Predicting Wind Farm Electricity Output: A Neural Network Empirical Modeling Approach

Wind energy is rapidly emerging as a substantial contributor to the electricity generation capacity of utilities around the world. While the use of wind power both adds to the electricity supply and offers significant environmental benefits as a renewable source of energy, the stochastic nature of forces that produce wind energy prevents relying on it to meet base load requirements. Intermittent availability also presents stability and control issues which grid operators must address before the potential benefits of wind energy can be fully realized. A fundamental requirement for successful control strategies is an accurate short-term prediction of wind farm output. Over the longer term output predictions also provide the foundation for revenue forecasts critical to enterprise operations. Inherent variability in key inputs suggests the use of empirical models. Neural networks comprise a collection of algorithms that yield robust empirical models. A neural network engine that incorporates a genetic algorithm for variable selection and employs cascade correlation to dynamically define the neural network architecture is introduced. Preliminary results obtained from prediction models for an operating wind farm are presented, along with directions for future work.