论文部分内容阅读
在杭长客专浙江段沉降评估实践中发现,用常规的4种拟合算法得出的相关系数普遍偏低,绝大部分测点小于0.92。分析了相关系数普遍偏低的原因,并引入了3种新的评估指标:平均绝对误差MAD、均方误差MSE和平均绝对百分误差MAPE。通过与相关系数的对比分析表明,即使在相关系数很低的情况下,沉降预测模型也可以获得足够高的精确度。因此,采用MAD、MSE和MAPE的组合模式来代替相关系数,从而建立针对沉降“小量级、大波动”特点的沉降预测精确度评价建议模式。对于沉降预测结果同时满足MAD≤0.11mm、MSE≤0.015 mm2和MAPE≤8%的观测点,可以认为其沉降预测精确度满足要求。研究表明,此建议模式在杭长客专浙江段具有很好的适用性,并显著提高了满足沉降预测精确度要求的测点百分比。
In Hangzhang passenger depot Zhejiang subsidence assessment practice found that the conventional four kinds of fitting algorithm to draw the correlation coefficient is generally low, most of the measuring point is less than 0.92. The reasons for the generally low correlation coefficient were analyzed and three new evaluation indexes were introduced: average absolute error MAD, mean square error MSE and mean absolute percentage error MAPE. The comparative analysis with the correlation coefficient shows that the settlement prediction model can obtain high enough precision even in the low correlation coefficient. Therefore, the combined model of MAD, MSE and MAPE is adopted instead of the correlation coefficient so as to set up a suggestion model for accuracy of settlement prediction for the characteristics of “small magnitude and large fluctuation” of settlement. For the observation points whose settlement prediction results satisfy both MAD≤0.11mm, MSE≤0.015 mm2 and MAPE≤8%, the accuracy of settlement prediction can be considered as satisfying the requirements. The research shows that this proposed model has good applicability in Zhejiang section of Hangzhou-Changchuan Project and significantly increases the percentage of measuring points that meet the accuracy requirements of settlement prediction.