A method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components
By using a short-term photovoltaic power prediction method based on weather scenarios and multi-scale components, weather scenario samples are collected and divided, multi-scale decomposition is performed, low-frequency trends and high-frequency fluctuation characteristics of photovoltaic power are extracted, and a prediction model is constructed. This solves the problem of inconsistent accuracy in short-term photovoltaic power prediction in existing technologies and achieves high-precision prediction under different weather conditions.
Patent Information
- Application Number
- CN202610947873.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-29
- Publication Date
- 2026-11-10
AI Technical Summary
Existing methods for short-term photovoltaic power generation forecasting fail to effectively distinguish the output patterns under different scenarios such as sunny, cloudy, and rainy days. They also struggle to separate the low-frequency trends and high-frequency fluctuations of photovoltaic power, resulting in inconsistent forecast accuracy, especially with significant errors under complex weather conditions.
By collecting weather scene samples, classifying typical weather scenes based on historical meteorological characteristics, performing multi-scale decomposition, extracting low-frequency trend terms and high-frequency fluctuation terms of photovoltaic power, and constructing an initial prediction model for superposition and reconstruction, short-term prediction values of photovoltaic power are obtained.
It has improved the accuracy of short-term photovoltaic power prediction under different weather scenarios. By using multi-scale component collaborative prediction, it adapts to the photovoltaic power change characteristics under various weather scenarios, reduces prediction errors, and improves the overall accuracy and scenario adaptability of photovoltaic power prediction.
Smart Images

Figure CN123073697A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of short-term photovoltaic power generation prediction technology, and more specifically, to a method for short-term photovoltaic power generation prediction based on weather scenarios and multi-scale components. Background Technology
[0002] Short-term photovoltaic (PV) power generation forecasting is a time-series forecasting method based on historical power generation data of PV power plants and real-time and forecast meteorological information. This method combines various meteorological factors that affect PV output, such as irradiance, ambient temperature, cloud cover, and wind speed, to analyze the inherent patterns of PV power changes with meteorological conditions and intraday time series, and to predict the output power of PV power plants in the near future. It is an important basic technology for new energy grid connection, grid dispatch, and power plant operation and management. It can help to know the changes in PV output in advance, providing data support for the stable operation of the power system and the daily operation and maintenance of PV power plants.
[0003] However, most existing short-term photovoltaic (PV) power forecasting methods employ a unified model for global prediction, lacking an integrated technical system that combines weather scenario segmentation, multi-scale power decomposition, sub-model feature extraction, and collaborative prediction. This makes it difficult to distinguish power output patterns under different scenarios such as sunny, cloudy, and rainy days, and also hinders the separation of low-frequency trends and high-frequency fluctuations in PV power. Long-term evolution characteristics and short-term dynamic characteristics cannot be specifically mined, resulting in inconsistent prediction accuracy for power components at different scales. Under complex weather conditions, the overall prediction results show significant errors, and the model's adaptability to various scenarios is insufficient. Therefore, achieving weather scenario adaptation and multi-scale component collaborative prediction to improve the accuracy of short-term PV power forecasting under various weather conditions is a key challenge facing the industry. Summary of the Invention
[0004] This application provides a method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components, which can achieve weather scenario adaptation and multi-scale component collaborative prediction, thereby improving the accuracy of short-term photovoltaic power prediction under various weather conditions.
[0005] In a first aspect, this application provides a short-term photovoltaic power prediction method based on weather scenarios and multi-scale components, the prediction method comprising: Weather scene samples are collected and divided according to historical meteorological characteristics to obtain multiple typical weather scenes; Multi-scale decomposition is performed on the photovoltaic power sequence under each typical weather scenario to obtain the photovoltaic power frequency characteristics at different scales. Based on each photovoltaic power frequency characteristic and the sample entropy characteristics corresponding to the weather scenario sample, the low-frequency trend term and high-frequency fluctuation term of photovoltaic power are determined. The periodic evolution characteristics of fixed fluctuations in photovoltaic power are extracted from the low-frequency trend term, and the local short-time dynamic characteristics of random fluctuations in photovoltaic power are extracted from the high-frequency fluctuation term. Based on the periodic evolution characteristics and the local short-term dynamic characteristics, the photovoltaic power components under various weather scenarios are predicted, and the initial prediction results are superimposed and reconstructed to obtain the short-term predicted value of photovoltaic power.
[0006] In this embodiment, the weather scene samples are divided according to historical meteorological characteristics to obtain several typical weather scenes, including: Obtain historical meteorological characteristics during the operation of photovoltaic power plants; Based on the historical meteorological characteristics, the weather information in the weather scene sample is divided to obtain multiple typical weather scenes.
[0007] In this embodiment, the photovoltaic power sequence under each typical weather scenario is decomposed into multiple scales to obtain the photovoltaic power frequency characteristics at different scales, specifically including: Determine the photovoltaic power sequence for each typical weather scenario; Modal decomposition is performed on the photovoltaic power sequence to obtain intrinsic photovoltaic power information; The intrinsic photovoltaic power information is classified by scale to obtain photovoltaic power frequency characteristics at different scales.
[0008] In this embodiment, performing mode decomposition on the photovoltaic power sequence to obtain intrinsic photovoltaic power information specifically includes: Extract the intrinsic mode components corresponding to the photovoltaic power from the photovoltaic power sequence; The intrinsic photovoltaic power information is obtained by identifying the components of the photovoltaic power sequence using the intrinsic mode components.
[0009] In this embodiment, determining the low-frequency trend term and high-frequency fluctuation term of photovoltaic power based on the frequency characteristics of each photovoltaic power and the sample entropy characteristics corresponding to the weather scene sample specifically includes: Determine the jump characteristics of photovoltaic power in power jumps based on the frequency characteristics of each photovoltaic power; The dynamic instability information of photovoltaic power when the weather scene changes is determined based on the sample entropy characteristics corresponding to the weather scene samples; The fluctuation limits of photovoltaic power are determined based on all the jump characteristics and the dynamic instability information, and then the low-frequency trend term and high-frequency fluctuation term of photovoltaic power are determined through the fluctuation limits.
[0010] In this embodiment, extracting the periodic evolution characteristics of fixed fluctuations in photovoltaic power from the low-frequency trend term specifically includes: The power input sequence of the low-frequency trend term is aligned and transformed to obtain the power aligned amplitude. The main periodic information for fixed fluctuations in photovoltaic power is determined based on the power alignment amplitude. The periodic evolution characteristics of fixed fluctuations in photovoltaic power are determined by the aforementioned key periodic information.
[0011] In this embodiment, determining the main periodic information for fixed fluctuations in photovoltaic power based on the power alignment amplitude specifically includes: The power alignment amplitude is divided into multiple amplitude variation segments; By marking the fluctuation trend of all amplitude variation segments, the main periodic information of photovoltaic power fluctuations with constant fluctuations can be obtained.
[0012] In this embodiment, extracting the local short-time dynamic features of photovoltaic power random fluctuations from the high-frequency fluctuation term specifically includes: Based on the volatility reduction factor for power fluctuations in the high-frequency fluctuation term, random fluctuation components for random fluctuations in photovoltaic power are generated through a fluctuation evolution model. The power interaction characteristics of photovoltaic power random fluctuations are determined based on the random fluctuation components. By performing fluctuation maintenance matching based on the power interaction characteristics, the local short-term dynamic characteristics of random fluctuations in photovoltaic power are obtained.
[0013] In this embodiment, determining the power interaction characteristics of photovoltaic power random fluctuations based on the random fluctuation components specifically includes: The random integral interval of the random fluctuation of photovoltaic power is determined based on the random fluctuation components. By integrating the random integration interval into the fluctuation space, the power fluctuation attribute of photovoltaic power under random fluctuation is obtained. The power interaction characteristics of random fluctuations in photovoltaic power are determined based on the power fluctuation attributes.
[0014] In this embodiment, predicting photovoltaic power components under various weather scenarios based on the periodic evolution characteristics and the local short-term dynamic characteristics specifically includes: Construct initial prediction models for photovoltaic power under various weather scenarios; The initial prediction model is used to predict the periodic evolution features and the local short-term dynamic features to obtain low-frequency initial prediction results and high-frequency initial prediction results. The low-frequency initial prediction result and the high-frequency initial prediction result are spliced together to obtain the initial prediction result, thus completing the prediction of photovoltaic power components under various weather scenarios.
[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: Weather scene samples are collected and divided according to historical meteorological characteristics to obtain multiple typical weather scenes. The photovoltaic power sequence under each typical weather scene is decomposed into multiple scales to obtain photovoltaic power frequency characteristics at different scales. Based on each photovoltaic power frequency characteristic and the sample entropy characteristics corresponding to the weather scene samples, the low-frequency trend term and high-frequency fluctuation term of photovoltaic power are determined. Periodic evolution characteristics of fixed fluctuations in photovoltaic power are extracted from the low-frequency trend term, and local short-term dynamic characteristics of random fluctuations in photovoltaic power are extracted from the high-frequency fluctuation term. The photovoltaic power components under various weather scenes are predicted based on the periodic evolution characteristics and the local short-term dynamic characteristics, and the initial prediction results are superimposed and reconstructed to obtain short-term predicted values of photovoltaic power.
[0016] Therefore, this application involves superimposing and reconstructing the initial prediction results to obtain short-term photovoltaic power predictions. Identifying the low-frequency trend term and the high-frequency fluctuation term allows for the separation of different time-scale variations in the photovoltaic power sequence, achieving multi-scale data decoupling. This partitioning method establishes objective discrimination criteria based on Pearson correlation coefficient and sample entropy, accurately distinguishing between the fundamental variation patterns of photovoltaic output and random fluctuations caused by meteorological disturbances. Based on this, differentiated processing logic can be matched to the two types of components, adapting to power variation characteristics under different weather scenarios such as sunny, cloudy, and rainy days. This lays a solid data foundation for grouped modeling and adaptive prediction of meteorological scenarios, effectively avoiding the shortcomings of a single data structure that cannot accommodate different fluctuation patterns, and improving the overall prediction accuracy from the data source level. Identifying the periodic evolution characteristics and local short-term dynamic characteristics allows for the extraction of the core temporal expression information of the two types of power components. The periodic evolution characteristics fully preserve the long-term evolution patterns of the low-frequency trend, while the local short-term dynamic characteristics accurately depict the instantaneous disturbance characteristics of high-frequency fluctuations, achieving professional extraction of feature dimensions. The two types of features can be directly connected to the appropriate deep learning models, giving full play to the technical advantages of different models, strengthening the ability of multi-scale component collaborative modeling, enabling the model to accurately capture force features in various weather scenarios, further reducing prediction errors, and comprehensively improving the overall accuracy and scenario adaptability of short-term photovoltaic power prediction.
[0017] In summary, the technical solution adopted in this application can achieve adaptive weather scenario prediction and multi-scale component collaborative prediction, thereby improving the accuracy of short-term photovoltaic power prediction under various weather conditions. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application 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 for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is an exemplary flowchart of a short-term photovoltaic power prediction method based on weather scenarios and multi-scale components provided in this application; Figure 2 This is a flowchart illustrating the process for determining local short-term dynamic characteristics provided in this application. Detailed Implementation
[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0021] This application provides a method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components. Its core is to collect weather scenario samples and divide the weather scenario samples according to historical meteorological characteristics to obtain multiple typical weather scenarios. Multi-scale decomposition is performed on the photovoltaic power sequence under each typical weather scenario to obtain the photovoltaic power frequency characteristics at different scales. Based on each photovoltaic power frequency characteristic and the sample entropy characteristics corresponding to the weather scenario sample, the low-frequency trend term and high-frequency fluctuation term of photovoltaic power are determined. The periodic evolution characteristics of fixed fluctuations in photovoltaic power are extracted from the low-frequency trend term, and the local short-time dynamic characteristics of random fluctuations in photovoltaic power are extracted from the high-frequency fluctuation term. Based on the periodic evolution characteristics and the local short-term dynamic characteristics, the photovoltaic power components under various weather scenarios are predicted, and the initial prediction results are superimposed and reconstructed to obtain the short-term predicted value of photovoltaic power.
[0022] To better understand the above technical solutions, a detailed description of the technical solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of a short-term photovoltaic power prediction method based on weather scenarios and multi-scale components according to this embodiment of the present application. The prediction method includes the following steps: In step S1, weather scene samples are collected, and the weather scene samples are divided according to historical meteorological characteristics to obtain multiple typical weather scenes.
[0023] In specific implementation, weather scenario samples can be collected in the following way: First, a meteorological monitoring device is deployed at the photovoltaic power station to synchronously collect meteorological observation data within a historical time window at a preset time interval. The preset time interval refers to the time resolution of the meteorological monitoring device collecting meteorological data and the power station recording output power. In the short-term forecasting scenario of photovoltaic power, in order to strictly match the time granularity assessment requirements of the power grid dispatch for short-term power forecasting, while taking into account the sample data volume and model training efficiency, the preset time interval is preferably 15 minutes. Then, the collected raw data is cleaned to remove outliers caused by equipment failure or transmission errors, and missing data is filled in using linear interpolation. Next, the processed data is aligned according to the time series to construct a structured dataset containing meteorological features and corresponding timestamps. Finally, the structured dataset is divided into a training set and a test set to complete the collection and preprocessing of weather scenario samples.
[0024] It should be noted that, in this application, the weather scene sample refers to a multi-dimensional meteorological feature data set including total irradiance, temperature, humidity, cloud cover, rainfall, and wind speed.
[0025] In this embodiment, dividing the weather scene samples based on historical meteorological characteristics to obtain multiple typical weather scenes can be achieved through the following steps: Obtain historical meteorological characteristics during the operation of photovoltaic power plants; Based on the historical meteorological characteristics, the weather information in the weather scene sample is divided to obtain multiple typical weather scenes.
[0026] In practice, the process begins with the deployment of meteorological monitoring terminals across the entire photovoltaic power station area. Monitoring data is continuously collected at fixed intervals, including total irradiance, temperature, humidity, cloud cover, rainfall, and wind speed. After collection, data preprocessing is performed. Abnormal monitoring data is manually removed using equipment logs, and missing data segments are supplemented using linear interpolation algorithms. All cleaned meteorological data are aligned along a unified timeline and integrated to form a standardized historical meteorological feature dataset. The dataset samples are derived from long-term continuous operation monitoring records of the power station. Next, the historical meteorological features are constructed into multi-dimensional feature vectors. A sphere-based K-means clustering algorithm is initialized, with three clusters corresponding to three weather types. Neighboring meteorological samples are first aggregated into independent spheres. Then, the distance between each sphere and the cluster center is calculated, iteratively updating the sphere's association with the cluster center, using minimum intra-cluster dispersion as the convergence condition. After algorithm convergence, the clustering results are categorized into three typical weather scenarios—sunny, cloudy, and rainy—based on irradiance, cloud cover, and photovoltaic power curves. The formula for calculating the sphere center is: ,in, For the first The center of each sphere; Number of samples within a granule: : The set of samples contained in the j-th particle; Algorithm objective function: C k : The set of particles contained in the k-th class; μk: The cluster center of the k-th class.
[0027] It should be noted that in this application, historical meteorological characteristics refer to the set of various monitoring indicators of the atmospheric environment around the photovoltaic power station; weather information refers to meteorological data consisting of total irradiance, temperature, humidity, cloud cover, rainfall, and wind speed; typical weather scenarios refer to the standardized weather classification results formed after clustering meteorological data, which are defined as three categories in this embodiment: sunny, cloudy, and rainy.
[0028] In step S2, the photovoltaic power sequence under each typical weather scenario is decomposed into multiple scales to obtain the photovoltaic power frequency characteristics at different scales. Based on each photovoltaic power frequency characteristic and the sample entropy characteristics corresponding to the weather scenario sample, the low-frequency trend term and high-frequency fluctuation term of photovoltaic power are determined.
[0029] In this embodiment, the photovoltaic power sequence under each typical weather scenario is decomposed into multiple scales to obtain the photovoltaic power frequency characteristics at different scales. This can be achieved in the following way: Determine the photovoltaic power sequence for each typical weather scenario; Modal decomposition is performed on the photovoltaic power sequence to obtain intrinsic photovoltaic power information; The intrinsic photovoltaic power information is classified by scale to obtain photovoltaic power frequency characteristics at different scales.
[0030] In practice, the process begins by matching power plant operation data for the corresponding time intervals to the three weather scenarios (sunny, cloudy, and rainy). From the historical operation database of the entire station, raw photovoltaic power data that perfectly corresponds to the timestamps of each weather scenario is selected. The selected data is then arranged chronologically to form independent solar power sequences for sunny, cloudy, and rainy days. Next, adaptive noise-complete set empirical mode decomposition (EMD) is employed. First, min-max normalization is performed on each photovoltaic power sequence to unify the data interval. Then, white noise of different amplitudes is adaptively matched to the sequences, and iterative operations of extreme point retrieval, upper and lower envelope fitting, and mean calculation are repeatedly performed until no new components can be extracted. Finally, the original sequence is split into multiple intrinsic mode components and a set of residual terms. All the obtained components together constitute the intrinsic photovoltaic power information. Finally, the Pearson correlation coefficient between each intrinsic mode component and the original photovoltaic power sequence is calculated, and the sample entropy corresponding to each component is also determined. By combining two indicators to comprehensively determine the fluctuation scale of the components, those with high correlation coefficients and low sample entropy are classified as low-frequency, while those with low correlation coefficients and high sample entropy are classified as high-frequency. After classifying all components, they are integrated according to scale to form multiple sets of photovoltaic power frequency characteristics at different scales.
[0031] It should be noted that, in this application, the photovoltaic power sequence refers to the output power dataset of a photovoltaic power station arranged in chronological order corresponding to a single typical weather scenario; intrinsic photovoltaic power information refers to the set of intrinsic mode components and residual terms generated after mode decomposition; scale classification refers to the processing flow of classifying the decomposed components according to the temporal fluctuation characteristics; and photovoltaic power frequency characteristics refer to the set of power components distinguished according to the speed of fluctuation.
[0032] In addition, in this embodiment, the intrinsic photovoltaic power information can be obtained by performing mode decomposition on the photovoltaic power sequence using the following steps: Extract the intrinsic mode components corresponding to the photovoltaic power from the photovoltaic power sequence; The intrinsic photovoltaic power information is obtained by identifying the components of the photovoltaic power sequence using the intrinsic mode components.
[0033] In practice, the photovoltaic power sequence is first subjected to min-max normalization to unify the numerical range of all data. Then, an adaptive noise-complete set empirical mode decomposition method is used to perform calculations, continuously adding white noise with matching amplitude to the original sequence. Iterative calculations of extreme point finding, upper and lower envelope fitting, and sequence mean calculation are performed repeatedly, separating an independent set of intrinsic modes in each iteration. When the remaining signal cannot be further decomposed into new components, the iteration process is terminated, ultimately extracting all intrinsic mode components from the original photovoltaic power sequence. Next, the Pearson correlation coefficient between each intrinsic mode component and the original photovoltaic power sequence is calculated sequentially, along with the corresponding sample entropy value. The fluctuation attributes of each component are determined by combining the two sets of calculation results, distinguishing whether the component belongs to the trend or fluctuation category. All intrinsic mode components with identified attributes, along with the residual terms remaining from the decomposition process, are integrated to form complete intrinsic photovoltaic power information. The entire set of data fully preserves the multi-scale variation characteristics of the original sequence.
[0034] It should be noted that, in this application, intrinsic mode components refer to individual time-series signal units separated after mode decomposition, each carrying the fluctuation pattern of photovoltaic power at different frequencies; component identification refers to the process of determining the attributes of the extracted intrinsic mode components.
[0035] In this embodiment, determining the low-frequency trend term and high-frequency fluctuation term of photovoltaic power based on the frequency characteristics of each photovoltaic power and the sample entropy characteristics corresponding to the weather scene sample can be achieved through the following steps: Determine the jump characteristics of photovoltaic power in power jumps based on the frequency characteristics of each photovoltaic power; The dynamic instability information of photovoltaic power when the weather scene changes is determined based on the sample entropy characteristics corresponding to the weather scene samples; The fluctuation limits of photovoltaic power are determined based on all the jump characteristics and the dynamic instability information, and then the low-frequency trend term and high-frequency fluctuation term of photovoltaic power are determined through the fluctuation limits.
[0036] In practice, firstly, based on the frequency characteristics of photovoltaic power corresponding to each intrinsic mode component, the time series data of the components are traversed segment by segment. The power value difference between adjacent time nodes is continuously compared, and the number and amplitude of power value abrupt changes per unit time are counted. The correlation between the component and the original power sequence is measured by combining the Pearson correlation coefficient. The higher the correlation, the more closely the jump fits the overall power output pattern. Based on the statistical results and correlation data, the power jump characteristics corresponding to each component are fully summarized. Then, for each typical weather scenario of sunny, cloudy, and rainy days, the sample entropy of the corresponding intrinsic mode component is calculated. The sample entropy calculation is carried out according to the standard algorithm. The time series vector is reconstructed and the vector matching probability is counted to finally obtain the entropy value. The higher the entropy value, the stronger the randomness of the sequence and the more obvious the disordered power fluctuation caused by the weather change. Combining the sample entropy values of each weather scenario, the degree of disordered fluctuation of photovoltaic power under different meteorological conditions is sorted out, thereby determining the complete dynamic instability information. Finally, by integrating all jump characteristics and dynamic instability information, and considering the power jump amplitude, frequency, and sample entropy values, a unified fluctuation judgment boundary is established. This fluctuation judgment boundary refers to a dual-index discrimination threshold system used to quantify and distinguish the attributes of different frequency components of photovoltaic power. Specifically, it consists of a Pearson correlation coefficient threshold and a sample entropy threshold. The correlation coefficient threshold measures the overall trend consistency between the intrinsic mode component (IMF) and the original power sequence, while the sample entropy threshold characterizes the complexity and randomness of the temporal fluctuations of this component. By setting these dual boundaries, the correlation coefficient... Components with correlation coefficients above the threshold and sample entropy below the threshold are accurately classified as low-frequency trend terms with strong regularity; components with low correlation coefficients or high sample entropy are classified as mid-to-high-frequency fluctuation terms reflecting meteorological disturbances; while pure noise disturbance terms with extremely low correlation coefficients and extremely high sample entropy are discarded. Components with small jump amplitudes, low frequency and low sample entropy values are classified as low-frequency trend terms, and components with large jump amplitudes, high frequency and high sample entropy values are classified as high-frequency fluctuation terms. This classification logic combines two dimensions and has been repeatedly verified by measured data from power plants. The boundary values are adapted to the photovoltaic output characteristics under various local weather conditions.
[0037] It should be noted that, in this application, power jump refers to the phenomenon of sudden numerical changes in photovoltaic power within a short period of time; jump characteristics refer to the characteristic information that quantitatively describes the amplitude, frequency and change law of power jump; sample entropy characteristics refer to the quantitative index that measures the complexity and randomness of time series; dynamic instability information refers to the relevant information of disordered fluctuations in photovoltaic power caused by weather changes; fluctuation boundary refers to the judgment threshold that distinguishes between stable changes and disordered fluctuations in photovoltaic power; low-frequency trend term refers to the main component with stable power changes, few jumps and low disorder; high-frequency fluctuation term refers to the disturbance component with frequent power jumps and high disorder.
[0038] In step S3, the periodic evolution characteristics of fixed fluctuations in photovoltaic power are extracted from the low-frequency trend term, and the local short-time dynamic characteristics of random fluctuations in photovoltaic power are extracted from the high-frequency fluctuation term.
[0039] In this embodiment, the extraction of the periodic evolution characteristics of fixed fluctuations in photovoltaic power from the low-frequency trend term can be achieved through the following steps: The power input sequence of the low-frequency trend term is aligned and transformed to obtain the power aligned amplitude. The main periodic information for fixed fluctuations in photovoltaic power is determined based on the power alignment amplitude. The periodic evolution characteristics of fixed fluctuations in photovoltaic power are determined by the aforementioned key periodic information.
[0040] In practical implementation, firstly, a complete low-frequency trend term power input sequence is selected as the processing object. Discrete Fourier Transform (DFT) is used for signal conversion. Following the standard algorithm rules, the inner product of each power sampling point in the time domain with the complex exponential basis function is calculated, and the complex results for the corresponding frequency points are solved sequentially. The magnitude of all complex results is extracted as the power-aligned amplitude. The entire calculation is based on measured low-frequency data from photovoltaic power plants. The transformation process follows the well-known principles of frequency domain analysis of time-series signals, ensuring that the amplitude results accurately correspond to the energy magnitude of each frequency component. Then, all power-aligned amplitudes are sorted from largest to smallest. Based on the energy proportion principle, the frequency points corresponding to the top-ranked amplitudes are selected. Combining the basic conversion relationship between frequency and period, the corresponding time period is calculated for each selected frequency point. All calculated periods are integrated and summarized to form the main periodic information corresponding to the fixed fluctuations in photovoltaic power. Then, according to the period length corresponding to the main period information, the one-dimensional low-frequency trend term sequence is cut into multiple groups of equal-length segments, which are then stacked in chronological order to form a two-dimensional tensor. A two-dimensional convolutional structure is used to perform feature extraction operations on the tensor corresponding to each group of periods, capturing the local changes within the period and the evolution relationship between different periods. The features extracted by multiple convolutions are weighted and fused to integrate the effective information corresponding to all periods, and finally generate complete periodic evolution features of fixed fluctuations in photovoltaic power.
[0041] It should be noted that, in this application, the power input sequence refers to the time-series data of the divided low-frequency components, which carries the long-term stable variation law of photovoltaic power and is the original input for periodic analysis; alignment transformation refers to the process of converting the time-domain power signal into a frequency-domain signal; power alignment amplitude refers to the frequency-domain energy value obtained after signal transformation; fixed fluctuation of photovoltaic power refers to the regular power change caused by stable factors such as solar altitude angle and intraday irradiance; main periodic information refers to the several sets of periodic parameters with the highest energy proportion in the power sequence; periodic evolution characteristics refer to the comprehensive time-series characteristics that integrate local changes within the period and cross-period evolution.
[0042] In addition, in this embodiment, determining the main periodic information of fixed fluctuations in photovoltaic power based on the power alignment amplitude can be achieved using the following steps: The power alignment amplitude is divided into multiple amplitude variation segments; By marking the fluctuation trend of all amplitude variation segments, the main periodic information of photovoltaic power fluctuations with constant fluctuations can be obtained.
[0043] In practice, the process begins by processing a complete power-aligned amplitude sequence. A segmentation threshold is set based on the continuous variation of the frequency domain signal. Following the frequency point arrangement, inflection points where amplitude increases then decreases and vice versa are identified as segmentation nodes. These inflection points serve as boundaries to divide the overall amplitude sequence into several continuous amplitude variation segments. All segmentation nodes are automatically identified based on measured frequency domain data, and the segmentation rules have been validated using multiple sets of photovoltaic low-frequency sequences. Next, each amplitude variation segment is iterated through, and the overall trend, fluctuation frequency, and amplitude of the amplitude within each segment are statistically analyzed to mark the fluctuation trend. Combining the conversion relationship between the frequency and time domains, the corresponding period length is calculated based on the characteristics of the marked segments. Segments with generally high amplitudes and strong trend regularity are selected, and their corresponding period parameters are integrated into the main period information.
[0044] It should be noted that, in this application, amplitude variation segment refers to a continuous data unit divided according to the amplitude fluctuation law, and each segment corresponds to a type of frequency energy change pattern; fluctuation trend marking refers to the operation of marking the amplitude rise and fall, stability and other change states within each segment.
[0045] Preferably, in this embodiment, local short-time dynamic features of random fluctuations in photovoltaic power are extracted from the high-frequency fluctuation term, with reference to... Figure 2 As shown in the figure, this is a flowchart illustrating the determination of local short-term dynamic features in some embodiments of this application. In this embodiment, the determination of local short-term dynamic features can be achieved by the following steps: In step S31, based on the volatility reduction factor for power fluctuations in the high-frequency fluctuation term, a random fluctuation component for random fluctuations in photovoltaic power is generated through a fluctuation evolution model. In step S32, the power interaction characteristics of the random fluctuation of photovoltaic power are determined based on the random fluctuation components; In step S33, fluctuation maintenance matching is performed based on the power interaction characteristics to obtain the local short-term dynamic characteristics of random fluctuations in photovoltaic power.
[0046] In practice, the process begins by iterating through all time-series data of the high-frequency fluctuation term, calculating the rate of change of power values segment by segment, and calibrating the volatility conversion factor by combining historical measured fluctuation data of the power plant. The original high-frequency fluctuation data and the conversion factor are then input into the fluctuation evolution model. The model extrapolates the power change state frame by frame according to a time-series progression rule, simulating the continuous fluctuation process caused by factors such as cloud disturbances. The calculation process uses model parameters trained with measured samples. After the extrapolation is completed, pure random fluctuation data without a basic trend is separated, which is the random fluctuation component. Then, using the generated random fluctuation component as the analysis object, continuous short-time data units are divided, and the numerical correlation between adjacent units is calculated one by one. The synchronous and lagging changes of different fluctuation units are statistically analyzed, and the mutual influence patterns between units are identified. Various linkage patterns are summarized and integrated, and the correlation strength between units is quantified. The entire statistical method has been verified through multi-scenario field tests, and finally, a complete power interaction characteristic is obtained. Finally, two matching standards were established: fluctuation duration and fluctuation amplitude. Power interaction characteristics were compared with the standard library one by one to complete the fluctuation maintenance matching, filter out instantaneous invalid clutter, and retain effective fluctuation information with continuous change characteristics. The filtered effective information was integrated and the features were refined to focus on the power change pattern under short time scale. This matching standard was established based on multiple sets of measured data on sunny, cloudy and rainy days, and finally output local short-term dynamic characteristics.
[0047] It should be noted that in this application, the fluctuation evolution model is constructed based on stochastic differential equations to describe the continuous stochastic evolution process of photovoltaic power under high-frequency fluctuation conditions. When constructing the fluctuation evolution model, historical time-series data of the high-frequency fluctuation term is first used as training samples to extract the volatility sequence of power fluctuations under different weather scenarios. A time-parameterized diffusion process is then used to model the time-varying characteristics of the volatility. Specifically, the volatility time series is decomposed into deterministic drift terms and random disturbance terms. The deterministic drift term characterizes the mean recovery trend of the volatility under clear-sky radiation boundary constraints, while the random disturbance term is driven by Brownian motion and characterizes cloud cover, etc. Unpredictable fluctuations caused by meteorological factors are addressed by using the maximum likelihood estimation method to calibrate the parameters of the fluctuation evolution model from historical fluctuation data. This ensures that the statistical characteristics of the random fluctuation components output by the model are consistent with the statistical characteristics of the measured power fluctuation sequence. During actual operation, the fluctuation rate conversion factor at the current moment is used as the initial state input of the fluctuation evolution model. Based on the calibrated drift coefficient and diffusion coefficient, the fluctuation evolution model extrapolates the power fluctuation state step by step according to the time-series recursive rule, generating random fluctuation components that match the current weather scenario. This fluctuation evolution model can simultaneously characterize the random fluctuation characteristics of photovoltaic power on minute-level and hour-level scales.
[0048] The fluctuation maintenance matching is used to identify fluctuation segments with continuously evolving characteristics from power interaction features, distinguishing between valid fluctuations and transient clutter. The specific matching process is as follows: First, a fluctuation maintenance standard library is constructed for matching. Using manually annotated valid fluctuation segments from historical operating data of photovoltaic power plants as samples, the duration distribution range and amplitude variation range of fluctuation events under various weather scenarios are statistically analyzed, and a multi-dimensional matching standard library is established accordingly. The standard library sets corresponding duration thresholds and amplitude tolerance ranges for three weather scenarios: sunny, cloudy, and rainy. Matching parameters for different scenarios are configured independently. During matching operations, fluctuation units in the power interaction features are input into the matcher one by one in chronological order. Each fluctuation unit contains the fluctuation duration, peak fluctuation amplitude, and fluctuation morphology description parameters for that time period. The matcher performs two steps of judgment in sequence: First, it determines whether the duration of the current fluctuation unit falls within the standard duration range of the corresponding weather scenario. If it does not meet the requirement, it is directly marked as invalid clutter and removed. Second, for the fluctuation unit that passes the judgment in the first step, it calculates the deviation rate between its peak fluctuation amplitude and the benchmark fluctuation amplitude of the corresponding scenario in the standard library. If the deviation rate falls within the preset tolerance range, it is judged as a valid fluctuation segment; otherwise, it is marked as an invalid fluctuation and removed.
[0049] It should be noted that, in this application, the volatility reduction factor refers to the correction parameter that quantifies the severity of high-frequency power fluctuations; the fluctuation evolution model refers to the time-series computational model that simulates short-term continuous fluctuations in photovoltaic power; the random fluctuation component refers to the data unit that purely reflects the random power changes caused by meteorological disturbances after the high-frequency fluctuation term is decomposed; the power interaction feature refers to the feature set of correlations between different time periods and different fluctuation units within the random fluctuation; the fluctuation maintenance matching refers to the computational process of matching the duration of power fluctuations against the interaction features; and the local short-term dynamic features refer to the dynamic laws that centrally embody short-term sudden changes and continuous disturbances in photovoltaic power, wherein the dynamic laws are the time-series evolution patterns exhibited by photovoltaic power during short-term continuous fluctuations, specifically including the fluctuation morphology change rules. The system employs two levels: the law of fluctuation pattern change and the law of fluctuation intensity evolution. The law of fluctuation pattern change describes the trend transformation characteristics of the power sequence from rising to falling or from falling to rising over a continuous period of time. It is characterized by extracting the amplitude difference and arrival time interval of adjacent extreme points in the fluctuation unit. The law of fluctuation intensity evolution describes the decay or divergence trend of fluctuation amplitude over time. It is quantified by calculating the first-order difference sequence of the amplitude of each sampling point in the fluctuation unit and its autocorrelation coefficient. The above dynamic laws are encoded in the form of feature vectors. Each group of fluctuation units corresponds to a feature vector. This feature vector, together with the original power values of all sampling points in the fluctuation unit, constitutes the local short-term dynamic features, which are input into the subsequent prediction model to complete the prediction of the change trend of high-frequency fluctuation terms.
[0050] In addition, in this embodiment, the power interaction characteristics of photovoltaic power random fluctuations can be determined based on the random fluctuation components using the following steps: The random integral interval of the random fluctuation of photovoltaic power is determined based on the random fluctuation components. By integrating the random integration interval into the fluctuation space, the power fluctuation attribute of photovoltaic power under random fluctuation is obtained. The power interaction characteristics of random fluctuations in photovoltaic power are determined based on the power fluctuation attributes.
[0051] In practice, the process begins by defining integration boundaries based on complete random fluctuation time-series data and the duration of short-term photovoltaic disturbances. Using measured power fluctuation duration data from the power plant, multiple continuous time calculation intervals are established, with interval lengths matching the typical duration of local short-term cloud disturbances. The integrity of data within each interval is verified, and intervals containing abnormal breakpoints are removed, leaving only continuous, valid intervals as random integration intervals. Then, spatial integration is performed on the power time-series data within each random integration interval. All power values within the interval are accumulated along the time dimension, following the general calculation rules for continuous time-series signal integration. The integration results for each interval are calculated sequentially, and the magnitude of the integral values is used to distinguish the strength of fluctuations. Simultaneously, the fluctuation patterns within the intervals are statistically analyzed, and a comprehensive summary is formed to establish power fluctuation attributes. Finally, the power fluctuation attributes of adjacent random integration intervals are compared, analyzing the continuity of fluctuation intensity and patterns between intervals. Synchronous and lagging changes in fluctuations across different intervals are statistically analyzed to quantify the degree of correlation between intervals. All correlation patterns are categorized and integrated, and unrelated discrete fluctuation information is removed, resulting in a complete power interaction feature summary.
[0052] It should be noted that, in this application, the random integration interval refers to the continuous calculation range defined for time series fluctuations; the fluctuation space integration refers to the integration operation along the time dimension for photovoltaic random fluctuation time series data within the defined random integration interval; and the power fluctuation attribute is a physical characteristic characterizing the fluctuation amplitude and cumulative intensity.
[0053] In step S4, the photovoltaic power components under various weather scenarios are predicted based on the periodic evolution characteristics and the local short-term dynamic characteristics, and the initial prediction results are superimposed and reconstructed to obtain the short-term prediction value of photovoltaic power.
[0054] In this embodiment, the prediction of photovoltaic power components under various weather scenarios based on the periodic evolution characteristics and the local short-term dynamic characteristics can be achieved through the following steps: Construct initial prediction models for photovoltaic power under various weather scenarios; The initial prediction model is used to predict the periodic evolution features and the local short-term dynamic features to obtain low-frequency initial prediction results and high-frequency initial prediction results. The low-frequency initial prediction result and the high-frequency initial prediction result are spliced together to obtain the initial prediction result, thus completing the prediction of photovoltaic power components under various weather scenarios.
[0055] In practical implementation, firstly, model combinations are built for three weather scenarios: sunny, cloudy, and rainy. For periodic evolution characteristics, a time-series two-dimensional variation modeling network (TimesNet) is constructed. This network adaptively mines multiple period lengths of the sequence using Fast Fourier Transform and reshapes one-dimensional data into two-dimensional tensors. Then, it combines the ability of multi-scale two-dimensional convolutional kernels to extract intra- and inter-periodic evolution features, adapting to the long-period pattern prediction of low-frequency trend terms. For the local short-term dynamic features of mid- and high-frequency fluctuation terms, a Modern Temporal Convolutional Network (ModernTCN) is built. Its ultra-large convolutional kernels expand the temporal receptive field, and combined with depthwise separable convolution and residual structures, it accurately captures short-term, drastic fluctuations caused by meteorological disturbances. Model training is conducted using corresponding components and real power data for each weather scenario. The mean squared error is used as the loss function to iteratively optimize the parameters. After multiple sets of experimental samples verify the model's convergence, the initial prediction model for the entire scenario is completed. Then, the low-frequency trend term is input into the time-series two-dimensional variation modeling network for the corresponding scenario. The model calls the trained parameters and combines them with the learned periodic evolution patterns to complete the calculation, outputting the initial low-frequency prediction result. The local short-term dynamic features are then input into a Modern Temporal Convolutional Network (ModernTCN) for the same scene. The model relies on its learned short-term fluctuation features to perform inferences and output high-frequency initial prediction results. The two sets of prediction operations are independent, and their operational logic follows the temporal inference rules established during model training. Finally, the low-frequency and high-frequency initial prediction results are aligned along the time dimension to ensure that the time steps and number of data points in both sequences are completely consistent. A feature-dimensional concatenation method is used to integrate the two prediction sequences bit by bit, forming a complete time-series data sequence. This concatenation method follows the general rules of time-series data fusion and does not change the original values or variation patterns of individual prediction data. After concatenation, the prediction of all photovoltaic power components under the current weather scenario is complete.
[0056] It should be noted that, in this application, the initial prediction model refers to a combination of dedicated deep learning models built separately for different weather scenarios; the low-frequency initial prediction result refers to the predicted numerical sequence of low-frequency trend terms output by the time-series two-dimensional change modeling network; the high-frequency initial prediction result refers to the predicted numerical sequence of high-frequency fluctuation terms output by the modern time convolutional network; and the initial prediction result refers to the complete photovoltaic power prediction sequence formed after splicing.
[0057] In this embodiment, the initial prediction results are superimposed and reconstructed to obtain the short-term prediction value of photovoltaic power, which can be achieved by the following steps: Determine the target weather scenario to which meteorological characteristics belong when forecasting photovoltaic power; The photovoltaic power prediction window is determined based on the target weather scenario. The initial prediction results are linearly superimposed based on the power change characteristics in the power prediction window to obtain the short-term prediction value of photovoltaic power.
[0058] In practice, the process begins by collecting and preprocessing complete meteorological data for the forecast day. After removing outliers and filling in missing information, a standard feature vector is constructed. A previously trained granular K-means clustering model is then invoked, using granules as the basic operational unit. The distance between the current meteorological features and each cluster center is calculated, and samples are categorized based on this distance. Combined with the calibration results of features such as irradiance and cloud cover, the target weather scenario corresponding to the current weather is ultimately determined. Next, based on the identified target weather scenario and referring to the time length set during model training for that scenario, continuous photovoltaic power data prior to the forecast time is extracted. This ensures that the extracted data duration and sampling interval remain consistent with the model training parameters, forming a standard photovoltaic power forecast window. Simultaneously, power trends and fluctuations within the window are extracted for reference in subsequent overlay calculations. The window selection rules have been validated through long-term operational data and are adapted to the power output characteristics under different weather conditions. Finally, based on the power change characteristics extracted within the power prediction window, the execution rules for the superposition operation are determined. The initial prediction results that have already been stitched together are linearly superimposed point by point in chronological order, with the low-frequency trend prediction values and high-frequency fluctuation prediction values added sequentially. The superposition process strictly follows the general rules for time-series data operations, without altering the original change patterns of individual components. After the operation is completed, complete time-series data is generated, which is the short-term prediction value of photovoltaic power.
[0059] It should be noted that, in this application, the target weather scenario refers to the actual sunny, cloudy, or rainy weather scenario corresponding to the forecast period; the photovoltaic power forecast window refers to the extracted continuous historical photovoltaic power time series segment; the power change characteristics refer to the trend and fluctuation state of power within the forecast window; linear superposition refers to the data fusion method of adding the component forecast values point by point according to the time series; and the short-term photovoltaic power forecast value refers to the final power forecast time series data obtained after superposition.
[0060] Therefore, this application involves superimposing and reconstructing the initial prediction results to obtain short-term photovoltaic power predictions. Identifying the low-frequency trend term and the high-frequency fluctuation term allows for the separation of different time-scale variations in the photovoltaic power sequence, achieving multi-scale data decoupling. This partitioning method establishes objective discrimination criteria based on Pearson correlation coefficient and sample entropy, accurately distinguishing between the fundamental variation patterns of photovoltaic output and random fluctuations caused by meteorological disturbances. Based on this, differentiated processing logic can be matched to the two types of components, adapting to power variation characteristics under different weather scenarios such as sunny, cloudy, and rainy days. This lays a solid data foundation for grouped modeling and adaptive prediction of meteorological scenarios, effectively avoiding the shortcomings of a single data structure that cannot accommodate different fluctuation patterns, and improving the overall prediction accuracy from the data source level. Identifying the periodic evolution characteristics and local short-term dynamic characteristics allows for the extraction of the core temporal expression information of the two types of power components. The periodic evolution characteristics fully preserve the long-term evolution patterns of the low-frequency trend, while the local short-term dynamic characteristics accurately depict the instantaneous disturbance characteristics of high-frequency fluctuations, achieving professional extraction of feature dimensions. The two types of features can be directly connected to the appropriate deep learning models, giving full play to the technical advantages of different models, strengthening the ability of multi-scale component collaborative modeling, enabling the model to accurately capture force features in various weather scenarios, further reducing prediction errors, and comprehensively improving the overall accuracy and scenario adaptability of short-term photovoltaic power prediction.
[0061] In summary, the technical solution adopted in this application can achieve adaptive weather scenario prediction and multi-scale component collaborative prediction, thereby improving the accuracy of short-term photovoltaic power prediction under various weather conditions.
[0062] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0063] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.
[0064] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
Claims
1. A method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components, characterized in that, The prediction method includes: Weather scene samples are collected and divided according to historical meteorological characteristics to obtain multiple typical weather scenes; Multi-scale decomposition is performed on the photovoltaic power sequence under each typical weather scenario to obtain the photovoltaic power frequency characteristics at different scales. Based on each photovoltaic power frequency characteristic and the sample entropy characteristics corresponding to the weather scenario sample, the low-frequency trend term and high-frequency fluctuation term of photovoltaic power are determined. The periodic evolution characteristics of fixed fluctuations in photovoltaic power are extracted from the low-frequency trend term, and the local short-time dynamic characteristics of random fluctuations in photovoltaic power are extracted from the high-frequency fluctuation term. Based on the periodic evolution characteristics and the local short-term dynamic characteristics, the photovoltaic power components under various weather scenarios are predicted to obtain initial prediction results. The initial prediction results are then superimposed and reconstructed to obtain short-term prediction values of photovoltaic power.
2. The method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components as described in claim 1, characterized in that, Based on historical meteorological characteristics, the weather scene samples were divided into several typical weather scenes, including: Obtain historical meteorological characteristics during the operation of photovoltaic power plants; Based on the historical meteorological characteristics, the weather information in the weather scene sample is divided to obtain multiple typical weather scenes.
3. The method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components as described in claim 1, characterized in that, Multi-scale decomposition of photovoltaic power sequences under each typical weather scenario yields photovoltaic power frequency characteristics at different scales, including: Determine the photovoltaic power sequence for each typical weather scenario; Modal decomposition is performed on the photovoltaic power sequence to obtain intrinsic photovoltaic power information; The intrinsic photovoltaic power information is classified by scale to obtain photovoltaic power frequency characteristics at different scales.
4. The method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components as described in claim 3, characterized in that, Modal decomposition of the photovoltaic power sequence to obtain intrinsic photovoltaic power information specifically includes: Extract the intrinsic mode components corresponding to the photovoltaic power from the photovoltaic power sequence; The intrinsic photovoltaic power information is obtained by identifying the components of the photovoltaic power sequence using the intrinsic mode components.
5. The method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components as described in claim 1, characterized in that, The low-frequency trend term and high-frequency fluctuation term of photovoltaic power are determined based on the frequency characteristics of each photovoltaic power and the sample entropy characteristics corresponding to the weather scenario samples. Specifically, these include: Determine the jump characteristics of photovoltaic power in power jumps based on the frequency characteristics of each photovoltaic power; The dynamic instability information of photovoltaic power when the weather scene changes is determined based on the sample entropy characteristics corresponding to the weather scene samples; The fluctuation limits of photovoltaic power are determined based on all the jump characteristics and the dynamic instability information, and then the low-frequency trend term and high-frequency fluctuation term of photovoltaic power are determined through the fluctuation limits.
6. The method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components as described in claim 1, characterized in that, Extracting the periodic evolution characteristics of fixed fluctuations in photovoltaic power from the low-frequency trend term specifically includes: The power input sequence of the low-frequency trend term is aligned and transformed to obtain the power aligned amplitude. The main periodic information for fixed fluctuations in photovoltaic power is determined based on the power alignment amplitude. The periodic evolution characteristics of fixed fluctuations in photovoltaic power are determined by the aforementioned key periodic information.
7. The method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components as described in claim 6, characterized in that, The main periodic information for determining the fixed fluctuation of photovoltaic power based on the power alignment amplitude specifically includes: The power alignment amplitude is divided into multiple amplitude variation segments; By marking the fluctuation trend of all amplitude variation segments, the main periodic information of photovoltaic power fluctuations with constant fluctuations can be obtained.
8. The method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components as described in claim 1, characterized in that, Extracting the local short-time dynamic features of random fluctuations in photovoltaic power from the high-frequency fluctuation term specifically includes: Based on the volatility reduction factor for power fluctuations in the high-frequency fluctuation term, random fluctuation components for random fluctuations in photovoltaic power are generated through a fluctuation evolution model. The power interaction characteristics of photovoltaic power random fluctuations are determined based on the random fluctuation components. By performing fluctuation maintenance matching based on the power interaction characteristics, the local short-term dynamic characteristics of random fluctuations in photovoltaic power are obtained.
9. The method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components as described in claim 8, characterized in that, The power interaction characteristics of photovoltaic power random fluctuations determined based on the random fluctuation components specifically include: The random integral interval of the random fluctuation of photovoltaic power is determined based on the random fluctuation components. By integrating the random integration interval into the fluctuation space, the power fluctuation attribute of photovoltaic power under random fluctuation is obtained. The power interaction characteristics of random fluctuations in photovoltaic power are determined based on the power fluctuation attributes.
10. The method for short-term photovoltaic power prediction based on weather scenarios and multi-scale components as described in claim 1, characterized in that, The prediction of photovoltaic power components under various weather scenarios based on the aforementioned periodic evolution characteristics and local short-term dynamic characteristics specifically includes: Construct initial prediction models for photovoltaic power under various weather scenarios; The initial prediction model is used to predict the periodic evolution features and the local short-term dynamic features to obtain low-frequency initial prediction results and high-frequency initial prediction results. The low-frequency initial prediction result and the high-frequency initial prediction result are spliced together to obtain the initial prediction result, thus completing the prediction of photovoltaic power components under various weather scenarios.
