A regional sub-seasonal wind energy resource prediction method based on dynamic climate model, electronic equipment, storage medium and computer program product

Abstract

The present invention discloses a regional sub-seasonal wind energy resource prediction method, electronic device, storage medium and computer program product based on a dynamic climate model, aiming to improve the accuracy of sub-seasonal prediction of wind energy resources, especially to reduce the error of the dynamic climate model in the prediction of wind speed at the hub height of the wind turbine. The method analyzes the deviation characteristics of the model output and uses quantile regression technology to correct the error to achieve accurate prediction of wind speed within the next 1 to 45 days. When the present invention is implemented, the prediction ability of the model for the wind speed in the target area and the deviation characteristics for different levels of wind speed at the sub-seasonal scale are first analyzed. Secondly, according to the deviation characteristics of the model for different levels of wind speed, a correction scheme that can effectively improve the model deviation is selected, thereby improving the prediction ability of regional sub-seasonal scale wind energy resources and improving the prediction accuracy of the future evolution of regional wind energy resources.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind energy resource prediction, and relates to the prediction of wind energy resources on a sub-seasonal time scale (the next 1 to 45 days). Specifically, it is a regional sub-seasonal wind energy resource percentile regression prediction and correction technology based on a dynamic climate model. By analyzing the historical observation data and model return data of wind speed at the hub height of wind turbines in the target prediction area, the quantile regression method is applied to correct the model deviation, thereby effectively improving its prediction accuracy and realizing accurate prediction of regional sub-seasonal wind energy resources. Background Art

[0002] Against the backdrop of global climate change and energy transition, the "dual carbon" goals (carbon peak and carbon neutrality) have necessitated a shift from traditional fossil energy to clean, renewable energy. Compared to traditional fossil fuels, renewable energy, particularly wind energy, is characterized by intermittent and fluctuating generation. Especially in the short term, fluctuations in wind speed can directly impact wind power generation capacity. In the future, the large-scale and high-proportion integration of wind power into the grid will inevitably pose serious challenges to the safe and stable operation of the grid. The safe and stable operation of energy systems, as well as fluctuations in energy supply and demand, are closely linked to climatic conditions. Especially during seasonal fluctuations and extreme weather conditions, the stability of power supply relies even more heavily on accurate forecasts of wind energy resources. Therefore, developing accurate short- to medium-term energy climate forecasts, especially of wind energy availability, has become a crucial component of energy security strategies.

[0003] Dynamical climate models are currently the most important technical tool for subseasonal forecasting. They simulate atmospheric dynamics and thermodynamics to predict future climate conditions. These models have achieved considerable success in predicting atmospheric circulation, temperature, precipitation, and other parameters, and are widely used in climate change research. However, current dynamical climate models have very limited forecasting capabilities for surface layer wind speeds. Even the most advanced global climate models currently offer forecasts with a precision of only approximately one week, and their performance decreases significantly with increasing forecast horizons. Forecasts beyond 10 days are no longer valuable for reference or application. Therefore, direct forecasts of surface layer wind speeds or average wind speed at wind turbine hub height on a subseasonal timescale (1-45 days in the future) based on climate models lack the accuracy needed to meet the requirements of the renewable energy power industry. With the increasing number of wind farms and the increasing proportion of wind power connected to the grid, model prediction errors will introduce potential risks and uncertainties to power system operations.

[0004] In summary, in the context of the current global energy transition, facing the growing scale of wind power grid integration and the demand for wind energy resource forecasting, the limitations of dynamic climate models in wind energy resource forecasting are becoming increasingly prominent. In particular, sub-seasonal wind speed forecasts face significant bias, leading to greater uncertainty in the dispatch and operation of wind power systems within the grid. Therefore, there is an urgent need to develop wind speed forecasting technologies that optimize dynamic models for average turbine hub height, improve the accuracy and service level of sub-seasonal regional wind energy forecasts, and thus meet the urgent need to ensure the safe and stable operation of energy systems. Summary of the Invention

[0005] (1) Purpose of the invention

[0006] In view of the limitations of existing dynamic climate models in wind energy resource forecasting, especially the limited ability to forecast wind speed at hub height on a sub-seasonal scale, the significant decrease in forecast accuracy over time, and the systematic deviation of the model for different levels of wind speed, in order to solve at least one of the above and other technical problems in the prior art, the present invention aims to propose a regional sub-seasonal wind energy resource forecasting method based on a dynamic climate model, an electronic device, a storage medium, and a computer program product. Although the direct forecasting ability of the climate dynamic model for wind speed is relatively low, the model deviation has certain regularity and distribution characteristics in different regions, which provides the possibility of model-based forecast correction, thereby further improving the regional wind speed forecast accuracy. For the target forecast area, the forecasting ability of the dynamic model for the average wind speed at hub height in the target area and the deviation characteristics of the model for different levels of wind speed in the sub-seasonal forecast are first analyzed. Secondly, based on the deviation characteristics of the target area model for different levels of wind speed, a correction scheme that can effectively improve the model deviation is selected, thereby improving the prediction ability of regional sub-seasonal wind energy resources, improving the prediction accuracy of the future evolution of regional wind energy resources, and enhancing the ability to ensure the safe and stable operation of the power grid.

[0007] (2) Technical solution

[0008] In order to achieve the purpose of the invention and solve the technical problems, the present invention adopts the following technical solutions:

[0009] The first object of the present invention is to provide a regional sub-seasonal wind energy resource forecasting method based on a dynamic climate model, which is used to improve the accuracy of wind energy resource forecasting, particularly to reduce the dynamic climate model error in the wind speed forecast at wind turbine hub height in the target forecast area, and to achieve accurate forecasting of wind energy resources within the sub-seasonal scale (the next 1 to 45 days). When implemented, the method comprises the following steps:

[0010] SS1. Acquisition and organization of historical wind speed observation data

[0011] Obtain historical wind speed observation data at the hub height of wind turbines from several representative wind farms in the target prediction area within a set time range. Remove outliers and missing data through cleaning and convert them into daily average wind speed data as the basic dataset for subsequent predictions.

[0012] SS2. Acquisition and organization of daily model output data

[0013] Obtain daily output data from a high-resolution dynamical climate model for the target prediction area within a set past time range, including daily rolling sub-seasonal scale forecast data for the next 1 to 45 days generated by feedback within the set past time range, and use spatial interpolation and time warping to match historical wind speed observations at least in space, altitude, and time;

[0014] SS3. Analysis of Bias Characteristics of Dynamical Climate Model Forecasts

[0015] Based on the collated historical wind speed observations and the corresponding daily model output data, the daily root mean square error (RMSE) was selected as the test and evaluation metric to assess the forecast capability and bias level of the dynamic climate model for wind speed at hub height at different lead times (LT) for the next 1 to 45 days on a subseasonal scale. Furthermore, the distribution characteristics of model bias were further analyzed for different lead times, wind farm sites, months, and / or wind speed levels.

[0016] SS4. Construction and Training of Quantile Regression Bias Correction Models

[0017] Based on the results of the analysis of the forecast bias characteristics of the dynamic climate model, the daily output data of the model within a set past time range is used as the independent variable, and the corresponding historical wind speed observation data is used as the dependent variable. The Quantile Regression (QR) method is selected and combined with the machine learning algorithm to establish and train quantile regression bias correction models for different forecast days, different wind farms, different months and / or different wind speed levels.

[0018] SS5. Error Correction and Application of Model Real-time Forecast Data

[0019] The original real-time forecast data of the dynamic climate model is obtained, and the corresponding quantile regression deviation correction model is selected according to its advance forecast days, target wind farm station, forecast month and / or forecast wind speed level. The quantile regression correction result is obtained by inputting the original real-time forecast data of the model into the selected model for error correction. The quantile regression correction result is weightedly fused with the original forecast result of the model to obtain the final sub-seasonal scale model-corrected wind speed forecast result of the target forecast area, providing decision support for wind energy resource prediction and actual operation and scheduling of wind farms.

[0020] The second invention object of the present invention is to provide an electronic device, comprising at least a processor and a memory communicatively connected to the processor, wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can execute the above-mentioned regional sub-seasonal wind energy resource prediction method based on the dynamic climate model.

[0021] The third invention object of the present invention is to provide a computer program product, including computer instructions, for executing the above-mentioned regional sub-seasonal wind energy resource prediction method based on dynamic climate model.

[0022] The fourth invention object of the present invention is to provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the regional sub-seasonal wind energy resource prediction method based on the dynamic climate model is implemented.

[0023] (3) Technical effects

[0024] Compared with the prior art, the present invention has the following beneficial and significant technical effects:

[0025] (1) The present invention fully considers the forecasting capability of the dynamic climate model for the average wind turbine hub height wind speed and the deviation characteristics of different wind speed levels. By adopting a model-based error correction technology, especially by introducing the quantile regression method and adopting a scheme of giving different regression models with different forecast durations, the error correction of the model direct output results can be effectively corrected for the forecast deviation of different wind speed levels. Through the comprehensive analysis of model return data and historical observation data, the present invention can significantly reduce the systematic error in the forecast, especially in the problem of low wind speeds (such as less than 6m / s) being too high and high wind speeds (such as more than 12m / s) being too low.

[0026] (2) Quantile regression technology has strong adaptability and robustness. Its significant advantage is that it avoids the problem of random disturbance term distribution assumption in traditional regression models and effectively handles heteroscedasticity. Since quantile regression can be performed at different quantile levels, the model can better reduce the impact of outliers on the prediction results and enhance the stability of the forecast model. In addition, using this method, quantile regression models can be established based on different forecast days, different wind farms and historical data of different months. This makes the present invention more adaptable and can perform accurate forecasts under diverse meteorological conditions.

[0027] (3) This paper establishes a model-based objective sub-seasonal forecast model for wind energy resources by correcting the errors in the direct output of the model. This model combines the simulation capabilities of the dynamic model with the error correction advantages of quantile regression. It can not only improve the accuracy of wind speed forecasts, but also more accurately predict wind speed evolution on a sub-seasonal scale of 1 to 45 days in the future. The application of this technology will significantly improve the operating efficiency of wind farms and regional power grids, meeting the energy and power industry's demand for high-precision forecasts of wind energy resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of the implementation of the regional sub-seasonal wind energy resource prediction method based on the dynamic climate model of the present invention;

[0029] Figure 2 Schematic diagram of the change in the average RMSE (vertical axis) of the selected nine wind power sites in October 2021 (yellow) and February 2022 (blue) as a function of the number of days of lead forecast LT (horizontal axis);

[0030] Figure 3 This is a schematic diagram of the deviation of the model forecast wind speed relative to the station observed wind speed in October 2021 and February 2022 under different lead forecast days LT. The horizontal axis in the figure is the observed wind speed range, and the vertical axis is the size of the model deviation;

[0031] Figure 4 Figure 2 shows the growth rate of the extended-period average RMSE of nine stations after corrections using the QR50 scheme and the control schemes LR and OTS relative to the CFS model forecast in (a) October 2022 and (b) October 2023. LR is the linear regression correction scheme, and OTS is the optimal TS score scheme.

[0032] Figure 5 Schematic diagram of the growth rate of the extended-period average RMSE of nine stations relative to the CFS model forecast after QR50, LR, and OTS corrections in (a) February 2023 and (b) February 2024. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the implementation of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in more detail below with reference to the drawings in the embodiments of the present invention. The described embodiments are part of the embodiments of the present invention, rather than all the embodiments. The embodiments described below with reference to the drawings are exemplary and are intended to be used to explain the present invention, and should not be understood as limiting the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0034] In response to the limitations of existing dynamic climate models in wind energy resource forecasting, especially the limited ability to predict wind speed on a sub-seasonal scale, the significant decrease in prediction accuracy over time, and the systematic deviation of the model for different levels of wind speed, the present invention proposes a regional sub-seasonal wind energy resource forecasting method, electronic equipment, storage medium and computer program product based on a dynamic climate model.

[0035] Example 1

[0036] As a specific example, Figure 1 The figure shows the implementation flow chart of the objective sub-seasonal prediction method of regional wind energy resources based on dynamic climate model of the present invention. Figure 1 As shown, the present invention is mainly divided into two stages according to the actual operation process of the technology, namely the preliminary data regularization and analysis stage and the modeling correction stage. The first stage includes the collation of the measured wind speed data within the available period of the selected typical wind farm station, the extraction and processing of the power model return data of the corresponding period, and the analysis of the model error characteristics. After integrating the required data, a suitable correction method is selected according to the distribution characteristics of the model error, and an implementation example based on percentile regression is specifically selected. This method can effectively reduce the deviation between the model forecast wind speed and the observed wind speed under different levels of wind speed thresholds. The method specifically includes the following steps when implemented:

[0037] SS1. Acquisition and organization of historical wind speed observation data

[0038] The historical wind speed observation data at the hub height of the wind turbines of several representative wind farms in the target prediction area are obtained within a set past time range. After cleaning, outliers and missing data are removed and converted into daily average wind speed data as the basic data set for subsequent predictions.

[0039] In some preferred embodiments, the set past time range can be the past 3-10 years. The historical wind speed observation data can be based on the wind farm station's wind tower, wind turbine SCADA system's measurement data, meteorological station, lidar and satellite remote sensing data, or can be obtained by fusion of these multi-source data. The time interval for wind speed observation is at the minute level, and when performing data cleaning processing on the wind speed observation data, an adaptive outlier removal algorithm is used to identify and remove outliers. At the same time, the observation days with a daily missing rate exceeding the set threshold are treated as missing for the entire day and the missing values ​​are supplemented by polynomial interpolation. According to the needs of sub-seasonal scale forecasting, the data of each observation day are averaged to obtain the daily average wind speed observation results, and finally a historical wind speed observation data set is generated.

[0040] SS2. Acquisition and organization of daily model output data

[0041] The daily output data of the model in the target prediction area of ​​the high-resolution dynamic climate model within the set past time range are obtained, including the daily return data within the set past time range and the daily rolling forecast data of the corresponding sub-seasonal scale for the next 1 to 45 days, and through spatial interpolation and time warping, they are matched with the historical wind speed observation data at least in space, height and time.

[0042] In some preferred embodiments, the forecast duration of the selected dynamic climate model must be equal to or greater than 45 days to ensure that the model can provide meteorological information within a sufficient time range and support sub-seasonal wind energy resource forecasts, and the daily report data of the model contains multiple meteorological variable parameters and at least includes latitudinal wind speed and longitudinal wind speed at a height of 10 meters, ground pressure, potential heights of multiple isobaric layers (such as 925hPa, 850hPa, 700hPa, 500hPa), air temperature, latitudinal wind speed and longitudinal wind speed, and the model data is high-resolution grid data containing geographic longitude and latitude information to facilitate subsequent spatial interpolation and matching, thereby ensuring that the model output is consistent with the wind farm data in the target prediction area.

[0043] Further preferably, when arranging the daily output data of the model: first, the height of each isobaric layer from the ground is calculated by the air temperature and potential height of each isobaric layer and combined with the ground pressure; then, the longitude and latitude wind speeds of each isobaric layer and the longitude and latitude wind speeds at a height of 10m are vertically interpolated to the height of the wind turbine hub by the linear interpolation method; then, the grid data output by the model are spatially interpolated to the locations of each selected wind farm station by the bilinear interpolation method; finally, the forecast time of the model output data is sorted, and the rolling forecast data output by the model on a daily basis is obtained by intercepting the forecast results of 1 to 45 days after each reporting date within the set past time range.

[0044] SS3. Analysis of Bias Characteristics of Dynamical Climate Model Forecasts

[0045] Based on the collated historical wind speed observation data and the corresponding daily model output data, the root mean square error (RMSE) was selected as the test and evaluation metric to assess the forecast capability and bias level of the dynamic climate model for wind speed at wind turbine hub height at different lead times (LT) on the sub-seasonal scale, i.e., within the next 1-45 days. Furthermore, the distribution characteristics of model bias were further analyzed for different lead times, wind farms, months, and / or wind speed levels.

[0046] In some preferred embodiments, different lead forecast days at sub-seasonal scales are used in the prediction of dynamic climate models. LT When the wind speed forecast deviation is 0.000, the root mean square error (RMSE) is used to characterize the test prediction index, which is used to quantify the deviation between the predicted wind speed and the actual observed wind speed. The mathematical calculation formula is:

[0047]

[0048] in, LT Forecast days in advance and LT ≤45 days, F i For the i The forecast daily average wind speed for the day, O i is the average wind speed of the observation day on the i-th day, 1≤ i ≤ LT .

[0049] In other preferred embodiments, when evaluating the wind speed forecast deviation of the dynamic climate model, multi-dimensional evaluation indicators such as bias and correlation coefficient can be further introduced to comprehensively analyze the spatiotemporal distribution characteristics of the model forecast deviation. At the same time, machine learning algorithms (such as random forests or gradient boosting trees, etc.) can be used to identify the key factors affecting the model forecast deviation, providing a scientific basis for the construction of subsequent deviation correction models.

[0050] SS4. Construction and Training of Quantile Regression Bias Correction Models

[0051] Based on the results of the forecast bias characteristics analysis of the dynamic climate model, the daily output data of the model within a set past time range is used as the independent variable, and the corresponding historical wind speed observation data is used as the dependent variable. The quantile regression (QR) method is selected and combined with the machine learning algorithm to establish and train quantile regression bias correction models for different advance forecast days, different wind farms, different months and / or different wind speed levels.

[0052] In some preferred embodiments, when constructing a quantile regression bias correction model, an adaptive quantile selection strategy can be adopted to automatically adjust the quantile value (such as 25%, 50%, 75% quantile) according to the error distribution characteristics of different advance forecast days, different wind farms, different months and / or different wind speed levels. At the same time, deep learning technology (such as long short-term memory network LSTM or attention mechanism Attention Mechanism, etc.) is introduced to capture the long-term dependence and periodic characteristics of wind speed time series, improve the model's ability to fit complex nonlinear relationships, and automatically tune the model hyperparameters through the Bayesian optimization algorithm to obtain the optimal model structure and training strategy.

[0053] In other preferred embodiments, when training the quantile regression bias correction model, adaptive learning rate and dynamic regularization strategy can be introduced to achieve differentiated training of different quantile regression models, thereby ensuring training convergence while avoiding model overfitting, thereby improving the correction accuracy and applicability of the model under different advance forecast days.

[0054] SS5. Error Correction and Application of Model Real-time Forecast Data

[0055] The original real-time forecast data of the dynamic climate model is obtained, and the corresponding quantile regression deviation correction model is selected according to its advance forecast days, target wind farm station, forecast month and / or forecast wind speed level. The quantile regression correction result is obtained by inputting the original real-time forecast data of the model into the selected model for error correction. The quantile regression correction result is weightedly fused with the original forecast result of the model to obtain the final sub-seasonal scale model-corrected wind speed forecast result of the target forecast area, providing decision support for wind energy resource prediction and actual operation and scheduling of wind farms.

[0056] SS6. Verify and optimize the model's revised wind speed forecast results

[0057] At least the daily root mean square error (RMSE) is used as an evaluation metric to evaluate the model-corrected wind speed forecast results obtained in step SS5. The accuracy of the model quantile regression correction results is verified by comparing them before and after correction and / or by comparing them with the results of other model correction schemes. Based on the verification results, the key parameters in the quantile regression bias correction model are optimized to further improve the correction accuracy and robustness of the model. The model performance is continuously improved through an iterative verification and optimization process to ensure the reliability and practicality of the model-corrected wind speed forecast results.

[0058] In some preferred embodiments, when optimizing key parameters in the quantile regression bias correction model, the accuracy and stability of the model are improved by adjusting the quantile threshold, optimizing the hyperparameters of the machine learning algorithm (such as learning rate, maximum depth, regularization coefficient, etc.) and combining improved weighted fusion strategies (such as dynamic weighting or weighting methods based on model uncertainty).

[0059] Example 2

[0060] Based on the above-mentioned Example 1, this Example 2 takes the wind speed at hub height of wind turbines selected from nine representative wind farms in a certain area as the research object. Utilizing the historical observation data of wind speed at hub height of wind turbines at wind farms and the return data of the power model, the prediction skill of the power model for wind speed at hub height of wind turbines on a sub-seasonal scale is predicted, and the model deviation characteristics of wind speeds at different levels are analyzed in detail. On this basis, through comparative screening, correction schemes that can effectively improve the deviations of wind speeds at different levels are selected, thereby establishing an objective sub-seasonal forecasting technology for wind energy resources based on model error correction. The daily root mean square error of wind speed for the next 45 days is significantly reduced compared to the direct output of the power model. Specifically:

[0061] SS1. Acquisition and organization of historical wind speed observation data

[0062] Step 1.1: Filter and read raw data. Wind farm observation data is obtained from observations at the wind towers at the wind farms. Unquality-controlled raw data is stored on the tower's corresponding server in a text file format (.txt). The first column contains the observation time (formatted as 20210101001500), and the second column contains the observed wind speed (in m / s). The observation time resolution is 15 minutes, and each day's observations are stored as a single file. First, a wind speed dataset at the hub height of wind turbines (using a height of 70m as an example) measured over the past three years at nine typical wind farms in a specific region is filtered and read from the server. This contains 10,953 text files. The file names are named based on the observation time and the wind farm serial number (e.g., 20220101_01.txt).

[0063] Step 1.2: Clean and organize site wind speed data. First, clean the 15-minute observation data, remove outliers, and treat days with a daily absence rate greater than 30% as missing (a missing value is defined as -999). Next, average the data for each observation day to obtain the daily average wind speed observation result. Combine the daily observation data from each wind tower to generate 10 text files named after the wind farm station number (e.g., 01.txt). The first column contains the observation time (formatted as 20210101), and the second column contains the daily average wind speed (in m / s).

[0064] SS2. Acquisition and organization of daily model output data

[0065] Step 2.1: Acquire and store daily model forecast data. Obtain the past three years of return data and real-time forecast data from a high-resolution dynamical model (using the NCEP_CFSv2 model as an example). This data is a rolling daily forecast for the next 1 to 45 days. The real-time forecast period is from January 1, 2022, to February 28, 2024. Parameters include 10-meter u-velocity, 10-meter v-velocity, U-velocity, V-velocity, geopotential height, surface pressure, and temperature. Model data is gridded, including longitude and latitude information, and stored in the grib2 format.

[0066] Step 2.2: Spatial Interpolation and Time Warping of Model Data. After preprocessing the model results using Python, spatial interpolation of the model data is performed. Specifically, the height of each isobaric layer from the ground is calculated using the temperature, geopotential height, and surface pressure of each isobaric layer. The wind speed of each isobaric layer and the 10-meter wind speed are then linearly interpolated to the height of the wind tower (70 meters in this example). Finally, the grid data are bilinearly interpolated to the locations of the nine observation stations to obtain the 70-meter wind speed forecast results for all nine stations.

[0067] Secondly, the forecast time of the model is sorted, and the results of 1 to 45 days after each reporting start date are intercepted to obtain a 45×9 array (for example, if the reporting start date is January 1, 2023, the forecast results from January 2 to February 15, 2023 are intercepted). The processed results are stored as a txt file with the naming rule of 20230101.txt.

[0068] SS3. Analysis of Bias Characteristics of Dynamical Climate Model Forecasts

[0069] Step 3.1: Evaluate the model's ability to forecast wind speed over the extended period. Using the processed historical observation data from the nine wind farms and the corresponding feedback data from the power model, evaluate the model's ability to predict wind speed at turbine hub height for the next 1 to 45 days. The root mean square error (RMSE) is used as the evaluation metric, and the average RMSE over the next 45 days is taken. A smaller RMSE value indicates a better model's forecasting ability. The RMSE calculation formula is:

[0070]

[0071] In formula (1), Nis the total number used for calculation, which is set to 45 days here. F i Is the number of days i The predicted daily average wind speed is O i Is the number of days i The average wind speed on the observation day.

[0072] The analysis found that when the lead time (LT) is less than 7 days, the RMSE shows a clear increasing trend as the LT increases. When the LT is less than one week, the RMSE is less than 2.0 m / s; when the LT exceeds one week, the error increases to more than 2.0 m / s. Subsequently, as the LT increases, the RMSE fluctuates around an average level above 2.0 m / s. Figure 2 Overall, the error level of the model forecast for the extended period wind speed at the selected stations exceeds 2.0 m / s, and there is considerable room for improvement in the forecast effect.

[0073] Step 3.2 Analysis of the distribution characteristics of model forecast errors. Using 2 m / s as a tier, the deviation ranges of model forecast results at different lead times were classified. It can be found that when the actual wind speed is below 6 m / s, the average error of the model over the five forecast time periods is positive. For wind speeds above 6 m / s, the model shows a negative deviation, and the degree of negative deviation increases significantly with increasing wind speed and longer forecast time periods ( Figure 3 At LT = 1 day, the model's average forecast error for actual wind speeds of 6 to 8 m / s was -1.2 m / s, and the average error for wind speeds of 12 to 14 m / s was -4.6 m / s. At LT = 30 days, the model's average forecast error for actual wind speeds of 6 to 8 m / s was -1.2 m / s, and the average error for wind speeds of 12 to 14 m / s reached -8.8 m / s. Therefore, the model's forecasts for wind speeds of 70 m / s exhibited an error pattern of "light winds predicted too high, and strong winds predicted significantly too low."

[0074] SS4. Build and train a quantile regression bias correction model

[0075] According to the error characteristics, after screening and analyzing multiple schemes, the quantile regression method was finally selected to correct the error. Quantile regression (QR) is to regress the forecast factor X through the conditional quantile of the forecast quantity Y, that is, . Similar to the traditional linear regression model, the quantile regression model can be expressed as:

[0076] (2)

[0077] Among them, Y is the forecast quantity, ; X is the predictor, ; is the parameter vector, ; is the corresponding error vector, ; satisfy , is a specific quantile. The parameter estimation in quantile regression is done by giving the quantile of the predicted quantity Y , and then use the weighted residual minimum to estimate the parameters, specifically:

[0078] (3)

[0079] in is the loss function, I ( μ ) is the characteristic function, and the characteristic function satisfies:

[0080] (4)

[0081] This scheme adopts , that is, the 50th percentile (median) is used as the correction for quantile regression, and the following formula is used for correction:

[0082] (5)

[0083] Among them, Correct QR is the percentile regression correction result, is the quantile regression result with percentile 50, and Ori is the model forecast result. The Quantile Regressor model in sklearn in Python is used to train the historical data (the quantile is selected as 50 percentile, that is, the median). The historical data of the same period of the forecast month are used to train the historical data. By training, we can obtain quantile regression models established for different observation stations with different advance forecast days.

[0084] SS5. Corrected forecast results based on the corrected quantile regression model.

[0085] Taking October and February as examples, we build quantile regression models for different lead forecast days in these two months, i.e., 45 models are built for each month, for a total of 90. The quantile regression correction results account for 50% and the original forecast accounts for 50%, that is, the calculation method according to formula (5) is used as the corrected model results.

[0086] SS6. Further validation and evaluation of model returns and real-time forecasts

[0087] The RMSE of the model correction results was calculated and compared with the results of the other two model correction schemes (e.g. Figure 4 、 5 The prediction results obtained by the model-based quantile regression correction model have lower RMSE at all stations compared with the direct model output results to varying degrees, and the overall improvement effect is more significant than the other two control correction schemes.

[0088] Example 3

[0089] Based on the above-mentioned embodiment 1, this embodiment 3 further provides an electronic device, comprising at least a processor and a memory communicatively connected to the processor, wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can execute the regional sub-seasonal wind energy resource prediction method based on the dynamic climate model shown in the above-mentioned embodiment 1.

[0090] Example 4

[0091] Based on the above embodiment 1, this embodiment 4 further provides a computer program product, including computer instructions for executing the regional sub-seasonal wind energy resource prediction method based on the dynamic climate model shown in the above embodiment 1.

[0092] Example 5

[0093] Based on the above-mentioned embodiment 1, this embodiment 5 further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, the regional sub-seasonal wind energy resource prediction method based on the dynamic climate model shown in the above-mentioned embodiment 1 is implemented.

[0094] The above embodiments fully and effectively achieve the objectives of the present invention. Those skilled in the art will appreciate that the present invention includes, but is not limited to, the contents described in the accompanying drawings and the above specific embodiments. Although the present invention has been described with reference to the embodiments currently considered to be the most practical and preferred, it should be understood that the present invention is not limited to the disclosed embodiments, and any modifications that do not deviate from the functional and structural principles of the present invention are intended to be included within the scope of the claims.

Claims

1. A regional sub-seasonal wind energy resource prediction method based on a dynamic climate model, characterized in that: The method comprises at least the following steps when implemented: SS1. Acquisition and compilation of historical wind speed observation data Obtain historical wind speed observation data at the hub height of wind turbines of several representative wind farms in the target prediction area within a set past time range, and organize them into daily average wind speed data after cleaning; SS2. Acquisition and organization of daily output data from the model Obtain the daily output data of the model in the target prediction area within the set past time range under the high-resolution dynamic climate model, including the daily rolling forecast data of the corresponding sub-seasonal scale for the next 1 to 45 days generated by the feedback method within the set past time range, and make it match the historical wind speed observation data at least in space, height and time through spatial interpolation and time warping; SS3. Analysis of the characteristics of forecast bias of dynamical climate models Based on the collated historical wind speed observation data and the corresponding daily output data of the model, the daily root mean square error is selected as the test and evaluation indicator to evaluate the deviation level of the dynamic climate model for the wind speed at the hub height of the wind turbine under different advance forecast days in the sub-seasonal scale. On this basis, the distribution characteristics of the model deviation under different advance forecast days, different wind farms, different months and / or different wind speed levels are further analyzed; SS4. Construction and training of quantile regression bias correction model Based on the analysis results of the forecast deviation characteristics of the dynamic climate model, the daily output data of the model is used as the independent variable and the corresponding historical wind speed observation data is used as the dependent variable. The quantile regression method is selected and combined with the machine learning algorithm. The quantile regression deviation correction model is established and trained for different forecast days, different wind farms, different months and / or different wind speed levels, where: The constructed quantile regression bias correction model adopts The median is used for error correction, and the forecast factor X is regressed through the conditional quantile of the forecast quantity Y. The mathematical expression is: Among them, Y is the forecast quantity, ; X is the prediction factor, ; is the parameter vector, ; is the corresponding error vector, ; is a quantile and satisfies , by giving the quantile of the predicted quantity Y , and estimate the parameters of the quantile regression by minimizing the weighted residuals : in, is the loss function and through the indicative function I ( μ ) is determined, and the indicative function I ( μ )satisfy ; In the practical application of the quantile regression bias correction model, the following formula is used for error correction: Among them, Correct QR is the percentile regression correction result, for The quantile regression results are: Ori For model forecast results; When constructing the quantile regression bias correction model, an adaptive quantile selection strategy is adopted to automatically adjust the quantile value according to the error distribution characteristics of different forecast days, different wind farms, different months and / or different wind speed levels. At the same time, deep learning technology is introduced to capture the long-term dependence and periodicity characteristics of wind speed time series, and the model hyperparameters are automatically tuned through the Bayesian optimization algorithm. When training the quantile regression bias correction model, adaptive learning rate and dynamic regularization strategy are introduced to achieve differentiated training of different quantile regression models, thus ensuring training convergence and avoiding model overfitting. SS5. Error correction and application of model real-time forecast data The original real-time forecast data of the dynamic climate model is obtained, and the corresponding quantile regression deviation correction model is selected according to its advance forecast days, target wind farm station, forecast month and / or forecast wind speed level. The quantile regression correction result is obtained by inputting the original real-time forecast data of the model into the selected model for error correction. The quantile regression correction result is weightedly fused with the original forecast result of the model to obtain the final sub-seasonal scale model-corrected wind speed forecast result of the target forecast area, which provides decision support for wind energy resource prediction and actual operation and scheduling of wind farms.

2. The regional sub-seasonal wind energy resource prediction method based on dynamic climate model according to claim 1 is characterized in that: In the above step SS1, the past time range is set to the past 3-10 years. The historical wind speed observation data is based on the wind farm station's wind tower, the measurement data of the wind turbine SCADA system, the meteorological station, the lidar and the satellite remote sensing data, or is obtained through multi-source data fusion. The time interval of wind speed observation is in minutes. When the wind speed observation data is cleaned, an adaptive outlier removal algorithm is used to identify and remove outliers. At the same time, the observation days with a daily missing measurement rate exceeding the set threshold are processed as whole-day missing measurements and the missing values ​​are supplemented by the polynomial interpolation method. According to the needs of sub-seasonal scale forecasting, the data of each observation day are averaged to obtain the daily average wind speed observation results, and finally a historical wind speed observation data set is generated.

3. The regional sub-seasonal wind energy resource prediction method based on dynamic climate model according to claim 1 is characterized in that: In the above step SS2, the forecast duration of the selected dynamic climate model must be equal to or greater than 45 days to ensure that the model can provide meteorological information within a sufficient time range and support sub-seasonal wind energy resource forecasts, and the model's daily report data contains a variety of meteorological variable parameters and at least includes 10-meter altitude latitudinal wind speed and longitudinal wind speed, ground pressure, potential heights of multiple isobaric layers, air temperature, latitudinal wind speed and longitudinal wind speed, and the model data is high-resolution grid data containing geographic longitude and latitude information to facilitate subsequent spatial interpolation and matching.

4. The regional sub-seasonal wind energy resource prediction method based on dynamic climate model according to claim 3 is characterized in that: In the above step SS2, when the daily output data of the model are sorted: first, the height of each isobaric layer from the ground is calculated by the air temperature and potential height of each isobaric layer and combined with the ground pressure; then, the longitude and latitude wind speeds of each isobaric layer and the longitude and latitude wind speeds at a height of 10m are vertically interpolated to the height of the wind turbine hub by the linear interpolation method; then, the grid data output by the model are spatially interpolated to the locations of each selected wind farm station by the bilinear interpolation method; finally, the forecast time of the model output data is sorted, and the rolling forecast data output by the model on a daily basis is obtained by intercepting the forecast results of 1 to 45 days after each reporting date within the set past time range.

5. The regional sub-seasonal wind energy resource prediction method based on dynamic climate model according to claim 1 is characterized in that: In the above step SS3, the dynamic climate model is used to predict the different lead forecast days at the sub-seasonal scale. LT When the wind speed forecast deviation is 0.004, the root mean square error (RMSE) is used to characterize the test evaluation index, which is used to quantify the deviation between the predicted wind speed and the actual observed wind speed. The mathematical calculation formula is: in, LT Forecast days in advance and LT ≤45 days, F i For the i The forecast daily average wind speed for the day, O i For the i The average wind speed observed on the day, 1≤ i ≤ LT .

6. The regional sub-seasonal wind energy resource prediction method based on dynamic climate model according to claim 1 is characterized in that: In the above step SS3, when evaluating the wind speed forecast deviation of the dynamic climate model, the deviation and / or correlation coefficient evaluation index is introduced to comprehensively analyze the spatiotemporal distribution characteristics of the model forecast deviation, and the machine learning algorithm is used to identify the key factors affecting the model forecast deviation.

7. The regional sub-seasonal wind energy resource prediction method based on dynamic climate model according to claim 1 is characterized in that: The regional sub-seasonal wind energy resource prediction method further includes step SS6 of verifying the model-corrected wind speed prediction result and optimizing the model, wherein: The model corrected wind speed forecast results obtained in step SS5 are evaluated using at least the daily root mean square error as an evaluation indicator, and the accuracy of the model quantile regression correction results is verified by comparing the results before and after correction and / or by comparing with the results of other model correction schemes. Then, based on the verification results, the key parameters in the quantile regression bias correction model are optimized, and the model performance is continuously improved through an iterative verification and optimization process.

8. The regional sub-seasonal wind energy resource prediction method based on dynamic climate model according to claim 7 is characterized in that: In the above step SS6, when optimizing the key parameters in the quantile regression bias correction model, the accuracy and stability of the model are improved by adjusting the quantile threshold, optimizing the hyperparameters of the machine learning algorithm and combining the improved weighted fusion strategy.

9. An electronic device comprising at least a processor and a memory connected to the processor in communication, wherein: The memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can execute the regional sub-seasonal wind energy resource prediction method based on the dynamic climate model as described in any one of claims 1 to 8 above.

10. A computer program product, comprising computer instructions for executing the regional sub-seasonal wind energy resource prediction method based on a dynamic climate model as described in any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for predicting regional sub-seasonal wind energy resources based on a dynamic climate model as described in any one of claims 1 to 8 is implemented.

Images