广义回归神经网络的改进及在延迟焦化建模中的应用

GENERALIZED REGRESSION NEURAL NETWORK AND ITS APPLICATION TO DELAYED COKING PROCESS

广义回归神经网络 (GRNN)具有明确的概率意义 ,其参数大多能自动确定 ,仅光滑因子参数需优化估值 .采用优进遗传算法 (EGA) ,将确定性与随机性寻优操作相融合 ,实现了高效全局搜优 ,它所基于的优进策略包括设计Powell寻优算子、改进交叉算子、自适应地调整交叉率和变异率等 .以推广能力作为优化目标 ,所建的GRNN有很强的非线性拟合能力和优良的预报性能 ,将其成功地为延迟焦化过程建模 ,与径向基网络(RBFN)等相比 ,显示了明显的优势 .

This paper describes an optimal generalized regression neural network(GRNN) in which training of the network is optimization of the smoothing factors. A eugenic evolution strategy genetic algorithm(EGA), which integrates random evolution operation and deterministic optimization operation, is adopted in this paper to realize global optimization in high efficiency. The eugenic evolution strategies used in this paper include adding new deterministic Powell searching operation, improving crossover operation, modifying adaptive crossover probability and mutation probability, and others. The GRNN-EGA, which is based on EGA and provides powerful capacity in non-linear modeling and predicting, is applied to modeling delayed coking process to predict the productivity of coke. The GRNN-EGA prediction results are compared to those obtained with the radial basis function network(RBFN) and the GRNN-Powell, which is based on Powell optimization. The GRNN-EGA model has better prediction performance and stability as compared with the latter two models.