不同优化类型组合预测模型在沙漠软土路基的应用

某沙漠重载铁路工程，地势低洼路段雨季易积水，加上原有高地下水位的长期浸泡和植物根系的有机分解积累，形成了淤泥质或泥炭质软土地基；运用理论分析、现场监测等方法，从权重分配合理性和子模型组合结构两方面探讨组合预测模型的精度优化效果，开展沙漠淤泥质软土路基沉降预测研究。结果表明：双曲线法、三点法、指数曲线法、泊松曲线法、BP神经网络5种预测模型均能达到较高水平的拟合程度；变权重组合预测模型、引入鲸鱼优化算法的自适应权重组合预测模型、滚动动态组合预测模型对于预测精度、效果的提升较小；引入BP神经网络的误差补偿组合预测模型，极大程度地降低了人为建立子预测模型组合结构所产生的精度影响，在沙漠淤泥质软土路基中具有更优的预测精度及效果。

Application of Combined Prediction Models of Different Optimization Types in Desert Soft Soil Subgrade

A railway project coupled with the long-term immersion of the original high groundwater level and the organic decomposition and accumulation of plant roots, a silty or peat soft soil foundation is formed. By theoretical analysis, on-site monitoring and other methods, the accuracy optimization effect of the combined prediction model from the two aspects of the rationality of weight distribution and the sub-model combined structure are discussed, and the settlement predictive research of desert silt soft soil subgrade is carried out. The results show that: the hyperbolic method, the three-point method, the exponential curve method, the Poisson curve method, and the BP neural network five prediction models can achieve a high level of fitting. The variable weight combined prediction model, adaptive weight combination prediction model based on the whale optimization algorithm and the rolling dynamic combination prediction model have little improvement in the prediction accuracy and effect; the error compensation combination prediction model of the BP neural network is introduced, which greatly reduces the accuracy caused by the artificial establishment of the sub-prediction model combination structure. It has better prediction accuracy and effect in desert muddy soft soil subgrade.