Abstract:

The combustion chamber temperature model of a gas-steam combined cycle unit is nonlinear and strongly coupled, so it is difficult to establish an accurate process control model.To solve the problem, an Elman neural network-based combustion chamber temperature model is proposed in this study. This model uses the output response under different inputs as training set data, and uses the advantage of Elman neural network to approximate non-linear systems with arbitrary accuracy to train Elman neural network. The BPTT algorithm is adopted to back-propagate errors over time, and the SGD algorithm is utilized to optimized network weights. The experimental results show that each index is better than the original transfer function model. The ITAE index of the Elman neural network model under the unit step input signal and the unit ramp input signal are 16.103 4 and 8.990 1, respectively, and the errors of the outputs tracking inputs are 0.039% and 0.035%.

Key words: Elman neural network, combined cycle, combustion chamber, temperature model, non-linear system, ITAE