Objectives : In this study, deep learning models of artificial neural network (ANN) and one-dimension convolutional neural networks (1D-CNN) were compared to predict nonlinear wind power generation at Yeongheung wind power plant.Methods : The study site was Yeongheung-do, which has a 46 MW wind power plant. Hourly wind power and meteorological data from January to December 2018 were collected. After pre-processing with standardscaler, the training data were 64%, the validation data were 16%, and the test data were 20%. The optimum input variables of the model were selected using literature, and trial and error method. Rectified linear unit was used as the activation function. Hyperparameters were adjusted by trial and error method to optimized models. To compare the optimized models, the coefficient of determination (R2), mean absolute error (MAE), and root mean square error (RMSE) were used as the performance efficiency. Both ANN, and 1D-CNN were imported from the Keras library, and all of the performance efficiency was imported from the Scikit-learn library.Results and Discussion : The optimized input variables in this study were wind speed, wind direction, temperature, and humidity. The optimized ANN performance was R2=0.848, MAE=1.054, and RMSE=1.616, and the hyperparameters were 8 hidden layers with 100 nodes in each layer. The optimized 1D-CNN (R2=0.875, MAE=0.982, and RMSE=1.583) had 4 convolutional layers and the number of filters were 64, 128, 64, and 32 in order from the first layer, and one hidden fully connected layer had 100 nodes. The 1D-CNN had higher R2, and lower MAE and RMSE than the ANN. Therefore, the 1D-CNN was selected as the optimized model to predict wind generation of the Yeongheung wind power plant.Conclusions : The optimized 1D-CNN model in this study was more effective in predicting the Yeongheung wind power plant than the ANN. The optimal input variables were wind speed, wind direction, temperature, and humidity.
A211 Construction of Wind Power Plant Model Utilizing Wind Blowing through Buildings and Assessment of Power Generation Performance
