Optimization of sensor location for improving wind power prediction accuracy

Wind power becomes more and more popular in renewables, which has a large proportion in total capacity of the grid system. Since operations of the grid system dependent a lot on the prediction accuracy of power generation, improving the prediction accuracy attracts a lot attention in researchers. Sensors, which can measure weather information, especially wind information, have been used in wind power forecasting to have a better prediction results. However, location of these sensors are usually decided imprudently, because there are no rules indicating where to place these sensors. This paper proposes a method to find an optimized place, where represents the characteristics of the whole wind farm, to install meteorological sensors. Firstly, a support vector machine (SVM) is used as a crude model to predict total wind power output of a wind farm. Then, a group of existing sensors are used as auxiliary inputs of the crude model to improve the prediction results. Improvements are calculated based on sample path analysis during the whole dynamic process. Finally, an optimized place is calculated according to the parameters fitting of a two dimensional normal distribution. A case study is given based on a real wind farm to demonstrate the proposed sensor location optimization method.