Research Of A Number Of Problems Of Combustion Control System In Power Plant

The boiler's combustion system is a multi-variables system with severe coupling , Disturbance ,time-variant and big amplitude property. The change of the parameters, such as blast capacity, induced-air amount, heat load , coal amount, steam amount can produce the direct disturbance to the combustion system. When the disturbance is big, there can be vibration in the whole combustion system ,which is a great threat to The safe running of the boiler。Unlike the conventional algorithm, Neural Network can be trained to accomplish the given task. If it is given some representative samples, namely, some groups of inputs and outputs, Neural Network can deduce the mapping relationship between inputs and outputs. After training, Neural Network can be used to identify the new data that is like any sample of training. Furthermore, Neural Network can even be used to identify the incomplete data or data with noise. This important character can be used in the area of prediction, diagnose and control.Considering two aspects of combustion control system identification and soft sensor of O2 content in flue gas , some fundamental research is carried out in this paper on optimization of combustion control system..Live data and RBF neural network are utilized in this paper on the offline identification of main variable of the combustion control system under settled process , which gained good effect and will present convenience for the further application of advanced control strategy.Nowadays , soft sensor is so important an area in process control industry that the measure to the leading process variable is necessary for the controlling ,optimization ,and testing in industry process, on which kinds of control method can be applied successfully. The accurate measure to the O2 content in flue gas is important to the economy of the combustion in boiler. This paper adopts the idea of Support Vector Machine which is popular in machine learning area to model and utilizes the live data sampling to the auxiliary variables associated with the O2 content in flue gas when a certain unit in power plant reduce its burden and try to carry on the soft sensor to the O2 content in flue gas. The results of experiment show that the method of Support Vector Machine by which soft sensor is carried on the O2 content in flue gas can gain the reserved precision, depends less on the samples and is powerful in generalization.