Powell-DE算法在软测量参数预估中的应用

提出一种新的全局集成优化算法(Powell-DE算法)．该算法在差分进化算法中引入具有强局部搜索能力的Powell算法，克服了DE算法收敛慢且易陷入局部最优等缺陷，几个典型测试函数的仿真实验和比较，验证了新算法的有效性，体现了新算法的优越性；然后，Powell-DE算法被用于基于LSSVM模型的蒸发过程出料溶液NK的浓度的软测量中，并与其它2种方法的比较，现场工业数据验证表明新模型体现出了更好的跟踪性能且精确度高，能够完全应用于出料溶液NK的浓度的在线预测.

Powell-DE Algorithm and its Application in Estimating Soft Sensor Parameter

A new global optimization algorithm (Powell-DE algorithm) was introduced. Powell algorithm with strong local search ability was integrated into the differential evolution algorithm to overcome shortcomings such as slow convergence and the tendency to fall into local optimum. Meanwhile, typical test function simulation was used to test and compare Powell-DE algorithm and to verify the effectiveness of the algorithm and reflect the superiority of the new algorithm. Then, Powell-DE algorithm was used to soft measure the NK concentration on the basis of LSSVM model in the evaporation process, and was compared with the other two methods. The new model shows better track performance and high accuracy, and can be applied to the online prediction of the NK concentration of the feed solution.