Improvement of an Elman neural network model for once-through boiler unit and its verification

Abstract

A supercritical (SC) or ultra-supercritical (USC) once-through boiler unit is a typical multi-variable strong coupling system with large time-delay, slow time-variant and nonlinear characteristics, which often makes the coordinated control quality deteriorate under wide-range load-changing conditions, and thus leads to the unit slow load response and large fluctuations of key operating parameters. Therefore, it is of vital significance to study the once-through boiler unit's operation characteristic by means of modeling method, and improve the coordinated control quality with model-based advanced intelligent control strategy. In this paper, the improved Elman neural network based on inverse system models for a 600MW once-through boiler unit was established. An original standard Elman neural network and an improved Elman neural network with time-delay inputs and time-delay outputs feedback were established and compared in the simulation. The training data of model is the operation data over wide-range load-changing conditions for a 600MW once-through boiler unit. The off-Line and on-Line verification tests for load-changing conditions showed that the improved model with time-delay inputs and outputs feedback can fit the complex non-linear, dynamic characteristics between three inputs(fuel, feedwater flow, turbine governing valve opening) and three outputs (unit's load, main steam pressure, intermediate points temperature). Compared with the original model, the improved model is stronger in simulation accuracy and predict stability. It can meet the engineering application requirements for intelligent coordinated controllers' design with simple structure, high precision and strong generalization ability.