变尺度粒子群算法及其在工业软测量中的应用

A novel mutative scale particle swarm optimization algorithm and its application in industrial soft sensor modeling

粒子群优化算法是一种基于群体智能的随机优化算法,具有收敛速度快、设置参数少、算法简单、容易实现等优点,其缺点是容易陷入局部最优解。变尺度法是一种可靠的局部快速寻优方法。为了解决了基本粒子群优化算法易陷入局部最优的问题,本文提出了一种基于变尺度方法的自适应变异粒子群优化算法。在本文算法中,粒子群每进化一代后,对所有粒子执行变尺度搜索,寻找更优个体,从而使算法具有动态自适应性,能够较容易地跳出局部最优。在延迟焦化生产过程中,汽油干点是衡量汽油的一个关键指标,建立汽油干点的软测量对延迟焦化生产实现卡边控制和提高装置的经济效益是有必要的。在实际生产过程中,无法在线测量延迟焦化汽油干点,只能采用离线实验室分析的方法获得,但离线分析不能满足控制的要求。基于软测量技术而开发的延迟焦化汽油干点软测量模型,使汽油干点的在线测量成为可能。目前,工程上一般采用BP神经网络来训练软测量模型。BP神经网络的学习算法是决定BP神经网络预测质量的关键。鉴于此,本文将所提出的变尺度粒子群优化算法用于BP神经网络学习过程中,并将本文方案的预测结果与文献方案进行了对比实验。实验结果表明,与文献方案相比,本文方案具有较好预测精度和良好的泛化能力,具有较好的应用价值。

The particle swarm optimization algorithm is a kind of stochastic optimization algorithm based on swarm intelligence, has some advantages such as fast convergence rate, less set parameters, simple and easy to implement, and etc, and its shortcoming is easy to fall into local optimal solution. Variable metric method is a reliable local optimization method. To solve the problem that particle swarm optimization algorithm is apt to trap in local optimum, a novel particle swarm optimization algorithm with mutative scale was proposed. The mutative scale learning factors were introduced in this new algorithm, so that proposed algorithm can easily jump out of local optimum. In the production process of delayed coking, the gasoline endpoint is a key indicator to measure the gasoline, the establishment of the soft sensor of gasoline endpoint is necessary for achieving card edge control of delayed coking production and improving the economic efficiency of the device. In the actual production process, we can＇t online measurement the delayed coking gasoline endpoint, only use offline laboratory analysis methods to obtain the gasoline endpoint, but the off-line analysis can not meet the control requirements. The development of soft sensor model for delayed coking gasoline endpoint based on soft measurement technology; make online measurements of the gasoline endpoint become possible. At present, the BP neural network is using to train the soft measurement model generally. The learning algorithm of BP neural network is the key to determine the performance of BP neural network, The proposed solution is compared with the literature solution by using experiments. The test experiments show that, compared with the literature solution, the proposed solution has better predict precision and better generalization performance, and has good application value.