三间房矿3.0Mta新井设计【含CAD图纸+文档】
三间房矿3.0Mta新井设计【含CAD图纸+文档】,含CAD图纸+文档,三间,mta,设计,cad,图纸,文档
附录A应用具有遗传算法和模糊选择模型的神经网络为全机械化采煤选择设备王新宇,吴瑞明,冯春华摘要:根据典型的工程样品,用遗传算法来优化权值的神经网络工作模式提出了预测回采工作面的生产能力和效率。通过这个模型,我们可以获得一定地质情况下综采工作面设备组合可能的结果。模糊选择理论被用于评估每组设备组合的各项效率。通过详细的实证分析,该模型结合预测回采工作面的数据和选择最佳的设备组合的功能,并且有利于完全机械化采煤设备组合的决策。关键词:遗传算法;人工神经网络;模糊选择;选择设备组合1前言在煤矿生产过程中,为采煤工作面选择适当的设备组合是最重要的决策任务之一。一般的解决方案,任务是通过专业知识和工程经验完成的。设备与采煤工作面的地质条件的匹配的重要性是为了使设备运行良好并且实现高输出性能。在本文中,我们应用人工神经网络模型,根据地质条件和回采设备,利用在典型的工程案例中学到的经验性的知识来预测回采工作面的输出和效率。模糊优选理论被用于评估每个组合设备在某些综采工作面的地质条件下的性能。使用模糊决策模型可以得到最令人满意的设备组合。在人工神经网络模型中,通过遗传算法来完成网络节点之间最优权的搜索,这种算法可以通过反向传播算法避免网络目标函数局部优化。通过详细的实证分析,该模型集成预测回采工作面的成绩和选择最佳的设备组合的功能,并有利于综采设备组合的的决策。2选择合适的设备组合系统模型2.1用人工神经网络和遗传算法预测输出和采煤工作面的效率遗传算法模拟生物进化的自然选择的原则,它已被用于解决各种工程和科学的优化问题来找到真正理想的最佳点。通过遗传算法随机运营,遗传算法控制对最佳点的搜索过程。大家都知道,反向传播算法是梯度递减算法,它不能得到节点之间的最佳权重。因此,我们应用遗传算法训练网络,而不是被称为GA-BP算法的BP算法。它可以提高网络收敛速度的非线性映射能力。在GA-BP算法中,用算法数字的字符串代替二进制数字的字符串来直接表达节点之间的权重。由此,在二进制编码的字符串中实施相应的编码和解码操作是可以避免的,因此收敛速度更高。另一个好处是,权重的计算精度要好得多。详细步骤如下:1)随机产生N组网络的权重;2)采取BP算法训练网络与上述初始N组权重。如果至少有一组权重达到目标函数规定的训练精度,算法结束;否则采取下一步骤;3)寻找权重属于上述训练有素的N组权重信息的可能的数字领域。我们在这个数值区域设新的RN组随机权重。所有组权重由整个基因集落组成;4)应用遗传操作,即选择,交叉和变异(R+1)N组权重;5)如果有至少一组权重在第4步中满足精度要求,算法结束;否则我们从(R1)N组权重中选择最好的权重,然后返回到步骤2。根据典型工程案例,人工神经网络的主要任务是学习地质条件和设备与产出和效率的映射关系的经验知识。然后,它被用来预测在一定地质条件下可用的设备组合的未来业绩。2.2根据模糊选择决策模型评估设备的性能可用的设备组合及其未来业绩,如通过人工神经网络构建一个多目标的决策系统来获得输出和效率。我们运用模糊优选理论来评估每个可用的设备组合。评价模型会帮我们找到最好的设备组合。假设有m个目标和n可用的项目,我们使用矩阵X=(Xi)MN表达基础数据集,其中Xij表示第i个目标第j个项目的值。对于输出和效率,我们采取下列公式得到相对隶属度: (i=1,2,m), (1)其中rij为第i个目标第j个项目的相对隶属度。由方程(1)我们得到的相对隶属度矩阵R=(rij)mn。最好的项目的相对隶属度是:g=(g1,g2,gm)T=(1,1,1)T,最差的项目的相对隶属度为b=(b1,b2,bm)T=(0,0,0)T,所有决策目标的权重向量w=(w1,w2,wm)T,。第j个项目可表示为:Rj=(r1j,r2j,rmj)T。对于第j个项目的最好的项目的一般距离为。根据模糊集合论,如果设第j个项目中最好的项目的相对隶属度为j,第j个项目中最差的项目的相对隶属度必须是。第j个项目中最好的项目的加权距离定义为;第j个项目的最差的项目的加权距离是。我们采取决策规则得到j:最小值,则有是第j个项目的最终评估值。j越大,第j个项目越好。3实证分析我们在山东省收集了一些采煤工作面的典型工程案例。考虑到综采的不同方法,我们将所有样品分成四组训练样本。3.1用人工神经网络预测输出和效率3.1.1输入和输出变量地质因素对选择合适的设备很重要,也对设备性能有影响。相应的主要地质因素是神经网络的输入对象包括:煤层的高度X1(米),煤层倾角X2(度),煤炭硬度系数X3,综采工作面的长度X4(米),老岩石峰的水平X5,直岩石峰的水平X6,气体的水平X7(立方米每吨每天),采煤工作面机械的类型X8,支架的类型X9。我们只考虑采煤工作面机械和支架两种煤机。神经网络的输出对象是输出Y1(t)和效率Y2(tI-1)。3.1.2神经网络的输入和输出对象的转型神经网络中的节点的核心功能往往是乙状结肠功能。当节点的输入接近0或1时,节点的输出改变速度是非常缓慢的。为了避免这种情况,我们通过线性变换把所有的输入和输出数据转换到0.5,0.95内。1)连续输入和输出对象输入变量X1,X2,X3,X4,X7,XJQ和输出变量Y1,Y2是连续的。以煤层的高X1为例,线性变换公式是:. (2)为了收回实际输出,我们采取另一种线性变换:. (3)2)离散输入和输出对象输入变量X5,X6,X8,X9是离散的。例如,变量X5包括四个层次:第I,II,III和IV。我们用0.05,0.35,0.65和0.95标志这些层次作为网络的输入。变量X8,X9代表设备。训练样本中有7种采煤工作面机械:MLS3-170,MXA-300,MG-300,AM-500,MG-150,MD-150,MXA-600用0.05,0.20,0.35,0.50,0.65,0.80,0.95编码。有10种支架:BY360-25/20,BC-480,QY320-13/32,QY200-14/31,ZY35,QY240-26/10,BY240-16/35,ZY28,ZY560K/ L.12,ZYQ1700,同样可以编码。3.1.3神经网络的预测结果建立一个有9个输入节点,30个隐藏节点和2个输出节点的神经网络。我们选择18个样本使用GA-BP算法训练网络,并使用其他的7个样品测试训练有素的网络的可靠性。预测结果如表1。所有样品的所有相对百分比误差在士10之内。这表明,该模型是有效的。3.1.4从网络学到的知识的存储我们用C语言开发GA-BP算法的计算机软件。当训练过程结束时,我们使用文本文件来存储网络的信息包括:网络结构,网络中任意两个节点之间的所有权重。当推理任务需要时,我们可以使用存储在文本文件中的网络信息重建的计算机网络,然后新的输入信息通过神经网络进行处理,以获得预测的输出结果。神经网络中的权重如表2。3.2设备组合的评价3.2.1原始地质条件有一个采煤工作面,它的地质条件是:煤层高2.5米,煤层倾角5度,煤炭硬度系数1.8,挖掘长度面临180米,老岩石峰的水平1,直岩石峰的水平2,气体水平6立方米/(吨天)。我们曾提出基于模糊信息分配理论的模型来选择采煤方法。通过这种模式,适当的地质条件下的煤炭开采方法是普通综采。3.2.2 可用的设备组合有两种采煤工作面机械:MXA-600,MLS3-170;三种支架:BY240-16/35,ZY35,QY200-14/31。所以,有六个可用的设备组合。3.2.3不同的设备组合的评价结果不同的设备组合评价结果如表3。从表3可以看到,第2个项目得分最高。所以,它是我们的最佳选择。这表明MXA-600和QY240-14/31是最好的设备组合。表1 神经网络的预测值编号X1X2X3X4X5X6X7X8X9Y1预测值Y1/t误差/%Y2预测值Y2/t误差/%12.6101.6180387188.94093.8765.5546.845.0-3.7522.282.5172277270.55564.840-8.1024.626.68.1032.481.8130183378.56977.830-0.9429.330.14.7042.0122.41301132448.10050.6015.2015.515.4-0.5553.162.2156274563.22366.4475.1028.426.6-6.3062.6101.5144151675.61776.1460.7030.229.6-2.0572.781.4138281476.17578.0032.4032.830.8-6.1583.051.81651103773.01974.9912.7029.228.7-1.7092.8111.5142161675.58475.463-0.1630.229.2-3.20103.272.0152242686.21683.543-3.1037.339.35.30112.252.21803111853.07952.798-0.5316.817.01.40122.6131.6165152786.40784.592-2.1037.438.52.90132.832.0149287586.17089.6514.0447.145.5-3.50142.0102.31502104946.34947.5082.5016.765.9-4.50152.892.01652125558.64257.938-1.2019.618.9-3.70162.0102.018021221054.30653.980-0.6017.417.92.70172.6141.9145366846.41150.6349.1016.717.44.70182.491.6180287494.30295.2921.0546.750.07.10192.5121.8170162178.02076.148-2.4025.326.54.90202.281.8148181468.74969.2650.7523.224.87.10212.4112.0190162690.54891.9061.5051.248.0-6.20222.281.8160285369.30267.362-2.8023.123.72.80232.552.516821161054.89854.678-0.4018.217.9-1.90241.8131.8154194169.07065.271-5.5022.823.00.90253.272.3158274562.34261.158-1.9027.128.44.80表2 神经网络的权值表3不同的设备组合的评价结果编号.X8X9Y1Y2评价结果1MXA-600ZY-3567.91328.40.19482MXA-600QY240-14/3182.44539.20.99773MXA-600BY240-16/3583.97136.40.88344MLS3-170ZY-3560.71826.10.02665MLS3-170QY240-14-3175.48230.30.59516MLS3-170BY240-16/3565.32922.60.02344结论本文提出了用遗传算法和模糊优选模型来选择基于神经网络的综采工作面设备的一个框架,这个框架可以整合工程知识,为煤炭开采提供智能的决策支持。这个框架可以整合工程知识,为煤炭开采提供的智能决策支持。在今后的工作中,为了使整个决策支持系统更可靠和有效的,我们可以接受更多典型的工程案例,并扩大输入和输出对象的数量,如经济因素。Applying Neural Network with Genetic Algorithm and Fuzzy Selection Models to Select Equipments for Fully-Mechanized Coal MiningWANG Xin-yu, WU Rui-ming, FENG Chun-huaAbstract: According to the typical engineering samples. A neural net work model with genetic algorithm to optimize weight values is put forward to forecast the productivities and efficiencies of mining faces. By this model we can obtain the possible achievements of available equipment combinations under certain geological situations of fully mechanized coal mining faces. Then theory of fuzzy selection is applied to evaluate the performance of each equipment combination. By detailed empirical analysis,this model integrates the functions of forecasting mining faces achievements and selecting optimal equipment combination and is helpful to the decision of equipment combination for fully-mechanized coal mining.Key words: genetic algorithm; artificial neural network; fuzzy selection; selection of equipment combination.1 PrefaceSelecting propr equeipment combinations for coal mining face is one of the most important decision-making tasks during the process of coal mine production. Ordinary solution to that task is completed by expert knowledge, engineering experience. The matching of equipments and the geological conditions in coal mining face is of importance in order to make the equipments work well and achieve high performance of output. In this paper,we apply artificial neural network model to predict the outputs and efficiency of mining faces according the geological conditions and mining equipments with the learned empirical knowledge from typical engineering cases. The theory of fuzzy selection is applied to evaluate the performance of each equipment combination under certain geological conditions of mining face. Using the fuzzy decision model the most satisfied equipment combinations can be obtained. In the artificial neural network model the search for optimal weights between nodes of networks is finished by genetic algorithm which can avoid the objective function of networks into the local optimization by back- propagation algorithm. By detailed empirical analysis, this model integrates the functions of forecasting mining faces achievements and selecting optimal equipment combination and is helpful to the decision of equipment combination for fully mechanized coal mining.2 Systematic Model of Selecting Proper Equipment Combination2.1 Predicting output and efficiency of mining face with artificial neural network and genetic algorithm. Genetic algorithm mimics the natural selection principle in evolution of biology. It has been used to solve all kinds of optimal problems of engineering and science to find the really and ideally optimal point. Genetic algorithm controls the search process towards the optimal points by genetic arithmetic operators randoml. As we all known,the back-propagation algorithm can not get the best weights between nodes for it is a gradient-decreasing algorithm. So we apply the genetic algorithm to train the networks instead of BP algorithm, which is called GA-BP algorithm. It can enhance the nonlinear mapping capability of networks and convergence speed.In GA-BP algorithm,the algorism string of number takes place of binary string of number to express the weights between nodes directly. By that the corresponding coding and decoding operations in binary coding string is avoided, so the convergence speed is higher. Another benefit is that the calculation precision of weights can be much better.The detailed steps are as following:1) Randomly producing N groups of weights of networks;2) Take BP algorithm to train the networks with above initial N groups of weights. If there is at least one group of weights that makes the prescribed training precision of objective function, the algorithm is over; or take the next steps;3) Finding the possible numerical areas which the weights are belong to with the above trained N groups of weights information. In this numerical area we crate other new rN groups of weights randomly. All the groups of weights make up of the whole gene colony.4) Apply genetic operations namely selection, crossover and mutation to (r+1)N groups of weights;5) If there is at least one groups of weights satisfies the precision demand in step 4, the algorithm is over; or we select the best N groups of weights from (r+1)N groups of weights and return to step 2.The main task of artificial neural network is to learn the empirical knowledge on the mapping relations from geological conditions and equipments to outputs and efficiency,according to the typical engineering cases. Then it is used to predict the available equipment combinations future performance in certain geological condition.2.2 Evaluating equipments performance based on fuzzy selection decision-making modelAvailable equipment combinations and their future performance such as output and efficiency obtained by artificial neural networks construct a multi-objective decision-making system. We apply fuzzy selection theory to evaluate each available equipment combination. The evaluation model can help us to find which equipment combination is best. Suppose that there are m objectives and n available projects, we use matrix X=(Xi)m*n to express the basis data set,in which Xij means the value of i-th objective of j-th project. For output and efficiency,we take the following formula to get the relatively subjection degree: (i=1,2,m), (1)where rij is relative subjection degree for i-th objective of j-th project.By equation (1) we get the relative subjection degree matrix R=(rij)m*n. The relative subjection degree of the best project is g=(g1,g2,,gm)T=(1,1,1)T,and the relative subjection degree of the worst project is b=(b1,b2,,bm)T=(0,0,0)T. The weight vector of all decision objectives is w=(w1,w2,wm)T,. The j-th project can be expressed as rj=(r1j,r2j,,rmj)T. The general distance to the best project for j-th project is .If we set the relative subjection degree to best project of j-th project as j according the fuzzy set theory,the relative subjection degree to the worst project of j-th project must be .The weighted general distance to the best project of j-th project is defined as; The weighted general distance to the worst project of j-th project is. We take the following decision rule to get j: min,then we have is the final evaluation values of j-th project. The bigger j,the better j-th project.3 Empirical analysis We collected some typical engineering cases of coal mining faces in Shandong province. In regard to the different methods of fully mechanized coal mining, we divided all samples into four groups of training samples.3.1 Predicting output and efficiency with artificial neural networksGeological factors are of importance to select proper equipments, and also have an effect on the performance of equipments. The main corresponding geological factors are input targets of neural network including: height of coal bed X1 (m), obliquity of coal bed X2(), rigidity coefficient of coal X3,length of mining face X4 (m),levels of old rock peak X5 levels of straight rock peak X6,levels of gas X7(m3t-1d-1),types of coal face machinery X8,types of bracket X9. We only consider two kinds of coal machines, which are coal face machinery and bracket. The output targets of neural networks are output Y1(t) and efficiency Y2(tI-1).3.1.2 Transformation of input and output targets in neural networksThe core function of nodes in neural networks is often sigmoid function. When the input of nodes is near to 0 or 1, the changing speed of output of nodes is very slow. In order to avoid this, we convert all input and output data into areas 0.5,0.95 by linear transformations. 1) Continuous input and output targetsThe input variables X1 X2 X3 X4 ,X7 XJQ and the output variables Y1,Y2 are continuous. Take height of coal bed X1 as an example,the linear transformation formula is . (2)In order to recover the factual output, we take another linear transformation:. (3)2) Discrete input and output targetsThe input variables X5 X6 X8 X9 are discrete. For example,variables X5 includes four levels: I, II, III, and IV. We code these levels with 0.05, 0.35, 0.65, and 0.95 as inputs of networks. Variables X8 X9 denote equipments. There are seven types of coal face machineries in training samples:MLS3-170, MXA-300, MG-300, AM-500, MG-150, MD-150, MXA-600, which are coded with 0.05, 0.20, 0.35, 0.50, 0.65, 0.80, 0.95. There are ten types of brackets: BY360-25/20BC-480, QY320-13/32, QY200-14/31, ZY35, QY240-26/10, BY240-16/35 ZY28, ZY560K/l.12, ZYQ1700, which can be coded similarly. 3.1.3 Predicting results of neural networksA neural network with 9 input nodes, 30 hidden nodes and 2 output nodes are built. We choose 18 samples to train the network by GA-BP algorithm, and use other 7 samples to test the reliability of trained networks. The predicting result is as Table 1. All relative percentage errors for all samples are within 10%. It indicates the model is effective.3.1.4 Storage of knowledge learned from networks We develop the computer software of GA-BP algorithm with C language. When the training process is over, we use a text file to store the networks information including: networks structure, all weights between any two nodes in networks. When a reasoning task is needed, we can make use of the stored network information in that text file to reconstruct the network in computer, and then the new input information is processed by neural networks to obtain the predicting output results. The weights in neural network are as Table 2.Table 1 Forecasting values of neural networkNo.X1X2X3X4X5X6X7X8X9Y1PredictedY1/tError/%Y2PredictedY2/tError/%12.6101.6180387188.94093.8765.5546.845.0-3.7522.282.5172277270.55564.840-8.1024.626.68.1032.481.8130183378.56977.830-0.9429.330.14.7042.0122.41301132448.10050.6015.2015.515.4-0.5553.162.2156274563.22366.4475.1028.426.6-6.3062.6101.5144151675.61776.1460.7030.229.6-2.0572.781.4138281476.17578.0032.4032.830.8-6.1583.051.81651103773.01974.9912.7029.228.7-1.7092.8111.5142161675.58475.463-0.1630.229.2-3.20103.272.0152242686.21683.543-3.1037.339.35.30112.252.21803111853.07952.798-0.5316.817.01.40122.6131.6165152786.40784.592-2.1037.438.52.90132.832.0149287586.17089.6514.0447.145.5-3.50142.0102.31502104946.34947.5082.5016.765.9-4.50152.892.01652125558.64257.938-1.2019.618.9-3.70162.0102.018021221054.30653.980-0.6017.417.92.70172.6141.9145366846.41150.6349.1016.717.44.70182.491.6180287494.30295.2921.0546.750.07.10192.5121.8170162178.02076.148-2.4025.326.54.90202.281.8148181468.74969.2650.7523.224.87.10212.4112.0190162690.54891.9061.5051.248.0-6.20222.281.8160285369.30267.362-2.8023.123.72.80232.552.516821161054.89854.678-0.4018.217.9-1.90241.8131.8154194169.07065.271-5.5022.823.00.90253.272.3158274562.34261.158-1.9027.128.44.80Table 2 Weight values of neural network3.2 Evaluation on equipment combinations3.2.1 Original geological conditionsThere is a coal face, which geological conditions are: height of coal bed 2.5m, obliquity of coal bed 5 radians, rigidity coefficient of coal 1.8; length of mining faces 180m; levels of old rock peak 1; levels of straight rock peak 2; levels of gas 6 m3/(td). We have ever put forth a model based on fuzzy information distribution theory to select coal mining methods. By that model the proper coal mining method for the geological conditions is general fully mechanized coal mining.3.2.2 Available equipment combinationsThere are two kinds of coal face machinery: MXA-600 MLS3-170 and three kinds of brackets: BY240-16/35 ZY35 QY200-14/31. so we have six available equipment combinations.3.2.3 The result of evaluation of different equipment combinationsThe evaluation results of different equipment combinations are as Table 3. From table 3, we can see that 2-th project has the highest scores. So we make it as the best choice. It indicates that MXA-600 and QY240-14/31 is the best equipment combination.Table 3 Evaluation results of different equipment combinationsNo.X8X9Y1Y2Evaluationvalue1MXA-600ZY-3567.91328.40.19482MXA-600QY240-14/3182.44539.20.99773MXA-600BY240-16/3583.97136.40.88344MLS3-170ZY-3560.71826.10.02665MLS3-170QY240-14-3175.48230.30.59516MLS3-170BY240-16/3565.32922.60.02344ConclusionsThis paper puts forth a framework to select equipment for coal faces based on neural network with genetic algorithm and fuzzy selection models. This framework can integrate the engineering Knowledge, and supply intelligent decision support for coal mining. In future work, we can embrace more typical engineering cases and expanded the number of input and output targets such as economic factors. In order to make the whole decision support system more reliable and effective.附录B小煤矿安全生产的开采方法和技术改进翟莘县,邵强,李永明,李华明摘要:河南省平顶山市新华区是中国国家重点煤炭生产县之一,其生产能力是每年超过0.6M
收藏