The application of rule-based knowledge to load forecasting of electrical power systems

Abstract

The thesis describes short-term load forecasting by an expert system approach based on knowledge engineering. Conventionally, short-term load prediction is based on mathematical models which either extract the mathematical properties in the time series of load data, or present the static causal relationships between the load demand and its effective factors. The conventional methods can predict the electrical demand under normal situations, but not for special events. The thesis proposes a new approach to estimate the loads for special events, such as time change-overs, public holidays, which is mainly based on knowledge about the system load. Based on the ARIMA model, modifications have been made to predict weekend loads, which take the weather effects into consideration. The thesis also proposes a method to disaggregate the overall load into its components in order to study the relationships between the components and the causal variables. The time change-over (from Greenwich Mean Time to British Summer Time and vice versa) effects can be considered by separately estimating the lighting load and the rest load. The thesis investigates the holiday load characteristics and presents different estimation methods for different public holidays ranging from normal Monday Bank Holidays to Christmas Day holiday periods. Knowledge about the load is represented in production rules. The proposed estimation methods are written in POP-11 which can be interfaced with FORTRAN in which the ARIMA model is programmed for the prediction of the load under normal situations.