Day-ahead wind power prediction method under combination of physics and data driving
The wind power prediction method that combines physics and data-driven approaches solves the problem of low accuracy of single prediction methods across multiple time scales, achieving high-precision wind power prediction that is suitable for precise scheduling and intelligent management of integrated energy systems.
Patent Information
- Application Number
- CN202511033957.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-25
- Publication Date
- 2025-11-11
AI Technical Summary
Existing single-method wind power prediction methods have limitations across multiple time scales and low prediction accuracy, especially in ultra-short-term and ultra-long-term predictions.
A combined physical and data-driven wind power forecasting method is adopted. The wind power sequence for the next 24 hours is predicted by both data-driven and physical-driven methods. The prediction results are corrected and fitted by the total prediction error weight. The advantages of the two methods are combined to form a combined prediction result.
It improves the accuracy and precision of wind power forecasting, especially in ultra-short-term and ultra-long-term forecasting, with a forecasting accuracy of over 90%, making it suitable for precise scheduling and intelligent management of integrated energy systems.
Smart Images
Figure CN120933915A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind power prediction. Background Technology
[0002] With the continuous expansion of wind power generation, the volatility and randomness of wind energy have increasingly impacted wind power grid connection. Accurate wind speed forecasting is considered one of the important means to address this issue, and accurate short-term wind power forecasting is of great significance for the further development of wind power. Currently, there are many methods for wind power forecasting, but most research focuses on three types of single forecasting methods: physical prediction models, data-driven models, and statistical prediction models. There is relatively little literature on combined forecasting models, and even those mentioned in published literature often lack detailed explanations or clear descriptions of their specific forecasting methods and fusion strategies. Traditional data-driven methods rely excessively on model algorithms, and conventional physical prediction models are mainly used for medium- and long-term wind power forecasting.
[0003] With the continuous expansion of wind power generation, the volatility and randomness of wind energy have increasingly highlighted their impact on wind power grid connection. Day-ahead wind power forecasting (accurately predicting the output power of wind farms 24 hours in advance) has become a crucial link in wind power consumption and dispatch optimization. Accurate day-ahead short-term wind power forecasting is of great significance for the further development of wind power. Wind power forecasting methods are divided into three main categories: physical mechanism-based numerical weather prediction models (such as WRF), statistical learning models based on historical data, and deep learning models. Although these methods have certain advantages at different time scales, single models still have limitations in dealing with wind power forecasting across multiple time scales, resulting in low prediction accuracy. For example, data-driven models can effectively capture the nonlinear fluctuation characteristics in wind power time series, but due to the lack of physical constraints, their prediction accuracy decreases with the extension of the prediction period. Physical models such as WRF have good physical consistency and medium-term trend prediction capabilities, but they suffer from response lag in ultra-short-term forecasting, limiting their application in this area. Summary of the Invention
[0004] The purpose of this invention is to address the limitations and low prediction accuracy of existing single-method wind power prediction methods when dealing with wind power prediction at multiple time scales, and to provide a day-ahead wind power prediction method based on a combination of physics and data-driven approaches.
[0005] A day-ahead wind power forecasting method based on a combination of physics and data-driven approaches, comprising:
[0006] S1. At the current moment, predict the wind power sequence of the target electric field for the next 24 hours at the hour using both data-driven and physical-driven methods respectively.
[0007] S2. Based on the target electric field wind power sequences predicted at the hourly times of the previous day and the two days prior, respectively, under data-driven and physical-driven methods for 6 to 24 hours, and the corresponding actual target electric field wind power sequences, obtain the total prediction error weight w. * ;
[0008] S3. Using the total prediction error weight w * After correcting and fitting the target farm wind power sequence predicted under data-driven and physical-driven methods for the next 6 to 24 hours respectively, the target farm wind power sequence for the next 6 to 24 hours under the combined method is obtained.
[0009] S4. After splicing the target farm wind power sequence for the next 1 to 5 hours under the data-driven method with the target farm wind power sequence for the next 6 to 24 hours under the combined method, the target farm wind power sequence for the next 24 hours under the combined method is obtained.
[0010] Preferably, in step S1, the method for predicting the wind power sequence of the target power field in the next 24 hours using a data-driven approach includes:
[0011] The trained wind power data-driven prediction model predicts the wind power sequence of the target wind farm for the next 24 hours based on the wind speed, wind direction, temperature, pressure, and wind power sequences of the target wind farm in the historical period prior to the current moment.
[0012] Preferably, the implementation methods of the prediction model driven by the trained wind power data include:
[0013] Dataset construction phase:
[0014] The wind speed, wind direction, temperature, pressure and wind power sequences of the target wind field collected in history are sampled at a preset time step of 1 hour to form a dataset. The input data of each sample in the dataset are the wind speed, wind direction, temperature, pressure and wind power sequences corresponding to the historical sampling period, and the output data of the sample is the actual wind power sequence corresponding to the prediction period.
[0015] Model training phase:
[0016] The wind power data-driven prediction model is trained using each sample to obtain the trained wind power data-driven prediction model.
[0017] Preferably, in step S1, the method of predicting the wind power sequence of the target electric field in the next 24 hours using the physical driving method includes the following steps:
[0018] S11. Obtain the wind speed sequence of the target wind farm area for the next 24 hours using the WRF meteorological model;
[0019] S12. Based on the wind speed sequence of the target wind farm area and the latitude and longitude coordinates of each wind turbine, determine the wind speed sequence at the hub of each wind turbine in the next 24 hours.
[0020] S13. Correct the wind speed sequence at the hub of each wind turbine for the next 24 hours to obtain the corrected wind speed sequence at the hub of each wind turbine for the next 24 hours.
[0021] S14. Based on the wind speed sequence at the hub of each wind turbine after correction for the next 24 hours, obtain the wind power sequence of each wind turbine for the next 24 hours, and combine the wind power sequences of all wind turbines at each time point to obtain the wind power sequence of the target wind farm for the next 24 hours.
[0022] Preferably, the method for correcting the wind speed sequence at the hub of each wind turbine in step S13 for the next 24 hours includes:
[0023] S131. In the direction of wind propagation, determine whether the current k-th wind turbine is affected by the wake of the i-th wind turbine located upstream. The result is yes. If the result is negative, proceed to step S132.
[0024] This is the corrected wind speed at the hub of the k-th wind turbine in the j-th hour. To correct the wind speed at the hub of the k-th wind turbine in the j-th hour prior;
[0025]
[0026] in, To correct the wind speed at the hub of the i-th wind turbine in the j-th hour prior, where i, k, and j are integers, and 1 ≤ i, k ≤ n, 1 ≤ j ≤ 24, where n is the total number of wind turbines, and H k Let be the hub height of the k-th wind turbine, α1 be the wind shear index, and C be the... T d is the thrust coefficient of the wind turbine, R is the rotor radius of the wind turbine, κ′ is the wake descent coefficient, and d k,i Let h be the horizontal distance between the k-th wind turbine and its corresponding upstream i-th wind turbine along the wind propagation direction. k,i Let be the altitude difference between the current k-th wind turbine and its corresponding upstream i-th wind turbine in the direction of wind propagation.
[0027] Preferably, if the i-th wind turbine located upstream of the k-th wind turbine is at the outermost edge of the target wind farm area, then κ′=0.04; otherwise, κ′=0.08.
[0028] Preferably, in step S131, when the current k-th wind turbine is affected by the wake of the i-th wind turbine located upstream, the following must be satisfied simultaneously: d i,k <15×(R×2), and
[0029] in, Let be the vector pointing from the origin of the region where the target wind farm is located to the location of the k-th wind turbine. Let be the vector pointing from the origin of the region where the target wind farm is located to the location of the i-th wind turbine. As an intermediate variable, To correct the wind speed at the hub of the i-th wind turbine in the j-th hour prior, d i,k Let be the horizontal distance between the current k-th wind turbine and the i-th wind turbine located upstream of it, and let α be the angle between the line connecting the current k-th wind turbine and the i-th wind turbine located upstream of it and the direction of wind propagation, where α < 15°.
[0030] Preferably, in step S2, the total prediction error weight w is obtained. * The implementation method is as follows:
[0031] S21. Obtain the prediction error and in,
[0032]
[0033] and These represent the prediction errors for the h-th hour of the previous day at the current moment, under data-driven and physical-driven methods, respectively; h is an integer, and 6≤h≤24;
[0034] and These represent the prediction errors for the h-th hour of the previous two days under data-driven and physical-driven methods, respectively; P d,h and P d-1,h These represent the actual target wind power output of the electric field at the h-th hour of the previous day and the two days prior, respectively;
[0035] and The target wind power output of the electric field is predicted in the data-driven manner for the h-th hour of the previous day and the two days prior, respectively;
[0036] and These are the target electric field wind power predicted under the physical drive mode at the h-th hour of the previous day and the previous two days, respectively;
[0037] S22, according to and Construct the combination matrix at hour h
[0038] S23. Based on the prediction error matrix in the data-driven mode at hour h. The prediction error matrix at hour h under physical drive mode calculate and The prediction error covariance matrix between
[0039] in,
[0040] Prediction error in data-driven approach variance Prediction error under physical driving method variance for and Inter-variance error;
[0041] S24, According to Cov(E) h The total prediction error weight matrix w is obtained. * .
[0042] Preferably, according to S24, according to Cov(E) h The total prediction error weight matrix w is obtained. * The implementation methods include:
[0043]
[0044] w * =[w *,6 w *,7 w *,8 ...w *,24 ] T ;
[0045] Among them, w *,h Here, T represents the prediction error weight for the h-th hour, and T is the transpose.
[0046] Preferably, S3 utilizes the total prediction error weight matrix w *After correcting and fitting the target farm wind power sequence predicted under both data-driven and physical-driven methods for the next 6 to 24 hours, the combined method for realizing the target farm wind power sequence for the next 6 to 24 hours is as follows:
[0047] in, This is the target wind power sequence for the next 6 to 24 hours under the combined configuration. The target electric field wind power sequence obtained through a data-driven approach over 6 to 24 hours. The target electric field wind power sequence for 6 to 24 hours is obtained by physical drive method.
[0048] Advantages of this invention:
[0049] The prediction method of this invention consists of two main parts. The first part (1-5 hour prediction): Due to the high correlation between wind power data and the prediction time, and the long computation time of the physical-driven method (WRF meteorological model) for wind power prediction, the data-driven results are directly used as the predicted wind power values during this period. The second part (6-24 hour prediction): As the prediction time increases, the correlation coefficient gradually decreases, limiting the accuracy of relying solely on data-driven methods. Therefore, during this period, a static weighted combination strategy is adopted to fuse the physical-driven and data-driven prediction results to improve the accuracy of wind power prediction. Finally, by combining the data-driven and physical-driven results, the wind power sequence of the target electric field for the next 24 hours is obtained under the combined method, improving prediction accuracy.
[0050] The prediction method of this invention has high computational accuracy, overcoming the drawbacks of data-driven or single-physical prediction. By combining the advantages of different prediction methods, the prediction accuracy can reach over 90%. This invention also overcomes the problem of traditional data-driven methods over-reliance on model algorithms and the limitation of physical prediction models to long-term forecasting. It plays a significant role in promoting precise scheduling and intelligent management of integrated energy systems. The combined prediction model integrates the advantages of multiple prediction models, improving both stability and accuracy.
[0051] The prediction method of this invention has a fast calculation speed. In the physical driving, a high-precision terrain is used to establish a calculation domain for the size of the wind farm. Compared with the traditional mesoscale WRF prediction model, its calculation domain area is smaller. The selection of the calculation domain is more purposeful, which is conducive to improving the prediction accuracy of mesoscale physical prediction.
[0052] By incorporating refined terrain features to consider the impact of terrain on wind speed, the influence of terrain undulations on wind speed distribution under complex terrain conditions is taken into account. Based on this, and combining the output results of the WRF meteorological model, a wake correction mechanism is further introduced to consider the wake effect of upstream wind turbines on wind speed at the hub height of downstream turbines, enabling dynamic adjustment of the effective value of the wind speed received by the turbine. This effectively improves the consistency and accuracy of wind speed and power prediction in wind farms with complex terrain, and suppresses systematic overestimation errors in the overall power prediction of wind farms.
[0053] The prediction method of this invention integrates the advantages of physical prediction models and data-driven models, which helps to improve the accuracy of wind power prediction. Ultimately, the accuracy of wind power prediction can reach 90%. Attached Figure Description
[0054] Figure 1 This is a logic block diagram of the day-ahead wind power prediction method based on the combination of physics and data-driven approaches of the present invention.
[0055] Figure 2 This is a schematic diagram showing the distribution of wind turbines within the target wind farm area.
[0056] Figure 3 This is a schematic diagram of wind speed distribution within the area where the target wind farm is located. Detailed Implementation
[0057] 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. 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.
[0058] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.
[0059] Specific Implementation Method 1: Combination Figure 1 As shown, this embodiment provides a day-ahead wind power prediction method based on a combination of physical and data-driven approaches. The method includes:
[0060] S1. At the current moment, predict the wind power sequence of the target electric field for the next 24 hours at the hour using both data-driven and physical-driven methods respectively.
[0061] S2. Based on the target electric field wind power sequences predicted at the hourly times of the previous day and the two days prior, respectively, under data-driven and physical-driven methods for 6 to 24 hours, and the corresponding actual target electric field wind power sequences, obtain the total prediction error weight matrix w.* ;
[0062] S3. Using the total prediction error weight matrix w * After correcting and fitting the target farm wind power sequence predicted under data-driven and physical-driven methods for the next 6 to 24 hours respectively, the target farm wind power sequence for the next 6 to 24 hours under the combined method is obtained.
[0063] S4. After splicing the target farm wind power sequence for the next 1 to 5 hours under the data-driven method with the target farm wind power sequence for the next 6 to 24 hours under the combined method, the target farm wind power sequence for the next 24 hours under the combined method is obtained.
[0064] The prediction method of this invention consists of two main parts. The first part (1-5 hour prediction): Due to the high correlation between wind power data and the prediction time, and the long computation time of the physical-driven method (WRF meteorological model) for wind power prediction, the data-driven results are directly used as the predicted wind power values during this period. The second part (6-24 hour prediction): As the prediction time increases, the correlation coefficient gradually decreases, limiting the accuracy of relying solely on data-driven methods. Therefore, during this period, a static weighted combination strategy is adopted to fuse the physical-driven and data-driven prediction results to improve the accuracy of wind power prediction. Finally, by combining the data-driven and physical-driven results, the wind power sequence of the target electric field for the next 24 hours is obtained under the combined method, improving prediction accuracy.
[0065] Specifically, in step S1, the implementation method of predicting the wind power sequence of the target power field in the next 24 hours through data-driven approach includes:
[0066] The trained wind power data-driven prediction model predicts the wind power sequence of the target wind farm for the next 24 hours based on the wind speed, wind direction, temperature, pressure, and wind power sequences of the target wind farm in the historical period prior to the current moment.
[0067] The implementation methods for using the trained wind power data to drive the prediction model include:
[0068] Dataset construction phase:
[0069] The wind speed, wind direction, temperature, pressure and wind power sequences of the target wind field collected in history are sampled at a preset time step of 1 hour to form a dataset. The input data of each sample in the dataset are the wind speed, wind direction, temperature, pressure and wind power sequences corresponding to the historical sampling period, and the output data of the sample is the actual wind power sequence corresponding to the prediction period.
[0070] Model training phase:
[0071] The wind power data-driven prediction model is trained using each sample to obtain the trained wind power data-driven prediction model.
[0072] This invention utilizes historical power generation data from wind farms and meteorological data such as wind speed, wind direction, temperature, humidity, and atmospheric pressure. A data-driven prediction model for wind power generation is established using deep learning algorithms. This model can be implemented using existing neural networks.
[0073] Furthermore, in step S1, the method for predicting the wind power sequence of the target electric field in the next 24 hours using the physical driving method includes the following steps:
[0074] S11. Obtain the wind speed sequence of the target wind farm area for the next 24 hours using the WRF meteorological model;
[0075] S12. Based on the wind speed sequence of the target wind farm area and the latitude and longitude coordinates of each wind turbine, determine the wind speed sequence at the hub of each wind turbine for the next 24 hours; see [link / reference]. Figure 2 and Figure 3 ;
[0076] S13. Correct the wind speed sequence at the hub of each wind turbine for the next 24 hours to obtain the corrected wind speed sequence at the hub of each wind turbine for the next 24 hours; specifically...
[0077] S131. In the direction of wind propagation, determine whether the current k-th wind turbine is affected by the wake of the i-th wind turbine located upstream. The result is yes. If the result is negative, proceed to step S132.
[0078] This is the corrected wind speed at the hub of the k-th wind turbine in the j-th hour. To correct the wind speed at the hub of the k-th wind turbine in the j-th hour prior;
[0079]
[0080] in, To correct the wind speed at the hub of the i-th wind turbine in the j-th hour prior, where i, k, and j are integers, and 1 ≤ i, k ≤ n, 1 ≤ j ≤ 24, where n is the total number of wind turbines, and H k Let be the hub height of the k-th wind turbine, α1 be the wind shear index, and C be the... T d is the thrust coefficient of the wind turbine, R is the rotor radius of the wind turbine, κ′ is the wake descent coefficient, and d k,iLet h be the horizontal distance between the k-th wind turbine and its corresponding upstream i-th wind turbine along the wind propagation direction. k,i Let be the altitude difference between the current k-th wind turbine and its corresponding upstream i-th wind turbine in the direction of wind propagation;
[0081] S14. Based on the wind speed sequence at the hub of each wind turbine after correction for the next 24 hours, obtain the wind power sequence of each wind turbine for the next 24 hours, and combine the wind power sequences of all wind turbines at each time point to obtain the wind power sequence of the target wind farm for the next 24 hours.
[0082] In this preferred embodiment, the influence of terrain on wind speed is considered by introducing refined terrain features, taking into account the impact of terrain undulations on wind speed distribution under complex terrain conditions. Based on this, and combined with the output results of the WRF meteorological model, a wake correction mechanism is further introduced to consider the wake effect of upstream wind turbines on wind speed at the hub height of downstream turbines, enabling dynamic adjustment of the effective value of the wind speed received by the current turbine. This effectively improves the consistency and accuracy of wind speed and power prediction in wind farms with complex terrain, and suppresses systematic overestimation errors in the overall power prediction of the wind farm. The WRF meteorological model is existing technology.
[0083] In practical applications, if the current k-th wind turbine is affected by the wake of the i-th wind turbine located upstream, then only one upstream wind turbine has an impact on the current k-th wind turbine.
[0084] Furthermore, if the i-th wind turbine located upstream of the k-th wind turbine is at the outermost edge of the target wind farm area, then κ′=0.04; otherwise, κ′=0.08.
[0085] In this preferred embodiment, different coefficients are used to consider the varying intensity of wake effects between the outer and inner wind turbines in the wind farm, allowing for a more accurate correction of the wind speed predicted by the WRF meteorological model received by the downstream wind turbines. This method can more accurately reflect the disturbance characteristics of wind turbines at different spatial locations within the wind farm, effectively reducing power prediction errors caused by wake effects.
[0086] Furthermore, in step S131, when the current k-th wind turbine is affected by the wake of the i-th wind turbine located upstream, the following must be satisfied simultaneously: d i,k <15×(R×2), and
[0087] in, Let be the vector pointing from the origin of the region where the target wind farm is located to the location of the k-th wind turbine. Let be the vector pointing from the origin of the region where the target wind farm is located to the location of the i-th wind turbine. As an intermediate variable, To correct the wind speed at the hub of the i-th wind turbine in the j-th hour prior, d i,k Let be the horizontal distance between the current k-th wind turbine and the i-th wind turbine located upstream of it, and let α be the angle between the line connecting the current k-th wind turbine and the i-th wind turbine located upstream of it and the direction of wind propagation, where α < 15°.
[0088] In this preferred embodiment, wind direction information predicted by the WRF meteorological model is combined with geometric quantification analysis of the dynamic wind direction relationship between the wind turbine's on-site layout and the upstream hub to determine whether the wind turbine is affected by the wake of the upstream turbine. Compared with traditional methods that only use fixed wind direction or simple projection relationships to determine the wake influence, this invention can more accurately consider the impact of wind direction disturbances and heterogeneous turbine layout on the wake propagation path in the wind field, enhance the wake model's adaptability to non-uniform wind field layouts, and make the early prediction of wind power more accurate.
[0089] Furthermore, in step S2, the total prediction error weight matrix w is obtained. * The implementation method is as follows:
[0090] S21. Obtain the prediction error and in,
[0091]
[0092] and These represent the prediction errors for the h-th hour of the previous day at the current moment, under data-driven and physical-driven methods, respectively; h is an integer, and 6≤h≤24;
[0093] and These represent the prediction errors for the h-th hour of the previous two days under data-driven and physical-driven methods, respectively; P d,h and P d-1,h These represent the actual target wind power output of the electric field at the h-th hour of the previous day and the two days prior, respectively;
[0094] and The target wind power output of the electric field is predicted in the data-driven manner for the h-th hour of the previous day and the two days prior, respectively;
[0095] and These are the target electric field wind power predicted under the physical drive mode at the h-th hour of the previous day and the previous two days, respectively;
[0096] S22, according to and Construct the combination matrix at hour h
[0097] S23. Based on the prediction error matrix in the data-driven mode at hour h. The prediction error matrix at hour h under physical drive mode calculate and The prediction error covariance matrix between
[0098]
[0099] in,
[0100] Prediction error in data-driven approach variance Prediction error under physical driving method variance for and Inter-variance error;
[0101] S24, According to Cov(E) h The total prediction error weight matrix w is obtained. * Specifically:
[0102]
[0103] w * =[w *,6 w *,7 w *,8 ...w *,24 ] T ;
[0104] Among them, w *,h Here, T represents the prediction error weight for the h-th hour, and T is the transpose.
[0105] In this preferred embodiment, the prediction error covariance matrix is constructed by introducing prediction errors under two different methods, thereby achieving dynamic optimization of the combined prediction weights of data-driven and physical prediction methods.
[0106] In calculating the total prediction error weight matrix w * During the process, the total prediction error weight matrix w *The variation of error weight coefficients over time is considered, reflecting the trend of prediction error over time, effectively avoiding the limitations of using fixed weights in traditional optimal combination methods. The weight calculation follows the "minimum combination variance" criterion, considering the variance information of the errors of the two driving methods and the correlation between their errors, thereby ensuring the overall prediction accuracy and stability of wind power combination prediction.
[0107] Furthermore, S3, using the total prediction error weight matrix w * After correcting and fitting the target farm wind power sequence predicted under both data-driven and physical-driven methods for the next 6 to 24 hours, the combined method for realizing the target farm wind power sequence for the next 6 to 24 hours is as follows:
[0108] in, This is the target wind power sequence for the next 6 to 24 hours under the combined configuration. The target electric field wind power sequence obtained through a data-driven approach over 6 to 24 hours. The target electric field wind power sequence for 6 to 24 hours is obtained by physical drive method.
[0109] This preferred embodiment comprehensively employs a time-series combination strategy and a multi-prediction combination method, effectively combining the advantages of data-driven and physical prediction methods. It fully leverages the high responsiveness and accuracy of data-driven methods in ultra-short-term prediction, while integrating the WRF physical model's 6-24h prediction stability and physical consistency, thus achieving an organic balance between accuracy and stability across different prediction periods. This effectively avoids the "start-up time error" and wasted computational resources caused by directly using WRF model prediction results. During the combination process, the prediction weights of the two models are dynamically adjusted based on error characteristics, ensuring that the prediction results not only possess temporal consistency but also good error robustness, improving the stability, adaptability, and overall accuracy of the day-ahead wind power prediction.
[0110] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.
Claims
1. A day-ahead wind power prediction method based on a combination of physics and data-driven approaches, characterized in that, The method includes: S1. At the current moment, predict the wind power sequence of the target electric field for the next 24 hours at the hour using both data-driven and physical-driven methods respectively. S2. Based on the target electric field wind power sequences predicted at the hourly times of the previous day and the two days prior, respectively, under data-driven and physical-driven methods for 6 to 24 hours, and the corresponding actual target electric field wind power sequences, obtain the total prediction error weight w. * ; S3. Using the total prediction error weight w * After correcting and fitting the target farm wind power sequence predicted under data-driven and physical-driven methods for the next 6 to 24 hours respectively, the target farm wind power sequence for the next 6 to 24 hours under the combined method is obtained. S4. After splicing the target farm wind power sequence for the next 1 to 5 hours under the data-driven method with the target farm wind power sequence for the next 6 to 24 hours under the combined method, the target farm wind power sequence for the next 24 hours under the combined method is obtained.
2. The day-ahead wind power prediction method based on a combination of physics and data-driven approaches according to claim 1, characterized in that, In step S1, the data-driven method for predicting the wind power sequence of the target power field in the next 24 hours includes: The trained wind power data-driven prediction model predicts the wind power sequence of the target wind farm for the next 24 hours based on the wind speed, wind direction, temperature, pressure, and wind power sequences of the target wind farm in the historical period prior to the current moment.
3. The day-ahead wind power prediction method based on a combination of physics and data-driven approaches according to claim 2, characterized in that, The methods for implementing a prediction model driven by trained wind power data include: Dataset construction phase: The wind speed, wind direction, temperature, pressure and wind power sequences of the target wind field collected in history are sampled at a preset time step of 1 hour to form a dataset. The input data of each sample in the dataset are the wind speed, wind direction, temperature, pressure and wind power sequences corresponding to the historical sampling period, and the output data of the sample is the actual wind power sequence corresponding to the prediction period. Model training phase: The wind power data-driven prediction model is trained using each sample to obtain the trained wind power data-driven prediction model.
4. The day-ahead wind power prediction method based on a combination of physical and data-driven approaches according to claim 1, characterized in that, In step S1, the method for predicting the wind power sequence of the target electric field in the next 24 hours using the physical driving method includes the following steps: S11. Obtain the wind speed sequence of the target wind farm area for the next 24 hours using the WRF meteorological model; S12. Based on the wind speed sequence of the target wind farm area and the latitude and longitude coordinates of each wind turbine, determine the wind speed sequence at the hub of each wind turbine in the next 24 hours. S13. Correct the wind speed sequence at the hub of each wind turbine for the next 24 hours to obtain the corrected wind speed sequence at the hub of each wind turbine for the next 24 hours. S14. Based on the wind speed sequence at the hub of each wind turbine after correction for the next 24 hours, obtain the wind power sequence of each wind turbine for the next 24 hours, and combine the wind power sequences of all wind turbines at each time point to obtain the wind power sequence of the target wind farm for the next 24 hours.
5. The day-ahead wind power prediction method based on a combination of physics and data-driven approaches according to claim 4, characterized in that, The implementation methods for correcting the wind speed sequence at the hub of each wind turbine in step S13 for the next 24 hours include: S131. In the direction of wind propagation, determine whether the current k-th wind turbine is affected by the wake of the i-th wind turbine located upstream. The result is yes. If the result is negative, proceed to step S132. This is the corrected wind speed at the hub of the k-th wind turbine in the j-th hour. To correct the wind speed at the hub of the k-th wind turbine in the j-th hour prior; S132, Order in, To correct the wind speed at the hub of the i-th wind turbine in the j-th hour prior, where i, k, and j are integers, and 1 ≤ i, k ≤ n, 1 ≤ j ≤ 24, where n is the total number of wind turbines, and H k Let be the hub height of the k-th wind turbine, α1 be the wind shear index, and C be the... T d is the thrust coefficient of the wind turbine, R is the rotor radius of the wind turbine, κ′ is the wake descent coefficient, and d k,i Let h be the horizontal distance between the k-th wind turbine and its corresponding upstream i-th wind turbine along the wind propagation direction. k,i Let be the altitude difference between the current k-th wind turbine and its corresponding upstream i-th wind turbine in the direction of wind propagation.
6. The day-ahead wind power prediction method based on a combination of physical and data-driven approaches according to claim 5, characterized in that, If the i-th wind turbine located upstream of the k-th wind turbine is at the outermost edge of the target wind farm area, then κ′=0.04; otherwise, κ′=0.
08.
7. The day-ahead wind power prediction method based on a combination of physical and data-driven approaches according to claim 5, characterized in that, In step S131, when the current k-th wind turbine is affected by the wake of the i-th wind turbine located upstream, the following must be satisfied simultaneously: d i,k <15×(R×2), and in, Let be the vector pointing from the origin of the region where the target wind farm is located to the location of the k-th wind turbine. Let be the vector pointing from the origin of the region where the target wind farm is located to the location of the i-th wind turbine. As an intermediate variable, To correct the wind speed at the hub of the i-th wind turbine in the j-th hour prior, d i,k Let be the horizontal distance between the current k-th wind turbine and the i-th wind turbine located upstream of it, and α be the angle between the line connecting the current k-th wind turbine and the i-th wind turbine located upstream of it and the direction of wind propagation, where α < 15°.
8. The day-ahead wind power prediction method based on a combination of physical and data-driven approaches according to claim 1, characterized in that, In step S2, the total prediction error weight w is obtained. * The implementation method is as follows: S21. Obtain the prediction error and in, and These represent the prediction errors for the h-th hour of the previous day at the current moment, under data-driven and physical-driven methods, respectively; h is an integer, and 6≤h≤24; and These represent the prediction errors for the h-th hour of the previous two days under data-driven and physical-driven methods, respectively; P d,h and P d-1,h These represent the actual target wind power output of the electric field at the h-th hour of the previous day and the two days prior, respectively; and The target wind power output of the electric field is predicted in the data-driven manner for the h-th hour of the previous day and the two days prior, respectively; and These are the target electric field wind power predicted under the physical drive mode at the h-th hour of the previous day and the previous two days, respectively; S22, according to and Construct the combination matrix at hour h S23. Based on the prediction error matrix in the data-driven mode at hour h. The prediction error matrix at hour h under physical drive mode calculate and The prediction error covariance matrix between in, Prediction error in data-driven approach variance Prediction error under physical driving method variance for and Inter-variance error; S24, According to Cov(E) h The total prediction error weight matrix w is obtained. * .
9. The day-ahead wind power prediction method based on a combination of physical and data-driven approaches according to claim 8, characterized in that, According to S24, according to Cov(E) h The total prediction error weight matrix w is obtained. * The implementation methods include: In * =[in *,6 In *,7 In *,8 ...In *,24 ] T ; Among them, w *,h Here, T represents the prediction error weight for the h-th hour, and T is the transpose.
10. The day-ahead wind power prediction method based on a combination of physical and data-driven approaches according to claim 1, characterized in that, S3. Using the total prediction error weight matrix w * After correcting and fitting the target farm wind power sequence predicted under both data-driven and physical-driven methods for the next 6 to 24 hours, the combined method for realizing the target farm wind power sequence for the next 6 to 24 hours is as follows: in, This is the target wind power sequence for the next 6 to 24 hours under the combined configuration. The target electric field wind power sequence obtained through a data-driven approach over 6 to 24 hours. The target electric field wind power sequence for 6 to 24 hours is obtained by physical drive method.