摘要
丙烯精馏塔关键组分塔顶的丙烷浓度和塔釜的丙烯浓度的准确测量是乙烯生产企业提高丙烯收率的关键。鉴于丙烷浓度和丙烯浓度分析仪经常出现故障,提出以RBF神经网络加协同随机粒子群优化(PSO)算法的软测量建模法,即利用RBF神经网络的局部逼近能力来获得模型的结构,利用协同随机PSO算法的全局搜索能力来优化模型的参数,提高模型的逼近能力和泛化能力。该方法克服了BP网络对初始值和网络结构敏感,容易陷入局部最优的缺陷,以及RBF网络全局逼近能力差的缺点。仿真结果表明,此方法所得软测量模型精度高,泛化能力强。
The accurate measurement of the concentration of propylene on the top of the propylene fraction column and propane on the bottom of the propylene fraction column is the key to increasing the propylene production.Based on radial basis function（RBF） neural network and cooperative random particle swarm optimization（CRPSO）,the soft sensor method is proposed to solve the problem that the analyzers,used to measure the concentration of propane and propylene,are frequently broken-down.In the proposed method,the local approximation capability of RBF neural network is utilized to obtain the structure of the model,and the global approximation capability of CRPSO is used to optimize the parameters of the model so as to improve the approximation ability and generalization ability of the model.The proposed model overcomes the drawbacks of BP neural network,such as its sensitivity to initialization and network structure,as well as being easily trapped into local optima.Furthermore,it also enhances the global approximation capability of the RBF neural network.The simulation results show that the developed method is effective in finding accurate and well-generalized soft sensor models.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2010年第4期557-560,共4页
Computers and Applied Chemistry
关键词
RBF神经网络
粒子群优化算法
协同随机粒子群优化算法
软测量
丙烯精馏塔
RBF neural network
particle swarm optimization
cooperative random particle swarm optimization
soft sensor
propylene fraction column
