A Single-Phase and Three-Phase Hybrid Microgrid Cluster Islanding Detection Method Based on Improved Adaboost

Abstract

An islanding detection method for single-phase and three-phase hybrid microgrid groups based on an improved Adaboost algorithm includes the following steps: collecting sample data of different electrical characteristics measured at the Point of Common Coupling (PCC) in the single-phase and three-phase hybrid microgrid group; performing correlation analysis on various electrical characteristics measured at the PCC based on Partial Least Squares (PLS), selecting strongly correlated electrical characteristics, and forming training and testing sets; improving the Adaboost learner based on the Slime Mould Algorithm (SMA) and GSN weight learning method to enhance classification ability, reduce the impact of disturbances, and shorten detection time; constructing an islanding detection model for the microgrid group based on the improved Adaboost learner, and outputting islanding detection results based on real-time test set data at the PCC. This method can accurately detect islanding regardless of disturbance signals and system three-phase imbalance, exhibiting high anti-interference capability, accuracy, and generalization ability.

Description

Technical Field

[0001] This invention relates to the field of microgrid islanding detection, and particularly to a method for detecting islanding in single-phase and three-phase hybrid microgrid groups based on an improved Adaboost. Background Technology

[0002] Microgrids enable effective coordinated control of distributed generation devices and have become one of the main forms of efficient utilization of distributed power sources. Microgrids can operate in both grid-connected and islanded modes, with islanded operation further divided into planned and unplanned islanding. When unplanned islanding occurs, the microgrid's voltage and frequency will deviate from the main grid's control and become unstable, potentially damaging power equipment performance and affecting stable operation. Therefore, researching islanding detection technology for microgrids, quickly identifying unplanned islanding states, and taking measures to change the inverter's control mode are crucial for ensuring the reliability and stability of both the microgrid and the main grid.

[0003] In existing technical literature: Reference [1]: "Hybrid Island Detection Method Based on Genetic Algorithm and BP Neural Network" (Yu Yunjun, Zhong Guoying, Wan Xiaofeng, et al. Hybrid Island Detection Method Based on Genetic Algorithm and BP Neural Network [J]. Renewable Energy, 2018, 36(5):701-706.) uses a genetic algorithm to improve the traditional BP neural network, and uses the output of the improved BP neural network as the result of system state determination. This method can maintain a fast detection speed and a low false detection rate under different load characteristic scenarios.

[0004] Reference [2]: "Interpretable Study on Island Detection and Integration of Photovoltaic Grid-Connected Systems Based on LightGBM Algorithm" (Zhu Chunlin, Yu Chengbo. Interpretable Study on Island Detection and Integration of Photovoltaic Grid-Connected Systems Based on LightGBM Algorithm [J]. Electric Power Automation Equipment, 2023, 43(7):80-86+150.) proposes an island detection method based on the Light Gradient Boosting Machine algorithm. This method uses K-fold cross-validation and Shapley value additive interpretation based on decision tree to perform island detection and analysis, thereby improving the credibility of the model.

[0005] However, the above studies did not consider the selection and extraction of various electrical features at the PCC related to the islanding state. Reference [3]: "Islanding Detection Method for Grid-connected Photovoltaic Power Generation System Based on Adaboost Algorithm" (Jia Ke, Xuan Zhenwen, Lin Yaoqi, et al. Islanding Detection Method for Grid-connected Photovoltaic Power Generation System Based on Adaboost Algorithm [J]. Journal of Electrical Engineering, 2018, 33(5): 1106-1113.) proposes an intelligent passive islanding detection method based on Adaboost learner. This method can avoid the problem of traditional detection methods being unable to determine the threshold of electrical quantities without introducing interference signals. However, this method still has problems such as being affected by disturbance factors and taking too long when there are many training sample features. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for islanding detection of single-phase and three-phase hybrid microgrid groups based on an improved Adaboost. This method can accurately detect islanding regardless of disturbance signals and system three-phase imbalance, and has high anti-interference capability, accuracy, and generalization ability.

[0007] The technical solution adopted in this invention is as follows:

[0008] The islanding detection method for single-phase and three-phase hybrid microgrid groups based on the improved Adaboost includes the following steps:

[0009] Step 1: Collect sample data of different electrical characteristics in single-phase and three-phase hybrid microgrid groups measured at the Point of Common Coupling (PCC);

[0010] Step 2: Based on the partial least squares (PLS) method, correlation analysis was performed on various electrical characteristic quantities measured at the common coupling point, and strongly correlated electrical characteristic quantities were selected to form training and test sets;

[0011] Step 3: Improve the Adaboost learner based on the Slime Mould Algorithm (SMA) and GSN weight learning method to enhance classification ability, reduce the impact of perturbation, and shorten detection time.

[0012] Step 4: Construct an island detection model for micronetwork clusters based on the improved Adaboost learner, and output the island detection results based on the real-time test set data at PCC.

[0013] In step 1, the islanded electrical characteristics are caused by the islanding effect and are affected by the system and its operating state. The islanded electrical characteristics considered in this invention include voltage, rate of change of voltage over time, total harmonic distortion (THD) of voltage, current, rate of change of current over time, THD of current, frequency, rate of change of frequency over time, impedance, rate of change of impedance over time, active power, rate of change of active power over time, reactive power, rate of change of reactive power over time, and power factor. Therefore, step 1 requires collecting sample data of the above electrical characteristics. Sample data for different electrical characteristics are shown in Table 1.

[0014] Table 1 Basic Islanding Electrical Characteristics of Single-Phase and Three-Phase Hybrid Microgrid Clusters

[0015]

[0016]

[0017] In step 2, the selection of island electrical features in the intelligent island detection method directly affects the accuracy and speed of island detection. In machine learning, the more features a training model contains, the longer the training time and the higher the computational requirements. Therefore, this invention uses the PLS method to rank the basic island electrical features of a microgrid cluster by importance, and selects electrical features with a Variable Importance in Projection (VIP) index greater than 1 as strongly relevant electrical features for microgrid cluster island detection.

[0018] Using the PLS method to process the collected basic electrical characteristic data, an empirical expression for multiple linear regression is established:

[0019]

[0020] In the formula: X is the dataset composed of basic islanded electrical characteristics; This refers to the k-th electrical characteristic quantity corresponding to the PCC of the first microgrid. This refers to the k-th electrical characteristic quantity corresponding to the PCC of the second microgrid; The kth electrical characteristic quantity corresponds to the PCC of the i-th microgrid. The nth microgrid PCC corresponds to the kth electrical characteristic.

[0021] Y represents the islanded state dataset; y1 represents the islanded state of the first microgrid, where y1 = 0 indicates a non-islanded state and y1 = 1 indicates an islanded state; y2 represents the islanded state of the second microgrid, where y2 = 0 indicates a non-islanded state and y2 = 1 indicates an islanded state; y j For the islanded state of the j-th microgrid, y j =0 indicates a non-island, y j=1 represents an isolated island; y n For the islanded state of the nth microgrid, y n =0 indicates a non-island, y n =1 represents an isolated island.

[0022] The predictor variable is the result predicted by a multiple linear regression model; This is the first principal component variable obtained through the PLS method; This is the second principal component variable obtained through the PLS method; Let a1, a2, a3 be the m-th principal component variables obtained by the PLS method. m These are the coefficients of the 1st, 2nd, and mth variables in the empirical expression of multiple linear regression, representing the variables. and The proportion among all variables;

[0023] a i The coefficients are the empirical expression for multiple linear regression; m is the total number of principal component electrical features extracted; r i k For variables The corresponding correlation coefficient; ω ih With axis ω h The i-th principal component coefficient component.

[0024] Based on the above formula, the variable projection importance VIP value is calculated using the following formula:

[0025]

[0026]

[0027] In the formula: For variables The corresponding VIP value; Rdy is the value of the h-th extracted electrical feature component on the dependent variable y in the islanded state dataset Y. j Explanatory power; CRdy is the electrical characteristic quantity of m principal components paired with y j The cumulative explanatory power of r(y); j ;t h ) represents the j-th dependent variable y j The electrical characteristic component t extracted from the h-th principal component h The correlation coefficient between them; n is the sample size.

[0028] All training and detection sample data in this invention are collected from the PCC (Planetary Control Center). Starting from the moment islanding occurs, the average electrical characteristic values ​​are collected over three consecutive cycles. 70% of the samples are randomly selected from the collected dataset as the training sample set to train a single-phase / three-phase hybrid microgrid group islanding classification model based on the improved Adaboost. The remaining 30% of the samples are grouped as the detection sample set to verify the classification accuracy of the trained islanding classification model.

[0029] Step 3 specifically includes:

[0030] (1): Slime Mould Algorithm (SMA):

[0031] While the traditional AdaBoost learner can improve the utilization of individual weak classifiers, it is still susceptible to perturbation factors and redundant features. Therefore, this invention improves the AdaBoost learner based on SMA to enhance its classification ability and reduce the impact of perturbation factors.

[0032] The Adaboost learner trains the weak classifier G(x,l,ψ,δ) to improve its classification error rate e m To minimize this, the most important parameters are the bias parameter ψ and the threshold parameter δ, where x is the test sample and l(x) is the function for calculating the feature value.

[0033] This invention effectively optimizes the bias parameter ψ and the threshold parameter δ based on the slime mold optimization algorithm SMA, which has dynamic search capabilities and is not prone to getting trapped in local optima.

[0034] First, initialize the parameters of the slime mold optimization algorithm (SMA) with the classification error rate e of the Adaboost learner. m To train the Adaboost learner for the objective function of the slime mold optimization algorithm (SMA), the following steps are taken:

[0035] The objective function is the classification error rate e of the AdaBoost learner. m Its expression is

[0036] In the formula: G represents the weight of the i-th data point in the m-th iteration of training; m (x i ) represents the current weak classifier's performance on the i-th sample x. i The prediction results; y i Let be the true label of the i-th sample.

[0037] Secondly, the fitness is calculated and ranked using the objective function, and the population position is updated based on the ranking results. This includes the following:

[0038] The fitness value of each individual is calculated using the objective function, and all individuals in the population are ranked according to their fitness values. The position of each individual is then updated based on the ranking, resulting in a new population position. For each individual i, its fitness value is its corresponding classification error rate, i.e.,

[0039] Next, calculate the population fitness value, update the optimal fitness value and the optimal individual location of the slime mold, specifically including the following:

[0040] The set of individual fitness values ​​is the population fitness value. After calculating the fitness value of each individual, the individual with the smallest fitness value is found from the set, and its fitness value and position are taken as the best fitness value and best individual position of the current population.

[0041]

[0042] In the formula: x best The optimal position for an individual in the current population; f best This represents the fitness value of the best individual position in the current population.

[0043] Then, after satisfying the optimal conditions, the best individual position is output, and the global optimal parameter optimization result of the Adaboost learner is obtained, which includes the following:

[0044] If the number of iterations reaches the maximum value, the algorithm is considered to have converged, and the position of the current best individual is output. The parameters in the position of the best individual are the global best parameters of the Adaboost learner.

[0045] T = 500

[0046] X best =[δ * ,ψ * ]

[0047] In the formula: T is the number of iterations; X best The optimal individual position vector is output; δ * ψ * These are the optimal threshold parameter and the optimal bias parameter, respectively.

[0048] Finally, the optimized parameters are input into the Adaboost learner to obtain the trained, highest-performing binary classifier, which includes the following:

[0049] The globally optimal parameters obtained above are input into the AdaBoost learner to retrain the model. After training, the highest-performing binary classifier is output.

[0050] G m (x i )=G(x,l,ψ * ,δ * )

[0051]

[0052]

[0053] In the formula: x is the test sample, l(x) is the function for calculating the feature value; ψ * The optimal bias parameter; δ * The optimal threshold parameter; G represents the weight of the i-th data point in the m-th iteration of training; m (x i ) represents the current weak classifier's performance on the i-th sample x. i The prediction results; y i e is the true label of the i-th sample; m Let α be the classification error rate during the m-th iteration of training; m y(x) represents the weights of the weak classifier in the m-th iteration of training; y(x) represents the final strong classifier; M represents the number of iterations; and sign represents the sign function.

[0054] (2): GSN weight learning method:

[0055] This invention fully considers the impact of sample weight changes on algorithm performance and improves the original normalization method of the Adaboost learner based on the GSN weight learning method. GSN is a weight learning method formed by integrating three normalization methods: Batch Normalization (BN), Instance Normalization (IN), and Group Normalization (GN). This weight learning method can achieve the effects of improved accuracy and reduced computational redundancy. The specific expression is as follows:

[0056]

[0057] In the formula: Let be the weight of the i-th sample in the (m+1)-th iteration of training; Let α be the weight of the i-th sample in the m-th iteration of training; m y represents the weights of the weak classifier in the m-th iteration of training; i G represents the true label of the i-th sample;m (x i ) represents the prediction result of the weak classifier for the i-th sample in the m-th iteration of training;

[0058] γ is the initial learnable scaling factor; ζ is the initial learnable translation factor; υ is the hyperparameter; μ GSN , These are the feature mean and feature variance obtained after GSN processing, respectively; μ BN , Let μ be the mean and variance of the feature with respect to BN, respectively; IN , These are the mean and variance of the feature with respect to IN, respectively; μ GN , These are the mean and variance of the feature with respect to GN, respectively; , where BN, IN, and GN are weighting coefficients, representing the degree of contribution of BN, IN, and GN to the particular mean and particular variance, respectively.

[0059] The formula for the weight learning method in the above equation is:

[0060]

[0061] In the formula: θ represents the weight values; Ω = {BN, IN, GN} is the set of normalization methods; q represents the selected normalization method; θ q This is a control parameter, and its value is greater than 0; ln(1+θ) q ) is the parameter θ q Transformation of the natural logarithm.

[0062] The use of the logarithmic function can effectively prevent the phenomenon that the weights of correctly classified samples in the original Adaboost learner decrease in a numerical trend.

[0063] The normalization method of the Adaboost learner is improved based on the GSN weight learning approach, specifically including the following:

[0064] The three methods of batch normalization, instance normalization, and group normalization are combined to form GSN, which is then introduced into the Adaboost learner to replace the original normalization method.

[0065] The sample weights for the next round of training are adjusted according to the following formula:

[0066]

[0067] In the formula: H (m+1) Let be the sample weight matrix for the (m+1)th iteration training; These are the weights of sample 1, sample 2, and sample n in the (m+1)th iteration of training; Let be the weight of the i-th sample in the (m+1)-th iteration of training; Let α be the weight of the i-th sample in the m-th iteration of training; m y represents the weights of the weak classifier in the m-th iteration of training; i G represents the true label of the i-th sample; m (x i ) represents the prediction result of the weak classifier for the i-th sample in the m-th iteration of training; γ is the initial learnable scaling factor; ζ is the initial learnable translation factor; υ is the hyperparameter; μ GSN , These are the feature mean and feature variance obtained after GSN processing, respectively.

[0068] In step 4, an island detection model for the microgrid cluster is constructed based on the strong classifier generated by the improved Adaboost. The island detection result is output based on the threshold of the electrical feature quantity of island detection.

[0069] An islanding detection model for microgrid clusters is constructed using an improved Adaboost learner to monitor and identify in real time whether a microgrid cluster has disconnected from the main grid and entered an islanded state. This model collects electrical characteristic data (voltage amplitude, frequency, and phase angle changes) at the PCC (Power Control Center) of the microgrid cluster and analyzes and classifies this data using the improved Adaboost algorithm. When an electrical characteristic exceeds a preset threshold, the model outputs an islanding detection result, indicating that the microgrid has entered an islanded state. The islanding detection model for microgrid clusters implements the islanding detection method flow, as follows: Figure 3 As shown.

[0070] The islanding detection result is output based on the threshold of the electrical characteristic quantities used for islanding detection. Details are as follows:

[0071] Thresholds are set based on the differences in electrical characteristics under normal operation and islanded conditions. This invention selects voltage, phase angle change, and frequency as the monitored electrical characteristics. The voltage threshold is set to a fluctuation range of ±10%, i.e., [198V, 242V]; the phase angle change threshold is set to [-5°, +5°]; and the frequency threshold is set to a fluctuation range of ±1%, i.e., [49.5Hz, 50.5Hz]. When one or more electrical characteristics exceed these set thresholds, the microgrid group can be determined to be in an islanded state.

[0072] This invention discloses a method for detecting islanding in single-phase and three-phase hybrid microgrid groups based on an improved Adaboost, with the following technical advantages:

[0073] 1) In step 1 of this invention, the electrical characteristic data collected at the PCC covers a variety of electrical parameters of the microgrid cluster, such as voltage, current, and power, which can provide a comprehensive understanding of the microgrid's operating status. This detailed electrical parameter data accurately reflects the actual operating status of the microgrid, helping to improve the accuracy and reliability of islanding detection.

[0074] 2) In step 2 of this invention, the PLS method is used to analyze the correlation between electrical features and islanding states, which can effectively filter out the features with the strongest correlation to islanding detection from a large number of features. This method reduces the dimensionality of the data, improves the quality of training and testing samples, and thus improves the accuracy and efficiency of the model in detecting islanding states.

[0075] 3) In step 3 of this invention, by introducing the SMA algorithm and GSN weight learning method into the Adaboost learner, the performance of the model can be significantly improved. Specifically, the SMA algorithm can improve the model's classification ability and reduce the impact of perturbations, while the GSN weight learning method can shorten the model's detection time.

[0076] 4) In step 4 of this invention, by integrating multiple optimized weak classifiers, the improved Adaboost algorithm can significantly improve classification accuracy and robustness, especially performing excellently when dealing with complex and variable electrical features. Simultaneously, the improved sample weight update mechanism enables the model to more effectively handle noise and outlier data, shortening detection time and improving detection efficiency. Therefore, using the improved Adaboost strong classifier to construct an island detection model can significantly improve the accuracy and response speed of island detection. Attached Figure Description

[0077] Figure 1 This is a flowchart of the island detection method proposed in this invention.

[0078] Figure 2(a) shows the islanding detection results when the three-phase unbalance is 2%;

[0079] Figure 2(b) shows the islanding detection results when the three-phase unbalance is 8%;

[0080] Figure 2(c) shows the islanding detection results when the three-phase imbalance is 15%.

[0081] Figure 3 The flowchart illustrates the island detection method for the island detection model in microgrids. Detailed Implementation

[0082] The islanding detection method for single-phase and three-phase hybrid microgrid groups based on the improved Adaboost includes the following steps:

[0083] S1: Collect sample data of different electrical characteristics of single-phase and three-phase hybrid microgrid groups measured at the point of common coupling (PCC);

[0084] S2: The collected sample data is processed using the Partial Least Squares (PLS) method to select a set of strongly correlated electrical features and form a training sample set and a test sample set.

[0085] S3: Initialize the parameters of the Slime Mould Algorithm (SMA), the training sample set and its corresponding weights, and calculate the current classification error rate;

[0086] S4: Train the model with the classification error rate as the objective function, update the population position, and calculate the fitness value of the current individual;

[0087] S5: Update the best fitness value and the best individual position based on the current fitness value and the historical best fitness value, and determine whether the optimal conditions are met. If they are met, output the global best optimization parameters of the Adaboost learner; otherwise, repeat S4.

[0088] S6: Input the optimized parameters into the Adaboost learner, and at the same time use the GroupSwitchable Normalization (GSN) weight learning method to update the weights of the original training sample set to obtain the best strong binary classifier.

[0089] The GSN weight learning method is used to update the weights of the training samples. Then, all weak classifiers are linearly weighted and combined with their corresponding updated weights to obtain the optimal strong binary classifier. Using the GSN weight learning method, the weights of the training samples can be effectively updated in the AdaBoost learner, thereby improving the classifier's performance.

[0090] S7: The improved Adaboost learner is used to train the generated set of strongly correlated electrical features to obtain the best strong classifier;

[0091] First, electrical characteristic data at the PCC are collected. Then, the correlation between these electrical characteristics and the islanding state of single-phase and three-phase hybrid microgrid groups is analyzed using the PLS method. The set of electrical characteristics strongly correlated with islanding detection is extracted, and training and testing datasets are constructed. Then, the Adaboost learner is improved based on the SMA algorithm and GSN weight learning method, and the model is trained using the training dataset to finally obtain the best strong classifier.

[0092] S8: Based on the strong classifier generated by the improved Adaboost, construct an island detection model for micro-network groups.

[0093] Figure 1 This is a flowchart of the islanding detection method proposed in this invention. This invention discloses an islanding detection method for single-phase and three-phase hybrid microgrid groups based on an improved Adaboost. First, sample data of different electrical characteristics measured at the PCC (Pressure Cell Control Center) in the single-phase and three-phase hybrid microgrid group are collected. Second, correlation analysis and extraction of various electrical characteristics measured at the PCC are performed based on PLS (Power Linear Sequence), key characteristics are selected, and training and testing sets are formed. Then, the Adaboost learner is improved based on SMA (Structured Multi-Aspect Ratio) and GSN (Gateway Sequence Number) weight learning methods to improve classification ability, reduce the impact of disturbances, and shorten detection time. Finally, the improved Adaboost learner is used to train the generated set of strongly correlated electrical characteristics to obtain the optimal strong classifier. Based on this optimal strong classifier, an islanding detection model for the microgrid group is constructed, and the islanding detection results are output based on the real-time test set data at the PCC.

[0094] Figures 2(a), 2(b), and 2(c) show the islanding detection results under different three-phase unbalance degrees. A detection signal of 0 indicates no islanding, and 1 indicates islanding. When a single-phase and three-phase hybrid microgrid group suddenly enters islanding mode at 0.35s, after a certain period of time, the grid-connected circuit breaker opens, the system forms a power island, the grid-connected current becomes 0, and then the voltage at the PCC point begins to decay until it reaches 0. Figures 2(a) to 2(c) It can be seen that the method proposed in this invention can obtain relatively accurate island detection results when the three-phase imbalance is 2%, 8% and 15% respectively.

[0095] Table 2 Detection time of different island detection methods

[0096]

[0097] Table 2 shows the detection time for different islanding detection methods. When the three-phase imbalance is 2%, the islanding detection time of the method proposed in this invention is reduced by 17.39%, 13.64%, and 11.63% compared to Comparative Method 1, Comparative Method 2, and Comparative Method 3, respectively. When the three-phase imbalance is 8%, the islanding detection time of the method proposed in this invention is reduced by 17.20%, 15.58%, and 12.16% compared to Comparative Method 1, Comparative Method 2, and Comparative Method 3, respectively. When the three-phase imbalance is 15%, the islanding detection time of the method proposed in this invention is reduced by 11.05%, 9.55%, and 8.52% compared to Comparative Method 1, Comparative Method 2, and Comparative Method 3, respectively. When the three-phase imbalance increased from 2% to 15%, the time spans of the four island detection methods were 0.047s, 0.049s, 0.052s, and 0.057s, respectively. The island detection time span of the method proposed in this invention is reduced by 4.08%, 9.62%, and 17.54% compared to comparative methods one, two, and three, respectively. Therefore, compared to comparative methods one, two, and three, the method proposed in this invention has the shortest detection time under different three-phase imbalance conditions, is least affected by three-phase imbalance, and has the least increase in time consumption.

Claims

1. A single three-phase hybrid microgrid cluster island detection method based on improved Adaboost, characterized in that Includes the following steps: Step 1: Collect sample data of different electrical characteristics in a single-phase or three-phase hybrid microgrid group measured at the common coupling point (PCC); Step 2: Based on the partial least squares method, correlation analysis was performed on various electrical characteristic quantities measured at the common coupling point, and strongly correlated electrical characteristic quantities were screened out and used to form training and test sets; Step 3: Improve the Adaboost learner based on the slime mold optimization algorithm and GSN weight learning method; Step 4: Construct an island detection model for micronetwork clusters based on the improved Adaboost learner, and output the island detection results based on the real-time test set data at PCC; In step 3, the Adaboost learner trains the weak classifiers to minimize their classification error rate , where the most important parameters are the bias parameter and the threshold parameter , for the test samples; the slime mold optimization (SMA) effectively optimizes the bias parameter and the threshold parameter ; First, the parameters of the slime mold optimization algorithm SMA are initialized, and the classification error rate of the Adaboost learner is calculated The Adaboost learner is trained for the objective function of the slime mold optimization algorithm SMA, and specifically includes: The objective function is the classification error rate of the AdaBoost learner. Its expression is ; In the formula: For the first In the first round of iterative training The weight of each data point; For the current weak classifier to the th Sample The prediction results; For the first The true label of each sample; Secondly, the fitness is calculated and ranked using the objective function, and the population position is updated based on the ranking results, specifically including the following: The fitness value of each individual is calculated using the objective function, and all individuals in the population are sorted according to their fitness values. The position of each individual is then updated based on the sorting results, thus obtaining a new population position. For each individual... Its fitness value is its corresponding classification error rate, i.e. ; Next, calculate the population fitness value, update the optimal fitness value and the optimal individual location of the slime mold, specifically including: The set of individual fitness values ​​is the population fitness value. After calculating the fitness value of each individual, the individual with the smallest fitness value is found from the set, and its fitness value and position are taken as the best fitness value and best individual position of the current population. ； ； In the formula: The optimal position for an individual within the current population; This represents the fitness value of the best individual position in the current population. Then, after satisfying the optimal conditions, the best individual position is output, yielding the globally optimal parameter optimization result of the Adaboost learner, specifically including: If the number of iterations reaches the maximum value, the algorithm is considered to have converged, and the position of the current best individual is output. The parameters in the position of the best individual are the global best parameters of the Adaboost learner. ； ； In the formula: This represents the number of iterations. This is the optimal individual position vector for output; , These are the optimal threshold parameter and the optimal bias parameter, respectively; Finally, the optimized parameters are input into the Adaboost learner to obtain the trained highest-performing binary classifier, which includes: ； ； ； ； In the formula: For test samples, A function for calculating eigenvalues; The optimal bias parameter; The optimal threshold parameter; For the first In the first round of iterative training The weight of each data point; For the current weak classifier to the th Sample The prediction results; For the first The true label of each sample; For the first The classification error rate during rounds of iterative training; For the first The weights of the weak classifiers in each round of iterative training; For the final strong classifier; This represents the number of iterations. It is a symbolic function; The normalization method of the Adaboost learner is improved based on the GSN weight learning approach, specifically including: The sample weights for the next round of training are adjusted according to the following formula: ； ； In the formula: For the first The sample weight matrix for each round of iterative training; , , Sample 1, Sample 2 and Sample 3 are respectively. In the Weights in round-by-round training; For the first The sample at the th Weights in round-by-round training; For the first The sample at the th Weights in round-by-round training; For the first The weights of the weak classifiers during rounds of iterative training; For the first The true label of each sample; For the first In the round of iterative training, the weak classifier is used for the first... Prediction results for each sample; To initialize the learnable scaling factor; To initialize the learnable translation factor; For hyperparameters; , These are the feature mean and feature variance obtained after GSN processing, respectively.

2. The islanding detection method for single-phase and three-phase hybrid microgrid groups based on improved Adaboost according to claim 1, characterized in that: In step 1, the islanded electrical characteristics considered include voltage, rate of change of voltage over time, total harmonic distortion of voltage, current, rate of change of current over time, total harmonic distortion of current, frequency, rate of change of frequency over time, impedance, rate of change of impedance over time, active power, rate of change of active power over time, reactive power, rate of change of reactive power over time, and power factor.

3. The islanding detection method for single-phase and three-phase hybrid microgrid groups based on improved Adaboost according to claim 1, characterized in that: In step 2, the basic islanding electrical features of the microgrid are ranked by importance based on the PLS method, and electrical features with a variable projection importance (VIP) index greater than 1 are selected as strongly correlated electrical features for microgrid islanding detection.

4. The islanding detection method for single-phase and three-phase hybrid microgrid groups based on improved Adaboost according to claim 3, characterized in that: Using the PLS method to process the collected basic electrical characteristic data, an empirical expression for multiple linear regression is established: ； ； ； ； In the formula: The dataset consists of basic islanded electrical features; For the first microgrid PCC, corresponding to the first One electrical characteristic quantity; For the second microgrid PCC, corresponding to the first One electrical characteristic quantity; No. The corresponding microgrid PCC is at the first One electrical characteristic quantity; No. The corresponding microgrid PCC is at the first One electrical characteristic quantity; The dataset is in an isolated state; This represents the islanded state of the first microgrid. Not an isolated island, It is an isolated island; This represents the islanded state of the second microgrid. Not an isolated island, It is an isolated island; For the first The islanding state of a microgrid Not an isolated island, It is an isolated island; For the first The islanding state of a microgrid Not an isolated island, It is an isolated island; The predictor variable is the result predicted by a multiple linear regression model; The first one obtained by PLS method One principal component variable; This is the second principal component variable obtained through the PLS method; The first one obtained by PLS method One principal component variable; , , These are the first, second, and third terms in the empirical expression for multiple linear regression, respectively. The coefficients of each variable, representing the variables , and The proportion among all variables; These are the coefficients of the empirical expression for multiple linear regression; This represents the total number of principal component electrical features extracted. For variables The corresponding correlation coefficient; as axis The Each principal component coefficient component; Based on the above formula, the variable projection importance VIP value is calculated using the following formula: ； ； ； In the formula: For variables The corresponding VIP value; For the first The extracted electrical feature components are used in the isolated state dataset. dependent variable Explanatory power; for Each principal component electrical characteristic quantity pair Cumulative explanatory power; For the first One dependent variable With the The extracted principal component electrical characteristic components The correlation coefficient between them; This represents the number of samples.

5. The islanding detection method for single-phase and three-phase hybrid microgrid groups based on improved Adaboost according to claim 1, characterized in that: GSN is a weight learning method that integrates three normalization methods: Batch Normalization (BN), Instance Normalization (IN), and Group Normalization (GN). The specific expression is as follows: ； ； ； In the formula: For the first The sample at the th Weights in round-by-round training; For the first The sample at the th Weights in round-by-round training; For the first The weights of the weak classifiers during rounds of iterative training; For the first The true label of each sample; For the first In the round of iterative training, the weak classifier is used for the first... Prediction results for each sample; To initialize the learnable scaling factor; To initialize the learnable translation factor; For hyperparameters; , These are the feature mean and feature variance obtained after GSN processing, respectively; , These are the mean and variance of the feature with respect to BN, respectively; , These are the mean and variance of the feature with respect to IN, respectively; , These are the mean and variance of the feature with respect to GN, respectively; , , These are weighting coefficients, representing the degree of contribution of BN, IN, and GN to the particular mean and particular variance, respectively. The formula for the weight learning method in the above equation is: ； In the formula: These are weight values; This is the set of normalization methods. The type of normalization method selected; This is a control parameter, and its value is greater than 0; For parameters Transformation of the natural logarithm.

6. The islanding detection method for single-phase and three-phase hybrid microgrid groups based on improved Adaboost according to claim 1, characterized in that: In step 4, an island detection model for the microgrid cluster is constructed based on the strong classifier generated by the improved Adaboost. The island detection result is output based on the threshold of the electrical feature quantities for island detection; specifically as follows: Thresholds are set based on the differences in electrical characteristics under normal operating conditions and islanded conditions, selecting voltage, phase angle change, and frequency as the monitored electrical characteristics; the voltage threshold is set to a fluctuation range of ±10%, i.e. The phase angle change threshold is set to... The frequency threshold is set to a fluctuation range of ±1%, i.e. When one or more electrical characteristics exceed the set threshold, it can be determined that the microgrid group is in an islanded state.

Images