Two-phase optimization modelling with swarm computation and biomimetic intelligence learning for neural network training

• Through introducing the superiorities of artificial neural network (ANN) and swarm intelligence (SI)-based approaches, this research combines self-organized map (SOM) neural network (SOMnt) mode and the integration of ant colony optimization (ACO)-inspired with artificial immune system (AIS)-inspired approaches (IACI) algorithm to propose a two-phase hybrid of SOMnt mode and IACI algorithm (HSACI) method to tune relevant parameters of radial basis function neural network (RNt). • Further, the IACI algorithm takes advantages from these two approaches, and it is able to conduct prospecting and development in parallel for the population solution space. It can search globally and locally for multiple and concentrated solution sets, and receives the global optimal solutions. • This research applies the proposed two-stage HSACI method on a verification experiment with five nonlinear basis functions for function approximation, as well as on an exercise of practical demand prediction for a laptop product (HP-Pavilion series). This study aims to enhance the tuning efficiency of radial basis function neural network (RNt) through the self-organized map (SOM) neural network (SOMnt) mode and several swarm intelligence (SI) algorithms. Next, ant colony optimization (ACO)-inspired approach and artificial immune system (AIS)-inspired approaches is integrated into the integration of ACO-inspired and AIS-inspired approaches (IACI) algorithm, which is then applied to RNt for modulation. The proposed two-phase hybrid of SOMnt mode and IACI algorithm (HSACI) method, offers diversity and incorporates intensive solutions to achieve optimized explication. The population variety characteristic demonstrates a higher success rate in reaching global extreme values in the five nonlinear function problems, replacing restricted local extreme values. The verification results indicate that the combination of SOMnt mode, ACO-inspired, and AIS-inspired approaches is a distinctive method and accordingly a two-phase HSACI method is proposed, which is capable to adjust to the best precision among relevant algorithms in this paper. The method is then evaluated on five nonlinear function problems as well as the results from an actual laptop demand forecasting exercise in Taiwan. The results demonstrate that the proposed two-phase HSACI method outperforms the relevant algorithms and the Box-Jenkins models in term of precision.