Abstract
A novel optimization algorithm called musical chairs algorithm (MCA) is introduced in this paper for a shorter convergence time and lower failure rate compared to other optimization algorithms. This idea is implemented by using several search agents at the beginning of optimization steps and reduces this number gradually by removing the worst solutions. This idea is inspired by the musical chairs game in which the players and chairs are reduced by one in each round of the game. The proposed methodology has been compared with 10 other optimization algorithms for 10 benchmark functions. The results obtained from this comparison study showed superior performance of the MCA compared to the other optimization algorithms for all benchmark functions and different numbers of search agents. The convergence time varied between 3.8% and 18.2% compared to the average convergence time for all optimization algorithms applied with all benchmark functions. At the same time, the failure rate of the MCA is 0% for all the benchmark functions, but other optimization algorithms give a percentage of the failure rate. Moreover, the MCA is applied to feature selection of load forecasting as a real-world application which is vital for smart grid applications. The MCA is modified from continuous to binary MCA (BMCA). The BMCA is compared to several optimization algorithms, where it outperforms other optimization algorithms to select the best set of features to quickly and correctly learn the forecasting model. The accuracy of the obtained results from BMCA is increased to 96.7% compared to 70% of other algorithms, and the convergence time is reduced to 0.096 s compared to 18 s for other optimization algorithms. The outstanding results of MCA compared to the other optimization algorithms prove its superiority for all types of applications even with very complex benchmark functions.