Optimum Turbine Number and Data Determination for Short Term Wind Power Forecasting in Complex Terrains: A case study in Turkey

Abstract

Bandirma power plant with an average altitude of 500 m is located at 3 km away from Marmara Sea. The farm has 29 wind turbines placed at intervals ranging from 300 m to 1 km. Short term (up to 48 hours) wind power forecasting is very difficult in complex terrains, due to sea-land interaction and wake interactions between turbines. Additionally, intervals between turbines are very small for mesoscale Numerical Weather Prediction (NWP) models. The purpose of this study is to develop a methodology to forecast short term wind power from the minimum number of input and to determine which turbines should be selected for NWP models in the Bandirma power plant. To predict the power production; directional equivalent plant power curve, regression, artificial neural networks, time series models and fuzzy logic algorithms are developed. Turbine based 10-minutes average wind speed, wind direction, temperature measurements, and the actual power production of the farm from March 2015 to September 2017 are used in model development. All algorithms are trained with more than 2 years historical data, and the performance of algorithms are tested with an untrained one-month measurement data. According to the results, the most precise algorithm, and optimum number of turbines and input data are selected. The locations of these turbines are used to determine the optimum resolution of domains in NWP model development study.