Information Hiding in QR Code

Abstract

When implementing a wind power system, an accurate forecasting model should be built in order to predict the wind power generated in the near future. This is crucial to regulate the exact power generated from base power plants which in turn prevents the waste in generating unnecessary energy, vice versa. In this project, a short to long term wind forecasting was built and evaluated based on the date, time, predicted wind attributes and normalized wind power across 7 unknown wind farms in the period between 1/7/2009 and 31/12/2010 using MATLAB. Among various modelling approaches, Multiple Linear Regression and K-Nearest Neighbours (KNN) were chosen to analyse according to the properties of dataset. To evaluate the performance of the predicted model, two actions were done. First, the data set was separated into training set and test set. Period between 1/7/2009 and 30/06/2010 was used as a training set to build the models while the remaining was treated as a test set. Second, the rooted mean squared error (RMSE) between the forecasted and actual wind power was calculated. RMSE is lower if the model is more accurate. The built models can predict any period of wind power forecasting using the date, time, predicted wind speed and predicted wind components of the target predicted day. In conclusion, satisfactory models can be built under limited information. The values of wind power for the next 24 hours across 7 wind farms are forecasted with 0.14 RMSE, while the prediction of wind power for the test set is under 0.2 RMSE. As both models have similar performances towards the datasets in this project, KNN is preferred in terms of simpler algorithm and assumptions. Due to the physical background, humidity and air density are key factors in determining wind power. If they are included, the model can be more accurate.