基于长度修正的预测算法优化

Optimization of Prediction Algorithm Based on Length Modified

当自变量之间存在多重相关性时,若利用最小二乘法建立预测模型,参数估计会存在误差。若应用偏最小二乘回归算法建立预测模型,可以克服自变量之间多重相关性问题,计算结果更为可靠。长度修正的偏最小二乘回归算法从预测的角度对偏最小二乘模型进行了改进。以芹山水电站的水平位移预测为例,验证了长度修正的偏最小二乘回归法比普通偏最小二乘回归法在预测方面效果更好。

When variables are multiple correlating, the estimation error of parameters exists if least-squares regression model is used to predict. The multiple correlating problems between the independent variables can be overcome by partial least squares(PLS) with a more reliable result. Length modified partial least squares(LMPLS) can improve the PLS from the prediction aspect. Taking the horizontal displacement prediction of Qinshan Hydropower for example, we verified LMPLS has better prediction results than the ordinary PLS.