Algoritma Long Short Term Memory untuk Peramalan Kecepatan Angin Wilayah Surabaya dan Sekitarnya

Abstract

Surabaya is a big city where the majority of the area is in the lowlands, but there are still coastal areas with higher wind speeds, so in this area it is necessary to do forecasting to find out future wind conditions, so that it can be used as a basis for studies in planning the development of PLTB in the Surabaya area and its surroundings. One algorithm that is often used for forecasting is LSTM. This algorithm has a memory cell to store information for a long period of time. LSTM components in the form of forget gates, input gates, and output gates that function to control the flow of information in different time periods. The parameters used will be carried out a hyperparameter tuning process first to find out the best parameters from several parameters that have been provided. The results of the best parameters show that smallest RMSE and MAPE values, namely 2.64 and 47.78%, with epoch of 150, batch size of 32, neurons of 64, splitting data of 90%:10%, and timesteps of 31 or t-31. Based on these parameters, forecasting is carried out for the next 10 days, and the graph of the forecasting results tends to fluctuate stationarily without showing any upward or downward trend.