Linear prediction of wind Speed variation in a wind power plant in Sri Lanka

Abstract

Wind speed-time series data typically exhibit autocorrelation, which can be defined as the degree of dependence on preceding values. This paper presents state of the art wind speed forecasting technique and compares the use of a standard class of statistical time-series model and dynamic neural network. Wind speed predictions are considered to improve the performances of optimum controls of wind turbines as well as it can be used to predict wind power feed-in to grid for evaluating grid security. In neural dynamic networks, the output depends on the current and previous inputs as well as the previous outputs of the network. The dynamic neural network method is more suitable for the time series prediction as it can be trained to learn sequential or time-varying patterns. For this study, a neural network incorporated with a tapped delay line and neurons in a hidden layer was used for wind speed predictions. The Levenberg-Marquardt algorithm was used to train this network by using selected data set. As well as present study shows how an autoregressive model can serve as predicting wind speed variation. Based on a number of historical data, pattern identification, parameter estimation, model checking are utilized to make a mathematical model of the time series data prediction. Statistical models have been used for time series analysis and an autoregressive model (AR) has introduced in this paper and compare with dynamic neural network method.