A feature selection integration method and system for large-scale wind power grid connection

Abstract

This disclosure belongs to the field of power system technology, specifically relating to a feature selection and integration method and system for large-scale wind power grid connection, including: acquiring multi-dimensional feature data containing wind power grid connection response features; performing incremental search on the acquired multi-dimensional data features based on preset optimal correlation and redundancy evaluation criteria to construct multiple nested candidate feature subsets; calculating the classification accuracy of the constructed multiple nested candidate feature subsets; recording the candidate feature subset with the highest classification accuracy, verifying the dimension of the recorded candidate feature subset, obtaining the optimal candidate feature subset, and completing feature selection and integration.

Description

Technical Field

[0001] This disclosure belongs to the field of power system technology, specifically relating to a feature selection integration method and system for large-scale wind power grid connection. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] After the integration of new energy units such as wind power, the system exhibits new characteristics not found in traditional synchronous units. These characteristics include weak stability, low inertia response, weak anti-interference ability, low overload capacity, and strong output fluctuations. These features bring new risks and challenges to classic stability problems and significantly increase the difficulty of using traditional methods for transient power angle stability assessment and control. Currently, traditional model-based analysis methods face difficulties in modeling new energy units and new power electronic equipment when large-scale integration of new energy sources are implemented. They also struggle to meet real-time requirements, and constructing Lyapunov energy functions for new power electronic components individually is quite challenging. Therefore, a fast, accurate, and online-applicable transient stability assessment method is urgently needed.

[0004] Leveraging the strong learning capabilities and high computational speed of big data technology to fully mine potentially useful information, while avoiding the difficulties of modeling new energy units, provides a new approach for transient power angle stability assessment of high-proportion new energy power systems. From a data-driven perspective, since detailed consideration of model structure and internal control logic is unnecessary, online assessment can be carried out by learning from measurement data offline. Furthermore, the complex control strategies and numerous component parameters of new energy units, coupled with continuous advancements in power grid measurement methods, objectively provide practical conditions for the application of data-driven methods.

[0005] Selecting appropriate input features is fundamental and crucial for subsequent transient stability assessment using machine learning models. Directly using the original feature set constructed from information on all generators, buses, etc., as input significantly increases the complexity and number of parameters in the subsequent model structure, placing a heavy burden on training. Compared to the optimal feature subset, training time would increase by tens of times, hindering rapid real-time application in the power grid and potentially leading to the "curse of dimensionality." Furthermore, the constructed feature set typically starts from physical mechanisms, relying on human experience to combine and calculate features. Humans often cannot identify deep-seated implicit relationships in the data, resulting in a large amount of redundant information. This redundancy not only wastes valuable computational resources but may also reduce model accuracy. Therefore, selecting a representative feature subset from the original feature set, removing redundant information, and achieving optimal dimensionality reduction are particularly important for further improving the performance of subsequent algorithms.

[0006] According to the inventor, the main methods for dimensionality reduction currently fall into two categories: feature extraction and feature selection. Feature extraction compresses the original data by mapping it to a new dimensional space, but loses the physical meaning represented by the features in the process. Feature selection, on the other hand, does not change the original expression of the features, but merely reduces their complexity through screening, which is more conducive to interpreting the subsequent classification results from the perspective of physical meaning. Summary of the Invention

[0007] To address the aforementioned issues, this disclosure proposes a feature selection integration method and system for large-scale wind power grid connection. The original feature set is processed through feature selection to meet the requirements for rapid evaluation.

[0008] According to some embodiments, the first solution of this disclosure provides a feature selection integration method for large-scale wind power grid connection, which adopts the following technical solution:

[0009] A feature selection integration method for large-scale wind power grid connection includes:

[0010] Obtain multidimensional feature data containing wind power grid connection response characteristics;

[0011] Based on the preset optimal correlation and redundancy evaluation criteria, incremental search is performed on the acquired multidimensional data features to construct multiple nested candidate feature subsets;

[0012] Calculate the classification accuracy of the constructed multiple nested candidate feature subsets;

[0013] Record the candidate feature subset with the highest classification accuracy, verify the dimensionality of the recorded candidate feature subset, obtain the optimal candidate feature subset, and complete the feature selection integration.

[0014] As a further technical limitation, the process for determining the preset optimal correlation and redundancy evaluation criteria is as follows:

[0015] The acquired multidimensional feature data containing wind power grid connection response characteristics were normalized and preprocessed, and the processed multidimensional feature data were divided into experimental dataset and test dataset.

[0016] The correlation criterion and redundancy were calculated and evaluated on the obtained experimental datasets to obtain the optimal feature subset;

[0017] Based on the obtained optimal feature subset, taking into account both dimensionality reduction and classification accuracy, the optimal relevance and redundancy assessments are obtained, and the preset optimal relevance and redundancy evaluation criteria are determined.

[0018] Furthermore, based on intensive and integrated comprehensive analysis, T-test and χ² were selected. 2The experimental datasets were evaluated using the following metrics: feature scores based on the Relief algorithm, Fisher-Score (a filtered feature selection algorithm), information gain ratio, Kruskal-Wallis test, and maximum correlation coefficient. The T-test was used to compare the significance of differences in features between the two experimental datasets using t-distribution theory. The χ² test was also employed. 2 The system employs several methods to assess the importance of features. These include: 1) testing the correlation between computational features and class labels to assign importance scores; 2) the Relief algorithm to score features by calculating intra-class and inter-class sample distances; 3) Fisher-Score, which selects features containing more discriminative information based on the principle of smaller intra-class distances and larger inter-class distances, with higher Fisher-Score values ​​indicating greater feature importance and relevance to the class; 4) Information gain ratio, which scores features by calculating the rate of change in information entropy before and after use, with higher values ​​indicating greater feature importance and relevance to the class; 5) Kruskal-Wallies test, which scores features by calculating whether there are differences in the distribution of different features; and 6) a novel method for detecting nonlinear correlations between features using the maximum correlation coefficient, which transforms mutual information values ​​into a new metric by finding the optimal discretization.

[0019] Furthermore, regarding feature redundancy, to balance correlation and redundancy and overcome the bias of mutual information towards attributes with more values, mutual information is improved based on normalization, i.e., the normalized form NMI is... Here, H(x) and H(y) are the entropies of x and y, respectively, which are used to measure the amount of information an event may have in multiple states, that is, the expected value of the amount of information with respect to the probability distribution of the event.

[0020] Furthermore, in the process of constructing the multiple nested candidate feature subsets, a search is performed based on different initial multidimensional data features. To divide the weights between relevance measures and redundancy, quantiles are introduced. By iteratively assigning values ​​to the introduced quantiles, the optimal feature sequence under different weights is obtained. The optimal feature sequence under different weights is the multiple nested candidate feature subsets.

[0021] As a further technical limitation, the classification accuracy of multiple nested candidate feature subsets obtained by support vector machine is used to verify the merits of each nested candidate feature subset based on the classification accuracy. The nested candidate feature subset with the best classification performance is selected, and the candidate feature subset with the highest classification accuracy is recorded.

[0022] As a further technical limitation, let X be a feature set consisting of N features, S be the selected feature set, and F be the candidate feature set. The specific steps to obtain the optimal candidate feature subset are as follows:

[0023] (1) Initialization; Set the selected feature set S to an empty set and the candidate feature set F to a full feature set, i.e. S→φ, F→X;

[0024] (2) Calculation The correlation measure between the label y and the label W(x) i ;y), sort all features according to W to form a new candidate feature set Fˊ, and denote the first k features as ;y), Take one feature in turn As an initial feature, that is, let

[0025] (3) Next, set the quantiles α of the correlation and redundancy measures. t Let x ∈ [0.1, 0.25, 0.5, 0.75, 0.9] (1 ≤ t ≤ T), and let x i ∈S m-1 x j ∈XS m-1 (m=2,...,N), from F′ m-1 Seeking to make maxφ(W(x) j ;y),G(x j ;x i ),α t Let x be the characteristic that holds true. vt ** , making F m-1 ′-{x vt **}→F m ′, S m-1 +{x vt **}→S m ;

[0026] (4) Repeat steps (2) to (3) until Fˊ is A set of features corresponding to different initial features is obtained T×N nested candidate feature subset matrix The total combination of feature sets is S = [S1 S2 … S v … S k ] T (1≤v≤k);

[0027] (5) For S v The corresponding nested candidate feature sets are used to successively verify each feature subset using SVM. Classification accuracy It also records the candidate feature subsets that achieve stable accuracy and are close to the global highest accuracy, along with their associated weights α. t * That is, j satisfies

[0028] (6) Calculate k results in S sequentially and record the results obtained in (5). make satisfy Furthermore, the subset of candidate features with smaller dimensions is the optimal subset of candidate features, where δ and δ′ both represent the set threshold.

[0029] According to some embodiments, the second solution of this disclosure provides a feature selection integrated system for large-scale wind power grid connection, which adopts the following technical solution:

[0030] A feature selection integrated system for large-scale wind power grid connection includes:

[0031] The acquisition module is configured to acquire multidimensional feature data containing wind power grid connection response characteristics;

[0032] The construction module is configured to perform incremental search on the acquired multidimensional data features based on preset optimal relevance and redundancy evaluation criteria, and construct multiple nested candidate feature subsets.

[0033] The computation module is configured to calculate the classification accuracy of the constructed multiple nested subsets of candidate features;

[0034] The selection integration module is configured to record the candidate feature subset with the highest classification accuracy, verify the dimensionality of the recorded candidate feature subset, obtain the optimal candidate feature subset, and complete the feature selection integration.

[0035] According to some embodiments, a third aspect of this disclosure provides a computer-readable storage medium, employing the following technical solution:

[0036] A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the feature selection integration method for large-scale wind power grid connection as described in the first aspect of this disclosure.

[0037] According to some embodiments, the fourth solution of this disclosure provides an electronic device that adopts the following technical solution:

[0038] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the feature selection integration method for large-scale wind power grid connection as described in the first aspect of this disclosure.

[0039] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0040] This disclosure employs an improved MMRMR ensemble selection method, deriving multiple MMRMR algorithms by introducing multiple correlation and redundancy evaluation criteria, and achieving intensive fusion through corresponding strategies. Simultaneously, a weight factor is introduced into the evaluation function, and the initial feature search space is expanded to obtain a set of nested candidate feature subsets. Then, a learning algorithm is used to verify each one to obtain the optimal feature subset, verifying the superiority of the selected optimal feature subset, thus achieving the requirement of rapid evaluation. Attached Figure Description

[0041] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0042] Figure 1 This is a flowchart of the feature selection integration method for large-scale wind power grid connection in Embodiment 1 of this disclosure;

[0043] Figure 2 This is a flowchart of the algorithm for determining the optimal evaluation criteria using the integrated MMRMR selection framework in Embodiment 1 of this disclosure;

[0044] Figure 3 This is a flowchart of the improved incremental search algorithm in Embodiment 1 of this disclosure;

[0045] Figure 4 This is a schematic diagram comparing the effects of different MMRMR feature selections in the integrated framework of Embodiment 1 of this disclosure;

[0046] Figure 5(a) is a schematic diagram comparing the feature selection effect of mRMR-R-NMI at different quantiles α when k=3 in Embodiment 1 of this disclosure;

[0047] Figure 5(b) is a schematic diagram comparing the feature selection effect of mRMR-R-NMI at different quantiles α when k=4 in Embodiment 1 of this disclosure;

[0048] Figure 6 This is a schematic diagram comparing the time complexity of the feature selection process in Embodiment 1 of this disclosure;

[0049] Figure 7 This is a flowchart of the feature selection integration system for large-scale wind power grid connection in Embodiment 2 of this disclosure. Detailed Implementation

[0050] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0051] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0052] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0053] Where there is no conflict, the embodiments and features described herein can be combined with each other.

[0054] Example 1

[0055] Embodiment 1 of this disclosure introduces a feature selection integration method for large-scale wind power grid connection.

[0056] like Figure 1 The illustrated feature selection integration method for large-scale wind power grid connection includes:

[0057] Obtain multidimensional feature data containing wind power grid connection response characteristics;

[0058] Based on the preset optimal correlation and redundancy evaluation criteria, incremental search is performed on the acquired multidimensional data features to construct multiple nested candidate feature subsets;

[0059] Calculate the classification accuracy of the constructed multiple nested candidate feature subsets;

[0060] Record the candidate feature subset with the highest classification accuracy, verify the dimensionality of the recorded candidate feature subset, obtain the optimal candidate feature subset, and complete the feature selection integration.

[0061] As one or more implementation methods, the acquired multidimensional feature data containing wind power grid connection response characteristics includes: voltage drop at the grid connection point of each wind farm at the time of fault, voltage at the grid connection point of each wind farm at the time of fault clearing, surge in output current at the grid connection port of each wind farm at the time of fault, change ratio of output current at the grid connection port of each wind farm at the time of fault clearing relative to the instant before fault clearing, reduction in active power output of each wind farm at the time of fault clearing relative to the time before fault, and increase in reactive power output of each wind farm at the time of fault clearing relative to the time before fault.

[0062] Specifically, the feature selection integration method for large-scale wind power grid connection introduced in this embodiment mainly includes two stages: selecting appropriate mRMR evaluation criteria and selecting the optimal feature set based on an improved mRMR search strategy.

[0063] In the process of selecting appropriate mRMR evaluation criteria, a variety of relevant measurement criteria are introduced.

[0064] Traditional mRMR algorithms typically use mutual information as a metric to measure the correlation between features and categories, which has poor adaptability and limits the improvement of feature search performance to some extent. Currently, there are various criteria for evaluating feature importance, but there is still a lack of theoretical support regarding which specific method is more suitable for which data type.

[0065] This embodiment adopts an intensive and integrated approach. After comprehensive analysis, T-test and χ² are selected. 2 Various metrics, including feature scores based on the Relief algorithm, Fisher-Score (a filtered feature selection algorithm), Information Gain Ratio (IGR), Kruskal-Wallis test, and Minimum Correlation Coefficient (MIC), are used to measure feature importance and can indirectly reflect the association between features and categories. Specifically, the T-test uses t-distribution theory to compare whether the difference between two features is significant; ChiSquare uses χ²... 2 The algorithm tests the correlation between computational features and class labels to achieve importance scoring; the Relief algorithm scores features by calculating intra-class and inter-class sample distances; Fisher-Score mainly selects features containing more discriminative information based on the principle of small intra-class distance and large inter-class distance, with larger values ​​indicating greater importance and relevance to the class; IGR scores features by calculating the rate of change of information entropy before and after feature usage, with larger values ​​indicating greater importance and relevance to the class; Kruskal-Wallis scores features by calculating whether there are differences in the distribution of different features; MIC is a new method for detecting nonlinear correlations between features, transforming mutual information values ​​into a new metric by finding the optimal discretization.

[0066] Regarding the redundancy between features, in order to better balance correlation and redundancy and overcome the disadvantage that mutual information tends to favor attributes with more values, mutual information is improved and its normalized form NMI is proposed, as shown in formula (1).

[0067]

[0068] In the formula, H(x) and H(y) are the entropies of x and y, respectively.

[0069] In this embodiment, the traditional MMRMR algorithm and seven improved MMRMR algorithms are used, with the traditional mRMR as a reference. The integrated framework is shown in Table 1.

[0070] Table 1. MMRMR Algorithm Integration Framework

[0071]

[0072] Eight selection algorithms were constructed and trained on the training set. Multiple nested candidate subsets were obtained based on the corresponding evaluation function maxφ(D,R). Through comparative analysis of classifier performance, the maximum classification accuracy and the corresponding optimal feature subsets for each algorithm were obtained. Considering both dimensionality reduction and classification accuracy, the optimal relevance and redundancy estimation methods within the overall MMRMR framework were determined and validated on the test set. The algorithm flowchart is shown below. Figure 2 As shown.

[0073] Because during the incremental search process, the value that makes max[I(x)] is always selected... i Using features that are true for the category as initial features may lead to overly limited subsequent feature searches. Therefore, this embodiment makes an improvement by no longer simply selecting the feature with the highest correlation to the category as the initial feature, but instead using the optimal correlation measure W(x) determined above. i The top k features of y) are sorted and used as initial features for incremental mRMR search to obtain multiple nested candidate subsets.

[0074] In addition, during the search process under different initial features, a quantile α is introduced to more precisely differentiate the weights between relevance measures and redundancy. By iteratively assigning values ​​to α, the optimal feature sequence under different weights is obtained. A support vector machine is then used to verify the merits of each nested feature subset, thereby selecting the feature subset that achieves the optimal classification performance. The specific correction criteria are as follows:

[0075] F = maxΦ(W,G)

[0076] Φ=αW-(1-α)G (2)

[0077] Here, W and G represent the optimal relevance measure and redundancy measure obtained in the preceding steps, respectively. When introducing new features, the search follows this strategy:

[0078]

[0079] The process of improving the incremental search algorithm is as follows: Figure 3 As shown, b is the variable step size of the quantile α.

[0080] Let X be a feature set consisting of N features, S be the selected feature set, and F be the candidate feature set. The specific steps to obtain the optimal candidate feature subset are as follows:

[0081] (1) Initialization; Set the selected feature set S to an empty set and the candidate feature set F to a full feature set, i.e. S→φ, F→X;

[0082] (2) Calculation The correlation measure between the label y and the label W(x) i ;y), sort all features according to W to form a new candidate feature set Fˊ, and denote the first k features as ;y), Take one feature in turn As an initial feature, that is, let

[0083] (3) Next, set the quantiles α of the correlation and redundancy measures. t Let x ∈ [0.1, 0.25, 0.5, 0.75, 0.9] (1 ≤ t ≤ T), and let x i ∈S m-1 x j ∈XS m-1 (m=2,...,N), from F′ m-1 Seeking to make maxφ(W(x) j ;y),G(x j ;x i ),α t Let x be the characteristic that holds true. vt ** , making F m-1 ′-{x vt **}→F m ′, S m-1 +{x vt **}→S m ;

[0084] (4) Repeat steps (2) to (3) until Fˊ is A set of features corresponding to different initial features is obtained T×N nested candidate feature subset matrix The total combination of feature sets is S = [S1 S2 … S v … S k ] T (1≤v≤k);

[0085] (5) For S v The corresponding nested candidate feature sets are used to successively verify each feature subset using SVM. Classification accuracy It also records the candidate feature subsets that achieve stable accuracy and are close to the global highest accuracy, along with their associated weights α. t * That is, j satisfies

[0086] (6) Calculate k results in S sequentially and record the results obtained in (5). make satisfy Furthermore, the subset of candidate features with smaller dimensions is the optimal subset of candidate features, where δ and δ′ both represent the set threshold.

[0087] The method proposed in this embodiment is analyzed using numerical examples, as follows:

[0088] The proposed feature selection integration framework based on the MMRR strategy was validated in an IEEE 39-bus power system containing wind power. The original feature set was a 63-dimensional feature set including wind power response features. By employing different relevance and redundancy evaluation criteria, the "value" of corresponding features in the entire incremental search process can be changed, thereby affecting their priority in the candidate subset.

[0089] To compare the aforementioned feature selection algorithms and select the optimal relevance and redundancy evaluation criteria, the goal is to minimize dimensionality and training time while maintaining classification accuracy, thereby improving model performance. SVM is used to classify and predict eight nested candidate subsets, with classification accuracy as the evaluation metric. This accuracy indirectly reflects the feature selection effect, as shown in the classification results. Figure 4 As shown.

[0090] During the process of adding features sequentially, the evaluation accuracy trend gradually increases from a certain value, then stabilizes or fluctuates slightly after reaching a certain extreme value. This indicates that there is indeed redundancy among features and confirms the necessity of selecting an appropriate number of features. An appropriate number of features can reduce information redundancy, while too many features will place a heavy burden on model training and reduce model evaluation performance. On the other hand, different MMRMR algorithms perform differently before reaching their peak. As the number of features increases, the accuracy of the mRMR-R-NMI algorithm reaches a stable value first, indicating that compared with other MMRMR algorithms, the mRMR-R-NMI algorithm can achieve relatively high classification accuracy with the fewest features. This is because the evaluation criterion based on Relief score is more applicable to data types under various penetration rates in this power grid, and NMI uses a normalized form, which can take into account features with smaller attribute values, effectively overcoming the shortcoming of biasing towards attributes with larger values ​​when selecting features. Therefore, mRMR-R-NMI is selected as the feature selection method here, that is, Relief score is used as the relevance measure and NMI is used as the redundancy measure.

[0091] By improving the incremental search algorithm, expanding the initial candidate features, and introducing quantiles to refine the search space, the screening level is further improved. The classification results for different values ​​of k are shown in Figure 5(a) and Figure 5(b), respectively:

[0092] By refining the search space using different initial features and weights, multiple nested candidate subsets can be obtained. Taking k=3 as an example, different feature sequences are generated under different quantile factors α. Comparison shows that the features at the head and tail do not change much, while the ranking of features near the optimal feature dimension changes significantly. This indicates that the intrinsic differences between these features are small, and their order can be changed by subtle weight adjustments, thus affecting the final selection of the optimal subset. Comprehensive comparison shows that when the quantile α is 0.25, the overall performance is the best in low to medium dimensions. When the feature dimension is only 14, the accuracy reaches 96.48%. Compared with the original feature set, this achieves efficient and simplified expression of information, avoids the decrease in accuracy that may be caused by too much redundant information, improves model performance, and compresses the dimensionality to 22% of the original.

[0093] To further verify the superiority of the proposed method and the selected feature subset in this embodiment, the selected feature subset A1 is compared with existing dimensionality reduction methods such as classical mRMR, principal component analysis (PCA), recursive feature elimination based on correlation bias reduction (SVM-RFE-CBR), the improved Relief algorithm ReliefF, Fisher's algorithm, LASSO, regularized discriminative feature selection UDFS for unsupervised learning, and the InfFS algorithm. When using the PCA algorithm, two compression methods are employed: one retains 99% of the original dataset's variance; the other compresses it to the same dimension as A1, constructing two feature subsets A2 and A3 respectively. All other algorithms are compared and verified using the same dimension as A1. The recursive feature elimination method uses SVM to score individual features, with C and g being 2. 0 and 2 -6 .

[0094] Least Squares Support Vector Machine (LSSVM) was used as the evaluation model, with the RBF kernel function selected. The model parameters, γ and σ, were optimized using grid search combined with ten-fold cross-validation. 2 The initial values ​​were set to 10 and 0.5 respectively, and then tests and verifications were conducted. In evaluating the results, this paper uses accuracy (ACC), Kappa statistic, and area under the ROC curve (AUC) as indicators. The average of these three values, η, is taken as the comprehensive evaluation index, and its expression is shown in formula (4):

[0095]

[0096] The final performance comparison of different feature selection algorithms is shown in Table 2. The time complexity of the feature selection process is as follows: Figure 6 As shown.

[0097] Table 2 Comparison of the optimal subset selection effects of different feature selection algorithms

[0098]

[0099]

[0100] Based on Table 2, it can be seen that, under the same dimensionality, the optimal subset selected by the method proposed in this embodiment has the best performance, exceeding that of traditional mRMR and other algorithms. PCA, while reducing dimensionality, loses the physical meaning of features, making the compressed subset lack interpretability. Regarding time complexity, [the following text appears to be incomplete and requires further context: "by..."] Figure 6It is evident that the SVM-RFE-CBR algorithm consumes the most time during feature selection, while other filter-type methods consume significantly less time than the Wrapper algorithm. This is primarily because the SVM-RFE-CBR algorithm, during the recursive feature elimination process, requires repeated training of the SVM to determine the optimal model parameters and scoring each feature based on the classification results, greatly increasing computation time. In contrast, the method proposed in this embodiment has a moderate time complexity among filter-type methods, less than the LASSO algorithm, thus meeting the requirements for rapid evaluation.

[0101] This embodiment employs an improved MMRMR ensemble selection method, deriving multiple MMRMR algorithms by introducing multiple correlation and redundancy evaluation criteria, and achieving intensive fusion through corresponding strategies. Simultaneously, a weight factor is introduced into the evaluation function, and the initial feature search space is expanded to obtain a set of nested candidate feature subsets. Then, a learning algorithm is used to verify each one to obtain the optimal feature subset, verifying the superiority of the selected optimal feature subset, and achieving the requirement of rapid evaluation.

[0102] Example 2

[0103] Embodiment 2 of this disclosure introduces a feature selection integrated system for large-scale wind power grid connection.

[0104] like Figure 7 The illustrated feature selection integrated system for large-scale wind power grid connection includes:

[0105] The acquisition module is configured to acquire multidimensional feature data containing wind power grid connection response characteristics;

[0106] The construction module is configured to perform incremental search on the acquired multidimensional data features based on preset optimal relevance and redundancy evaluation criteria, and construct multiple nested candidate feature subsets.

[0107] The computation module is configured to calculate the classification accuracy of the constructed multiple nested subsets of candidate features;

[0108] The selection integration module is configured to record the candidate feature subset with the highest classification accuracy, verify the dimensionality of the recorded candidate feature subset, obtain the optimal candidate feature subset, and complete the feature selection integration.

[0109] The detailed steps are the same as those of the feature selection integration method with large-scale wind power grid connection provided in Example 1, and will not be repeated here.

[0110] Example 3

[0111] Embodiment 3 of this disclosure provides a computer-readable storage medium.

[0112] A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the feature selection integration method for large-scale wind power grid connection as described in Embodiment 1 of this disclosure.

[0113] The detailed steps are the same as those of the feature selection integration method with large-scale wind power grid connection provided in Example 1, and will not be repeated here.

[0114] Example 4

[0115] Embodiment 4 of this disclosure provides an electronic device.

[0116] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the feature selection integration method for large-scale wind power grid connection as described in Embodiment 1 of this disclosure.

[0117] The detailed steps are the same as those of the feature selection integration method with large-scale wind power grid connection provided in Example 1, and will not be repeated here.

[0118] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

[0119] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A feature selection integration method for large-scale wind power grid connection, characterized in that, include: Obtain multidimensional feature data containing wind power grid connection response characteristics; Based on the preset optimal correlation and redundancy evaluation criteria, incremental search is performed on the acquired multidimensional data features to construct multiple nested candidate feature subsets; Calculate the classification accuracy of the constructed multiple nested candidate feature subsets; Record the candidate feature subset with the highest classification accuracy, verify the dimensionality of the recorded candidate feature subset, obtain the optimal candidate feature subset, and complete the feature selection integration; Let X be a common value. N A feature set consisting of 1 feature, S For the selected feature set, F Given the set of features to be selected, the specific steps to obtain the optimal subset of candidate features are as follows: (1) Initialization; Initialize the selected feature set S Let it be an empty set, candidate feature set F Let it be the full feature set, i.e. , ; (2) Calculation With tags y Correlation measure between All features according to W Sort to form a new candidate feature set Fˊ ,forward k Each feature is denoted as _ ... , , …, Take one feature in turn As an initial feature, that is, let , ; (3) Next, set the quantiles of the correlation and redundancy measures. At the same time, set , ,from China seeks to make The characteristics of establishment are denoted as ,make , ; (4) Repeat steps (2) to (3) until Fˊ for This yields a set of features corresponding to different initial characteristics. T×N nested candidate feature subset matrix The total feature set combination is ; (5) The corresponding nested candidate feature sets are used to successively verify each feature subset using SVM. Classification accuracy It also records the candidate feature subsets that achieve stable accuracy and are close to the global highest accuracy, along with their associated weights. ,Right now j satisfy ; (6) Calculate in sequence middle k The results are recorded in (5). ,make ,satisfy Furthermore, the subset of candidate features with smaller dimensionality is the optimal subset of candidate features. and Both indicate the setting of a threshold.

2. The feature selection integration method for large-scale wind power grid connection as described in claim 1, characterized in that, The process for determining the preset optimal correlation and redundancy evaluation criteria is as follows: The acquired multidimensional feature data containing wind power grid connection response characteristics were normalized and preprocessed, and the processed multidimensional feature data were divided into experimental dataset and test dataset. The correlation criterion and redundancy were calculated and evaluated on the obtained experimental datasets to obtain the optimal feature subset; Based on the obtained optimal feature subset, taking into account both dimensionality reduction and classification accuracy, the optimal relevance and redundancy assessments are obtained, and the preset optimal relevance and redundancy evaluation criteria are determined.

3. The feature selection integration method for large-scale wind power grid connection as described in claim 2, characterized in that, Based on the comprehensive analysis of intensive fusion, T-test and χ² were selected. 2 The experimental datasets were evaluated using the following metrics: feature scores based on the Relief algorithm, filtered feature selection algorithm, information gain ratio, Kruskal-Wallis test, and maximum correlation coefficient. The T-test utilized t-distribution theory to compare the significance of differences in features between the two experimental datasets. The χ² test was also used to evaluate the statistical significance of these differences. 2 The algorithm tests the correlation between computational features and class labels to achieve importance scoring; the Relief algorithm scores features by calculating intra-class and inter-class sample distances; the filtering feature selection algorithm selects features containing more discriminative information based on the principle of smaller intra-class distances and larger inter-class distances; the larger the value of the filtering feature selection algorithm, the more important the feature and the greater its correlation with the class; the information gain ratio scores features by calculating the rate of change of information entropy before and after the feature is used; the larger the value, the more important the feature and the greater its correlation with the class; the Kruskal-Wallies test scores features by calculating whether there are differences in the distribution of different features; and a new method for detecting nonlinear correlation between features using the maximum correlation coefficient transforms the mutual information value into a new metric by finding the optimal discretization.

4. The feature selection integration method for large-scale wind power grid connection as described in claim 3, characterized in that, Regarding feature redundancy, to balance correlation and redundancy and overcome the bias of mutual information towards attributes with more values, mutual information is improved based on normalization, i.e., the normalized form NMI is: ,in, and They are respectively x and y Entropy is used to measure the amount of information in an event with multiple states, that is, the expected value of the amount of information with respect to the probability distribution of the event.

5. The feature selection integration method for large-scale wind power grid connection as described in claim 2, characterized in that, In the process of constructing the multiple nested candidate feature subsets, a search is performed based on different initial multidimensional data features. To divide the weights between relevance measures and redundancy, quantiles are introduced. By iteratively assigning values ​​to the introduced quantiles, the optimal feature sequence under different weights is obtained. The optimal feature sequence under different weights is the multiple nested candidate feature subsets.

6. The feature selection integration method for large-scale wind power grid connection as described in claim 1, characterized in that, The classification accuracy of multiple nested candidate feature subsets obtained by using support vector machines is used to verify the merits of each nested candidate feature subset based on the classification accuracy. The nested candidate feature subset with the best classification performance is selected, and the candidate feature subset with the highest classification accuracy is recorded.

7. A feature selection integration system for large-scale wind power grid connection, employing a feature selection integration method for large-scale wind power grid connection as described in any one of claims 1-6, characterized in that, include: The acquisition module is configured to acquire multidimensional feature data containing wind power grid connection response characteristics; The construction module is configured to perform incremental search on the acquired multidimensional data features based on preset optimal relevance and redundancy evaluation criteria, and construct multiple nested candidate feature subsets. The computation module is configured to calculate the classification accuracy of the constructed multiple nested subsets of candidate features; The selection integration module is configured to record the candidate feature subset with the highest classification accuracy, verify the dimensionality of the recorded candidate feature subset, obtain the optimal candidate feature subset, and complete the feature selection integration.

8. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the feature selection integration method for large-scale wind power grid connection as described in any one of claims 1-6.

9. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the feature selection integration method for large-scale wind power grid connection as described in any one of claims 1-6.

Images