Wind power generation amount prediction system reflecting multi-period wind power generation amount tendency

Abstract

The present invention provides a wind power generation amount prediction system comprising: an input data generation unit that generates, for respective institutions from institution-specific numerical prediction data, input data including wind speed data, wind direction data, and air density data; a wind power generation amount prediction unit that predicts, for the respective institutions, initial wind power generation amounts of a wind power plant by applying, for the respective institutions, the input data generated for the respective institutions to respective four-dimensional power curve models; and a wind power generation amount correction unit that predicts a final wind power generation amount by correcting the initial wind power generation amounts so as to reflect a wind power generation amount tendency in a multi-period including a short period and a long period, and integrating the initial wind power generation amounts corrected for the respective institutions.

Description

A wind power generation prediction system that reflects multi-period wind power generation trends.

[0001] The present invention relates to a wind power generation prediction system, and more specifically, to a wind power generation prediction system that predicts wind power generation reflecting wind power generation trends.

[0002] The present invention was made under the support of the Ministry of Trade, Industry and Energy of the Republic of Korea under the project number 2021202090053A (project unique number 1415186830). The research management specialized institution of the project is the Korea Institute of Energy Technology Evaluation and Planning, the research project name is 'Development of Core Technology for Energy Demand Management (E-Tech)', the research project name is 'Development of Core Technology and Demonstration Research for Cloud Energy Management System for Building Distributed Business Sites', the project performing institution is Haejum Co., Ltd., and the research period is from January 1, 2023 to December 31, 2023.

[0003] This patent application claims priority to Republic of Korea Patent Application No. 10-2023-0184517, filed with the Korean Intellectual Property Office on December 18, 2023, the disclosure of which is incorporated herein by reference.

[0004] Wind power is a major source of renewable energy, and its importance is being recognized worldwide. However, accurate predictions of wind power generation are difficult because it is heavily dependent on various factors, particularly weather conditions. Accurate power generation forecasts are essential for maintaining the stability of the power grid, maximizing economic benefits, and efficiently managing energy storage and distribution.

[0005] The most critical factor in wind power generation forecasting is the wind power of the wind farm. Wind power is based on numerical weather prediction (NWP) data collected by various agencies. However, there are differences between the NWP data provided by various agencies, leading to significant differences in the accuracy of wind power generation forecasts depending on the NWP data used.

[0006] Therefore, to overcome these problems, more accurate and reliable wind power generation prediction technology is needed.

[0007] The technical problem to be achieved by the present invention is to provide a wind power generation prediction system capable of predicting wind power generation with reliability by reflecting multi-period wind power generation trends.

[0008] In order to solve these problems, a wind power generation prediction system according to an embodiment of the present invention includes an input data generation unit that generates input data including wind speed data, wind direction data, and air density data for each agency from numerical forecast data for each agency; a wind power generation prediction unit that applies the input data generated for each agency to a four-dimensional power curve model to predict the initial wind power generation of a wind power plant for each agency; and a wind power generation correction unit that corrects the initial wind power generation so that a multi-period wind power generation tendency including long-term and short-term periods is reflected, and predicts the final wind power generation by integrating the corrected initial wind power generation for each agency.

[0009] The above input data generation unit may include a numerical forecast data collection unit that collects numerical forecast data for each institution from a plurality of institutions; and a numerical forecast data correction unit that generates wind speed data, wind direction data, and air density data for each institution from the numerical forecast data by taking into account the topography of the wind power plant and the height of the wind turbine.

[0010] The above numerical forecast data correction unit separates the wind speed and wind direction collected from the institution into east wind (U) wind speed and north wind (V) wind speed, and then corrects them by considering the topography and wind turbine height of the area where the wind power plant is located, and then corrects the wind speed deviation using a wind speed correction artificial intelligence model, and then uses trigonometric functions to generate integrated wind speed data and wind direction data using the corrected east wind (U) wind speed and north wind (V) wind speed.

[0011] The above four-dimensional power curve model may be a model that represents the relationship between wind power generation according to wind speed, wind direction, and air density.

[0012] The above wind power generation correction unit may include a first power generation correction unit that corrects the initial wind power generation amount to reflect a long-term wind power generation tendency and predicts the corrected power generation amount, and a second power generation correction unit that corrects the corrected power generation amount to reflect a short-term wind power generation tendency and predicts the final wind power generation amount.

[0013] The first power generation correction unit may include a numerical forecast data conversion unit that converts the numerical forecast data for each agency into periodic variable data for a preset time before and after a power generation prediction time, and a corrected power generation prediction unit that predicts the first corrected power generation through the periodic variable data and the initial wind power generation amount through a first post-processing model.

[0014] The above first post-processing model can apply an ensemble technique to multiple machine learning models.

[0015] The above first post-processing model can be trained using the initial wind power generation for each agency over the past year, wind speed data at time t in the past, wind speed data at time t-1 in the past, wind speed data at time t-2 in the past, the prediction cycle for each agency, and actual wind power generation for the past year as learning data based on the time point of power generation prediction.

[0016] The above first post-processing model can be generated using a boosting method for a polynomial regression model and a LightGBM model.

[0017] The second power generation correction unit may include a wave function transform unit that generates the corrected power generation and the actual power generation by using a wavelet transform model in a short-term time domain in the past based on the power generation prediction time point of the numerical forecast data for each agency by decomposing them into high-frequency values ​​and low-frequency values, and a final power generation prediction unit that predicts the final wind power generation at the power generation prediction time point using the low-frequency values ​​of the corrected power generation and the low-frequency values ​​of the actual power generation through a second post-processing model.

[0018] The above wave function transformation unit can generate the actual power generation at the time of power generation prediction used as learning data of the second post-processing model using LSTM (Long short time memory) and then decompose it into high-frequency values ​​and low-frequency values.

[0019] The above second post-processing model may use at least one of GBR (Gradiant Boost Regression), LGBM (Light Gradiant Boost Machine), and MLP (Multi-layer Perceptron).

[0020] The above final power generation prediction unit can select a model with the smallest power generation prediction error among GBR (Gradiant Boost Regression), LGBM (Light Gradiant Boost Machine), and MLP (Multi-layer Perceptron) according to a preset period of time during the power generation prediction date, depending on the power generation prediction time.

[0021] In addition to the technical problems of the present invention mentioned above, other features and advantages of the present invention are described below or may be clearly understood by a person skilled in the art to which the present invention pertains from such description and explanation.

[0022] According to the present invention as described above, the following effects are achieved.

[0023] The present invention can predict reliable wind power generation by using numerical forecast data from multiple institutions.

[0024] The present invention can derive a reliable initial wind power generation amount by using a four-dimensional power curve model that represents the relationship between wind power generation amount according to wind speed, wind direction, and air density of a wind power plant.

[0025] The present invention can predict reliable wind power generation by correcting the initial wind power generation amount so that the multi-period wind power generation amount trend is reflected.

[0026] In addition, other features and advantages of the present invention may be newly discovered through the embodiments of the present invention.

[0027] FIG. 1 is a conceptual block diagram of a wind power generation prediction system according to one embodiment of the present invention.

[0028] FIG. 2 is a conceptual block diagram of an input data generation unit according to one embodiment of the present invention.

[0029] Figure 3 is a graph showing monthly wind speed deviation correction by a numerical forecast data correction unit according to one embodiment of the present invention.

[0030] FIG. 4 is a diagram showing the relationship between wind speed, air density, and wind power generation to explain a four-dimensional power curve model according to one embodiment of the present invention.

[0031] FIG. 5 is a diagram showing the relationship between wind speed, wind direction, and wind power generation to explain a four-dimensional power curve model according to one embodiment of the present invention.

[0032] Figure 6 is a conceptual block diagram of a wind power generation compensation unit according to one embodiment of the present invention.

[0033] FIG. 7 is a conceptual block diagram of a first power generation compensation unit according to one embodiment of the present invention.

[0034] Figure 8 is a conceptual block diagram of a second power generation compensation unit according to one embodiment of the present invention.

[0035] When adding reference numbers to components of each drawing in this specification, it should be noted that identical components are given the same numbers as much as possible even if they are shown in different drawings.

[0036] Meanwhile, the meanings of the terms described in this specification should be understood as follows.

[0037] Singular expressions should be understood to include plural expressions unless the context clearly defines otherwise, and terms such as “first”, “second”, etc. are intended to distinguish one component from another, and the scope of rights should not be limited by these terms.

[0038] The terms "include" or "have" should be understood as not excluding in advance the presence or addition of one or more other features, numbers, steps, operations, components, parts or combinations thereof.

[0039] Hereinafter, preferred embodiments of the present invention designed to solve the above problems will be described in detail with reference to the attached drawings.

[0040] FIG. 1 is a conceptual block diagram of a wind power generation prediction system according to one embodiment of the present invention, and FIG. 2 is a conceptual block diagram of an input data generation unit according to one embodiment of the present invention.

[0041] Referring to FIG. 1, a wind power generation prediction system (1000) according to one embodiment of the present invention includes an input data generation unit (100), a wind power generation prediction unit (200), and a wind power generation correction unit (300).

[0042] The input data generation unit (100) can collect numerical forecast data, which is a forecast value for the weather, from multiple organizations, and then generate input data for each organization, including wind speed data, wind direction data, and air density data for each organization.

[0043] Referring to FIG. 2, an input data generation unit (100) according to one embodiment of the present invention may include a numerical forecast data collection unit (110) that collects numerical forecast data for each institution from a plurality of institutions, and a numerical forecast data correction unit (120) that generates wind speed data, wind direction data, and air density data for each institution from the numerical forecast data by taking into account the height of a wind turbine of a wind power plant.

[0044] Numerical forecast data for multiple institutions according to one embodiment can be collected through the Global Forcast System (GFS), the Local Data Assimilation and Prediction System (LDAPS), the Korea Meteorological Administration's local forecast, and the European Centre for Medium-Range Weather Forecasts (ECMWF).

[0045] Numerical forecast data from each agency may include wind speed, wind direction, pressure, temperature, relative humidity, absolute humidity, and dew point. Because each agency uses different variables, forecast cycles, and spatial resolutions, numerical forecast data for the same time period and region can differ. This means that the numerical forecast data used can significantly affect wind power generation forecasts.

[0046] In order to solve this problem, the wind power generation prediction system (1000) according to an embodiment of the present invention generates input data for each of a plurality of agency-specific numerical forecast data, applies a four-dimensional power curve model to each agency-specific input data to predict the initial wind power generation amount for each agency, and then corrects the initial wind power generation amount for each agency to reflect the multi-period wind power generation amount trend to predict the integrated final wind power generation amount, thereby making it possible to predict reliable wind power generation amount even when using a plurality of different agency-specific numerical forecast data.

[0047] Hereinafter, a process of predicting wind power generation using numerical forecast data collected from a single institution by a wind power generation prediction system (1000) according to an embodiment is described. Unless otherwise specifically described, the process of predicting wind power generation using numerical forecast data collected from multiple other institutions is also the same.

[0048] The numerical forecast data correction unit (120) can generate wind speed data, wind direction data, and air density data from numerical forecast data by taking into account the height of the wind turbine of the wind power plant.

[0049] Since each institution provides different numerical forecast data at a height between 10 m and 100 m from the ground, the numerical forecast data correction unit (120) can generate wind speed data, wind direction data, and air density data by correcting the data to a value that takes into account the height of the wind turbine of the wind power plant to be predicted.

[0050] The numerical forecast data correction unit (120) can calculate air density using a physical calculation formula based on pressure, temperature, relative humidity, absolute humidity, and dew point. At this time, air density data can be generated by varying the weight according to the condition of the ground surface in the area where the wind power plant is located. Here, the condition of the ground surface can be various, such as whether snow is accumulated or whether it is in a vegetation state.

[0051] As an example, the state of the ground surface of a wind power plant photographed using a satellite is analyzed to determine the state of the ground surface, and then the air density is calculated using a physical calculation formula based on pressure, temperature, and relative humidity by considering weights, and then converted to the air density at the height of the wind turbine of the wind power plant.

[0052] The numerical forecast data correction unit (120) separates the wind speed and wind direction collected from the agency into east wind (U) wind speed and north wind (V) wind speed, and then corrects them by considering the topography of the area where the wind power plant is located and the height of the wind turbine. Afterwards, the wind speed deviation can be corrected using a wind speed correction artificial intelligence model. The east wind (U) wind speed and north wind (V) wind speed corrected in this way can be used to generate integrated wind speed data and wind direction data using trigonometric functions.

[0053] The numerical forecast data correction unit (120) can generate wind speed data at the height of the wind turbine of a wind power plant by taking into account the topography of the area where the wind power plant is located after dividing the terrain into three types: flat area, highland area, and uneven area.

[0054] At this time, since the predicted wind speed by institution shows different deviations by month and hour for each institution, the deviation can be corrected by separating periods with similar characteristics and then using a wind speed correction artificial intelligence model.

[0055] The wind speed correction AI model can be trained based on numerical forecast data and actual wind speed data from each institution that exist monthly and hourly, and various machine learning models such as GBR (Gradiant Boost Regression), LGBM (Light Gradient Boost Machine), polynomial regression, and MLP (Multi-layer Perceptron) can be used.

[0056] GBR and LGBM are tree-based ensemble models that demonstrate strong performance in ensemble learning, while polynomial regression can effectively model nonlinear relationships. Multi-Layer Progressive Models (MLPs) are versatile models that can be applied to a variety of problems by adjusting the depth and width of the neural network.

[0057] Figure 3 is a graph showing monthly wind speed deviation correction by a numerical forecast data correction unit according to one embodiment of the present invention.

[0058] Referring to Figure 3, the X-axis represents actual wind speed, and the Y-axis represents wind speed predicted by one of the numerical forecasting agencies. When a trend line is drawn based on the distribution of predicted wind speeds versus actual wind speeds, it can be seen that there are monthly variations.

[0059] The numerical forecast data correction unit (120) according to one embodiment of the present invention can correct wind speed by dividing it by month.

[0060] The wind power generation prediction unit (200) can predict the initial wind power generation of a wind power plant by applying input data to a four-dimensional power curve model.

[0061] FIG. 4 is a diagram showing the relationship between wind speed, air density, and wind power generation to explain a four-dimensional power curve model according to one embodiment of the present invention, and FIG. 5 is a diagram showing the relationship between wind speed, wind direction, and wind power generation to explain a four-dimensional power curve model according to one embodiment of the present invention.

[0062] The four-dimensional power curve model is a model that represents the relationship between wind power generation according to wind speed, wind direction, and air density. However, since it can only be represented in three dimensions in a diagram, Fig. 4 represents wind speed (X-axis), air density (Y-axis), and wind power generation (Z-axis), and Fig. 5 represents wind speed (X-axis), wind direction (Y-axis), and wind power generation (Z-axis).

[0063] Referring to Figures 4 and 5, if cut in multiple directions based on a specific wind direction or specific air density, the shape of the conventional power curve can be obtained.

[0064] The wind power generation prediction unit (200) can predict wind power generation by inputting wind speed data, wind direction data, and air density data for each agency into a four-dimensional power curve model.

[0065] The 4D power curve model can be created using various machine learning models such as GBR (Gradiant Boost Regression), LGBM (Light Gradient Boost Machine), polynomial regression, and MLP (Multi-layer Perceptron), and among them, using GBR may be the most desirable.

[0066] A wind power generation prediction system (1000) according to one embodiment of the present invention can predict a reliable initial wind power generation amount by applying wind speed, wind direction, and air density, which have the greatest influence on wind power generation prediction, to a four-dimensional power curve model after correcting the wind speed deviation and considering the topography of a wind power plant.

[0067] The wind power generation correction unit (300) corrects the initial wind power generation for each institution so that the long-term wind power generation trend is reflected, and can predict the final wind power generation so that the short-term wind power generation trend is reflected based on the corrected initial wind power generation for each institution.

[0068] FIG. 6 is a conceptual block diagram of a wind power generation correction unit (300) according to one embodiment of the present invention, FIG. 7 is a conceptual block diagram of a first generation correction unit (310) according to one embodiment of the present invention, and FIG. 8 is a conceptual block diagram of a second generation correction unit (320) according to one embodiment of the present invention.

[0069] Referring to FIGS. 6 and 7, a wind power generation correction unit (300) according to one embodiment of the present invention may include a first generation correction unit (310) and a second generation correction unit (320).

[0070] The first power generation correction unit (310) can correct the initial wind power generation amount so that the long-term wind power generation amount trend is reflected, including a numerical forecast data conversion unit (311) and a corrected power generation amount prediction unit (312).

[0071] Here, long-term wind power generation trends roughly mean wind power generation trends over a period of one year.

[0072] Because the operating time and forecast cycle of each NWP model used to produce numerical forecast data for each agency are different, the numerical forecast data with high hourly accuracy differs for each agency, so it is necessary to correct them by considering long-term wind power generation trends.

[0073] The present invention can use a first post-processing model learned from data converted into a trigonometric function to become a variable with time periodicity in order to increase the reliability of initial wind power generation that is different for each institution.

[0074] A numerical forecast data conversion unit (311) according to one embodiment of the present invention can convert numerical forecast data for each institution into periodic variable data for a preset time before and after a power generation prediction time point.

[0075] At this time, the first post-processing model can be trained using the initial wind power generation for each agency over the past year based on the power generation prediction time, wind speed data at time t in the past, wind speed data at time t-1 in the past, wind speed data at time t-2 in the past, prediction cycle for each agency, and actual wind power generation as learning data.

[0076] Previously, there was a problem that the accuracy of power generation predictions differed by institution because the prediction cycles were different for each institution.

[0077] The present invention can reduce power generation prediction errors according to different prediction periods for each agency by training the first post-processing model with the initial wind power generation for each agency over the past year, wind speed data at a point in time t in the past, wind speed data at a point in time t-1 in the past, wind speed data at a point in time t-2 in the past, prediction cycles for each agency, and actual wind power generation as learning data.

[0078] The corrected power generation prediction unit (312) can predict the corrected power generation through the first post-processing model using periodic variable data and the initial wind power generation amount.

[0079] At this time, the first post-processing model can be generated using an ensemble technique for multiple machine learning models, and as an example, can be generated using a boosting method for a polynomial regression model and a LightGBM model.

[0080] In general, when the wind speed is slow, the amount of wind power generation is small, and when the wind speed is fast, the amount of wind power generation is large. Therefore, the tendency to predict wind speed by each agency is different, so the mixing ratio of multiple models used in the first post-processing model can be adjusted by weighting.

[0081] For example, the weights are set differently for each agency between 1 and 4, and the initial wind power generation of each agency is multiplied by the weight value to the nth power, added, and then raised to the 1 / (n+a) power and returned. Using the above method, the general and conservatively designed numerical forecast data that predicts close to the average value is spread more widely above and below the average value, which can reduce the tendency of under- or over-simulation.

[0082] At this time, the weights can be determined by finding appropriate values ​​through experimental methods, and the weights that can minimize the prediction error can be determined using one year of past data.

[0083] When the average wind speed at time t-1 is 3 m / s based on the power generation prediction time (t), the probability of having a wind speed that does not deviate significantly from 3 m / s is greater than the probability of the average wind speed at time t suddenly changing to 15 m / s or 20 m / s.

[0084] Numerical forecast data has time-series characteristics because it calculates changes in the atmosphere and ocean, and since wind turbines in wind power plants generate power by rotation, they have mechanical inertia and resistance, so time-series characteristics are even more evident.

[0085] The present invention is designed to recognize time-series characteristics as learning variables of the first post-processing model, thereby enabling better derivation of changes in a preset time before and after the power generation prediction time.

[0086] Referring to FIGS. 6 and 8, the second power generation correction unit (320) according to one embodiment of the present invention includes a wave function conversion unit (321) and a final power generation prediction unit (322).

[0087] The second power generation correction unit (320) can correct the corrected power generation amount so that the short-term wind power generation amount trend is reflected.

[0088] Here, short-term wind power generation trends refer to wind power generation trends over a period of approximately one week to one month.

[0089] The wave function transformation unit (321) can generate numerical forecast data for each agency by decomposing the corrected power generation and actual power generation for each agency in the short-term time domain in the past based on the power generation prediction time into high-frequency values ​​and low-frequency values ​​using a wavelet transform model.

[0090] The wave function transformation unit (321) can generate high-frequency values ​​and low-frequency values ​​by decomposing the actual wind power generation and the corrected power generation within a preset short period of time based on the power generation prediction time using a wavelet transform model and converting them into high-frequency and low-frequency bands.

[0091] At this time, if the learning period of the transform model (wavelet transform) is set from D-7 to D-1 based on the wind power generation prediction date (D), the time from D-7 to D-day can be viewed as a single continuous domain and the predicted power generation values ​​can be listed and analyzed. However, the actual power generation only exists from D-7 to D-1, and the actual power generation on the wind power generation prediction date (D) does not exist yet, so if only the time from D-7 to D-1 is viewed as a continuous time domain and wavelet decomposition is performed, the problem arises that it is not accurate.

[0092] Accordingly, the wave function conversion unit (321) according to one embodiment of the present invention can generate the actual wind power generation amount at the wind power generation prediction date (D) or prediction time using LSTM (Long short time memory), and then convert it into high frequency and low frequency bands to generate high frequency values ​​and low frequency values.

[0093] The final power generation prediction unit (322) can predict the final wind power generation by integrating the predicted date decomposition values ​​by each institution through the second post-processing model.

[0094] At this time, the second post-processing model according to one embodiment of the present invention may use at least one of GBR (Gradiant Boost Regression), LGBM (Light Gradiant Boost Machine), and MLP (Multi-layer Perceptron).

[0095] The final power generation prediction unit (322) according to one embodiment of the present invention can improve the reliability of power generation prediction by selecting a model with the smallest prediction error of power generation according to a preset period of the day of power generation prediction among GBR (Gradiant Boost Regression), LGBM (Light Gradiant Boost Machine), and MLP (Multi-layer Perceptron) depending on the power generation prediction time.

[0096] In this way, the present invention can predict reliable wind power generation by correcting the initial wind power generation amount so that the multi-period wind power generation amount trend is reflected.

[0097] It will be apparent to those skilled in the art that the present invention described above is not limited to the above-described embodiments and the attached drawings, and that various substitutions, modifications, and changes are possible within the scope that does not depart from the technical spirit of the present invention.

[0098] The present invention relates to a wind power generation prediction system, and more specifically, to a wind power generation prediction system that predicts wind power generation reflecting wind power generation trends.

Claims

1. Input data generation unit that generates input data including wind speed data, wind direction data, and air density data from numerical forecast data for each institution; A wind power generation prediction unit that applies the above input data generated by each institution to a four-dimensional power curve model to predict the initial wind power generation of a wind power plant by institution; and A wind power generation prediction system including a wind power generation correction unit that corrects the initial wind power generation to reflect multi-period wind power generation trends including long-term and short-term periods, and predicts the final wind power generation by integrating the initial wind power generation corrected by each institution.

2. In paragraph 1, The above input data generation unit, A numerical forecast data collection unit that collects numerical forecast data from multiple institutions; and A wind power generation prediction system including a numerical forecast data correction unit that generates wind speed data, wind direction data, and air density data for each agency from the numerical forecast data, taking into account the terrain of the wind power plant and the height of the wind turbine.

3. In paragraph 2, The above numerical forecast data correction unit separates the wind speed and wind direction collected from the institution into east wind (U) wind speed and north wind (V) wind speed, and then corrects them by considering the topography of the area where the wind power plant is located and the height of the wind turbine, and then corrects the wind speed deviation through a wind speed correction artificial intelligence model, and then generates integrated wind speed data and wind direction data by using trigonometric functions on the corrected east wind (U) wind speed and north wind (V) wind speed. A wind power generation prediction system.

4. In paragraph 1, A wind power generation prediction system characterized in that the above four-dimensional power curve model is a model representing the relationship between wind power generation according to wind speed, wind direction, and air density.

5. In paragraph 2, The above wind power generation compensation unit is, A wind power generation prediction system comprising a first generation amount correction unit for predicting a corrected generation amount by correcting the initial wind power generation amount so that a long-term wind power generation tendency is reflected, and a second generation amount correction unit for predicting the final wind power generation amount by correcting the corrected generation amount so that a short-term wind power generation tendency is reflected.

6. In paragraph 5, The above first power generation compensation unit, A wind power generation prediction system including a numerical forecast data conversion unit that converts the above-mentioned agency-specific numerical forecast data into periodic variable data for a preset time before and after a power generation prediction time point, and a corrected power generation prediction unit that predicts a first corrected power generation amount through the periodic variable data and the initial wind power generation amount through a first post-processing model.

7. In paragraph 6, A wind power generation prediction system characterized in that the first post-processing model applies an ensemble technique to a plurality of machine learning models.

8. In paragraph 6, A wind power generation prediction system characterized in that the first post-processing model is trained using the initial wind power generation for each agency over the past year, wind speed data at time t in the past, wind speed data at time t-1 in the past, wind speed data at time t-2 in the past, prediction cycle for each agency, and actual wind power generation for the past year as learning data based on the time point of power generation prediction.

9. In paragraph 8, A wind power generation prediction system characterized in that the first post-processing model is generated using a boosting method for a polynomial regression model and a LightGBM model.

10. In paragraph 5, The second power generation compensation unit is, A wind power generation prediction system including a wave function transform unit that generates high-frequency values ​​and low-frequency values ​​by decomposing the above agency-specific numerical forecast data into the above-mentioned corrected power generation and actual power generation by agency in the short-term time domain in the past based on the power generation prediction time point using a wavelet transform model, and a final power generation prediction unit that predicts the final wind power generation at the power generation prediction time point using the low-frequency values ​​of the above-mentioned corrected power generation and the above-mentioned actual power generation through a second post-processing model.

11. In paragraph 10, A wind power generation prediction system characterized in that the above wave function transformation unit generates the actual power generation at the time of power generation prediction used as learning data of the second post-processing model using LSTM (Long short time memory) and then generates it by decomposing it into high-frequency values ​​and low-frequency values.

12. In paragraph 10, A wind power generation prediction system characterized in that the second post-processing model uses at least one of GBR (Gradiant Boost Regression), LGBM (Light Gradiant Boost Machine), and MLP (Multi-layer Perceptron).

13. In paragraph 11, The above final power generation prediction unit is a wind power generation prediction system characterized in that it selects a model with the smallest prediction error of power generation among GBR (Gradiant Boost Regression), LGBM (Light Gradiant Boost Machine), and MLP (Multi-layer Perceptron) according to a preset period of time during the power generation prediction date.

Images