Rapid productivity prediction method for frac hits affected wells based on gas reservoir numerical simulation and probability method

Jie Nie, Hao Wang, Yuexiang HaoOpen Physics 2023 · DOI

Abstract

As an important unconventional resource, shale gas can alleviate energy shortage, and its efficient development ensures the long-term growth of oil and gas. The prediction of production levels and estimated ultimate recovery with high accuracy is necessary for shale gas development. Conventional methods are widely applied in the oil and gas industry owing to their simplicity and effectiveness; however, none of them can accurately predict the results for frac hits affected wells. In this work, a probability method based on the numerical model of shale gas reservoir has been formed. In view of the impact of frac hits on the productivity of production wells during the development of shale gas reservoirs, an embedded discrete fractured numerical simulation method for gas reservoirs is proposed to simulate the geological engineering parameter range of wells before frac. And aiming at the established numerical model of shale gas reservoir, this method adopts the ensemble smoother with multiple data assimilation automatic history matching technology to carry out the history matching process of the model. Based on the probability theory and numerical simulation results, this study analyses the influence of different distribution functions of parameters on the calculation results of reserves, and obtains the expected curve of reserves through combination calculation. Besides, the effectiveness of this method was verified by comparing with other traditional predicted method.

Introduction

In the development of unconventional shale gas resources, "frac hits"—interference caused by hydraulic fracturing of adjacent wells—represent a significant operational challenge. These events often lead to pressure channeling, increased water production, and physical damage to existing fracture networks, causing abrupt and often permanent changes in production behavior.

Traditional empirical methods, such as Decline Curve Analysis (DCA), rely on the assumption of stable reservoir physics. When a frac hit occurs, these assumptions are violated, rendering standard predictions inaccurate. While full-scale numerical simulations can account for these physical changes, they are traditionally too computationally intensive for rapid field-wide application. This paper presents a hybrid workflow that combines high-fidelity numerical simulation with probabilistic modeling to provide rapid and accurate productivity predictions for wells affected by these interference events.

Methodology

The proposed workflow balances physical rigor with computational efficiency through a multi-stage process:

  1. Embedded Discrete Fracture Modeling (EDFM): To accurately represent the complex, multi-scale fracture networks altered by frac hits, the researchers utilized EDFM. This approach explicitly models fracture geometries rather than averaging them into a grid, allowing for a more realistic simulation of post-interference flow paths.
  2. Dimensionality Reduction via Autoencoder: History matching in complex reservoirs often involves hundreds of parameters, leading to the "curse of dimensionality." To address this, an unsupervised autoencoder neural network was employed to compress 368 reservoir parameters into a lower-dimensional latent space, significantly accelerating the optimization process.
  3. ES-MDA Algorithm: The Ensemble Smoother with Multiple Data Assimilation (ES-MDA) was used for automated history matching. Unlike sequential filters, ES-MDA allows for parallel computing and does not require model restarts at every time step, making it ideal for rapid iteration.
Workflow integrating autoencoder-based dimensionality reduction with the ES-MDA algorithm for rapid history matching.
  1. Probabilistic Forecasting: Recognizing the inherent uncertainty in subsurface geometry, the method employs Monte Carlo sampling. By assigning probability distributions to uncertain parameters (e.g., fracture conductivity in the reformed zone), the workflow generates a range of potential outcomes, including P10, P50, and P90 production curves.

Case Study and Validation

The effectiveness of the method was validated using production data from Well W5 in the Weiyuan shale gas block, China. This well experienced a severe frac hit after 212 days of initial production, which caused gas production to drop to zero and significantly increased water production (the "water lock" effect).

Observed gas and water production data for Well W5, illustrating the production cessation following a frac hit.

The model accounted for 368 parameters, including matrix permeability and stress sensitivity. By adjusting regional permeability and fracture openings during the history-matching phase, the researchers were able to reconstruct the post-hit reservoir state.

Key Results

The proposed method demonstrated high accuracy and efficiency:

  • EUR Prediction: The model predicted a P50 Estimated Ultimate Recovery (EUR) of 134 million cubic meters.
  • Comparative Accuracy: When compared against established empirical models (Duong, YM-SPED, and WK methods) applied after production resumed, the proposed method showed remarkably low error rates, ranging from 0.75% to 4.96%.
  • Computational Speed: By combining the autoencoder with the ES-MDA approach, a stable, matched state was achieved in just 200 total simulations across two iterations, a significant improvement over traditional manual matching.

Conclusion

This research introduces a "grey box" approach that respects the underlying physics of reservoir interference while acknowledging subsurface uncertainty. By integrating deep learning for dimensionality reduction with robust data assimilation, the workflow enables engineers to rapidly re-evaluate well potential following frac hits. Future work will focus on integrating real-time microseismic data to further constrain fracture geometry during the interference event itself, moving from reactive to proactive productivity management.

Summary updated

Abstract

Abstract As an important unconventional resource, shale gas can alleviate energy shortage, and its efficient development ensures the long-term growth of oil and gas. The prediction of production levels and estimated ultimate recovery with high accuracy is necessary for shale gas development. Conventional methods are widely applied in the oil and gas industry owing to their simplicity and effectiveness; however, none of them can accurately predict the results for frac hits affected wells. In this work, a probability method based on the numerical model of shale gas reservoir has been formed. In view of the impact of frac hits on the productivity of production wells during the development of shale gas reservoirs, an embedded discrete fractured numerical simulation method for gas reservoirs is proposed to simulate the geological engineering parameter range of wells before frac. And aiming at the established numerical model of shale gas reservoir, this method adopts the ensemble smoother with multiple data assimilation automatic history matching technology to carry out the history matching process of the model. Based on the probability theory and numerical simulation results, this study analyses the influence of different distribution functions of parameters on the calculation results of reserves, and obtains the expected curve of reserves through combination calculation. Besides, the effectiveness of this method was verified by comparing with other traditional predicted method.
BibTeX
@inproceedings{nie2023rapid,
  author = {Jie Nie and Hao Wang and Yuexiang Hao},
  title = {{Rapid productivity prediction method for frac hits affected wells based on gas reservoir numerical simulation and probability method}},
  booktitle = {Open Physics},
  year = {2023},
  doi = {10.1515/phys-2022-0233},
}
Figure