【作者】 王天凤；

【导师】 满春涛；

【作者基本信息】 哈尔滨理工大学 ， 导航、制导与控制， 2011， 硕士

【摘要】 保证过程安全生产的前提下使系统稳定运行在最佳工况,获取最大经济效益。随着现代工业生产过程日趋复杂,过程呈现出强关联性、严重非线性及不确定性,再加上十分苛刻的生产条件和环境,很难保持生产过程在最佳工况运行。本文以稳态工业生产过程为背景,研究补偿模糊神经网络(Compensatory Fuzzy Neural Network,CFNN)学习算法的改进和网络权值的优化;针对基本遗传算法在优化方面的不足做了有效地改进。最后,将改进的GA应用于CFNN参数的估计和优化中,并取得了良好的效果。首先,本文分析了能够执行补偿模糊推理的补偿模糊神经网络,该网络解决了模糊神经网络设计过程中的一些问题,针对初始网络模型建立难的问题引入基于减法的模糊c-均值聚类算法确定模糊规则数。基本遗传算法(SGA)在寻优过程中,存在易于陷入局部最优、不同编码方式有解码误差、收敛速度慢等缺点。本文应用实数编码使计算简单并极大的节约了运算空间。采用稳态复制,非均匀算术交叉和非均匀变异。针对传统自适应遗传算法的可能陷入局部最优的不足,提出一种改进的自适应遗传算法,该改进GA算法避免了“早熟”问题并保证了算法的收敛性,提高了收敛精度和速度,通过对Shubert函数仿真分析,验证了改进GA的有效性。用改进的GA和BP算法的混合算法优化调整CFNN的权值,该混合算法综合了GA的全局收敛性和BP算法较强的局部搜索能力,极大提高了CFNN的全局逼近能力和收敛速度。最后以过氧化氢异丙苯(CHP)分解过程为控制实例,采用IAGA-CFNN建立CHP分解模型及改进的GA求取最优解,仿真结果分析表明,该算法是一种更有效的稳态优化方法。更多还原

【Abstract】 The steady-state optimization of complicated industrial process is the effec-tive mean for increasing economic benefit, the ultimate end is made system can run stably at optimal operating condition by the premise of guarantee of process safety production for the best economic benefit. With the modern industrial proc-esses becoming more and more complex, the processes show strong associations, nonlinearity and uncertainty, and it is hard to keep the working condition at the best state. Thus, the steady-state industrial production process is taken as a back-ground. This paper mainly studies on improvement of the learning algorithms of CFNN and optimization of network weights. GA algorithm has been effectively improved in accordance with the deficiencies of optimization ability, and using the improved GA algorithm to parameter estimation and optimization in CFNN, it has achieved good results.First, this paper analyses compensatory fuzzy neural network to perform fuzzy reasoning solves the difficulties in the process of designing fuzzy neural network. Initial network model is established by using fuzzy c-means clustering based on subtraction clustering.In the searching process of simple genetic algorithm, there is easy to get into local optimum, the decoding error in encoding methods, and slow convergence speed etc. The paper uses real coding with simple computation and to save com-puting space greatly. Uses steady reproduction, non-uniformity crossover and mutation, in accordance with the deficiencies of getting in local optimization of conditional AGA, the paper presents a improved AGA, the improved GA algo-rithm prevent premature problem and increase the convergence speed of the algo-rithm. Finally, its effectiveness is verified by simulation Shubert function.A hybrid algorithm based on improved GA and BP algorithm for optimizing and adjusting CFNN weights, it colligate the global convergence of GA and strong ability to search locally of BP algorithm, greatly improved global ap- proximation ability and convergence speed of CFNN. Finally, cumene hydroper-oxide (CHP) decomposition process is used as the control example, using im-proved IAGA-CFNN to build CHP decomposition model and the IAGA to obtain the optimal solution. The simulation results shows that the algorithm is a more feasible and efficient Steady-state optimization method.更多还原