A light-load combined typical scene generation method and device

Abstract

The present invention discloses a method and device for generating typical solar-load scenarios, relating to the technical field of distribution network planning. The method comprises: determining target solar-load and meteorological-photovoltaic datasets through correlation coefficient analysis based on meteorological-photovoltaic datasets and meteorological-load datasets, inputting these datasets into an improved time-series generative adversarial network model for processing to obtain enhanced solar-photovoltaic and meteorological-load datasets; performing dimensionality reduction and interactive clustering on the enhanced meteorological data of the enhanced meteorological-photovoltaic and meteorological-load datasets to obtain photovoltaic-side meteorological scenarios and load-side meteorological scenarios; and performing cluster analysis on the enhanced meteorological-photovoltaic and meteorological-load datasets, respectively, and then performing global-local time-series matching on the obtained photovoltaic clusters for each photovoltaic-side meteorological scenario and the load clusters for each load-side meteorological scenario to generate a set of solar-load typical scenarios. This scheme improves the reliability of the generated solar-load typical scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network planning, and in particular to a method and device for generating a typical light-load combined scenario. Background Art

[0002] With the rapid development of distributed photovoltaic power generation in my country, the amount of photovoltaic power connected to distribution networks has increased rapidly in recent years. In distribution network planning projects that include photovoltaics, the presence of photovoltaic power generation transforms traditional distribution networks into active distribution networks. At the same time, photovoltaic output is significantly affected by environmental factors, resulting in randomness and volatility. Distribution network load is not only closely related to people's production and living cycles, but is also subject to temperature, humidity, precipitation, and holidays. Therefore, load and photovoltaic power generation scenarios are highly variable. Distribution networks are subject to the combined effects of distributed photovoltaic and load uncertainties, resulting in numerous conflicts and overlaps between scenarios. Therefore, in order to scientifically describe the current status of distribution networks and enable accurate predictions of regional photovoltaic-load combined scenarios, scenario generation technology that accurately models the combined behavior of photovoltaics and power loads is crucial.

[0003] In order to accurately describe the original characteristics of photovoltaic output and power load time series scenarios, some studies have generated light-load joint scenarios through parametric and non-parametric methods. The existing scenario generation method based on parameter methods makes assumptions about the probability distribution of data and then uses sampling to generate scenarios, but simple statistical assumptions are difficult to describe the high-dimensional nonlinear characteristics of the data. The existing non-parametric methods are usually machine learning and deep learning methods, which can learn the true distribution of uncertain variables based on data-driven models, but have difficulties in learning the dynamic characteristics, fluctuations, trends and periodic factors within the time series data, resulting in low reliability of the generated typical light-load joint scenarios. Summary of the Invention

[0004] The present invention provides a method and device for generating a typical light-load joint scene, which solves the technical problem that in the prior art, when generating a light-load joint scene, the scene generation method based on the parametric method is difficult to describe the high-dimensional nonlinear characteristics of the data, and the scene generation method based on the non-parametric method has difficulty in learning the dynamic characteristics, fluctuations, trends and periodic factors within the time series data, resulting in low reliability of the generated light-load joint typical scene.

[0005] A first aspect of the present invention provides a light-load combined typical scene generation method, comprising:

[0006] Obtain target photovoltaic output data, multiple target meteorological data, and target load data within a preset time range, and perform multi-source data fusion on each target meteorological data with each target photovoltaic output data and each target load data to obtain a meteorological photovoltaic data set and a meteorological load data set;

[0007] Based on the correlation coefficients among the target photovoltaic output data, the target meteorological data, and the target load data, redundancy of meteorological data is eliminated from the meteorological photovoltaic dataset and the meteorological load dataset to determine a target meteorological photovoltaic dataset and a target meteorological load dataset;

[0008] An improved temporal generative adversarial network model is used to perform data enhancement on the target meteorological photovoltaic dataset and the target meteorological load dataset, respectively, to obtain an enhanced meteorological photovoltaic dataset and an enhanced meteorological load dataset;

[0009] Performing dimensionality reduction on the enhanced meteorological data in the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset, and performing interactive clustering to obtain a photovoltaic side meteorological scene and a load side meteorological scene;

[0010] Performing cluster analysis on the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset respectively to obtain photovoltaic clusters of each of the photovoltaic-side meteorological scenarios and load clusters of each of the load-side meteorological scenarios;

[0011] Based on the global-local time series matching of the meteorological day scenes of each photovoltaic cluster and each load cluster, a set of typical photovoltaic-load combined scenes is generated.

[0012] Optionally, the acquiring of target photovoltaic output data, multiple target meteorological data, and target load data within a preset time range, and performing multi-source data fusion on each target meteorological data with each target photovoltaic output data and each target load data to obtain a meteorological photovoltaic dataset and a meteorological load dataset, include:

[0013] Obtain initial photovoltaic output data, various initial meteorological data, and initial load data within a preset time range;

[0014] performing data cleaning on each of the photovoltaic output data, each of the meteorological data, and each of the load data to obtain target photovoltaic output data, target meteorological data, and target load data;

[0015] fusing the target meteorological data with the target photovoltaic output data according to time to generate a meteorological photovoltaic data set;

[0016] The target meteorological data and the target load data are fused according to time to determine a meteorological load data set.

[0017] Optionally, the performing meteorological data redundancy elimination on the meteorological photovoltaic dataset and the meteorological load dataset based on the correlation coefficients among the target photovoltaic output data, the target meteorological data, and the target load data, and correspondingly determining the target meteorological photovoltaic dataset and the target meteorological load dataset, includes:

[0018] Calculating the peak-to-valley difference of each target meteorological data on a daily basis to obtain a plurality of corresponding peak-to-valley difference meteorological data;

[0019] The Pearson correlation coefficient method is used to calculate the correlation coefficient between each target photovoltaic output data, each target load data and each peak-to-valley difference meteorological data;

[0020] Selecting peak-valley difference meteorological data having a correlation coefficient greater than a first correlation threshold with both the target photovoltaic output data and the target load data as target peak-valley difference meteorological data;

[0021] If there are multiple categories of the target peak-to-valley difference meteorological data, then based on the correlation coefficients between the various target peak-to-valley difference meteorological data, one category of the target peak-to-valley difference meteorological data having a correlation coefficient greater than a second correlation threshold is selected as the dimensionality reduction meteorological scene;

[0022] Target meteorological data associated with the dimension-reduced meteorological scene are respectively removed from the meteorological photovoltaic dataset and the meteorological load dataset to obtain a target meteorological photovoltaic dataset and a target meteorological load dataset.

[0023] Optionally, the dimensionality reduction of the enhanced meteorological data in the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset is performed respectively, and interactive clustering is performed to obtain a photovoltaic side meteorological scene and a load side meteorological scene, including:

[0024] Performing peak-to-valley difference calculation and average value calculation on the enhanced meteorological data in the enhanced meteorological photovoltaic dataset on a daily basis to obtain a plurality of photovoltaic-side dimension-reduced meteorological data;

[0025] Performing peak-to-valley difference calculation and average value calculation on the enhanced meteorological data in the enhanced meteorological load data set on a daily basis to obtain a plurality of load-side dimension-reduced meteorological data;

[0026] Clustering the photovoltaic-side dimension-reduced meteorological data to obtain a plurality of photovoltaic-side meteorological scenarios and photovoltaic clustering parameters;

[0027] The photovoltaic clustering parameters are used as load clustering parameters to perform cluster analysis on the load-side dimension-reduced meteorological data to obtain multiple load-side meteorological scenarios.

[0028] Optionally, performing cluster analysis on the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset respectively to obtain photovoltaic clusters of each photovoltaic-side meteorological scenario and load clusters of each load-side meteorological scenario includes:

[0029] According to each photovoltaic side meteorological scene, corresponding enhanced photovoltaic data are extracted from the enhanced meteorological photovoltaic data set for clustering to obtain photovoltaic clusters of each photovoltaic side meteorological scene;

[0030] According to each of the load-side meteorological scenarios, corresponding enhanced load data are extracted from the enhanced meteorological load data set for clustering to obtain load clusters of each of the load-side meteorological scenarios.

[0031] Optionally, performing global-local time series matching of meteorological day scenarios based on each of the photovoltaic clusters and each of the load clusters to generate a set of typical solar-load combined scenarios includes:

[0032] taking the enhanced photovoltaic data closest to the corresponding cluster center in each photovoltaic cluster as the photovoltaic typical scene of each photovoltaic cluster;

[0033] Taking the enhanced load data closest to the corresponding cluster center in each load cluster as the load typical scenario of each load cluster;

[0034] Taking the PV side meteorological scene and the load side meteorological scene in the same meteorological scene as a unit, the enhanced meteorological data of each typical PV scene on the day and the corresponding enhanced meteorological data of each typical load scene on the day are used to calculate the dynamic time warping distance and matrix profile distance;

[0035] Performing comprehensive processing on the dynamic time warping distances and the associated matrix profile distances to obtain a plurality of comprehensive distances for the typical photovoltaic scenarios;

[0036] Traversing each of the comprehensive distances, and combining each of the photovoltaic typical scenarios with the corresponding load typical scenarios associated with the minimum comprehensive distance into a photovoltaic-load combined typical scenario pair;

[0037] The light-load combined typical scene pairs are used to construct a light-load combined typical scene set according to the meteorological scene.

[0038] A second aspect of the present invention provides a light-load combined typical scene generation device, comprising:

[0039] A data processing module is used to obtain target photovoltaic output data, multiple target meteorological data, and target load data within a preset time range, and perform multi-source data fusion on each target meteorological data with each target photovoltaic output data and each target load data to obtain a meteorological photovoltaic data set and a meteorological load data set;

[0040] a data elimination module, configured to eliminate meteorological data redundancy from the meteorological photovoltaic dataset and the meteorological load dataset based on correlation coefficients among the target photovoltaic output data, the target meteorological data, and the target load data, and determine a target meteorological photovoltaic dataset and a target meteorological load dataset;

[0041] a data enhancement module, configured to perform data enhancement on the target meteorological photovoltaic dataset and the target meteorological load dataset respectively using an improved temporal generative adversarial network model to obtain an enhanced meteorological photovoltaic dataset and an enhanced meteorological load dataset;

[0042] A dimensionality reduction clustering module is used to reduce the dimensionality of the enhanced meteorological data in the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset, and perform interactive clustering to obtain the photovoltaic side meteorological scene and the load side meteorological scene;

[0043] A photovoltaic load clustering module is used to perform cluster analysis on the enhanced meteorological photovoltaic data set and the enhanced meteorological load data set respectively to obtain photovoltaic clusters of each photovoltaic-side meteorological scenario and load clusters of each load-side meteorological scenario;

[0044] The typical scenario generation module is used to perform global-local time series matching of meteorological day scenarios based on each photovoltaic cluster and each load cluster to generate a set of typical photovoltaic-load combined scenarios.

[0045] A third aspect of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the light-load combined typical scene generation method as described in any one of the above items.

[0046] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the light-load combined typical scene generation method as described in any one of the above items.

[0047] A fifth aspect of the present invention provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the light-load combined typical scene generation method as described in any one of the above items.

[0048] It can be seen from the above technical solutions that the present invention has the following advantages:

[0049] The above-mentioned solution of the present invention provides a method for generating typical solar-load combined scenarios, including: obtaining target photovoltaic output data, multiple target meteorological data, and target load data within a preset time range, performing multi-source data fusion on each target meteorological data with each target photovoltaic output data and each target load data to obtain a meteorological photovoltaic dataset and a meteorological load dataset; eliminating meteorological data redundancy from the meteorological photovoltaic dataset and the meteorological load dataset based on correlation coefficients among the target photovoltaic output data, target meteorological data, and target load data to determine a target meteorological photovoltaic dataset and a target meteorological load dataset; using an improved time series generative adversarial network model to perform data enhancement on the target meteorological photovoltaic dataset and the target meteorological load dataset to obtain an enhanced meteorological photovoltaic dataset and an enhanced meteorological load dataset; performing dimensionality reduction on the enhanced meteorological data in the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset, and performing interactive clustering to obtain a photovoltaic-side meteorological scenario and a load-side meteorological scenario; performing cluster analysis on the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset to obtain a photovoltaic cluster for each photovoltaic-side meteorological scenario and a load cluster for each load-side meteorological scenario; and performing global-local time series matching of meteorological day scenarios based on each photovoltaic cluster and each load cluster to generate a set of solar-load combined typical scenarios. Based on the above scheme, the dynamic and static characteristics of the target meteorological photovoltaic dataset and the target meteorological load dataset are fully exploited through the improved time series generative adversarial network model, allowing the enhanced meteorological data on the photovoltaic side and the load side to interact with each other in the clustering process, ensuring the consistency of the clustering results on both sides while improving the clustering efficiency. The global-local time series matching method takes into account both the global characteristics and local details of the time series data, and can more accurately match the photovoltaic and load scenarios. In the entire process from correlation analysis to scenario matching, objective meteorological data is fully utilized to couple the photovoltaic and load scenarios, reducing the influence of subjective factors to a certain extent, improving the authenticity and representativeness of the generated scenarios as a whole, and improving the reliability of the generated typical photovoltaic-load joint scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 A flowchart of a method for generating a typical light-load combined scene according to an embodiment of the present invention;

[0052] Figure 2 A data processing flow chart of a light-load combined typical scene generation method provided by an embodiment of the present invention;

[0053] Figure 3 A schematic diagram of the structure of the Autoformer-TimeGAN model provided in an embodiment of the present invention;

[0054] Figure 4 This is a structural block diagram of a light-load combined typical scene generation device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The embodiments of the present invention provide a method and device for generating a typical light-load joint scene, which is used to solve the technical problem that in the prior art, when generating a light-load joint scene, the scene generation method based on the parametric method is difficult to describe the high-dimensional nonlinear characteristics of the data, and the scene generation method based on the non-parametric method has difficulty in learning the dynamic characteristics, fluctuations, trends and periodic factors within the time series data, resulting in low reliability of the generated light-load joint typical scene.

[0056] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0057] See also Figure 1 , Figure 1 A flowchart of the steps of a light-load combined typical scene generation method provided by an embodiment of the present invention.

[0058] The present invention provides a light-load combined typical scene generation method, comprising:

[0059] Step 101: Obtain target photovoltaic output data, multiple target meteorological data, and target load data within a preset time range, and perform multi-source data fusion on each target meteorological data with each target photovoltaic output data and each target load data to obtain a meteorological photovoltaic data set and a meteorological load data set.

[0060] Step 101 includes the following sub-steps:

[0061] S11, obtaining initial photovoltaic output data, various initial meteorological data, and initial load data within a preset time range;

[0062] S12, performing data cleaning on each photovoltaic output data, each meteorological data, and each load data to obtain target photovoltaic output data, target meteorological data, and target load data;

[0063] S13, fusing each target meteorological data with each target photovoltaic output data according to time to generate a meteorological photovoltaic data set;

[0064] S14. Fusing each target meteorological data with each target load data according to time to determine a meteorological load data set.

[0065] It should be noted that if Figure 2 As shown, first, initial photovoltaic output data, multiple initial meteorological data, and initial load data are obtained according to a preset time range. This time range can be a range of multiple days, one or more months, or one or more years, and can be set as needed. The initial meteorological data includes multiple objective meteorological factors that affect photovoltaic output and load, such as temperature, humidity, precipitation, and wind speed;

[0066] In order to lay the foundation for the subsequent comprehensive analysis of the three types of data, it is necessary to ensure that the three types of data are aligned in time and length. In specific implementation, the time alignment is reflected in that the time of any type of data among meteorological, photovoltaic output and load data should be consistent. For example, when the time range of meteorological data is the whole year of 2023, the time range of photovoltaic output and load data should also be the whole year of 2023. The length alignment is reflected in that the length of any type of data among meteorological, photovoltaic output and load data should be consistent. For example, when meteorological data is collected at 96 points per day and 365 days per year, the length is 365*96. Then, when photovoltaic output and load data are collected at 96 points per day and 365 days per year, the length is 365*96.

[0067] In order to achieve consistency among the three types of data on a time scale, it is necessary to perform data cleaning on the initial photovoltaic output data, initial meteorological data, and initial load data, mainly including processing duplicate values, missing values, and outliers. In specific implementation, processing duplicate values ​​means identifying and deleting duplicate records in the data. For example, if duplicates occur at the data point at 12:00 on January 1, 2023, the latest or earliest record is retained according to actual needs, and other duplicates are deleted. Processing missing values ​​means filling or deleting blanks or invalid values ​​in the data. For example, if the initial photovoltaic output Force data is missing between 14:00 and 16:00 on July 15, 2023. Linear interpolation can be used to fill in the missing data using data before and after this period. If the missing time is longer than the preset missing time range, it is considered that the missing time is long, and historical data from the same time period of the adjacent data can be considered for filling in the missing data. Processing outliers means identifying and processing data that does not conform to the normal range. For example, if the initial load data of a certain day suddenly reaches an abnormally high point that far exceeds the historical peak, statistical methods (such as the triple standard deviation method) can be used to identify this outlier and replace it with the historical average or median of the period.

[0068] After data cleaning is completed, the target photovoltaic output data, target meteorological data and target load data are output. According to the time of each data, the target meteorological data and the target photovoltaic output data are fused to obtain a photovoltaic output data set containing meteorological factors, namely, a meteorological photovoltaic data set. The target meteorological data and the target load data are fused to generate a load data set containing meteorological factors, namely, a meteorological load data set.

[0069] Step 102: Based on the correlation coefficients among the target photovoltaic output data, the target meteorological data, and the target load data, redundancy of meteorological data is eliminated from the meteorological photovoltaic dataset and the meteorological load dataset to determine the target meteorological photovoltaic dataset and the target meteorological load dataset.

[0070] Step 102 includes the following sub-steps:

[0071] S21. Calculate the peak-to-valley difference of various target meteorological data on a daily basis, and obtain a plurality of corresponding peak-to-valley difference meteorological data.

[0072] It should be noted that in this embodiment, the characteristics of weather, PV output, and load are considered. To determine which weather factors are highly correlated with PV output and load, the peak-valley difference (PV diff) of daily data within the multi-day data contained in various target weather data is used as a feature. This data can well capture the fluctuation characteristics of the data, which is crucial for understanding the dynamic relationship between weather, PV output, and load. The peak-valley difference calculation process includes the following:

[0073] ;

[0074] Where, Represents the j-th day data of the i-th meteorological factor.

[0075] S22. Use the Pearson correlation coefficient method to calculate the correlation coefficients between each target photovoltaic output data, each target load data, and each peak-to-valley difference meteorological data.

[0076] It should be noted that the process of calculating the correlation coefficient using the Pearson correlation coefficient method includes:

[0077] ;

[0078] Where, is the first variable, is the second variable, is the covariance between the variables, is the standard deviation between the variables, is the correlation coefficient between variables;

[0079] According to relevant theories, when When the correlation is around 0.5, the two variables are considered to have a strong correlation.

[0080] In specific implementation, after using the peak-valley difference target meteorological data for dimensionality reduction, the Pearson correlation coefficient method is used to calculate the correlation coefficient between the peak-valley difference meteorological data and the target photovoltaic output data and the target load data. For example, the correlation coefficient matrix can be constructed: as follows:

[0081] ;

[0082] In the correlation coefficient matrix, from top to bottom and from left to right, they are temperature, humidity, precipitation, photovoltaic output and load.

[0083] S23 , selecting peak-valley difference meteorological data whose correlation coefficient with the target photovoltaic output data and the target load data is greater than a first correlation threshold as the target peak-valley difference meteorological data.

[0084] It should be noted that, according to the correlation coefficient matrix, temperature, humidity, and precipitation are all highly correlated with photovoltaic output and load. The degree of this high correlation can be determined by setting the first correlation threshold. Therefore, it can be considered that these three types of meteorological data have a strong correlation with photovoltaic output and load, and can be used as target peak-to-valley difference meteorological data.

[0085] S24. If there are multiple categories of target peak-valley difference meteorological data, based on the correlation coefficients between the various target peak-valley difference meteorological data, one category of target peak-valley difference meteorological data having a correlation coefficient greater than a second correlation threshold is selected as a dimensionality reduction meteorological scenario.

[0086] It should be noted that if there are multiple target peak-valley difference meteorological data, then it is necessary to consider whether there is redundancy between the various target peak-valley difference meteorological data. From the correlation coefficient matrix, it can be seen that the correlation coefficient between temperature and humidity is 0.86, and the correlation is very high. The degree of this high correlation can be determined by setting a second correlation threshold. From this, it can be determined that the two are redundant with each other, and one of the target peak-valley difference meteorological data that needs to be redundantly eliminated can be selected as a dimensionality reduction meteorological scenario.

[0087] S25. Target meteorological data associated with the dimension-reduced meteorological scene are respectively removed from the meteorological photovoltaic dataset and the meteorological load dataset to obtain a target meteorological photovoltaic dataset and a target meteorological load dataset.

[0088] It should be noted that if Figure 2As shown, according to the correlation coefficient analysis, the target meteorological data associated with the dimensionality reduction meteorological scene is eliminated from the meteorological photovoltaic dataset and the meteorological load dataset. For example, the target meteorological data excluding humidity can be selected, thereby selecting the two meteorological factors of temperature and precipitation and using the peak-to-valley difference as the feature, which can characterize the correlation and meteorological correlation between the meteorological factors and the photovoltaic output and load, and obtain a photovoltaic dataset containing mutually non-redundant meteorological data, namely the target meteorological photovoltaic dataset, and a load dataset containing mutually non-redundant meteorological data, namely the target meteorological load dataset.

[0089] Step 103: Use the improved time series generative adversarial network model to perform data enhancement on the target meteorological photovoltaic dataset and the target meteorological load dataset respectively to obtain an enhanced meteorological photovoltaic dataset and an enhanced meteorological load dataset.

[0090] It should be noted that in this embodiment, the improved time-series generative adversarial network model refers to the TimeGAN (Time-series generative adversarial networks, time-series generative adversarial network) model (Autoformer-TimeGAN) improved based on Autoformer. For photovoltaic output, the photovoltaic output size in different regions and environments is a static feature, and dynamic factors such as season, weather, and time constitute dynamic feature influencing factors. The power load is closely related to factors such as people's living habits and economic level, so it constitutes a static feature. Changes with time, weekdays and holidays, and weather are dynamic feature influencing factors. The original TimeGAN model combines the autoregressive model and GAN (Generative adversarial networks). In addition to the unsupervised adversarial loss of the real sequence and the generated sequence in the traditional GAN, it also introduces the original data supervision loss. By training the above two losses, the generated sequence can fully restore the static and dynamic features of the time series data. On this basis, Autoformer-TimeGAN uses Autoformer to replace the LSTM (Long short-term Compared to LSTM, which processes time series as a whole without explicit data decomposition, Autoformer introduces a time series decomposition module. This feature enables the model to accurately learn the short-term fluctuations, long-term trend changes, and multi-scale periodicity of photovoltaic output and load time series data. This enables Autoformer-TimeGAN to fully learn the internal characteristics of the original time series data. It then generates enhanced meteorological photovoltaic datasets and enhanced meteorological load datasets based on the target meteorological photovoltaic dataset and target meteorological load dataset through an improved time series generative adversarial network model.

[0091] The Autoformer-TimeGAN model consists of four parts: generator, discriminator, embedding function and recovery function. The specific structure is as follows Figure 3 shown; represents all vectors in the dynamic feature space, represents all vectors in the static feature space, Represents the discriminator’s discrimination result, represents Gaussian noise, represents the static feature space after dimensionality reduction by the embedding function, Represents the dynamic feature space at time t after dimensionality reduction; the structures of each part of the Autoformer-TimeGAN model are described as follows:

[0092] The embedding function processes static and dynamic features through recursion. For static features, the embedding function projects them into a low-dimensional space. For dynamic features, the relationship between time points is mined and projected into a low-dimensional space: , , where represents the processing of static features by the embedding function, represents the processing of dynamic features by the embedding function, represents the static features after dimensionality reduction by the embedding function, represents the dynamic characteristics at time t after dimensionality reduction, Represents the high-dimensional dynamic features at time t;

[0093] The reconstruction function reconstructs the low-dimensional vector after dimensionality reduction into a high-dimensional vector in the original space. The process is as follows: , , where Represents the processing of low-dimensional static features by the reconstruction function, Represents the processing of dynamic features by the reconstruction function, Represents the static features after reconstruction, Represents the dynamic characteristics at time t after reconstruction;

[0094] The generator randomly extracts vectors from the vector space of static and dynamic features and combines the input conditions and inputs them into the low-dimensional latent space. The process is as follows: , , where A generative network representing static features, A generative network representing dynamic features, represents the sampling of the vector space to which the static features belong, represents the sampling of the vector space to which the dynamic features belong, represents the generated static feature vector set, represents the dynamic feature vector set generated at time t, Represents the dynamic feature vector set generated at time t-1;

[0095] The output of the generator and the output of the embedding function are jointly encoded and input into the discriminator. The discriminator will then judge the truth or falsehood based on the real data, the generated data, and the input conditions. If the discriminator outputs 1, it is true, and 0, it is false. The process is as follows: , , where Represents the discrimination result of the static features of the input data, Represents the discriminant result of the dynamic characteristics of the input data, The discriminant network representing static features, The discriminant network representing dynamic features, when using a bidirectional recurrent network with a feedforward output layer, represents the forward hidden state sequence, represents the reverse hidden state sequence;

[0096] Autoformer-TimeGAN has the following three loss functions during training:

[0097] For the embedding function and reconstruction function, their goal should be to generate low-dimensional latent space and reconstruct high-dimensional original feature space as accurately as possible, introducing the first loss:

[0098] ;

[0099] Where, represents the expected distribution;

[0100] There is a classic game between the generator and the discriminator of the model, and the second loss is introduced as:

[0101] ;

[0102] Where, Represents the discrimination result of the static features of the original data, Represents the discrimination result of the dynamic characteristics of the original data, Represents the discriminant result of the static features of the generated data, Indicates the discriminant results of the dynamic features of the generated data;

[0103] Introducing unsupervised loss The model focuses on describing the overall probability distribution of time series data, but has not yet learned the point-by-point conditional probability distribution. Therefore, a third loss is introduced to achieve this:

[0104] ;

[0105] The Autoformer-TimeGAN model training process is as follows:

[0106] 1. Initialize the parameters of the embedding function e, reconstruction function r, generator g and discriminator d;

[0107] 2. For the original data Perform forward propagation:

[0108] 1) Computing latent representations through embedding networks: ;

[0109] 2) Reconstruct data by restoring network calculation: ;

[0110] 3. Calculate reconstruction loss ;

[0111] 4. Update the parameters of the embedding function e and the reconstruction function r to minimize ;

[0112] 5. Random noise vector Perform forward propagation and calculate the generated potential representation through the generator and ;

[0113] 6. Input the real and generated potential representations and conditional information into the discriminator and calculate the unsupervised loss ;

[0114] 7. Update the parameters of the generator g and the discriminator d ( represents the parameter set of the discriminator d):

[0115] 1) Minimize the generator's objective: ;

[0116] 2) Maximize the discriminator's goal: ;

[0117] 8. Using the conditional information c and the potential representation of the original sequence as input, the generator calculates the potential representation of the next prediction ;

[0118] 9. Calculate supervision loss ;

[0119] 10. Update the parameters of the embedding function e and the generator g to minimize the supervision loss ;

[0120] 11. Repeat steps 2-10 until convergence or the predetermined number of training times is reached;

[0121] In the specific implementation, it is necessary to set the hyperparameters of the Autoformer-TimeGAN model. The hyperparameter settings in the embodiment of this article are shown in the following table:

[0122] Table 1 Hyperparameters of the Autoformer-TimeGAN model

[0123]

[0124] Step 104 : Dimensionality reduction is performed on the enhanced meteorological data in the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset, and interactive clustering is performed to obtain the photovoltaic side meteorological scene and the load side meteorological scene.

[0125] Step 104 includes the following sub-steps:

[0126] S31, performing peak-to-valley difference calculation and average value calculation on the enhanced meteorological data in the enhanced meteorological photovoltaic data set on a daily basis to obtain a plurality of photovoltaic-side dimension-reduced meteorological data;

[0127] S32, performing peak-to-valley difference calculation and average value calculation on the enhanced meteorological data in the enhanced meteorological load data set on a daily basis to obtain a plurality of load-side dimension-reduced meteorological data;

[0128] S33, clustering the dimensionality-reduced meteorological data on the photovoltaic side to obtain multiple photovoltaic side meteorological scenarios and photovoltaic clustering parameters;

[0129] S34. Using the photovoltaic clustering parameters as the load clustering parameters, cluster analysis is performed on the load-side dimensionality reduction meteorological data to obtain multiple load-side meteorological scenarios.

[0130] It should be noted that after the enhanced meteorological photovoltaic dataset and enhanced meteorological load dataset are generated by the improved time series generative adversarial network model, Figure 2 As shown in the figure, considering meteorological factors, photovoltaic output and distribution network load characteristics, the peak-to-valley difference and average value are used to reduce the dimension of the enhanced meteorological data to obtain reduced-dimensional meteorological data. The peak-to-valley difference can characterize the fluctuation characteristics of the curve, and the average value can characterize the size level of the curve. This multi-dimensional feature extraction method helps to more comprehensively capture the characteristics of the data and improve the accuracy of subsequent analysis and generation. The average value calculation process includes:

[0131] ;

[0132] Where, represents the total number of data points for that day, represents the value of the j-th day data of the i-th meteorological factor at time t;

[0133] After completing the dimensionality reduction of the enhanced meteorological data, the K-Means algorithm is used to perform interactive K-Means clustering on the dimensionality reduction meteorological data. First, the PV side dimensionality reduction meteorological data is clustered to obtain N types of PV side meteorological scenes and the corresponding PV clustering parameters. The clustering parameters can be understood as including the K value, i.e., the optimal number of clusters, and the cluster center, i.e., the coordinates of the center point of each cluster. Then, the PV clustering parameters are used as the load clustering parameters for clustering the load side dimensionality reduction meteorological data to perform clustering analysis, and N types of load side meteorological scenes are obtained. By using the PV clustering parameters to initialize and guide the clustering of the load side dimensionality reduction meteorological data, the computational complexity of the clustering stage can be effectively reduced, thereby accelerating the convergence of the algorithm and ensuring the photovoltaic The clustering results on the load side and the energy side are "aligned" to some extent, that is, their clustering structures are similar, and the categories after classification are consistent, which is conducive to the subsequent generation of typical solar-load joint scenarios. It can be understood that since the meteorological data of the target meteorological photovoltaic dataset and the target meteorological load dataset come from the same source, their data distribution is basically the same. Therefore, the enhanced meteorological data distribution in the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset learned and generated by the improved time series generative adversarial network model is also basically the same. This means that the number of meteorological scene categories in the two datasets is equal, which we denote as N categories. The specific process of K-Means clustering includes:

[0134] 1) Specify the number of clusters K for clustering: In the specific implementation, this can be determined based on the Davies-Bouldin Index (DBI). DBI is an indicator used to evaluate clustering effectiveness. It measures the compactness and dispersion of clusters. A smaller DBI value indicates a good clustering effect, that is, the points within the cluster are compact and the points between clusters are dispersed. In the K-Means algorithm, people always expect the points within each cluster to be as close as possible, while the points in different clusters are as far apart as possible. DBI is based on this idea to evaluate clustering effectiveness. DBI is defined as follows:

[0135] ;

[0136] Where, represents the number of clusters, represents the average distance from all sample points in cluster a to the cluster center, represents the average distance from all sample points in the jth cluster to the cluster center, represents the distance between the cluster center of cluster a and the cluster center of cluster j, Indicates the similarity between cluster a and cluster b; when clustering is actually performed, by setting Perform multiple clustering (where c is the number of sample points) and select the K value corresponding to the minimum DBI as the optimal number of clusters K;

[0137] 2) Initialize the cluster centers based on the selected optimal number of clusters K: In the specific implementation, K data points are randomly selected from the dataset as the initial cluster centers;

[0138] 3) Assign samples to clusters: For each sample, calculate its distance to each cluster center and assign it to the cluster with the closest distance;

[0139] 4) Update the cluster center: For each cluster, calculate the average value of all samples in the cluster and use the average value as the new cluster center;

[0140] 5) Repeat steps 3) and 4) until the stopping condition is reached; the stopping condition is that the cluster center no longer changes significantly;

[0141] 6) Output clustering results: Finally, K clusters are obtained, and each sample is assigned to a cluster.

[0142] Step 105 : Perform cluster analysis on the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset respectively to obtain photovoltaic clusters for each photovoltaic-side meteorological scenario and load clusters for each load-side meteorological scenario.

[0143] Step 105 includes the following sub-steps:

[0144] S41, extracting corresponding enhanced photovoltaic data from the enhanced meteorological photovoltaic data set according to each photovoltaic side meteorological scenario, and clustering them to obtain photovoltaic clusters for each photovoltaic side meteorological scenario;

[0145] S42 , extracting corresponding enhanced load data from the enhanced meteorological load data set according to each load-side meteorological scenario, and clustering the data to obtain load clusters of each load-side meteorological scenario.

[0146] It should be noted that the enhanced photovoltaic data corresponding to each photovoltaic side meteorological scenario in the enhanced meteorological photovoltaic data set are clustered using the K-Means algorithm, and then N types of photovoltaic side meteorological scenarios are divided into Q photovoltaic clusters. The enhanced load data corresponding to each load side meteorological scenario are clustered using the K-Means algorithm, and N types of load side meteorological scenarios are divided into U load clusters.

[0147] Step 106: Perform global-local time series matching of meteorological day scenes based on each photovoltaic cluster and each load cluster to generate a typical solar-load joint scene set.

[0148] Step 106 includes the following sub-steps:

[0149] S51. Taking the enhanced photovoltaic data closest to the corresponding cluster center in each photovoltaic cluster as the photovoltaic typical scene of each photovoltaic cluster.

[0150] It should be noted that the data point closest to the cluster center in each photovoltaic cluster is defined as the typical scene of the photovoltaic cluster, so there are Q typical scenes.

[0151] S52. The enhanced load data closest to the corresponding cluster center in each load cluster is used as the typical load scenario of each load cluster, so there are U typical scenarios.

[0152] It should be noted that the data point closest to the cluster center in each load cluster is defined as the typical scenario of the load cluster.

[0153] S53. Taking the photovoltaic side meteorological scene and the load side meteorological scene in the same meteorological scene as a unit, the enhanced meteorological data of the day where each photovoltaic typical scene is located and the enhanced meteorological data of the day where each corresponding load typical scene is located are used to calculate the dynamic time warping distance and the matrix profile distance.

[0154] It should be noted that in order to achieve accurate matching of typical PV and load scenarios, the DTW (Dynamic Time Warping) and MPdist (Matrix Profile Distance) of the meteorological time series data corresponding to the day of the same type of PV and load typical scenarios in the enhanced dataset are calculated through global-local time series matching. DTW mainly captures the global characteristics and overall trends of the time series, while MPdist focuses on identifying local patterns and detailed features.

[0155] Among them, DTW achieves the curvature of the time axis by extending and shortening the time series, thereby evaluating the morphological similarity of two time series; for two given time series and , construct the F×V distance matrix , the elements in the matrix are calculated as follows:

[0156] ;

[0157] Where, for point and The Euclidean distance between For time series length, For time series length;

[0158] The purpose of DTW is to find an optimal curved path , satisfying the point pair arrive There exists a minimum cumulative distance, where q is the number of point pairs in the path W:

[0159] ;

[0160] Where, is the dynamic time warping distance, is the path element at position o in the distance matrix and usually needs to meet the requirements of boundary, continuity and monotonicity;

[0161] Among them, MPdist is a distance measurement method used to measure the similarity of time series. The calculation formula is as follows:

[0162] ;

[0163] in, and are two time series to be compared, It is from arrive The matrix profile distance, It is from arrive The matrix profile distance is calculated as follows:

[0164] ;

[0165] Where, for and The distance matrix between them, where each element express subsequence of and a subsequence of B The Euclidean distance between is the length of the subsequence; the core idea of ​​this formula is to find the most similar subsequence pair between two time series and use it as their distance measure.

[0166] S54. Perform comprehensive processing on each dynamic time warping distance and the associated matrix profile distance to obtain multiple comprehensive distances for each typical photovoltaic scene.

[0167] S55. Traverse each comprehensive distance and combine each photovoltaic typical scenario with the corresponding load typical scenario associated with the minimum comprehensive distance into a photovoltaic-load combined typical scenario pair.

[0168] S56. Use each light-load joint typical scenario pair to construct a light-load joint typical scenario set according to the meteorological scenario.

[0169] It should be noted that by combining the two measurement methods, the similarity of time series data can be evaluated more comprehensively. During the matching process, this embodiment takes the same type of meteorological scene as a unit, traverses each photovoltaic typical scene, and finds the one with the smallest absolute difference with the load typical scene under the comprehensive score of dynamic time warping distance and matrix profile distance as the similar weather day, thereby forming a light-load joint typical scene pair under this type of meteorological scene and constructing a light-load joint typical scene set under N types of meteorological scenes. Each type of meteorological scene contains multiple matching light-load joint typical scene pairs. This global-local combined matching strategy can simultaneously consider the overall similarity of the meteorological scene and the matching degree of local features. Since the matched photovoltaic and load typical scenes have similar global trends and local details, we can regard them as light-load joint typical scenes under the same meteorological conditions; in specific implementation, the dynamic time warping distance and the associated matrix profile distance can be weighted and then comprehensively calculated to obtain the comprehensive distance, and the weight can be set to 0.5.

[0170] In this embodiment, the dynamic and static characteristics of the target meteorological photovoltaic dataset and the target meteorological load dataset are fully exploited through the improved time series generative adversarial network model, so that the enhanced meteorological data on the photovoltaic side and the load side can interact with each other in the clustering process, ensuring the consistency of the clustering results on both sides while improving the clustering efficiency. The global-local time series matching method takes into account the global characteristics and local details of the time series data at the same time, and can more accurately match the photovoltaic and load scenarios. In the entire process from correlation analysis to scenario matching, objective meteorological data is fully utilized to couple the photovoltaic and load scenarios, which reduces the influence of subjective factors to a certain extent, improves the authenticity and representativeness of the generated scenarios as a whole, and improves the reliability of the generated typical light-load combined scenarios.

[0171] See also Figure 4 , Figure 4 This is a structural block diagram of a light-load combined typical scene generation device provided by an embodiment of the present invention.

[0172] The present invention provides a light-load combined typical scene generation device, comprising:

[0173] The data processing module 401 is used to obtain target photovoltaic output data, multiple target meteorological data, and target load data within a preset time range, and perform multi-source data fusion on each target meteorological data with each target photovoltaic output data and each target load data to obtain a meteorological photovoltaic data set and a meteorological load data set;

[0174] A data elimination module 402 is configured to eliminate meteorological data redundancy from the meteorological photovoltaic dataset and the meteorological load dataset based on correlation coefficients among the target photovoltaic output data, the target meteorological data, and the target load data, and determine the target meteorological photovoltaic dataset and the target meteorological load dataset;

[0175] The data enhancement module 403 is used to perform data enhancement on the target meteorological photovoltaic dataset and the target meteorological load dataset respectively using the improved time series generative adversarial network model to obtain an enhanced meteorological photovoltaic dataset and an enhanced meteorological load dataset;

[0176] Dimensionality reduction and clustering module 404 is used to reduce the dimension of the enhanced meteorological data in the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset, and perform interactive clustering to obtain the meteorological scene on the photovoltaic side and the meteorological scene on the load side;

[0177] The photovoltaic load clustering module 405 is used to perform cluster analysis on the enhanced meteorological photovoltaic data set and the enhanced meteorological load data set to obtain photovoltaic clusters for each photovoltaic-side meteorological scenario and load clusters for each load-side meteorological scenario;

[0178] The typical scenario generation module 406 is used to perform global-local time series matching of meteorological day scenarios based on each photovoltaic cluster and each load cluster to generate a set of photovoltaic-load combined typical scenarios.

[0179] Optionally, the data processing module 401 is specifically configured to:

[0180] Obtain initial photovoltaic output data, various initial meteorological data, and initial load data within a preset time range;

[0181] Perform data cleaning on each photovoltaic output data, each meteorological data and each load data to obtain target photovoltaic output data, target meteorological data and target load data;

[0182] The meteorological data of each target and the photovoltaic output data of each target are integrated according to time to generate a meteorological photovoltaic data set;

[0183] The target meteorological data and target load data are fused according to time to determine the meteorological load data set.

[0184] Optionally, the data elimination module 402 is specifically configured to:

[0185] Calculate the peak-valley difference of various target meteorological data on a daily basis, and obtain multiple peak-valley difference meteorological data;

[0186] The Pearson correlation coefficient method is used to calculate the correlation coefficients between each target photovoltaic output data, each target load data and each peak-to-valley difference meteorological data;

[0187] Select peak-valley difference meteorological data whose correlation coefficient with target photovoltaic output data and target load data is greater than a first correlation threshold as target peak-valley difference meteorological data;

[0188] If there are multiple categories of target peak-valley difference meteorological data, based on the correlation coefficients between the various target peak-valley difference meteorological data, a category of target peak-valley difference meteorological data having a correlation coefficient greater than a second correlation threshold is selected as a dimensionality reduction meteorological scenario;

[0189] The target meteorological data associated with the dimension-reduced meteorological scene are removed from the meteorological photovoltaic dataset and the meteorological load dataset to obtain the target meteorological photovoltaic dataset and the target meteorological load dataset.

[0190] Optionally, the dimensionality reduction clustering module 404 is specifically configured to:

[0191] The enhanced meteorological data in the enhanced meteorological photovoltaic dataset are subjected to peak-to-valley difference and average value calculations on a daily basis to obtain multiple photovoltaic-side dimension-reduced meteorological data.

[0192] The enhanced meteorological data in the enhanced meteorological load data set are calculated on a daily basis for peak-to-valley difference and average value, and multiple load-side dimension-reduced meteorological data are obtained.

[0193] Clustering the PV-side dimensionality-reduced meteorological data to obtain multiple PV-side meteorological scenarios and PV clustering parameters;

[0194] Using photovoltaic clustering parameters as load clustering parameters, cluster analysis is performed on the load-side dimensionality-reduced meteorological data to obtain multiple load-side meteorological scenarios.

[0195] Optionally, the photovoltaic load clustering module 405 is specifically configured to:

[0196] Clustering is performed on the enhanced photovoltaic data to obtain photovoltaic clusters of each photovoltaic side meteorological scene;

[0197] According to each load-side meteorological scenario, the corresponding enhanced load data are extracted from the enhanced meteorological load data set for clustering to obtain the load clusters of each load-side meteorological scenario.

[0198] Optionally, the typical scenario generating module 406 is specifically configured to:

[0199] The enhanced photovoltaic data closest to the corresponding cluster center in each photovoltaic cluster is used as the photovoltaic typical scene of each photovoltaic cluster;

[0200] The enhanced load data closest to the corresponding cluster center in each load cluster is used as the load typical scenario of each load cluster;

[0201] Taking the PV side meteorological scene and the load side meteorological scene in the same meteorological scene as a unit, the enhanced meteorological data of each typical PV scene on the day and the corresponding enhanced meteorological data of each typical load scene on the day are used to calculate the dynamic time warping distance and matrix profile distance;

[0202] The dynamic time warping distances and the associated matrix profile distances are used for comprehensive processing to obtain multiple comprehensive distances for each typical photovoltaic scenario;

[0203] Traverse each comprehensive distance and combine each photovoltaic typical scenario with the corresponding load typical scenario associated with the minimum comprehensive distance into a photovoltaic-load joint typical scenario pair;

[0204] The typical light-load joint scenario pairs are used to construct a light-load joint typical scenario set according to the meteorological scenario.

[0205] An embodiment of the present invention further provides a computer device comprising a memory and a processor, wherein a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the steps of the light-load combined typical scene generation method as described in any of the above embodiments.

[0206] An embodiment of the present invention further provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the steps of the light-load combined typical scene generation method as described in any of the above embodiments are implemented.

[0207] An embodiment of the present invention further provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the light-load combined typical scene generation method as described in any of the above embodiments.

[0208] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0209] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0210] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0211] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0212] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the 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 can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0213] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A light-load combined typical scene generation method, characterized in that: include: Obtain target photovoltaic output data, multiple target meteorological data, and target load data within a preset time range, and perform multi-source data fusion on each target meteorological data with each target photovoltaic output data and each target load data to obtain a meteorological photovoltaic data set and a meteorological load data set; Based on the correlation coefficients among the target photovoltaic output data, the target meteorological data, and the target load data, redundancy of meteorological data is eliminated from the meteorological photovoltaic dataset and the meteorological load dataset to determine a target meteorological photovoltaic dataset and a target meteorological load dataset; An improved temporal generative adversarial network model is used to perform data enhancement on the target meteorological photovoltaic dataset and the target meteorological load dataset, respectively, to obtain an enhanced meteorological photovoltaic dataset and an enhanced meteorological load dataset; Performing dimensionality reduction on the enhanced meteorological data in the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset, and performing interactive clustering to obtain a photovoltaic side meteorological scene and a load side meteorological scene; Performing cluster analysis on the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset respectively to obtain photovoltaic clusters of each of the photovoltaic-side meteorological scenarios and load clusters of each of the load-side meteorological scenarios; Performing global-local time series matching of meteorological day scenes based on each of the photovoltaic clusters and each of the load clusters to generate a set of typical photovoltaic-load combined scenes; The enhanced meteorological data in the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset are respectively reduced in dimension, and interactively clustered to obtain a photovoltaic side meteorological scene and a load side meteorological scene, including: Performing peak-to-valley difference calculation and average value calculation on the enhanced meteorological data in the enhanced meteorological photovoltaic dataset on a daily basis to obtain a plurality of photovoltaic-side dimension-reduced meteorological data; Performing peak-to-valley difference calculation and average value calculation on the enhanced meteorological data in the enhanced meteorological load data set on a daily basis to obtain a plurality of load-side dimension-reduced meteorological data; Clustering the photovoltaic-side dimension-reduced meteorological data to obtain a plurality of photovoltaic-side meteorological scenarios and photovoltaic clustering parameters; The photovoltaic clustering parameters are used as load clustering parameters to perform cluster analysis on the load-side dimension-reduced meteorological data to obtain multiple load-side meteorological scenarios.

2. The light-load combined typical scene generation method according to claim 1, characterized in that: The step of acquiring target photovoltaic output data, multiple target meteorological data, and target load data within a preset time range, and performing multi-source data fusion on each target meteorological data with each target photovoltaic output data and each target load data to obtain a meteorological photovoltaic data set and a meteorological load data set includes: Obtain initial photovoltaic output data, various initial meteorological data, and initial load data within a preset time range; performing data cleaning on each of the photovoltaic output data, each of the meteorological data, and each of the load data to obtain target photovoltaic output data, target meteorological data, and target load data; fusing the target meteorological data with the target photovoltaic output data according to time to generate a meteorological photovoltaic data set; The target meteorological data and the target load data are fused according to time to determine a meteorological load data set.

3. The light-load combined typical scene generation method according to claim 1, characterized in that: The step of performing meteorological data redundancy elimination on the meteorological photovoltaic dataset and the meteorological load dataset based on the correlation coefficients among the target photovoltaic output data, the target meteorological data, and the target load data, and correspondingly determining the target meteorological photovoltaic dataset and the target meteorological load dataset, includes: Calculating the peak-to-valley difference of each target meteorological data on a daily basis to obtain a plurality of corresponding peak-to-valley difference meteorological data; The Pearson correlation coefficient method is used to calculate the correlation coefficient between each target photovoltaic output data, each target load data and each peak-to-valley difference meteorological data; Selecting peak-valley difference meteorological data having a correlation coefficient greater than a first correlation threshold with both the target photovoltaic output data and the target load data as target peak-valley difference meteorological data; If there are multiple categories of the target peak-to-valley difference meteorological data, then based on the correlation coefficients between the various target peak-to-valley difference meteorological data, one category of the target peak-to-valley difference meteorological data having a correlation coefficient greater than a second correlation threshold is selected as the dimensionality reduction meteorological scene; Target meteorological data associated with the dimension-reduced meteorological scene are respectively removed from the meteorological photovoltaic dataset and the meteorological load dataset to obtain a target meteorological photovoltaic dataset and a target meteorological load dataset.

4. The light-load combined typical scene generation method according to claim 1, characterized in that: The cluster analysis is performed on the enhanced meteorological photovoltaic data set and the enhanced meteorological load data set respectively to obtain the photovoltaic clusters of each photovoltaic-side meteorological scene and the load clusters of each load-side meteorological scene, including: According to each photovoltaic side meteorological scene, corresponding enhanced photovoltaic data are extracted from the enhanced meteorological photovoltaic data set for clustering to obtain photovoltaic clusters of each photovoltaic side meteorological scene; According to each of the load-side meteorological scenarios, corresponding enhanced load data are extracted from the enhanced meteorological load data set for clustering to obtain load clusters of each of the load-side meteorological scenarios.

5. The light-load combined typical scene generation method according to claim 1, characterized in that: The global-local time series matching of meteorological day scenes based on each photovoltaic cluster and each load cluster is performed to generate a set of typical solar-load combined scenes, including: taking the enhanced photovoltaic data closest to the corresponding cluster center in each photovoltaic cluster as the photovoltaic typical scene of each photovoltaic cluster; Taking the enhanced load data closest to the corresponding cluster center in each load cluster as the load typical scenario of each load cluster; Taking the PV side meteorological scene and the load side meteorological scene in the same meteorological scene as a unit, the enhanced meteorological data of each typical PV scene on the day and the corresponding enhanced meteorological data of each typical load scene on the day are used to calculate the dynamic time warping distance and matrix profile distance; Performing comprehensive processing on the dynamic time warping distances and the associated matrix profile distances to obtain a plurality of comprehensive distances for the typical photovoltaic scenarios; Traversing each of the comprehensive distances, and combining each of the photovoltaic typical scenarios with the corresponding load typical scenarios associated with the minimum comprehensive distance into a photovoltaic-load combined typical scenario pair; The light-load combined typical scene pairs are used to construct a light-load combined typical scene set according to the meteorological scene.

6. A light-charge combined typical scene generation device, characterized in that: include: A data processing module is used to obtain target photovoltaic output data, multiple target meteorological data, and target load data within a preset time range, and perform multi-source data fusion on each target meteorological data with each target photovoltaic output data and each target load data to obtain a meteorological photovoltaic data set and a meteorological load data set; a data elimination module, configured to eliminate meteorological data redundancy from the meteorological photovoltaic dataset and the meteorological load dataset based on correlation coefficients among the target photovoltaic output data, the target meteorological data, and the target load data, and determine a target meteorological photovoltaic dataset and a target meteorological load dataset; a data enhancement module, configured to perform data enhancement on the target meteorological photovoltaic dataset and the target meteorological load dataset respectively using an improved temporal generative adversarial network model to obtain an enhanced meteorological photovoltaic dataset and an enhanced meteorological load dataset; A dimensionality reduction clustering module is used to reduce the dimensionality of the enhanced meteorological data in the enhanced meteorological photovoltaic dataset and the enhanced meteorological load dataset, and perform interactive clustering to obtain the photovoltaic side meteorological scene and the load side meteorological scene; A photovoltaic load clustering module is used to perform cluster analysis on the enhanced meteorological photovoltaic data set and the enhanced meteorological load data set respectively to obtain photovoltaic clusters of each photovoltaic-side meteorological scenario and load clusters of each load-side meteorological scenario; A typical scenario generation module is used to perform global-local time series matching of meteorological day scenarios based on each of the photovoltaic clusters and each of the load clusters to generate a set of typical solar-load combined scenarios; The dimensionality reduction clustering module is specifically used for: Performing peak-to-valley difference calculation and average value calculation on the enhanced meteorological data in the enhanced meteorological photovoltaic dataset on a daily basis to obtain a plurality of photovoltaic-side dimension-reduced meteorological data; Performing peak-to-valley difference calculation and average value calculation on the enhanced meteorological data in the enhanced meteorological load data set on a daily basis to obtain a plurality of load-side dimension-reduced meteorological data; Clustering the photovoltaic-side dimension-reduced meteorological data to obtain a plurality of photovoltaic-side meteorological scenarios and photovoltaic clustering parameters; The photovoltaic clustering parameters are used as load clustering parameters to perform cluster analysis on the load-side dimension-reduced meteorological data to obtain multiple load-side meteorological scenarios.

7. A computer device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the light-load combined typical scene generation method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the light-load combined typical scene generation method according to any one of claims 1 to 5 are implemented.

9. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the light-load combined typical scene generation method according to any one of claims 1 to 5 are implemented.

Images