Hyperparameter Tuning for Cycle Reservoir with Regular Jumps

Enhanced MFO, CO, and PSO with crossover, mutation, and Lévy flights optimize CRJ hyperparameters for accurate time series prediction.

Metaheuristic Hyperparameter Tuning for Cycle Reservoir with Regular Jumps (CRJ) Using Enhanced MFO, CO, and PSO Algorithms

This MATLAB project focuses on efficient hyperparameter optimization of the Cycle Reservoir with Regular Jumps (CRJ) model—a variant of reservoir computing closely related to Echo State Networks (ESNs)—tailored for nonlinear time series prediction tasks. The CRJ model exhibits complex dynamics governed by multiple interdependent parameters, which necessitate advanced tuning strategies to achieve optimal performance.

To address this, the code integrates and compares three population-based metaheuristic optimization algorithms: Moth–Flame Optimization (MFO), Cheetah Optimizer (CO), and Particle Swarm Optimization (PSO), alongside their enhanced variants. These enhanced versions incorporate genetic crossover, mutation mechanisms, and Lévy flight operators to improve exploration and exploitation trade-offs during the search process. By embedding these stochastic operators, the algorithms exhibit improved convergence behavior and robustness against local optima.

The optimization framework is tested across multiple benchmark time series datasets, aiming to minimize error metrics such as Normalized Mean Squared Error (NMSE), Root Mean Square Error (RMSE), and maximize the prediction accuracy R2. The results demonstrate that the enhanced metaheuristics significantly outperform their baseline versions, making them suitable candidates for tuning CRJ models in real-world predictive modeling scenarios.