A study on the application of new approach to the forecasting of electric power demand and nuclear power share optimization

A new methodologies using an Artificial Neural Network (ANN), a Genetic Programming (GP) and Genetic Algorithms (GAs) are proposed to forecast long-term electric power demand and to optimize the share of nuclear power in Korean electric power system. Based on the results presented in this study, it is concluded that the ANN is suitable for long-term demand forecasting when it was trained by proposed strategy, and the GP can be successfully used to forecast electric power demand only using historical data. From the analysis of GAs, it is concluded that GAs can be used for system optimization more effectively than traditional methods. An ANN and a GP are proposed as a methodology for long-term forecasting of electric power demand. They are able to combine both time series and regressional approaches. They do not require assumptions for any functional relationship between dependent and independent variables. Moreover, since the result of the GP has a form of equation, it can be directly used in any computational codes for future electric power demand. The economic variables are used as a independent variables in long-term forecasting of electric power demand. The ANN can not make long-term forecast only using its historic data. In order to overcome this limitation of neural network, a new strategy is suggested to train the ANN. On the other hand, the GP can make forecasts only using the historic data. In addition, the GP is easier to use the result because it produces mathematical expression as the results. Among economic variables, only two variables (population and GDP) are used as independent variables for long-term forecasting of electric power demand, since annual electric power demand is mostly affected by those. Using Critical Heat Flux data, we validated that the GP also can be used for complex non-linear system. The GAs are suggested as a methodology to optimize the share of nuclear power in electric power system. The GAs can find optimal solution faster than traditional approaches. In addition, the GAs do not need any pre-process of matrice which are consisted of objective function and constraints. The results of GAs were compared with ones of linear programming to validate that GAs can find optimal solution