AI based mechanistic modeling and probabilistic forecasting of hybrid low salinity chemical flooding
مدل سازی مکانیکی مبتنی بر هوش مصنوعی و پیش بینی احتمالی سیلاب شیمیایی ترکیبی با شوری کم-2020
Over the past decades, it has been widely shown that Low Salinity Waterflooding (LSW) outperformed High Salinity Waterflooding (HSW) in terms of higher oil recovery, particularly in combining with other conventional Enhanced Oil Recovery (EOR) methods such as chemical flooding to benefit from their synergies. This paper presents a novel approach to mechanistically model Hybrid Low Salinity Chemical Flooding, with: (1) development of a hybrid EOR concept from past decades; (2) utilizing a Multilayer Neural Network (ML-NN) artificial intelligent technique in a robust Equation-of-State reservoir simulator fully coupled with geochemistry; (3) systematic validation with laboratory data; and (4) uncertainty assessment of the LSW process at the field scale. Various parameters such as polymer, surfactant, and salinity can affect on the relative permeability simultaneously during hybrid recovery processes. To overcome this problem, the ML-NN technique was applied for multidimensional interpolation of the relative permeability. Additionally, ML-NN was used within a Bayesian workflow to capture the uncertainties in both history matching and forecasting stages of LSW at field scale. The proposed model indicated good agreements with various coreflooding experiments including HSW, LSW, and Low Salinity Surfactant flooding (LSS), where it can efficiently capture the complex geochemistry, wettability alteration, microemulsion phase behavior, and the synergies occurring in these hybrid processes.
Keywords: Low salinity waterflooding | Chemical flooding | Hybrid EOR | Artificial intelligence | Probabilistic forecasting
Bayesian deep learning based method for probabilistic forecast of day-ahead electricity prices
روش مبتنی بر یادگیری عمیق بیزی برای پیش بینی احتمالی قیمت برق روز پیش رو-2019
The availability of accurate day-ahead energy prices forecasts is crucial to achieve a successful participation to liberalized electricity markets. Moreover, forecasting systems providing prediction intervals and densities (i.e. probabilistic forecasting) are fundamental to enable enhanced bidding and planning strategies considering uncertainty explicitly. Nonetheless, the vast majority of available approaches focus on point forecast. Therefore, we propose a novel methodology for probabilistic energy price forecast based on Bayesian deep learning techniques. A specific training method has been deployed to guarantee scalability to complex network architectures. Moreover, we developed a model originally supporting heteroscedasticity, thus avoiding the common homoscedastic assumption with related preprocessing effort. Experiments have been performed on two dayahead markets characterized by different behaviors. Then, we demonstrated the capability of the proposed method to achieve robust performances in out-of-sample conditions while providing forecast uncertainty indications..
Keywords: Electricity price forecasting | Probabilistic forecasting | Deep learning | Bayesian learning | Neural network