Development of a very short-term prediction system for windpower generation

Accurate forecasts are vital to efficient power system operation. This makes power systems forecasting research an important task. Windpower is presently the fastest growing power generation sector in the world; however, windpower is intermittent. Accurate forecasts are required to make the best use of this rapidly emerging technology. In fact, before large-scale penetration of windpower into a power system can be successfully managed it is important to be able to predict the rapid fluctuations of wind on a short time scale. This thesis concentrates upon wind parameter prediction on the \very short-term\" time scale. In traditional (meteorological) forecasting very short-term wind prediction has been relatively unnecessary. Very short-term forecasts of wind parameters were difficult to accurately obtain and were considered to be a low priority. However as windpower penetration increases the requirement for accurate forecasts also increases. The research for this thesis covered a variety of possible techniques which are discussed and critiqued. It was found that a forecasting system built using intelligent techniques showed great promise and the majority of the work went into developing and improving upon an Adaptive Neural Fuzzy Inference System (ANFIS) forecasting model. ANFIS is an intelligent system that is a hybrid between an artificial neural network and a fuzzy expert system that is capable of learning patterns and developing fuzzy rules and membership functions. A large number of forecasting systems were developed and tested and subsequently changes to the ANFIS model were instituted that enable more accurate very short-term wind parameter prediction. During this research one of the major obstacles was the lack of fine resolution time series data. This was possibly the largest single obstacle and it could not be remedied with any traditional techniques. The present probabilistic modelling techniques could not produce adequate results from a single time series due to the inherent data correlation between subsequent data points. Thus a novel data generation system needed to be developed. This thesis considers why windpower is important what effect it has on a power system the present techniques used for wind prediction and how an intelligent system may be more beneficial. It also provides two separate case studies that were used to test and develop prediction models. The models were all developed on data that would be readily available at any windfarms (without the need for additional equipment installation) and yet the results still manage to significantly outperform the present industry practice."