Method for selecting optimal wind power generation prediction model on basis of similarity of time-series data of wind power generation prediction models, and apparatus for performing same
The DTW technique optimizes wind power generation predictions by selecting models tailored to each wind farm's unique characteristics, improving accuracy and grid stability through customized forecasting.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- KOREA ELECTRONICS TECH INST
- Filing Date
- 2025-09-30
- Publication Date
- 2026-06-18
Smart Images
Figure KR2025015572_18062026_PF_FP_ABST
Abstract
Description
Method for selecting an optimal wind power generation prediction model based on the similarity of time series data of wind power generation prediction models and apparatus for performing the same
[0001] This specification relates to a technology for selecting an optimal wind power generation prediction model based on the similarity of time series data of wind power generation prediction models.
[0002] Over the past few decades, wind energy has grown rapidly, occupying a significant position as a renewable energy source. It is emerging as a key alternative to address global climate change and greenhouse gas emissions. According to the International Renewable Energy Agency (IRENA), the installed capacity of renewable energy worldwide is expected to increase to over 10,000 GW by 2030, with wind energy accounting for a significant share of this total. In particular, wind energy has established itself as a core component of power supply in major countries such as the United States, China, and Europe, with the global installed wind power capacity reaching approximately 93.6 GW as of 2021.
[0003] Due to the inherent irregularity and high variability of wind power generation as a natural energy source, it can cause instability in the power system. Managing this variability is particularly important in power grid operation and planning. The International Energy Agency (IEA) recommends the introduction of generation forecasting technology when the share of renewable energy, such as solar and wind power, reaches 3–15% of total power production. This is intended to minimize the impact of renewable energy variability on the power grid and to enable efficient energy management and operation.
[0004] Wind power forecasting is essential for the operation and management of power systems, and its role varies depending on the forecast timeframe. Long-term forecasting predicts power generation over a period of several months to several years to assess the variability of wind resources and is used to establish long-term grid expansion plans and investment strategies. On the other hand, short-term forecasting predicts power generation within a few hours to a few days, assisting in responding to real-time price fluctuations in the power market and aiding in short-term grid operation and maintenance planning. Ultra-short-term forecasting is a prediction within minutes to a few hours for real-time grid operation, focusing on maintaining grid stability by responding to sudden fluctuations in power generation.
[0005] The forecasting method is also a crucial element of wind power forecasting. Since wind power generation exhibits significant variability due to environmental factors such as topography and weather, it is difficult to guarantee high accuracy using a single forecasting method. Therefore, various forecasting techniques—including physical, statistical, artificial intelligence (AI), and hybrid methods—are utilized depending on the situation. Physical methods predict wind power generation based on atmospheric dynamics and weather data; in particular, they utilize NWP models to numerically simulate meteorological elements such as pressure, temperature, and humidity to calculate wind speed and direction. This method demonstrates strengths in long-term forecasting as it can model complex terrain or interference effects between turbines. Statistical methods predict future power generation through time-series analysis based on past generation data; they are widely used for relatively short-term forecasting and have the advantage of low computational costs. AI-based forecasting utilizes machine learning and deep learning technologies to learn and predict non-linear and complex data patterns, demonstrating excellent performance, especially with large-scale data. Finally, hybrid methods maximize forecasting performance by combining the strengths of physical, statistical, and AI models, providing high accuracy across diverse environments.
[0006] Traditionally, wind power generation has been predicted by individually utilizing physical, statistical, and artificial intelligence (AI) models. Physical models rely on meteorological data for prediction; while suitable for long-term forecasting, they have limitations in real-time prediction. Physical models suffer from complex design requirements and high costs associated with acquiring and analyzing meteorological data, as well as low accuracy in regional forecasting rather than comprehensive climate prediction. Data-driven models designed to address these issues have limitations, such as significant performance variability depending on the type, quantity, and characteristics of the acquired dataset. Statistical models are primarily used for short-term forecasting based on historical data, but they may not be suitable for complex non-linear data. While AI models are advantageous for learning non-linear data, they require large-scale training data and computing resources. Furthermore, because the same prediction model is applied to all wind farms during the selection process, unique characteristics of each farm, such as specific meteorological conditions and geographical features, are not reflected.
[0007] While research has been conducted on models suitable for various situations, accurately predicting wind power remains a challenging problem due to its high variability and irregularity. Ultimately, since topography, weather conditions, and turbine characteristics differ for each wind farm, there are limitations to applying a single model to all sites. Therefore, it is necessary to select and apply the most suitable model based on the specific characteristics of each site. In particular, to improve prediction accuracy and enhance stability, an approach is required that involves utilizing a variety of models and selecting the optimal model tailored to the specific characteristics of the site.
[0008] The above description is intended solely to aid in understanding the background technology regarding the technical concepts of the present invention, and therefore, it should not be understood as prior art known to those skilled in the art of the present invention.
[0009] This specification aims to solve the aforementioned problems. One embodiment of this specification aims to improve prediction accuracy and enhance power grid stability and energy efficiency by analyzing time-series data of wind power complexes and selecting an optimal prediction model suitable for each complex, thereby effectively managing the variability of renewable energy generation and contributing to the establishment of bidding strategies and real-time operations in the power market.
[0010] The problems that the present invention aims to solve are not limited to those mentioned above, and other unmentioned problems will be clearly understood by a person skilled in the art from the description below.
[0011] The present specification presents a method for selecting an optimal wind power generation prediction model to achieve the aforementioned objectives. The method for selecting an optimal wind power generation prediction model may include: a step of detecting weather prediction data with the most similar pattern by comparing weather prediction data for future time intervals with weather prediction data for past time intervals; a step of calculating each predicted power generation amount according to a plurality of power generation prediction models for the weather prediction data with the most similar pattern; and a step of selecting a prediction model as the optimal wind power generation prediction model by comparing the predicted power generation amount according to each of the plurality of power generation prediction models with the actual power generation amount in the time interval of the weather prediction data with the most similar pattern, and calculating the predicted power generation amount closest to the actual power generation amount.
[0012] In addition, the present specification presents an optimal wind power generation prediction model selection device to achieve the aforementioned objectives. The optimal wind power generation prediction model selection device comprises: a memory for storing one or more instructions; and a processor for executing said instructions. When said instructions are executed, the processor may detect weather prediction data with the most similar pattern by comparing weather prediction data for future time intervals with weather prediction data for past time intervals, calculate each predicted power generation amount according to a plurality of power generation prediction models for said weather prediction data with the most similar pattern, and select a prediction model that calculates a predicted power generation amount closest to the actual power generation amount by comparing the predicted power generation amount according to each of said power generation prediction models with the actual power generation amount in the time interval of said weather prediction data with the most similar pattern as the optimal wind power generation prediction model.
[0013] The embodiments disclosed in this specification have the effect of improving prediction accuracy and enhancing power grid stability and energy efficiency by analyzing time-series data of wind power complexes and selecting an optimal prediction model suitable for each complex, thereby effectively managing the variability of renewable energy generation and contributing to the establishment of bidding strategies and real-time operations in the power market.
[0014] In addition, the embodiments disclosed in this specification have the effect of enabling stable and accurate predictions even under highly variable weather conditions by selecting a prediction model suitable for weather patterns that change moment by moment through DTW-based similarity analysis.
[0015] In addition, the embodiments disclosed in this specification have the effect of providing customized prediction performance optimized for individual complexes by analyzing the time-series data characteristics of each power generation complex and selecting an optimal prediction model.
[0016] Furthermore, the embodiments disclosed in this specification can be extended and applied to various renewable energy resources, such as solar power generation, in addition to wind power generation. This ensures the universality of renewable energy forecasting, thereby having the effect of simultaneously improving the forecasting performance of various power generation resources.
[0017] Meanwhile, the effects obtainable from the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art to which the present invention belongs from the description below.
[0018] The following drawings attached to this specification illustrate preferred embodiments of the present invention and serve to further enhance understanding of the technical concept of the present invention along with specific details for implementing the invention; therefore, the present invention should not be interpreted as being limited only to the matters described in such drawings.
[0019] Figure 1 is a diagram illustrating a reverse distance weighted power generation prediction model.
[0020] Figure 2 is a diagram illustrating a transfer learning-based power generation prediction model.
[0021] Figure 3 is a diagram illustrating a power generation prediction model based on wind speed change rate characteristics.
[0022] Figure 4 is a diagram illustrating a wind speed correction-based power generation prediction model.
[0023] Figure 5 is a diagram illustrating a power generation prediction model based on turbine state information.
[0024] Figure 6 is a diagram illustrating an optimal wind power generation prediction model selection algorithm based on the DTW technique.
[0025] FIG. 7 is a block diagram of an optimal wind power generation prediction model selection device that performs an optimal wind power generation prediction model selection method according to one embodiment.
[0026] Figure 8 is a diagram illustrating an AI processing unit applied to the optimal wind power generation prediction model selection method by the optimal wind power generation prediction model selection device.
[0027] It should be noted that technical terms used in this specification are used merely to describe specific embodiments and are not intended to limit the scope of the technology disclosed herein. Furthermore, unless specifically defined otherwise in this specification, technical terms used in this specification shall be interpreted in the sense generally understood by those skilled in the art to which the technology disclosed herein belongs, and shall not be interpreted in an overly broad or overly narrow sense. Additionally, if a technical term used in this specification is an incorrect technical term that fails to accurately express the concept of the technology disclosed herein, it shall be understood as being replaced by a technical term that can be correctly understood by those skilled in the art to which the technology disclosed herein belongs. Furthermore, general terms used in this specification shall be interpreted according to their prior definitions or according to the context, and shall not be interpreted in an overly narrow sense.
[0028] The embodiments disclosed in this specification will be described in detail with reference to the attached drawings. Identical or similar components, regardless of drawing symbols, will be assigned the same reference number, and redundant descriptions thereof will be omitted. The suffixes "module" and "part" for components used in the following description are assigned or used interchangeably solely for the ease of drafting the specification and do not have distinct meanings or roles in themselves. Furthermore, the attached drawings are intended only to facilitate understanding of the embodiments disclosed in this specification; the technical concept disclosed in this specification is not limited by the attached drawings and should be understood to include all modifications, equivalents, and substitutions that fall within the spirit and technical scope of the invention.
[0029] Terms including ordinal numbers, such as first, second, etc., as used in this specification may be used to describe various components, but said components should not be limited by said terms. Such terms are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component.
[0030] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.
[0031] A singular expression includes a plural expression unless the context clearly indicates otherwise.
[0032] In this application, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.
[0033]
[0034] The present invention relates to a technology for predicting wind power generation, and more specifically, to an apparatus and method for improving prediction accuracy by selecting an optimal model based on the similarity of time-series data among various models used for predicting wind power generation. The apparatus and method disclosed herein can be utilized for short-term or ultra-short-term power generation prediction required for power grid operation, power market bidding, and renewable energy management.
[0035] The present invention selects a prediction model based on time-series data similarity by utilizing a Dynamic Time Warping (DTW) technique (hereinafter referred to as the DTW technique) to reflect the unique characteristics of each wind power complex. This overcomes the limitations of a single model, which is sensitive to variability in weather conditions and causes prediction errors, and enables the selection and utilization of an optimal prediction model in real time according to the unique characteristics of the power complex, such as weather conditions or geographical characteristics. Therefore, it is possible to effectively respond to various weather patterns and non-linear fluctuations.
[0036] In this specification, several models suitable for short-term and ultra-short-term forecasting, which are essential for power market bidding and real-time operation of wind power farms, are presented, and an optimal wind power forecasting model selection device based on a method of selecting the optimal wind power model using the Dynamic Time Warping (DTW) technique is proposed.
[0037] For short-term forecasting models, we propose an inverse distance weighting model suitable for large-scale wind power complexes and a transfer learning model for new power plants where sufficient operational data has not yet been secured. For ultra-short-term forecasting models, we propose a wind speed change rate characteristic model reflecting the spatial and time-series characteristics of meteorological data, a wind speed correction model that generates meteorological information suitable for ultra-short-term forecasting using Numerical Weather Prediction (NWP), and a turbine characteristic model reflecting the state information of each turbine.
[0038] Each short-term and ultra-short-term prediction model was trained using a sufficient amount of data totaling approximately 31 months, and monthly and seasonal characteristics were verified through a 10-month validation period. Subsequently, experiments were conducted comparing each prediction model with the optimal model utilizing DTW, and based on nMAPE, short-term predictions showed a maximum error reduction of 0.99%, and ultra-short-term predictions showed a maximum error reduction of 1.38%.
[0039] The optimal model selection method utilizing DTW techniques plays a crucial role in improving prediction accuracy by selecting a prediction model suited to the unique characteristics of each wind farm. This prediction system can be utilized as an important tool for establishing efficient bidding strategies in the power market and ensuring the operational stability of wind farms in real time.
[0040] Wind Power Forecasting Model
[0041] The temporal scope of wind power forecasting varies depending on the needs and purposes. This specification studies models applicable to power markets where renewable energy sources, such as wind power, are available for bidding. To this end, it develops a short-term power generation forecasting model that predicts the next day in 1-hour increments for 24 hours, and an ultra-short-term power generation forecasting model that predicts one hour in 10-minute increments from the current point in time. Furthermore, since no specific model delivers the best performance for all complexes, this specification describes an optimal model selection method based on the DTW technique to select and apply the most suitable model according to the characteristics of each complex.
[0042] Short-term Forecasting Model
[0043] Figure 1 is a diagram illustrating a reverse distance weighted power generation prediction model.
[0044] Referring to Figure 1, weather data such as NWP consists of spatial grids of 1.5 km or more. However, since the actual scale of a wind farm is several kilometers or more, the weather data for each wind turbine belonging to a number of spatial grids is different, so various weather conditions may exist. Therefore, NWPs from adjacent areas were collected, and an inverse distance weighting preprocessing process was added to assign a greater weight as the distance between adjacent areas increases. Then, using a CNN model with the preprocessed data, a process for predicting the amount of power generated for a target wind farm was performed.
[0045] Specifically, weather data from NWP coordinates including turbine locations (NWP #1) and data from three adjacent locations (NWP #2, NWP #3, NWP #4) were additionally collected and incorporated into the prediction model. To enable the model to learn the spatial characteristics of each NWP location, only the variables were preprocessed; the model was then constructed by inputting the preprocessed variables into the CNN without reducing the spatial dimension. Instead of manually setting weights, the CNN was configured to calculate weights based on the data.
[0046] Figure 2 is a diagram illustrating a transfer learning-based power generation prediction model.
[0047] Transfer learning is a method of optimizing existing models by utilizing self-trained data and pre-trained models; it refers to a technique where the weights of a model trained on a large dataset are readjusted to suit the problem to be solved.
[0048] In this specification, a power generation prediction model for a target wind farm is generated by using a model that extracts features using an autoencoder from a portion of meteorological data condensed into a lower dimension—that is, the front part of the pre-trained model—in order to predict power generation from a power generation prediction model (pre-trained model) trained using data from other wind farms other than the target wind farm. At this time, the front part (200) of the pre-trained model is fixed as is, and a neural network model for predicting power generation of the target farm is attached to the back part in a small size for training.
[0049] A pre-trained model was created to extract the characteristics of the relationship between meteorological data and power generation using data from wind power complexes other than the target power complex, where there is a large amount of accessible data.
[0050] Since it utilizes a pre-trained model based on a large amount of weather data and power generation, an unbiased and objective model is quickly generated even if the target complex does not have much data.
[0051] Ultra Short-term Forecasting Model
[0052] Figure 3 is a diagram illustrating a power generation prediction model based on wind speed change rate characteristics.
[0053] Because each wind power complex has different natural environments and wind turbine layouts, diverse wind flows are created spatially and temporally. In particular, since the output of a wind turbine changes in proportion to the cube of the wind speed, it is necessary to predict power generation using meteorological data tailored to the specific characteristics of the complex in order to reduce deviations in output caused by changes in wind speed.
[0054] Accordingly, to construct a model that utilizes both spatial characteristics measured by wind power turbine (tur #1, tur #2, …, tur #n) and temporal characteristics at 10-minute intervals in the weather data, a model combining CNN and LSTM was designed. CNN (Convolutional Neural Network) is specialized among neural networks for extracting local features that can reflect spatial characteristics, while LSTM (Long Short-Term Memory) is characterized by its ability to effectively identify and process long-term dependencies in time-series data. By combining the respective features of the two models, a model was constructed that reflects both the spatial and time-series characteristics of the weather data.
[0055] As described above, this model first inputs weather data and power generation data at predetermined time intervals (e.g., 10 minutes) for each turbine into a CNN model to extract spatial features, and then inputs the results into an LSTM model to derive temporally changing features to predict ultra-short-term power generation.
[0056] Figure 4 is a diagram illustrating a wind speed correction-based power generation prediction model.
[0057] Wind speed, which has the greatest impact on wind power generation, is highly variable and difficult to predict, so errors in weather forecast data such as NWP are always present. Furthermore, since NWP is fundamentally a model suitable for forecasting over a day, it is difficult to directly utilize it for ultra-short-term power generation forecasting within a few hours. To address this, a new model was designed with the following structure.
[0058] In the first step, a data-based wind speed prediction model (410) is constructed to predict wind speed based on actual wind speed data. In the second step, a data-based wind power generation prediction model (420) is constructed by combining the wind speed data (421) predicted in the first step, weather prediction data (422), and actual data (421).
[0059] The wind speed and wind power generation prediction models are trained using XGBoost, a boosting-type machine learning model. Through this model, the problem of overfitting is appropriately regulated, and optimizations are performed via various parameter adjustments to obtain suitable results for each power generation complex.
[0060] Figure 5 is a diagram illustrating a power generation prediction model based on turbine state information.
[0061] Wind power complexes exhibit different power output characteristics even under the same wind conditions, depending on factors such as the inertia of wind turbine blade rotation and the update status of control variables. Although the initial state of the wind turbine may differ, if the same wind conditions persist, the influence of the initial state diminishes, resulting in identical output characteristics over the long term; however, the influence of the initial state is significant in ultra-short-term power generation forecasts of 10 minutes or less.
[0062] Considering that the conditions of each turbine differ even under the same weather conditions, a model was developed to predict the power generation of the entire power complex by predicting the power generation of each turbine individually and summing them, rather than predicting the power generation of the entire complex.
[0063] A state information-based power generation prediction model for wind turbines was constructed using XGBoost, a boosting-type machine learning model. Through this model, a pruning algorithm capable of effectively reflecting turbine-specific state data was applied, and various parameters were adjusted to obtain appropriate results for each power plant.
[0064] Optimal Model Selection
[0065] The DTW technique is an algorithm used to measure similarity between time-series data, providing a powerful method to compare two sequences even if their speeds or times differ. One of the various advantages of the DTW technique is its non-linear time alignment capability. Even though two sequences may occur at different times or have different speeds, the DTW technique compensates for these differences and can calculate similarity by aligning the two data.
[0066] In the embodiments of this specification, the most suitable model among the various models presented in the aforementioned short-term and ultra-short-term prediction models is selected using the characteristics of such DTW.
[0067] Through the DTW technique, future weather information (e.g., wind speed, wind direction, etc.) is compared with each segment of a past data set, and the most similar segment is identified. In this specification, by using the DTW technique to correct distortion on the time axis, segments with future weather information similar to specific past weather patterns are identified, and the data of those segments can be utilized.
[0068] The core of the DTW-based model selection technique lies in identifying similar past weather patterns and comparing the prediction accuracy of existing models. The prediction performance of each model is evaluated by comparing it with actual power generation in past periods, and the model demonstrating higher prediction accuracy is ultimately selected to forecast future power generation. Through this process, overall prediction performance can be improved by utilizing a more suitable model for each weather condition.
[0069] In the embodiments of this specification, weather data for a future time interval to be predicted is collected, and a weather pattern similar to the pattern of the corresponding weather data is identified. Next, power generation amounts according to various prediction models are calculated for past time intervals exhibiting similar weather patterns, and the actual power generation amount in the past time intervals is compared with the predicted power generation amount according to each prediction model to select the prediction model that produces the predicted power generation amount closest to the actual power generation amount. Finally, the predicted power generation amount for the future time interval is calculated using the finally selected prediction model.
[0070] Figure 6 is a diagram illustrating an optimal wind power generation prediction model selection algorithm based on the DTW technique.
[0071] The algorithm described in Fig. 6 is an Optimal Model Selection Algorithm that utilizes the Dynamic Time Warping (DTW) technique to compare time series data and select the optimal model.
[0072] An algorithm according to one embodiment uses DTW to find the most similar pattern in given time series data (e.g., weather forecast data or power generation forecast data) and selects an optimal model based on this information. Ultimately, it is used to select a model among several power generation forecast models to minimize the error in power generation forecasting. This algorithm is used for power generation forecasting of wind power based on weather data (NWP). NWP data represents numerical forecast data reflecting weather conditions, and based on this, similar past patterns are found to select a suitable optimal model among multiple forecast models.
[0073] For example, if today's weather data (NWP) exhibits a specific weather pattern, similar weather patterns are identified from past weather data. Subsequently, power data generated by various models based on similar past weather patterns is evaluated by comparing them with actual power data; finally, the model that predicts the value closest to the actual power data is selected as today's power generation prediction model.
[0074] These algorithms enable non-linear time series comparison using DTW, improve the reliability of predictions based on similar past patterns, and allow for the selection of the optimal model by evaluating various models.
[0075] This algorithm focuses on utilizing DTW to find the optimal reference point from historical data and, based on this, comparing multiple prediction models to select the best model. This algorithm can be applied to any power generation prediction model that uses meteorological data, including not only wind power but also solar power prediction models where meteorological data can be utilized. In the case of solar power prediction models, meteorological data such as sunshine duration data, hourly solar radiation data, and hourly cloud cover data can be used.
[0076] The description of the key variables and the structure of the algorithm in the described algorithm are as follows.
[0077] (1) Explanation of major variables
[0078] f DTW : A function that calculates a matching score between two time series data using DTW, where the matching score indicates the similarity between the two time series data.
[0079] v NWP,today : Refers to tomorrow's NWP (Numerical Weather Prediction) data predicted today.
[0080] v NWP,t : Refers to the NWP data at time t+1 predicted at time t.
[0081] t max : Refers to the point in time when the DTW matching score is highest.
[0082] x max : Refers to the highest DTW matching score.
[0083] E: Refers to a function that calculates the power generation prediction error.
[0084] P m (t): Refers to the amount of power generated on day t+1 predicted on day t using model m.
[0085] P truth (t+1): Refers to the actual amount of power generated on (t+1).
[0086] (2) Composition of the algorithm
[0087] The algorithm of the embodiment consists of a DTW matching score calculation process (610) and a prediction error calculation process (620).
[0088] In the DTW matching score calculation process (610), f DTW Using today's predicted tomorrow's NWP data (v NWP,today ) and predicted data at past time t (v NWP,tCalculate the matching score between ). Also, repeat for all t to find the highest matching score x max and the corresponding time point t max It stores . That is, it calculates the DTW matching score between past data and today's NWP data, and finds t with the most similar pattern. max Find it and determine whether today's data is similar to a point in the past.
[0089] In the prediction error calculation process (620), t max Based on the data predicted in, the predicted power generation values for various models m are calculated. Next, the predicted value P for all models m (t max ) and actual value P truth (t max Calculate the error E between ), and the model m with the smallest error max Select . That is, t max Power generation is predicted using various prediction models based on similar data. The error between the prediction result of each model and the actual data is calculated, and the model m with the smallest error max Select .
[0090] Hereinafter, an apparatus for performing the above-described method will be described in detail with reference to the attached drawings.
[0091] FIG. 7 is a block diagram of an optimal wind power generation prediction model selection device that performs an optimal wind power generation prediction model selection method according to one embodiment.
[0092] Referring to FIG. 7, the optimal wind power generation prediction model selection device (100) includes a communication unit (110), a user interface device (120), a display device (130), a storage medium (140), a processor (150), a system memory (160), and an AI processing unit (170). The illustrated components are not essential, so an optimal wind power generation prediction model selection device may be implemented with more or fewer components. These components may be implemented in hardware or software, or through a combination of hardware and software.
[0093] The communication unit (110) can transmit and receive signals between the optimal wind power generation prediction model selection device (100) and an external weather data server (not shown) via a network. The weather data server can extract and provide weather data (NWP) for a specific location, date, and time requested by the optimal wind power generation prediction model selection device (100).
[0094] The user interface device (120) receives user input to control the operations of the optimal wind power generation prediction model selection device (100) or the processor (150). The user interface device (120) may include a key pad, a dome switch, a touch pad (static / capacitive), a jog wheel, a jog switch, a finger mouse, etc.
[0095] The display device (130) operates in response to the control of the processor (150). The display device (130) displays information processed by the optimal wind power generation prediction model selection device (100) or the processor (150). For example, the display device (130) can display an image according to the control of the processor (150).
[0096] The storage medium (140) may be at least one of flash memory, hard disk, solid state disk type (SSD), multimedia card memory, random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, and optical disk. The storage medium (140) is configured to write and read data in response to the control of the processor (150).
[0097] The storage medium (140) can store the various power generation prediction models described above. Additionally, the storage medium (140) can store current, future, and past weather data (NWP) received from a weather data server. Furthermore, the storage medium (140) can store power generation calculation values by each power generation prediction model calculated by the processor (150), as well as actual power generation data by wind power complex and wind power turbine at a specific point in the past.
[0098] The processor (150) may include either a general-purpose or dedicated processor and controls the operations of the communication unit (110), user interface device (120), display device (130), storage medium (140), system memory (160), and AI processing unit (170).
[0099] The processor (150) is configured to load program codes containing instructions that provide various functions when executed from the storage medium (140) into the system memory (160) and to execute the loaded program codes. The processor (150) can load an optimal model selection module (161) containing instructions and / or program codes from the storage medium (140) into the system memory (160) and execute the loaded optimal model selection module (161). The optimal model selection module (161) can process data received from a weather data server and display it on a display device (130), and detect related user input. Additionally, the optimal model selection module (161) can visualize an additional user interface on the display device (130) and detect user input through it.
[0100] The processor (150) can implement all functions of the optimal wind power generation prediction model selection device (100) described with reference to FIGS. 1 to 7 and all processes of the optimal wind power generation prediction model selection method by the optimal wind power generation prediction model selection device (100) through the optimal model selection module (161) or through a program loaded from the storage medium (140).
[0101] System memory (160) may be provided as working memory of the processor (150). In the drawing, system memory (160) is shown as a component separate from the processor (150), but this is exemplary and at least a portion of system memory (160) may be integrated within the processor (150). System memory (160) may include at least one of Random Access Memory (RAM), Read Only Memory (ROM), and other types of computer-readable storage media.
[0102] The AI processing unit (170) is for processing data based on artificial intelligence, and according to one embodiment of the present specification, it can train various prediction models using weather data, wind power generation amount data, and status data of wind power generators. The AI processing unit (170) may be implemented as a single module with the processor (150) that controls the entire system, or as an independent module. The AI processing unit (170) may also be implemented as a separate module or device, etc., separated from the optimal wind power generation prediction model selection device (100).
[0103] The processor (150) can perform data processing necessary for the execution of the aforementioned optimal wind power generation prediction model selection method in cooperation with the AI processing unit (170).
[0104] Figure 8 is a diagram illustrating an AI processing unit applied to the optimal wind power generation prediction model selection method by the optimal wind power generation prediction model selection device.
[0105] Referring to FIG. 8, the AI processing unit (170) may include an electronic device including an AI module capable of performing AI processing or a server including an AI module. Additionally, the AI processing unit (170) may be configured to be included as at least part of the configuration of the optimal wind power generation prediction model selection device (100) to perform at least part of the AI processing together.
[0106] AI processing may include all operations related to the processor or control unit of the optimal wind power generation prediction model selection device (100).
[0107] The AI processing unit (170) may be a client device that directly uses the AI processing results, or a device in a cloud environment that provides the AI processing results to other devices. The AI processing unit (170) is a computing device capable of training a neural network and can be implemented as various electronic devices such as an AI chip, an AI module, a server, a desktop PC, a laptop PC, a tablet PC, a smartpad, etc.
[0108] The AI processing unit (170) may include an AI processor (171), memory (175), and a communication unit (177).
[0109] The AI processor (171) can train a neural network using a program stored in memory (175). In particular, the AI processor (171) can train a neural network for recognizing data related to a surveillance camera. Here, the neural network for recognizing data related to a surveillance camera can be designed to simulate the structure of the human brain on a computer and may include multiple network nodes having weights that simulate neurons of a human neural network. Multiple network modes can exchange data according to their respective connection relationships to simulate the synaptic activity of neurons exchanging signals through synapses. Here, the neural network may include a deep learning model that has evolved from a neural network model. In the deep learning model, multiple network nodes can be located in different layers and exchange data according to convolutional connection relationships. Examples of neural network models include various deep learning techniques such as Deep Neural Networks (DNN), Convolutional Deep Neural Networks (CNN), Recurrent Neural Networks (RNN), Restricted Boltzmann Machines (RBM), Deep Belief Networks (DBN), and Deep Q-Networks, and can be applied in fields such as computer vision (CV), speech recognition, natural language processing, and speech / signal processing.
[0110] Meanwhile, an AI processor that performs the functions described above may be a general-purpose processor (e.g., CPU), but may also be an AI-dedicated processor for artificial intelligence learning (e.g., GPU).
[0111] The memory (175) can store various programs and data required for the operation of the AI processing unit (170). The memory (175) can be implemented as non-volatile memory, volatile memory, flash memory, hard disk drive (HDD), or solid-state drive (SDD). The memory (175) is accessed by the AI processor (171), and the AI processor (171) can perform reading / writing / modification / deletion / updating of data. Additionally, the memory (175) can store a neural network model (e.g., a deep learning model (176)) generated through a learning algorithm for data classification / recognition according to an embodiment of the present invention. That is, the deep learning model (176) can store the aforementioned various power generation prediction models for each wind power complex and wind power turbine.
[0112] Meanwhile, the AI processor (171) may include a data learning unit (172) that learns a neural network for data classification / recognition. The data learning unit (172) can learn criteria regarding which training data to use to determine data classification / recognition, and how to classify and recognize data using the training data. The data learning unit (172) can learn a deep learning model by acquiring training data to be used for learning and applying the acquired training data to a deep learning model.
[0113] The data learning unit (172) may be manufactured in the form of at least one hardware chip and mounted on the AI processing unit (170). For example, the data learning unit (172) may be manufactured in the form of a dedicated hardware chip for artificial intelligence (AI), or may be manufactured as part of a general-purpose processor (CPU) or a dedicated graphics processor (GPU) and mounted on the AI processing unit (170). Additionally, the data learning unit (172) may be implemented as a software module. If implemented as a software module (or a program module containing instructions), the software module may be stored on a non-transitory computer-readable media. In this case, at least one software module may be provided by an operating system (OS) or by an application.
[0114] The data learning unit (172) may include a learning data acquisition unit (173) and a model learning unit (174).
[0115] The learning data acquisition unit (173) can acquire learning data required for a neural network model for classifying and recognizing data.
[0116] The model learning unit (174) can learn to have a judgment criterion regarding how a neural network model classifies a predetermined data using acquired training data. At this time, the model learning unit (174) can train the neural network model through supervised learning, which uses at least a portion of the training data as a judgment criterion. Alternatively, the model learning unit (174) can train the neural network model through unsupervised learning, which discovers the judgment criterion by learning on its own using training data without supervision. Additionally, the model learning unit (174) can train the neural network model through reinforcement learning using feedback on whether the result of the situation judgment based on the learning is correct. Furthermore, the model learning unit (174) can train the neural network model using a learning algorithm including error back-propagation or gradient descent.
[0117] When the neural network model is trained, the model training unit (174) can store the trained neural network model in memory (175). Additionally, the model training unit (174) can store the trained neural network model in memory of a server connected to the AI processing unit (170) via a wired or wireless network.
[0118] The data learning unit (172) may further include a learning data preprocessing unit (not shown) and a learning data selection unit (not shown) to improve the analysis results of the recognition model or to save resources or time required for the creation of the recognition model.
[0119] The learning data preprocessing unit can preprocess the acquired data so that the acquired data can be used for learning to make situational judgments. For example, the learning data preprocessing unit can process the acquired data into a pre-set format so that the model learning unit (174) can use the acquired learning data for learning to recognize images.
[0120] Additionally, the training data selection unit may select data required for training from among the training data obtained from the training data acquisition unit (173) or the training data preprocessed from the preprocessing unit. The selected training data may be provided to the model training unit (174).
[0121] Additionally, the data learning unit (172) may further include a model evaluation unit (not shown) to improve the analysis results of the neural network model.
[0122] The model evaluation unit inputs evaluation data into a neural network model, and if the analysis result output from the evaluation data does not satisfy a predetermined standard, it may have the model learning unit (174) learn again. In this case, the evaluation data may be predefined data for evaluating a recognition model. For example, the model evaluation unit may evaluate that the predetermined standard is not satisfied if, among the analysis results of the recognition model learned for the evaluation data, the number or ratio of evaluation data for which the analysis result is inaccurate exceeds a preset threshold.
[0123] The communication unit (177) can transmit the AI processing results by the AI processor (171) to an external electronic device. For example, the external electronic device may include a grid control server, a power exchange server, etc.
[0124] Meanwhile, although the AI processing unit (170) illustrated in FIG. 8 is described as being functionally divided into an AI processor (171), memory (175), and communication unit (177), it should be noted that the aforementioned components may be integrated into a single module and referred to as an AI module.
[0125] The networks disclosed in this specification may be, for example, wireless networks, wired networks, public networks such as the Internet, private networks, Global System for Mobile communication network (GSM) networks, General Packet Radio Network (GPRN), Local Area Network (LAN), Wide Area Network (WAN), Metropolitan Area Network (MAN), cellular networks, Public Switched Telephone Network (PSTN), Personal Area Network, Bluetooth, Wi-Fi Direct, Near Field communication, Ultra-Wide band, combinations thereof, or any other networks, but are not limited thereto.
[0126]
[0127] The embodiments described above are combinations of the components and features of the present invention in a specific form. Each component or feature should be considered optional unless otherwise explicitly stated. Each component or feature may be implemented in a form not combined with other components or features. Additionally, it is possible to construct embodiments of the present invention by combining some components and / or features. The order of operations described in the embodiments of the present invention may be changed. Some components or features of one embodiment may be included in another embodiment, or may be replaced with corresponding components or features of another embodiment. It is obvious that embodiments may be constructed by combining claims that do not have an explicit citation relationship in the claims, or that they may be included as new claims through amendments made after filing.
[0128] Embodiments according to the present invention may be implemented by various means, for example, hardware, firmware, software, or a combination thereof. In the case of implementation by hardware, an embodiment of the present invention may be implemented by one or more ASICs (application specific integrated circuits), DSPs (digital signal processors), DSPDs (digital signal processing devices), PLDs (programmable logic devices), FPGAs (field programmable gate arrays), processors, controllers, microcontrollers, microprocessors, etc.
[0129] In the case of implementation by firmware or software, an embodiment of the present invention may be implemented in the form of a module, procedure, function, etc., that performs the functions or operations described above. The software code may be stored in memory and executed by a processor. The memory may be located inside or outside the processor and may exchange data with the processor by various known means.
[0130] Additionally, terms such as "front," "back," "top," "top," "bottom," "bottom," "on top," and "below" in the detailed description and claims are used for illustrative purposes but are not necessarily used to describe permanent relative positions. It is understood that such terms are interchangeable under appropriate circumstances to allow the embodiments of the invention described herein to operate in directions other than those shown or otherwise described herein.
[0131] For the sake of simplicity and clarity, it should be understood that the elements depicted in the drawings are not necessarily drawn in a fixed proportion. For example, the dimensions of some elements may be exaggerated compared to others for clarity. Additionally, where deemed appropriate, reference numbers may be repeated between the drawings to indicate corresponding or similar elements.
[0132] It is obvious to those skilled in the art that the present invention may be embodied in other specific forms without departing from the essential features of the invention. Accordingly, the foregoing detailed description should not be interpreted restrictively in all respects but should be considered exemplary. The scope of the invention shall be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the invention are included within the scope of the invention.
Claims
1. A step of detecting the weather forecast data with the most similar pattern by comparing weather forecast data for future time intervals with weather forecast data for past time intervals; A step of calculating each predicted power generation amount according to a plurality of power generation prediction models for the weather forecast data with the most similar pattern above; and The method comprises the step of selecting, as the optimal wind power generation prediction model, a prediction model that calculates a predicted power generation amount closest to the actual power generation amount by comparing the predicted power generation amount according to each of the plurality of power generation prediction models with the actual power generation amount in the time interval of the weather prediction data with the most similar pattern. Method for Selecting the Optimal Wind Power Prediction Model.
2. In Paragraph 1, The step of detecting the weather forecast data with the most similar pattern by comparing the weather forecast data of the future time interval with the weather forecast data of past time intervals is: The above-mentioned virtual similar pattern weather forecast data includes the step of determining that the matching score is the maximum value through the calculation of a matching score by a Dynamic Time Warping technique; Method for Selecting the Optimal Wind Power Prediction Model.
3. In Paragraph 2, The above plurality of power generation prediction models include an inverse distance weighting model and a transfer learning model, and The above inverse distance weighting model It is a model that predicts power generation based on a CNN model for a target wind power complex, using weather forecast data for a reference spatial grid and weather forecast data preprocessed with distance-based inverse distance weights for adjacent spatial grids. The above transfer learning model A model in which the corresponding layer in the power generation prediction model for the aforementioned target wind power complex is replaced with a weather forecast data feature extraction model for power generation prediction in a power generation prediction model pre-trained based on data from other wind power complexes other than the aforementioned target wind power complex, and Method for Selecting the Optimal Wind Power Prediction Model.
4. In Paragraph 2, The above plurality of power generation prediction models include a wind speed change rate characteristic model, a wind speed correction-based power generation prediction model, and a turbine state information-based power generation prediction model, and The above wind speed change rate characteristic model is, This is a model generated by inputting weather forecast data and power generation data at predetermined time intervals for each wind turbine of a target wind power complex into a CNN model to extract spatial features, and inputting the extracted spatial feature results for each turbine into an LSTM model to derive temporal change features. The above wind speed correction-based power generation prediction model is, A wind speed prediction model that calculates wind speed prediction data through a data-based wind speed prediction model that predicts wind speed based on actual wind speed data of the above-mentioned target wind power generation complex, and a wind power generation prediction model based on the calculated wind speed prediction data, weather prediction data, and actual data, and the wind speed prediction model and the wind power generation prediction model are models based on a boosting-type machine learning model. The above turbine state information-based power generation prediction model is, A model that predicts the total power generation of the target wind power complex by integrating power generation prediction models for each turbine within the target wind power complex based on a boosting-type machine learning model, Method for Selecting the Optimal Wind Power Prediction Model.
5. Memory for storing one or more instructions; and A processor for executing the above instruction; including, When the above instruction is executed, the processor, By comparing weather forecast data for future time intervals with weather forecast data for past time intervals, the most similar pattern of weather forecast data is detected, and Calculate each predicted power generation amount according to multiple power generation prediction models for the weather forecast data with the most similar pattern mentioned above, and Among the plurality of power generation prediction models mentioned above, the prediction model that calculates the predicted power generation closest to the actual power generation by comparing the predicted power generation according to each model with the actual power generation in the time interval of the weather forecast data with the most similar pattern is selected as the optimal wind power generation prediction model. Optimal wind power generation prediction model selection device.
6. In claim 5, the processor, Determining the weather forecast data of the virtual similar pattern as the one with the maximum matching score through the calculation of a matching score using a dynamic time warping technique on the weather forecast data of the future time interval and the weather forecast data of the past time intervals. Optimal wind power generation prediction model selection device.
7. In Paragraph 6, The above plurality of power generation prediction models include an inverse distance weighting model and a transfer learning model, and The above inverse distance weighting model It is a model that predicts power generation based on a CNN model for a target wind power complex, using weather forecast data for a reference spatial grid and weather forecast data preprocessed with distance-based inverse distance weights for adjacent spatial grids. The above transfer learning model A model in which the corresponding layer in the power generation prediction model for the aforementioned target wind power complex is replaced with a meteorological data feature extraction model for power generation prediction in a power generation prediction model pre-trained based on data from other wind power complexes other than the aforementioned target wind power complex, Optimal wind power generation prediction model selection device.
8. In Paragraph 6, The above plurality of power generation prediction models include a wind speed change rate characteristic model, a wind speed correction-based power generation prediction model, and a turbine state information-based power generation prediction model, and The above wind speed change rate characteristic model is, This is a model generated by inputting weather forecast data and power generation data at predetermined time intervals for each wind turbine of a target wind power complex into a CNN model to extract spatial features, and inputting the extracted spatial feature results for each turbine into an LSTM model to derive temporal change features. The above wind speed correction-based power generation prediction model is, A wind speed prediction model that calculates wind speed prediction data through a data-based wind speed prediction model that predicts wind speed based on actual wind speed data of the above-mentioned target wind power generation complex, and a wind power generation prediction model based on the calculated wind speed prediction data, weather prediction data, and actual data, and the wind speed prediction model and the wind power generation prediction model are models based on a boosting-type machine learning model. The above turbine state information-based power generation prediction model is, A model that predicts the total power generation of the target wind power complex by integrating power generation prediction models for each turbine within the target wind power complex based on a boosting-type machine learning model, Optimal wind power generation prediction model selection device.