(1. 北京科技大學(xué) 金屬礦山高效開采與安全教育部重點實驗室，北京 100083；
2. 金屬礦山安全與健康國家重點實驗室，馬鞍山 243000；
3. 北京科技大學(xué) 土木與資源工程學(xué)院，北京 100083；
4. 綠色化工過程教育部重點實驗室，武漢 430205；
5. 首鋼集團(tuán)有限公司礦業(yè)公司，唐山 064400)
摘 要: 膏體充填能夠改變礦區(qū)周圍巖石的應(yīng)力狀態(tài)，減少礦區(qū)周圍的環(huán)境問題，在國內(nèi)外地下礦山得到廣泛應(yīng)用。膏體合理配比的確定需要進(jìn)行大量實驗，亟需尋找一種簡單有效的方法去對膏體的性能進(jìn)行預(yù)測。對此，本文開展了27組配比的強度和坍落度測試，采用PSO、Grid和GA算法優(yōu)化后的支持向量機(SVM)進(jìn)行訓(xùn)練，對比分析預(yù)測結(jié)果對算法進(jìn)行優(yōu)選。基于算法模型進(jìn)行膏體性能預(yù)測，采用極差分析手段揭示了不同影響因素對充填體性能的影響，最后采用線性規(guī)劃原理得到了最佳配比參數(shù)。結(jié)果表明：核函數(shù)的參數(shù)c和懲罰系數(shù)g隨算法不同而變化，不能作為評判預(yù)測精度的依據(jù)。Grid算法和PSO算法可以分別對坍落度和強度進(jìn)行精準(zhǔn)預(yù)測。由于膏體中自由水含量的影響導(dǎo)致坍落度受粗骨料-尾砂比和濃度的影響較大。不同齡期條件下水泥摻量和濃度對充填體強度的影響始終處于主導(dǎo)地位，隨著齡期增加粗骨料“骨架”效應(yīng)逐漸顯現(xiàn)。以金川礦山實際充填需求為例，獲得其最低成本配比條件為水泥摻量為280 kg/m3，粗骨料-尾砂質(zhì)量比為1∶1，廢石-棒磨砂比為1∶1.44。
關(guān)鍵字: 膏體充填；支持向量機；極差分析；線性規(guī)劃；配比優(yōu)化
（1. Key Laboratory of High-Efficient Mining and Safety of Metal Mines, Ministry of Education, University of Science and Technology Beijing, Beijing 100083, China;
2. State Key Laboratory of Safety and Health for Metal Mines, Maanshan 243000, China;
3. School of Civil and Resources Engineering, University of Science and Technology Beijing, Beijing 100083, China;
4. Key Laboratory of Green Chemical Engineering Process, Ministry of Education, China;
5. Shougang Ming Corporation, Tangshan 064400, China）
Abstract:Paste filling can change the stress state of the rocks around the mine and reduce the environmental problems around the mine, which is widely used in underground mines at home and abroad. The determination of paste''''s reasonable ratio requires a lot of experiments, and there is an urgent need to find a simple and effective method to predict the paste performance. Thus, strength and slump tests of 27 sets were conducted in this paper, and the SVMs optimized by PSO, Grid, and GA algorithms were trained to compare the prediction results for algorithm optimization. Then the paste performance prediction was carried out based on the corresponding model, and the influence degree of different factors on the paste performance was revealed using extreme difference analysis. Finally, the optimal proportioning parameters were obtained by using the linear programming principle. The results show that the parameter c and the penalty coefficient g vary with the algorithm and cannot be used as a basis for judging the prediction accuracy. Grid algorithm and PSO algorithm can make accurate predictions of slump and strength respectively. Because of free water content in paste, the slump is mainly influenced by the coarse aggregate-to-tailing ratio and concentration. At the same time, the influence of cement content and concentration on the strength of the filler under different age conditions is always dominant. Moreover, the “skeleton” effect of coarse aggregates gradually appears with the curing age increase. The lowest cost ratio to meet the actual filling requirements of Jinchuan mine is 280 kg/m3 of cement, a coarse aggregate-tailing mass fraction ratio of 1∶1 , and a waste rock-rod mill sand ratio of 1∶1.44.
Key words: paste backfill; support vector machine (SVM); range analysis; linear programming; proportion optimization
