Abstract
Shale oil is mainly extracted by fracturing. However, it is difficult to determine the optimum construction parameters to obtain maximum productivity. In this paper, a fuzzy comprehensive production evaluation model for fractured shale oil horizontal wells based on random forest algorithm and coordinated principal component analysis is proposed. The fracturing parameters of the target wells are optimized by combining this model with an orthogonal experimental design. The random forest algorithm was used to calculate the importance of data sample factors. The main controlling factors of the production of fractured horizontal wells in shale oil were obtained. To reduce the noise of the sample data, principal component analysis was used to reduce the dimensions of the main control factors. Furthermore, the random forest algorithm was used to determine the weight of the principal components after reducing the dimensionality. The membership function of the main control factors after reducing dimensionality was established by combining the fuzzy statistics and assignment methods. In addition, the membership matrix of the effect prediction of fractured horizontal wells in shale oil was determined. The fuzzy comprehensive evaluation method is used to score and evaluate the effect of fractured horizontal wells. Combined with the orthogonal experimental design method, the optimized parameter design of a fractured horizontal well considering the comprehensive action of multiple parameters is realized. After construction according to the optimized parameters, production following fracturing increases significantly. This verifies the rationality of the optimization method that is proposed in this paper.
Introduction
Extracting oil from shale formations is a complex engineering challenge. Because shale is a "tight" rock with extremely low permeability, oil is trapped in microscopic pores that do not allow for natural flow. To overcome this, engineers utilize hydraulic fracturing (fracking) in horizontal wells to create artificial pathways for the oil. However, determining the optimal construction parameters—balancing geological constraints with engineering inputs—is notoriously difficult. Traditional simulation models often struggle with the inherent "fuzziness" and high dimensionality of oilfield data, leading to suboptimal production. This research proposes a hybrid model combining machine learning, dimensionality reduction, and fuzzy logic to streamline this optimization process.
Methodology
The researchers developed a multi-stage workflow designed to identify and optimize the primary drivers of well productivity.
Feature Screening and Noise Reduction
Initially, a Random Forest (RF) algorithm was employed to rank 16 different geological and engineering factors based on their impact on production. To handle the redundancy and noise inherent in large datasets, Principal Component Analysis (PCA) was used to compress these factors into a smaller set of "principal components."
As shown in the figure above, the first few principal components capture the vast majority of the variance in the dataset, allowing the model to focus on the most significant signals while discarding noise.
Fuzzy Comprehensive Evaluation (FCE)
Recognizing that geological boundaries are often ambiguous, the team integrated Fuzzy Logic. By establishing membership functions, the model can evaluate well performance with a degree of nuance that traditional "hard" classification lacks. This approach better accounts for the uncertainty and overlapping characteristics of different reservoir layers.
Orthogonal Experimental Design
To find the optimal parameter set without exhaustive testing, the study used an orthogonal design. This statistical method allowed the researchers to identify the "global optimum" from over 16,000 possible combinations by simulating only 216 representative schemes, drastically reducing computational requirements.
Datasets and Field Application
The study focused on Area W in the Bohai Bay Basin, China. A comprehensive dataset of 23 fractured horizontal wells was used to train and weight the model.
The model's predictive power was first verified against Well W24 as a blind test. Subsequently, the optimized parameters were applied to Well W25 in a real-world field application to compare performance against non-optimized neighboring wells.
Key Results and Insights
The model identified the Brittleness Index (geological) and the Water-Sand Ratio (engineering) as the two most critical factors for production. Key findings include:
- High Accuracy: The model predicted 12-month cumulative production for the test well with a relative error of just 4.8%.
- Production Increase: Well W25, constructed using the optimized parameters, produced 2.34 times more oil than its non-optimized neighbor, W24, despite having similar horizontal lengths.
- Optimized Recipe: For the target area, the optimal configuration included a fluid volume of 42 $m^3/m$, a slick water ratio of 0.8, and a cluster spacing of 8 meters.
Conclusion
The integration of Random Forest, PCA, and Fuzzy Comprehensive Evaluation provides a robust framework for optimizing shale oil production. By accounting for geological uncertainty and focusing on high-impact engineering parameters, this method offers a significant improvement over traditional trial-and-error or brute-force simulation approaches. While the specific parameters are tailored to the Bohai Bay Basin, the methodology itself is highly adaptable to other unconventional reservoirs globally. Future work will look to expand the training dataset and incorporate economic cost-benefit analyses to further refine the optimization goals.
Summary updated