Tracking the state and diagnosing downhole permanent sensors in intelligent-well completions with artificial neural network

Permanent downhole sensors are extensively used in Intelligent Well completions. Typically, their numbers increase with the number of controllable zones. Replacement of a failed sensor rarely occurs in practice even when data is known to be incorrect (or completely missing) due to the cost implication of a workover. However, these sensors serve as the optic nerve for achieving the vision of Real Time Optimization. They measure parameters of interest (Monitoring) which is the first stage of the production optimization loop frequently known as "Measure, Analyze and Control", which will manage the reservoir and well performance. The reliability of permanent downhole gauges is not 100% throughout the well life. It is therefore imperative to authenticate whenever possible, the operational state of these downhole sensors. Such authentication will allows greater confidence to be placed in the data they supply. This paper discusses how authentication is achieved by comparing the soft (computed) sensor value predicted by an Artificial Intelligence technique with the actual hard data. This approach is demonstrated using a real case data. An Artificial Neural Network (ANN) is trained with multi-dimensional historical downhole data. The ANN is shown to be able to validate continued proper sensor operation or diagnose a future malfunction of any sensor. It will also be shown that the trained ANN can predict parameters such as the zonal BHP, with sufficient accuracy and that they serve as back up to maintain reservoir management operations. Copyright 2007, Society of Petroleum Engineers.