Hybrid Hiking Sparse Particle Swarm Optimization for Dynamic System Identification

Presently, system identification has become a key approach for dynamic and nonlinear systems. Dynamic system identification is the creation or upgrading of a mathematical model based on observed data. Still, the identification of dynamic system behaviors with a huge number of variables is computationally complex and leads to overfitting. To tackle such complexities, this paper proposes a new efficient model named Hiking Sparse Particle Swarm Optimization (HiSPSO) for dynamic system identification. The HiSPSO model is the integration of Hiking Optimization (HOA) and Sparse‐Based Particle Swarm Optimisation Algorithm (Sparse‐PSO). The iterative process for dynamic system identification is designed and validated using two nonlinear processes, like Cascaded Tanks and Coupled Electric Drives, as well as three linear dynamic processes, such as Hairdryer, Heat Exchanger, and Glass Tube Manufacturing. The execution of structured sparsity regularization is carried out using Neural Networks (NNs) with the incorporation of group‐sparsity inducing priors. The training of the model is carried out by the proposed HiSPSO, and the proposed approach effectively trains the linear and nonlinear system identification. Moreover, the effectiveness of the model is validated using the Normalized Mean Square Error (MSE), Normalized Root Mean Square Error (RMSE), Normalized Nash–Sutcliffe Efficiency (NSE), Normalized R‐Squared ( R 2 ) parameters, Percent Bias (PBIAS), and Willmott's Index (WI), with corresponding values of 0.526, 0.725, 0.527, 0.547, 0.568, and 0.558 attained. The results demonstrate the superior generalization and robustness of the proposed HiSPSO‐based model.