High spatio-temporal resolution grid wind speed correction method based on error dynamic change

Abstract

The application discloses a high space-time resolution grid wind speed correction method based on error dynamic change, belongs to the technical field of wind speed correction, and comprises the following steps: obtaining initial mode predicted wind speed and initial real wind speed; calculating sliding average mode predicted wind speed sequences and actual error sequences of each grid point under different sliding window steps and different data set lengths respectively, and forming a parameter multimodal training set; establishing a nonlinear fitting relationship between the sliding average mode predicted wind speed sequences and the fitting error sequences; introducing a determination coefficient to screen the optimal error correction model of each grid point; estimating the wind speed error estimation value of the corresponding grid point based on the optimal error correction model, correcting the original wind speed prediction value of the corresponding grid point, and calculating the final wind speed prediction value of the corresponding grid point. Therefore, the application can accurately depict and correct the dynamic change and nonlinear error in the original wind speed prediction value, and retain the original spatial distribution characteristics of the real wind speed.

Description

Technical Field

[0001] This invention relates to a high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes, belonging to the technical field of wind speed correction. Background Technology

[0002] In fields such as wind farm power forecasting and power grid dispatch management, predictions of key parameters (such as wind speed) based on numerical model systems are the core basis for decision-making. However, inherent model errors and initial value errors cause systematic biases in the directly output prediction results, which must be corrected to meet the requirements of accurate applications. For example, in wind power forecasting, small errors in wind speed prediction can be amplified into significant deviations in power generation prediction, directly affecting electricity market trading decisions, stable grid operation, and the economic benefits of wind farms.

[0003] Existing error correction techniques are mainly divided into two categories:

[0004] The first category consists of traditional correction methods based on statistical assumptions, such as sliding bias correction and Kalman filtering. The core principle of these methods is based on the statistical assumption of "error consistency," which assumes that the error characteristics exhibited by the numerical model in historical periods remain unchanged in future prediction periods. Therefore, they statistically determine the distribution or magnitude of errors in historical samples and directly apply this as a constant or a simple linear relationship to future predictions.

[0005] The second category: black-box correction methods based on complex machine learning. These methods utilize models such as deep neural networks and long short-term memory networks, taking a large amount of historical pattern prediction data and real-world data as input to train an end-to-end mapping model. The core principle of this type of method is to rely on the model's powerful nonlinear fitting ability to implicitly learn the complex mapping relationship from predicted values ​​to real-world values.

[0006] Both of the aforementioned technical approaches have revealed serious limitations when addressing the correction problem of high-resolution gridded wind speed predictions:

[0007] For the first type of traditional method:

[0008] (1) The basic assumptions are invalid: The target system (such as the atmosphere and wind resource field) is a typical non-stationary dynamic system. The prediction error of key parameters is affected by a variety of factors such as the evolution of the system state and the characteristics of the local environment, and has significant dynamic time-varying and nonlinear characteristics. For example, the relationship between error and prediction value varies greatly at different spatial locations. The static assumption of "error consistency" is difficult to hold in reality, which leads to a sharp decline in the correction effect of such methods during system state transitions or extreme processes, and even introduces the opposite effect.

[0009] (2) Unable to characterize complex relationships: Simple linear or statistical models cannot effectively capture and express the above-mentioned complex nonlinear error dynamics, and the correction accuracy has a ceiling.

[0010] For the second type of machine learning method:

[0011] (1) Data and computing power bottlenecks: High-quality gridded historical data acquisition is costly, and deep learning models require massive amounts of data for training, making it difficult to deploy and apply in scenarios with limited historical data or to new models quickly.

[0012] (2) Low computational efficiency and high deployment cost. If a unified complex model is trained to cover a massive number of grid points, the model will be too large, computation will be time-consuming and deployment will be difficult; if a model is trained separately for each grid point, the problem of exponential growth of computing resources will be faced, which will not meet the timeliness requirements of business forecasting.

[0013] (3) Smoothing of spatial features, loss of local details. When a unified global model is corrected, it is easy to smooth out localized and subtle spatial distribution features, causing the corrected data field to lose the real local change information and become too smooth, which is particularly fatal in areas with significant local characteristics.

[0014] (4) Poor interpretability affects the credibility of decision-making. As a "black box" model, its correction process lacks clear physical or statistical significance, making it difficult for domain experts and decision-makers to understand and trust its correction results, which is not conducive to its application in actual business management and decision-making. Summary of the Invention

[0015] The purpose of this invention is to provide a high spatiotemporal resolution grid-based wind speed correction method based on dynamic error changes. The core technical problem to be solved is: how to accurately characterize and correct the dynamic and nonlinear errors in the original wind speed prediction values ​​of grid points g output by numerical models without relying on a large amount of historical data and high computational costs, while preserving the original spatial distribution characteristics of the initial actual wind speed, so that the correction process is more in line with the physical nature of atmospheric motion and greatly improves the accuracy of wind speed correction.

[0016] To achieve the above-mentioned technical objectives, the present invention will adopt the following technical solution:

[0017] A high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes includes the following steps:

[0018] Obtain initial model to predict wind speed Compared with the initial actual wind speed ;

[0019] Predict wind speed based on the obtained initial model Compared with the initial actual wind speed For each grid point, a corresponding set of values ​​is constructed using different preset dataset lengths d; where: dataset length d is in days;

[0020] For each grid point, a different sliding window step size is used. Calculate the moving average sequence of the initial model-predicted wind speed and the initial actual wind speed in each value set to obtain the corresponding sliding window step size. Predicting wind speed sequences using a moving average model with a corresponding dataset length d. and moving average real wind speed sequence ;

[0021] For each grid point, predict the wind speed sequence based on the obtained moving average model. and moving average real wind speed sequence Construct the actual error sequence for each sliding window step size k and dataset length d. ;

[0022] Predict wind speed sequences based on the obtained moving average model and actual error sequence Construct a multimodal training set of parameters for any grid point g. ;

[0023] Based on the constructed multimodal training set of parameters corresponding to each grid point g. A nonlinear fitting model for the fitting error is constructed by establishing a nonlinear fitting relationship between the wind speed predicted by the moving average model and the actual error sequence, and then the step size of each moving window is calculated. Fitting error sequence of corresponding grid points g under dataset length d ;

[0024] Based on the calculated fitting error sequence of each grid point g and actual error sequence Calculate the determination coefficients of the corresponding fitting error nonlinear model. Then, the optimal error correction model for each grid point g is selected from the nonlinear models of the fitting errors. ;

[0025] The optimal error correction model based on the selected grid points g Estimate the wind speed error at the corresponding grid point g. The original wind speed prediction value for the corresponding grid point g. After correction, the final predicted wind speed value for the corresponding grid point g can be calculated. .

[0026] As a further improvement to the present invention, a moving average model predicts wind speed sequences. the i-th element in and the sliding average actual wind speed sequence the i-th element in are respectively expressed as:

[0027] ;

[0028] ;

[0029] In the above formulas: represents the j-th initial mode predicted wind speed in the value set ; represents the step size of the sliding window; represents the j-th initial actual wind speed in the value set .

[0030] As a further improvement of the present invention, in the sliding average mode predicted wind speed sequence and the sliding average actual wind speed sequence , the boundaries of the sliding window corresponding to each element are processed in the following manner:

[0031] when i≤k / 2, the corresponding boundary of the sliding window is [1,i+k / 2];

[0032] when i≥l-k / 2, the corresponding boundary of the sliding window is [i-k / 2,l];

[0033] when k / 2 <i<l-k / 2, the corresponding boundary of the sliding window is symmetric (i-k / 2,i+k / 2), and l represents the length of the initial sequence.

[0034] As a further improvement of the present invention, for any actual error in the actual error sequence is calculated by the following formula: .

[0035] As a further improvement of the present invention, the fitting error sequence is expressed as:

[0036] ;

[0037] ;

[0038] In the above formulas: represents the nonlinear model of fitting error; represents the sliding average mode predicted wind speed sequence under the sliding window step size and the data set length d; represents the quadratic term coefficient in the nonlinear fitting error model ; Represents the nonlinear model of fitting error The coefficient of the linear term in the equation; Represents the nonlinear model of fitting error The constant term in it.

[0039] As a further improvement to the present invention, the determination coefficient The time-varying weighting determination coefficients are calculated using the following formula:

[0040] ;

[0041] ;

[0042] = ;

[0043] In the above formula: Represents the actual error sequence The i-th actual error; Represents the fitting error sequence The i-th fitting error; This represents the average value of the actual error. For time decay weight, The time decay constant, This represents the effective sequence length of the moving average sequence when the sliding window step size is k and the dataset length is d.

[0044] As a further improvement to the present invention, the optimal error correction model The determination coefficients were obtained by comparing different sliding window step sizes k and dataset length d. Select the coefficient of determination Maximum fitting error nonlinear model The optimal error correction model for this grid point .

[0045] As a further improvement to the present invention, the final wind speed prediction value of grid point g... Calculated using the following formula: .

[0046] As a further improvement of the present invention, when new historical wind speed prediction data and actual wind speed data are accumulated, the parameter multimodal training set of each grid point g is updated to obtain the updated parameter multimodal training set. ;

[0047] Based on the updated parameter multimodal training set The coefficients of the quadratic term in the nonlinear model of the fitting error are obtained by refitting. coefficient of the first term and constant terms That is, to obtain the updated fitting error nonlinear model. ;

[0048] Based on the updated fitting error nonlinear model Recalculate the determination coefficients of the corresponding fitting error nonlinear model. To obtain the updated coefficient of determination ;

[0049] Based on the updated determination coefficient This allows for the selection of new optimal error correction models for each grid point. .

[0050] Another technical objective of the present invention is to provide an electronic device including a memory, a processor, and a computer program stored in the memory and running on the processor, the computer program being executed to perform the above-described high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes.

[0051] Based on the above technical solution, compared with the prior art, the technical breakthrough and beneficial effects brought about by the present invention are comprehensive and multi-layered, as specifically manifested as follows:

[0052] 1. A disruptive innovation in the principle of wind speed correction:

[0053] The grid-point wind speed correction method of this invention first constructs a sliding window step size for each grid point. And the multimodal training set of parameters with dataset length d And based on the constructed parameter multimodal training set By establishing a nonlinear model of the fitting error, the step size of each sliding window can be calculated. Fitting error sequence of corresponding grid points g under dataset length d Then, based on the fitting error sequence of the corresponding grid point g... and actual error sequence The determination coefficients of the corresponding fitting error nonlinear model are calculated, thereby selecting the optimal error correction model for the corresponding grid point g. And based on the selected optimal error correction model This allows us to estimate the wind speed error at the corresponding grid point g. The original wind speed prediction value for the corresponding grid point g. Corrections are then made. It is evident that this invention elevates traditional wind speed correction from static error elimination to dynamic error function prediction, fundamentally solving the scientific problem of error inconsistency. This makes the grid point wind speed correction process more consistent with the physical nature of atmospheric motion, resulting in a qualitative leap in the correction accuracy of grid point wind speeds.

[0054] 2. Lightweight Model:

[0055] The nonlinear fitting error model described in this invention uses a quadratic equation with the wind speed predicted by the moving average mode as the independent variable and the actual error sequence as the dependent variable as the core model. Its computational cost for fitting and prediction is lower than that for training and running deep learning models.

[0056] 3. Parallelization architecture:

[0057] The wind speed correction scheme for each grid point is independent of each other, making the entire scheme naturally suitable for parallel computing. It can utilize high-performance computing resources to simultaneously correct tens of thousands of grid points, with extremely fast computing speed, meeting the minute-level timeliness requirements of business forecasting.

[0058] 4. Transparency of the "white box":

[0059] The grid wind speed correction method disclosed in this invention uses a fitting error nonlinear model with a clear mathematical form and physical meaning (the actual error sequence changes nonlinearly with the wind speed sequence predicted by the moving average model), which is highly interpretable and greatly enhances the trust and acceptance of grid wind speed correction results by business personnel.

[0060] 5. Perfect preservation and refined treatment of spatial features:

[0061] By employing a strategy of constructing a nonlinear model of fitting error for each grid point, this invention can almost completely avoid the smoothing effect of the global model on spatial features. Each grid point can have a tailored optimal correction scheme based on its own geographical location and underlying surface characteristics, thereby preserving the original spatial details and localized features of the wind field to the greatest extent, especially in complex terrain areas where the advantages are particularly obvious.

[0062] 6. Low data requirements:

[0063] The fitting error nonlinear model described in this invention requires only a small amount of historical data (e.g., dataset length d is the most recent 7 to 30 days) to establish an effective model, which is friendly to new patterns or regions with scarce data.

[0064] 7. Highly adaptive dynamic learning:

[0065] The determination coefficient described in this invention By employing time-varying weighted coefficients, the concept of error memory time is quantified into a mathematical model. The recent fitting accuracy of the fitting error nonlinear model is considered. The fitting error nonlinear model can be updated regularly (e.g., daily) using the latest data to continuously track the evolution of the weather and climate system. It has strong online learning and adaptive capabilities, ensuring the stability and robustness of long-term applications.

[0066] 8. Low technical barriers and low deployment costs:

[0067] The grid wind speed correction method described in this invention has a clear principle, is simple to implement, does not rely on expensive AI training platforms and complex parameter tuning, and is easy to quickly promote and deploy in business units at all levels, thus having extremely high commercial value and strategic significance. Attached Figure Description

[0068] Figure 1 : Overall flowchart of the method of the present invention.

[0069] Figure 2 The root mean square error of hourly 10m wind speed prediction for a certain region in 2024 by the CMA-MESO model before and after correction. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and should not be regarded as any limitation on the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise specifically stated, the relative arrangement, expressions, and values ​​of components and steps set forth in these embodiments do not limit the scope of the present invention. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0071] The high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes described in this invention abandons the linear assumption of "directly applying historical errors to the future." Instead, it dynamically establishes a nonlinear functional relationship between the model-predicted wind speed (i.e., the moving average model-predicted wind speed sequence) and the prediction error (i.e., the fitting error) for each grid point, and uses this relationship to directly correct the original wind speed prediction values ​​at future times. By introducing multi-time-window moving averages, multi-dataset length variations, and an optimal model selection mechanism that considers time-varying weights, it ensures that the model can learn and adapt to the characteristics of recent Earth system processes, thereby achieving dynamic and nonlinear error correction. Figure 1 As shown, the specific steps are as follows:

[0072] Step 1, Data Acquisition and Preprocessing:

[0073] The model-predicted wind speed (1-hour interval, 3km x 3km spatial resolution) from high-precision numerical model prediction data (such as prediction data output by numerical models like CMA-MESO) and the actual wind speed from the actual grid data output by the data assimilation system (such as the CLDAS assimilation system) are obtained respectively.

[0074] The obtained model-predicted and observed wind speeds are preprocessed to unify the spatial resolution of the predicted and observed data. Specifically, the obtained observed wind speeds are spatially interpolated using a double-spline function to achieve the same spatial resolution as the numerical model, thus unifying the spatial resolution of the model-predicted and observed wind speeds and obtaining the initial model-predicted wind speed. Compared with the initial actual wind speed ,in: , l is the initial sequence length, and its index j takes the values ​​1, 2, 3, ... l.

[0075] Step 2, Construction of multimodal training set:

[0076] The multimodal training set (also known as the parametric multimodal training set) described in this invention is constructed through the following steps:

[0077] Step 2.1, set the possible values ​​for the sliding window step size and the dataset length:

[0078] For each grid point, predict the wind speed based on the obtained initial model. Compared with the initial actual wind speed By constructing corresponding value sets with different preset dataset lengths d, where the dataset length d is in days (e.g., d = 7, 8, 9... 30 days), different dataset lengths d can form a dataset length set D. Different preset sliding window steps are used... Construct a corresponding set of values, for example, k = 3, 5, 7, 9, 11 hours, with different sliding window step sizes. A set K of sliding window step sizes can be constructed.

[0079] Step 2.2, calculate the moving average sequence for different sliding window step sizes and dataset lengths:

[0080] Calculate the sliding window step size The moving average sequences of the initial model-predicted wind speed and the initial actual wind speed in the set of values ​​for dataset length d are used to obtain different sliding window step sizes. Predicting wind speed sequences using a moving average model with a corresponding dataset length d. and moving average real wind speed sequence .

[0081] Therefore, for each sliding window step size and data set length , a set of moving average pattern predicted wind speed sequences can be obtained and moving average actual wind speed sequences , wherein represents the effective sequence length of moving average sequences (including moving average pattern predicted wind speed sequences and moving average actual wind speed sequences) when the sliding window step size is k and the data set length is d, and the value of the sequence index i is .

[0082] The i-th element in the moving average pattern predicted wind speed sequence under the sliding window step size and data set length d is denoted as : , which represents the moving average sequence of the initial pattern predicted wind speed in the value set under the i-th sliding window; the i-th element in the actual wind speed sequence under the sliding window step size and data set length d is denoted as : ; which represents the moving average sequence of the initial actual wind speed in the value set under the i-th sliding window.

[0083] In the moving average pattern predicted wind speed sequence and the moving average actual wind speed sequence , the boundary conditions of the sliding window corresponding to each element are processed as follows:

[0084] When i≤k / 2, the corresponding sliding window boundary is [1,i+k / 2];

[0085] When i≥l-k / 2, the corresponding sliding window boundary is [i-k / 2,l];

[0086] When k / 2 <i<l-k / 2, the corresponding sliding window boundary is symmetric (i-k / 2,i+k / 2), wherein l represents the initial sequence length.

[0087] Step 2.3: Calculate actual error sequences for different sliding window step sizes and data set lengths:

[0088] For the moving average pattern predicted wind speed sequence obtained in step 2.2 and the moving average actual wind speed sequence , calculate the actual error : , so as to obtain the actual error sequence under each sliding window step size k and data set length d .

[0089] Therefore, for grid point g, the wind speed sequence is predicted based on the moving average model obtained in step 2.2. and the actual error sequence obtained in step 2.3 This allows for the multimodal training of parameters for grid point g. , can be represented as:

[0090] .

[0091] In the above formula, K represents the set of all sliding window step sizes k; This represents the set of dataset lengths consisting of all dataset lengths d.

[0092] Step 3, construct the optimal grid point model library:

[0093] The purpose of this step is to construct an optimal nonlinear correction model for each grid point, thereby building an optimal model library for each grid point. This includes the following steps:

[0094] Step 3.1, initialize the optimal determination coefficients;

[0095] Initially, the optimal coefficient of determination is 0.

[0096] Step 3.2, Traverse the multimodal training set of parameters :

[0097] Traversing the multimodal training set of parameters Each element in the model is used to construct a nonlinear model of the fitting error. Specifically, for each grid point g, each sliding window step size k, and each dataset length d, wind speed sequences are predicted using a moving average model. As the independent variable, the actual error sequence Using the least squares method to fit a quadratic equation as the dependent variable, a nonlinear model of the fitting error can be obtained. Based on the nonlinear model of fitting error The fitting error sequence can be calculated. This invention directly uses the obtained fitting error nonlinear model As the fitting error sequence The computational model can be represented as follows:

[0098] ;

[0099] in, , , Let g be the polynomial coefficients of grid point g with respect to the sliding window step size k and dataset length d, where: Represents the nonlinear model of fitting error The coefficient of the quadratic term in the equation; Represents the nonlinear model of fitting error The coefficient of the linear term in the equation; Represents the nonlinear model of fitting error The constant term in it.

[0100] Step 3.3, calculate the coefficient of determination:

[0101] Calculate the nonlinear model for each fitting error The coefficient of determination, in this invention, is the coefficient of determination. The time-varying weighting determination coefficient is selected, and the calculation formula is as follows:

[0102] ;

[0103] In the above formula: This represents the actual error within each sliding window i, which corresponds to the actual error sequence under the sliding window step size k and the dataset length d. The i-th actual error; Represents the fitting error sequence The i-th fitting error is calculated based on the nonlinear model of fitting error. The average value of the actual error is represented by the actual error sequence. Calculated; The time decay weights are applied using an exponential decay method: , The time decay constant, This represents the effective sequence length of the moving average sequence when the sliding window step size is k and the dataset length is d, controlling the relative importance of recent data. = This indicates that the weight of the most recent data is approximately twice that of the data from d days ago.

[0104] The numerator of the coefficient of determination represents the time-weighted regression sum of squares, which is the variation of the predicted value of the fitted model (representing the wind speed fitting error calculated based on the nonlinear model of the fitting error), while the denominator represents the sum of the regression sum of squares and the residual sum of squares (representing the variation that the nonlinear model of the fitting error could not explain).

[0105] Step 3.4, Determination and Update of Coefficient of Decision:

[0106] The purpose of this step is to select the optimal error correction model for each grid point. And it will be updated, specifically:

[0107] Step 3.4.1, Determining the coefficient of determination:

[0108] Compare the determination coefficients of various fitting error nonlinear models under different sliding window step sizes k and dataset length d. Select the coefficient of determination The nonlinear model with the largest fitting error is used as the optimal error correction model for that grid point. Therefore, an optimal nonlinear correction model (i.e., the optimal error correction model) is determined for each grid point in the entire grid field. ), forming the global optimal correction model matrix :

[0109] .

[0110] Step 3.4.2, update the coefficient of determination:

[0111] When new historical wind speed forecast data and actual wind speed data are accumulated, the parameter multimodal training set for each grid point g is updated to obtain the updated parameter multimodal training set. Based on the updated parameter multimodal training set Refit all sliding window step sizes and the length d of all datasets Fitting error That is, to obtain the updated fitting error nonlinear model. Then, based on the updated fitting error nonlinear model Recalculate the coefficient of determination To obtain the updated coefficient of determination Based on the updated determination coefficient Screening for new optimal error correction models And update the optimal correction model matrix accordingly. .

[0112] Step 3.5, Output the optimal model: Output the obtained globally optimal correction model matrix. Outputting the data completes the construction of the optimal grid point model library.

[0113] Step 4, Dynamically correct the application:

[0114] Step 4.1, Input the wind speed forecast for the future time:

[0115] When making wind speed predictions for future moments, the grid wind speed prediction value corresponding to any uncorrected grid point g output by the numerical model (denoted as the original wind speed prediction value) is used. The data is input into the optimal model library. The optimal model library integrates the globally optimal correction model matrix. .

[0116] Step 4.2: Calculate the error estimation field using the model library.

[0117] Call the optimal error correction model corresponding to grid point g in the optimal model library. Based on the input raw wind speed prediction value The estimated value of the prediction error (denoted as the wind speed error estimate) can then be calculated. ).

[0118] Step 4.3, correct the original wind speed prediction error:

[0119] From the original wind speed forecast Subtract the estimated wind speed error from the middle. This allows for the prediction of the original wind speed value. Error correction is performed. The final wind speed prediction value after correction for grid point g. It can be represented as:

[0120] .

[0121] Finally, the corrected predicted wind speed is output. The corrected predicted wind speed is the final predicted wind speed value. .

[0122] The present invention also provides a storage medium, wherein the computer program stored in the storage medium executes the above-described high spatiotemporal resolution grid wind speed correction method based on dynamic error changes during runtime.

[0123] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-described high spatiotemporal resolution grid wind speed correction method based on dynamic error changes through the computer program.

[0124] The sequence numbers of the above embodiments of the present invention are merely for description and do not represent the superiority or inferiority of the embodiments. In the above embodiments of the present invention, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units can be a logical functional division, and in actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces; the indirect coupling or communication connection of units or modules can be electrical or other forms. The units described as separate components may or may not be physically separated; the components shown as units may or may not be physical units, that is, they can be located in one place or distributed across multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of this embodiment.

[0125] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part 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 the present invention. The aforementioned storage medium includes: USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, optical disks, and other media capable of storing program code.

[0126] The following is combined Figure 1 The flowchart shown illustrates the invention in detail, using the correction of the 10m model predicted wind speed for the next 12-35 hours in the 2024 CMA-MESO numerical model as a specific embodiment, so that those skilled in the art can implement the invention.

[0127] 1. Data preparation:

[0128] The 10m model-predicted wind speed for a certain region with a lead time of 12-35 hours was obtained from the meteorological data center, starting from 00UTC in 2024 (grid data, hourly temporal resolution, 3km spatial resolution).

[0129] Simultaneously acquire the real-time wind speed from the real-time grid data output by the CLDAS assimilation system for the corresponding time period and region.

[0130] By using a bispline function to spatially interpolate the actual wind speed output from the CLDAS assimilation system to the latitude and longitude grid points of the CMA-MESO numerical model, the spatial resolution of the two systems is unified, thereby obtaining the initial model-predicted wind speeds accordingly. Compared with the initial actual wind speed .

[0131] 2. Model building (performed independently for each grid point):

[0132] For all grid points with latitude and longitude coordinates within the corresponding region (30.53°-35.47°N, 116.03°-121.97°E, a total of 7680 grid points), hourly model-predicted wind speeds and actual wind speeds for the past 7, 8, 9...30 days are read to construct a set of values ​​for the corresponding dataset length d. Then, with sliding window steps of 3, 5, 7, 9, and 11 hours respectively, the moving average sequence of the initial model-predicted wind speed and the initial actual wind speed in each set of values ​​is calculated to obtain the corresponding sliding window step size. Predicting wind speed sequences using a moving average model with a corresponding dataset length d. and moving average real wind speed sequence .

[0133] Based on the least squares method, for each grid point g, a "predicted value (moving average model predicted wind speed sequence)" is established for each of the above 5 sliding window step sizes. Error (actual error sequence) The fitting error nonlinear model of “)” is used, and the fitting coefficients of each model are calculated. ,in: Represents the nonlinear model of fitting error The coefficient of the quadratic term in the equation; Represents the nonlinear model of fitting error The coefficient of the linear term in the equation; Represents the nonlinear model of fitting error The constant term in it.

[0134] Calculate the coefficients of determination for these nonlinear models of fitting error. For example, for a grid point (35.0°N, 115.5°E), the nonlinear model of the fitting error with a sliding window step size of 7 hours and a dataset length of 14 days has the highest determination coefficient (0.89). Therefore, this nonlinear model of the fitting error (i.e., its fitting coefficient) is recorded as the optimal error correction model for this grid point.

[0135] 3. Forecast Correction:

[0136] When the latest model-predicted wind speed for the next 12-35 hours is received from the CMA-MESO numerical model, the model-predicted wind speed corresponding to the grid point (35.0°N, 115.5°E) is selected as the original wind speed prediction value. .

[0137] The original wind speed prediction value Substituting the values ​​into the optimal error correction model, the estimated prediction error for the grid point (35.0°N, 115.5°E) can be calculated. : , where: k=7, d=14.

[0138] Calculate the final wind speed forecast : .

[0139] This operation is repeated for each grid point within the region, ultimately generating a corrected hourly grid wind speed forecast map for the next 12-35 hours. This map can be compared with the actual wind speed output from the CLDAS assimilation system to calculate the root mean square error before and after correction. Figure 2 Looking at the overall performance throughout the year, the root mean square error after correction decreased significantly compared to before correction. This indicates that the gridded wind speed correction method provided by this invention effectively eliminates systematic biases and significantly improves prediction / simulation accuracy. Furthermore, this correction strategy is robust and efficient across all four seasons, demonstrating good generalization ability.

[0140] Through the above implementation methods, dynamic nonlinear error correction for high-resolution grid wind speed prediction can be completed efficiently and with high quality. Those skilled in the art can implement the entire process of this invention using programming languages ​​such as Python, Fortran, and C++, combined with standard scientific computing libraries (such as NumPy for Python), based on the above description.

[0141] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes, characterized in that, Comprising the following steps: Obtain initial model to predict wind speed Compared with the initial actual wind speed ; Predict wind speed based on the obtained initial model Compared with the initial actual wind speed For each grid point, a corresponding set of values ​​is constructed using different preset dataset lengths d; where: dataset length d is in days; For each grid point, a different sliding window step size is used. Calculate the moving average sequence of the initial model-predicted wind speed and the initial actual wind speed in each value set to obtain the corresponding sliding window step size. Predicting wind speed sequences using a moving average model with a corresponding dataset length d. and moving average real wind speed sequence ; For each grid point, predict the wind speed sequence based on the obtained moving average model. and moving average real wind speed sequence Construct the actual error sequence for each sliding window step size k and dataset length d. ; Predict wind speed sequences based on the obtained moving average model and actual error sequence Construct a multimodal training set of parameters for any grid point g. ; Based on the constructed multimodal training set of parameters corresponding to each grid point g. A nonlinear fitting model for the fitting error is constructed by establishing a nonlinear fitting relationship between the wind speed predicted by the moving average model and the actual error sequence, and then the step size of each moving window is calculated. Fitting error sequence of corresponding grid points g under dataset length d ; Based on the calculated fitting error sequence of each grid point g and actual error sequence Calculate the determination coefficients of the corresponding fitting error nonlinear model. Then, the optimal error correction model for each grid point g is selected from the nonlinear models of the fitting errors. ; The optimal error correction model based on the selected grid points g Estimate the wind speed error at the corresponding grid point g. The original wind speed prediction value for the corresponding grid point g. After correction, the final predicted wind speed value for the corresponding grid point g can be calculated. .

2. The high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes according to claim 1, characterized in that, Moving average model predicts wind speed sequence The i-th element and moving average real-time wind speed sequence The i-th element They are represented as follows: ； ； In the above formula: This represents the predicted wind speed of the j-th initial model in the set of values. ; Indicates the step size of the sliding window; This represents the j-th initial actual wind speed in the set of values. .

3. The high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes according to claim 2, characterized in that, Moving average model predicts wind speed sequence and moving average real-time wind speed sequence In the middle, the boundaries of the sliding windows corresponding to each element are handled as follows: when i ≤ k / 2, the boundary of the corresponding sliding window is [1, i+k / 2]; when i ≥ l-k / 2, the boundary of the corresponding sliding window is [i-k / 2, l]; when k / 2 < i < l-k / 2, the boundary of the corresponding sliding window is symmetric (i-k / 2, i+k / 2), and l represents the length of the initial sequence.

4. The high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes according to claim 2 or 3, characterized in that, Actual error sequence In, any actual error Calculated using the following formula: .

5. The high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes according to claim 4, characterized in that, Fitting error sequence Represented as: ； ； In the above formula: This represents a nonlinear model of the fitting error; Indicates the sliding window step size Predicting wind speed sequences using moving average patterns with dataset length d; Represents the nonlinear model of fitting error The coefficient of the quadratic term in the equation; Represents the nonlinear model of fitting error The coefficient of the linear term in the equation; Represents the nonlinear model of fitting error The constant term in it.

6. The high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes according to claim 5, characterized in that, Coefficient of determination The time-varying weighting determination coefficients are calculated using the following formula: ； ； = ； In the above formula: Represents the actual error sequence The i-th actual error; Represents the fitting error sequence The i-th fitting error; This represents the average value of the actual error; For time decay weight, The time decay constant, This represents the effective sequence length of the moving average sequence when the sliding window step size is k and the dataset length is d.

7. The high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes according to claim 6, characterized in that, Optimal Error Correction Model The determination coefficients were obtained by comparing different sliding window step sizes k and dataset length d. Select the coefficient of determination Maximum fitting error nonlinear model The optimal error correction model for this grid point .

8. The high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes according to claim 7, characterized in that, Final wind speed prediction at grid point g Calculated using the following formula: .

9. The high spatiotemporal resolution gridded wind speed correction method based on dynamic error changes according to claim 6, characterized in that, When new historical wind speed forecast data and actual wind speed data are accumulated, the parameter multimodal training set for each grid point g is updated to obtain the updated parameter multimodal training set. ; Based on the updated parameter multimodal training set The coefficients of the quadratic term in the nonlinear model of the fitting error are obtained by refitting. coefficient of the first term and constant terms That is, to obtain the updated fitting error nonlinear model. ; Based on the updated fitting error nonlinear model Recalculate the determination coefficients of the corresponding fitting error nonlinear model. To obtain the updated coefficient of determination ; Based on the updated coefficient of determination This allows for the selection of new optimal error correction models for each grid point. .

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, the computer program runs to implement the high spatio-temporal resolution grid wind speed correction method based on error dynamic change described in any one of claims 1 to 9.

Images