结合GM(1,1)和LSSVM的多效蒸发过程参数预测

摘要

A parameter prediction method was proposed for solving the time-series prediction problem on the parameters of the multi-effect evaporation process with high noise and non-stationary, combining GM (1,1) and least squares support vector machines (LSSVM) based on the wavelet transform model. Firstly, the Mallat algorithm was used to decompose and reconstruct the time series of parameters, in order to separate low frequency and high-frequency sequence. Next, the GM (1,1) model was designed by using a low frequency and high-frequency information sequence based on the LSSVM. Finally, a result of the prediction on all models was analyzed to determine the final prediction results. Production data of a multi-effect evaporation process in alumina production were tested in the experiment; and the results show the prediction algorithm is feasible and superior to a single GM (1, 1). The test demonstrated the LSSVM method had a good generalization performance and powerful robustness; and could be used for operation of an optimal e-vaporation process in the alumina production.%为了解决多效蒸发过程具有高噪声和非平稳等特性的参数时间序列预测问题,提出了一种基于小波变换结合GM(1,1)和LSSVM的蒸发过程参数预测方法.该方法首先利用Mallat算法对参数时间序列进行分解和重构,分离出序列中的低频信息和高频信息；然后对低频信息构建GM(1,1)模型,对高频信息则用最小二乘支持向量机进行拟合；最后将各模型的预测结果进行叠加,从而得到最终的预测结果.以氧化铝多效蒸发过程的生产数据进行了实验验证,结果表明,该预测算法切实可行且优于单一的GM(1,1)和LSSVM方法,具有较好的泛化性能和较强的鲁棒性,可用于氧化铝生产蒸发过程的优化控制.