Multi-source data fusion radiation prediction revision method, system, device and medium

Abstract

The application provides a multi-source data fusion radiation prediction correction method, system, device and medium, relates to the cross technical field of weather forecast and new energy power generation, and comprises obtaining numerical mode radiation prediction data, satellite radiation data and ground radiation station observation data; for each photovoltaic point, the distance between the photovoltaic point and the nearest ground radiation station is calculated, and the distance is compared with a preset validity distance threshold; according to the comparison result, the ground radiation station observation data and / or the satellite radiation data are matched, and the numerical mode radiation prediction data is corrected. The application improves the accuracy and stability of distributed photovoltaic short-term and medium-term radiation prediction under the wide-area discrete layout.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of meteorological forecasting and new energy power generation, and in particular to a method, system, device and medium for multi-source data fusion radiation forecast correction. Background Technology

[0002] Distributed photovoltaic (PV) systems are geographically dispersed and have small individual capacities, making their power generation highly dependent on the intensity of solar radiation at each location. Short- and medium-term radiation forecasts are typically based on numerical weather prediction models, but these models have systematic biases and require correction using measured data from ground-based radiation stations.

[0003] However, ground-based radiation stations are costly to construct and maintain, and their spatial distribution is extremely sparse. For photovoltaic sites far from the nearest radiation station, directly using the correction parameters from that site will introduce significant errors due to differences in geographical and meteorological conditions. While satellite remote sensing radiation data has the advantage of continuous spatial coverage, its inversion accuracy is uncertain under different weather conditions.

[0004] In the existing technology, the correction method based on ground station and the correction method based on satellite data are independent of each other, and do not provide differentiated correction strategies according to the differences in the spatial relationship between different photovoltaic points and ground stations. Summary of the Invention

[0005] The purpose of this application is to provide a method, system, device and medium for correcting radiation forecasts using multi-source data fusion, so as to alleviate the above-mentioned technical problems existing in the prior art.

[0006] In a first aspect, the present invention provides a method for correcting radiation forecasts using multi-source data fusion, comprising: Acquire numerical model radiation forecast data, satellite radiation data, and ground radiation station observation data; For each photovoltaic (PV) point, the distance between the PV point and the nearest ground-based radiation station is calculated, and the distance is compared with a preset effective distance threshold. Based on the comparison results, ground-based radiation station observation data and / or satellite radiation data are matched to correct the numerical model radiation forecast data.

[0007] In an optional implementation, before calculating the distance between the photovoltaic point and the nearest ground-based radiation station for each photovoltaic point, the method further includes: Based on at least one of the factors of terrain complexity, underlying surface type, and climate characteristics of the area to be predicted, the effective distance threshold is determined by the correlation attenuation relationship between ground radiation station observation data and the actual radiation of surrounding points at different distances.

[0008] In an optional implementation, the numerical model radiation forecast data is corrected by matching ground-based radiation station observation data and / or satellite radiation data based on the comparison results, including: When the distance is less than or equal to the effective distance threshold, the numerical model radiation forecast data are corrected based on ground radiation station observation data; When the distance is greater than the effective distance threshold, the satellite radiation data is first calibrated using ground radiation station observation data to obtain calibrated satellite radiation data, and then the numerical model radiation forecast data is corrected based on the calibrated satellite radiation data.

[0009] In an optional implementation, the numerical model radiation forecast data is corrected based on observation data from ground-based radiation stations, including: Based on the location of the nearest ground radiation station, obtain historical numerical model radiation forecast sequences and ground radiation station observation sequences for the same period; Numerical model radiation forecast data of the grid where the photovoltaic point is located is input into the pre-trained first machine learning correction model to obtain the corrected radiation forecast value; wherein, the first machine learning correction model is trained based on the historical numerical model radiation forecast sequence and the ground radiation station observation sequence.

[0010] In an optional implementation, satellite radiation data is calibrated using observation data from ground-based radiation stations, including: Using ground radiation stations in the region as anchor points, satellite radiation data and ground radiation station observation data are obtained at each anchor point location; Construct a calibration model that integrates satellite radiation data with ground-based radiation station observation data; The calibration model is applied to the entire spatial range of satellite radiation data to generate calibrated satellite radiation data.

[0011] In an optional implementation, a calibration model is constructed from satellite radiation data to ground-based radiation station observation data, including: A calibration model is obtained by fitting satellite radiation data at each anchor point as the independent variable and ground radiation station observation data as the dependent variable, using one of the linear regression, geographically weighted regression, or Kriging interpolation methods.

[0012] In an optional implementation, the numerical model radiation forecast data is corrected based on the calibrated satellite radiation data, including: For each photovoltaic point grid, obtain the historical numerical model radiation forecast sequence and the calibrated satellite radiation sequence for the same period; Numerical model radiation forecast data of the grid where the photovoltaic point is located is input into a pre-trained second machine learning correction model to obtain corrected radiation forecast values; wherein, the second machine learning correction model is trained based on the numerical model radiation forecast sequence and the calibrated satellite radiation sequence.

[0013] Secondly, the present invention provides a multi-source data fusion radiation forecast correction system, comprising: The data acquisition module is used to acquire numerical model radiation forecast data, satellite radiation data, and ground radiation station observation data; The distance comparison module is used to calculate the distance between each photovoltaic point and the nearest ground radiation station, and compare the distance with a preset effective distance threshold. The correction module is used to correct the numerical model radiation forecast data by matching the ground radiation station observation data and / or the satellite radiation data according to the comparison results.

[0014] Thirdly, the present invention provides a multi-source data fusion radiation forecast correction device, including a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the method of any of the aforementioned embodiments.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the method of any of the foregoing embodiments.

[0016] The multi-source data fusion radiation forecast correction method, system, equipment, and medium provided in this application expand the correction basis from single-source to multi-source complementarity by acquiring three types of data: numerical model, satellite, and ground station data, overcoming the limitations of insufficient coverage or accuracy of single data. It calculates the distance between the photovoltaic point and the nearest ground station and compares it with an effective distance threshold, transforming spatial location relationships into actionable decision-making criteria and avoiding regional mismatch caused by applying a uniform strategy to all points. Based on the comparison results, it dynamically selects to use ground station data, satellite data, or a combination of both for correction, achieving on-demand coupling of physical mechanisms and observational constraints. These features collectively solve key problems in existing technologies such as inaccurate correction in sparse observation areas, fragmented use of multi-source data, and lack of scenario adaptability of strategies. Overall, it improves the accuracy and stability of distributed photovoltaic short- and medium-term radiation forecasts in wide-area discrete layouts. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a multi-source data fusion radiation forecast correction method provided in this application embodiment; Figure 2 A schematic diagram of the overall structure of a system framework provided in an embodiment of this application; Figure 3 A path correction decision flowchart is provided for embodiments of this application; Figure 4 A structural diagram of a multi-source data fusion radiation forecast correction system provided in this application embodiment; Figure 5 This is a structural diagram of a multi-source data fusion radiation forecast correction device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0020] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0022] This application provides a method for correcting radiation forecasts using multi-source data fusion. See also... Figure 1 As shown, the method mainly includes the following steps: S110 acquires numerical model radiation forecast data, satellite radiation data, and ground radiation station observation data.

[0023] Numerical model radiation forecast data refers to gridded downwave radiation forecast products output by numerical weather prediction models, providing a basic forecast field for subsequent corrections. Satellite radiation data refers to surface solar radiation data retrieved from geostationary meteorological satellites, characterized by continuous spatial coverage. Ground-based radiation station observation data refers to solar radiation values ​​measured by ground-based radiation meteorological stations, serving as a high-precision reference value.

[0024] In practice, the data acquisition and preprocessing module is responsible for collecting the three types of data mentioned above. Specifically, numerical model radiation forecast data is acquired from numerical forecast products containing radiation elements released by the operational meteorological forecasting center, with a time horizon of 72 hours. Satellite radiation data is acquired from radiation inversion products of geostationary meteorological satellites, with the time horizon matching the start of the model forecast. Ground radiation station observation data is acquired from historical and real-time observations from at least one ground radiation meteorological station within the target area. All data need to be standardized in terms of spatiotemporal resolution and coordinate system, and quality control measures are implemented, such as removing outliers and filling in missing values, to form standardized input.

[0025] S120, For each photovoltaic point, calculate the distance between the photovoltaic point and the nearest ground radiation station, and compare the distance with a preset effective distance threshold.

[0026] The effective distance threshold is a pre-set spatial distance critical value based on regional environmental characteristics, used to determine the representativeness of ground station observation data for distant photovoltaic sites. The distance between the photovoltaic site and the nearest ground station is the actual spherical distance calculated using coordinates.

[0027] In practice, the correction path decision-maker performs the following operations: For any distributed photovoltaic point to be predicted, it calculates the spherical distance between the point and all ground-based radiation stations within the target area based on its geographical coordinates, and finds the minimum value. This minimum distance is then compared with a preset effective distance threshold. The method for determining this threshold will be explained in detail later. The comparison result determines which correction path to adopt subsequently.

[0028] S130, Based on the comparison results, match the ground radiation station observation data and / or the satellite radiation data to correct the numerical model radiation forecast data.

[0029] This step provides a higher-level overview of the correction methods. Depending on the distance comparison results, one can choose to correct the model forecast using only ground-based radiation station observation data, or one can choose to first calibrate the satellite data using ground-based station data, and then use the calibrated satellite data to correct the model forecast. In other words, the most suitable combination of data sources is dynamically matched based on the actual spatial location relationships.

[0030] In practice, the core intelligent correction engine executes the corresponding correction operations based on the path instructions output by the decision-maker. Specifically, when the photovoltaic point is close to the nearest ground station (not exceeding the threshold), the model forecast is directly corrected using only ground station observation data; when the photovoltaic point is far from all ground stations (exceeding the threshold), the satellite radiation field is first calibrated using ground station observation data, and then the model forecast is corrected using the calibrated satellite radiation field. These two specific implementation methods will be discussed in detail later.

[0031] The multi-source data fusion radiation forecast correction method provided in this application integrates three types of data: numerical model forecasts, satellite remote sensing observations, and ground station measurements. It adaptively matches the most suitable data source combination based on the distance between the photovoltaic site and the nearest ground station (directly using the site correction mode for nearby sites, and calibrating the satellite first and then using the satellite correction mode for distant sites). This achieves personalized and accurate correction for each site, effectively solving the problem of insufficient forecast accuracy in sparsely observed areas of distributed photovoltaic scenarios. It also takes into account the high accuracy preservation of nearby sites and the full spatial coverage of areas without stations, significantly improving the overall reliability and accuracy of wide-area distributed photovoltaic radiation forecasts.

[0032] For ease of understanding, the methods provided in this application are described in detail below.

[0033] The aforementioned effective distance threshold is not a fixed value, but is set based on at least one of the following factors: topographic complexity, underlying surface type, and climate characteristics of the area to be predicted. Topographic complexity includes different landforms such as mountains and plains; underlying surface type includes cities, farmland, and water bodies; and climate characteristics include continental and maritime climates.

[0034] Specifically, the threshold can be quantified by analyzing the correlation decay patterns between ground-based radiation station observations and the actual radiation at surrounding locations at different distances in historical data. For example, in flat plains or farmland areas with uniform underlying surfaces, the spatial correlation decay of ground-based station observations is slower, and the effective distance threshold can be set at 15 to 25 kilometers. In mountainous or urban areas with complex terrain and drastically changing underlying surfaces, the correlation decays rapidly, and the threshold should be set at 5 to 10 kilometers. This dynamic setting method makes the selection of correction paths more regionally adaptable.

[0035] This embodiment mainly focuses on matching the corresponding correction method based on the comparison results of the above distances. Two specific correction methods are provided below.

[0036] Method 1: When the distance between the photovoltaic (PV) point and the nearest ground station is less than or equal to the effective distance threshold, a direct correction method based on ground station observation data is adopted. In specific implementation, using the location of the nearest ground station as a reference, at least 1 to 3 years of historical data for the same period are collected, including historical forecast sequences from numerical models at that location and historical observation sequences from the ground station. A machine learning algorithm is used to train the correction model. Using the location of the nearest ground station as a reference, historical numerical model radiation forecast sequences and ground radiation station observation sequences for the same period are obtained. The numerical model radiation forecast data of the grid where the PV point is located is input into the pre-trained first machine learning correction model to obtain the corrected radiation forecast value; wherein, the first machine learning correction model is trained based on the historical numerical model radiation forecast sequences and ground radiation station observation sequences for the same period.

[0037] When training the first machine learning correction model, it can be constructed using one of the following methods: random forest, gradient boosting machine, support vector regression, or artificial neural network. The numerical model radiation forecast sequence is used as input features, and multiple factors such as temperature, humidity, and cloud cover from the model forecast, as well as static features such as topographic elevation and season, can be added as auxiliary features. The ground radiation station observation sequence is used as the training target to fit the model parameters, resulting in the trained first machine learning correction model. During the online operation phase, only real-time model forecast data needs to be input, and the model can quickly output correction results.

[0038] Method 2: When the distance between the photovoltaic point and the nearest ground station is greater than the effective distance threshold, satellite is used. The site joint correction method is divided into two phases.

[0039] The first stage involves calibrating satellite radiation data using observation data from ground-based radiation stations. Specifically, using ground-based radiation stations within the region as anchor points, satellite radiation data and ground-based radiation station observation data are acquired at each anchor point. A calibration model is then constructed to convert the satellite radiation data into ground-based radiation station observation data. This calibration model is applied to the entire spatial range of the satellite radiation data to generate calibrated satellite radiation data. When constructing the calibration model, one of the following methods can be used: linear regression, geographic weighted regression, or kriging interpolation. The satellite radiation data at each anchor point is used as the independent variable, and the ground-based radiation station observation data is used as the dependent variable to fit the calibration model.

[0040] The second stage involves correcting the numerical model radiation forecast data based on calibrated satellite radiation data. Specifically, for each grid cell containing a photovoltaic (PV) point, historical numerical model radiation forecast sequences and calibrated satellite radiation sequences for the same period are acquired. The numerical model radiation forecast data for the grid cell containing the PV point is then input into a pre-trained second machine learning correction model to obtain the corrected radiation forecast value. The second machine learning correction model is trained based on the numerical model radiation forecast sequences and the calibrated satellite radiation sequences.

[0041] The training method for the second machine learning correction model is similar to that of the first model. It can be constructed using one of the following: random forest, gradient boosting machine, support vector regression, or artificial neural network. It uses numerical model radiation forecast sequences as input features and calibrated satellite radiation sequences as training targets to fit model parameters.

[0042] Furthermore, this application provides several alternative implementation methods. For example, the core machine learning model in the correction method can be replaced with other supervised learning algorithms such as gradient boosting machines, support vector regression, or artificial neural networks. Alternatively, instead of a strict binary path, a distance-based weighted fusion scheme can be designed: weights are calculated based on the distance between the photovoltaic point and the nearest ground station, with closer points receiving larger weights. The final correction result is a weighted average of the results from the two paths, achieving a smooth transition. Furthermore, in the satellite product calibration step, in addition to geographic weighted regression, Kriging interpolation can be used to interpolate ground station observations into a continuous field, which is then fused and corrected with the original satellite field. These alternative schemes all achieve the same inventive objective and fall within the scope of protection of this application.

[0043] In practical applications, the high-precision, point-based radiation forecasts output by the embodiments of this application can be directly input into distributed photovoltaic power generation prediction models. The prediction results serve as a crucial basis for grid dispatch departments to formulate daily power generation plans and for distributed photovoltaic aggregators to participate in electricity market transactions. The aforementioned random forest and other models require offline training using historical data of a sufficiently long time series (typically 1 to 3 years). During the online operation phase, only real-time model forecast products and the latest observation data need to be input to quickly output correction results. The technical solution of this application is essentially a hybrid paradigm of "numerical model physics-driven + multi-source observation data fusion correction," using physical model results as a basis for error correction, combining physical consistency with the accuracy of data-driven approaches.

[0044] In one specific embodiment of the present invention, a multi-source data fusion intelligent correction technology for short- and medium-term radiation forecasts suitable for distributed photovoltaic scenarios is provided. The embodiment will now be described in detail with reference to the accompanying drawings.

[0045] First, the data sources and types are explained. The data used in this embodiment includes three categories: the first category is numerical weather prediction radiation products, such as numerical forecast products containing radiation elements released by operational meteorological forecasting centers; the second category is satellite radiation data, such as radiation inversion products from geostationary meteorological satellites; and the third category is ground radiation observation data, derived from actual radiation observations at ground-based radiation meteorological stations. In addition, the geographical coordinates of each small photovoltaic system within the wide-area distributed photovoltaic system to be forecasted are also required.

[0046] like Figure 2 As shown, the system in this embodiment comprises a data acquisition and preprocessing module, a path correction decision-maker, a core intelligent correction engine, and a forecast result integration and output module.

[0047] The data acquisition and preprocessing module is responsible for collecting and standardizing multi-source input data. Specific tasks include: acquiring the gridded downlink shortwave radiation forecast data for the next 72 hours from the numerical model output, denoted as... ; Obtain satellite-retrieved surface solar radiation data for the covered area that matches the start time of the model forecast, denoted as This data has spatial continuity; it acquires the latest radiation observation data from at least one ground-based radiation station within the region, both historically and in real-time, denoted as... Obtain the coordinates of the distributed photovoltaic power station to be predicted, denoted as... All data is standardized to the same spatiotemporal resolution and coordinate system, and undergoes quality control (such as outlier removal and missing value imputation). The preprocessed data is then fed into the correction path decision-maker and the core intelligent correction engine.

[0048] The path correction decision-maker is the core of the adaptive capability in this embodiment. For example... Figure 3 As shown, the decision-making process is as follows. First, a key parameter, "effectiveness distance," needs to be determined. K ". K It is not a fixed value, but rather a value determined comprehensively based on the topographic complexity (e.g., mountains, plains), underlying surface homogeneity (e.g., cities, farmland, water bodies), and climatic characteristics (e.g., continental, maritime climate) of the target area. It can be determined by analyzing historical data to examine the correlation attenuation patterns between ground station observations at different distances and the actual radiation at surrounding points. For example, in flat, homogeneous areas... K It can be set to 15 to 25 kilometers in complex mountainous areas. K It can be set to 5 to 10 kilometers.

[0049] Decision-making logic such as Figure 3 As shown: For any photovoltaic point to be predicted P First, input its coordinates, calculate its distance to all ground stations in the area, and determine the minimum distance. Then With threshold K Compare. If Then it is considered that the observation of this station is correct. P With good representativeness, the correction method (1) was selected, namely, direct correction of site-pattern; if If the site is deemed insufficiently representative, correction method (2) is selected, namely satellite-site joint correction. The decision-maker outputs correction path instructions and sends them to the core intelligent correction engine.

[0050] The core intelligent correction engine comprises two parallel correction technology routes, respectively corresponding to... Figure 1 The correction methods (1) and (2) are described in the text.

[0051] Correction method (1) is a direct site-model correction, suitable for points close to the ground station. This method uses the location of the nearest ground station S as a reference and collects historical data, including the model's forecast sequence F at that location. s and site observation sequence O s A correction model M1 is trained using machine learning algorithms such as random forest, and its mapping relationship can be expressed as follows: OtherFeatures can include multiple factors such as model-predicted temperature, humidity, and cloud cover, as well as static features such as terrain elevation and season. After training, the model-predicted values ​​F of the grid containing the target photovoltaic point P are... p Input model M1 to obtain the corrected prediction value C at point P. p1 This method essentially uses high-precision station observations to perform "point-to-point" deviation correction on the model.

[0052] Correction method (2) is a satellite-site joint correction, applicable to points far from the ground station. This method consists of two steps. The first step is satellite product calibration: using all ground stations S in the area... i The location is used as the anchor point to measure the satellite's radiation at that location. Compared with ground station observation value O i By comparing the results, a calibration model (such as linear regression or geographic weighted regression) is established to obtain a calibrated, more accurate regional satellite radiation field S. calibrated This step uses the true values ​​of ground station "points" to correct the satellite "area" data. The second step is model-satellite field correction: the calibrated satellite radiation field S... calibrated This is considered a "near-true value". For each grid point, the model forecast value F from historical data is collected. grid and S in the same period calibrated The value is used to train another random forest model M2, whose mapping relationship is as follows: For the target photovoltaic point P, its grid model prediction value F pInput model M2 to obtain the corrected forecast value C p2 This method essentially uses calibrated, spatially continuous satellite data as an intermediary to perform "surface-to-surface" constraint corrections on the model.

[0053] The forecast results integration and output module receives correction results (partially C) from all photovoltaic points in the correction engine. p1 Part of it is C p2 This data is integrated into a complete set of corrected radiation forecast products covering all distributed photovoltaic (PV) sites, and then output to downstream PV power prediction systems, such as... Figure 1 The final result obtained is the accurate radiation forecast of distributed photovoltaic sites.

[0054] In this embodiment, some alternative solutions can also be used to achieve the same inventive objective. For example, the core machine learning model "random forest" in correction methods (1) and (2) can be replaced by other supervised learning algorithms such as gradient boosting machine, support vector regression, or artificial neural network. Alternatively, instead of a strict "either / or" path, a distance-based weighted fusion scheme can be designed: the weight w is calculated based on the distance D (e.g., w = e^(-1 / 2)). (D / K)^2 (The smaller D is, the larger w is), final correction result This allows for a smooth transition between the results of the two paths. For example, in the satellite product calibration process, in addition to geographic weighted regression, Kriging interpolation can be used to interpolate ground station observations into a continuous field, which is then fused and corrected with the original satellite field.

[0055] The high-precision, point-based radiation forecast output in this embodiment can be directly input into the distributed photovoltaic power generation prediction model. Its prediction results are a key basis for grid dispatch departments to formulate daily power generation plans and for distributed photovoltaic aggregators (virtual power plants) to participate in the electricity spot market and ancillary service market bidding. Random forest and other models require offline training using historical data of a sufficiently long time series (typically 1 to 3 years). During the online operation phase, only real-time model forecast products and the latest observation data need to be input to quickly output correction results. This embodiment is essentially a hybrid paradigm of "numerical model physics-driven + multi-source observation data fusion correction," unlike pure AI models that learn forecast patterns entirely from end-to-end data. It uses physical model results as a basis for error correction, combining physical consistency with the accuracy of data-driven approaches.

[0056] In summary, the multi-source data fusion radiation forecast correction method provided in this application integrates three types of observation data: numerical model, satellite, and ground station data. It adaptively matches the correction path based on the distance between the photovoltaic site and the nearest ground radiation station: when the distance is short, it directly uses the site's actual measurements to correct the model forecast; when the distance is long, it first calibrates the satellite data using the ground station data, and then uses the calibrated satellite field to correct the model forecast. This scheme balances high-precision preservation in near-station areas with spatial coverage in station-free areas, solving the problem of insufficient spatial representativeness in correction caused by sparse observations in distributed photovoltaic scenarios, and improving the overall accuracy and reliability of wide-area radiation forecasts.

[0057] Based on the above method embodiments, this application also provides a multi-source data fusion radiation forecast correction system, see [link to relevant documentation]. Figure 4 As shown, the system includes the following components: Data acquisition module 410 is used to acquire numerical model radiation forecast data, satellite radiation data and ground radiation station observation data; The distance comparison module 420 is used to calculate the distance between each photovoltaic point and the nearest ground radiation station, and compare the distance with a preset effective distance threshold. The correction module 430 is used to correct the numerical model radiation forecast data by matching the ground radiation station observation data and / or the satellite radiation data according to the comparison results.

[0058] In one feasible implementation, a threshold determination module is also included, which is used to determine the effective distance threshold based on at least one of the following factors: topographic complexity, underlying surface type, and climate characteristics of the area to be predicted, by means of the correlation attenuation relationship between ground radiation station observation data at different distances and the actual radiation of surrounding points.

[0059] In one feasible implementation, the aforementioned correction module is specifically used to: correct the numerical model radiation forecast data based on ground radiation station observation data when the distance is less than or equal to the effective distance threshold; and when the distance is greater than the effective distance threshold, first calibrate the satellite radiation data using ground radiation station observation data to obtain calibrated satellite radiation data, and then correct the numerical model radiation forecast data based on the calibrated satellite radiation data.

[0060] In one feasible implementation, the aforementioned correction module is specifically used to: obtain historical numerical model radiation forecast sequences and ground radiation station observation sequences based on the location of the nearest ground radiation station; input the numerical model radiation forecast data of the grid where the photovoltaic point is located into a pre-trained first machine learning correction model to obtain corrected radiation forecast values; wherein, the first machine learning correction model is trained based on historical numerical model radiation forecast sequences and ground radiation station observation sequences.

[0061] In one feasible implementation, the aforementioned correction module is specifically used to: acquire satellite radiation data and ground radiation station observation data at each anchor point location, using ground radiation stations in the region as anchor points; construct a calibration model from satellite radiation data to ground radiation station observation data; and apply the calibration model to the entire spatial range of satellite radiation data to generate calibrated satellite radiation data.

[0062] In one feasible implementation, the aforementioned correction module is specifically used to: fit a calibration model by using satellite radiation data at each anchor point location as the independent variable and ground radiation station observation data as the dependent variable, based on one of linear regression, geographically weighted regression, or Kriging interpolation methods.

[0063] In one feasible implementation, the aforementioned correction module is specifically used to: for each grid where a photovoltaic point is located, acquire the historical numerical model radiation forecast sequence and the calibrated satellite radiation sequence for the same period; input the numerical model radiation forecast data of the grid where the photovoltaic point is located into a pre-trained second machine learning correction model to obtain the corrected radiation forecast value; wherein, the second machine learning correction model is trained based on the numerical model radiation forecast sequence and the calibrated satellite radiation sequence.

[0064] The system provided in this application embodiment has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the embodiment of the device can be referred to the corresponding content in the aforementioned method embodiment.

[0065] This application also provides an electronic device, such as... Figure 5 The diagram shows the structure of the electronic device 100, which includes a processor 51 and a memory 50. The memory 50 stores computer-executable instructions that can be executed by the processor 51. The processor 51 executes the computer-executable instructions to implement any of the methods described above.

[0066] exist Figure 5 In the illustrated embodiment, the electronic device further includes a bus 52 and a communication interface 53, wherein the processor 51, the communication interface 53, and the memory 50 are connected via the bus 52.

[0067] The memory 50 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 53 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 52 may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 52 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0068] The processor 51 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 51 or by instructions in software form. The processor 51 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in the memory, and the processor 51 reads the information in the memory and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiment.

[0069] This application also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions cause the processor to implement the above-described method. For specific implementation details, please refer to the foregoing method embodiments, which will not be repeated here.

[0070] The computer program products of the multi-source data fusion radiation forecast correction method, system, device and medium provided in the embodiments of this application include a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0071] Unless otherwise specifically stated, the relative steps, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application.

[0072] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0073] In the description of this application, it should be noted that the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0074] Finally, it should be noted that 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 or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for correcting radiation forecasts using multi-source data fusion, characterized in that, include: Acquire numerical model radiation forecast data, satellite radiation data, and ground radiation station observation data; For each photovoltaic (PV) point, the distance between the PV point and the nearest ground-based radiation station is calculated, and the distance is compared with a preset effective distance threshold. The numerical model radiation forecast data is corrected by matching the ground radiation station observation data and / or the satellite radiation data based on the comparison results.

2. The method according to claim 1, characterized in that, Before calculating the distance between each photovoltaic (PV) point and the nearest ground-based radiation station, the process also includes: The effective distance threshold is determined by considering at least one of the following factors: topographic complexity, underlying surface type, and climate characteristics of the area to be predicted. This is achieved through the correlation attenuation relationship between ground radiation station observation data at different distances and the actual radiation at surrounding locations.

3. The method according to claim 1, characterized in that, Based on the comparison results, the numerical model radiation forecast data is corrected by matching the ground-based radiation station observation data and / or the satellite radiation data, including: When the distance is less than or equal to the effective distance threshold, the numerical model radiation forecast data is corrected based on the ground radiation station observation data; When the distance is greater than the effective distance threshold, the satellite radiation data is first calibrated using the observation data from the ground radiation station to obtain calibrated satellite radiation data, and then the numerical model radiation forecast data is corrected based on the calibrated satellite radiation data.

4. The method according to claim 3, characterized in that, Correcting the numerical model radiation forecast data based on the ground-based radiation station observation data includes: Based on the location of the nearest ground radiation station, obtain historical numerical model radiation forecast sequences and ground radiation station observation sequences for the same period; The numerical model radiation forecast data of the grid where the photovoltaic point is located is input into the pre-trained first machine learning correction model to obtain the corrected radiation forecast value; wherein, the first machine learning correction model is trained based on the historical numerical model radiation forecast sequence and the ground radiation station observation sequence.

5. The method according to claim 1, characterized in that, The satellite radiation data is calibrated using observation data from the ground radiation station, including: Using ground radiation stations in the region as anchor points, satellite radiation data and ground radiation station observation data are obtained at each anchor point location; Construct a calibration model that translates satellite radiation data into ground-based radiation station observation data; The calibration model is applied to the entire spatial range of the satellite radiation data to generate calibrated satellite radiation data.

6. The method according to claim 5, characterized in that, Constructing a calibration model from satellite radiation data to ground-based radiation station observation data includes: The calibration model is obtained by fitting satellite radiation data at each anchor point location as the independent variable and ground radiation station observation data as the dependent variable, using one of the linear regression, geographically weighted regression, or Kriging interpolation methods.

7. The method according to claim 3, characterized in that, Correcting the numerical model radiation forecast data based on the calibrated satellite radiation data includes: For each photovoltaic point grid, obtain the historical numerical model radiation forecast sequence and the calibrated satellite radiation sequence for the same period; The numerical model radiation forecast data of the grid where the photovoltaic point is located is input into a pre-trained second machine learning correction model to obtain the corrected radiation forecast value; wherein, the second machine learning correction model is trained based on the numerical model radiation forecast sequence and the calibrated satellite radiation sequence.

8. A multi-source data fusion radiation forecast correction system, characterized in that, include: The data acquisition module is used to acquire numerical model radiation forecast data, satellite radiation data, and ground radiation station observation data; The distance comparison module is used to calculate the distance between each photovoltaic point and the nearest ground radiation station, and compare the distance with a preset effective distance threshold. The correction module is used to correct the numerical model radiation forecast data by matching the ground radiation station observation data and / or the satellite radiation data according to the comparison results.

9. A multi-source data fusion radiation forecast correction device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method according to any one of claims 1 to 7.

Images