Short-term and ultra-short-term combined power prediction method and system considering wind turbine wake flow influence

Abstract

The invention discloses a short-term-ultra-short-term combined power prediction method and system considering wind turbine wake flow influence, and belongs to the technical field of new energy power generation power prediction. The method comprises the following steps: acquiring the overall forecast power of a wind power plant to be subjected to power forecast and the forecast wind speed at the height of each wind turbine hub of the wind power plant based on a numerical weather forecast model; training the short-term wind power prediction model by using the forecast wind speed and the forecast power; acquiring the historical real-time output power of the wind power plant, and training the ultra-short-term wind power prediction model by using the historical real-time output power, the forecast wind speed and the output of the short-term wind power prediction model; and finally, the short-term wind power prediction model and the ultra-short-term wind power prediction model are utilized to obtain the overall short-term and ultra-short-term forecast power of the wind power plant. The method improves the accuracy and stability of multi-time scale wind power prediction, and is suitable for wind power plant operation scheduling and power system safety control.

Description

A method and system for short-term and ultra-short-term combined power prediction that takes into account the effect of wind turbine wake. Technical Field

[0001] This invention belongs to the field of wind power prediction and new energy grid-connected operation technology, specifically involving a short-term and ultra-short-term combined power prediction method and system that takes into account the influence of wind turbine wake. Background Technology

[0002] With the rapid development of new energy power generation technologies, the installed capacity and penetration rate of wind power in the power system continue to increase. Affected by meteorological conditions, wind power exhibits significant randomness, volatility, and uncertainty. Its prediction accuracy directly impacts the safe and stable operation of the power system, power dispatch decisions, and the economic viability of electricity market transactions. Therefore, conducting research on high-precision, multi-timescale wind power prediction has significant engineering application value.

[0003] Numerical weather prediction models, as an important reference for wind farm power forecasting, can effectively depict large-scale meteorological evolution trends. However, their spatial resolution is limited, making it difficult to accurately reflect the complex flow field structure within wind farms, especially the wake effect caused by wind turbine interactions. In actual wind farm operation, the operation of upstream wind turbines leads to reduced wind speeds and enhanced turbulence in downstream areas, resulting in a significant deviation between the overall power output of the wind farm and the theoretical value under free flow conditions. Directly using the output results of numerical weather prediction models for power forecasting can easily introduce systematic errors, affecting forecast accuracy. Furthermore, existing wind farm power forecasting methods mostly focus on single-time-scale forecasting. Short-term and ultra-short-term forecasting models are usually independent of each other, lacking effective information transmission and coordination mechanisms. Short-term forecast results are often treated as independent outputs, failing to play a priori constraint role in the ultra-short-term forecasting stage; while ultra-short-term forecasting models can reflect the current operating status using real-time power data, their ability to perceive future meteorological trends is limited.

[0004] In summary, the existing technology still has the following shortcomings: (1) Numerical weather forecast results are difficult to directly and accurately reflect the actual wind conditions inside the wind farm that take into account the wake effect; (2) Short-term wind power forecast and ultra-short-term wind power forecast are independent of each other and lack a multi-time scale coordination mechanism; (3) It fails to effectively integrate meteorological forecast information and real-time operation data of wind farms, making it difficult to balance forecast accuracy and stability.

[0005] Therefore, it is necessary to propose a short-term and ultra-short-term joint wind power prediction method and system that takes into account the influence of wind turbine wake, so as to realize the synergistic fusion of numerical weather forecast information and real-time power information, thereby improving the prediction accuracy and engineering applicability of wind power at different time scales. Summary of the Invention

[0006] To address the problems in the prior art, this invention provides a short-term and ultra-short-term combined power prediction method and system that takes into account the influence of wind turbine wake.

[0007] The technical solution of the present invention is as follows: In a first aspect, the present invention discloses a short-term-ultra-short-term joint power prediction method that takes into account the influence of wind turbine wake, for wind farm power prediction; the method includes the following steps: 1) acquiring the physical information, historical operating data, and historical numerical weather forecast data of the wind farm to be predicted, and preprocessing the historical operating data, and then using the numerical weather forecast model to obtain the predicted wind speed at the hub height of each wind turbine in the wind farm and the predicted power of the wind farm as a whole within a preset time range; 2) using the predicted wind speed and predicted power obtained in step 1) to predict the short-term wind power. The model is trained; the output of the short-term wind power prediction model is the corrected forecast power of the wind farm as a whole; 3) the historical real-time output power of the wind farm is obtained, and the forecast wind speed of step 1), the forecast power of step 2), and the historical real-time output power are used to train the ultra-short-term wind power prediction model; 4) based on the current operating data of the wind farm, the numerical weather forecast data of the wind farm location at the current time, and the real-time output power of the wind farm at the previous time, the short-term and ultra-short-term forecast power of the wind farm as a whole is obtained using the trained short-term wind power prediction model and the trained ultra-short-term wind power prediction model.

[0008] Further, in step 1), the preprocessing of historical operating data includes: firstly, identifying and correcting anomalies in the wind speed sequence of the historical operating data to obtain a corrected wind speed sequence; then, based on the corrected wind speed sequence, identifying and correcting anomalies in the actual output power sequence of the historical operating data to obtain a corrected actual output power sequence; wherein, the actual output power sequence of the historical operating data is corrected using the theoretical power characteristics of the wind turbine.

[0009] Further, in step 2), the short-term wind power prediction model is an iTransformer model; when training the short-term wind power prediction model, the actual output power of the wind farm as a whole within the time range corresponding to the predicted power in step 1) is also obtained, and the actual output power is used as a label for training the short-term wind power prediction model.

[0010] Further, in step 3), the ultra-short-term wind power prediction model is an iTransformer model; the step of training the ultra-short-term wind power prediction model using the predicted wind speed from step 1), the predicted power from step 2), and the historical real-time output power includes: acquiring partial predicted power data from the corrected predicted power from step 2) that corresponds to the preset time range predicted by the ultra-short-term wind power prediction model, and acquiring partial predicted wind speed data from the predicted wind speed from step 1) that corresponds to the preset time range predicted by the ultra-short-term wind power prediction model; training the ultra-short-term wind power prediction model using the historical real-time output power, partial predicted power data, and partial predicted wind speed data from step 3) to obtain a trained short-term wind power prediction model; wherein, the output of the ultra-short-term wind power prediction model is the overall predicted power of the wind farm within the preset time range.

[0011] Step 4) includes: based on the physical information of the wind farm, the current operating data of the wind farm, and the current numerical weather forecast data of the wind farm's location, using a numerical weather forecast model to obtain the forecast wind speed at the hub height of each wind turbine in the wind farm within a preset time range, as well as the forecast power of the wind farm as a whole; based on the forecast wind speed and forecast power, using a trained short-term wind power prediction model to obtain the corrected short-term forecast power of the wind farm as a whole; based on the real-time output power of the wind farm at the previous moment, the forecast wind speed, and the corrected forecast power, using a trained ultra-short-term wind power prediction model to obtain the ultra-short-term forecast power of the wind farm as a whole.

[0012] Furthermore, the short-term forecast power of the wind farm as a whole is the forecast power of the wind farm as a whole for the next 24 hours; the ultra-short-term forecast power of the wind farm as a whole is the forecast power of the wind farm as a whole for the next 4 hours.

[0013] Secondly, the present invention also discloses a short-term-ultra-short-term joint power prediction system that takes into account the influence of wind turbine wakes to implement the aforementioned method, comprising a model training module and a power prediction module; the model training module is used to acquire the physical information, historical operating data, and historical numerical weather prediction data of the wind farm to be predicted, and to preprocess the historical operating data, and then use the numerical weather prediction model to obtain the predicted wind speed at the hub height of each wind turbine in the wind farm and the predicted power of the wind farm as a whole within a preset time range; then, the obtained predicted wind speed and predicted power are used to perform short-term wind power prediction model. The model is trained using the following methods: the output of the short-term wind power prediction model is the corrected forecast power of the entire wind farm; the historical real-time output power of the wind farm is obtained, and the ultra-short-term wind power prediction model is trained using the forecast wind speed, the corrected forecast power, and the historical real-time output power; the power prediction module is used to obtain the short-term and ultra-short-term forecast power of the entire wind farm based on the current operating data of the wind farm, the current numerical weather forecast data of the wind farm's location, and the real-time output power of the wind farm at the previous moment, using the trained short-term wind power prediction model and the trained ultra-short-term wind power prediction model.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention introduces the influence of wind farm wake in the numerical weather prediction model stage, which can effectively characterize the influence of the interaction of wind turbines inside the wind farm on wind speed and power; the results of the short-term wind power prediction model are used to constrain the ultra-short-term prediction of the ultra-short-term wind power prediction model, so as to realize the synergistic fusion of prediction information at different time scales; it is applicable to complex terrain and large-scale wind farms, and has strong engineering practicality. Attached Figure Description

[0015] Figure 1 is a flowchart of the short-term-ultra-short-term combined power prediction method of the present invention that takes into account the influence of wind turbine wake; Figure 2 is a comparison diagram between the results of the short-term wind power prediction model of the present invention and the corresponding actual values; Figure 3 is a comparison diagram between the results of the ultra-short-term wind power prediction model of the present invention and the corresponding actual values. Detailed Implementation

[0016] The present invention will be further described and illustrated below with reference to specific embodiments. The embodiments described are merely examples of the content of this disclosure and do not limit the scope of the invention. The technical features of each embodiment in the present invention can be combined accordingly, provided that there is no mutual conflict.

[0017] To address the shortcomings of existing wind power forecasting methods, such as insufficient accuracy in meteorological characterization at the wind farm scale, inadequate consideration of wind turbine wake effects, and limited synergy between short-term and ultra-short-term forecasts, this invention proposes a combined short-term and ultra-short-term wind power forecasting method and system that takes into account the influence of wind turbine wake. This invention utilizes a numerical weather prediction model incorporating a parameterized wind farm model and a wind turbine wake model to perform dynamic downscaling of numerical weather prediction results at the wind farm scale. Furthermore, it combines this with an artificial intelligence model to construct both short-term and ultra-short-term wind power forecasting models, achieving the synergistic fusion of prior meteorological information, wind farm physical information, and real-time operational data, thereby improving the accuracy and stability of wind power forecasting.

[0018] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0019] As shown in Figure 1, the short-term-ultra-short-term combined power prediction method for wind farm power prediction of the present invention, which takes into account the influence of wind turbine wake, includes the following steps: S1: Data acquisition and processing: acquiring the physical information and historical operating data of the wind farm to be predicted, and preprocessing the historical operating data; acquiring historical numerical weather forecast data of the location of the wind farm; using a numerical weather forecast model to perform meteorological simulation based on the physical information, the preprocessed historical operating data and the historical numerical weather forecast data, to obtain the forecast wind speed at the hub height of each wind turbine in the wind farm and the forecast power of the wind farm as a whole within a preset time range.

[0020] In this embodiment, relevant data of the target wind farm are first acquired and processed to construct the data foundation required for wind power prediction. The data includes the wind farm's physical information, historical operational data, and historical numerical weather forecast data for the wind farm's location.

[0021] The physical information of the wind farm is used to drive the parameterized calculation of the wind farm in the numerical weather prediction model. Specifically, it includes parameters such as the type of each wind turbine in the wind farm, rated power, power curve, thrust curve, hub diameter and hub height, and the number of wind turbines in the wind farm.

[0022] Historical operational data includes the actual power output sequence of the wind farm and the corresponding wind speed sequence, which can be obtained from the wind farm monitoring system or operational database. Because historical operational data may contain missing data, abrupt changes, or records that do not conform to physical laws during acquisition, communication, or equipment status changes, outlier processing should be performed before inputting it into the numerical weather prediction model. Specifically, the wind speed sequence in the historical operational data is first identified and corrected for anomalies. Wind speed anomalies include at least abnormal states that remain constant for extended periods, anomalies that significantly exceed reasonable physical ranges, and anomalies that exhibit negative values ​​or sharp spikes. Identified abnormal wind speed points are then removed, interpolated, or replaced to obtain a corrected wind speed sequence. Subsequently, based on the corrected wind speed sequence, the actual power output sequence in the historical operational data is identified for anomalies. Power anomalies include at least negative power, power exceeding the wind farm's physical capacity limit, and abnormal power points caused by power-limited operation within the normal wind speed range. To avoid abnormal power points interfering with model learning, this embodiment uses the theoretical power characteristics of wind turbines to correct power anomalies. Specifically, a reasonable power range is constructed under the constraint of wind speed-power relationship. Power data falling outside the reasonable range is replaced with the theoretical power value under the corresponding wind speed conditions, while power data falling within the reasonable range is retained. This results in historical wind speed and historical power data samples with stronger physical consistency and better continuity.

[0023] Historical numerical weather forecast data and the historical operational data are contemporaneous data, which are the original archived numerical weather forecast data. When obtaining the original archived numerical weather forecast data contemporaneous with the historical operational data of the target wind farm, considering that the wind farm needs to submit the wind power forecast results from 0:00 to 24:00 of the next day to the dispatch department before 6:00 on the same day, this embodiment selects the numerical weather forecast data released at 12 UTC on the previous day as the input for meteorological simulation to ensure that the forecast results can be generated and submitted before the business deadline. Further, the numerical weather forecast data is input into the numerical weather forecast model (i.e., the WRF (Weather Research and Forecasting) model) for dynamic downscaling. The numerical weather forecast model is equipped with a wind farm parameterized model and a wind turbine wake model, which are used to characterize the momentum and energy impact of wind turbines on the local wind field and the impact of wake effects on the spatial distribution of wind speed during meteorological simulation. Simultaneously, the physical information of the wind farm and preprocessed historical operational data are input into the numerical weather prediction model. The numerical weather prediction model performs continuous meteorological simulation calculations according to a preset time configuration. Its operation includes a preheating stage to eliminate the influence of initial field imbalances and a formal forecast stage to output target prediction results. The formal forecast stage covers a prediction time range of 0-24 hours to support the short-term wind power prediction needs for the following day. The predicted wind speed time series at the hub height of each wind turbine in the wind farm is obtained through the numerical weather prediction model, and the overall predicted power time series of the wind farm is obtained based on the wind farm parameterized model. That is, the predicted wind speed at the hub height of each wind turbine and the overall predicted power of the wind farm are obtained within the preset time range. The predicted wind speed at the hub height of each wind turbine and the overall predicted power of the wind farm obtained within the preset time range serve as the input data basis for subsequent short-term and ultra-short-term wind power prediction models.

[0024] S2: Construction and training of short-term wind power prediction model: Obtain a short-term wind power prediction model, and use the predicted wind speed and predicted power obtained in S1 to train the short-term wind power prediction model to obtain a trained short-term wind power prediction model; wherein, the output of the short-term wind power prediction model is the corrected overall predicted power of the wind farm.

[0025] After completing the data acquisition and processing in S1, a short-term wind power prediction model is constructed and trained to correct the error of the overall wind farm forecast power output by the numerical weather prediction model and generate the corrected overall wind farm forecast power.

[0026] In this embodiment, the short-term wind power prediction model adopts the iTransformer model. During the model training phase, the actual output power of the entire wind farm within the time range corresponding to the predicted wind speed of S1 is also obtained. This actual output power, along with the predicted wind speed and predicted power obtained from S1, are input into the short-term wind power prediction model to train the model with the goal of minimizing the loss function. This actual output power serves as the training label for the short-term wind power prediction model, and the output of the short-term wind power prediction model is the corrected predicted power of the entire wind farm. The loss function of the short-term wind power prediction model is the mean squared error loss function, which minimizes the error between the predicted power and the actual power. Through the above training process, the short-term wind power prediction model can learn the systematic deviation pattern between numerical weather prediction results and actual wind power, thereby outputting the corrected predicted power of the entire wind farm while maintaining consistency with the weather forecast trend, i.e., the corrected short-term wind power prediction result. The short-term wind power prediction result serves as one of the important inputs for the subsequent ultra-short-term wind power prediction model, guiding the ultra-short-term prediction in judging future power change trends.

[0027] S3: Construction and Training of Ultra-Short-Term Wind Power Prediction Model: Obtain the ultra-short-term wind power prediction model, obtain the historical real-time output power of the wind farm, obtain the partial forecast power data corresponding to the preset time range predicted by the ultra-short-term wind power prediction model from the corrected forecast power of S2, and obtain the partial forecast wind speed data corresponding to the preset time range predicted by the ultra-short-term wind power prediction model from the forecast wind speed of S1; use the historical real-time output power, partial forecast power data, and partial forecast wind speed data of S3 to train the ultra-short-term wind power prediction model to obtain the trained short-term wind power prediction model; wherein, the output of the ultra-short-term wind power prediction model is the overall forecast power of the wind farm within the preset time range.

[0028] After obtaining the corrected overall forecast power of the wind farm from the S2 output, an ultra-short-term wind power prediction model is further constructed and trained to make high-precision predictions of future power changes in the wind farm over a shorter time scale.

[0029] In this embodiment, the ultra-short-term wind power prediction model also adopts the iTransformer model architecture. During the model training phase, the historical real-time output power, partial forecast power data, and partial forecast wind speed data of S3 are used as inputs to the ultra-short-term wind power prediction model. The ultra-short-term wind power prediction model outputs the overall forecast power of the wind farm within a preset time range. The preset time range of the ultra-short-term wind power prediction model is shorter than that of the short-term wind power prediction model. The historical real-time output power is used to characterize the short-time autocorrelation characteristics of wind power, the short-term wind power prediction results are used to provide trend-based prior information on future power changes, and the forecast wind speed at the hub height of each wind turbine in S1 is used to reflect the evolution trend of meteorological conditions within the ultra-short-term prediction period. The label used for model training is the actual output power of the wind farm within the corresponding time period, and the training loss function is the mean squared error loss function to minimize the error between the predicted power and the actual power.

[0030] Suppose we want to use an ultra-short-term wind power prediction model to predict the power of a wind farm within the next 4 hours at a certain moment. The input of the ultra-short-term wind power prediction model is the real-time output power of the previous moment, the predicted power data for the next 4 hours at that moment, and the predicted wind speed data for the next 4 hours at that moment. The actual output power of the wind farm within the next 4 hours at that moment is used as the label.

[0031] S4: Power Prediction: Obtain the current operational data of the wind farm and the current numerical weather forecast data for the wind farm's location. Based on the wind farm's physical information, operational data, and numerical weather forecast data, use a numerical weather forecast model to obtain the predicted wind speed at the hub height of each wind turbine and the overall predicted power of the wind farm within a preset time range. Based on the predicted wind speed and predicted power, use a trained short-term wind power prediction model to obtain the corrected overall predicted power of the wind farm. Obtain the wind farm's real-time output power at the previous moment, and combine it with the predicted wind speed and corrected predicted power to obtain the overall predicted power of the wind farm within a preset time range using a trained ultra-short-term wind power prediction model. The predicted power of the wind farm obtained by the short-term wind power prediction model is the predicted power of the wind farm for the next 24 hours; the predicted power of the wind farm obtained by the ultra-short-term wind power prediction model is the predicted power of the wind farm for the next 4 hours.

[0032] This invention, based on a numerical weather prediction model incorporating a wind farm parameterization model and a wind turbine wake model, performs dynamic downscaling simulations of the wind farm's location. It can characterize the momentum extraction effect of wind turbines on the local wind field and the impact of wake effects on wind speed distribution at a meteorological scale, thereby obtaining more physically consistent hub height-predicted wind speed and overall wind farm power forecasts. Compared to numerical weather prediction results that do not consider wind farm parameterization and wake effects, the wind speed and power forecasts obtained by this invention are more closely aligned with the actual operating characteristics of wind farms, providing reliable physical prior constraints for the subsequent correction of short-term wind power prediction models and the wind power prediction of ultra-short-term wind power prediction models.

[0033] Furthermore, this invention corrects the errors in numerical weather prediction models by constructing a short-term wind power prediction model, and incorporates short-term prediction results and real-time wind farm operation information into the ultra-short-term prediction stage, establishing a collaborative mechanism between short-term and ultra-short-term wind power prediction. Compared with existing technical solutions that model short-term and ultra-short-term predictions independently, this invention can fully utilize the meteorological evolution trend information contained in the short-term prediction, and incorporate historical power data from wind farms into the ultra-short-term prediction stage, enabling the prediction model to reflect the current operating status of wind farms in a timely manner and enhancing the responsiveness of the prediction results to instantaneous power fluctuations.

[0034] In a specific embodiment of the present invention, the present invention also discloses a short-term-ultra-short-term joint power prediction system that takes into account the influence of wind turbine wake, for implementing the short-term-ultra-short-term joint power prediction method that takes into account the influence of wind turbine wake. The system includes a model training module and a power prediction module. The model training module acquires the physical information, historical operating data, and historical numerical weather forecast data of the wind farm to be predicted, and preprocesses the historical operating data. Then, it uses the numerical weather forecast model to obtain the predicted wind speed at the hub height of each wind turbine and the overall predicted power of the wind farm within a preset time range. The system then uses the obtained predicted wind speed and predicted power to train a short-term wind power prediction model. The output of the short-term wind power prediction model is the corrected overall predicted power of the wind farm. The system also acquires the historical real-time output power of the wind farm and uses the predicted wind speed, the corrected predicted power, and the historical real-time output power to train an ultra-short-term wind power prediction model. The power prediction module uses the current operating data of the wind farm, the current numerical weather forecast data at the wind farm's location, and the previous real-time output power of the wind farm to obtain the overall short-term and ultra-short-term predicted power of the wind farm using the trained short-term and ultra-short-term wind power prediction models. In one embodiment, a specific offshore wind farm is used as the research object. The wind farm includes multiple wind turbines, each with known parameters such as spatial arrangement, hub height, and rated power. The dataset is obtained from the offshore wind farm's SCADA (Supervisory Control and Data Acquisition) system and contains sample data from December 2023 to November 30, 2024, with 15-minute intervals. The training set spans from 00:00 on December 1, 2023 to 23:45 on September 30, 2024; followed by the validation set from October 1, 2024 to November 1, 2024; and the test set from November 1, 2024 to November 30, 2024.

[0035] In the example short-term forecast, the WRF model uses a forecast dataset archived by the European Centre for Medium-Range Weather Forecasts (ECMWF) (spatial resolution 0.125°×0.125°, temporal resolution 1 hour) to provide initial fields and boundary conditions for dynamic downscaling simulations. The WRF model employs a double-nested region configuration: the outermost region (D01) covers a grid with a resolution of 9 km (80×80 grid points), nested within it a middle region with a resolution of 3 km (D02, 100×100 grid points). Furthermore, the WRF model sets up a total of 78 layers in the vertical direction to accurately resolve wind field characteristics within the wind turbine rotor range, with 24 of these layers densely distributed below 200 meters above the ground.

[0036] In both the short-term and ultra-short-term wind power prediction models in this embodiment, the iTransformer model is used as the predictor, and the PSO algorithm is used to adjust the hyperparameters.

[0037] Figure 2 shows a comparison between the short-term wind power forecast results and the actual power values ​​for November (30 days). As can be seen from Figure 2, the constructed short-term wind power forecast model can accurately depict the overall trend of wind farm power changes throughout the forecast period. It remains consistent with the actual power changes during both the power increase and decrease phases, indicating that the short-term forecast model can effectively reflect the power evolution characteristics of wind farms on a daily scale, providing stable and reliable trend information input for subsequent ultra-short-term forecasts.

[0038] Figure 3 shows a comparison between the prediction results and actual power values ​​of the ultra-short-term wind power prediction model at the 16th prediction step within the same time range, where all power values ​​have been normalized. As can be seen from Figure 3, based on the short-term prediction results as prior information, the ultra-short-term prediction model further incorporates real-time operational information to effectively correct for rapid fluctuations and local changes in power, and the prediction curve closely follows the actual power changes. Therefore, the prediction method proposed in this invention significantly improves the response capability to power fluctuations on short timescales while ensuring overall trend consistency.

[0039] Combining Figures 2 and 3, it can be concluded that the present invention achieves effective transmission of prediction information at different time scales by synergistically coupling short-term wind power prediction results with ultra-short-term prediction models. This enables the prediction results to have both long-term trend stability and short-term dynamic correction capabilities, thereby improving the accuracy of wind power prediction and the reliability of engineering applications.

[0040] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. Those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A short-term-ultra-short-term combined power prediction method considering the influence of wind turbine wake, used for wind farm power prediction; characterized in that, The method includes the following steps: 1) Obtaining the physical information, historical operating data, and historical numerical weather prediction data of the wind farm to be predicted for power, and preprocessing the historical operating data, then using a numerical weather prediction model to obtain the predicted wind speed at the hub height of each wind turbine and the predicted power of the wind farm as a whole within a preset time range; 2) Training a short-term wind power prediction model using the predicted wind speed and predicted power obtained in step 1); wherein the output of the short-term wind power prediction model is the corrected... 3) Obtain the historical real-time output power of the wind farm, and train the ultra-short-term wind power prediction model using the predicted wind speed in step 1), the predicted power in step 2), and the historical real-time output power; 4) Based on the current operating data of the wind farm, the numerical weather forecast data at the current location of the wind farm, and the real-time output power of the wind farm at the previous moment, obtain the short-term and ultra-short-term predicted power of the wind farm as a whole using the trained short-term wind power prediction model and the trained ultra-short-term wind power prediction model.

2. The method according to claim 1, characterized in that, In step 1), the physical information includes the model type, rated power, power curve, thrust curve, hub diameter and hub height of each wind turbine in the wind farm, as well as the number of wind turbines in the wind farm; the historical operating data includes the actual output power sequence of the wind farm and the corresponding wind speed sequence; the historical operating data and the historical numerical weather forecast data are data from the same period.

3. The method according to claim 2, characterized in that, In step 1), the preprocessing of historical operating data includes: firstly, identifying and correcting anomalies in the wind speed sequence of the historical operating data to obtain a corrected wind speed sequence; then, based on the corrected wind speed sequence, identifying and correcting anomalies in the actual output power sequence of the historical operating data to obtain a corrected actual output power sequence; wherein, the actual output power sequence of the historical operating data is corrected using the theoretical power characteristics of the wind turbine.

4. The method according to claim 1, characterized in that, In step 1), the numerical weather prediction model is the WRF model, which is equipped with a wind farm parameterization model and a wind turbine wake model.

5. The method according to claim 1, characterized in that, In step 2), the short-term wind power prediction model is the iTransformer model. When training the short-term wind power prediction model, the actual output power of the wind farm as a whole within the time range corresponding to the predicted power in step 1) is also obtained, and the actual output power is used as a label for training the short-term wind power prediction model.

6. The method according to claim 1, characterized in that, In step 3), the ultra-short-term wind power prediction model is an iTransformer model; training the ultra-short-term wind power prediction model using the predicted wind speed from step 1), the predicted power from step 2), and the historical real-time output power includes: acquiring partial predicted power data from the corrected predicted power from step 2) that corresponds to the preset time range predicted by the ultra-short-term wind power prediction model, and acquiring partial predicted wind speed data from the predicted wind speed from step 1) that corresponds to the preset time range predicted by the ultra-short-term wind power prediction model; training the ultra-short-term wind power prediction model using the historical real-time output power, partial predicted power data, and partial predicted wind speed data from step 3) to obtain a trained short-term wind power prediction model; wherein, the output of the ultra-short-term wind power prediction model is the overall predicted power of the wind farm within the preset time range.

7. The method according to claim 1, characterized in that, Step 4) includes: based on the physical information of the wind farm, the current operating data of the wind farm, and the current numerical weather forecast data of the wind farm's location, using a numerical weather forecast model to obtain the forecast wind speed at the hub height of each wind turbine in the wind farm within a preset time range, as well as the forecast power of the wind farm as a whole; based on the forecast wind speed and forecast power, using a trained short-term wind power prediction model to obtain the corrected short-term forecast power of the wind farm as a whole; based on the real-time output power of the wind farm at the previous moment, the forecast wind speed, and the corrected forecast power, using a trained ultra-short-term wind power prediction model to obtain the ultra-short-term forecast power of the wind farm as a whole.

8. The method according to claim 7, characterized in that, The overall short-term forecast power of the wind farm is the forecast power of the wind farm over the next 24 hours; the overall ultra-short-term forecast power of the wind farm is the forecast power of the wind farm over the next 4 hours.

9. A short-term-ultra-short-term joint power prediction system that implements the method of any one of claims 1-8 and takes into account the effect of wind turbine wake, characterized in that, The system includes a model training module and a power prediction module. The model training module acquires the physical information, historical operating data, and historical numerical weather forecast data of the wind farm's location for power prediction. It preprocesses the historical operating data and then uses the numerical weather forecast model to obtain the predicted wind speed at the hub height of each wind turbine and the overall predicted power of the wind farm within a preset time range. The system then uses the obtained predicted wind speed and predicted power to train a short-term wind power prediction model. The output of the short-term wind power prediction model is the corrected overall predicted power of the wind farm. The system also acquires the historical real-time output power of the wind farm and uses the predicted wind speed, corrected predicted power, and historical real-time output power to train an ultra-short-term wind power prediction model. The power prediction module uses the current operating data of the wind farm, the current numerical weather forecast data at the wind farm's location, and the previous real-time output power of the wind farm to obtain the overall short-term and ultra-short-term predicted power of the wind farm using the trained short-term and ultra-short-term wind power prediction models.

Images