Ultra-short-term photovoltaic power prediction method, system, device and medium based on multi-element data fusion

Abstract

The present application aims at the problem that the photovoltaic power prediction accuracy is not high, and proposes a kind of super short-term photovoltaic power prediction method, system, equipment and medium based on multi-element data fusion, method includes the underlying surface data, ground meteorological monitoring data and satellite monitoring data are input WRF-Solar model, predict solar irradiance;Solar irradiance, historical weather data and photovoltaic power station historical data are input LSTM power prediction model, predict super short photovoltaic power.Through the fusion of underlying surface data, ground real-time meteorological monitoring data, satellite monitoring data, photovoltaic module operating state and other multi-source heterogeneous data, combined with the collaborative optimization of machine learning model and physical model, the accuracy of super short numerical weather prediction can be effectively improved, the short-term prediction error under extreme weather and complex scene is reduced, thereby the accuracy of super short photovoltaic prediction is improved, the adaptability of power grid to photovoltaic fluctuation is enhanced, and the energy dispatching efficiency and stability are optimized.

Description

Technical Field

[0001] This invention relates to photovoltaic power prediction technology, specifically to an ultra-short-term photovoltaic power prediction method, system, device, and medium based on multi-source data fusion. Background Technology

[0002] The WRF-Solar model is the first mesoscale professional meteorological model developed based on the WRF (Weather Research and Forecasting) modeling framework, specifically designed for solar energy applications.

[0003] The highly volatile output of new energy sources such as wind and solar power, and their mismatch with the spatiotemporal distribution of load, significantly exacerbates the pressure on power grid frequency regulation and peak shaving. Against this backdrop, solar power forecasting has become the technological cornerstone for the safe, economical, and low-carbon operation of new power systems. Its core value lies in balancing the randomness of new energy sources with the controllability of the power grid, promoting the transformation of the energy structure from "source follows load" to "source interacts with load." With the continuous increase in the proportion of installed solar power capacity, accurate solar power forecasting technology is crucial.

[0004] Current short-term photovoltaic (PV) forecasting technologies primarily rely on machine learning models (such as Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BLSTM), and Temporal Convolutional Networks (TCN)) combined with meteorological data correction (numerical weather prediction (NWP), convolutional neural networks (CNN), grey relational analysis, etc.), multi-timescale analysis, and data preprocessing techniques to predict PV power output. However, the meteorological data correction stage suffers from significant uncertainty, numerical weather prediction accuracy is insufficient, and localized meteorological abrupt changes (such as short-duration severe convection) are difficult to capture. Under extreme weather conditions, the absolute error increases significantly, thus affecting the accuracy of PV power forecasting. Summary of the Invention

[0005] This invention addresses the shortcomings of existing technologies by providing a method, system, device, and medium for ultra-short-term photovoltaic power prediction based on multi-source data fusion. By integrating heterogeneous data from multiple sources, such as underlying surface data, real-time ground meteorological monitoring data, satellite monitoring data, and photovoltaic module operating status (e.g., temperature, degradation rate), and combining machine learning models with physical models for collaborative optimization, it can effectively improve the accuracy of ultra-short-term numerical weather forecasts, reduce short-term prediction errors under extreme weather conditions (e.g., short-term strong convection, aerosol mutations) and complex scenarios (e.g., cloudy, dust storms), thereby improving the accuracy of ultra-short-term photovoltaic predictions, enhancing the grid's adaptability to photovoltaic fluctuations, and optimizing energy dispatch efficiency and stability.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for predicting ultra-short-term photovoltaic power based on multi-source data fusion includes the following steps:

[0008] The underlying surface data, surface meteorological monitoring data, and satellite monitoring data are input into the WRF-Solar model to predict solar irradiance.

[0009] By inputting solar irradiance, historical weather data, and historical data of photovoltaic power plants into the LSTM power prediction model, ultra-short-term photovoltaic power can be predicted.

[0010] To optimize the above technical solution, the specific measures also include:

[0011] Furthermore, the WRF-Solar model is developed based on the WRF model, and the input of underlying surface data, surface meteorological monitoring data, and satellite monitoring data into the WRF-Solar model specifically includes:

[0012] The coordinate system of the latest underlying surface data is converted to the geographic coordinate system required by the WRF model through a geographic information system. The latest underlying surface data is reclassified according to the WRF classification system, and the classified latest underlying surface data replaces the original underlying surface data in the WRF model.

[0013] The WRF model integrates surface meteorological monitoring data and FNL global reanalysis data based on the four-dimensional variational assimilation method and the Spectral Nudging method.

[0014] Based on the WRFDA system, radiation data monitored by the Himawari-8 satellite were integrated into the WRF model.

[0015] Furthermore, the underlying surface data specifically refers to a global land cover classification dataset, and the method for generating the global land cover classification dataset is as follows:

[0016] Reflectance data from MODISTerra and Aqua satellites were collected, and the reflectance data were classified and post-processed to generate a global land cover classification dataset.

[0017] The classification methods include supervised decision tree classification, maximum likelihood method, support vector machine or spectral angle mapping method. The post-processing includes noise removal, classification result optimization, multi-temporal data fusion and quality control.

[0018] Furthermore, the four-dimensional variational assimilation method adopts an incremental form, and the cost function of the four-dimensional variational assimilation method is specifically as follows:

[0019]

[0020] In the formula, J(δx0) is the cost function of the four-dimensional variational assimilation method, δx0 is the analysis increment, and x0 is the value of x0. b and x0 g Let B0 represent the background field and the initial guess field at the initial time, respectively. The background error covariance matrix at the initial time is B0. The observation field is divided into N sub-time windows within a time window, where n is the index of the sub-time window, and d... n For the increase in new information, H n and M n These are the linearized observation operator and the forward mode, respectively, and R0 represents the observation error covariance matrix at the initial time.

[0021] The expression for the increment is as follows:

[0022]

[0023] In the formula, x0 is the analysis field;

[0024] The expression for the new interest increment is as follows:

[0025]

[0026] In the formula, y n 0 H represents the observation field of the nth sub-time window. n and M n These are the nonlinear observation operator and the forward mode, respectively;

[0027]

[0028] In the formula, H n and M n These are the linearized observation operator and the forward mode, x0, respectively. g For the initial guess, δx0 is the analysis increment, and H n and M n These represent the nonlinear observation operator and the forward mode, respectively, with x0 representing the analysis field.

[0029] Furthermore, the SpectralNudging method specifically comprises:

[0030] The FNL global reanalysis data is converted into a spectrum using Fast Fourier Analysis (FFT). Spectral analysis is then performed, and a wavenumber is set. High-frequency waves exceeding the set wavenumber are filtered out in space, retaining large-scale low-frequency waves. These low-frequency waves are then added to the model's forecast field according to set weights, thereby continuously bringing the simulated state closer to the large-scale driving state. This can be expressed by the following formula:

[0031]

[0032] In the formula, α represents horizontal wind, potential temperature, or geopotential height, and its value is obtained by interpolation from analysis values ​​at adjacent times. t represents time, X(α) is the physical forcing term of the model, and G... α Here, w(η) is the nudging coefficient, and w(η) is the weighting coefficient of nudging in the vertical direction.

[0033] α 0,pq Let α be the analysis field when the initial beams are p and q. pq Let i represent the analysis field for beams p and q, where i represents time, x represents the background field, y represents the observation field, and K represents the observation field. pq K represents the approximation coefficients for different scales; p K q represents the wave vector components in the meridional and zonal directions, respectively; p and q represent the wave numbers in the meridional and zonal directions, respectively; P and Q are the nudging wave numbers in the meridional and zonal directions, whose values ​​are calculated based on grid spacing, grid number, and wavelength, and are used to adjust the large-scale circulation field of the model to make it consistent with the actual forcing field.

[0034] Furthermore, the LSTM power prediction model includes an input layer, an LSTM layer, and an output layer. The LSTM layer is used to process the time dependence in the time series data. The LSTM layer has a two-layer structure, and the hidden layer dimension is used to enhance the LSTM power prediction model's ability to remember and capture time series patterns.

[0035] This invention also proposes an ultra-short-term photovoltaic power prediction system based on multi-source data fusion, comprising:

[0036] The WRF-Solar model is used to predict solar irradiance based on underlying surface data, surface meteorological monitoring data, and satellite monitoring data.

[0037] The LSTM power prediction model is used to predict ultra-short-time photovoltaic power based on solar irradiance, historical weather data, and historical data of photovoltaic power plants.

[0038] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the ultra-short-term photovoltaic power prediction method based on multi-source data fusion as described above.

[0039] The present invention also proposes a computer-readable storage medium storing a computer program that enables a computer to execute the ultra-short-term photovoltaic power prediction method based on multi-source data fusion as described above.

[0040] The beneficial effects of this invention are as follows: Based on the WRF-Solar model, this invention integrates underlying surface data of the research area and various real-time monitoring data (such as ground meteorological monitoring data, satellite monitoring data, etc.), and superimposes machine learning methods on the basis of considering atmospheric cloud microphysical changes. It can capture local meteorological abrupt changes (such as short-term strong convection), effectively improve the accuracy of short-term numerical weather forecasts, and thus improve the accuracy of ultra-short-term photovoltaic forecasts. It has good application prospects and promotion significance. Attached Figure Description

[0041] Figure 1 This is an overall flowchart of the ultra-short-term photovoltaic power prediction method based on multi-source data fusion proposed in this invention. Detailed Implementation

[0042] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0043] Example 1

[0044] This invention proposes an ultra-short-term photovoltaic power prediction method based on multi-source data fusion. The overall process of this method is as follows: Figure 1 As shown, it includes the following steps:

[0045] The underlying surface data, surface meteorological monitoring data, and satellite monitoring data are input into the WRF-Solar model to predict solar irradiance.

[0046] By inputting solar irradiance, historical weather data, and historical data of photovoltaic power plants into the LSTM power prediction model, ultra-short-term photovoltaic power can be predicted.

[0047] The WRF-Solar model is developed based on the WRF model. Specifically, the underlying surface data, surface meteorological monitoring data, and satellite monitoring data are input into the WRF-Solar model as follows:

[0048] Underlying surface data fusion: The latest high-resolution underlying surface data (based on reflectance data from MODISTerra and Aqua satellites, generated through supervised classification (using a global prior knowledge base and field survey data, selecting spectrally representative training samples; subsequently employing supervised decision tree classification as the primary method, combined with maximum likelihood, support vector machine, and spectral angle mapping) and post-processing techniques (including noise removal, classification result optimization, multi-temporal data fusion, and quality control) to create the global land cover classification dataset: MODISMCD12Q1)) was selected. The original high-resolution underlying surface data coordinate system was converted to the geographic coordinate system required by the mesoscale numerical weather prediction model (WRF) using ArcGIS. Reclassification was then performed according to the WRF classification system (USGS-24), replacing the original underlying surface data in the WRF model (currently, the underlying surface data used in the WRF is based on underlying surface distribution from 20 years ago). Updating the underlying surface data improves the accuracy of weather forecasts.

[0049] Ground meteorological monitoring data fusion: Based on the four-dimensional variational data assimilation (4DVar) and Spectral Nudging (SN) methods, real-time meteorological monitoring data (temperature, humidity, air pressure, wind speed, wind direction, etc.) and FNL (Final Operational Global Analysis) global reanalysis data are fused into WRF. While correcting the large-scale bias of the WRF model, the meteorological data at the time of analysis are also corrected, further improving the accuracy of short-term weather forecasts.

[0050] Satellite monitoring data fusion: Based on the WRFDA (Weather Research and Forecasting Model Data Assimilation System), the radiation data of the Himawari-8 satellite is fused into the WRF to correct the meteorological data at the time of analysis and further improve the accuracy of short-term weather forecasts.

[0051] MODIS: Moderate-resolution Imaging Spectroradiometer.

[0052] Terra is a morning orbit satellite in the U.S. Earth Observation System (EOS) program, primarily used for land and atmospheric observation.

[0053] Aqua is an afternoon orbit satellite in the EOS program, focusing on ocean and atmospheric water cycle research.

[0054] ArcGIS: ArcGeographic Information System.

[0055] The four-dimensional variational assimilation method adopts an incremental form, and its cost function is as follows:

[0056]

[0057] In the formula, J(δx0) is the cost function of the four-dimensional variational assimilation method, δx0 is the analysis increment, and x0 is the value of x0. b and x0 g Let B0 represent the background field and the initial guess field at the initial time, respectively. The background error covariance matrix at the initial time is B0. The observation field is divided into N sub-time windows within a time window, where n is the index of the sub-time window, and d... n For the increase in new information, H n and M n These are the linearized observation operator and the forward mode, respectively, and R0 represents the observation error covariance matrix at the initial time.

[0058] The expression for the increment is as follows:

[0059]

[0060] In the formula, x0 is the analysis field;

[0061] The expression for the new interest increment is as follows:

[0062]

[0063] In the formula, y n 0 H represents the observation field of the nth sub-time window. n and M n These are the nonlinear observation operator and the forward mode, respectively;

[0064]

[0065] In the formula, H n and M n These are the linearized observation operator and the forward mode, x0, respectively. g For the initial guess, δx0 is the analysis increment, and H n and M n These represent the nonlinear observation operator and the forward mode, respectively, with x0 representing the analysis field.

[0066] The incremental form of 4DVar linearizes the observation operator and forward mode, which reduces the computational cost and improves the mathematical conditions for solving the cost function.

[0067] The SpectralNudging method is as follows:

[0068] FNL global reanalysis data is transformed into a spectrum using Fast Fourier Analysis (FFT). Spectral analysis is then performed, and a wavenumber is set to filter out high-frequency waves exceeding the set wavenumber, retaining large-scale low-frequency waves. These low-frequency waves are then added to the model's forecast field according to set weights, thereby continuously bringing the simulated state closer to the large-scale driving state. This effectively reduces large-scale errors and avoids excessive over-correction of high-frequency data by the large-scale analysis field. Model variables can freely evolve into small- and medium-scale processes while simultaneously reflecting smaller-scale characteristics. This can be expressed by the following formula:

[0069]

[0070] In the formula, α represents horizontal wind, potential temperature, or geopotential height, and its value is obtained by interpolation from analysis values ​​at adjacent times. t represents time, X(α) is the physical forcing term of the model, and G... α Here, w(η) is the nudging coefficient, and w(η) is the weighting coefficient of nudging in the vertical direction.

[0071] α 0,pq Let α be the analysis field when the initial beams are p and q. pq Let i represent the analysis field for beams p and q, where i represents time, x represents the background field, y represents the observation field, and K represents the observation field. pq K represents the approximation coefficients for different scales; p K q represents the wave vector components in the meridional and zonal directions, respectively; p and q represent the wave numbers in the meridional and zonal directions, respectively; P and Q are the nudging wave numbers in the meridional and zonal directions, whose values ​​are calculated based on grid spacing, grid number, and wavelength, and are used to adjust the large-scale circulation field of the model to make it consistent with the actual forcing field.

[0072] The LSTM power prediction model is a deep learning model used to process time series data. Its core principle is to improve the accuracy of photovoltaic power generation prediction by capturing long-term dependencies in the time series. The LSTM power prediction model consists of an input layer, an LSTM layer, and an output layer. The LSTM layer is used to handle the temporal dependencies in the time series data. The LSTM layer has a two-layer structure, and the hidden layer dimension (hidden_size = 64) enhances the LSTM power prediction model's ability to memorize and capture time series patterns.

[0073] Example 2

[0074] This invention proposes an ultra-short-term photovoltaic power prediction system based on multi-source data fusion, corresponding to the method in Embodiment 1, comprising:

[0075] The WRF-Solar model is used to predict solar irradiance based on underlying surface data, surface meteorological monitoring data, and satellite monitoring data.

[0076] The LSTM power prediction model is used to predict ultra-short-time photovoltaic power based on solar irradiance, historical weather data, and historical data of photovoltaic power plants.

[0077] The implementation methods of each module and its function in the system are completely consistent with the steps of the method in Implementation Example 1, so they will not be repeated here.

[0078] Example 3

[0079] This invention proposes an electronic device, comprising: 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 ultra-short-term photovoltaic power prediction method based on multi-source data fusion as described in Embodiment 1.

[0080] Example 4

[0081] This invention proposes a computer-readable storage medium storing a computer program that enables a computer to execute the ultra-short-term photovoltaic power prediction method based on multi-source data fusion as described in Embodiment 1.

[0082] In the embodiments disclosed in this application, a computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0083] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application 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.

[0084] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for predicting ultra-short-term photovoltaic power based on multi-source data fusion, characterized in that, Includes the following steps: The underlying surface data, surface meteorological monitoring data, and satellite monitoring data are input into the WRF-Solar model to predict solar irradiance. The WRF-Solar model is developed based on the WRF model, and the specific steps of inputting the underlying surface data, surface meteorological monitoring data, and satellite monitoring data into the WRF-Solar model include: The coordinate system of the latest underlying surface data is converted to the geographic coordinate system required by the WRF model through a geographic information system. The latest underlying surface data is reclassified according to the WRF classification system, and the classified latest underlying surface data replaces the original underlying surface data in the WRF model. The surface meteorological monitoring data and FNL global reanalysis data are fused into the WRF model based on the four-dimensional variational assimilation method and the Spectral Nudging method; Based on the WRFDA system, radiation data monitored by the Himawari-8 satellite were integrated into the WRF model; The four-dimensional variational assimilation method adopts an incremental form, and the cost function of the four-dimensional variational assimilation method is specifically as follows: In the formula, J (δ) x 0) is the cost function of the four-dimensional variational assimilation method, δ x 0 represents the analysis increment. x 0 b and x 0 g Let B0 represent the background field and the initial guess field at the initial time, respectively. Let B0 represent the background error covariance matrix at the initial time. The observation field is divided into N sub-time windows within a time window. n d is the sequence number of the sub-time window. n For the increase in new information, H n and M n These are the linearized observation operator and the forward mode, respectively, and R0 represents the observation error covariance matrix at the initial time. The expression for the increment is as follows: In the formula, x 0 represents the analysis field; The expression for the new interest increment is as follows: In the formula, y n 0 This represents the observation field of the nth sub-time window. H n and M n These are the nonlinear observation operator and the forward mode, respectively; In the formula, H n and M n These are the linearized observation operator and the forward mode, respectively. x 0 g For the initial guessing round, δ x 0 represents the analysis increment. H n and M n These are the nonlinear observation operator and the forward mode, respectively. x 0 represents the analysis field; The Spectral Nudging method is specifically as follows: The FNL global reanalysis data is converted into a spectrum using Fast Fourier Analysis (FFT). Spectral analysis is then performed, and a wavenumber is set. High-frequency waves exceeding the set wavenumber are filtered out in space, retaining large-scale low-frequency waves. These low-frequency waves are then added to the model's forecast field according to set weights, thereby continuously bringing the simulated state closer to the large-scale driving state. This can be expressed by the following formula: In the formula, α It refers to horizontal wind, potential temperature, or geopotential height. α The value is obtained by interpolation of the analysis values ​​at adjacent time points, where t is time. X ( α ) represents the physical forcing term of the pattern. G α The nudging coefficient, w ( η ) represents the weighting coefficient for nudging in the vertical direction; The initial beam is p and q Analysis field at time, α pq For beam p and q Analysis field at time, i Indicates time, x Indicates the background field. y Indicates the observation field. K pq Represents the approximation coefficients for different scales; K p , K q These represent the wave vector components in the meridional and latitudinal directions, respectively; p , q These represent the wavenumbers in the longitude and latitudinal directions, respectively. P , Q Nudging wavenumbers in the meridional and zonal directions are calculated based on grid spacing, grid number, and wavelength. They are used to adjust the large-scale circulation field of the model to make it consistent with the actual forcing field. By inputting solar irradiance, historical weather data, and historical data of photovoltaic power plants into the LSTM power prediction model, ultra-short-term photovoltaic power can be predicted.

2. The ultra-short-term photovoltaic power prediction method based on multi-source data fusion as described in claim 1, characterized in that, The underlying surface data is specifically a global land cover classification dataset, and the method for generating the global land cover classification dataset is as follows: Reflectance data from MODIS Terra and Aqua satellites were collected, and the reflectance data were classified and post-processed to generate a global land cover classification dataset. The classification methods include supervised decision tree classification, maximum likelihood method, support vector machine or spectral angle mapping method. The post-processing includes noise removal, classification result optimization, multi-temporal data fusion and quality control.

3. The ultra-short-term photovoltaic power prediction method based on multi-source data fusion as described in claim 1, characterized in that, The LSTM power prediction model includes an input layer, an LSTM layer, and an output layer. The LSTM layer is used to process the time dependence in time series data. The LSTM layer has a two-layer structure, and the hidden layer dimension is used to enhance the LSTM power prediction model's ability to remember and capture time series patterns.

4. A short-term photovoltaic power prediction system based on multi-source data fusion, characterized in that, Implemented by the method described in claim 1, comprising: The WRF-Solar model is used to predict solar irradiance based on underlying surface data, surface meteorological monitoring data, and satellite monitoring data. The LSTM power prediction model is used to predict ultra-short-time photovoltaic power based on solar irradiance, historical weather data, and historical data of photovoltaic power plants.

5. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the ultra-short-term photovoltaic power prediction method based on multi-source data fusion as described in any one of claims 1-3.

6. A computer-readable storage medium storing a computer program, characterized in that, The computer program causes the computer to execute the ultra-short-term photovoltaic power prediction method based on multi-source data fusion as described in any one of claims 1-3.

Images