Studies On Hybrid Multi-objective Optimization Method And Its Applications To Boiler Load Distribution In Power Plant

The multi-objective optimization problems (MOPs) have multiple conflicting objectives, and in general, there is no an optimal solution that could optimize all objectives simultaneously. The traditional methods are more susceptible to the influences of MOPs’ characteristics as well as human subjective factors. However, evolutionary algorithms (EAs) could solve MOPs more effectively due to their superior ability in global search. In addition, EAs could generate a set of solutions in each optimization run because of its population based mechanism, and the diversity of solutions. A considerable amount of references have demonstrated the effectiveness of multi-objective EAs. But many EAs are easily trapped into local optimum, and get convergence prematurely when dealing with MOPs, especially to some complex problems, such as the problems with multi-modal functions, numerous decision variables, multiple local Pareto optimal fronts, discontinuous or high-dimensional search space, etc. Now, people tend to face variety of complex MOPs, it is high demand to develop more advanced MOP solutions with high accuracy, stability, robustness, applicability and efficiency, etc.When solving MOPs by particle swarm optimization (PSO), how to avoid premature convergence, and get higher accuracy and efficiency is the focus of this study. Introducing extremal optimization (EO), a hybrid multi-objective optimization algorithm called PSO-EO was proposed. The performance of PSO-EO solution was verified through a series of unconstrained benchmark problems. Moreover, the approach to deal with constraints and discrete decision variables has been developed and applied to a number of engineering design MOPs, the simulation results show that PSO-EO has good performance. Finally, the PSO-EO solutions with the relevant software were successfully applied to load dispatching for multi-boilers systems in production-scale thermal power plants.The specific content of the thesis could be summarized as follows:(1) Based on a brief introduction to evolutionary computation, the research activities on the multi-objective evolutionary optimization algorithms are reviewed with their shortcomings.(2) To avoid premature convergence and improve the efficiency of PSO, in the study a hybrid multi-objective optimization algorithm was proposed, which relies on PSO for its fast and powerful search ability and relies on EO for its ability to escape from the local optima in search dynamics. The marriage of PSO and EO is able to improve both search speed and capabilities. To compare with some other popular MOP algorithms, the proposed PSO-EO soltions are tested with a number of unconstrained benchmark MOPs. The simulation results show the efficiency, applicability and diversity of the proposed solutions.(3) In practice, there are some constraints and/or discrete decision variables in real-world engineering design MOPs. In this study, the relevant PSO-EO solution with dealing both constraints and discrete decision variables has been developed and tested with a number of popular benchmark problems. The simulation results show the PSO-EO solutions may provide better performance in search dynamics for both convergence and diversity.(4) The proposed PSO-EO has also been tested and applied in the load dispatch for multi-boiler systems in production-scale thermal power plants. The integrated system consists of three main parts:energy efficiency monitoring and evaluating, short-term load demand forecasting and load dispatching. The energy efficiency and emissions monitoring and evaluation is based on the combustion system energy balance of boilers; the short-term load demand forecasting is based on BP neural network; according to the forecasting results, the load dispatching of multi-boilers system is solved by PSO-EO to provide decision supports for the operation of the plant. The study was developed into an integration system software, and passed the software evaluation of the Zhejiang Software Testing Center (2010No.5262) in April2010. The system was carried out offline test and online test run based on the actual production data of one project demonstration (a thermal power plant in Dongguan), and achieved good results.