基于KPCA-SVM的预测模型在铀矿堆浸中的应用

The Application of Heap Leaching of Uranium Based on KPCA-SVM

针对影响堆浸工艺铀矿浸出率的因素较多且具有非线性的特点,提出一种利用核主成分分析(KPCA)进行参数处理,整合冗余,降低维数,并将处理后得到的6个主成分作为支持向量机(SVM)测量模型输入的预测方法。在此过程中,利用粒子群算法(PSO)优化核主成分分析和支持向量机的参数,使模型具有较高的训练精度。在此基础上,对铀矿堆浸进行建模仿真,并进行预测。结果表明,基于KPCA-SVM的铀矿累计浸出率模型与BP神经网络方法相比,具有有效降低数据维数、在小样本条件下学习更加有效、建模采样过程更快、预测精度更高的优点。

Aiming at the features of factors which is non-linear and influence heap leaching of uranium leaching rate,then we proposed a measure that using the kernel principal component analysis（ KPCA） to process parameters. It can reduce redundancy and lower dimensions,then we can get 6principal components as a predictive method to measure inputs of model by support vector machines（ SVM）. In this process,the particle swarm optimization（ PSO） is used to optimize the parameters of kernel principal component analysis and support vector machines,so that the model has higher training accuracy. Based on that,we built a model of the heap leaching uranium to simulate and forecast.The results show that,compared KPCA-SVM cumulative uranium leaching rate model with BP neural network method,the former can effectively reduce data dimension,learn more effectively in small samples,model the sampling process faster and has a higher predictive accuracy.