基于模拟退火-粒子群算法优化支持向量机参数的连铸漏钢预报

Breakout Prediction for Continuous Casting Based on SA-PSO to Optimize Parameters of SVM

针对连铸漏钢预报神经网络模型在小样本训练数据情况下难以获得较高预报准确率的问题,提出了一种基于模拟退火-粒子群(SA-PSO)算法优化支持向量机(SVM)参数的连铸漏钢预报算法。将粒子群优化算法引入支持向量机的训练过程中,利用其调整参数少、寻优速度快的优点,有效地提高了漏钢预报模型的寻优速度;利用模拟退火算法对粒子群算法迭代更新后粒子的新位置加以评价,来决定新位置是否被接受,避免了粒子群算法在迭代寻优过程中陷入局部极值的问题。结合某钢厂连铸现场历史数据对提出的连铸漏钢预报算法进行了测试,测试结果表明,所提算法的连铸漏钢预报准确率可达98.8%。

In order to overcome the problems that the neural network model was difficult to obtain a high accurate breakout prediction for continuous casting under the conditions of small sample training data, a breakout prediction for continuous casting was proposed based on SA-PSO algorithm to optimize the parameters of SVM. Firstly, the PSO algorithm was introduced into the training proces- ses of SVM, increasing the optimization speeds of breakout prediction model by using the advantages of less parameters and fast optimization speeds. Secondly, SA algorithm was used to evaluate the new positions of updated particles, and to determine whether the new positions were accepted, which could avoid the PSO algorithm steped into the local extremum in optimization processes. Finally, the break- out prediction for continuous casting was tested by the history data of continuous casting. The results show that the proposed algorithm may make the breakout prediction accuracy reach 98.8%.