Dynamic soft sensor of boiler drum stress based on multi-agent system frame

On-line monitoring the stress of the boiler drum, is a very important content of power plant safety management. For traditional methods do many simplifications in computing drum stress, accuracy of the results can not be ensured. Finite element method can make fine accuracy, but it can not be used in on-line computing for its large computational quantities. So finite element method is only used in off-line optimal design. The nonlinearity of neural network is used in constructing the mapping relation between drum stress and temperature series of drum wall and pressure of drum. For pressure and temperature of drum vapor vary in a wide range, multi-agent system (MAS) design method was introduced. A big neural network can be divided into many small sub neural networks by this method, so the structure of sub neural network is smart and has good learning and generalization performance. Model of MAS can decide which sub neural network to response for the inputs, according to the pressure and temperature of the drum via classification of the perception part, therefore resource of system is spared and response speed is enhanced. By simulation, the accuracy of results can meet the industrial demand and this model can be used in practice.