Search

Fluid rate quantification is important for field development and well planning. Existing physical and data-driven models require a copious amount of reservoir information and history to improve the capability of prediction. In this paper, we introduce a new approach of using a deep meta learning approach using Neural Basis Expansion Analysis for Interpretable Time Series Forecasting (N-BEATS) model to forecast the oil rate, eliminating the need of large historical training datasets to overcome the limitation of conventional models. Two deep learning models, the Long- Short-Term Memory (LSTM) and N-BEATS, were used to forecast the oil rate data of two wells from Norne Field in Norway. The historical bottom hole pressure data with only 500 data points were used as input in the LSTM model. To overcome the dataset limitation, we developed and tested a pretrained N-BEATS transfer model. The N-BEATS model was initially trained on a large and heterogeneous dataset, widely known as the M4 series, with an excess of 100,000 time series. This pretrained model was then used to forecast the oil rates using a meta learning approach. It was challenging to fit the LSTM model, as it demanded complex hyperparameter tuning and grid search to solve this nonlinear multivariate time-series forecasting problem. Even after we tuned the model's parameters, the model could not generalize the testing data attributed to the limited and small dataset available during training. Our work showed that the newly proposed N-BEATS meta learning model was better able to capture the dynamic time series features from the M4 dataset and transfer the learned knowledge. The N-BEATS approach was able to forecast oil rates in Norne field for both wells without supplying the model with any features or input variables, such as the reservoir pressure. The Mean Squared Error (MSE) was used as an evaluation metric to benchmark the prediction results from both models. The results clearly demonstrated that the N-BEATS meta learning approach outperformed the LSTM model performance when comparing the difference between the model forecast and actual data. We concluded that using the pretrained N-BEATS model overcomes the previous disadvantages in LSTMs model requiring feature selection and abundant training history. To the best of our knowledge, our work presents a novel approach of using N-BEATS meta learning model for oil rate forecasting without any prior knowledge of the field. Most previous machine and deep learning attempts were based mainly on deriving predictions from an abundant historical field data. This research showed promising results using meta learning and N-BEATS model in the petroleum industry, especially for new or green fields with limited historical data.

Accurate well rate forecasting is critical for field development. In the oil industry, recurrent-based deep learning models have been used for production forecasting. Modern research is shifting towards using the novel transformer architecture in natural language and time series applications to better handle long-term temporal dependencies using attention layers. In this paper, we present a novel approach of applying a transformer-based deep learning model for probabilistic production forecasting. The Temporal Fusion Transformer (TFT) model, a modern transformer-based model for time series forecasting, was used to provide a better oil rate prediction in the Norwegian Volve field. The historical bottomhole pressure, wellhead pressure, wellhead temperature and choke size opening real time series data were used as past input features in the TFT model to forecast ahead the oil rate of two wells. The model was trained with a quantile loss function to produce a probabilistic prediction with both upper and lower bounds of uncertainties. After optimization the TFT model was blindly tested on the last 20% of the data to evaluate its prediction performance compared Block Recurrent Neural Network architectures (BlockRNN). The real-time production data used in this multivariate forecasting problem was found to be complex with no clear trend or seasonality. The BlockRNN model failed to produce a good prediction of oil rate compared to the TFT model. The TFT model was better at encoding the input features including, the choke size, wellhead pressure, and temperature, and understanding their long-range dependencies with the oil rate.The model was able to minimize the testing Mean Squared Error (MSE) for the two wells F-11H and F-12H reaching values of 0.08 and 0.03, respectively. In addition, the model was able to forecast a prediction bandwidth in between the 10th and 90th quantiles to account for uncertainty ranges which laid in-between the blind test intervals. Overall, the TFT model was proven successful in accurately forecasting complex trends of oil rates overcoming the limitations of conventional, BlockRNN uses a memory to understand and make predictions about new data models of information loss over long-term multivariate time series prediction. Our work presents a novel approach of using the TFT probabilistic deep learning model for multivariate oil rate forecasting in the oil and gas industry. The model showed very promising results outperforming conventional BlockRNN-based models in addition to providing a range of the forecasting uncertainty using quantile regression. Knowing the uncertainty range helps in making critical decision particularly in the well intervention and field development.