Nonlinear control of an oil well

Recent studies of waterflood and steamdrive in petroleum reservoirs have revealed that unwanted extension of hydrofractures is caused primarily by aggressive actions taken by conventional PI or PID controllers during injector start-up, or by operating near hydrofracturing pressure. The extension has resulted in reservoir damage and irreversible lost oil production. In this paper, we consider CalResources Phase III steam injection pilot in the South Belridge field of California. For each injector in this pilot, a neural network model is identified by using historical data on injection-fluid flow rate, well-head pressure, depth to the top of perforation, and the length of perforated interval. Differential geometric well-head pressure controllers are synthesized by using the neural network models. The satisfactory performance of the neural network model-based controllers is demonstrated, via numerical simulations.