Volume 55

Dynamic Production Forecasting Method Based on the EMD-iTransformer Model Lizhe Li; Fujian Zhou

Abstract

In the development of unconventional oil and gas resources, dynamic production forecasting plays a crucial role in guiding production optimization and resource allocation. However, production variations are influenced by a combination of multi-source heterogeneous factors, including geological conditions, well control parameters, and historical production data. Consequently, forecasting models must handle high-dimensional, nonlinear, and temporally complex data while maintaining strong generalization ability and adaptability.
To address these challenges, this study proposes a dynamic production forecasting method tailored for multi-source heterogeneous data. The approach applies Empirical Mode Decomposition (EMD) to perform multi-scale decomposition of the original time series, enabling the extraction of key features across different frequency levels. These features are then integrated into an iTransformer-based predictive model, enhancing the model’s capability to process non-stationary and multivariate dynamic data. This chapter first establishes and preprocesses the multi-source dataset, covering key dimensions such as well control parameters, geological attributes, and production data, followed by data cleaning and normalization. Subsequently, the structure and optimization process of the dynamic forecasting model are presented in detail, with a focus on the EMD decomposition strategy and enhancements to the iTransformer architecture. Experiments are conducted under both single-well and multi-well scenarios. The results demonstrate that the proposed method achieves high prediction accuracy and strong robustness in dynamic production forecasting tasks. Notably, it exhibits excellent trend-capturing capability in forecasting newly drilled wells. This study provides a novel approach and technical foundation for multi-source data-driven production prediction, offering promising potential for digital modeling and intelligent forecasting of complex reservoir production processes.

Keywords AI-Driven Production Forecasting, Multi-Source Data Fusion, Nonlinear Time Series Modeling, Energy Intelligence for Sustainable Development