• 목록

[Munhwa News Reporter Sunhye Park] Turbine Crew Co., Ltd. announced that it presented a paper titled "A Study on a Local-Based Ultra-Short-Term Wind Power Generation Prediction System Using CGAN-LSTM" in collaboration with Professor Jeonghun Park of Ajou University at the "2023 Korean Intelligent Systems Society Autumn Conference."

The presentation focused on a system designed to predict local-based ultra-short-term wind power generation using CGAN-LSTM. The paper detailed the process of collecting precise local weather data—such as temperature, wind direction, wind speed, humidity, pressure, and one-minute rainfall—using Mini Weather Station (MWS) sensors. The collected data was used to design a CGAN-LSTM model for predicting wind power generation. The system aims to perform ultra-short-term and real-time wind power generation predictions by reflecting the spatiotemporal characteristics of weather data through a cloud platform.

The study also incorporated several advanced techniques, including the Sliding Window method, Self-Attention layers, and the use of batch normalization and Mean Squared Error (MSE) loss to prevent overfitting. The addition of a Self-Attention layer at the initial input stage of the model was used to apply weights according to the importance of input features, and batch normalization after convolutional layers improved the model's normalization performance and prevented overfitting.

In comparison to similar studies, the research highlighted that the weather data used in this model has low data variance, which can often make it difficult to train the model and improve performance. However, Turbine Crew's developed model successfully reflected the trend for the next six hours, effectively predicting the total power generation value, a key point emphasized during the presentation.

The research also plans to consider utilizing weather prediction models provided by the Korea Meteorological Administration for future predictions. In addition, while there are multiple wind power plants in South Korea, using deep learning models for prediction is still in its early stages. The study highlighted that not utilizing data from local weather sensors near the plants often leads to accuracy issues, and the approach taken by Turbine Crew represents an advancement in addressing these challenges by incorporating temporal characteristics of weather and power generation data.

Through this achievement, Turbine Crew has received recognition for its differentiated technological capabilities and plans to accelerate its growth further with AWS software partner approval.

Additionally, the company is expanding its global reach by establishing a branch in Delaware, USA, and continuing to develop its technology through collaboration with Silicon Valley. Turbine Crew boasts an impressive track record, including 6th place in the AI Tech GuangZhen IT competition, 1st place in the Minds & Company AI Simulation Competition, a silver medal in the 2023 Kaggle competition, and numerous other coding competition awards.

Furthermore, Turbine Crew, in collaboration with Haerum Drone Co., Ltd., is conducting free Python training courses for Jeonnam residents from September to November, leveraging the technology presented in this conference.