Photovoltaic short-term power prediction method and device of multi-scale decomposition-aggregation architecture

Abstract

The invention is suitable for the technical field of new energy and artificial intelligence, and provides a photovoltaic short-term power prediction method and device of a multi-scale decomposition-aggregation architecture, and the method comprises the steps: obtaining first meteorological data; performing multi-scale characteristic decomposition on the first meteorological data to obtain a plurality of second meteorological data; the scales of the second meteorological data are different; performing multiple times of feature processing on each piece of second meteorological data to obtain a first feature component and a second feature component corresponding to each piece of second meteorological data after each time of feature processing; the first feature component is used for representing meteorological periodic features in the second meteorological data; the second feature component is used for representing meteorological trend features in the second meteorological data; performing feature fusion according to the plurality of first feature components and the second feature components to obtain third meteorological data corresponding to each second meteorological data; and performing photovoltaic power prediction according to the plurality of third meteorological data to obtain a first prediction result. According to the method, the prediction precision of the photovoltaic power can be improved.

Description

A method and apparatus for short-term photovoltaic power prediction based on a multi-scale decomposition-aggregation architecture Technical Field

[0001] This application belongs to the fields of new energy and artificial intelligence technology, and in particular relates to a method and device for short-term photovoltaic power prediction with a multi-scale decomposition-aggregation architecture. Background Technology

[0002] Against the backdrop of my country's energy system transitioning towards cleaner and lower-carbon energy, the photovoltaic new energy industry is developing rapidly in terms of industrialization and scale. However, photovoltaic power output is significantly unstable and volatile due to environmental influences. High grid connection ratios pose challenges to the dynamic balance of power source and load and the safe and economical operation of the power system. Short-term power prediction technology can provide data support for power plant planning and grid dispatch, improve photovoltaic absorption capacity, and reduce the phenomenon of "curtailment," thus possessing significant application value.

[0003] Current mainstream photovoltaic power prediction is based on a data-driven paradigm, mainly including statistical methods and deep learning methods. Statistical methods can extract shallow and stable mapping relationships in data and make stable predictions under good weather conditions, but they are difficult to capture the deep-seated trend and seasonal dynamic characteristics of meteorological and power sequences, and cannot meet the needs of long-term high-precision prediction. Although deep learning methods are good at mining complex patterns and dependencies, they are affected by the complex temporal characteristics of photovoltaic power sequences, making it difficult to achieve long-term high-precision prediction of photovoltaic power. Summary of the Invention

[0004] This application provides a method and apparatus for predicting short-term photovoltaic power using a multi-scale decomposition-aggregation architecture, which can improve the accuracy of photovoltaic power prediction.

[0005] In a first aspect, embodiments of this application provide a method for short-term photovoltaic power prediction, comprising: acquiring first meteorological data; wherein the first meteorological data is weather forecast sequence data within a preset time period; performing multi-scale feature decomposition on the first meteorological data to obtain multiple second meteorological data; wherein each second meteorological data has a different scale; performing multiple feature processing on each second meteorological data to obtain a first feature component and a second feature component corresponding to each second meteorological data after each feature processing; wherein the first feature component is used to represent the meteorological periodicity feature in the second meteorological data; the second feature component is used to represent the meteorological trend feature in the second meteorological data; performing feature fusion based on the multiple first feature components and second feature components to obtain third meteorological data corresponding to each second meteorological data; and performing photovoltaic power prediction based on the multiple third meteorological data to obtain a first prediction result.

[0006] In this embodiment, weather forecast sequence data within a preset time period is first acquired as the basic input. Then, multi-scale feature decomposition is performed on this meteorological data to obtain subdivided meteorological data at different scales. Subsequently, for each scale of meteorological data, multiple feature processing steps are performed to extract a first feature component representing the periodic variation pattern of meteorological data and a second feature component representing the long-term evolution trend of meteorological data. Next, the two types of feature components at the same scale are fused to generate the fused meteorological feature (third meteorological data) corresponding to that scale. Finally, the fused meteorological features from all scales are combined to complete the photovoltaic power prediction and output the final prediction result. This method separates meteorological features at different time scales through multi-scale feature decomposition, avoiding mutual interference between different scale patterns; it extracts features with clear physical meaning through periodic-trend component separation, strengthening the correlation between meteorological factors and photovoltaic power; and it achieves information complementarity of features from different dimensions through multi-component feature fusion, taking into account both the short-term fluctuation characteristics of photovoltaic power and supporting the fitting of long-term variation trends. It progresses step-by-step from the three key stages of feature extraction, characterization, and integration, reducing invalid feature interference and prediction bias, ultimately achieving a significant improvement in the accuracy of photovoltaic power prediction.

[0007] In one possible implementation of the first aspect, feature fusion is performed based on multiple first feature components and second feature components to obtain third meteorological data corresponding to each second meteorological data, including: after each feature processing, fusing the first feature components corresponding to every two adjacent scales of second meteorological data to obtain periodic fused data corresponding to each second meteorological data; fusing the second feature components corresponding to every two adjacent scales of second meteorological data to obtain trend fused data corresponding to each second meteorological data; and obtaining third meteorological data corresponding to each second meteorological data based on the periodic fused data and trend fused data corresponding to each second meteorological data.

[0008] In this embodiment, by fusing periodic and trend components across dimensions and scales, fine-scale details are added to periodic features, and coarse-scale macro-guidance is injected into trend features. This also achieves deep integration of the two types of core features, effectively mining the deep dynamic correlation between meteorological and power sequences, solving the problem of insufficient modeling of single-scale features, and significantly improving the accuracy and stability of photovoltaic short-term power prediction.

[0009] In one possible implementation of the first aspect, the step of fusing the target feature components corresponding to every two adjacent scales of second meteorological data to obtain target fused data corresponding to each second meteorological data includes: converting the second meteorological data corresponding to the first scale in every two adjacent scales of second meteorological data into fourth meteorological data; wherein the scale of the fourth meteorological data is the same as the scale of the second meteorological data at the second scale; superimposing the second meteorological data at the second scale and the fourth meteorological data to obtain target fused data corresponding to the second meteorological data at the second scale; wherein, when the target feature component is the first feature component, the target fused data is recorded as periodic fused data, and the first scale is a fine scale and the second scale is a coarse scale; when the target feature component is the second feature component, the target fused data is recorded as trend fused data, and the first scale is a coarse scale and the second scale is a fine scale.

[0010] In this embodiment, by first converting the corresponding meteorological data in adjacent scales into a unified scale and then superimposing and fusing them, detailed supplementation from high scale to low scale is achieved for periodic components, and macroscopic guidance from low scale to high scale is achieved for trend components. This accurately realizes cross-scale information complementarity of different feature components, effectively improving the completeness and accuracy of feature expression, and providing better feature support for subsequent photovoltaic power prediction.

[0011] In one possible implementation of the first aspect, the third meteorological data corresponding to each second meteorological data is obtained based on the periodic fusion data and trend fusion data corresponding to each second meteorological data, including: after each feature processing, performing feature fusion on the periodic fusion data and trend fusion data corresponding to each second meteorological data to obtain the first fusion data corresponding to each second meteorological data; according to the feature processing order, fusing every two adjacent first fusion data to obtain multiple second fusion data; and obtaining multiple third meteorological data based on the multiple second fusion data.

[0012] In this embodiment, by first fusing periodic and trend data within a single round, and then integrating adjacent fusion results across rounds, a deep aggregation of multi-round and multi-dimensional features is achieved in a progressive manner. This not only fully preserves the core meteorological information at each scale, but also strengthens the correlation and integrity of features, effectively improving the modeling ability for complex time series patterns and providing more accurate and reliable feature support for short-term photovoltaic power prediction.

[0013] In one possible implementation of the first aspect, each pair of adjacent first fused data is fused according to the feature processing order to obtain multiple second fused data, including: in each pair of adjacent first fused data, the first fused data corresponding to each second meteorological data in the first feature processing sequence number is superimposed with the first fused data corresponding to each second meteorological data in the second feature processing sequence number to obtain the second fused data corresponding to each second meteorological data in the second feature processing sequence number; wherein, the first feature processing sequence number is the preceding sequence number of the second feature processing sequence number.

[0014] In this embodiment, by first fusing periodic and trend data within a single round, and then integrating adjacent fusion results across rounds, a deep aggregation of multi-round and multi-dimensional features is achieved in a progressive manner. This not only fully preserves the core meteorological information at each scale, but also strengthens the correlation and integrity of features, effectively improving the modeling ability for complex time series patterns and providing more accurate and reliable feature support for short-term photovoltaic power prediction.

[0015] In one possible implementation of the first aspect, obtaining multiple third meteorological data based on multiple second fused data includes: determining a third feature processing sequence number from the feature processing sequence, and determining the second fused data of each of the second meteorological data corresponding to the third feature processing sequence number as the third meteorological data; wherein the third feature processing sequence number is the last feature processing sequence number in the feature processing sequence.

[0016] In this embodiment, the second fused data corresponding to the last number in the feature processing sequence is directly selected as the third meteorological data. This fully preserves the complete feature information after multiple rounds and cross-scale fusion, avoids information loss caused by additional processing, and ensures that the output feature data has both depth and reliability, providing high-quality and standardized core input for subsequent photovoltaic power prediction.

[0017] In one possible implementation of the first aspect, photovoltaic power prediction is performed based on multiple third meteorological data to obtain a first prediction result, including: predicting each third meteorological data to obtain a first sub-prediction result corresponding to each third meteorological data; and obtaining the first prediction result based on multiple first sub-prediction results.

[0018] In this embodiment, the first sub-prediction result is obtained by independently predicting the third meteorological data at each scale, and then the multi-sub-prediction results are integrated to obtain the final first prediction result. This not only gives full play to the prediction advantages of the characteristics of each scale (capturing short-term changes at a fine scale and grasping macro trends at a coarse scale), but also realizes the complementary fusion of multi-scale prediction information, effectively improving the accuracy and stability of photovoltaic short-term power prediction and making up for the limitations of single-scale prediction.

[0019] In one possible implementation of the first aspect, obtaining a first prediction result based on multiple first sub-prediction results includes: assigning each of the multiple first prediction results a corresponding first weight; multiplying each first prediction result with its corresponding first weight to obtain a corresponding second sub-prediction result; and superimposing the multiple second sub-prediction results to obtain the first prediction result.

[0020] In this embodiment, by first dynamically allocating the first weight according to the reliability of each first sub-prediction result, then strengthening the advantageous prediction and weakening the inefficient prediction through weighted multiplication, and finally superimposing and integrating the multi-dimensional weighted sub-prediction results, the complementary advantages of predictions at each scale are fully integrated, effectively improving the accuracy and stability of photovoltaic short-term power prediction and making up for the limitations of single-scale prediction.

[0021] Secondly, embodiments of this application provide a photovoltaic power prediction device, comprising: a meteorological data acquisition module for acquiring first meteorological data; wherein the first meteorological data is weather forecast sequence data within a preset time period; a multi-scale decomposition module for performing multi-scale feature decomposition on the first meteorological data to obtain multiple second meteorological data; wherein each second meteorological data has a different scale; a feature processing module for performing multiple feature processing on each second meteorological data to obtain a first feature component and a second feature component corresponding to each second meteorological data after each feature processing; wherein the first feature component is used to represent the meteorological periodicity feature in the second meteorological data; and the second feature component is used to represent the meteorological trend feature in the second meteorological data; a fusion processing module for performing feature fusion based on the multiple first feature components and second feature components to obtain third meteorological data corresponding to each second meteorological data; and a power prediction module for performing photovoltaic power prediction based on the multiple third meteorological data to obtain a first prediction result.

[0022] Thirdly, embodiments of this application provide a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the photovoltaic short-term power prediction method as described in any of the first aspects above.

[0023] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the photovoltaic short-term power prediction method as described in any of the first aspects above.

[0024] Fifthly, embodiments of this application provide a computer program product that, when run on a terminal device, causes the terminal device to execute the photovoltaic short-term power prediction method described in any of the first aspects above.

[0025] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 is a flowchart illustrating the photovoltaic short-term power prediction method provided in this application embodiment; Figure 2 is a flowchart illustrating the generation of third meteorological data provided in this application embodiment (first); Figure 3 is a flowchart illustrating the generation of target fusion data provided in this application embodiment; Figure 4 is a flowchart illustrating the generation of third meteorological data provided in this application embodiment (second); Figure 5 is a flowchart illustrating the obtaining of prediction results provided in this application embodiment (first); Figure 6 is a flowchart illustrating the obtaining of prediction results provided in this application embodiment (second); Figure 7 is a schematic diagram of the system structure for photovoltaic power prediction provided in this application embodiment; Figure 8 is a schematic diagram of the structure for periodic component fusion provided in this application embodiment; Figure 9 is a schematic diagram of the structure for trend component fusion provided in this application embodiment; Figure 10 is a schematic diagram of the structure for multi-scale prediction aggregation provided in this application embodiment; Figure 11 is a structural block diagram of the photovoltaic power prediction device provided in this application embodiment; Figure 12 is a schematic diagram of the structure of the terminal device provided in this application embodiment. Detailed Implementation

[0028] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0029] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0030] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0031] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0032] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0033] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0034] Against the backdrop of my country's energy system transitioning towards cleaner and lower-carbon energy, the photovoltaic new energy industry is developing rapidly in terms of industrialization and scale. However, photovoltaic power output is significantly unstable and volatile due to environmental influences. High grid connection ratios pose challenges to the dynamic balance of power source and load and the safe and economical operation of the power system. Short-term power prediction technology can provide data support for power plant planning and grid dispatch, improve photovoltaic absorption capacity, and reduce the phenomenon of "curtailment," thus possessing significant application value.

[0035] Current mainstream photovoltaic power prediction is based on a data-driven paradigm, mainly including statistical methods and deep learning methods. Statistical methods can extract shallow and stable mapping relationships in data and make stable predictions under good weather conditions, but they are difficult to capture the deep-seated trend and seasonal dynamic characteristics of meteorological and power sequences, and cannot meet the needs of long-term high-precision prediction. Although deep learning methods are good at mining complex patterns and dependencies, they are affected by the complex temporal characteristics of photovoltaic power sequences, making it difficult to achieve long-term high-precision prediction of photovoltaic power.

[0036] To address the issues of unstable photovoltaic (PV) output, the difficulty of extracting deep-seated trend and periodic features using existing statistical methods, and the inadequacy of mainstream deep learning models in modeling multivariate coupling and complex temporal dependencies, this invention proposes a short-term PV power prediction algorithm model, DA-SolarPower, based on a decomposable multi-scale aggregation architecture. This model first performs mean downsampling on the input numerical weather prediction (NWP) to construct a multi-scale sequence. Subsequently, a temporal decomposable fusion (TDF) module is used to decouple periodic and trend features from the multi-scale sequence, and the periodic and trend feature components at different scales are fused using a combination of bottom-up and top-down approaches. Finally, a multi-predictor aggregation (MPA) module coordinates the temporal representations and spatial dependencies at each scale to perform complementary predictions, thereby generating refined PV power prediction results.

[0037] Referring to Figure 1, which is a flowchart of the photovoltaic short-term power prediction method provided in the embodiment of this application, and is intended as an example rather than a limitation, the method may include the following steps: S101, obtaining first meteorological data; wherein, the first meteorological data is weather forecast sequence data within a preset time period.

[0038] In this embodiment, the first meteorological data is weather forecast sequence data within a preset time period. This refers to the primary step in the short-term photovoltaic power prediction process: collecting a set of continuous numerical weather forecast data (i.e., "weather forecast sequence data") containing multiple meteorological observation variables within a specific time range (i.e., the "preset time"). This data is the core input for algorithm modeling and must reflect the temporal variation patterns of meteorological conditions, providing fundamental data support for subsequent analysis of the dynamic correlation between meteorology and photovoltaic power, and for achieving power prediction for the next 24-72 hours. The acquisition of the first meteorological data must focus on "data source reliability, variable completeness, time scale adaptability, and sequence continuity."

[0039] For example, by connecting to authoritative meteorological data release platforms (such as data interfaces of the National Meteorological Department, application interfaces of professional meteorological service providers, etc.), the data collection range is accurately located based on the geographical coordinates (latitude, longitude, altitude, etc.) of the photovoltaic power station. A preset time window is set (usually historical meteorological data 1-7 days before the forecast, which needs to be matched with the forecast period of 24-72 hours). Data on key meteorological variables affecting photovoltaic output, such as solar irradiance, temperature, wind speed, humidity, and cloud cover, are collected in batches. After data cleaning (removing missing values ​​and outliers) and format standardization (unifying timestamps and unit conversions), the data is organized into a continuous sequence of data arranged in chronological order. This ensures that each time step (such as every 15 minutes or 1 hour) corresponds to complete multi-dimensional meteorological observations, and finally forms the first meteorological data (numerical weather forecast sequence) that meets the algorithm input requirements.

[0040] S102, perform multi-scale feature decomposition on the first meteorological data to obtain multiple second meteorological data; among them, each second meteorological data has a different scale.

[0041] In this embodiment, multi-scale feature decomposition of the first meteorological data is the core preprocessing step of the DA-SolarPower algorithm. This involves using the input weather forecast sequence (first meteorological data) within a preset time period as a basis, and through specific downsampling and feature mapping operations, decomposing the original single-scale meteorological time-series data into multiple sub-sequences (i.e., second meteorological data) with different time granularities. The second meteorological data at different scales carry information of different dimensions: fine-scale data focuses on high-frequency fluctuations and detailed changes in meteorological conditions, while coarse-scale data highlights the macroscopic trends and long-term patterns of the meteorological sequence. This lays the foundation for subsequent feature deconstruction and fusion in the TDF module and multi-scale prediction aggregation in the MPA module, ultimately achieving accurate modeling of the complex correlation between meteorological and photovoltaic power.

[0042] For example, the model first processes the input sequence through multi-scale building blocks. This process is then performed. Specifically, average pooling is used to... Downsampling to multiple different scales forms a multi-scale sequence set. , of which The sequence of each scale is represented as , In this set of sequences, the lowest-scale sequence It carries the most subtle time patterns and high-frequency fluctuations, while the highest-scale sequences This leads to changes in the macroscopic trend of the dominant sequence. Subsequently, the multi-scale sequence is integrated through a shared embedding layer. Mapping to deep feature representation The process is formally defined as follows: At this point, the model has completed the multi-scale feature decomposition of the original input.

[0043] S103, perform multiple feature processing on each second meteorological data to obtain the first feature component and the second feature component corresponding to each second meteorological data after each feature processing; wherein, the first feature component is used to represent the meteorological periodicity feature in the second meteorological data; and the second feature component is used to represent the meteorological trend feature in the second meteorological data.

[0044] In this embodiment, performing multiple feature processing steps on each second meteorological data point is the core operation of the Temporal Deconstruction and Fusion (TDF) module in the DA-SolarPower algorithm. The TDF module includes multiple TDF modules, each performing feature processing on each second meteorological data point. This processing involves repeatedly performing feature deconstruction steps on each scale of the second meteorological data (meteorological sequences at different time granularities) obtained after multi-scale feature decomposition to separate two independent features from each second meteorological data point: a first feature component (corresponding to meteorological periodic features, depicting short-term stable periodic changes) and a second feature component (corresponding to meteorological trend features, reflecting long-term non-stationary macroscopic change trends). This lays the foundation for subsequent cross-scale feature fusion and accurate modeling of the dynamic correlation between meteorology and photovoltaic power.

[0045] Specifically, during feature processing for each TDF module, for a single second meteorological data point, it can be decomposed into a first feature component (meteorological periodicity feature) using Autoformer's sequence decomposition module. The second characteristic component (meteorological trend characteristics) ,Right now Periodic features capture short-term recurring fluctuations in meteorological data (such as periodic changes in daily light intensity), while trend features extract long-term overall trends (such as trends in sunshine duration brought about by seasonal changes).

[0046] S104, feature fusion is performed based on multiple first feature components and second feature components to obtain the third meteorological data corresponding to each second meteorological data.

[0047] In this embodiment, "multiple first feature components" refers to the set of all meteorological periodic features separated after each feature processing, and "multiple second feature components" refers to the set of all corresponding meteorological trend features. By using a differentiated fusion strategy, the two types of components are fused across scales and hierarchically. Then, the fused periodic features and trend features are integrated to finally obtain comprehensive feature data (third meteorological data) with complete periodic details and macro trend patterns at each scale (i.e., corresponding to each second meteorological data). This provides high-quality and strongly correlated feature support for subsequent multi-scale prediction aggregation.

[0048] In one embodiment, referring to Figure 2, which is a schematic diagram of the process for generating third meteorological data provided in this application embodiment, as shown in Figure 2, step S104 includes: S201, after each feature processing, the first feature components corresponding to every two adjacent scales of second meteorological data are fused to obtain periodic fused data corresponding to each second meteorological data.

[0049] In this embodiment, the periodic component, or first feature component, describes the dynamic characteristics of the periodic changes in the historical sequence. The detailed information contained therein is crucial for accurately predicting future power periodic fluctuations. The periodic component fusion adopts a detail-to-macro approach, gradually merging information from low-level fine-scale sequences upwards to supplement detailed information for coarse-scale periodic modeling.

[0050] The "first feature component" refers to the meteorological periodicity (characterizing short-term stable periodic changes), and "every two adjacent scales" refers to the continuous scale hierarchy formed after multi-scale decomposition (such as scale 1 and scale 2, scale 2 and scale 3, etc.). By selectively fusing the periodicity of adjacent scales, detailed information from fine scales is supplemented to coarse scales, and periodic patterns interact across scales. Ultimately, a periodic fusion data integrating the correlation information of adjacent scales is generated from the second meteorological data at each scale, strengthening the integrity and correlation of periodicity, and providing support for subsequent overall feature fusion and prediction.

[0051] S202, fuse the second feature components corresponding to every two adjacent scales of the second meteorological data to obtain the trend fused data corresponding to each second meteorological data.

[0052] In this embodiment, unlike periodic components, changes in detailed information introduce noise when capturing macro trends, and higher-level coarse-scale sequences can provide clearer macro trend information. Therefore, the trend component, i.e., the second characteristic component, adopts a macro-to-detail fusion approach, using coarse-scale macro information to guide trend modeling at the fine scale.

[0053] The "second feature component" refers to meteorological trend characteristics (reflecting long-term non-stationary macroscopic change trends), and "every two adjacent scales" refers to the continuous scale levels after multi-scale decomposition (such as scale 2 and scale 3, scale 3 and scale 4, etc.). By selectively fusing the trend characteristics of adjacent scales, the coarse-scale macroscopic trend information is transmitted to the fine-scale scale, and cross-scale trend patterns are coordinated. Finally, a trend fusion data integrating the correlation information of adjacent scales is generated for the second meteorological data of each scale, which strengthens the coherence and accuracy of trend characteristics and provides support for subsequent overall feature fusion and prediction.

[0054] In one embodiment, referring to Figure 3, which is a schematic diagram of the process of generating target fusion data provided by the embodiment of this application, as shown in Figure 3, the step of fusing the target feature components corresponding to the second meteorological data at every two adjacent scales to obtain the target fusion data corresponding to each second meteorological data includes: S301, converting the second meteorological data corresponding to the first scale in every two adjacent scales of the second meteorological data into fourth meteorological data; wherein, the scale of the fourth meteorological data is the same as the scale of the second meteorological data at the second scale.

[0055] In the embodiments of this application, "every two adjacent scales" refers to the continuous scale hierarchy after multi-scale decomposition (such as scale 1 and scale 2, scale 2 and scale 3, etc.). The definition of "first scale" and "second scale" is not fixed, but dynamically determined by the fusion type: during periodic fusion, the first scale is a fine scale and the second scale is an adjacent coarse scale (i.e., bottom-up fusion logic); during trend fusion, the first scale is a coarse scale and the second scale is an adjacent fine scale (i.e., top-down trend guidance logic). The fourth meteorological data generated through scale transformation is completely consistent with the specifications of the second meteorological data corresponding to the second scale, ensuring that the cross-scale interaction and merging of the two types of features can be effectively carried out.

[0056] For example, for a multi-scale periodic component sequence set A residual join can be used to construct a Layers, the core of this operation is the "cross-scale fusion formula of fine-scale and coarse-scale features", that is, through... It is possible to convert fine-scale second-scale meteorological data into coarse-scale fourth-scale meteorological data. Essentially, this involves integrating features of different granularities through scale adaptation and information interaction to simultaneously preserve details and trends.

[0057] Another example is for multi-scale trend components. Trend fusion employs residual connections The layer enables top-down information interaction, and the core of this operation is the "cross-scale fusion formula of coarse-scale and fine-scale features," that is, through... Coarse-scale second-level meteorological data can be transformed into fine-scale fourth-level meteorological data. Essentially, this involves integrating features of different granularities through scale adaptation and information interaction to simultaneously preserve details and trends.

[0058] S302, superimpose the second meteorological data and the fourth meteorological data at the second scale to obtain the target fusion data corresponding to the second meteorological data at the second scale; wherein, when the target feature component is the first feature component, the target fusion data is recorded as periodic fusion data, and the first scale is a fine scale and the second scale is a coarse scale; when the target feature component is the second feature component, the target fusion data is recorded as trend fusion data, and the first scale is a coarse scale and the second scale is a fine scale.

[0059] In the embodiments of this application, "overlaying the second meteorological data and the fourth meteorological data at the second scale to obtain the target fused data corresponding to the second meteorological data at the second scale" is a key execution step of cross-scale feature fusion in the DA-SolarPower algorithm. The core is to achieve complementary integration of features at different scales through "data overlay". The fusion logic needs to dynamically adjust the attributes of the first and second scales according to the target feature components (periodicity / trend).

[0060] When the target feature component is the first feature component (periodic feature), the first scale is defined as a fine scale (carrying details of high-frequency fluctuations), and the second scale is defined as a coarse scale (carrying macro-periodic patterns). The target fusion data generated after superposition is the "periodic fusion data," which supplements the coarse-scale periodic modeling with fine-scale details. The calculation process of periodic fusion data is as follows:

[0061] in, The layer consists of two linear layers and a GELU activation function along the time dimension in between, with an input dimension of . The output dimension is .

[0062] When the target feature component is the second feature component (trend feature), the first scale is defined as a coarse scale (carrying a clear macro trend), and the second scale is defined as a fine scale (susceptible to noise interference). The target fusion data generated after superposition is the "trend fusion data," which guides the modeling of fine-scale trends based on coarse-scale trends. The calculation process for obtaining trend fusion data by fusing trend components is as follows:

[0063] in, The layer also consists of two linear layers and a GELU activation function along the time dimension in between, with an input dimension of... The output dimension is .

[0064] Driven by a hybrid mechanism of periodicity and trend, the TDM module can gradually aggregate periodic details from fine to coarse scales, and deeply mine macro trend information with the help of prior knowledge from the coarse scale, ultimately realizing the perception of correlations and time series modeling between multi-scale sequences.

[0065] In the above method, by first converting the corresponding meteorological data in adjacent scales into a unified scale and then superimposing and fusing them, detailed supplementation from high scale to low scale is achieved for periodic components, and macroscopic guidance from low scale to high scale is achieved for trend components. This accurately realizes cross-scale information complementarity of different feature components, effectively improving the completeness and accuracy of feature expression, and providing better feature support for subsequent photovoltaic power prediction.

[0066] S203. Based on the periodic fusion data and trend fusion data corresponding to each second meteorological data, the third meteorological data corresponding to each second meteorological data is obtained.

[0067] In this embodiment of the application, "periodic fusion data" is a set of periodic features supplemented by fine-scale details at the coarse scale, and "trend fusion data" is a set of trend features guided by coarse-scale trends at the fine scale. By integrating the features of the two types of fusion data of the same second meteorological data, the limitations of a single feature dimension are eliminated, and comprehensive feature data (third meteorological data) with both "high-frequency detail periodicity" and "macroscopic stable trend" is generated, providing complete and high-quality feature support for the accurate prediction of the subsequent multi-scale prediction aggregation module (MPA).

[0068] The above method integrates periodic and trend components across dimensions and scales, which not only adds fine-scale details to periodic features and injects coarse-scale macro guidance into trend features, but also achieves deep integration of the two types of core features. This effectively explores the deep dynamic correlation between meteorological and power sequences, solves the problem of insufficient modeling of single-scale features, and significantly improves the accuracy and stability of photovoltaic short-term power forecasting.

[0069] In one embodiment, referring to Figure 4, is a schematic diagram of the second process for generating third meteorological data provided in this application embodiment. As shown in Figure 4, step S203 includes: S401, after each feature processing, performing feature fusion on the periodic fusion data and trend fusion data corresponding to each second meteorological data to obtain the first fusion data corresponding to each second meteorological data.

[0070] In this embodiment, "second meteorological data" refers to meteorological sequences at different time granularities after multi-scale decomposition (e.g., fine-scale 15-minute data, coarse-scale 1-hour data), "periodic fusion data" refers to short-term fluctuation characteristics after bottom-up cross-scale fusion (e.g., hourly periodic details of intraday irradiance), and "trend fusion data" refers to long-term change characteristics after top-down cross-scale fusion (e.g., seasonal sunshine duration trends). Through targeted fusion, the limitations of a single feature dimension are eliminated, generating comprehensive feature data (first fusion data) with both "short-term dynamic cycles" and "long-term stable trends" for each scale of second meteorological data. This lays a complete feature foundation for the accurate prediction of the subsequent multi-scale prediction aggregation module (MPA). The fusion formula is as follows:

[0071] in, It contains two linear layers, with the GELU activation function used in between for information exchange between channels. , These represent periodic and trend-based multi-scale fusion, respectively, which are periodic fusion data and trend fusion data.

[0072] S402, according to the feature processing order, every two adjacent first fusion data are fused to obtain multiple second fusion data.

[0073] In this embodiment, "feature processing order" refers to the iterative rounds of multiple feature processing (such as the processing of the 1st TDF to the Kth TDF, where multiple TDFs are stacked from top to bottom, and the feature data order is also from top to bottom). "Adjacent first fusion data" is the comprehensive feature data generated after two consecutive rounds of feature processing (such as the first fusion data obtained from the K-1th and Kth processing). By fusing the first fusion data of adjacent rounds, the accumulation and enhancement of feature processing information from multiple rounds are realized, so that the final second fusion data retains the effective features of each round and integrates the cross-round correlation information, further improving the depth and completeness of the features, and providing better input for the subsequent multi-scale prediction aggregation module (MPA).

[0074] In one embodiment, step S402 includes: in every two adjacent first fused data, superimposing the first fused data corresponding to each second meteorological data in the first feature processing sequence number with the first fused data corresponding to each second meteorological data in the second feature processing sequence number to obtain the second fused data corresponding to each second meteorological data in the second feature processing sequence number; wherein, the first feature processing sequence number is the previous sequence number of the second feature processing sequence number.

[0075] In this embodiment, the "first feature processing sequence number" is the previous round of the "second feature processing sequence number" (e.g., the K-1th and Kth rounds), and the "first fusion data corresponding to each second meteorological data" is the comprehensive feature of each scale (fine / coarse scale) after a single round of feature processing, integrating periodic fusion data and trend fusion data. By superimposing the first fusion data of the same scale from the previous round with the first fusion data of the same scale in the current round, second fusion data with multi-round feature information is generated for each scale in the current round, realizing the complementarity and enhancement of cross-round features and improving the deep expression capability of features. The calculation formula for the second fusion data is as follows:

[0076] Right now

[0077] in, It is the total number of layers in the TDF module stack. This is the first fused data corresponding to the (k-1)th TDF module. This is the second fusion data corresponding to the kth TDF module.

[0078] The above method first integrates periodic and trend data within a single round, and then integrates adjacent fusion results across rounds, achieving deep aggregation of multi-round and multi-dimensional features in a progressive manner. This not only fully preserves the core meteorological information at each scale, but also strengthens the correlation and integrity of features, effectively improving the modeling ability for complex time series patterns and providing more accurate and reliable feature support for short-term photovoltaic power prediction.

[0079] S403, multiple third meteorological data are obtained based on multiple second fusion data.

[0080] In this embodiment, "multiple second fusion data" refers to multiple sets of comprehensive feature data obtained by fusing adjacent first fusion data across rounds according to the feature processing order. Each set carries periodic and trend correlation information across multiple rounds and scales. By performing final feature purification and dimensional unification on these second fusion data, a high-quality feature data (third meteorological data) with both "multi-round feature enhancement" and "multi-scale information complementarity" is generated for each original second meteorological data. This completely solves the problem of insufficient modeling of complex meteorological-power correlation by single-scale and single-round features, and provides core input for accurate prediction of the subsequent multi-scale prediction aggregation module (MPA).

[0081] The above method first integrates periodic and trend data within a single round, and then integrates adjacent fusion results across rounds, achieving deep aggregation of multi-round and multi-dimensional features in a progressive manner. This not only fully preserves the core meteorological information at each scale, but also strengthens the correlation and integrity of features, effectively improving the modeling ability for complex time series patterns and providing more accurate and reliable feature support for short-term photovoltaic power prediction.

[0082] In one embodiment, step S403 includes: determining a third feature processing sequence number from the feature processing sequence, and determining the second fused data of each second meteorological data corresponding to the third feature processing sequence number as the third meteorological data; wherein the third feature processing sequence number is the last feature processing sequence number in the feature processing sequence.

[0083] In the embodiments of this application, “the third feature processing sequence number is the last feature processing sequence number in the feature processing order” is the final feature output definition step of the Temporal Deconstruction and Fusion (TDF) module in the DA-SolarPower algorithm.

[0084] Here, "feature processing order" refers to the sequence of rounds of feature processing (e.g., round 1, round 2, ..., round K), and "third feature processing sequence number" is the last round (round K) in this sequence. The corresponding second fused data is the final comprehensive feature data generated after multiple rounds of feature processing, cross-scale fusion (periodicity + trend), and cross-round integration, which has fully aggregated the core meteorological information from multiple scales and rounds. The second fused data corresponding to each second meteorological data in this round is directly defined as its corresponding third meteorological data, that is, after... After stacking several TDF modules, multi-scale temporal deconstruction information, i.e., the third meteorological data, was obtained. , Indicates having Dimensional fusion features.

[0085] The third meteorological data is the final solidification of the feature processing results of the TDF module, ensuring that the output feature data has completeness, depth and practicality, and providing standardized input for the accurate prediction of the subsequent multi-scale prediction aggregation module (MPA).

[0086] In the above method, the second fused data corresponding to the last number in the feature processing sequence is directly selected as the third meteorological data. This fully preserves the complete feature information after multiple rounds and cross-scale fusion, avoids information loss caused by additional processing, and ensures that the output feature data has both depth and reliability, providing high-quality and standardized core input for subsequent photovoltaic power prediction.

[0087] S105, based on multiple third-party meteorological data, photovoltaic power prediction is performed to obtain the first prediction result.

[0088] In this embodiment, high-quality multi-scale feature data processed by the Time Series Deconstruction and Fusion (TDF) module is used to achieve accurate prediction of short-term photovoltaic power. The "multiple third-level meteorological data" are comprehensive feature data generated from second-level meteorological data at various scales after multiple rounds of feature processing, cross-scale fusion (periodic detail supplementation and trend guidance), and cross-round information integration. Each data point corresponds to a specific time granularity (fine / coarse scale) and simultaneously carries high-frequency fluctuation details, macro-trend patterns, and multi-round reinforcement information of the meteorological sequence. During the prediction phase, the model uses the MPA module to aggregate this deconstructed multi-scale contextual information. A complementary prediction strategy is used to generate the final sequence prediction result:

[0089] in, This represents the model's final prediction result. Through the above design, DA-SolarPower can effectively extract core features from decoupled multi-scale observation data and achieve accurate prediction of future power sequences by leveraging collaborative modeling of multi-scale information.

[0090] The above method first acquires weather forecast sequence data within a preset time period as basic input, then performs multi-scale feature decomposition on the meteorological data to obtain subdivided meteorological data at different scales. Subsequently, for each scale of meteorological data, multiple feature processing steps are performed to extract the first feature component representing the periodic variation law of meteorology and the second feature component representing the long-term evolution trend of meteorology. Next, the two types of feature components at the same scale are fused to generate the fused meteorological feature (third meteorological data) corresponding to that scale. Finally, the fused meteorological features of all scales are combined to complete the photovoltaic power prediction and output the final prediction result. This method separates meteorological features at different time scales through multi-scale feature decomposition to avoid mutual interference between different scale laws; it extracts features with clear physical meaning through periodic-trend component separation to strengthen the correlation between meteorological factors and photovoltaic power; and it achieves information complementarity of features in different dimensions through multi-component feature fusion, taking into account both the short-term fluctuation characteristics of photovoltaic power and supporting the fitting of long-term variation trends. It progresses step by step from the three key links of feature extraction, characterization, and integration, reducing invalid feature interference and prediction bias, and ultimately achieving a significant improvement in the accuracy of photovoltaic power prediction.

[0091] In one embodiment, referring to Figure 5, which is a flowchart of obtaining the prediction result provided by the embodiment of this application, as shown in Figure 5, step S105 includes: S501, predicting each third meteorological data to obtain the first sub-prediction result corresponding to each third meteorological data.

[0092] In this embodiment, "each third meteorological data" is a high-quality comprehensive feature data generated after multiple rounds of processing and cross-scale and cross-round fusion of the second meteorological data at various scales by the Time Series Deconstruction and Fusion (TDF) module. Each data corresponds to a specific time granularity (fine / coarse scale) and carries complete periodic details and macro-trend information at that scale. By configuring a dedicated predictor for each third meteorological data, the independent mapping of features at each scale to photovoltaic power is realized. The generated first sub-prediction result will reflect the prediction advantages of the corresponding scale (fine scale captures short-term mutations, coarse scale grasps the overall trend), laying the foundation for subsequent adaptive weighted aggregation. The prediction formula for its first sub-prediction result is as follows:

[0093] in, Indicates the first Future prediction results for the scale sequence (first sub-prediction result). Indicates the first The scale-corresponding predictor uses a single linear layer to directly map future power sequences from deconstructed multi-scale information.

[0094] S502, the first prediction result is obtained based on multiple first sub-prediction results.

[0095] In the embodiments of this application, "multiple first sub-prediction results" are independent power prediction results generated by a dedicated single linear layer predictor from third meteorological data at each scale. Each result corresponds to the prediction advantages of a specific scale (fine scale captures short-term power mutations, coarse scale grasps the overall trend). Through an adaptive weighted fusion strategy, these heterogeneous sub-prediction results are integrated into a unified prediction consensus (first prediction result), giving full play to the complementarity of multi-scale prediction, solving the limitations of single-scale prediction in complex time series modeling, and finally outputting high-precision photovoltaic short-term power prediction results for the next 24-72 hours.

[0096] In the above method, the first sub-prediction result is obtained by independently predicting the third meteorological data at each scale, and then the multi-sub-prediction results are integrated to obtain the final first prediction result. This not only gives full play to the prediction advantages of the characteristics of each scale (capturing short-term changes at the fine scale and grasping macro trends at the coarse scale), but also realizes the complementary fusion of multi-scale prediction information, effectively improving the accuracy and stability of photovoltaic short-term power prediction and making up for the limitations of single-scale prediction.

[0097] In one embodiment, referring to Figure 6, which is a flowchart of obtaining prediction results provided by an embodiment of this application, as shown in Figure 6, step S105 includes: S601, assigning each of the multiple first prediction results a corresponding first weight.

[0098] In the embodiments of this application, "multiple first prediction results" refer to independent photovoltaic power prediction results generated by a dedicated single linear layer predictor from third meteorological data at each scale. Each result corresponds to the prediction advantage at a specific scale (fine scale captures short-term mutations, coarse scale grasps macro trends). "First weight" is a weight coefficient dynamically allocated based on the historical prediction performance of each first prediction result. Its core purpose is to highlight the contribution of prediction results at advantageous scales, suppress the interference of low-reliability prediction results, lay the foundation for subsequent weighted aggregation to generate high-precision final prediction results, and solve the problem that traditional fixed weights cannot adapt to complex meteorological scenarios.

[0099] For example, historical data corresponding to each first prediction result can be collected (including predicted values ​​and actual photovoltaic power values ​​under different weather types in the past 3 months), covering typical scenarios such as sunny, cloudy, partly cloudy, and rainy days, to ensure data integrity. Then, the root mean square error is used as the core indicator to calculate the historical prediction error of each first prediction result. Basic weights are allocated based on the reciprocal of the error to ensure that the sum of all weights is 1. Finally, the weights are optimized in combination with real-time weather type. For example, the weight of coarse-scale prediction results is increased by 5%~10% when it is sunny (to enhance trend stability), and the weight of fine-scale prediction results is increased by 5%~10% when it is cloudy (to highlight the accuracy of details), adapting the prediction advantages under different scenarios.

[0100] S602, multiply each first prediction result with its corresponding first weight to obtain the corresponding second sub-prediction result.

[0101] In this embodiment, the "first prediction result" refers to the independent photovoltaic power prediction result generated by a dedicated single linear layer predictor from third meteorological data at each scale (such as prediction values ​​that capture short-term mutations at a fine scale and grasp macro trends at a coarse scale). The "first weight" is a weight coefficient dynamically allocated based on the historical performance (error level) of each first prediction result and the real-time weather scenario (the smaller the error and the stronger the scenario adaptability, the higher the weight). Through element-wise multiplication, the prediction results at each scale are bound to their own weights, thereby strengthening the superior prediction results and weakening the low-reliability prediction results, laying the foundation for the subsequent aggregation to generate a high-precision final prediction result. The calculation formula for the second sub-prediction result is as follows:

[0102] in, For each first sub-prediction result The corresponding weight is the first weight.

[0103] S603, multiple second sub-prediction results are superimposed to obtain the first prediction result.

[0104] In this embodiment, "multiple second sub-prediction results" are weighted sub-prediction sequences obtained by multiplying the first prediction results at each scale with their corresponding first weights (dynamically allocated based on historical performance and real-time scenarios) element by element. Each result incorporates the prediction advantage weight of the corresponding scale (scales with smaller errors and stronger scenario adaptability have higher weights). Through superposition processing, the weighted sub-prediction information at each scale is summarized and merged to generate a unified photovoltaic power prediction result (first prediction result) that takes into account both "short-term detail accuracy" and "long-term trend stability." This ultimately achieves a comprehensive integration of the advantages of multi-scale prediction, solving the limitations of single-scale prediction. The calculation formula for the first prediction result is as follows:

[0105] in, This is the system's final prediction output, i.e., the first prediction result.

[0106] In the above method, the first weight is dynamically allocated according to the reliability of each first sub-prediction result, then the advantageous prediction is strengthened and the inefficient prediction is weakened by weighted multiplication, and finally the multi-dimensional weighted sub-prediction results are superimposed and integrated. This fully integrates the complementary advantages of predictions at different scales, effectively improves the accuracy and stability of photovoltaic short-term power prediction, and makes up for the limitations of single-scale prediction.

[0107] Referring to Figure 7, which is a schematic diagram of the photovoltaic power prediction system provided in the embodiment of this application, as shown in Figure 7, the core consists of three parts connected in sequence: "multi-scale construction module → stacked temporal deconstruction and fusion module (TDF) → multi-scale prediction aggregation module (MPA)". The specific process steps are as follows: 1) First, the multi-scale construction module constructs a multi-scale sequence set (covering low-scale sequences carrying high-frequency fluctuations and high-scale sequences highlighting macro trends) by average pooling downsampling of the input numerical weather forecast (NWP) sequence, and then maps it into deep features through a shared embedding layer to complete the multi-scale feature decomposition.

[0108] 2) Next, the stacked TDF module performs hierarchical processing on the output features of the previous layer. First, it decouples the periodic and trend features through the Autoformer sequence decomposition module, and then achieves cross-scale fusion in the bottom-up (periodic component supplements details) and top-down (trend component guides modeling) ways respectively to generate multi-scale fused features; 3) Finally, the MPA module independently predicts the fused features of each scale through multiple parallel single linear layer predictors, and then gathers the prediction insights of each scale through an adaptive weighted fusion strategy to output the final photovoltaic short-term power prediction result.

[0109] Referring to Figure 8, which is a schematic diagram of the periodic component fusion structure provided in the embodiment of this application, as shown in Figure 7, the core presents a "bottom-up" detail supplementation fusion logic: First, a Fusion layer containing two linear layers and a time-dimensional GELU activation function is constructed using a residual connection method. The input is the periodic component sequence at each scale (from fine to coarse). Through intra-layer operations, the periodic detail information at the fine scale is gradually transferred and merged to the coarse scale, supplementing the coarse-scale periodic modeling with high-frequency fluctuation details (such as short-term irradiance fluctuations and periodic changes caused by cloud cover). Finally, the output is a fusion feature that integrates cross-scale periodic information at each scale, solving the problem of missing details in single-scale periodic modeling and providing more accurate periodic feature support for subsequent predictions.

[0110] Referring to Figure 9, which is a schematic diagram of the trend component fusion structure provided in the embodiment of this application, as shown in Figure 9, the core presents a "top-down" trend-guided fusion logic: First, a Fusion layer containing two linear layers and a time-dimensional GELU activation function is constructed using a residual connection method. The input is the trend component sequence at each scale (from coarse to fine). Through intra-layer operations, the clear macro trend information at the coarse scale is gradually transmitted and guided to the fine scale, filtering noise interference (such as the impact of short-term irrelevant fluctuations on long-term trend judgment) for fine-scale trend modeling. Finally, the fusion feature integrating cross-scale trend information at each scale is output, solving the problem of single-scale trend modeling being easily interfered with and lacking stability, and providing more reliable trend feature support for subsequent predictions.

[0111] Referring to Figure 10, which is a schematic diagram of the multi-scale prediction aggregation provided in the embodiment of this application, the core presents the prediction logic of "multi-scale independent prediction + adaptive weighted aggregation": On the left, a single linear layer predictor is set up in parallel, corresponding one-to-one with the third meteorological data of each scale. Each predictor independently maps the first sub-prediction result specific to that scale from the deep fusion features of the corresponding scale (the fine-scale sub-prediction result focuses on capturing short-term power mutations, while the coarse-scale sub-prediction result focuses on grasping macro trends); in the middle, the adaptive weight allocation module dynamically calculates the weight coefficient based on the historical error of each first sub-prediction result (the smaller the error, the higher the weight); on the right, each first sub-prediction result is multiplied with its corresponding weight and then superimposed step by step, finally outputting a unified first prediction result of photovoltaic short-term power that integrates the complementary prediction advantages of multiple scales, achieving a dual improvement in prediction accuracy and stability.

[0112] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0113] Corresponding to the photovoltaic short-term power prediction method in the above embodiments, Figure 11 is a structural block diagram of the photovoltaic power prediction device provided in the embodiments of this application. For ease of explanation, only the parts related to the embodiments of this application are shown.

[0114] Referring to Figure 11, the device includes: a meteorological data acquisition module 111, used to acquire first meteorological data; wherein the first meteorological data is weather forecast sequence data within a preset time period; a multi-scale decomposition module 112, used to perform multi-scale feature decomposition on the first meteorological data to obtain multiple second meteorological data; wherein each second meteorological data has a different scale; a feature processing module 113, used to perform multiple feature processing on each second meteorological data to obtain a first feature component and a second feature component corresponding to each second meteorological data after each feature processing; wherein the first feature component is used to represent the meteorological periodicity feature in the second meteorological data; and the second feature component is used to represent the meteorological trend feature in the second meteorological data; a fusion processing module 114, used to perform feature fusion based on multiple first feature components and second feature components to obtain third meteorological data corresponding to each second meteorological data; and a power prediction module 115, used to perform photovoltaic power prediction based on multiple third meteorological data to obtain a first prediction result.

[0115] Optionally, the fusion processing module 114 is further configured to: after each feature processing, fuse the first feature components corresponding to every two adjacent scales of the second meteorological data to obtain periodic fused data corresponding to each second meteorological data; fuse the second feature components corresponding to every two adjacent scales of the second meteorological data to obtain trend fused data corresponding to each second meteorological data; and obtain the third meteorological data corresponding to each second meteorological data based on the periodic fused data and trend fused data corresponding to each second meteorological data.

[0116] Optionally, the fusion processing module 114 is further configured to: convert the second meteorological data corresponding to the first scale in every two adjacent scales of the second meteorological data into fourth meteorological data; wherein the scale of the fourth meteorological data is the same as the scale of the second meteorological data at the second scale; superimpose the second meteorological data at the second scale and the fourth meteorological data to obtain the target fused data corresponding to the second meteorological data at the second scale; wherein, when the target feature component is the first feature component, the target fused data is recorded as periodic fused data, and the first scale is a fine scale and the second scale is a coarse scale; when the target feature component is the second feature component, the target fused data is recorded as trend fused data, and the first scale is a coarse scale and the second scale is a fine scale.

[0117] Optionally, the fusion processing module 114 is further configured to: after each feature processing, perform feature fusion on the periodic fusion data and trend fusion data corresponding to each second meteorological data to obtain the first fusion data corresponding to each second meteorological data; fuse every two adjacent first fusion data according to the feature processing order to obtain multiple second fusion data; and obtain multiple third meteorological data based on the multiple second fusion data.

[0118] Optionally, the fusion processing module 114 is further configured to: in every two adjacent first fusion data, superimpose the first fusion data corresponding to each second meteorological data in the first feature processing sequence number with the first fusion data corresponding to each second meteorological data in the second feature processing sequence number to obtain the second fusion data corresponding to each second meteorological data in the second feature processing sequence number; wherein, the first feature processing sequence number is the previous sequence number of the second feature processing sequence number.

[0119] Optionally, the fusion processing module 114 is further configured to: determine the third feature processing sequence number from the feature processing sequence, and determine the second fused data of each second meteorological data corresponding to the third feature processing sequence number as the third meteorological data; wherein the third feature processing sequence number is the last feature processing sequence number in the feature processing sequence.

[0120] Optionally, the power prediction module 115 is further configured to: predict each third meteorological data to obtain a first sub-prediction result corresponding to each third meteorological data; and obtain a first prediction result based on multiple first sub-prediction results. Optionally, the power prediction module 115 is further configured to: assign a first weight to each of the multiple first prediction results; multiply each first prediction result with its corresponding first weight to obtain a second sub-prediction result; and superimpose the multiple second sub-prediction results to obtain a first prediction result.

[0121] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0122] In addition, the photovoltaic power prediction device shown in Figure 11 can be a software unit, a hardware unit, or a combination of software and hardware built into existing terminal equipment, or it can be integrated into the terminal equipment as an independent component, or it can exist as an independent terminal equipment.

[0123] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0124] Figure 12 is a schematic diagram of the structure of the terminal device provided in an embodiment of this application. As shown in Figure 12, the terminal device 12 of this embodiment includes: at least one processor 120 (only one is shown in Figure 12), a memory 121, and a computer program 122 stored in the memory 121 and executable on at least one processor 120. When the processor 120 executes the computer program 122, it implements the steps in any of the above embodiments of the photovoltaic short-term power prediction method.

[0125] The terminal device can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. This terminal device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that Figure 12 is merely an example of terminal device 12 and does not constitute a limitation on terminal device 12. It may include more or fewer components than shown, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0126] The processor 120 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0127] In some embodiments, memory 121 may be an internal storage unit of terminal device 12, such as a hard disk or memory of terminal device 12. In other embodiments, memory 121 may be an external storage device of terminal device 12, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on terminal device 12. Furthermore, memory 121 may include both internal and external storage units of terminal device 12. Memory 121 is used to store operating system, application programs, boot loader, data, and other programs, such as program code of computer programs. Memory 121 may also be used to temporarily store data that has been output or will be output.

[0128] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the steps in the above-described method embodiments.

[0129] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0130] 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, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium can include at least: any entity or device capable of carrying computer program code to a device / terminal equipment, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0131] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0132] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0133] In the embodiments provided in this application, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0134] The units described as separate components may or may not be physically separate. 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 the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0135] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for short-term photovoltaic power prediction, characterized in that, The method includes: acquiring first meteorological data; wherein the first meteorological data is weather forecast sequence data within a preset time period; performing multi-scale feature decomposition on the first meteorological data to obtain multiple second meteorological data; wherein each second meteorological data has a different scale; performing multiple feature processing on each second meteorological data to obtain a first feature component and a second feature component corresponding to each second meteorological data after each feature processing; wherein the first feature component is used to represent the meteorological periodic characteristics in the second meteorological data; the second feature component is used to represent the meteorological trend characteristics in the second meteorological data; performing feature fusion based on the multiple first feature components and second feature components to obtain third meteorological data corresponding to each second meteorological data; and performing photovoltaic power prediction based on the multiple third meteorological data to obtain a first prediction result.

2. The photovoltaic short-term power prediction method as described in claim 1, characterized in that, The step of fusing features based on multiple first and second feature components to obtain third meteorological data corresponding to each second meteorological data includes: after each feature processing, fusing the first feature components corresponding to every two adjacent scales of the second meteorological data to obtain periodic fused data corresponding to each second meteorological data; fusing the second feature components corresponding to every two adjacent scales of the second meteorological data to obtain trend fused data corresponding to each second meteorological data; and obtaining third meteorological data corresponding to each second meteorological data based on the periodic fused data and trend fused data corresponding to each second meteorological data.

3. The photovoltaic short-term power prediction method as described in claim 2, characterized in that, The step of fusing the target feature components corresponding to every two adjacent scales of the second meteorological data to obtain target fused data corresponding to each second meteorological data includes: converting the second meteorological data corresponding to the first scale in every two adjacent scales of the second meteorological data into fourth meteorological data; wherein the scale of the fourth meteorological data is the same as the scale of the second meteorological data at the second scale; superimposing the second meteorological data at the second scale and the fourth meteorological data to obtain target fused data corresponding to the second meteorological data at the second scale; wherein, when the target feature component is the first feature component, the target fused data is recorded as the periodic fused data, and the first scale is a fine scale and the second scale is a coarse scale; when the target feature component is the second feature component, the target fused data is recorded as the trend fused data, and the first scale is a coarse scale and the second scale is a fine scale.

4. The photovoltaic short-term power prediction method as described in claim 3, characterized in that, The step of obtaining the third meteorological data corresponding to each second meteorological data based on the periodic fusion data and trend fusion data corresponding to each second meteorological data includes: after each feature processing, performing feature fusion on the periodic fusion data and trend fusion data corresponding to each second meteorological data to obtain the first fusion data corresponding to each second meteorological data; fusing every two adjacent first fusion data according to the feature processing order to obtain multiple second fusion data; and obtaining multiple third meteorological data based on the multiple second fusion data.

5. The photovoltaic short-term power prediction method as described in claim 4, characterized in that, The step of fusing every two adjacent first fused data according to the feature processing order to obtain multiple second fused data includes: in every two adjacent first fused data, superimposing the first fused data corresponding to each second meteorological data in the first feature processing sequence number with the first fused data corresponding to each second meteorological data in the second feature processing sequence number to obtain the second fused data corresponding to each second meteorological data in the second feature processing sequence number; wherein, the first feature processing sequence number is the previous sequence number of the second feature processing sequence number.

6. The photovoltaic short-term power prediction method as described in claim 5, characterized in that, The step of obtaining multiple third meteorological data based on multiple second fusion data includes: determining a third feature processing sequence number from the feature processing sequence, and determining the second fusion data of each second meteorological data corresponding to the third feature processing sequence number as the third meteorological data; wherein the third feature processing sequence number is the last feature processing sequence number in the feature processing sequence.

7. The photovoltaic short-term power prediction method as described in claim 5, characterized in that, The step of predicting photovoltaic power based on multiple third meteorological data to obtain a first prediction result includes: predicting each of the third meteorological data to obtain a first sub-prediction result corresponding to each of the third meteorological data; and obtaining the first prediction result based on multiple first sub-prediction results.

8. The photovoltaic short-term power prediction method as described in claim 7, characterized in that, The step of obtaining the first prediction result based on multiple first sub-prediction results includes: assigning each of the multiple first prediction results a corresponding first weight; performing a multiplication operation on each first prediction result and its corresponding first weight to obtain a corresponding second sub-prediction result; and superimposing the multiple second sub-prediction results to obtain the first prediction result.

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

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 8.

Images