An Improved Particle-Swarm-Optimized Neural Network For Thermal System

Firstly, the PSO algorithm is analyzed systematically. In order to avoid premature convergence, the Hybrid PSO (HPSO) algorithm is presented by introducing adaptive mutation operation, chaos, simulated annealing and niche technology etc. Simulation results show that the HPSO has better convergent speed and accuracy than standard PSO on four benchmark function optimization problems. Secondly, a hybrid algorithm combining PSO algorithm with BP algorithm, also referred to as PSO-BP algorithm, is proposed to train the initial weights of neural network controller (NNC). The simulation results show that NNC based on PSO-BP algorithm can effectively control main steam temperature system and make the controlled object have a quite good disturbance resistance and a strong robustness. Lastly, the simplified subtractive clustering and PSO-BP algorithm are proposed for Takagi-Sugeno fuzzy model on structure identification and parameters optimization. T-S model is built according to the field data of the main steam temperature system. The experimental results demonstrate that T-S modeling by means of the proposed methods has better identification accuracy and generalization ability.