【摘要】 为实现水泥窑尾分解率的实时在线检测,利用软测量技术在解决工业在线测量问题中的优势,提出一种改进的粒子群参数优化的支持向量回归机算法(IPSO-SVR),即通过粒子群算法对支持向量机模型核心参数进行优化选择,并在粒子群算法中引入自适应惯性权重的思想,克服粒子群算法容易出现早熟收敛、陷入局部极值的缺点,最终建立起基于IPSO-SVR的窑尾分解率软测量模型。将其与基于交叉验证法(CV)和未改进粒子群算法优化SVR参数的软测量模型进行仿真对比实验,实验表明:该IPSO-SVR模型具有更佳的预测能力,窑尾分解率预测相关系数达0.857 5,预测最大相对误差不超过1.14%,平均相对误差为0.75%,可进一步运用到诸如水泥生产等大型工业的产品分解率预测中。更多还原

【Abstract】 A support vector regression( SVR) algorithm optimized by improved particle swarm optimization( IPSO) is proposed in this paper to realize the real-time online detection of cement kiln tail resolution ratio. IPSO-SVR is based on soft sensor measurement technique which solves the problem that some important industrial process parameters cannot be directly online measured. Adaptive inertia weight is introduced into PSO to avoid premature convergence and getting trapped in local extremum. The cement kiln tail resolution ratio soft sensor measurement model based on IPSO-SVR is built with the core parameter selection of SVR optimized by using IPSO. Simulation experiment proves that IPSO-SVR has a better forecasting ability compared with model based on SVR optimized by cross validation and PSO-SVR. The correlation coefficient is 0. 857 5. The maximum relative error is 1. 14%,and the average relative error is 0. 75%. It shows that IPSO-SVR can be further applied to the prediction of product resolution ratio in large scale industries such as cement production.更多还原