利用啟發式演算法建立可灌性預測公式

Abstract

本研究的目的是要建構一個粉土質砂層超微粒水泥滲透灌漿可灌性的預測公式。因為本研究之粉土質砂層富含較高的細粒料含量以及所使用的超微粒水泥粒徑遠小於傳統的卜特蘭水泥，遂傳統可灌性相對粒徑比經驗公式無法有效的預測。雖然應用倒傳遞與幅狀基底函數類神經網路亦是一個能夠預測可灌性的方法，然因其無法提供一個明確的預測公式，所以於實際工程應用上有其限制。因此，本研究藉由所蒐集之240筆現地資料以啟發式演算法(禁忌演算法)，建構一可灌性之預測公式。除了參考傳統相對粒徑比可灌性經驗公式的格式外，土壤有效粒徑(d10)、土壤粒徑(d15)、細粒料含量(FC)與水灰比(W/C) 4個影響可灌性的因子亦納入考慮，其預測準確率可高達94.58%。再者，由本研究可灌性預測公式之良好預測結果顯示，應用禁忌演算法所建立之可灌性預測公式為相當可行的方法。此外，本研究所建構之預測公式與傳統相對粒徑比可灌性經驗公式有相似的格式，將更方便於工程師之使用也更易於實際工程上的普遍應用。

The goal of this study is to provide an accurate formula to predict the groutability (N) of silty sand soils using microfine cement grouts in a permeation grouting. Because the fines content (FC) of the silty sand soils studied is relatively high, and the size of the grouts used is significantly smaller than the Portland cement, the existing empirical formulas cannot deliver a promising prediction of N. Artificial neural networks such as BPNN or RBFNN are alternative tools used to predict N. However, ANNs do not provide an explicit formula, which creates an obstacle for practical engineers. Thus, a heuristic algorithm (the Tabu search, TS) was used to build the prediction formula. A total of 240 in-situ data samples were analyzed to ensure the accuracy of the proposed formula. The format of the existing empirical formula, which is commonly used in practice, was adopted in the proposed TS-based formula. Four parameters were considered in our TS models: the effective soil particle size (d10), the soil particle size (d15), the water-to-cement ratio (W/C) and the FC. The prediction accuracy of the TS-based formula was approximately 94.58%. With such a promising result, it is evident that the proposed formula is a suitable tool for predicting N. Because the proposed formula has a similar format to that of formulas that are typically used, the proposed approach can be implemented readily in practical engineering settings.