A method and system for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and forecasting
By constructing a dynamic quantification method for multi-precision, multi-source irradiance data and employing a three-dimensional variational or ensemble Kalman filter data assimilation framework, the problem of dynamic quantification and uncertainty distribution of multi-source observation data in photovoltaic resource assessment is solved. This enables probabilistic risk assessment and decision support for photovoltaic power generation, and improves the scientificity and robustness of grid dispatching and investment decisions.
Patent Information
- Application Number
- CN202611370401.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-04
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies lack dynamic quantification of multi-source observation data and processing of spatiotemporal uncertainty distribution in photovoltaic resource assessment, making it difficult to quantify the uncertainty of photovoltaic power generation risk assessment and dispatch decisions, and failing to meet the needs of probabilistic prediction.
By constructing a dynamic quantization method for multi-precision, multi-source irradiance data, and employing a three-dimensional variational or ensemble Kalman filter data assimilation framework, combined with a Gaussian distribution probability model, confidence interval fields and over-threshold probability fields are generated, thereby achieving optimal fusion and uncertainty quantification of multi-source data.
It provides a spatiotemporal uncertainty distribution with physical meaning and statistical consistency, supports probabilistic risk assessment and decision-making for photovoltaic power generation, and improves the scientificity and robustness of grid dispatch and investment decisions.
Smart Images

Figure CN123069702A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of new energy meteorology and data assimilation technology, specifically to a method and system for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction. Background Technology
[0002] Irradiance spatial analysis fields are gridded irradiance data products covering a target area, serving as crucial foundational data for photovoltaic power plant site selection assessment, power generation prediction, and grid dispatch decisions. Currently, the mainstream method for generating this type of data product is deterministic spatial interpolation. This method uses measured data from ground meteorological stations as the sole input and employs algorithms such as inverse distance weighting, Kriging interpolation, and spline function interpolation to extrapolate discrete station observation data to obtain global spatial grid data.
[0003] With the continuous development of geostationary meteorological satellite technology, existing technologies are beginning to incorporate satellite-derived irradiance products as auxiliary data sources to compensate for the sparse distribution of ground meteorological stations. These schemes mostly employ weighted averaging and regression fusion methods, linearly combining satellite data and station observation data according to fixed weights. In the field of solar numerical weather prediction, there are also related studies on assimilating satellite cloud and aerosol information into numerical models.
[0004] However, existing technical solutions still have significant shortcomings. First, traditional deterministic interpolation and simple weighted fusion strategies only output the optimal estimate, lacking a quantifiable and physically meaningful spatiotemporal uncertainty distribution. They treat satellite and site data uniformly as noisy true-value observations, lacking a mathematical framework for dynamic quantification and propagation of various observation errors, and thus cannot output uncertainty quantification results. This results in power dispatching and photovoltaic investment decisions being limited to a single deterministic value, making it difficult to conduct probabilistic risk assessments.
[0005] Secondly, existing fusion methods generally assign fixed weights to different observation sources based on distance or experience, failing to dynamically adjust weights according to the actual reliability of the data in the spatiotemporal dimensions, and lacking the ability to model the qualitative differences in errors from multi-source observations. Especially for satellite indirect inversion data, no error model has been established to follow dynamic changes in atmospheric conditions, making it impossible to adaptively adjust contribution weights based on data quality, and thus insufficiently utilizing multi-source observation information.
[0006] Furthermore, photovoltaic power generation risk assessment and grid-connected dispatch are shifting from deterministic to probabilistic prediction. However, existing irradiance analysis fields only output single-point values and cannot provide probabilistic information on irradiance within specific ranges. This makes it difficult to assess operational risks under extreme conditions such as sudden drops in irradiance caused by rapid cloud movement. Existing technologies interrupt the flow of uncertain information during data fusion, lacking a processing step to transmit uncertainty to the final product. Subsequent power generation simulation and risk assessment can only use deterministic inputs, making it impossible to conduct ensemble probabilistic analysis and failing to meet the practical needs of probabilistic decision-making. Summary of the Invention
[0007] In order to overcome the shortcomings of the existing technology, the present invention aims to provide a method and system for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction, so as to solve the technical problem of how to construct a unified mathematical framework to achieve optimal fusion of multi-precision multi-source irradiance data, and dynamically quantify spatiotemporal uncertainty in the fusion process, thereby supporting the probabilistic risk assessment and decision-making of photovoltaic power generation.
[0008] This invention is achieved through the following technical solution: This invention provides a method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction, comprising: Satellite-inverted gridded irradiance data and ground-station measured irradiance data of the target area were collected. Anomaly control, validity screening and spatial matching resampling were performed on the two types of multi-source data in sequence to obtain spatiotemporally aligned and qualified multi-source observation data. Using multi-source observation data as input, the differential error characteristics of high-precision ground station observations and low-precision satellite inversion observations are distinguished. Fixed parameter ground station observation error models and time-varying and space-varying satellite dynamic error models are constructed respectively. The observation bias is statistically analyzed and the error variance is updated in real time by combining cloud influence factors to construct the global observation error covariance matrix. Using satellite-inverted gridded irradiance data as the background field and ground station measured irradiance data as the true observation value, and constrained by the constructed global observation error covariance matrix, the optimal analytical field of global gridded irradiance is obtained by iterative optimization through a three-dimensional variational or ensemble Kalman filter data assimilation framework. Based on the error statistics characteristics generated by the iterative optimal solution process of the assimilation framework, the spatiotemporal error variance distribution of the global gridded irradiance optimal analysis field is extracted. The uncertainty is quantified by combining the Gaussian distribution probability model, generating two types of probabilistic analysis fields: confidence interval field and over-threshold probability field, thus completing the construction of the irradiance spatial analysis field.
[0009] Preferably, the specific process of collecting satellite-inverted gridded irradiance data and ground-based measured irradiance data for the target area, and sequentially performing anomaly control, validity screening, and spatial matching resampling on the two types of multi-source data to obtain spatiotemporally aligned and qualified multi-source observation data includes: Visible and infrared channel data from geostationary meteorological satellites within the target area are collected, and satellite irradiance grid data are obtained through a radiative transfer inversion algorithm. Collect measured irradiance data from ground-based radiation observation stations in the target area; Remove invalid grid points from satellite data that are obscured by clouds or fail to be retrieved; perform extreme value verification and nighttime zeroing time consistency verification on station data to remove outliers. By interpolating satellite grid data to the locations of each station using bilinear interpolation, spatial registration of satellite data and station measured data is achieved, resulting in standardized observation data with spatiotemporal alignment.
[0010] Preferably, the construction process of the ground station observation error model is as follows: Based on the obtained ground station measured data, the standard deviation and variance of the station observation error are set according to the instrument calibration accuracy. Assuming that the observation errors of each ground station are independent, a diagonal matrix form of the station observation error covariance matrix is constructed.
[0011] Preferably, the process of constructing the satellite dynamic error model is as follows: Based on the spatially matched multi-source observation data, the deviation between the satellite inversion data and the measured data of each station location is calculated using the following formula: d(r i ,t)=I sat (r i ,t) I sta (r i ,t) in, I sat (r i ,t) The equivalent irradiance of the satellite at each station; I sta (r i ,t) Gridded data of incident solar irradiance on the Earth's surface; Based on satellite infrared window brightness temperature T 11 (r,t), visible light channel albedo A vis (r,t), constructing a cloud impact factor that can quantify changes in cloud cover using an empirical function: C(r,t)=f(T 11 ,A vis ) in, f This is an empirical function; the greater the cloud cover, the more... C The higher the value, C ∈[0,1]; Based on the deviation sample of the historical sliding window over the past 30 days, combined with the stratified statistics of real-time cloud impact factors, the variance of satellite observation error is calculated in real time. The calculation formula is as follows: s sat 2 (r,t)=Var [ d(r,τ)∣C(r,τ)≈C(r,t) ] in, d(r,τ) The satellite-site irradiance deviation at historical time τ and location r; Real-time satellite error variance at each grid point s sat 2 (r,t) Construct a diagonal matrix R sat Combined with the site error matrix R sta The global observation error covariance matrix R is obtained by integrating the components into a block-diagonal form.
[0012] Preferably, the specific process of obtaining the optimal analytical field of global gridded irradiance through iterative optimal solution using a three-dimensional variational assimilation framework is as follows: Using satellite data as background field x b The site data is represented by the observation vector y. Using the constructed global observation error covariance matrix R as a constraint, the fusion cost function is constructed as follows: J(x) = ½(xx) b ) T B -1 (xx b )+½(Hx-y) T R -1 (Hx-y) Where x is the analytical field to be solved, x b Background field, i.e., satellite-retrieved irradiance gridded data I sat B is the background error covariance matrix, representing the spatial correlation error structure of the satellite inversion field; H is the observation operator, i.e., the spatial interpolation operator from the grid space to the station location; y is the observation vector, containing the measured values I of all stations. sta R is the observation error covariance matrix; Construct the background error covariance matrix: B = Σ b ×C×Σ b ; Where, Σ b Let C be the satellite error variance diagonal matrix, and let C be the spatial correlation matrix. The elements of C are calculated using the Gaussian correlation function, and the correlation length scale L is 50km. The optimal irradiance analysis field x is obtained by iteratively minimizing the cost function J(x) using either the steepest descent method or the conjugate gradient method. a .
[0013] Preferably, the specific process of obtaining the optimal global gridded irradiance analysis field through iterative optimization using an ensemble Kalman filter data assimilation framework is as follows: Based on satellite background field x b The constructed background error covariance matrix B is used to represent the satellite background field x. b Add a Gaussian perturbation conforming to N(0,B) to generate N e A group of background field members x b (k) , k=1,2,...,N e ; Background error covariance matrix P based on ensemble estimation b Calculate the Kalman gain matrix using the global observation error covariance matrix R, as shown in the following formula: K=P b H T HP b H T +R) -1 Among them, P b H is the set of background error covariance matrices, which characterizes the spatial error discreteness of the satellite background field; T is the transpose matrix of the observation operator; H is the spatial interpolation observation operator used to map the grid field to the observation location of the station; R is the global observation error covariance matrix; Based on site observation data, perform status updates for each set member: x a (k) =x b (k) +K(y (k) -Hx b (k) ); The optimal analytical field x is obtained by taking the ensemble average. a The statistical characteristics of the analysis error are obtained by measuring the dispersion of the set members.
[0014] Preferably, the three-dimensional variational assimilation scheme achieves efficient solution through multi-source observation data and global observation error covariance matrix, and is used in operational irradiance field production scenarios with limited computational resources; The ensemble Kalman filter data assimilation scheme achieves high-precision fusion based on dynamic error characteristics, and is used in photovoltaic risk assessment research scenarios where high precision in uncertainty quantification is required.
[0015] Preferably, the specific process for extracting the spatiotemporal error variance distribution of the optimal analytical field of global gridded irradiance based on the error statistical characteristics generated by the iterative optimal solution process of the assimilation framework includes: When using a three-dimensional variational assimilation scheme, the formula for approximating the analysis error covariance matrix using the inverse Hessian matrix is as follows: P a ≈ (B -1 + H T R -1 H) -1 Where B is the background error covariance matrix, representing the spatial error characteristics of the satellite-retrieved background field; the optimal estimation error covariance matrix P is extracted. a The diagonal elements provide the variance of the analysis error at each grid point. s a 2 (r) ; When using an ensemble Kalman filter data assimilation scheme, the variance of the analysis error at each grid point is directly calculated using the discreteness of the ensemble members, as shown in the following formula: s a 2 (r)=Var [ x a (k) (r) ] in, x a (k) (r) This is the assimilated irradiance analysis field of the k-th set member at spatial location r within the Kalman filter data assimilation framework.
[0016] Preferably, uncertainty quantification is achieved by combining a Gaussian distribution probability model to generate two types of probabilistic analysis fields: a confidence interval field and a threshold probability field. Specifically, these include: Confidence interval calculation: Given a confidence level α, based on the quantiles of the standard normal distribution, the grid-point confidence interval is obtained, as shown in the following formula: CI 1-α (r)=[x a (r)-z 1-α / 2 ·s a (r), x a (r)+z 1-α / 2 ·s a (r)] in, x a (r) This is the estimated value of the optimal irradiance analysis field at spatial location r after assimilation; z1-α / 2 These are the two-tailed quantiles corresponding to the standard normal distribution; Calculation of probability field exceeding threshold: setting an irradiance threshold I thr Based on the cumulative distribution function Φ of the standard normal distribution, the probability of grid point irradiance exceeding the threshold is calculated, and the probabilistic product output is completed.
[0017] Secondly, the present invention also provides a system for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction, comprising: The data preprocessing module is used to collect satellite-inverted gridded irradiance data and ground station measured irradiance data of the target area. It sequentially performs anomaly quality control, validity screening and spatial matching resampling on the two types of multi-source data to obtain spatiotemporally aligned and qualified multi-source observation data. The error modeling module is used to take multi-source observation data as input, distinguish the differentiated error characteristics of high-precision ground station observations and low-precision satellite inversion observations, construct a ground station observation error model with fixed parameters and a time-varying and space-varying satellite dynamic error model, respectively, statistically analyze the observation bias and update the error variance in real time with cloud influence factors, and construct the global observation error covariance matrix. The data assimilation module is used to obtain the optimal analysis field of global gridded irradiance by taking satellite-inverted gridded irradiance data as the background field and ground station measured irradiance data as the observation true value. It uses the constructed global observation error covariance matrix as a constraint and iteratively solves the optimal solution through a three-dimensional variational or ensemble Kalman filter data assimilation framework. The probability output module is used to extract the spatiotemporal error variance distribution of the global gridded irradiance optimal analysis field by extracting the error statistical characteristics generated by the iterative optimal solution process based on the assimilation framework. It combines the Gaussian distribution probability model to complete the uncertainty quantification and generate two types of probabilistic analysis fields: confidence interval field and over-threshold probability field, thus completing the construction of the irradiance spatial analysis field.
[0018] Compared with the prior art, the present invention has the following beneficial technical effects: This invention provides a method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction. By constructing a dynamic observation error covariance matrix, the data fusion weights can be adjusted in real time according to actual atmospheric conditions such as cloud cover and aerosols, overcoming the empirical and blind nature of traditional fixed-weight methods. Under the optimal linear unbiased estimation criterion, the optimality of multi-source observation data fusion results is guaranteed. Simultaneously, this scheme quantifies uncertainty during the irradiance analysis field construction process. Based on the analysis error covariance matrix or set dispersion output by the data assimilation framework, a physically meaningful and statistically consistent spatiotemporal uncertainty distribution is obtained, changing the traditional... The technology has limitations in outputting only a single, definite value. The confidence interval field and over-threshold probability field generated by the scheme can be directly supplied to the downstream photovoltaic power generation probability prediction model, providing complete uncertainty information for grid dispatch and photovoltaic investment decisions, effectively improving the scientificity and robustness of the actual decision-making process. In addition, this method is compatible with two assimilation implementation paths: three-dimensional variational and ensemble Kalman filtering. Three-dimensional variational calculation has high computational efficiency and can meet the needs of operational high-frequency product production. Ensemble Kalman filtering can adapt to nonlinear observation operators and is suitable for scientific research and complex application scenarios. The two implementation methods complement each other, and the overall algorithm framework has good scalability. Attached Figure Description
[0019] Figure 1 This is a flowchart of the irradiance spatial analysis field construction method in this invention; Figure 2 This is an overall flowchart of the irradiance spatial analysis field construction method in an embodiment of the present invention; Figure 3 This is a schematic diagram of the irradiance spatial analysis field construction system in this invention; In the diagram: 1. Data preprocessing module; 2. Error modeling module; 3. Data assimilation module; 4. Probability output module. Detailed Implementation
[0020] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0021] This invention discloses a method and system for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction. The core innovation lies in reconstructing the traditional irradiance spatial interpolation modeling problem into a multi-precision observation data assimilation problem, breaking through the limitations of traditional fixed-weight fusion that only outputs a single, definitive value. This invention distinguishes between the low-precision, time-varying observation characteristics of satellite-derived irradiance and the high-precision observation characteristics of ground stations, constructing a dynamic and adaptive multi-source error modeling mechanism. It achieves optimal statistical fusion of multi-source data based on a three-dimensional variational and ensemble Kalman filter dual data assimilation framework, and simultaneously quantifies spatiotemporal uncertainty during the fusion process, outputting probabilistic products such as confidence interval fields and over-threshold probability fields. This provides comprehensive data support for photovoltaic power generation probabilistic risk assessment, grid dispatching, and photovoltaic investment decisions. To facilitate understanding of the technical solution of this invention, the abbreviations and key terms involved are first defined.
[0022] Among them, Ensemble Kalman Filtering (EnKF) is a sequential data assimilation method based on Monte Carlo ensemble forecasting. It dynamically estimates the background error covariance matrix by the statistical dispersion of ensemble members, adapting to nonlinear dynamic system scenarios. Three-dimensional variational assimilation (3D-Var) is a data assimilation method based on variational theory. It solves for the optimal irradiance state estimate by minimizing the cost function containing the background and observation terms. Optimal linear unbiased estimation (BLUE) is the core criterion followed by the assimilation scheme of this invention. Under the condition of known error covariance and linear observation model, it achieves the estimation of error... Minimizing variance ensures the statistical optimality of the fusion result; the irradiance fusion analysis field is the core output product of this invention, which is gridded irradiance data obtained by solving the data assimilation framework, possessing strict statistical optimality and supporting error statistics information, unlike traditional simple interpolation fields; the background field specifically refers to the high spatial resolution, low precision time-varying irradiance gridded data retrieved from geostationary meteorological satellites, serving as the prior initial field for assimilation and fusion; uncertainty quantification refers to the quantitative characterization of the error range and fluctuation probability of each grid point value in the fusion analysis field, realizing the technical upgrade from deterministic values to probabilistic values.
[0023] The present invention will now be described in further detail with reference to the accompanying drawings: Example 1 See Figure 1 and Figure 2 In one embodiment of the present invention, a method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction is provided. The method adopts a progressive technical process, in which the output of the preceding step serves as the sole input for the subsequent step. Specifically, it includes multi-source data preprocessing, dynamic modeling of multi-source errors, optimal fusion of dual-frame data assimilation, uncertainty quantification, and output of probability products, as detailed below: Step 1: Collect satellite-inverted gridded irradiance data and ground station measured irradiance data for the target area. Perform anomaly control, validity screening and spatial matching resampling on the two types of multi-source data in sequence to obtain spatiotemporally aligned and qualified multi-source observation data. Specifically, visible light and infrared channel observation data from Himawari-8 / 9 and FY-4 series geostationary meteorological satellites are collected within the target area. Using existing mature radiative transfer inversion algorithms, gridded surface solar irradiance data are generated at 15-minute or hourly intervals. I sat (r,t) The data has a spatial resolution of 0.05°×0.05°, an equivalent spatial scale of approximately 5km, and fully covers the target area, exhibiting strong spatial continuity and high temporal update frequency. Simultaneously, minute-by-minute or hourly measured total horizontal irradiance data from all ground-based meteorological radiation observation stations within the target area were collected. I sta (r i ,t) ,in r i The instrument, which represents the geographical location of the i-th ground observation station, has an accuracy better than ±5% and high data accuracy. However, it suffers from the defects of sparse station distribution and poor regional coverage. The distance between national-level radiation stations can reach hundreds of kilometers.
[0024] After data acquisition, hierarchical quality control was performed: For satellite-derived irradiance data, outlier grid data that failed due to cloud obstruction or were marked as unusable were removed, along with invalid null values and distorted values. For ground station measured data, extreme value verification and time consistency verification were performed, removing extreme outlier data with irradiance values exceeding the solar constant. Simultaneously, the nighttime irradiance zeroing characteristic was verified, deleting non-zero outlier data at night. This completed the anomaly screening and validity filtering for both types of data. Finally, a bilinear interpolation algorithm was used to interpolate and match the global satellite irradiance grid data to the geographical locations of various ground stations, obtaining the equivalent satellite irradiance for each station. I sat (r i ,t) This method achieves precise spatiotemporal registration between satellite grid data and on-site measured data, ultimately obtaining standardized multi-source observation data that is spatiotemporally aligned and of acceptable quality, providing reliable data input for subsequent error modeling.
[0025] Step 2: Using multi-source observation data as input, distinguish the differentiated error characteristics of high-precision ground station observations and low-precision satellite inversion observations, construct fixed-parameter ground station observation error models and time-varying and space-varying satellite dynamic error models respectively, statistically analyze the observation bias and update the error variance in real time with cloud influence factors, and construct the global observation error covariance matrix. Specifically, this embodiment constructs a ground station error model with fixed parameters and a time-varying and space-varying satellite dynamic error model based on the differences in the accuracy of the observation sources, so as to accurately characterize the qualitative differences in errors.
[0026] Specifically, the modeling of ground station observation errors involves using measured data from ground stations as a high-precision observation benchmark. Errors originate solely from instrument noise and regional representativeness errors. The standard deviation of station observation errors is set based on the instrument calibration accuracy. s sta =5%× I sta The corresponding error variance is s sta 2 In this embodiment, the observation errors of each ground station are assumed to be independent and without spatial correlation. Based on this, a diagonal matrix form of the station observation error covariance matrix is constructed. R sta All off-diagonal elements of the matrix are set to zero, while only the diagonal elements correspond to the fixed error variance of each station.
[0027] Satellite inversion data is affected by atmospheric factors such as cloud cover, aerosol optical thickness, and solar zenith angle, resulting in errors that exhibit strong time- and space-varying characteristics. Therefore, a dynamic satellite error model is constructed, and the specific process is as follows: First, based on the spatially matched observation data, the deviation between the satellite equivalent irradiance and the measured irradiance at each station location is calculated to quantify the real-time deviation characteristics of the satellite data. The calculation formula is as follows: d(r i ,t)=I sat (r i ,t) I sta (r i ,t) in, I sat (r i ,t) The equivalent irradiance of the satellite at each station; I sta (r i,t) Gridded data of incident solar irradiance on the Earth's surface; Secondly, satellite atmospheric observation parameters are extracted, based on the satellite infrared window region brightness temperature T. 11 (r,t), visible light channel albedo A vis (r,t), constructing a cloud impact factor that can quantify changes in cloud cover using an empirical function: C(r,t)=f(T 11 ,A vis ) The value of this factor is in the range of [0,1], and the value increases synchronously with the increase of cloud cover in the region, accurately quantifying the degree of cloud interference at different times and locations. Finally, based on the historical sliding window deviation samples from the past 30 days, and constrained by the real-time cloud impact factor, deviation samples with similar historical cloud conditions and regions were stratified and filtered. The satellite observation error variance was then calculated in real-time using variance statistics, with the formula being σ. sat 2 (r,t)=Var[d(r,τ)|C(r,τ)≈C(r,t)] Where d(r,τ) represents the satellite-site irradiance deviation at historical time τ and location r; τ represents the historical observation time. By statistically analyzing historical similar operating conditions, the satellite error is adaptively updated according to atmospheric conditions.
[0028] After completing the modeling of the two types of errors, the global observation error covariance matrix is constructed by integrating them, and the real-time satellite error variance of each grid point is included. s sat 2 (r,t) Construct a diagonal matrix R sat Combined with the site error diagonal matrix R sta The global observation error covariance matrix R is integrated into a block-diagonal structure, completing the unified quantitative expression of multi-source error information and providing dynamic weight constraints for subsequent data fusion. The expression for the global observation error covariance matrix R is as follows:
[0029] Among them, R sat It is a diagonal matrix, with diagonal elements representing the grid points. s sat 2 (r,t) .
[0030] Step 3: Using satellite-inverted gridded irradiance data as the background field and ground station measured irradiance data as the true observation value, and constrained by the constructed global observation error covariance matrix, iterative optimal solution is obtained through a three-dimensional variational or ensemble Kalman filter data assimilation framework to obtain the global gridded irradiance optimal analysis field. Specifically, this invention reconstructs the irradiance fusion problem into a standard data assimilation problem, using satellite irradiance gridded data as the prior background field x. b The ground station measured data is the true observation value y. The global observation error covariance matrix R is constructed as the core constraint. Following the optimal linear unbiased estimation criterion, it provides two alternative assimilation schemes: three-dimensional variational and ensemble Kalman filtering, to adapt to different business scenarios.
[0031] Solution A is a three-dimensional variational (3D-Var) assimilation scheme, suitable for high-frequency production scenarios with limited computing resources. First, the fusion cost function is constructed, as shown in the following formula: J(x) = ½(xx) b ) T B -1 (x x b )+½(Hx-y) T R -1 (Hx-y) Where x is the analytical field to be solved, x b Background field, i.e., satellite-retrieved irradiance gridded data I sat B is the background error covariance matrix, representing the spatial correlation error structure of the satellite inversion field; H is the observation operator, i.e., the spatial interpolation operator from the grid space to the station location; y is the observation vector, containing the measured values I of all stations. sta R is the observation error covariance matrix.
[0032] Subsequently, the background error covariance matrix was modeled, and recursive filtering or diffusion operators were used to reduce computational complexity. The matrix decomposition formula is as follows: B=Σ b ×C×Σ b ; Where, Σ b Let C be the satellite error variance diagonal matrix, and let C be the spatial correlation matrix. The elements of C are calculated using the Gaussian correlation function, as shown in the following formula:
[0033] in, L For the relevant length scale (e.g., 50km).
[0034] The matrix elements are calculated using the Gaussian correlation function, and the spatial correlation length scale is set to L=50km to characterize the spatial correlation features of satellite errors. Finally, the cost function is iteratively minimized using either the steepest descent method or the conjugate gradient method to seek the optimal balance between the spatial structure of the satellite background field and the high-precision true values measured at the stations, thus obtaining the global gridded optimal irradiance analysis field x. a The fusion weights are adaptively determined by the relative magnitudes of background error and observation error, avoiding the empirical shortcomings of traditional fixed weights. The optimal analysis field x...a The formula is as follows:
[0035] Scheme B is an ensemble Kalman filter (EnKF) assimilation scheme, suitable for scientific research and photovoltaic risk assessment scenarios with high requirements for uncertainty quantification accuracy. Firstly, it is based on the satellite background field x... b Add a random perturbation conforming to a Gaussian distribution N(0,B) to the background error covariance matrix B to generate N. e =50 background field members x b (k); secondly, the background error covariance matrix P is solved based on set statistical characteristics. b Combining the global observation error covariance matrix R, the formula K=P b H T HP b H T +R) -1 Calculate the Kalman gain matrix to achieve adaptive weight allocation for the fusion of the two types of data sources; then perform state updates for each set member one by one, with the update formula being x. a (k) =x b (k) +K(y (k) -Hx b (k) Finally, the average value of all set members is taken as the optimal irradiance analysis field x. a Meanwhile, it retains complete error statistics by relying on the discreteness of set members. Compared to 3D-Var, EnKF can characterize dynamic flow-dependent error features and is suitable for nonlinear observation scenarios, but its computational cost is relatively higher.
[0036] The optimal estimate of the analysis field is taken as the ensemble average:
[0037] The analysis error covariance matrix is estimated from the dispersion of the set members:
[0038] The comparison and selection of scheme A and scheme B in this embodiment is shown in Table 1.
[0039] Table 1 Comparison and Selection of Scheme A and Scheme B
[0040] Step 4: Based on the error statistics characteristics generated by the iterative optimal solution process of the assimilation framework, extract the spatiotemporal error variance distribution of the global gridded irradiance optimal analysis field, combine it with the Gaussian distribution probability model to complete the uncertainty quantification, generate two types of probabilistic analysis fields: confidence interval field and over-threshold probability field, and complete the construction of the irradiance spatial analysis field.
[0041] Specifically, based on the error statistics of the assimilation process, the spatiotemporal error variance distribution is extracted, and two types of core probabilistic products are generated by combining them with a Gaussian probability model. First, the analysis error variance field extraction is completed, with different solution methods corresponding to the two assimilation schemes: When using the 3D-Var scheme, the analysis error covariance matrix is approximated by the cost function and the inverse of the Hessian matrix, as shown in the formula P. a ≈ (B -1 + H T R -1 H) -1 Extracting the diagonal elements of the matrix yields the variance of the analysis error at each grid point. s a 2 (r) ; When using the EnKF scheme, the dispersion of set members is calculated directly using the formula: s a 2 (r)=Var [ x a (k) (r) [This allows for the rapid acquisition of the global error variance distribution.]
[0042] Based on the optimal analysis field x a (r) and the standard deviation of error σ a (r), combining the characteristics of the standard normal distribution to generate probability products: First, the confidence interval field, given a 95% conventional confidence level α, is generated through the formula... CI 1-α (r)=[x a (r)-z 1-α / 2 ·s a (r), x a (r)+ z 1-α / 2 ·s a (r)] Calculate the reliable fluctuation range of irradiance at each grid point, z 1-α / 2 The first is the two-sided quantile of the standard normal distribution; the second is the over-threshold probability field, where the irradiance threshold I is set according to the photovoltaic operation requirements. thrBased on the cumulative distribution function Φ of the standard normal distribution, the probability of irradiance exceeding the threshold at each grid point is calculated, and the risk of extreme irradiance fluctuations is accurately quantified.
[0043] For a given irradiance threshold I thr (For example, the lower limit of irradiance used to determine whether a photovoltaic panel is activated), calculate the probability that the irradiance at each grid point exceeds this threshold:
[0044] Where Φ is the cumulative distribution function of the standard normal distribution.
[0045] Finally, it outputs a series of products such as optimal irradiance field, error standard deviation field, confidence interval field, and over-threshold probability field in NetCDF and GeoTIFF formats, completing the construction of a high-precision, quantifiable, and probabilistic irradiance spatial analysis field.
[0046] There are multiple alternative implementation methods in this embodiment, all of which fall within the protection scope of this invention.
[0047] Among the alternatives to observation error modeling, Alternative A: Replace the empirical stratification method based on cloud impact factors with a Bayesian hierarchical model. Treating error parameters as random variables and estimating their posterior distribution yields more rigorous statistical inferences. Alternative B: Employ machine learning methods (such as random forests or gradient boosting trees) to directly predict the error variance of satellite-retrieved irradiance using cloud parameters, solar zenith angle, and aerosol optical thickness as input features, replacing the statistical stratification-based error model.
[0048] In alternative sources of the background field, besides geostationary satellite inversion products, the background field can also be derived from the short-term forecast irradiance field of numerical weather prediction models (such as the radiation forecast field of ECMWF). In this case, the background error covariance matrix should be constructed from the statistical characteristics of the model forecast error.
[0049] Among the alternatives to the assimilation algorithm, a hybrid approach can be adopted: a hybrid approach of 3D-Var and EnKF (Hybrid3D-EnKF) can be used, which linearly combines the covariance of static climatological background error with the covariance of flow-dependent ensemble estimation according to certain weights, taking into account both computational efficiency and the ability to characterize nonlinear error structures.
[0050] In an alternative to uncertainty quantification, the explicit calculation of the analysis error covariance can be omitted. Instead, a conformal prediction framework is used to construct calibration confidence intervals for the predicted values of the analysis field based on historical validation samples. This method does not rely on the Gaussian error assumption and has a more robust statistical coverage guarantee.
[0051] This embodiment explicitly defines satellite data as the "background field" and site data as "high-precision observations," rather than using equal weighting as in traditional methods. Specifically, for satellite-derived irradiance data, it constructs an error variance estimation method that changes in real-time with cloud influence factors, enabling the fusion weights to have spatiotemporal adaptive capabilities. While solving for the optimal estimate, it outputs a complete spatiotemporal uncertainty distribution by analyzing the error covariance matrix or ensemble dispersion, unlike the "truncated" uncertainty information processing in existing technologies. The analysis transforms error statistics into confidence interval fields and over-threshold probability fields, directly serving the probabilistic risk assessment of photovoltaic power generation.
[0052] Example 2 according to Figure 3 As shown in the figure, this embodiment also provides a system for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction, including a data preprocessing module 1, an error modeling module 2, a data assimilation module 3, and a probability output module 4, with the following specific functions: Data preprocessing module 1 is used to collect satellite-inverted gridded irradiance data and ground station measured irradiance data of the target area. It sequentially performs anomaly quality control, validity screening, and spatial matching resampling on the two types of multi-source data to obtain spatiotemporally aligned and qualified multi-source observation data. It is responsible for collecting visible light and infrared channel data from geostationary meteorological satellites and measured irradiance data from ground radiation stations in the target area. It performs targeted quality control on satellite data and station data respectively, removes abnormal and invalid data, and completes spatial matching resampling of satellite gridded data and station locations through bilinear interpolation. It outputs spatiotemporally aligned and qualified standardized multi-source observation data, providing compliant input data for backend error modeling and data assimilation.
[0053] Error modeling module 2 is used to distinguish the differentiated error characteristics of high-precision ground station observations and low-precision satellite inversion observations from multi-source observation data. It constructs a fixed-parameter ground station observation error model and a time-varying and space-varying satellite dynamic error model, respectively. It statistically analyzes the observation bias and updates the error variance in real time with the cloud influence factor to construct a global observation error covariance matrix. This module constructs a fixed-parameter ground station observation error model and a time-varying and space-varying satellite dynamic error model, automatically calculates the satellite-site irradiance deviation, cloud influence factor, and real-time satellite error variance, and integrates them to generate a block-diagonal global observation error covariance matrix. This realizes the dynamic quantification and unified modeling of the qualitative differences in multi-source observation errors and provides adaptive weight constraints for the data assimilation process.
[0054] Data assimilation module 3 uses satellite-retrieved gridded irradiance data as the background field and ground-based measured irradiance data as the true observation value. Constrained by a constructed global observation error covariance matrix, it iteratively optimizes the solution using a 3D variational or ensemble Kalman filter data assimilation framework to obtain the optimal global gridded irradiance analysis field. This module can automatically adapt to or manually switch between 3D-Var and EnKF assimilation schemes depending on the operating scenario. It performs operations such as cost function construction, error matrix modeling, iterative optimization, and ensemble updating, outputting the optimal global gridded irradiance analysis field. Simultaneously, it fully preserves statistical information such as background error, observation error, and ensemble dispersion during the assimilation process, providing data support for uncertainty quantification.
[0055] The probability output module 4 is used to extract the spatiotemporal error variance distribution of the global gridded irradiance optimal analysis field from the error statistical characteristics generated during the iterative optimal solution process based on the assimilation framework. It then combines this with a Gaussian probability model to quantify uncertainty, generating two types of probabilistic analysis fields: a confidence interval field and a threshold probability field, thus completing the construction of the irradiance spatial analysis field. This module can adaptively match the error variance extraction algorithm according to the assimilation scheme type, calculate and generate the confidence interval field and threshold probability field based on the Gaussian distribution model, and uniformly output deterministic and probabilistic irradiance products in a standardized format. This upgrades the output from single numerical values to probabilistic risk information, supporting downstream photovoltaic power generation probability prediction, grid dispatch, and photovoltaic investment decisions.
[0056] In summary, this invention provides a method and system for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction, achieving adaptive optimal fusion of multi-source observations: by constructing a dynamic observation error covariance matrix, the fusion weights change in real time with atmospheric conditions such as cloud cover and aerosols, overcoming the empirical and blind nature of traditional fixed-weight methods and theoretically guaranteeing optimality under the BLUE criterion. The system introduces uncertainty quantification in the irradiance analysis field construction: through the analysis error covariance matrix or set discreteness of the assimilation framework, it provides a spatiotemporal uncertainty distribution with physical meaning and statistical consistency, overcoming the fatal flaw of existing deterministic methods that "only provide a single number." It directly supports probabilistic risk assessment: the output confidence interval and over-threshold probability field can be directly input into downstream photovoltaic power generation probabilistic prediction models, providing complete uncertainty information for grid dispatch and investment decisions, improving the scientific nature and robustness of decision-making. The algorithm framework has good scalability: the 3D-Var scheme has high computational efficiency and is suitable for operational high-frequency production; the EnKF scheme can handle nonlinear observation operators and is suitable for scientific research and complex scenarios. The two schemes can complement each other.
[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction, characterized in that, include: Satellite-inverted gridded irradiance data and ground-station measured irradiance data of the target area were collected. Anomaly control, validity screening and spatial matching resampling were performed on the two types of multi-source data in sequence to obtain spatiotemporally aligned and qualified multi-source observation data. Using multi-source observation data as input, the differential error characteristics of high-precision ground station observations and low-precision satellite inversion observations are distinguished. Fixed parameter ground station observation error models and time-varying and space-varying satellite dynamic error models are constructed respectively. The observation bias is statistically analyzed and the error variance is updated in real time by combining cloud influence factors to construct the global observation error covariance matrix. Using satellite-inverted gridded irradiance data as the background field and ground station measured irradiance data as the true observation value, and constrained by the constructed global observation error covariance matrix, the optimal analytical field of global gridded irradiance is obtained by iterative optimization through a three-dimensional variational or ensemble Kalman filter data assimilation framework. Based on the error statistics characteristics generated by the iterative optimal solution process of the assimilation framework, the spatiotemporal error variance distribution of the global gridded irradiance optimal analysis field is extracted. The uncertainty is quantified by combining the Gaussian distribution probability model, generating two types of probabilistic analysis fields: confidence interval field and over-threshold probability field, thus completing the construction of the irradiance spatial analysis field.
2. The method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction according to claim 1, characterized in that, The specific process of collecting satellite-inverted gridded irradiance data and ground-station measured irradiance data for the target area, and sequentially performing anomaly control, validity screening, and spatial matching resampling on the two types of multi-source data to obtain spatiotemporally aligned, high-quality multi-source observation data includes: Collect visible light and infrared channel data from geostationary meteorological satellites within the target area, and obtain satellite irradiance grid data through a radiative transfer inversion algorithm; Collect measured irradiance data from ground-based radiation observation stations in the target area; Remove invalid grid points from satellite data that are obscured by clouds or fail to be retrieved; perform extreme value verification and nighttime zeroing time consistency verification on station data to remove outliers. By interpolating satellite grid data to the locations of each station using bilinear interpolation, spatial registration of satellite data and station measured data is achieved, resulting in standardized observation data with spatiotemporal alignment.
3. The method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction according to claim 1, characterized in that, The process of constructing the ground station observation error model is as follows: Based on the obtained ground station measured data, the standard deviation and variance of the station observation error are set according to the instrument calibration accuracy. Assuming that the observation errors of each ground station are independent, a diagonal matrix form of the station observation error covariance matrix is constructed.
4. The method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction according to claim 1, characterized in that, The process of constructing the satellite dynamic error model is as follows: Based on the spatially matched multi-source observation data, the deviation between the satellite inversion data and the measured data of each station location is calculated using the following formula: d(r i ,t)=I sat (r i ,t) I sta (r i ,t) in, I sat (r i ,t) The equivalent irradiance of the satellite at each station; I sta (r i ,t) Gridded data of incident solar irradiance on the Earth's surface; Based on satellite infrared window brightness temperature T 11 (r,t), visible light channel albedo A vis (r,t), constructing a cloud impact factor that can quantify changes in cloud cover using an empirical function: C(r,t)=f(T 11 ,A vis ) in, f This is an empirical function; the greater the cloud cover, the more... C The higher the value, C ∈[0,1]; Based on the deviation sample of the historical sliding window over the past 30 days, combined with the stratified statistics of real-time cloud impact factors, the variance of satellite observation error is calculated in real time. The calculation formula is as follows: σ sat 2 (r,t)=Var [ d(r,τ)∣C(r,τ)≈C(r,t) ] in, d(r,τ) The satellite-site irradiance deviation at historical time τ and location r; Real-time satellite error variance at each grid point σ sat 2 (r,t) Construct a diagonal matrix R sat Combined with the site error matrix R sta The global observation error covariance matrix R is obtained by integrating the components into a block-diagonal form.
5. The method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction according to claim 1, characterized in that, The specific process of obtaining the optimal analytical field of global gridded irradiance through iterative optimal solution using a three-dimensional variational assimilation framework is as follows: Using satellite data as background field x b The site data is represented by the observation vector y. Using the constructed global observation error covariance matrix R as a constraint, the fusion cost function is constructed as follows: J(x)=½(xx b ) T B -1 (xx) b )+½(Hx-y) T R -1 (Hx-y) Where x is the analytical field to be solved, x b Background field, i.e., satellite-retrieved irradiance gridded data I sat B is the background error covariance matrix, representing the spatial correlation error structure of the satellite inversion field; H is the observation operator, i.e., the spatial interpolation operator from the grid space to the station location; y is the observation vector, containing the measured values I of all stations. sta R is the observation error covariance matrix; Construct the background error covariance matrix: B = Σ b ×C×Σ b ; Where, Σ b Let C be the satellite error variance diagonal matrix, and let C be the spatial correlation matrix. The elements of C are calculated using the Gaussian correlation function, and the correlation length scale L is 50km. The optimal irradiance analysis field x is obtained by iteratively minimizing the cost function J(x) using either the steepest descent method or the conjugate gradient method. a .
6. The method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction according to claim 1, characterized in that, The specific process of obtaining the optimal global gridded irradiance analysis field through iterative optimization using an ensemble Kalman filter data assimilation framework is as follows: Based on satellite background field x b The constructed background error covariance matrix B is directed towards the satellite background field x. b Add a Gaussian perturbation conforming to N(0,B) to generate N e A group of background field members x b (k) , k=1,2,...,N e ; Background error covariance matrix P based on ensemble estimation b Calculate the Kalman gain matrix using the global observation error covariance matrix R, as shown in the following formula: K=P b H T (HP b H T +R) -1 Among them, P b H is the set of background error covariance matrices, which characterizes the spatial error discreteness of the satellite background field; T is the transpose matrix of the observation operator; H is the spatial interpolation observation operator used to map the grid field to the observation location of the station; R is the global observation error covariance matrix; Based on site observation data, perform status updates for each set member: x a (k) =x b (k) +K(y (k) -Hx b (k) ); The optimal analytical field x is obtained by taking the ensemble average. a The statistical characteristics of the analysis error are obtained by measuring the dispersion of the set members.
7. The method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction according to claim 1, characterized in that, The three-dimensional variational assimilation scheme achieves efficient solution through multi-source observation data and global observation error covariance matrix, and is used in operational irradiance field production scenarios with limited computational resources. The ensemble Kalman filter data assimilation scheme achieves high-precision fusion based on dynamic error characteristics, and is used in photovoltaic risk assessment research scenarios where high precision in uncertainty quantification is required.
8. The method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction according to claim 1, characterized in that, Based on the error statistics generated by the iterative optimal solution process of the assimilation framework, the specific process of extracting the spatiotemporal error variance distribution of the global gridded irradiance optimal analysis field includes: When using a three-dimensional variational assimilation scheme, the formula for approximating the analysis error covariance matrix using the inverse Hessian matrix is as follows: P a ≈ (B -1 + H T R -1 H) -1 Where B is the background error covariance matrix, representing the spatial error characteristics of the satellite-retrieved background field; the optimal estimation error covariance matrix P is extracted. a The diagonal elements provide the variance of the analysis error at each grid point. σ a 2 (r) ; When using an ensemble Kalman filter data assimilation scheme, the variance of the analysis error at each grid point is directly calculated using the discreteness of the ensemble members, as shown in the following formula: σ a 2 (r)=Var [ x a (k) (r) ] in, x a (k) (r) This is the assimilated irradiance analysis field of the k-th set member at spatial location r within the Kalman filter data assimilation framework.
9. The method for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction according to claim 1, characterized in that, Uncertainty quantification is achieved by combining the Gaussian distribution probability model, generating two types of probabilistic analysis fields: confidence interval field and over-threshold probability field. Specifically, these include: Confidence interval calculation: Given a confidence level α, based on the quantiles of the standard normal distribution, the grid-point confidence interval is obtained, as shown in the following formula: CI 1-α (r)=[x a (r)-z 1-α / 2 ·σ a (r), x a (r)+z 1-α / 2 ·σ a (r)] in, x a (r) This is the estimated value of the optimal irradiance analysis field at spatial location r after assimilation; z 1-α / 2 These are the two-tailed quantiles corresponding to the standard normal distribution; Calculation of probability field exceeding threshold: setting an irradiance threshold I thr Based on the cumulative distribution function Φ of the standard normal distribution, the probability of grid point irradiance exceeding the threshold is calculated, and the probabilistic product output is completed.
10. A system for constructing a spatial analysis field of irradiance for photovoltaic resource assessment and prediction, characterized in that, include: The data preprocessing module is used to collect satellite-inverted gridded irradiance data and ground station measured irradiance data of the target area. It sequentially performs anomaly quality control, validity screening and spatial matching resampling on the two types of multi-source data to obtain spatiotemporally aligned and qualified multi-source observation data. The error modeling module is used to take multi-source observation data as input, distinguish the differentiated error characteristics of high-precision ground station observations and low-precision satellite inversion observations, construct a ground station observation error model with fixed parameters and a time-varying and space-varying satellite dynamic error model, respectively, statistically analyze the observation bias and update the error variance in real time with cloud influence factors, and construct the global observation error covariance matrix. The data assimilation module is used to obtain the optimal analysis field of global gridded irradiance by taking satellite-inverted gridded irradiance data as the background field and ground station measured irradiance data as the observation true value. It uses the constructed global observation error covariance matrix as a constraint and iteratively solves the optimal solution through a three-dimensional variational or ensemble Kalman filter data assimilation framework. The probability output module is used to extract the spatiotemporal error variance distribution of the global gridded irradiance optimal analysis field by extracting the error statistical characteristics generated by the iterative optimal solution process based on the assimilation framework. It combines the Gaussian distribution probability model to complete the uncertainty quantification and generate two types of probabilistic analysis fields: confidence interval field and over-threshold probability field, thus completing the construction of the irradiance spatial analysis field.
