Comparison the forecasting performance of wavelet activation functions in wind speed prediction with wavelet neural networks

The integration of increasing renewable energy-based energy production into the existing electricity grid has become of great importance. The biggest difficulty to integration into the power grid is variability and discontinuity of wind energy. To deal with this situation, the best approach is to predict future values of wind power production. Wind speed estimation methods with high accuracy are an effective tool that can be used to minimize these problems. Thus, several wind power or wind speed forecasting methods have been reported in the literature over the past few years. This study presents a very short – term wind speed prediction using wavelet neural network (WNN) and compares the performance of different Polywog activation functions in this network. Data are collected from a weather station located in Ondokuz Mayis University in ten minute resolution for a period of one year. Wind speed predictions are presented within a period of 24-hours for 10 minute ahead. The root mean square error (RMSE) and the mean squared error (MSE) values have been selected as performance criteria. According to forecasts will be achieved, provides wind energy systems, accurate and rapid adaptation to changing climatic conditions are be attained. In this study, all WNN structures use the same topology. The selected architecture is in the form of an input layer, an output layer, and a hidden layer network structure, with one each input for pressure, temperature and humidity, one output for wind speed. There is no specific rule for the number of wavelon in the hidden layer. For this reason, 10 wavelon architects have been decided by trial and error method. Each Polywog family functions (Polywog1, Polywog2, Polywog3, Polywog4 and Polywog5) are used as an activation function in the hidden layer’s wavelons. The best prediction results were obtained with Polywog1 function.