Forecasting of Wind Speed in Malang City of Indonesia using Adaptive Neuro-Fuzzy Inference System and Autoregressive Integrated Moving Average Methods

Abstract

Wind energy power (WEP) is currently one of the generating technologies that could be implemented massively due to its low environmental impact and abundant resources. However, the availability of the wind always changes depending on the weather condition, such that the power system should be designed properly to adopt the intermittence of the power injection of WEP. Therefore, the wind speed forecasting is very prominent to be performed to ensure the WEP could be incorporated into the existing power system. In this paper, the proposed methods to predict the wind speed are developed based on the artificial intelligence methods i.e. Adaptive Neuro Fuzzy Inference System (ANFIS) and based on the conventional methods i.e. Auto Regressive Integrated Moving Average (ARIMA). Case study in Malang city of Indonesia has been taken to compare the performances of both methods. Some membership functions (MF) have been studied to show the performance of ANFIS. The mean absolute error (MAE) and root mean square error (RMSE) have been used as standard statistical metrics to measure the performance of ANFIS and ARIMA methods. The results show that the optimal ANFIS architecture was obtained with 85% training data and 15% testing data by using the Generalized Bell membership function with MAE of 2.1354 km/h and RMSE of 2.6333 km/h. In addition, the wind forecasting result using ARIMA has been obtained with MAE of 2.8383 km/h and RMSE of 3.4628 km/h. The ANFIS method offers better performance than ARIMA does for short-term forecasting of wind speed in terms of MAE and RMSE values.