Mutation based improved dragonfly optimization algorithm for a neuro-fuzzy system in short term wind speed forecasting

Abstract

The Dragonfly algorithm (DA) is a heuristic optimization algorithm that is commonly used for complex optimization problems. Despite its widespread application, the abundance of social behaviors in its construct can lead to poor accuracy in solutions and an imbalance between exploration and exploitation phases. To overcome these issues, this paper proposes a mutation-based Dragonfly optimization algorithm (MIDA). In order to increase the solution accuracy of the original DA and reduce its handicaps, the proposed model includes three procedures, namely, mutation operation, boundary control, and greedy selection mechanisms. The mutation operator helps to find global optima by avoiding getting stuck at the local optimum point, while the boundary control and greedy selection mechanisms update the dragonflies at each iteration and thus use the better fitness value of the updated ones. The performance of the proposed MIDA is tested and shown to be superior to the original DA using several analyses such as convergence, search history, trajectory, average distance, computational complexity, diversity and balance. To validate the MIDA algorithm, it is tested on the 10, 30, 50 and 100 dimensions of the CEC2014 benchmark functions. Furthermore, a comparison against twelve state-of-the-art (SOTA) meta-heuristic optimization algorithms is performed. The performance of the proposed MIDA is also compared with the performances of some improved dragonfly algorithms taken from the literature. The statistical results obtained by the original DA and MIDA for ten minimization problems with 5, 10, 15 and 20 dimensions from the CEC2020 test suite are presented. Finally, the proposed algorithm is applied to optimize the ANFIS model parameters in order to be used for short-term wind forecasting as a real-world problem. The results obtained for the different dimensions of CEC2014 and CEC2020 test problems in the study show that the search performance of proposed MIDA is better than that of the original DA. MIDA outperformed the original DA by 91.33% for CEC2014 benchmarks and 94.25% for CEC2020 benchmarks. In addition, MIDA came first in comparison with twelve different SOTAs from the literature. In comparisons with different DA versions, MIDA took the second place in terms of statistical performance. Computational complexity was examined in the CEC2020 benchmarks and it was seen that MIDA has a more effective run time than DA. Finally, the short-term wind speed forecasting results of the ANFIS-DA hybrid model according to four different error metrics are more successful than those of the ANFIS-DA hybrid model.