Abstract:

In the production of finished gasoline，octane number as the key indicators of quality of finished product gasoline and foundation of formula model，it is import to accurate measuring its content. However, due to the existing measurement technology and the constraints of complex conditions, it is difficult to obtain the component data effectively. Considering the problems of genetic algorithm (GA), particle swarm optimization (PSO), and optimizing back propagation (BP) network, a serial hybrid particle swarm genetic algorithm (serial hybrid PSO-GA, SHPSO-GA) is proposed to optimize the BP network, and it is used to predict and model the octane number. In this method, the output of PSO algorithm is divided into two populations according to the fitness value, and the bad ones are discarded and the good ones are retained. Then the optimized population was further optimized by GA crossover and mutation, the optimal population was applied to BP network parameter optimization. Finally, based on the method and the industrial historical data, a prediction model of octane number of hydrogenated gasoline was established. The simulation results showed that, compared with the traditional BP, GA-BP, PSO-BP, PSO-GA-BP and other methods, SHPSO-GA-BP had better prediction performance due to the deeper integration of PSO and GA, it could be used for octane number prediction.

Key words: octane number, prediction, neural network, genetic algorithm, particle swarm optimization, PSO-GA-BP neural network, optimization

摘要：

针对成品汽油调和配方建模中加氢汽油组分辛烷值难以实时获取，考虑遗传算法（genetic algorithm，GA）、粒子群算法（particle swarm optimization，PSO）优化反向传播（back propagation，BP）网络存在的问题，提出了一种串行混合粒子群遗传算法（serial hybrid PSO-GA，SHPSO-GA）优化BP网络，并用于辛烷值的预测建模。该方法首先将PSO算法的输出依据适应度值分为优劣2个种群，弃劣留优；然后对留优种群再进行GA的交叉变异操作，进一步优化种群，经过每一代PSO和GA的交替优化，并将最优种群用于BP网络参数优化；最后基于该方法和工业历史数据，建立了加氢汽油组分辛烷值的预测模型，仿真结果表明，较传统BP，以及改进的GA-BP、PSO-BP、PSO-GA-BP等方法，SHPSO-GA-BP由于将PSO与GA进行更优的深度融合，具有更好的预测性能，可以用于辛烷值的预测。

关键词: 辛烷值, 预测, 神经网络, 遗传算法, 粒子群算法, SHPSO-GA-BP神经网络, 优化