Abstract:

The boiler combustion system in coal-fired power plant is a so complicated and important system that it is difficult to build a precise and adaptable model for it. First, the input and output data of combustion system collected from the scene was processed and selected for the combustion system modeling. Second, the Echo State Network （ESN）, which was a new type of Recurrent Neural Network, was improved and its precision and adaptability were increased. Then, the improved ESN was applied to building static model of combustion system, which had best adaptability compared with models founded with other four Neural Networks. Last, the improved ESN was applied to building dynamic model of combustion system, which had better adaptability and was more suitable for long-time prediction compared with static model.

Key words: coal-fired power plant, boiler combustion system, echo state network（ESN）, static model, dynamic model