New energy station typical output scene extraction method with or without supervision combination
By combining unsupervised and supervised learning methods, using a support vector machine with comprehensive similarity indicators and kernel function mapping, the high-dimensional nonlinear feature processing problem of new energy station output scenario classification is solved, and more efficient and accurate classification results are achieved.
Patent Information
- Application Number
- CN202510473859.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-16
- Publication Date
- 2025-08-01
AI Technical Summary
The traditional new energy station output scenario classification method cannot effectively deal with high-dimensional nonlinear features, and there is a separation between the unsupervised and supervised learning frameworks, resulting in inefficient classification efficiency and insufficient accuracy.
The method of combining supervision and unsupervised is adopted to calculate the comprehensive similarity indicators for unsupervised pre-classification, and the support vector machine for kernel function mapping is used to perform supervised corrections, and finally optimize the classification results by evaluating the indicators.
It improves the accuracy and efficiency of the classification of force output scenarios in new energy stations, can better reflect the timing characteristics and coupling relationship of the force curve, and improves classification performance.
Smart Images
Figure CN120408338A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power system optimization, and particularly relates to a method for extracting typical output scenarios of new energy power stations by combining unsupervised and supervised methods. Background Art
[0002] Currently, with the continuous increase in the penetration rate of new energy in the power system, the volatility and uncertainty of new energy output have increased significantly. Its randomness and diversity not only pose great challenges to the stable operation of the power system, but also make it difficult for traditional optimization methods based on fixed output patterns to adapt. Facing a large number of new energy output scenarios, optimizing and analyzing them one by one is not only inefficient but also computationally burdensome.
[0003] Traditional similarity evaluation methods mostly use absolute difference metrics such as Euclidean distance to evaluate curve similarity. However, this method treats multi-dimensional vectors as a simple set of independent scalars, ignoring the internal coupling relationship between features, and is insensitive to phase shifts and morphological fluctuations of output curves, and cannot distinguish scenarios with time-series misalignment but similar patterns.
[0004] Existing scenario classification technologies usually independently adopt unsupervised or supervised learning frameworks, which have inherent limitations. Unsupervised classification methods rely on preset distance or density thresholds to divide samples, and it is difficult to characterize high-dimensional non-linear features. Supervised learning requires relying on complete labeled data to train the model, and it is easy to cause overfitting due to label sparsity or noise. Existing hybrid strategies usually adopt a serial mode of "first unsupervised rough classification, then supervised consolidation", resulting in the separation of the advantages of the two types of learning. Summary of the Invention
[0005] In view of the above problems, the present invention is proposed.
[0006] To solve the above technical problems, the present invention provides the following technical solution: A method for extracting typical output scenarios of new energy power stations by combining unsupervised and supervised methods, which includes the following steps:
[0007] Vectorize and normalize the output data of the new energy power station, and calculate the comprehensive similarity index;
[0008] Perform unsupervised pre-classification on the sample data to generate initial classification labels;
[0009] Construct a supervised multi-classification model of support vector machine based on kernel function mapping, and correct the initial classification labels to output the final classification result;
[0010] Calculate evaluation indicators for the final classification result to evaluate the compactness and separability of the final classification result.
[0011] As a preferred solution of a method for extracting typical output scenarios of new - energy power stations combining unsupervised and supervised learning, the calculation steps of the comprehensive similarity index include:
[0012] Vectorize the output data of the new - energy power station to form sample vector data;
[0013] Normalize the sample vector data to the range of the interval (0, 1] with the extreme values of each dimension as the boundaries to obtain sample normalized data;
[0014] Calculate the amplitude - feature distance and the shape - feature distance between different sample normalized data, and the weighted sum of the two is the comprehensive similarity index.
[0015] As a preferred solution of a method for extracting typical output scenarios of new - energy power stations combining unsupervised and supervised learning, the amplitude - feature distance is the Euclidean distance, and the shape - feature distance is the Pearson correlation - coefficient distance.
[0016] As a preferred solution of a method for extracting typical output scenarios of new - energy power stations combining unsupervised and supervised learning, the method for generating the initial classification labels includes:
[0017] Select the range of the number of clusters, and perform multiple unsupervised clusterings on the sample normalized data based on the comprehensive similarity index;
[0018] Calculate the cluster evaluation index, and determine the number of clusters corresponding to the optimal cluster evaluation index as the optimal number of clusters based on the quality of the cluster evaluation index;
[0019] Use the unsupervised clustering algorithm to pre - classify the samples based on the optimal number of clusters to generate initial classification labels.
[0020] As a preferred solution of a method for extracting typical output scenarios of new - energy power stations combining unsupervised and supervised learning, the method for obtaining the final classification result includes:
[0021] Use the kernel function to map the samples to a high - dimensional space, and construct an optimal separating hyperplane in this high - dimensional space for the binary - classification model of the support vector machine;
[0022] Adopt the one - against - one multi - classification strategy to extend the binary - classification model of the support vector machine to a multi - classification model;
[0023] Use the initial classification labels and the sample normalized data to train the support vector machine multi - classification model;
[0024] Use the trained support vector machine multi - classification model to correct the initial classification labels and output the final classification result.
[0025] As a preferred solution of a method for extracting typical output scenarios of new energy power stations combining unsupervised and supervised learning, the steps for evaluating the compactness and separability of the final classification results include:
[0026] Construct the within-class scatter and between-class scatter of the final classification results;
[0027] The within-class scatter characterizes the degree of within-group aggregation after classification, and the between-class scatter characterizes the degree of between-group separation after classification;
[0028] Calculate the CH coefficient based on the within-class scatter and between-class scatter, and comprehensively evaluate the compactness and separability of the final classification results through the CH coefficient.
[0029] As a preferred solution of a method for extracting typical output scenarios of new energy power stations combining unsupervised and supervised learning, the kernel function includes at least one of the hyperbolic tangent kernel function, polynomial kernel function, and Gaussian kernel function.
[0030] Another object of the present invention is to provide a system for extracting typical output scenarios of new energy power stations.
[0031] To solve the above technical problems, the present invention provides the following technical solution: A system for extracting typical output scenarios of new energy power stations, including: a data preprocessing module, an unsupervised label generation module, a supervised classification correction module, and a classification effect evaluation module;
[0032] The data preprocessing module vectorizes and normalizes the output data of the new energy power station, and calculates a comprehensive similarity index;
[0033] The unsupervised label generation module performs unsupervised pre-classification on the sample data to generate initial classification labels;
[0034] The supervised classification correction module constructs a supervised multi-classification model of a support vector machine based on kernel function mapping, corrects the initial classification labels, and outputs the final classification results;
[0035] The classification effect evaluation module calculates evaluation indicators for the final classification results to evaluate the compactness and separability of the final classification results.
[0036] The present invention provides a computer device, including a memory and a processor, the memory stores a computer program, and it is characterized in that when the processor executes the computer program, the steps of the method for extracting typical output scenarios of new energy power stations combining unsupervised and supervised learning are implemented.
[0037] The present invention provides a computer-readable storage medium, on which a computer program is stored, characterized in that when the computer program is executed by a processor, the steps of a method for extracting typical output scenarios of a new energy power station with combined unsupervised and supervised learning are implemented.
[0038] Advantages of the present invention: Based on the traditional Euclidean distance, the Pearson correlation coefficient is introduced as a similarity measure in the present invention, comprehensively considering the distance proximity and shape similarity of the output curves, so as to more accurately reflect the time-series characteristics of the curves and the coupling relationship between active power and reactive power. Secondly, unsupervised learning is used for pre-classification to give play to its advantage of high computational efficiency; then the classification results are corrected through supervised learning, making use of its ability to describe complex features. This combination effectively improves the performance and accuracy of classification. Description of the Drawings
[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0040] Figure 1 It is the overall flowchart of a method for extracting typical output scenarios of a new energy power station with combined unsupervised and supervised learning provided by an embodiment of the present invention.
[0041] Figure 2 It is the algorithm structure diagram of a method for extracting typical output scenarios of a new energy power station with combined unsupervised and supervised learning provided by an embodiment of the present invention. Detailed Embodiments
[0042] To make the above objects, features and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0043] Example 1, referring to Figures 1 to 2 , which is the first embodiment of the present invention. This embodiment provides a method for extracting typical output scenarios of a new energy power station with combined unsupervised and supervised learning, including the following steps:
[0044] S1. Vectorize and normalize the output data of the new energy power station, and calculate a comprehensive similarity index that takes into account both amplitude and shape.
[0045] The output data of new energy power stations consists of the new energy power generation in different periods. The collected output information needs to be converted into the sample form of vectors for subsequent algorithm processing. The length of the vector is related to the time resolution of the sample. The original time resolution of the output data is determined by relevant sensor devices, and the sample time resolution to be set should not exceed the original time resolution. The data type of the vector is related to the energy type of the new energy power station. For a single wind power or photovoltaic power station, the output data of the corresponding energy is directly used. For a hybrid wind-solar new energy power station, the sample vector of one energy needs to be connected after the sample vector of the other energy to form a complete sample.
[0046] It should be known that the preliminary processing of the output data of new energy power stations includes
[0047] vectorizing the output data of new energy power stations to form sample vector data;
[0048] normalizing the sample vector data with the extreme values of each dimension as the boundaries to the range of the interval (0, 1] to obtain sample normalized data;
[0049] In an optional embodiment, taking a hybrid wind-solar new energy power station with a set sample time resolution at the hourly level as an example, its sample data can be a 48-dimensional vector with the first 24 dimensions being the hourly wind power output data and the last 24 dimensions being the hourly photovoltaic power output data. Since the output intervals and characteristics of wind power and photovoltaic power are very different and belong to different types of data, normalization processing is required to eliminate this difference and normalize both types of data to the interval (0, 1] for subsequent algorithm processing.
[0050] Calculate the amplitude feature distance and morphological feature distance between different sample normalized data, and the weighted sum of the two is the comprehensive similarity index.
[0051] Normalization of wind power output data: p w，t ′ = (p w，t - p w，min ) / (p w，max - p w，min )
[0052] Normalization of photovoltaic power output data: p pv，t ′ = (p pv,t - p pv,min ) / (p pv，max - p pv，min ),t ∈ T pv
[0053] where, T pv represents the period with sunlight in a day, p w，t ′ represents the wind power output data at time t after normalization processing, p w，trepresents the wind power output data at time t before normalization, p w，max and p w，min represent the maximum and minimum values of the wind power output data at this moment; p pv,t ′ represents the photovoltaic power output data at time t after normalization, p pv，t represents the photovoltaic power output data at time t before normalization, p pv，max and p pv，min the maximum and minimum values of the photovoltaic power output data at this moment.
[0054] In this embodiment, a comprehensive similarity metric is used to measure the similarity between different samples. The comprehensive similarity metric adopts the weighted sum of the Euclidean Distance and the Pearson Correlation Coefficient distance, comprehensively considering the amplitude and shape of the samples. Among them, the Euclidean distance represents the straight-line distance between two samples, highlighting the amplitude characteristics of the samples, and the Pearson correlation coefficient can measure the linear relationship between two samples and reflect the shape characteristics of the samples.
[0055] In this embodiment, it should be noted that: the amplitude feature distance is the Euclidean distance, and the shape feature distance is the Pearson correlation coefficient distance.
[0056] Euclidean distance:
[0057] Pearson correlation coefficient distance:
[0058] Comprehensive similarity metric:
[0059] Among them, are the average values of samples x i and x j respectively, and λ is the weighting coefficient.
[0060] S2. Perform unsupervised pre-classification on the sample data to generate initial classification labels.
[0061] The generation method of the initial classification label includes
[0062] selecting the range of the number of clusters, and performing multiple unsupervised clusterings on the normalized sample data based on the comprehensive similarity index;
[0063] calculating the cluster evaluation index, and determining the number of clusters corresponding to the optimal cluster evaluation index as the optimal number of clusters based on the quality of the cluster evaluation index;
[0064] Using the unsupervised clustering algorithm, pre-classify the samples based on the optimal number of clusters to generate initial classification labels.
[0065] Specifically, in this embodiment, unsupervised divisive hierarchical clustering is used to classify the sample data. The Calinski-Harabasz Index (CH coefficient) is selected as the clustering evaluation index to obtain the optimal number of clusters, and the labeled data obtained from the unsupervised classification is used as the basis for subsequent supervised classification.
[0066] In an alternative embodiment, the steps for determining the optimal number of clusters of the new energy power station output samples include:
[0067] In this embodiment, the method for determining the optimal number of clusters of the output samples using the CH coefficient (Calinski-Harabasz Index) is adopted. The CH coefficient is a statistic for evaluating the quality of clustering results. The clustering quality is evaluated by quantifying the ratio of the within-class scatter to the between-class scatter. The calculation formula of CH is as follows:
[0068]
[0069] where k c represents the number of clusters, and N is the total number of samples;
[0070] The within-class scatter S W represents the degree of aggregation of samples within a single cluster after clustering, which is the sum of the squared distances from the samples in each cluster to the cluster center. The formula is as follows:
[0071]
[0072] where C c is the set of clusters, and μ c is the cluster center of cluster C.
[0073] The between-class scatter S B represents the overall degree of dispersion between different clusters after clustering, which is the sum of the squared distances from the cluster centers to the global mean. The formula is as follows:
[0074]
[0075] where μ is the global mean vector, and N c is the number of samples in cluster C.
[0076] In this embodiment, the specific steps for selecting the optimal number of clusters using the CH coefficient determination method are as follows:
[0077] ① Select the range of the number of clusters k c ∈[k min , k max , usually k min = 2,
[0078] ② For each k within this rangec Perform clustering using an unsupervised clustering method and calculate the CH coefficient;
[0079] ③ Obtain the function of CH with respect to k c Select the value that maximizes the CH coefficient as the optimal number of clusters That is
[0080] Unsupervised divisive hierarchical clustering:
[0081] In this embodiment, the divisive hierarchical clustering method is used as the unsupervised classification method, and the steps are as follows:
[0082] Regard all samples as an initial cluster;
[0083] Calculate the average distance between all samples within each cluster and the center of the cluster. The one with the largest value is the cluster to be split, denoted as cluster C. The distance uses a comprehensive similarity metric;
[0084] Calculate the average distance between all samples in cluster C and other samples within the cluster. The sample with the largest value is selected as the split point of the cluster to be split, serving as the seed of sub-cluster C1, and the center (centroid) of other samples within the cluster is the seed of sub-cluster C2;
[0085] Perform cluster splitting: Assign the remaining samples in the original cluster C to C1 or C2. If the average distance of a sample to C1 is less than the average distance to C2, it remains in C1; otherwise, it is moved to C2;
[0086] Iteratively repeat the above steps except for initialization until the total number of clusters meets the set requirements.
[0087] In this embodiment, the result of unsupervised clustering using the optimal number of clusters will be saved as labeled data for subsequent supervised classification.
[0088] S3. Construct a supervised multi-classification model of a support vector machine based on kernel function mapping, correct the initial classification labels, and output the final classification result.
[0089] Specifically, in this embodiment, the result of clustering the new energy output data using unsupervised classification is used as the label for supervised classification, and the supervised classification is implemented using the support vector machine (SVM) algorithm.
[0090] In an alternative embodiment, the method for obtaining the final classification result includes
[0091] Use a kernel function to map the samples to a high-dimensional space, and construct an optimal separating hyperplane in this high-dimensional space for the binary classification model of the support vector machine;
[0092] The binary classification model of the support vector machine is extended to a multi-classification model by adopting a one-to-one multi-classification strategy;
[0093] The support vector machine multi-classification model is trained using the initial classification labels and sample normalized data;
[0094] The trained support vector machine multi-classification model is used to correct the initial classification labels and output the final classification results.
[0095] In another set of alternative embodiments, the SVM binary classification model based on kernel function mapping
[0096] The output data classification of new energy power stations exhibits non-linearly separable characteristics. Therefore, the support vector machine (SVM) maps the samples to a high-dimensional space through a kernel function and constructs an optimal separating hyperplane in this space.
[0097] The kernel function includes at least one of the hyperbolic tangent kernel function, polynomial kernel function, and Gaussian kernel function. In this embodiment, the Gaussian kernel function is adopted in this method to achieve non-linear mapping:
[0098] K(u,v) = exp(-σ∥u - v∥ 2 );
[0099] Among them, the parameter σ controls the local influence range of the samples. When the value of σ is small, the kernel action range is wider, and vice versa, it focuses more on neighboring samples.
[0100] The core goal of SVM is to construct a hyperplane with the maximum margin, while allowing a small number of samples to be misclassified to improve the generalization ability. The corresponding optimization problem is:
[0101]
[0102] In the formula, φ(x i ) is the high-dimensional mapping vector of the sample x i , ω is the hyperplane normal vector, β is the bias term, ζ i is the slack variable, and the parameter D controls the misclassification penalty intensity. In the objective function, the first term maximizes the classification margin, and the second term constrains the number of misclassified samples. Solving this quadratic programming problem can obtain the optimal classification hyperplane.
[0103] The SVM extended model based on the multi-classification strategy:
[0104] For the multi-classification problem of new energy power station output data, the basic binary SVM needs to be extended to a multi-classification model. Common strategies include: One-vs-Rest (OvR) strategy and One-vs-One (OvO) strategy. Let the optimal number of classifications be k. Among them, the One-vs-Rest strategy constructs k binary classifiers. The i-th classifier distinguishes the i-th class from all other classes. During prediction, the sample is classified as the class corresponding to the classifier with the largest output value. The One-vs-One strategy constructs k(k - 1) / 2 binary classifiers, and each classifier only distinguishes two types of samples. During prediction, all classifiers vote on the sample, and the class with the most votes is the final result. Considering that the One-vs-Rest strategy is prone to blurred classification boundaries due to unbalanced class samples, and there may be a contradiction where multiple classifiers accept the sample at the same time, the One-vs-One strategy is adopted here to improve the classification robustness.
[0105] Classification implementation steps and boundary optimization:
[0106] There are N samples {x1, x2, …, x N} and labels {y1, y2, …, y } of category N . The supervised classification steps are as follows:
[0107] ① Kernel parameter initialization and multi-classification model construction: Initialize the Gaussian kernel parameter σ and the penalty coefficient D, and generate binary classifiers according to the One-vs-One strategy;
[0108] ② Model training: Independently solve the optimization problem for each binary classifier through the samples and labels to obtain the hyperplane parameters {ω ij , β ij} (i, j are category indices), and form a trained SVM classifier;
[0109] ③ Use the trained classifier for supervised classification: Put the N samples {x1, x2, …, x N} into the trained SVM classifier for classification to obtain the final classification result.
[0110] S4. Calculate the evaluation index for the final classification result to evaluate the compactness and separability of the final classification result.
[0111] The core of evaluating the classification effect lies in quantifying the similarity of samples within the group and the difference of patterns between groups. To meet the engineering requirements, it is also necessary to be independent of prior labels and be able to be compatible with vector-type data of different dimensions. Based on the above principles, the CH coefficient (Calinski-Harabasz Index) is selected as the evaluation criterion for the classification effect, which objectively measures the compactness and separability of the clustering result through the ratio of the within-class scatter to the between-class scatter.
[0112] Construct the within-class scatter and between-class scatter of the final classification result;
[0113] The within-class scatter characterizes the degree of within-group aggregation of each group after classification, and the between-class scatter characterizes the degree of between-group separation of each group after classification;
[0114] Calculate the CH coefficient based on the within-class scatter and between-class scatter, and comprehensively evaluate the compactness and separability of the final classification result through the CH coefficient.
[0115] In an alternative embodiment, use the within-class scatter S W to evaluate the degree of aggregation of samples within the same category. The smaller S W , the more compact the spatial distribution of samples of the same class, and the more ideal the classification result, where the Euclidean distance is replaced by the comprehensive similarity metric D ij , denoted as There is:
[0116]
[0117] where D(x i , μ c ) is the comprehensive similarity degree between the sample x i and the cluster center μ c .
[0118] Use the between-class scatter S B to characterize the degree of separation between different categories. The larger S B , the more dispersed the centroid distribution of different clusters, and the clearer the classification boundary, where the Euclidean distance is replaced by the comprehensive similarity metric D ij , denoted as There is:
[0119]
[0120] where D(μ c , μ) is the comprehensive similarity degree between the cluster center μ c and the global mean vector.
[0121] The CH coefficient comprehensively evaluates the classification effect through the normalized ratio of the between-class scatter to the within-class scatter. The larger the CH coefficient, the higher the within-group compactness and the stronger the between-group separability of the classification result corresponding to the current number of clusters:
[0122]
[0123] where the denominator k c -1 and N - k c are the degrees of freedom correction terms to eliminate the scale effect of the number of clusters on the scatter.
[0124] Example 2, the second example of the present invention, is different from the previous example in that:
[0125] If the above functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.
[0126] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
[0127] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion (electronic device) having one or more wirings, a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, a computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a computer memory.
[0128] It should be understood that each part of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logic functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.
[0129] Embodiment 3 is the third embodiment of the present invention. This embodiment provides a system for extracting typical output scenarios of a new energy power station, including a data preprocessing module, an unsupervised label generation module, a supervised classification correction module, and a classification effect evaluation module;
[0130] The data preprocessing module vectorizes and normalizes the output data of the new energy power station, and calculates a comprehensive similarity index;
[0131] The unsupervised label generation module performs unsupervised pre-classification on the sample data to generate initial classification labels;
[0132] The supervised classification correction module constructs a supervised multi-classification model of a support vector machine based on kernel function mapping, corrects the initial classification labels, and outputs the final classification results;
[0133] The classification effect evaluation module calculates evaluation indicators for the final classification results to evaluate the compactness and separability of the final classification results.
[0134] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.
Claims
1. A method for extracting typical output scenarios of new energy power stations with unsupervised combination, characterized in that: It includes the following steps: Vectorize and normalize the output data of the new energy power station, and calculate a comprehensive similarity index that takes into account both amplitude and morphology; Perform unsupervised pre-classification on the sample data to generate initial classification labels; Construct a supervised multi-classification model of support vector machine based on kernel function mapping, and correct the initial classification labels to output the final classification results; Calculate evaluation indicators for the final classification results to evaluate the compactness and separability of the final classification results.
2. The method for extracting typical output scenarios of new energy power stations with unsupervised combination according to claim 1, wherein: The calculation steps of the comprehensive similarity index include: Vectorize the output data of the new energy power station to form sample vector data; Normalize the sample vector data to the range of the interval (0, 1] with the extreme values of each dimension as the boundaries to obtain sample normalized data; Calculate the amplitude feature distance and the morphology feature distance between different sample normalized data, and the weighted sum of the two is the comprehensive similarity index.
3. The method for extracting typical output scenarios of a new energy power station with unsupervised combination according to claim 2, characterized in that: The amplitude feature distance is the Euclidean distance, and the morphology feature distance is the Pearson correlation coefficient distance.
4. The method for extracting typical output scenarios of new energy power stations with unsupervised combination according to claim 3, characterized in that: The generation method of the initial classification labels includes: Select the range of the number of clusters, and perform multiple unsupervised clusterings on the sample normalized data based on the comprehensive similarity index; Calculate the clustering evaluation indicators, and determine the number of clusters corresponding to the optimal clustering evaluation indicator as the optimal number of clusters based on the pros and cons of the clustering evaluation indicators; Use the unsupervised clustering algorithm to pre-classify the samples based on the optimal number of clusters to generate initial classification labels.
5. A method for extracting typical output scenarios of new energy power stations with unsupervised combination as claimed in claim 4, characterized in that: The method for obtaining the final classification results includes: Use the kernel function to map the samples to a high-dimensional space, and construct an optimal separating hyperplane in this high-dimensional space for the binary classification model of the support vector machine; Adopt a one-to-one multi-classification strategy to expand the binary classification model of the support vector machine into a multi-classification model; Use the initial classification labels and the sample normalized data to train the support vector machine multi-classification model; Use the trained support vector machine multi-classification model to correct the initial classification labels and output the final classification results.
6. A method for extracting typical output scenarios of new energy power stations with unsupervised combination as claimed in claim 5, characterized in that: The steps for evaluating the compactness and separability of the final classification results include: Construct the within-class scatter and between-class scatter of the final classification results; The within-class scatter characterizes the within-group aggregation degree of each group after classification, and the between-class scatter characterizes the between-group separation degree of each group after classification; Calculate the CH coefficient based on the within-class scatter and the between-class scatter, and comprehensively evaluate the compactness and separability of the final classification results through the CH coefficient.
7. A method for extracting typical output scenarios of new energy power stations with unsupervised combination as claimed in claim 5, characterized in that: The kernel function includes at least one of the hyperbolic tangent kernel function, the polynomial kernel function, and the Gaussian kernel function.
8. A typical output scenario extraction system for a new energy power station, which applies a method for extracting typical output scenarios of a new energy power station combining unsupervised and supervised learning as described in any one of claims 1 to 7, characterized in that, It includes: A data preprocessing module, an unsupervised label generation module, a supervised classification correction module, and a classification effect evaluation module; The data preprocessing module vectorizes and normalizes the output data of the new energy power station, and calculates the comprehensive similarity index; The unsupervised label generation module performs unsupervised pre-classification on the sample data to generate initial classification labels; The supervised classification correction module constructs a supervised multi-classification model of support vector machine based on kernel function mapping, and corrects the initial classification labels to output the final classification results; The classification effect evaluation module calculates evaluation indicators for the final classification results to evaluate the compactness and separability of the final classification results.
9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of a method for extracting typical output scenarios of a new energy power station with combined supervised and unsupervised learning as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a method for extracting typical output scenarios of a new energy power station with combined supervised and unsupervised learning as claimed in any one of claims 1 to 7.