Machine Learning for NGL Plant Compressor Train Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
NGL plants face uncertainties in predicting incoming feed gas volumes, leading to deviations from optimal compressor recycle rates, missed maintenance opportunities, and operational urgency due to inaccurate estimates.
Innovation Solution
The implementation of supervised machine learning algorithms, such as regression and decision tree models, to predict incoming feed gas volumes and determine the optimal number of running compressor trains and recycle rates, advising operators on train operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional estimation methods are used to predict incoming feed gas volumes, then operational simplicity is maintained, but prediction accuracy deteriorates leading to deviations from optimal compressor recycle rates
Solution Approach 1:
The patent replaces traditional mechanical estimation methods with machine learning algorithms that process upstream flow volumes, input flows, and operating conditions to predict incoming feed gas volumes. This substitution of predictive methodology dramatically improves accuracy while the automated nature of ML models keeps operational complexity manageable.
Solution Approach 2:
The patent introduces machine learning models as an intermediary between raw operational data and decision-making processes. The ML models process multiple input parameters (upstream flow volumes, input flows, operating conditions) and translate them into accurate predictions of incoming feed gas volumes, serving as a mediator that improves prediction accuracy without requiring operators to directly complex calculations.
2Reliability
If compressor trains are operated with high recycle rates to ensure sufficient capacity, then operational reliability is improved, but energy consumption increases
Solution Approach 1:
The patent implements dynamic adjustment of compressor recycle rates based on predicted incoming feed gas volumes. Instead of maintaining static high recycle rates for reliability, the system dynamically optimizes recycle rates according to actual feed conditions, ensuring sufficient capacity when needed while reducing energy consumption when feed volumes are lower.
Solution Approach 2:
The patent establishes a feedback loop where predicted incoming feed gas volumes inform compressor operational decisions. The ML model continuously predicts feed volumes, and these predictions feed back into operational adjustments of recycle rates and train configurations, creating a closed-loop system that balances reliability and energy efficiency.
3Stability of the object's composition
If more compressor trains are kept running to handle potential feed gas surges, then operational stability is improved, but maintenance opportunities are missed
Solution Approach 1:
The patent uses machine learning models to predict incoming feed gas volumes in advance, enabling proactive operational planning. By knowing predicted feed volumes beforehand, operators can proactively schedule maintenance during periods of lower predicted feed gas flow, rather than reactively keeping all trains running and missing maintenance opportunities.
Solution Approach 2:
The system performs preliminary prediction of feed gas volumes to inform advance scheduling decisions. This allows the plant to proactively plan maintenance activities during predicted low-feed periods, preventing the loss of maintenance time that occurs when trains are kept running continuously without accurate predictive information.
4Productivity
If accurate prediction of incoming feed gas volumes is achieved using machine learning, then operational optimization is improved, but computational requirements increase
Solution Approach 1:
The patent applies machine learning models to predict incoming feed gas volumes, which improves operational optimization. The system processes upstream flow volumes, input flows, and operating conditions through ML algorithms to achieve accurate predictions, balancing computational investment with operational benefits.
Data Source
AI summary
Systems and methods for operating a natural gas liquids (NGL) plant can include obtaining upstream flow volumes, input flows, and operating conditions of a refinery complex including the NGL plant for a first time period and a second time period. One or more features can be extracted from the upstream flow volumes, input flows, and operating conditions for multiple first time periods and used to form multiple feature vectors. A machine learning model trained with labeled data (e.g., labeled data associating upstream flow volumes, input flows, and operating conditions with incoming feed gas volumes) representing incoming feed gas of the NGL can be used to process the feature vectors to determine predicted incoming feed gas volumes.