<p>Drilling is one of the most crucial operations in the petroleum industry, hence the prediction of drilling performance is important for cost and efficiency estimation in project designing. Based on the field data collected from a vertical well in Xinjiang province, a drilling performance prediction function was trained by BP-ANN (back-propagation artificial neural network). The input parameters of this function consist of mineral composition and operating factors. Predicting drilling performance from the combination of mineral composition and operating factors is a typical multivariate nonlinear fitting problem. BP - ANN has extremely high accuracy in solving similar problems, so the widely used back - propagation artificial neural network was selected. According to the training and testing results, the introduction of mineral composition can effectively improve the training speed and testing accuracy. Through the established prediction function, a dynamic optimization algorithm combined with DOE (Design of Experiments) theory was also developed. This algorithm sets independent optimization strategies for different drilling performances (ROP(rate of penetration), drill diameter, inclination, azimuth) and reduces the number of tests for finding the best drilling performance. The optimization effect is obvious, with the ROP increased by 408% and the inclination decreased by 67%.</p>