Research On The Optimization Methods Of Coal Blending In Power Plants

In china, coal is a primary energy source for power generation. And the optimization management for the coal burning is very important for the profit and safety of the power plants. On one hand, coal has occupied over 70% of the overall cost. On the other hand, the stability of coal quality means much for the safety of the operation. These years, many power plants have made much progress in optimization of coal blending.Currently, some power plants have established the simple linear model for the optimization of coal blending. However, there are some problems in the application: 1) Due to the nonlinear relationship between the blended coal and its components, there are deviations between the calculation results of the linear model and the real results; 2) Randomness lies in the coal property. As the linear model can not give out the right description of the real situation, it's urgent to improve such model.This paper focuses on the following problem:1) Establish the multi-objective nonlinear model of coal blending Due to the strong nonlinear property of the blended coal, and its deviation from the simple linear model used currently, it's necessary to build a new model. This paper improves the linear model, and establishs a multi-objective nonlinear model, which is more suitable for the real application.2) Research on the regression estimation methods of blended coalDue to the nonlinear property of the blended coal, this paper introduces methods like BP Network, Support Vector Machines for the regression estimation experiments. Focusing on the drawbacks of the experiments, this paper proposes some improvement suggestions. And this paper also does clustering analysis with Self Organized Networks and Fuzzy Clustering, and estimate the property of blended coal. Finally, we show the comparison between the estimation results.3) Multi-objective model solution with Genetic AlgorithmThe model become very complex because of the training methods like BP Network and SVM, and it is difficult to solve it. So this paper introduces the Genetic Algorithm for the solution. What's more, we integrate GA and BP...