Research On Thermal Process Modeling Of Large Thermal Power Units Based On Running Data

With the informatization level's improvement of power plant, the system research which is based on the data is gradually aroused people's concern. The complex thermal system modeling based on the running data is one of research hot spots. This method is effective to overcome the deficiency of mechanism modeling, and the heavy workload of routine test modeling. On the basis of introducing modeling method and the analysis of thermal running data's characteristics, this article studies the principal component analysis method and extreme learning machine algorithm, and applies them to do the thermal process modeling research. Taking superheated steam temperature system of supercritical unit as an example, a neural network model based on extreme learning machine is established by using running data.Firstly, for the complexity and redundancy of the running data, this article uses the principal component analysis method for the choice of auxiliary variables model. This method can transform the correlation of the original input variables into unrelated principal component, which effectively overcomes the system redundancy between the input variables, optimizes the model input variables, and ensures the accuracy and effectiveness of the model.Secondly, put forward a kind of modeling method based on extreme learning machine and kernel extreme learning machine with the analysis of the selection of model parameters and model validation method. Use principal component analysis and extreme learning machine algorithm for Box- Jenkins gas data modeling and the simulation results is analyzed and compared with the BP neural network model and support vector machine(SVM) model.Finally, the neural network modeling method based on extreme learning machine is applied to the superheated steam temperature system modeling. The corresponding model is established based on supercritical unit running data. The results show that extreme learning machine algorithm has less parameters and fast identification; kernel extreme learning machine with speed advantage, also has the characteristics of high precision and stability, and the model can reflect the dynamic and static characteristics of the system. Extreme learning machine algorithm provides a new way for system modeling.