Subgrade Settlement Based On Measured Data, Prediction Methods Of Research And Engineering Applications

Excessive settlement or uneven settlement of the highways and high-speed railway embankment will deteriorate the road condition, reduce the comfort of the passengers, and even endanger traffic safety. Therefore, the settlement after construction will become more stringent with the improvement of the operating speed. Accurate and timely forecast the settlement will provide certain guidance for reasonably planning construction schedule, arranging paving time, scientifically planning the height in the staged construction, the time of the intermittent and the pre-pressure time. Based on the exploration survey material and the measured settlement data of the Foshan-Kaiping highway, The grey GM (1,1) model and the BP neural network to forecast settlement were used when the loads had been stable. What's more, the deformation mechanism of the settlement was analyzed under the step loading and the curve fitting -genetic algorithm united modeling to predict by skipping a grade the settlement was used. The issues conducted in this paper were generalized as follows:(1) Based on analysis of the research status of the settlement prediction under the stable load and the step loading, with the characteristics of settlement, the theoretical basis, conditions of application, and the advantages and disadvantages of the current commonly settlement prediction methods were analyzed and compared. These analyses provide theoretical basis for choosing settlement prediction method reasonably.(2) The settlement prediction model which based on the artificial neural network was established. with the FORTRAN languages, the relevant calculating program was developed and the measured settlement data were used to verify the procedure. Through comparative analysis for the measured data and prediction results, the impact of the number of the input vector, the time interval of the measured data and the forecast period for the forecasting accuracy were discussed.(3) Based on the characteristics which the grey GM (1,1) model requires the measured data to be an equal time interval data sequence, the straight line interpolation, the cubic spline interpolation and BP neural networks was adopted respectively to change the unequal time interval data sequence into an equal time interval data sequence and applied the model to the settlement prediction of the practical engineering. The results showed that: the cubic spline interpolation and BP neural networks are more reasonable in the unequal time interval conversion, the method of the BP neural network is the most precise.(4) The BP neural network - grey system united model was adopted to predict the settlement when the loads have been stable, and the impact of the modeling periods for the model has been discussed. The applicability of the model was verified through the project examples.(5) The curve fitting - genetic algorithm united mode was used to predict the settlement under the step loading. The method can used to predict by skipping a grade the settlement, and provide an effective way to predict the total settlement after construction according to the measured data at the period of the filling and the prepressing. The impact of the incremental value of load△p_k and the radiio of the incremental value of load△p_k to the former incremental value of load△P_k-1, for the accuracy of model were discussed, and the reasons for the error and model limitation were analyzed.(6) Through Project examples the impact of the number of the measured data at all levels of load which were used for modeling was discussed when we used the curve fitting - genetic algorithm united mode to predict the settlement.