Electricity demand forecasting in Turkey and Indonesia using linear and nonlinear models based on real-value genetic algorithm and extended Nelder-Mead local search

Electricity demand patterns have many variables related to uncertainty behaviour such as gross domestic product, population, import and export. The characteristics of these variables lead to two problems in forecasting the electricity demand. The first problem is the fitness evaluation in the electricity demand
forecasting model in which more than one variable are included which leads to increase the sum of squared deviations. The second problem is the use of a single algorithm that failed to solve local optima. These problems resulted in estimation errors and high computational cost. Hybrid genetic algorithm (GA) and Nelder-Mead local search mode 1 has been used to minimize demand estimation errors.
However, hybrid GA and Nelder-Mead local search failed to reach the global optimum solution and involve high number of iteration. Hence, an electricity demand forecasting model that reflects the characteristics of electricity demand has been developed in this research. The model is known as the hybrid Real-Value
GA and Extended Nelder-Mead (RVGA-ENM). The GA has been enhanced to accept real value while the Nelder-Mead local search is extended to assist in overcoming the local optima problem. The actual electricity demand data of Turkey and Indonesia were used in the experiments to evaluate the performance of the proposed model. Results of the proposed model were compared to the hybrid GA and Nelder-Mead local search, Real Code Genetic Algorithm and Particle Swarm
Optimisation. The findings indicate that the proposed model produced higher accuracy for electricity demand estimation. The proposed RVGA-ENM model can be used to assist decision-makers in forecasting electricity demand.