A NEW DIAGNOSTIC TOOL FOR PERFORMANCE EVALUATION OF MATURE HEAVY OIL WATERFLOODS: CASE OF WESTERN CANADIAN HEAVY OIL RESERVOIRS

Waterflooding is traditionally considered as the least favorable recovery method in heavy oil reservoirs. Despite this common belief, there have been reported cases of successful waterfloods with recovery factors as high as 40% in Western Canada. On this basis, a comprehensive statistical study was conducted to determine the effects of various reservoir and operational parameters on the performance of waterfloods and potential differences of heavy oil waterfloods with medium oil waterfloods. Collected database in this study includes 120 operational and reservoir parameters for 177 waterfloods in Alberta and Saskatchewan. Statistical analysis of collected database and 15 different performance indices based on studied injection‑production history was conducted using partial least square technique. This study revealed the significance of operational parameters on performance of heavy oil waterfloods. It also provided a ranking of various operational and reservoir parameters on performance of waterfloods which were used for dimension reduction of input parameters. In the next step, artificial neural network technique was applied to develop performance predictive models based on selected parameters after dimension reduction. Error analysis of developed neural network models showed average absolute error of 17% from calculated performance indices based on production and injection history of studied waterfloods. This paper provides details of successful application of partial least square approach and neural network for developing a new diagnostic tool for evaluation and prediction of performance of waterfloods in heavy oil reservoirs based on more than 50 years experience of heavy oil waterflooding in Western Canada. The developed tool in this study uses combination of 38 easily obtainable operational and reservoir parameters to predict the performance of heavy oil waterfloods.