CN-122026333-A - Photovoltaic power generation power prediction method and device, electronic equipment and storage medium

CN122026333ACN 122026333 ACN122026333 ACN 122026333ACN-122026333-A

Abstract

The application relates to the technical field of photovoltaic power generation, in particular to a photovoltaic power generation power prediction method, a device, electronic equipment and a storage medium, wherein multi-source heterogeneous data such as historical power generation, NWP, satellite remote sensing, cloud layer height and the like corresponding to a prediction time point are obtained and preprocessed, the preprocessed data are input into four deep learning sub-models, namely a historical power analysis model, a numerical prediction analysis model, a cloud area analysis model and a cloud height analysis model, corresponding future power prediction sequences are respectively output, and the sequences are fused to obtain a photovoltaic power generation power prediction result. According to the photovoltaic power generation power prediction method, local cloud cluster change can be more accurately captured through comprehensive multi-source data and multi-model analysis fusion, deep features of multi-model data are excavated, physical process influence is considered, and the accuracy and reliability of photovoltaic power generation power prediction are effectively improved.

Inventors

YANG SHU
Xia Ruitao
XU CHEN

Assignees

深圳织算科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260409

Claims (10)

1. A method for predicting photovoltaic power generation power, the method comprising: The method comprises the steps of obtaining multi-source heterogeneous data corresponding to a prediction time point, and preprocessing the multi-source heterogeneous data, wherein the multi-source heterogeneous data comprises historical power generation data, NWP data, satellite remote sensing data and cloud layer height data; The method comprises the steps of inputting preprocessed multi-source heterogeneous data into a deep learning sub-model, and outputting future power prediction sequences of all sub-models, wherein the deep learning sub-model comprises a historical power analysis model, a numerical forecasting analysis model, a cloud area analysis model and a cloud height analysis model, the historical power analysis model is used for predicting the future power prediction sequences corresponding to the historical power generation data, the numerical forecasting analysis model is used for predicting the future power prediction sequences corresponding to the NWP data, the cloud area analysis model is used for predicting the future power prediction sequences corresponding to the fusion of the historical power generation data and the satellite remote sensing data, and the cloud height analysis model is used for predicting the future power prediction sequences corresponding to the fusion of the satellite remote sensing data and the cloud height data; and fusing the future power prediction sequences to obtain a photovoltaic power generation power prediction result.
2. The photovoltaic power generation power prediction method according to claim 1 is characterized in that the Fusion of the future power prediction sequences to obtain a photovoltaic power generation power prediction result comprises the steps of carrying out self-adaptive Fusion on the future power prediction sequences output by the historical power analysis model, the numerical prediction analysis model, the cloud area analysis model and the cloud altitude analysis model through a Fusion layer, wherein the Fusion layer adopts a parameter matrix with the shape of [4,16], the number 4 represents the number of sub-models of a deep learning sub-model, the number 16 represents the dimension of a time sequence prediction value, the Fusion layer automatically adjusts weight distribution after training, and realizes self-adaptive precision contribution distribution according to data driving to fuse the future power prediction sequences output by the deep learning sub-model.
3. The photovoltaic power generation power prediction method according to claim 2, characterized in that the method further comprises: freezing the Fusion layer, using all parameters of 1 and being non-learnable Instead of the original fusion layer, training is carried out on the historical power analysis model, the numerical forecast analysis model, the cloud area analysis model and the cloud height analysis model only; and freezing the historical power analysis model, the numerical forecast analysis model, the cloud area analysis model and the cloud height analysis model, replacing the historical power analysis model, the numerical forecast analysis model, the cloud area analysis model and the cloud height analysis model with trainable Fusion layers, and training weight distribution parameters of the Fusion layers.
4. The method of claim 1, further comprising constructing the historical power analysis model based on iTransformer framework, constructing the numerical forecast analysis model based on FA-MLP structure, constructing the cloud area analysis model based on U-Net, TCN structure and cross Attention, and constructing the cloud height analysis model based on Prior_Attention and TCN structure.
5. The photovoltaic power generation power prediction method according to claim 1, characterized in that the method further comprises: Selecting the maximum value in the pre-processed tbb _13 channel data as a ground temperature reference in the lattice point area; calculating the difference between each pixel point value in the grid point area and the ground temperature reference ; According to the decreasing factor, by the formula Calculating pixel coordinates Cloud height value corresponding to grid point, wherein Representing pixel coordinates The cloud height value corresponding to the grid point, Is a decreasing factor.
6. The method of claim 1, wherein the preprocessing the multi-source heterogeneous data comprises: Taking the acquired instant power generation value at the t moment as the historical power generation data, and normalizing the historical power generation data through the following formula: ; Wherein, the Represents the normalized historical generated power data, Represents the instantaneous power generation value acquired at the time t, The starting capacity of the inverter at the moment t is represented; Normalizing the satellite remote sensing data according to the channel type by the following formula: ; Wherein, the Representing the normalized satellite remote sensing data, Representing c-channel pixel coordinates Is used for the gray-scale value of (c), Represents the effective upper limit value of the c-channel gray data, Representing the effective lower limit value of the c-channel gray data; The NWP data is normalized by the following formula: ; Wherein, the Represents the NWP data after normalization, Values representing the corresponding weather elements when the weather macrocode generates NWP data, Represents a reasonable maximum value of the corresponding weather element in the NWP data, A reasonable minimum value of the corresponding weather element in the NWP data is indicated.
7. The method of claim 6, wherein the preprocessing the multi-source heterogeneous data further comprises: Thinning the normalized historical power data from a time sequence interval of 1 step length/15 minutes to 1 step length/5 minutes through interpolation operation; When the time sequence interval of the satellite remote sensing data is 15 minutes, thinning the normalized satellite remote sensing data from the time sequence interval of 1 step length/15 minutes to 1 step length/5 minutes through interpolation operation; The normalized NWP data is thinned from a1 step/1 hour timing interval to 1 step/5 minutes by an interpolation operation.
8. A photovoltaic power generation power prediction apparatus, the apparatus comprising: The data processing module is used for acquiring multi-source heterogeneous data corresponding to a predicted time point and preprocessing the multi-source heterogeneous data, wherein the multi-source heterogeneous data comprises historical power generation data, NWP data, satellite remote sensing data and cloud layer height data; The system comprises a single prediction module, a deep learning sub-model, a cloud area analysis model and a cloud area analysis model, wherein the single prediction module is used for inputting preprocessed multi-source heterogeneous data into the deep learning sub-model and outputting future power prediction sequences of all sub-models, the deep learning sub-model comprises a historical power analysis model, a numerical prediction analysis model, a cloud area analysis model and a cloud area analysis model, the historical power analysis model is used for predicting the future power prediction sequences corresponding to the historical power generation data, the numerical prediction analysis model is used for predicting the future power prediction sequences corresponding to the NWP data, the cloud area analysis model is used for predicting the future power prediction sequences corresponding to the integration of the historical power generation data and the satellite remote sensing data, and the cloud area analysis model is used for predicting the future power prediction sequences corresponding to the integration of the satellite remote sensing data and the cloud area height data; And the fusion prediction module is used for fusing the future power prediction sequences to obtain a photovoltaic power generation power prediction result.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the photovoltaic power generation power prediction method of any of claims 1 to 7 when the computer program is executed.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the photovoltaic power generation power prediction method of any one of claims 1 to 7.

Description

Photovoltaic power generation power prediction method and device, electronic equipment and storage medium Technical Field The present application relates to the field of photovoltaic power generation technologies, and in particular, to a photovoltaic power generation power prediction method, a device, an electronic apparatus, and a storage medium. Background In the related technical fields of energy prediction, especially photovoltaic power prediction, accurate prediction plays a key role in ensuring stable operation and efficient allocation of an energy system. The traditional photovoltaic power prediction method depends on numerical weather forecast, has low space-time resolution, is not fine in space-time perception, and is difficult to accurately capture the phenomenon of rapid movement of local cloud clusters causing power mutation. And the current deep learning model is used for fusing multi-mode data and is stopped at a simple splicing level, and deep feature collaborative extraction is lacked. In the energy prediction, the related data types are various, such as meteorological data, satellite cloud image data, historical power data and the like, and the data in different modes contains rich information. However, most of the existing deep learning models simply splice the data, but do not fully mine deep association and features between different data, and cannot realize effective feature collaborative extraction, so that the improvement of prediction accuracy is limited. Meanwhile, a pure data-driven 'black box' model is not used for effectively constructing a physical process of 'cloud cover-irradiance reduction-power output reduction', and has poor interpretability, and a single-time convolution network (Temporal Convolutional Networks, TCN) or Long-Short-Term Memory neural network (LSTM) model is difficult to capture Long-Term time sequence dependence of power, cloud picture space characteristics and weather condition future influence simultaneously. In addition, although prediction research based on satellite cloud images exists, most schemes of simply inputting a single-channel cloud image into a convolutional neural network (Convolutional Neural Network, CNN) and combining time sequence processing, undermining time sequence and physical information (such as cloud height), and respectively designing deep learning submodels aiming at four dimensions of power time sequence, future weather, two-dimensional cloud area and three-dimensional cloud height and effectively fusing the deep learning submodels have not been reported. Disclosure of Invention In order to overcome the defects of the prior art, the invention provides a photovoltaic power generation power prediction method, a device, electronic equipment and a storage medium, which are used for realizing collaborative modeling of historical rules, future weather, cloud layer plane distribution and a vertical structure by means of depth feature extraction and fusion, so that prediction precision and interpretability are improved. A first aspect of the present application provides a photovoltaic power generation power prediction method, the method comprising: The method comprises the steps of obtaining multi-source heterogeneous data corresponding to a prediction time point, and preprocessing the multi-source heterogeneous data, wherein the multi-source heterogeneous data comprises historical power generation data, NWP data, satellite remote sensing data and cloud layer height data; The method comprises the steps of inputting preprocessed multi-source heterogeneous data into a deep learning sub-model, and outputting future power prediction sequences of all sub-models, wherein the deep learning sub-model comprises a historical power analysis model, a numerical forecasting analysis model, a cloud area analysis model and a cloud height analysis model, the historical power analysis model is used for predicting the future power prediction sequences corresponding to the historical power generation data, the numerical forecasting analysis model is used for predicting the future power prediction sequences corresponding to the NWP data, the cloud area analysis model is used for predicting the future power prediction sequences corresponding to the fusion of the historical power generation data and the satellite remote sensing data, and the cloud height analysis model is used for predicting the future power prediction sequences corresponding to the fusion of the satellite remote sensing data and the cloud height data; and fusing the future power prediction sequences to obtain a photovoltaic power generation power prediction result. In an optional implementation manner, the Fusion of the future power prediction sequences to obtain the photovoltaic power generation power prediction result comprises the steps of adaptively fusing the future power prediction sequences output by the historical power analysis model, the numerical prediction analysis model, the cl