基于混沌粒子群支持向量回归的高炉铁水硅含量预测

参数的优化选择对支持向量回归算法(SVR)的预测精度和泛化能力影响显著，提出混沌粒子群优化算法(CPSO)寻优一种改进支持向量回归算法(v-SVR)参数的新方法，在此基础上建立高炉铁水硅含量预测模型(CPSO -vSVR)用于对某钢铁厂3号高炉铁水硅含量的实际数据进行预测，研究结果表明：基于CPSO确定的最优参数建立的铁水硅含量粒子群支持向量回归预测模型的预测效果最佳，平均相对误差为5.32%。与使用粒子群优化算法训练的神经网络(PSO-NN)、v-SVR、最小二乘支持向量回归(LS-SVR)进行比较，CPSO -vSVR模型对铁水硅含量进行预测时预测绝对误差小于0.03的样本数占总测试样本数的百分比达到90%以上，预测效果明显优于PSO-NN，且比v-SVR稳定性更强，可用于高炉铁水硅含量的实际预测。

Prediction of Silicon Content in Hot Metal Based on SVR Optimized by Chaos Particle Swarm Optimization

The regression accuracy of the support vector regression (SVR) models depends on a proper setting of its parameters to a great extent. An optimal selection approach of v-SVR parameters was put forward based on chaos particle swarm optimization (CPSO) algorithm. Then a model based on v-SVR to predict the silicon content in hot metal was established, and the optimal parameters of the model was searched by CPSO. The data of the model were collected from No.3 BF in Panzhihua Iron and Steel Group Co.. The results showed that the proposed prediction model has better prediction results than neural network trained by particle swarm optimization and v-SVR, the percentage of samples whose absolute prediction error is less than 0.03 when predicting silicon content by v-SVR model is higher than 90%, it indicated that the prediction precision can meet the requirement of practical production.