feature engineering process, realizes the automatic selection and construction of features, improves the intelligent level of feature selection, and ensures the representativeness and effectiveness of the selected features; and by constructing a multi-criterion fusion model, realizes the effective fusion of multiple criteria, enhances the robustness and generalization ability of the model, and finally improves the accuracy and reliability of island detection.

[0083] Embodiment 5 In another preferred embodiment, based on the above-mentioned Embodiment 4, to further verify the technical solution of the present invention, this embodiment provides experimental simulation data of the active distribution network islanding detection method based on multi-criterion fusion. Refer to Table 1.

[0084] The preparation stage of the experiment includes obtaining historical operation data of the past year from the database and preprocessing it, including removing error data points, filling in missing values, and selecting and constructing features using statistical methods and professional knowledge in the power field. Subsequently, features are extracted through time series analysis methods, and PCA is used for feature dimensionality reduction processing. In addition, a variety of sensors including current transformers, voltage transformers, and temperature sensors are installed in the substation, the data acquisition frequency of once per second is set, and the data is transmitted to the central data processing server through a wireless sensor network. The data preprocessing also includes cleaning, normalization, and using the Z-Score method to identify and replace outliers, ensuring the accuracy and reliability of subsequent analysis.

[0085] First, power features are extracted from the historical operation data, and the data quality is ensured through preprocessing steps; secondly, the extracted features are used to train a convolutional neural network model to construct a multi-criterion fusion model for calculating the islanding state probability; then, real-time operation data is collected in real time through sensors and preprocessed, and power features are extracted from it and input into the trained model to calculate a comprehensive score, so as to obtain the judgment result of the islanding state; finally, based on the judgment result, the historical operation data is analyzed to identify the islanding occurrence pattern, and the probability of future islanding events is predicted using an adaptive filter, thereby achieving early warning and protection.

[0086] Specifically, as shown in Table 1 below: Table 1 Comparison Table of Grid States

[0087] Through the analysis of the data in the above table, it can be clearly seen that in the non-island state, the voltage, current, frequency, and power of the power grid are consistent with the historical data, indicating that the normal operating state can be accurately identified. Compared with the traditional method of judging the island state only by relying on the change of a single parameter, the multi-criterion fusion model of the present invention comprehensively considers the change trends of multiple power parameters, thereby improving the accuracy and reliability of detection. For example, in the island state, although the voltage drops slightly to 220.3V and the frequency drops to 59.8Hz, the current increases significantly to 11.5A and the power also rises substantially to 2533.45W. This change pattern is in sharp contrast to the data in the non-island state, indicating that the present invention can detect anomalies in the early stage of the islanding effect in a timely manner, so as to take protective measures in advance and avoid potential safety risks. By comparing the historical data with the measured values in different states, it can be clearly seen that the present invention not only performs excellently in terms of detection accuracy, but also can predict the probability of the occurrence of the islanding event through an adaptive filter, further enhancing the forward-looking and safety of power grid management. This enables the power grid management to respond more intelligently and efficiently to potential islanding effects in the face of distributed energy access, thus ensuring the stable operation of electricity.

[0088] In a preferred embodiment, the comprehensive power characteristics in Step 1.2 include mean, variance, standard deviation, maximum value, minimum value, and range; the dynamic power characteristics in Step 1.3 include linear trend, periodic seasonality, and autocorrelation coefficient; the above settings are aimed at comprehensively capturing the static and dynamic characteristics of power data, providing a solid foundation for subsequent anomaly detection and pattern recognition, and ensuring that the system can efficiently identify power fluctuations and respond to potential problems in a timely manner.

[0089] In a preferred embodiment, the real-time operation data in Step 3 includes real-time voltage, real-time current, real-time frequency, real-time power, and real-time energy; the preprocessing includes data cleaning, data normalization, and outlier identification and replacement; the above settings effectively improve the data quality and provide a reliable basis for subsequent data analysis. In the data cleaning stage, invalid and redundant information is removed; data normalization ensures the comparability of data with different dimensions; outlier identification and replacement further enhance the accuracy and stability of the data.

[0090] In a preferred embodiment, the detection method further includes: when the predicted probability of the occurrence of an island exceeds a preset threshold, an early warning notice is immediately issued and a protection mechanism is activated; the above settings are aimed at preventing the impact of the islanding effect on the safety of the power grid in advance, ensuring that necessary measures are taken before potential failures occur, reducing the risk of system damage, and at the same time enhancing the stability and reliability of the power grid operation.

[0091] In a preferred solution, the computer device further includes a communication interface for data exchange with external devices. With the above settings, the computer device can access a local area network or the Internet to achieve remote data transmission and control. In addition, the communication interface supports multiple communication protocols, enhancing the compatibility and expandability of the device and facilitating seamless docking with different types of external devices.

[0092] In a preferred solution, the memory is provided with a readable storage medium on which a computer program is stored. The processor executes the computer program to implement the steps of the island detection method. The storage medium is a solid-state drive, a flash memory, or an optical disc. With the above settings, the island detection system can quickly respond and accurately judge the power grid state, improving the efficiency and accuracy of detection. At the same time, the use of a solid-state drive, a flash memory, or an optical disc ensures the stability and persistence of data, providing reliable data support for island detection.

[0093] In summary, the active distribution network islanding detection method based on multi-criterion fusion provided by the present invention successfully solves the problems of insufficient intelligence in power feature selection and extraction and failure to fully utilize the advantages of multi-source data during the model training process. Aiming at the problems of traditional islanding detection methods being vulnerable to environmental influences or possibly introducing additional interference, the present invention effectively overcomes the technical limitations that existing data-driven methods mostly focus on single criteria, thus restricting their applicability in complex scenarios. Different from traditional methods, the present invention uses a convolutional neural network (CNN) model as the basis to construct a multi-criterion fusion model, which comprehensively considers the change trends of multiple power parameters, such as voltage, current, frequency, power, and energy, etc., so as to be able to more comprehensively evaluate the operating state of the power grid. The method of multi-criterion fusion is applied for the first time in the field of islanding detection, providing new ideas and means for islanding detection. The present invention extracts historical power features through an automated feature engineering process, realizing the automatic selection and construction of features, which not only significantly improves the efficiency and accuracy of feature selection, but also ensures the representativeness and effectiveness of the selected features, which is relatively rare in existing islanding detection methods. In addition, the present invention also introduces an adaptive filter to predict the probability of islanding occurrence, realizing early warning, more accurately capturing the change trend of the power grid state, and timely discovering potential islanding events, providing a strong guarantee for the safe and stable operation of the power grid. The present invention creatively applies the convolutional neural network model to the field of islanding detection, and significantly improves the accuracy and reliability of islanding detection by comprehensively considering the change trends of multiple power parameters. At the same time, this model also has strong learning ability and generalization ability, and can adapt to the islanding detection requirements under different power grid environments. The feature extraction and selection method based on an automated feature engineering process proposed by the present invention not only improves the efficiency and accuracy of feature extraction, but also avoids the feature selection deviation caused by human factors, providing new technical support for the field of islanding detection. By introducing an adaptive filter to predict the probability of islanding occurrence, the present invention realizes the early warning of islanding detection. This technological breakthrough not only improves the safety and foresight of power grid management, but also provides a new direction for the research in the field of islanding detection. By continuously optimizing and improving the algorithm and parameter settings of the adaptive filter, the accuracy and reliability of island prediction can be further improved. In addition, the present invention also has the ability of early warning. By predicting the probability of islanding events through an adaptive filter, the foresight and safety of power grid management are further enhanced, and this conclusion has been fully verified by detailed experimental data and theoretical derivation.

Claims

1. An active distribution network islanding detection method based on multi-criterion fusion, characterized in that It includes the following steps: Step1: Extract historical power operation characteristics from historical operation data; Step2: Input the historical power operation characteristics into a convolutional neural network model for training to construct a multi-criterion fusion model; Step3: Collect real-time operation data and preprocess the real-time operation data; Step4: Extract real-time power operation characteristics from the preprocessed real-time operation data; Step5: Calculate the comprehensive score of the islanding state probability based on the multi-criterion fusion model to obtain the islanding state judgment result; Step6: Analyze the islanding occurrence mode based on the islanding state judgment result and combined with historical operation data; Step7: Based on the islanding occurrence mode, use an adaptive filter to predict the islanding occurrence probability.

2. The active distribution network islanding detection method based on multi-criterion fusion according to claim 1, characterized in that The specific steps of Step1 include: Step1.1: Obtain historical operation data from the database; Step1.2: Use statistical methods to extract comprehensive power characteristics; Step1.3: Apply time series analysis methods to extract dynamic power characteristics; Step1.4: Adopt a dimensionality reduction technique to reduce the dimensions of the comprehensive power characteristics and dynamic power characteristics to obtain historical power operation characteristics.

3. The active distribution network islanding detection method based on multi-criterion fusion according to claim 2, wherein, The comprehensive power characteristics in Step1.2 include mean, variance, standard deviation, maximum value, minimum value, and range; the dynamic power characteristics in Step1.3 include linear trend, periodic seasonality, and autocorrelation coefficient.

4. A method for detecting islanding in an active distribution network based on multi-criterion fusion according to claim 1, characterized in that, The specific steps of Step2 include: Step2.1: Use a convolutional neural network model as the basic model; Step2.2: Convert the historical power operation characteristics into a time series through timestamps and input them into the input layer; Step2.3: Pass through multiple convolutional layers, each layer containing multiple convolutional kernels, for extracting local features; Step2.4: Add a pooling layer after the convolutional layer; Step2.5: After flattening the features after convolution and pooling, further process them through a fully connected layer; Step2.6: Use the softmax activation function as the output layer to output the comprehensive score representing the islanding state probability; Step2.7: Use the cross-entropy loss function and the gradient descent method to train and optimize the performance of the multi-criterion fusion model.

5. A method for detecting islanding in an active distribution network based on multi-criterion fusion according to claim 1, characterized in that: The real-time operation data in Step3 includes real-time voltage, real-time current, real-time frequency, real-time power, and real-time energy; the preprocessing includes data cleaning, data normalization, and outlier identification and replacement.

6. A method for detecting islanding in an active distribution network based on multi-criterion fusion according to claim 1, characterized in that, The specific expression for calculating the comprehensive score of the islanding state probability based on the multi-criterion fusion model in Step5 is: （1）； In the formula, is the comprehensive score of the island state probability, is the activation function, is the weight vector from the hidden layer to the output layer, is the weight matrix from the input layer to the hidden layer, is the data matrix after normalization, is the previous real-time power operation feature matrix corresponding to the principal components, is the number of principal components, is the bias vector of the hidden layer, is the bias term of the output layer.

7. A method for detecting islanding in an active distribution network based on multi-criterion fusion according to claim 1, characterized in that, The specific steps for obtaining the islanding state judgment result in Step5 include: Step5.1: Set a threshold based on historical data , and compare the comprehensive score of the islanding state probability with the threshold ; Step 5.2: When it indicates that the active distribution network is in an island state; Step5.3: When it indicates that the active distribution network is in a non-island state.

8. A method for detecting islanding in an active distribution network based on multi-criterion fusion according to claim 1, characterized in that, The specific steps of Step6 include: Step6.1: When the active distribution network is in a non-islanding state, analyze the historical operation data through a rolling window analysis method to identify the typical features before the occurrence of an islanding event; Step6.2: Use statistical techniques to identify outliers and periodic changes in the historical operation data; Step6.3: Based on the identified outliers and periodic changes, determine the islanding occurrence mode.

9. A method for detecting islanding in an active distribution network based on multi-criterion fusion according to claim 1, characterized in that, The specific expression for predicting the islanding occurrence probability using an adaptive filter in Step7 based on the islanding occurrence mode is as follows: （2）； In the formula, is the probability of islanding occurrence, is a constant vector, is the number of real-time power operation characteristics, is an index vector, is the importance weight of the th real-time power operation characteristic, is the normalized value of the th real-time power operation characteristic, is the coefficient vector of the adaptive filter, is the transpose of the vector is the real-time power operation characteristic sequence.

10. A method for detecting islanding in an active distribution network based on multi-criterion fusion according to claim 9, characterized in that, The specific steps for predicting the islanding occurrence probability in Step7 include: Step 7.1: Set the threshold through historical data , and compare the islanding occurrence probability with the threshold ; Step 7.2: When occurs, it indicates that the probability of islanding is very high, and the early warning notification is immediately taken to activate the protection mechanism; Step 7.3: When occurs, it indicates that the probability of islanding is relatively low.

11. A method for detecting islanding in an active distribution network based on multi-criterion fusion according to any one of claims 1 to 10, characterized in that, The detection method further includes: when the predicted islanding occurrence probability exceeds a preset threshold, immediately issue a warning notice and activate the protection mechanism.

Images