A photovoltaic power generation typical scene extraction method based on multi-dimensional similarity and adaptive graph fusion

Abstract

The application discloses a photovoltaic power generation typical scene extraction method based on multi-dimensional similarity and adaptive graph fusion, and belongs to the technical field of new energy consumption and random optimization of power systems. The method firstly carries out standardization and time aggregation pretreatment on historical photovoltaic output data of multiple power stations; secondly, a comprehensive similarity (PEST) measurement model integrating four dimensions of power, energy, form and time sequence is constructed; then, an adaptive graph fusion clustering (MAGFC) unified optimization model is established, each power station is regarded as an independent view, the similarity subgraph, fusion weight, global consensus graph and spectral embedding matrix of each view are automatically learned through optimization, and the optimal clustering number is automatically determined based on graph theory; finally, an alternating direction optimization algorithm is used to solve the model, and representative daily curves and their probabilities are extracted according to the spectral embedding result to form a multi-dimensional typical scene set. The application overcomes the defects of single similarity measurement, poor multi-station cooperation and preset clustering number of traditional methods, and significantly improves the physical fidelity, adaptive ability and engineering practicability of scene extraction.

Description

[0001] This invention relates to the field of power system planning and operation optimization technology, and in particular to a method applicable to high-proportion photovoltaic grid connection, which automatically extracts highly representative and probabilistic typical daily scenarios from massive historical output data of multiple power plants. Background Technology

[0002] The randomness and volatility of photovoltaic power generation output are the main challenges for its large-scale grid connection. Scenario-based stochastic optimization is an effective means to deal with photovoltaic uncertainty. Its core lies in extracting a few highly representative typical daily scenarios from historical data for subsequent probabilistic power flow, unit combination and other analyses.

[0003] Currently, typical scene extraction methods suffer from the following drawbacks: First, the similarity metric is singular. Traditional clustering methods (such as k-means) typically only use Euclidean distance, ignoring multi-dimensional key electrical features such as daily power generation, curve shape, and fluctuation time series, resulting in insufficient scene fidelity. Second, multi-power station coordination is poor. Existing methods often cluster each power station independently or simply piece them together, destroying the inherent spatiotemporal correlation of regional photovoltaic power station clusters and failing to accurately characterize the aggregated power output characteristics. Finally, automation and robustness are weak. Existing methods often require pre-setting the number of clusters, are sensitive to parameters, and produce unstable results that rely on expert experience, limiting engineering applications.

[0004] Therefore, there is an urgent need for a new photovoltaic scene extraction method that can comprehensively consider multi-dimensional electrical characteristics, adaptively fuse information related to multiple power plants, and automatically determine the optimal number of scenes. Summary of the Invention

[0005] To overcome the shortcomings of the existing technology, this invention provides a high-fidelity, adaptive, and robust method for extracting typical photovoltaic power generation scenarios. This invention addresses the three core issues sequentially by constructing a unified optimization framework, ultimately automatically outputting a set of typical scenarios with clearly defined probabilities.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for extracting typical photovoltaic power generation scenarios based on multidimensional similarity and adaptive graph fusion, comprising the following steps: S10: standardizing and time-scale aggregation preprocessing of historical daily power output data from multiple photovoltaic power plants; S20: constructing a four-dimensional photovoltaic power output comprehensive similarity measurement model to calculate the weighted distance between any two daily curves; S30: establishing an adaptive graph fusion clustering unified optimization model, taking each power plant as a view, optimizing and learning the global consensus similarity graph and spectral embedding matrix, and automatically determining the number of clusters; S40: using an alternating direction optimization algorithm to iteratively solve the unified optimization model, clustering all dates according to the solution results, selecting representative days for each cluster, combining data from multiple power plants to form a set of typical scenarios and calculating their probabilities.

[0007] The beneficial effects of this invention are as follows: by introducing a four-dimensional similarity metric, the physical fidelity of the scene is significantly improved; through an adaptive graph fusion mechanism, the spatiotemporal correlation between multiple power stations is effectively characterized; and by automatically determining the number of clusters through graph theory, the entire process is automated and highly robust. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is an overall flowchart of the photovoltaic power generation typical scenario extraction method described in this invention.

[0010] Figure 2 This is a schematic diagram of the four-dimensional photovoltaic power output comprehensive similarity (PEST) model.

[0011] Figure 3 This is a structural diagram of the Adaptive Graph Fusion Clustering (MAGFC) model.

[0012] Figure 4 The flowchart shows the iterative solution process of the Alternating Direction Optimization (ADMM) algorithm. Detailed Implementation

[0013] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0014] Unless otherwise defined, 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 invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0015] like Figure 1 As shown, this application provides a method for extracting typical photovoltaic power generation scenarios based on multidimensional similarity and adaptive graph fusion. The method includes:

[0016] S10: Standardize and time-scale aggregate preprocessing the historical daily power output data of multiple photovoltaic power plants; S20: Construct a four-dimensional photovoltaic power output comprehensive similarity measurement model to calculate the weighted distance between any two daily curves; S30: Establish an adaptive graph fusion clustering unified optimization model, using each power plant as a view, and automatically determine the number of clusters by optimizing the global consensus similarity graph and spectral embedding matrix; S40: Iteratively solve the unified optimization model using an alternating direction optimization algorithm, cluster all dates according to the solution results, select representative days for each cluster, combine data from multiple power plants to form a typical scenario set, and calculate its probability.

[0017] Specifically, the first step is to collect historical daily power output data from multiple photovoltaic power plants in the region, typically a year's worth of sampling data (365 days), with a sampling frequency of 5 or 15 minutes. This raw data forms the basis for all subsequent analysis and modeling.

[0018] As a preferred embodiment of the above embodiments, data standardization preprocessing is a key step in implementing step S10 to ensure the effectiveness of subsequent models. For the... Each power station, its raw data forming a matrix ,in For the number of valid days (e.g., 365). This represents the number of sampling points within a day. First, data cleaning is performed to correct or remove outliers that are significantly outside the physical limits (such as negative output or exceeding rated capacity). Then, to eliminate differences in installed capacity between different power plants, the data for each power plant is independently subjected to min-max normalization.

[0019]

[0020] In formula (1), Indicates the first The power station, the first Heaven, the First Normalized output values ​​of each sampling point. and These represent taking the minimum and maximum values ​​of all elements in the matrix, respectively. This operation scales the data from each power station independently to the same interval, facilitating a subsequent focus on the relative shape and variation patterns of the output curves. To meet the long-term planning requirements of the power system for hourly resolution, high-frequency normalized data needs to be aggregated into hourly data:

[0021]

[0022] In formula (2), Indicates the first The power station, the first Heaven, the First Average normalized output per hour; It is the number of sampling points within the aggregation window; Indicates the first This is the set of indices of all original sampling times corresponding to each hour. Through this step, the final standardized data matrix for each power station is obtained. Each row of this matrix It represents a 24-dimensional standardized daily power curve and serves as the basic input unit for subsequent models.

[0023] As a preferred embodiment of the above, constructing a reasonable similarity metric model is the core of extracting high-fidelity scenes during step S20. The limitation of traditional Euclidean distance is that it only focuses on the power difference at the same time. To comprehensively measure the two diurnal curves... and To measure the differences between them, this invention measures them from four physically and engineeringally significant dimensions: power, energy, shape, and timing, and combines them in a weighted manner to form a comprehensive distance:

[0024] In formula (3), It is the first The learnable weight vector corresponding to each power station satisfies and The model automatically learns these weights through an optimization process to reflect the relative importance of distances in different dimensions within a specific power plant dataset. For example, for power plants with varied output curve shapes, shape weights are used. It may learn larger values. , , , Representing four dimensions respectively The basic distance function is defined as follows:

[0025] (1) Power dimension distance The squared Euclidean distance is used to measure the overall similarity of the power values ​​of the curves at each corresponding hour point.

[0026]

[0027] (2) Energy dimensional distance The area under the curve (AUC) measures the relative difference in the total daily power generation represented by the two curves, which is crucial for system energy balance analysis.

[0028] ,in

[0029] (3) Morphological dimension distance Based on normalized cross-correlation (NCC), this invention measures the similarity of two curves in shape and trend, even if they are shifted on the time axis (e.g., different peak times). This is one of the key innovations of this invention. First, the displacement is calculated. (-twenty three 23) Cross-correlation Then, the normalized cross-correlation sequence was calculated. The morphological distance is defined as:

[0030]

[0031] When the two curves have exactly the same shape, the maximum When the value is 1, the distance is 0; when the shape is completely reversed, the distance is 2.

[0032] (4) Temporal dimension distance This measures the difference in the timing of major fluctuations (such as sudden increases or decreases in output) between the two curves. This is of great significance for assessing the system's peak-shaving pressure.

[0033] ,in

[0034] The absolute value of the displacement that maximizes the cross-correlation reflects the time offset required to align one curve to another.

[0035] As a preferred embodiment of the above, the key to this invention is establishing a model capable of fusing information from multiple power plants and automatically discovering cluster structures during step S30. Each photovoltaic power plant is considered an independent "view," whose data contains partial observations of the system state. The goal of this model is to fuse information from all views and learn a global similarity graph with a clear cluster structure. And automatically determine the number of clusters. Define decision variables: (No. (Similarity subgraph of individual power stations) (Global consensus graph) (Adaptive fusion weight vector). (Spectral embedding matrix). (picture (The Laplace matrix). Establish the following unified optimization problem:

[0036] Constraints such as probability normalization, nonnegativity, symmetry, and orthogonality are applied. The fourth term of the objective function... It enables automatic determination of the number of clusters. The core mechanism. According to Ky Fan's theorem, minimizing this term is equivalent to driving... The former The smallest eigenvalues ​​tend to zero. Graph theory shows that graphs... The number of connected components is equal to The multiplicity of zero eigenvalues. Therefore, the optimization process adjusts... and Forced Having just Connected components (i.e.) (individual clusters), thus enabling... It becomes the implicit output of the model and does not need to be specified in advance.

[0037] As a preferred embodiment of the above, designing an efficient solution algorithm is fundamental to ensuring the practicality of the method during step S40. Model (8) is a complex, multivariable, constrained non-convex optimization problem. This invention designs an efficient solution algorithm based on the Alternating Direction Multiplier Method (ADMM) framework, which solves the convex subproblem with respect to each variable group by fixing other variables.

[0038] Initialization: Set the number of iterations Uniform initialization Let the initial fusion weights be... The k-nearest neighbor method, based on simple Euclidean distance, is used to initially construct a similarity subgraph for each power station. Calculate the initial consensus graph. .right Perform eigenvalue decomposition and take the first few features. The eigenvectors corresponding to the smallest eigenvalues ​​are arranged in columns, and the spectral embedding matrix is ​​initialized. .

[0039] Iterative update (for) (until convergence)

[0040] 1. Update , :fixed , , For each power station Solve independently. (Updated) No. Row elements have closed-form solutions:

[0041]

[0042] in For comprehensive measurement, These are Lagrange multipliers, which can be efficiently obtained through sorting and thresholding.

[0043] 2. Update fusion weights :fixed The problem simplifies to a linear programming problem, and its optimal solution is a closed-form solution:

[0044]

[0045] This solution intuitively reflects the adaptive principle that "the more consistent the subgraph is with the current consensus graph, the greater its weight."

[0046] 3. Update the global consensus graph :fixed The problem can be decomposed into element-wise optimization, and its optimal solution is:

[0047]

[0048] Then, symmetry is applied: .

[0049] 4. Update the spectral embedding matrix :fixed The problem degenerates into minimizing the Rayleigh quotient:

[0050]

[0051] Optimal solution Depend on The former The eigenvectors are formed by the smallest eigenvalues.

[0052] 5. Convergence Criterion: Calculate the changes .like If the threshold is not met, the iteration stops; otherwise, let... continue.

[0053] As a preferred embodiment of the above, in implementing step S40, the extraction of typical scenarios needs to ensure their representativeness and practicality. After the algorithm converges, the final spectral embedding matrix is ​​obtained. ,in The final number of clusters is automatically determined by the model.

[0054] 1. Final clustering: For Perform k-means clustering (set) ),Will Tianfen was divided into Mutually exclusive clusters middle.

[0055] 2. Calculate the scenario probability: for each cluster The probability of occurrence of the corresponding typical scenario is: ,in The number of days within the cluster.

[0056] 3. Select representative days: For each cluster In the original output curve space (i.e., 24-dimensional space), calculate the PEST combined distance between all curves within a cluster and the cluster center (mean curve). Select the day with the smallest distance. As a typical day for this cluster.

[0057] 4. Constructing multi-dimensional typical scenarios: For typical days Collect all Standardized 24-hour output curve of a photovoltaic power station on that day Together they constitute the first A multi-dimensional typical scenario .

[0058] 5. Output: The final output is a collection of typical scenarios. and their corresponding probability distribution This output can be directly used for advanced applications such as probabilistic power flow calculations, stochastic unit combination, and reactive power optimization in power systems with a high proportion of photovoltaic power.

[0059] Through the above complete process, this invention realizes a fully automated, closed-loop solution from raw data processing, unified model construction, efficient optimization and solution to final scene extraction, which significantly improves the accuracy, adaptability and engineering practicality of typical photovoltaic power generation scene extraction.

[0060] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and accompanying drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. It is obvious that those skilled in the art can make various alterations and modifications to this application without departing from the scope of this application. Thus, if such modifications and modifications of this application fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

[0061] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for extracting typical photovoltaic power generation scenarios based on multidimensional similarity and adaptive graph fusion, characterized in that, The process includes the following steps: S10: Standardize and time-scale aggregate preprocessing the historical daily power output data of multiple photovoltaic power plants; S20: Construct a four-dimensional photovoltaic power output comprehensive similarity measurement model to calculate the weighted distance between any two daily curves; S30: Establish an adaptive graph fusion clustering unified optimization model, using each power plant as a view, and automatically determine the number of clusters by optimizing and learning the global consensus similarity graph and spectral embedding matrix; S40: Iteratively solve the unified optimization model using an alternating direction optimization algorithm, cluster all dates according to the solution results, select representative days for each cluster, combine data from multiple power plants to form a typical scenario set, and calculate its probability.

2. The method according to claim 1, characterized in that, Step S10 includes: S11: cleaning the raw power output data of each power station, correcting or removing outliers that exceed the physical range; S12: performing minimum-maximum normalization on the cleaned power station data independently, mapping it to the [0,1] interval; S13: aggregating the normalized high-frequency data into hourly average power output data, forming a standardized power output curve matrix with each curve representing one day and having a dimension of 24.

3. The method according to claim 1, characterized in that, In step S20, the comprehensive similarity measurement model is the PEST model. One power station, two daily curves and The comprehensive distance between them is defined as: in, For the first The learnable weight vector of each power station satisfies and ; The squared Euclidean distance is the power dimension. The distance represents the relative difference in energy dimension. Based on the morphological distance of normalized cross-correlation, The time-series distance is based on the maximum cross-correlation displacement.

4. The method according to claim 3, characterized in that, The morphological distance The calculations include: calculating the two curves at different displacements. The following cross-correlation ; Calculate the normalized cross-correlation sequence ; morphological distance is defined as The time-series distance Defined as ,in .

5. The method according to claim 1, characterized in that, In step S30, the objective function of the adaptive graph fusion clustering unified optimization model is: in, For the first Similarity subgraph of each power station The global consensus graph to be sought. For adaptive fusion weight vector, For the spectral embedding matrix, For the image The Laplace matrix; by minimizing drive The former Each eigenvalue tends to zero, thus forcing With just Connected components, achieving clustering number Automatic determination.

6. The method according to claim 1, characterized in that, Step S40 specifically employs the alternating direction multiplier method framework, sequentially and alternately updating the following variable groups until convergence: the similarity subgraph of each power station. With PEST weight Adaptive fusion weights Global consensus graph Spectral embedding matrix .

7. The method according to claim 1, characterized in that, Step S40 includes: S41: Performing k-means clustering on the row vectors of the converged spectral embedding matrix to divide all historical days into... Within each cluster; S42: The ratio of the number of days in each cluster to the total number of days is used as the probability of occurrence of the corresponding typical scenario; S43: Within each cluster, the PEST comprehensive distance between all daily curves and the cluster center curve is calculated in the original power output curve space, and the date with the smallest distance is selected as the representative day of the cluster; S44: The standardized power output curves of all power plants on the representative day are summarized to form a multi-dimensional typical scenario.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images