Improved AEA Algorithm And Its Application To Process Modeling

Alopex (Algorithm of Pattern Extraction) was firstly put forward by the medical experts, which was used for visual receptive field research. Later, it was discovered that Alopex also can be used for engineering optimization and obtained good results. The information of provious change of independent variables with respect to the objective function value are used as heuristic information, at the same time, annealing and probability are introduced. Therefore, Alopex possessed the characteristics of both gradient descent and simulated annealing. Based on the basic Alopex algorithm, an AEA (Alopex-based evolutionary algorithm) algorithm was perfected in this paper. Furthermore, on the basis of AEA, EDA (Estimation of Distribution Algorithm) was embedded into AEA for the purpose of improving the population diversity and distribution, owing to a fast convergence speed and unique evolutionary model owned by EDA. Eventually, a hybrid evolutionary algorithm EDA-Alopex was proposed. In AEA, two populations need to be produced for Alopex operation iteratively and the evolutionary information contained in these two populations largely determines the performance of AEA. In EDA-Alopex, inspired by the unique evolution pattern adopted by EDA, the two populations required by AEA are generated by EDA firstly, and then these two populations are carried on Alopex implementation. By the introduction of EDA, the population includes not only the micro-level correlation information between variables containd in the original AEA, but also contains the global probability information which descirbes the solution space distribution. In this way, the population can realize evolution in both global and local levels.Firstly, AEA was tested on large number of benchmark functions and compared with the basic genetic algorithm, particle swarm algorithm and differential evolutionary algorithm according to convergence rate, convergence accuracy and test functions optimization results and so on. The simulation results demonstrate that AEA is superior to these basic evolutionary algorithms. AEA not only possesses a fast convergence speed, but also obtains relatively high convergence accuracy. Later, a comparison test between EDA-Alopex, EDA and AEA was carried out on a series of high-demensional functions. The comparison results show that the EDA-embedded AEA is obviously better than AEA, and the performance of AEA is improved significantly by the introduction of EDA, both the convergence speed and the solution quality are all improved. After, EDA-Alopex algorithm was applied to model the depth of Ethylene Cracking for the ethylene cracking furnace; simulation results demonstrate that the soft sonsor model well reflects the actual operation situation of the production facilities. The model effectively extracts the interrelation between operational variables of the cracking furnace and cracking depth. The error between the output of the model and that of gas analysis instrument is minor. Finaly, EDA-Alopex was used for the estimation of the chemical kinetic parmeters for three lumped model of heavy oil pyrolysis; the reaction kinetics model obtained by our method compares with others reported in the literature, our model can fit the experiment data with less deviation.