基于神经网络的地质钻探工况判别模型研究

发布时间：2018-06-05 05:28

本文选题：工况判别 + 钻探参数　；参考：《中国地质大学(北京)》2016年硕士论文

【摘要】：地质钻探工程孔内事故会造成严重的经济损失、时间浪费,甚至会威胁人的生命安全,地质钻探工程工况判别系统能在第一时间判断孔内事故类型,为处理事故争取更多时间,以遏制事故朝着更严重的方向发展。鉴于地质钻探参数与工况类型间的不确定性关系,神经网络具有非线性映射能力,且建模快捷方便,采用神经网络建立工况判别模型事半功倍,相比其他智能化手段,优势明显。根据地质钻探钻参仪所能获得的参数,对常见工况模式类型与钻探参数间的相互关系进行了分析,排除了对工况类型判别影响不大、非独立的参数,保留了能直接反应工况类型的钻探参数:转速、称重、返浆流量、泥浆密度、泵压、钻速,使得工况类型与钻探参数间的映射关系简单有效,并将此6个钻探参数作为神经网络的输入参数。Matlab提供算法编程、调用函数名、界面化设计三种神经网络模型创建方法,与前两者相比,界面化设计的建模过程更简单,参数优化更方便,提供了多种神经网络工具,通过对比其中各种神经网络工具的优缺点和适用范围,排除了线性神经网络、竞争型神经网络、反馈型神经网络,优选出2种BP神经网络和4种RBF神经网络,BP、RBF网络反应迅速、非线性映射能力强,非常适合工况模式判别。在此基础上,利用甘肃阳山矿区实际施工的历史数据检验了BP、RBF网络性能。首先,对钻探施工过程中钻探参数的变化特征进行了研究,对获取的历史数据进行了降噪处理;其次,对钻探事故发生时钻探参数的变化趋势进行了分析,将参数变化趋势作为神经网络的输入数据,提高了神经网络模型的适应性和推广能力;最后,分别对不同BP、RBF神经网络建立模型,BP神经网络中表现最好的是基于LM、BR算法的Cascade-forward backprop,RBF神经网络中表现最好的是PNN网络,通过对比两者的性能差异,最终确定以PNN网络建立工况判别模。PNN网络反应迅速、稳定性高、网络结构简单、无需训练,经过试验仿真,模型判断准确率可达91%。
[Abstract]:The accident in the geological drilling engineering can cause serious economic loss, time waste and even threaten the life safety of people. The geological drilling engineering condition discriminant system can judge the type of the accident in the first time and strive for more time to deal with the accident, so as to prevent the accident from developing in a more serious direction. The neural network has the ability of nonlinear mapping, the neural network has the ability of nonlinear mapping, and the modeling is fast and convenient. It is more advantageous to use the neural network to establish the working condition discriminant model. Compared with other intelligent means, it has obvious advantages. According to the parameters obtained by the geological drilling drilling parameter, the relationship between the common working mode type and the drilling parameters is related to the parameters obtained by the drilling parameter. The system has been analyzed, and the non independent parameters have been excluded, and the drilling parameters that can react directly to the type of working conditions are retained, such as speed, weight, back slurry flow, mud density, pump pressure and drilling speed, which make the mapping between the working conditions type and drilling parameters simple and effective, and the 6 drilling parameters are used as neural network. The input parameter.Matlab provides algorithm programming, call function name and interface design three neural network model building methods. Compared with the previous two, the modeling process of the interface design is simpler, the parameter optimization is more convenient, and a variety of neural network tools are provided. By comparing the advantages and disadvantages of various neural network tools and the scope of application, the method is excluded. Linear neural network, competitive neural network, feedback neural network, 2 BP neural networks and 4 RBF neural networks, BP, RBF network respond quickly, nonlinear mapping ability is strong, and it is very suitable for working mode discrimination. On this basis, the performance of BP and RBF network is tested by the historical data of the actual construction of Yangshan mining area in Gansu. First, the performance of RBF network is tested. The variation characteristics of drilling parameters during the drilling construction are studied, and the acquired historical data are de-noised. Secondly, the variation trend of drilling parameters during drilling accidents is analyzed, and the variation trend of the parameters is taken as the input data of the neural network, and the adaptability and extension ability of the neural network model are raised. Finally, the model of different BP and RBF neural networks is set up respectively. The best performance of BP neural network is based on LM, BR algorithm Cascade-forward BackProp, and the best performance of RBF neural network is PNN network. By comparing the performance differences of the two networks, the final determination of the PNN network built mode.PNN network is rapid and stable, and the network is stable. The simulation results show that the accuracy of the model is 91%..
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：P634

【相似文献】