Research On Short-Term Wind Power Prediction

The randomness and intermittent of nature wind can not guarantee a stable electric power output, which has brought a lot of problems and challenges to the operation of power grid. Short-term wind power forecast is an effective way to reduce adverse effects of wind power and increase the proportion of installed capacity of wind power in power system. The length of 0-5 hour and 0-24 hour prediction of wind power are studied respectively.Time series method based on robust estimation is introduced to study 0-5 hour prediction of wind power. Firstly preprocess the data, and then use the least squares method and robust estimation method respectively to build an autoregressive integrated moving average model, finally forecast the wind power in the next 30 minutes, and repeat 10 times. The results show that use the time series model based on robust estimation to predict wind power, the forecast percentage error of most points is under 5 percent, except one point of 10.1 percent. The forecast error is significantly smaller than conventional time series. It proves robust estimation methods can get a better forecast accuracy when the data have few outliers.The generalized regression neural network is used to predict the 24 hour wind farm output. The both cases whether numerical weather prediction is added to predict wind power or not have been compared. Firstly, train the wind power data of first 15 days, model through cross-validation, and then predict wind power of the next day. Secondly, add numerical weather prediction information, train the history wind speed and power, and then enter wind speed of the next day forecasted by the numerical weather prediction (NWP) to pretect wind power of the next day. An example shows that using the GRNN model to predict wind power of the next day is effective, predictions can track the actual wind power. When numerical weather prediction information is added into the neural network to predict, forecast accuracy is higher.