A Gradient-Based Particle-Bat Algorithm for Stochastic Configuration Network

Stochastic configuration network (SCN) is a mathematical model of incremental generation under supervision mechanism, which has universal approximation property and advantages in data modeling. However, the efficiency SCN affected by some parameters. An optimized searching algorithm for input weights biases proposed this paper. optimization with constraints first established based on convergence theory inequality mechanism SCN; Then, hybrid bat-particle swarm (G-BAPSO) gradient information framework PSO algorithm, mainly uses local adaptive adjustment characterized pulse emission frequency to improve ability. The optimizes rate network. Simulation results over datasets demonstrate feasibility validity algorithm. training RMSE G-BAPSO-SCN increased 5.57×10−5 3.2×10−3 compared that two regression experiments, recognition accuracy 0.07% average classification experiments.