A ROP prediction approach based on improved BP neural network

Effective prediction of ROP (Rate of Penetration) is a crucial part of successful well drilling process. Due to the penetration complexities and the formation heterogeneity, traditional way such as ROP equations and regression analysis are confined by their limitations in the drilling practices. With the accumulation of the geology data and drilling logs, data-based modelling methods like ANN become powerful tools in modern drilling engineering. This paper proposed a ROP prediction approach based on improved BP neural network technologies. The main idea is to build prediction model of target well from well logs through the improved BP neural network modelling method. During the training process, the traditional BP training algorithm is improved by introducing momentum factor and the dynamical learning rate, which are able to notably increase the speed of converging and obtain better generalization performance. We collect and analyze the well log of the No.104 well in Yuanba, China. The experiment results show that the proposed approach is able to effectively utilize the engineering data, and provide accurate ROP prediction in the areas which have certain amount of data collection.