Geological Drilling Accident Discrimination Model Based On Neural Network

Drilling accidents of Geological drilling engineering will cause serious economic loss and time waste, even threaten the safety of human beings life. Geological drilling accident discrimination model can judge the type of drilling accidents in time, and create more time to handle accidents to stop it become more serious. With consideration of uncertainty relations between drilling parameters and the type of Geological drilling accidents, the discrimination model is established using neural network, which have nonlinear mapping ability and can model quickly and conveniently compared with other intelligent models.Based on the study of the parameters obtained from Geological drilling parameter instrument, this paper analyses the relations between the common types of Geological drilling accidents and drilling parameters. After removing some unaffected parameters to discrimination model, this paper obtained some independent drilling parameters, such as rotary speed、string suspending weight、return flow rate、mud density、pump pressure、drilling speed, which can reflect the type of Geological drilling accidents easily.Matlab provides 3 methods to build a neural network model. Those methods are algorithm programming, calling function name and graphical user interface(GUI). Compared with the former two methods, GUI can model easier and more convenient. GUI can also provide many neural network tools. By comparing the advantages and disadvantages of these neural network tools and the scope of application of them, this paper chooses 2 kinds of BP networks and 4 kinds of RBF networks after removing linear neural networks, competitive neural networks and feedback neural networks. BP and RBF networks have a swift response with strong nonlinear mapping ability, and they are very suitable for discrimination model. After collecting and analyzing some historical data of drilling accidents in Yangshan mining area, Gansu province, this paper tests the performance of RBF and BP networks through 3 steps. First step, variation characteristics of drilling parameters are analyzed and de-noised. Second step, changing trend of drilling parameters is analyzed and used as input parameters of these neural networks. Third step, discrimination models are built by different RBF and BP networks. Contrasting the performance differences between these models, cascade- forward BP based on LM and BR training algorithm is the best in all BP networks, and PNN is the best in all RBF networks. After comparing the performance of cascade- forward BP and PNN, finally, this paper determines PNN network to build Geological drilling accident discrimination model. The accuracy of estimation is up to 91% in the simulation test of PNN model with the performance of high stability、fast response and simple structure.