【正文】
isolated stope databases were each trained to 10 percent training error. These neural s were then used to predict dilution on new unseen data and pared with the respective formula dilution estimates. The differences in the actual dilution from the neural and formula predictions were pared for each stope and the bined averages of the stopes for each stope configuration. Figure 1 charts the average percent error between the neural and formula predictions. AVERAGE PERCENT ERROR OF NEURAL NET PREDICTION (SERIES 1) AND FORMULA PREDICTION (SERIES 2) Fig. 1. Average Neural Net / Formula Error Over Actual Average Dilution For the unseen rib stope data, the neural had an average error of percent dilution and the formula had an average error of percent dilution. For the echelon unseen stope data the neural had an average error of percent while the formula had an average error of percent. For the two unseen isolated stopes the neural had an average error of percent while the formula had an average error of percent. As the rib stope and echelon unseen databases were significantly larger than the isolated stope unseen database the neural work showed a clear improvement over the formula estimates. The improved performance of the neural predictions in this example over the statistically derived formulas suggests that neural s can have better predictions than conventional formula predictions. 5. CONCLUSION The Goldcorp/Canmet example shows that neural works can provide an effective tool for predicting rockbursts. Further work varying error, the number of nodes, layers and cycles could improve the work using this database. However, a larger database with more input factors would make a neural work a more effective burst predicting tool. Additional inputs for each failure may include: induced stress (map3d), hydraulic radius, presence of raises, microseismic data, faults or dikes, ground support。 Sons. 4. Barton, Lien, Lunde, 1974, Classification of Rock Masses for the Design of Tunnel Support, Roc Mechanics Vol. 6, No. 4,7 pp. 5. Lang, B., Pakalnis, R., Vongpaisal, S., 1991, Spa n Design in Wide Cut and Fill Stopes at Detour Lake. Mine, 93rd. AGM – CIMM, paper 142, Vancouver. 6. Pakalnis, R, 1986, Empirical Stope Design at the Ruttan Mine, Sherritt Gordon Mines Ltd., University of British Columbia, Canada, 276 pp. 中 文 譯 文 采礦工業(yè)中實用的神經網絡應用程序 米勒 .L泰特, (不列顛哥倫比亞大學 采礦與礦物加工工程學院,加拿大 范庫弗峰) 摘要 :采礦工業(yè)很大程度上依賴以觀察或試驗為依據的分析從而進行設計和預測。 232607668. 引言 礦井設計的許多方面都是基于實際觀察或試驗數(shù)據。 ? 通過使用多重隱藏的層次和節(jié)點,神經網絡可以審查輸入間的聯(lián)合影響。 神經網絡的優(yōu)點在一個巖爆預測和一個露天采場貧化例子中得到闡明。它大約占了同時期井下重大事故死亡人數(shù)的 10%。這個研究的一部分就是收集 1992~1995 年間發(fā)生在 、掘進、地面楔和放頂事故的輸入信息。輸入元素或穩(wěn)定性可能是以下四個失敗之一: PUNRF(頂板垮落的潛在不穩(wěn)定性 ), PUNGW(地面楔的潛在不穩(wěn)定性 ), BUR(巖爆 ), CAV(采掘 )。 這些因素被賦予了一個數(shù)值并且被計算到一起從而得到一個 RMR值。它是由來自 Norwegian地質技術學院的 巴頓、李恩和魯恩德共同提出的,建立于以下六個要素之上: ? RQD— 巖石質量指標; ? Jn— 節(jié)理面數(shù)目; ? Jr— 節(jié)理粗糙程度; ? Ja— 節(jié)理變化數(shù)目; ? Jw— 節(jié)理處水減少因素; ? SRF— 應力減小因素。 Q因素在一個從 1000的范圍內變化,其中 表示品質極差的巖石, 1000表示實質上品質非常理想的巖石。它并不直接代表使用在 Q 計算公式中的那個 SRF。 PUNRF 指考慮到頂板垮落的潛在不穩(wěn)定地域。 ? 地表運動和升高的頻率可能會增加。 232607670. 神經網絡分析 上面的輸入和輸出都運行于神經網絡中,以觀察一個神經網絡能否從輸入數(shù)據中預測出結果,同時也要觀察哪些輸入對輸出預測有最大影響。 神經網絡的結果顯示出網絡可以正確地預測出所有來自于培訓的輸出。沖擊狀況每次均可以成功預測的事實做出了這樣的承諾:就潛力而言,神經網絡有可能是一個預測巖爆的有效工具。一個包含穩(wěn)定開度的更大數(shù)據庫有必要建立,以取得神經網絡在預測方面和輸入因素影響作用的更大信心。 232607671. 神經網絡 /公式貧化預測比較 通過比較神經網絡和 常規(guī)公式,神經網絡預測和三個由數(shù)據庫中發(fā)展出來的公式進行了比較,數(shù)據庫取自魯坦礦。神經網絡在不可見數(shù)據上預測和公式評估進行了比較。對應于每一個采場配置的每個神經網絡都在整個原始數(shù)據庫上進行了訓練。 圖 1表示了神經網絡與公式預測間的平均百分錯誤。由于拱形采場和梯形采場的不可見數(shù)據庫明顯比獨立采場不可見數(shù)據庫大,故神經網絡顯示出較之公式評估的明顯改進。然而一個帶有更多輸入因素的更大數(shù)據庫 可以使用神經網絡