|
Abstract:
|
Today , internet becomes one of the most important resources for useful information . However , since the authentic of the information is difficult to verify , one also has to take precaution when getting information from the internet . Utility companies need to forecast their load for unit commitment scheduling and system planning . The traditional approach for neural network forecasting relies on the temperature forecasting information from a single source . The customer loads are closely correlated with the temperature . Therefore , the accuracy of the load forecasting is affected by the temperature forecasting errors . The objective of this thesis is to reduce the temperature forecasting errors by using artificial neural network (ANN ) to preprocessing the temperature forecasting information from various resources . In this thesis , temperature information from five (5 ) websites have been used fro this process . Each website provides hourly forecast temperature of 15 days . A JAVA program is designed to extract the useful temperature information from each individual web site and record them into the database (MySQL ) . Depends upon the available data , an ANN is then used to forecast hour ahead and day ahead temperatures up to seven days . Through this preprocessing , better weather information is obtained to have more accurate load forecasting results . |