【作者】 张志军；

【导师】 丁德馨；

【作者基本信息】 中南大学 ， 安全技术及工程， 2008， 博士

【摘要】 应用自适应神经模糊推理系统(ANFIS)研究采矿工程中巷道和采场失稳、露天矿边坡滑坡、开采地面沉陷以及导水裂隙扩展淹井这四大岩土工程灾害时所遇到的最大难题是,ANFIS的训练参数和模型结构的确定具有人为性、随意性和无规则性,这使得它的收敛需要经过长时间的反复调试才可能实现,而且难以收敛到所要求的精度。针对这些问题,本文首次采用正交试验设计法和模式搜索算法研究了ANFIS的训练参数和模型结构的优化方法,提出和建立了基于模式搜索算法的自适应神经模糊推理方法(PSA-ANFIS),并进一步采用这一方法研究了采矿工程中上述四大岩土工程灾害的预测问题,取得了如下研究成果:(1)采用一多峰函数进行离散,构建了训练数据对、检测数据对和预测数据对,首次采用正交试验设计法、模式搜索算法和自适应神经模糊推理系统,建立了PSA-ANFIS方法。(2)采用同一多峰函数,对ANFIS和PSA-ANFIS进行了比较研究,结果表明,无论是在拟合精度和预测精度上,还是在训练参数的调整、模型结构的建立和训练过程上,PSA-ANFIS均优于ANFIS。(3)采用一标准弹塑性问题进行解析计算,构建了训练数据对、检测数据对和预测数据对,采用正交试验设计法、模式搜索算法和自适应神经模糊推理系统,建立了围岩力学参数位移反分析的PSA-ANFIS方法,并将其应用于湘西金矿老采区围岩力学参数的反演,进而采用反演结果建立数值计算模型,对采场围岩的变形进行了模拟计算,与此同时,还对采场围岩的变形进行了监测,结果表明,模拟计算结果与监测结果吻合较好,从而验证了这种围岩力学参数位移反分析新方法的工程适用性。(4)搜集了大量圆弧滑动边坡工程实例,构建了训练数据对、检测数据对和预测数据对,采用正交试验设计法、模式搜索算法和自适应神经模糊推理系统,建立了圆弧滑动边坡稳定性预测的PSA-ANFIS方法,并将这一方法应用于前河露天矿1号滑坡的稳定性预测,预测结果与边坡的实际稳定状态吻合良好。(5)搜集了大量开采地面沉陷工程实例,构建了训练数据对、检测数据对和预测数据对,采用正交试验设计法、模式搜索算法和自适应神经模糊推理系统,建立了开采地面沉陷预测的PSA-ANFIS方法,并采用这一方法,对新汶矿业集团翟镇煤矿一采区西翼二层煤采区地面沉陷进行了预测,预测结果与实测结果吻合良好。(6)搜集了大量导水裂隙带发育高度工程实例,构建了训练数据对、检测数据对和预测数据对,采用正交试验设计法、模式搜索算法和自适应神经模糊推理系统,建立了导水裂隙带发育高度预测的PSA-ANFIS方法,并应用这一方法对水口山矿务局康家湾铅锌金矿水体下Ⅲ-1采场的导水裂隙带发育高度进行了预测,同时还对其进行了实测,结果表明,预测结果与实测结果吻合较好。进而根据预测结果,为Ⅲ-1采场设计了上向水平分层胶结充填采矿法,且已实现了采场的安全回采,取得了巨大的经济效益和社会效益。研究ANFIS的训练参数和模型结构的优化方法,是人工智能领域的一个前沿研究课题。本文率先采用正交试验设计法和模式搜索算法实现了对ANFIS的训练参数和模型结构的优化,建立了基于模式搜索算法的自适应神经模糊推理方法,丰富和发展了ANFIS的理论和方法。不仅如此,本文还将这一方法应用于采矿工程中四大灾害预测问题的研究,丰富和发展了它们的研究方法和预测理论。更多还原

【Abstract】 The most difficult problem to analyse and precdict the four geotechnical engineering disasters in mining engineering by using the adaptive neuro-fuzzy inference system (ANFIS), the failure of tunnuls and stopes, the slope failure of open-pit mine, the mining induced surface subsidence and the underground flooding by extension of transmissive fractured belt, is that the setting of the training parameters and model of ANFIS has many disadvantages of artificiality, randomness and unregularity. Therefore, the convergence process of training ANFIS is lengthy and trivial and repetitive. For this reason, this thesis studied the approach for setting the training parameters and model of ANFIS by using the orthogonal test design method and the pattern search algorithm(PSA), and put forward and established the approach for adaptive neuro-fuzzy inference system based on pattern search algorithm (PSA-ANFIS). Furthermore, PSA-ANFIS was used to establish the analysis and precdiction approaches for the four geotechnical engineering disasters in mining engineering. And the following achievements have been abtained.(1) The data pairs for training, data pairs for checking and data pairs for prediction were built by using a multi-peaks function. The PSA-ANFIS was established by using the orthogonal test design method and PSA and ANFIS.(2) A comparison study was conducted of the training process and the fitting and predicting capability of the PSA-ANFIS and the ANFIS by using the same multi-Peaks function. The results show that the PSA-ANFIS is far superior to the ANFIS in its parameters adjustment and model establishment and training process, and that is an excellent approach.(3) The data pairs for training, data pairs for checking and data pairs for prediction were built by analytical calculation of a prescribed elasto-plastic problem. A PSA-ANFIS based approach for back analysis of displacements for the mechanical parameters of rock mass was established by using the orthogonal test design method and PSA and ANFIS. The approach has been used to inverse the mechanical parameters of the rock mass surrounding the mining stopes in Xiangxi Gold Mine in Hunan Province. The inversion results were used to build the numerical calculation model to simulate and calculate the deformation of the rock mass and assess the stability of the stopes. At the same time, in-situ measurements were carried out for the deformation of the rock mass. It has been proved that the in-situ measurements are in good agreement with the simulation results and that the new approach is suitable for back analysis of displacements.(4) Based on quite a few cases of circular failure slopes, the data pairs for training, data pairs for checking and data pairs for prediction were built, and the PSA-ANFIS based approach for predicting the stability of circular failure slopes was established by using the orthogonal test design method and PSA and ANFIS. And the approach was used to predict the stability of No.1 circular failure slope in Qianhe open-pit mine. The prediction result is in good agreement with the practical state.(5) Based on quite a few cases of mining induced surface subsidence, the data pairs for training, data pairs for checking and data pairs for prediction were built, and the PSA-ANFIS based approach for mining induced surface subsidence was established by using the orthogonal test design method and PSA and ANFIS. And the approach was used to predict the mining induce subsidence in the west side of No.1 mining area in coal layer.2 in Zhaizhen coal mine of Xiwen Mining Group. The prediction results are in good agreement with the practical measurement results.(6) Based on quite a few cases of the height of the transmissive fractured belt, the data pairs for training, data pairs for checking and data pairs for prediction were built, and the PSA-ANFIS based approach for the height of the transmissive fractured belt was established by using the orthogonal test design method and PSA and ANFIS. And the predicting approach was used to predict the height of the transmissive fractured belt for No.Ⅲ-1 mining panel at Kangjiawan lead, zinc and gold mine in the Bureau of Mines of Shuikoushan. At the same time, the in-situ measurement is done, and the research result shows that the predicting result is in good agreement with the measurement result in-situ. Finally, in order to mine No.Ⅲ-1 ore body safely, a mining system of upward slicing and filling with cemented tailings was designed. The system was used to mine the ore body safely and great economic benefits were obtained.It is a new research subject in the domain of artificial intelligence to study the optimization method on setting the training parameters and model of ANFIS. This thesis accomplished this research by using the orthogonal test design method and PSA, and established the PSA-ANFIS. It is the enrichment and development of the ANFIS theory. Furthermore, the PSA-ANFIS was used to study the application of the prediction problems on the four geotechnical engineering disasters in mining engineering. This enriches and develops the research method and prediction theory of the four prediction problems.更多还原