Shi Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Shi Wang has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
- Hai Qu
School of Energy Resources, China University of Geosciences, Beijing, China
2 shared publications - Ying Liu
Chongqing University of Science and Technology
2 shared publications - Zhitian Lu
Chongqing University of Science and Technology
2 shared publications - Dameng Liu
School of Energy Resources, China University of Geosciences, Beijing 100083, China
1 shared publication - Gensheng Li
National Key Laboratory of Petroleum Resources and Engineering, China University of Petroleum (Beijing), Beijing, China
1 shared publication Junxiang Tong
Department of Liver Surgery, Peking Union Medical College Hospital, Beijing, China
1 shared publication
Latest Publications
- 10 November 2025
- Published by Society of Petroleum Engineers (SPE) in SPE Journal
- Vol. 31 (01) , 40-56
An Artificial Intelligence–Driven Methodology for Predicting Proppant Bed Morphology in Hydraulic Fractures
Summary: Predicting the distribution morphology of proppants in hydraulic fractures is a key focus in hydraulic fracturing research. Guided by the principle of similarity, various experimental and numerical models have been developed to analyze proppant bed morphology under different injection parameters. However, despite decades of practical experience and experimental studies, a robust and universally applicable model for proppant transport or a reliable simulation tool for accurately predicting proppant placement has yet to be fully realized. Based on the similarity principles governing narrow fracture geometry and the geometric self-similarity observed in proppant bed evolution, a numerical model was developed to simulate the transport and deposition of proppants within hydraulic fractures. A feature extraction method was used to process time-series simulation data and obtain six geometrical parameters of the bed. The extracted parameters were compiled into a standardized parameter data set. Based on the training data, a deep learning (DL) model was developed to predict bed evolution under varying conditions. The study shows that Open Source Computer Vision Library (OpenCV) techniques can accurately extract geometric feature parameters of numerically simulated sand beds, and the mean error is less than 2%. The bidirectional long short-term memory (BiLSTM) module effectively captures the temporal dependencies in the proppant bed evolution process, making it well-suited for dynamic prediction of bed morphology. The multihead self-attention (MSA) module enhances the model’s ability to identify key timesteps during proppant accumulation, reducing the impact of nonstationary data and improving prediction stability. The Bayesian optimization (BO) module performs global optimization of model hyperparameters, further improving performance and generalization capability. The BiLSTM-MSA-BO model can accurately predict the bed morphology in planar fractures. Based on the model analysis results, the bed morphology evolution follows three patterns under specific injection parameters, including layer upward, mixed deposition, and stack forward. Fluid velocity and proppant particle size influence equilibrium height most, while fluid viscosity has the least impact. The novel method can quantitatively analyze and predict proppant bed morphology within hydraulic fractures, offering an effective tool for completion engineers to optimize injection parameters.
- 10 November 2025
- Published by Society of Petroleum Engineers (SPE) in SPE Journal
- Vol. 31 (01) , 40-56
An Artificial Intelligence–Driven Methodology for Predicting Proppant Bed Morphology in Hydraulic Fractures
Summary: Predicting the distribution morphology of proppants in hydraulic fractures is a key focus in hydraulic fracturing research. Guided by the principle of similarity, various experimental and numerical models have been developed to analyze proppant bed morphology under different injection parameters. However, despite decades of practical experience and experimental studies, a robust and universally applicable model for proppant transport or a reliable simulation tool for accurately predicting proppant placement has yet to be fully realized. Based on the similarity principles governing narrow fracture geometry and the geometric self-similarity observed in proppant bed evolution, a numerical model was developed to simulate the transport and deposition of proppants within hydraulic fractures. A feature extraction method was used to process time-series simulation data and obtain six geometrical parameters of the bed. The extracted parameters were compiled into a standardized parameter data set. Based on the training data, a deep learning (DL) model was developed to predict bed evolution under varying conditions. The study shows that Open Source Computer Vision Library (OpenCV) techniques can accurately extract geometric feature parameters of numerically simulated sand beds, and the mean error is less than 2%. The bidirectional long short-term memory (BiLSTM) module effectively captures the temporal dependencies in the proppant bed evolution process, making it well-suited for dynamic prediction of bed morphology. The multihead self-attention (MSA) module enhances the model’s ability to identify key timesteps during proppant accumulation, reducing the impact of nonstationary data and improving prediction stability. The Bayesian optimization (BO) module performs global optimization of model hyperparameters, further improving performance and generalization capability. The BiLSTM-MSA-BO model can accurately predict the bed morphology in planar fractures. Based on the model analysis results, the bed morphology evolution follows three patterns under specific injection parameters, including layer upward, mixed deposition, and stack forward. Fluid velocity and proppant particle size influence equilibrium height most, while fluid viscosity has the least impact. The novel method can quantitatively analyze and predict proppant bed morphology within hydraulic fractures, offering an effective tool for completion engineers to optimize injection parameters.
- 14 August 2024
- Published by Society of Petroleum Engineers (SPE) in SPE Journal
- Vol. 29 (10) , 5286-5304
Experimental and Simulation Investigations of Proppant Transport and Distribution Between Perforation Clusters in a Horizontal Well
Summary: Multistage hydraulic fracturing is widely used to stimulate tight reservoirs by means of plug and perforation technology. The proppant distribution between perforation clusters significantly impacts fracture conductivity and well productivity. Uneven slurry distribution is often the norm rather than the exception. Proppant transport behaviors and distribution characteristics are still poorly understood in a horizontal wellbore with clusters, especially at field scales. The objective is to propose an innovative and feasible method to quantitatively evaluate the distribution uniformity of proppant between clusters. In this work, we systematically investigate proppant migration and placement by means of laboratory tests and numerical simulation. Computational fluid dynamics (CFD) and the discrete element method (DEM) are coupled to analyze proppant-fluid flow. The experimental observation and results validate the numerical model and calibrate critical parameters. The transport efficiency (E) and normalized standard deviation (NSD) are used to evaluate proppant distribution. The effects of nine parameters on the E and NSD are investigated at field ranges. The calibrated CFD-DEM model is accurate in studying proppant distribution between multiple clusters. The toe bias is the primary distribution between clusters because of the large inertia originating from high injection rates. Fluid distribution and perforation configuration are critical factors that significantly change the toe bias at the cluster level. Fluid redistribution changes proppant distribution toward the heel. The inline up pattern has the best uniformity, followed by the 180° up-down pattern. The secondary characteristic is bottom-biased within a cluster. Increasing fluid viscosity, using small and light proppants, and pumping high-concentration slurry can improve proppant distribution. The slurry diversion into perforations is hardly changed unless external conditions change. The combination of high-concentration slurry and a large bed quickly induces premature screenout at the toe-side cluster, especially when injecting large and high-density particles. Slurry redistributes toward the heel if the toe-side cluster is blocked. The investigation provides a rational and feasible method for operators to understand proppant transport between clusters and optimize pumping parameters under field situations.
- 14 August 2024
- Published by Society of Petroleum Engineers (SPE) in SPE Journal
- Vol. 29 (10) , 5286-5304
Experimental and Simulation Investigations of Proppant Transport and Distribution Between Perforation Clusters in a Horizontal Well
Summary: Multistage hydraulic fracturing is widely used to stimulate tight reservoirs by means of plug and perforation technology. The proppant distribution between perforation clusters significantly impacts fracture conductivity and well productivity. Uneven slurry distribution is often the norm rather than the exception. Proppant transport behaviors and distribution characteristics are still poorly understood in a horizontal wellbore with clusters, especially at field scales. The objective is to propose an innovative and feasible method to quantitatively evaluate the distribution uniformity of proppant between clusters. In this work, we systematically investigate proppant migration and placement by means of laboratory tests and numerical simulation. Computational fluid dynamics (CFD) and the discrete element method (DEM) are coupled to analyze proppant-fluid flow. The experimental observation and results validate the numerical model and calibrate critical parameters. The transport efficiency (E) and normalized standard deviation (NSD) are used to evaluate proppant distribution. The effects of nine parameters on the E and NSD are investigated at field ranges. The calibrated CFD-DEM model is accurate in studying proppant distribution between multiple clusters. The toe bias is the primary distribution between clusters because of the large inertia originating from high injection rates. Fluid distribution and perforation configuration are critical factors that significantly change the toe bias at the cluster level. Fluid redistribution changes proppant distribution toward the heel. The inline up pattern has the best uniformity, followed by the 180° up-down pattern. The secondary characteristic is bottom-biased within a cluster. Increasing fluid viscosity, using small and light proppants, and pumping high-concentration slurry can improve proppant distribution. The slurry diversion into perforations is hardly changed unless external conditions change. The combination of high-concentration slurry and a large bed quickly induces premature screenout at the toe-side cluster, especially when injecting large and high-density particles. Slurry redistributes toward the heel if the toe-side cluster is blocked. The investigation provides a rational and feasible method for operators to understand proppant transport between clusters and optimize pumping parameters under field situations.
- Open Access
- 26 January 2015
- Published by Vilnius Gediminas Technical University in Transport
- Vol. 33 (1) , 22-31
Application of nonparametric regression in predicting traffic incident duration
Predicting the duration time of incidents is important for effective real-time Traffic Incident Management (TIM). In the current study, the k-Nearest Neighbor (kNN) algorithm is employed as a nonparametric regression approach to develop a traffic incident duration prediction model. Incident data from 2008 on the third ring expressway mainline in Beijing are collected from the local Incident Reporting and Dispatching System. The incident sites are randomly distributed along the mainline, which is 48.3 km long and has six two-way lanes with a single-lane daily volume of more than 10000 veh. The main incident type used is sideswipe and the average incident duration time is 32.69 min. The most recent one-fourth of the incident records are selected as testing set. Vivatrat method is employed to filter anomalous data for the training set. Incident duration time is set as the dependent variable in Kruskal–Wallis test, and six attributes are identified as the main factors that affect the length of duration time, which are ‘day first shift’, ‘weekday’, ‘incident type’, ‘congestion’, ‘incident grade’ and ‘distance’. Based on the characteristics of duration time distribution, log transformation of original data is tested and proven to improve model performance. Different distance metrics and prediction algorithms are carefully investigated. Results demonstrate that the kNN model has better prediction accuracy using weighted distance metric based on decision tree and weighted prediction algorithm. The developed prediction model is further compared with other models based on the same dataset. Results show that the developed model can obtain reasonable prediction results, except for samples with extremely short or long duration. Such a prediction model can help TIM teams estimate the incident duration and implement real-time incident management strategies.
- 26 January 2015
- Published by Vilnius Gediminas Technical University in Transport
- Vol. 33 (1) , 22-31
Open AccessApplication of nonparametric regression in predicting traffic incident duration
Predicting the duration time of incidents is important for effective real-time Traffic Incident Management (TIM). In the current study, the k-Nearest Neighbor (kNN) algorithm is employed as a nonparametric regression approach to develop a traffic incident duration prediction model. Incident data from 2008 on the third ring expressway mainline in Beijing are collected from the local Incident Reporting and Dispatching System. The incident sites are randomly distributed along the mainline, which is 48.3 km long and has six two-way lanes with a single-lane daily volume of more than 10000 veh. The main incident type used is sideswipe and the average incident duration time is 32.69 min. The most recent one-fourth of the incident records are selected as testing set. Vivatrat method is employed to filter anomalous data for the training set. Incident duration time is set as the dependent variable in Kruskal–Wallis test, and six attributes are identified as the main factors that affect the length of duration time, which are ‘day first shift’, ‘weekday’, ‘incident type’, ‘congestion’, ‘incident grade’ and ‘distance’. Based on the characteristics of duration time distribution, log transformation of original data is tested and proven to improve model performance. Different distance metrics and prediction algorithms are carefully investigated. Results demonstrate that the kNN model has better prediction accuracy using weighted distance metric based on decision tree and weighted prediction algorithm. The developed prediction model is further compared with other models based on the same dataset. Results show that the developed model can obtain reasonable prediction results, except for samples with extremely short or long duration. Such a prediction model can help TIM teams estimate the incident duration and implement real-time incident management strategies.
- 1 October 2010
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 1 (19482914) , 203-207
Drug-metabolism-based automatic segmentation of dynamic scintigraphic images
Quantitative dynamic scintigraphic image studies provide in vivo measurements of dynamic physiological and biochemical processes in humans. Segmentation of scintigraphic images is one of the most important steps to distinguish organs of interest from the background. However, poor spatial resolution and signal-to-noise ratio (SNR) of scintigraphic images make accurate organ boundary segmentation difficult. In scintigraphic image studies, areas of specific binding are usually segmented by manually drawing regions of interest (ROIs), which is a time-consuming and subjective process. In this work, we describe a method to automatically segment 99mTc-GSA dynamic scintigraphic images based on the prior knowledge of drug metabolism, including the drug flow mechanism and the time of activity curves (TACs). Our results suggest that drug-metabolism-based segmentation can automatically segment tissues in dynamic scintigraphic images studies and has the potential to replace manual ROI segmentation for our application.
- 1 October 2010
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 1 (19482914) , 203-207
Drug-metabolism-based automatic segmentation of dynamic scintigraphic images
Quantitative dynamic scintigraphic image studies provide in vivo measurements of dynamic physiological and biochemical processes in humans. Segmentation of scintigraphic images is one of the most important steps to distinguish organs of interest from the background. However, poor spatial resolution and signal-to-noise ratio (SNR) of scintigraphic images make accurate organ boundary segmentation difficult. In scintigraphic image studies, areas of specific binding are usually segmented by manually drawing regions of interest (ROIs), which is a time-consuming and subjective process. In this work, we describe a method to automatically segment 99mTc-GSA dynamic scintigraphic images based on the prior knowledge of drug metabolism, including the drug flow mechanism and the time of activity curves (TACs). Our results suggest that drug-metabolism-based segmentation can automatically segment tissues in dynamic scintigraphic images studies and has the potential to replace manual ROI segmentation for our application.