Adaptive MINLP Process Control With Oracle-Based Feasibility Pruning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for continuous process optimization, such as those in refineries and petrochemical plants, face inefficiencies due to the complexity and time-consuming nature of Mixed Integer Nonlinear Programming (MINLP) solvers, which are not optimized for processor and memory usage, leading to suboptimal decision-making in switching operations.
Innovation Solution
The implementation of online first-principles simulation techniques in conjunction with a MINLP solver, utilizing an oracle system with adaptive algorithms to manage infeasible solutions, reduce the search space, and automatically generate feasible paths, thereby enhancing the robustness and efficiency of the optimization process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional MINLP solvers search all possible regions for optimal solutions, then solution completeness is improved, but computational time and complexity increase significantly
Solution Approach 1:
The patent performs preliminary feasibility analysis using unit models and oracle components before the main MINLP optimization. This preliminary action identifies and eliminates infeasible regions in advance, so the solver doesn't waste time exploring them during the main optimization process, thus reducing computational time while maintaining solution completeness.
Solution Approach 2:
The patent extracts and separates the feasibility checking function into a distinct oracle component that operates independently from the main MINLP solver. This extracted feasibility analysis prunes infeasible search spaces before they enter the main optimization, reducing the search space without compromising the solver's ability to find complete optimal solutions.
2Manufacturing precision
If conventional MINLP solvers explore all possible switching states, then optimal switching decisions are improved, but processor and memory usage increase
Solution Approach 1:
The system performs preliminary feasibility screening using unit models and oracle components to identify promising switching states before submitting them to the MINLP solver. This preliminary action filters out infeasible switching combinations, reducing the number of states the solver must evaluate, thus lowering processor and memory requirements while maintaining optimization accuracy.
Solution Approach 2:
The patent introduces unit models and oracle components as intermediary layers between the problem definition and the MINLP solver. These intermediaries pre-process the switching state space, eliminating infeasible states before they reach the solver, thereby reducing the computational burden on processors and memory without affecting the final optimization accuracy.
3Productivity
If the search space is reduced based on initial feasibility, then computational efficiency is improved, but risk of missing optimal solutions increases
Solution Approach 1:
The patent implements feedback mechanisms where the oracle component continuously monitors feasibility and provides information back to guide the search process. This feedback ensures that the search space reduction is based on actual feasibility constraints rather than arbitrary reductions, maintaining solution optimality while improving computational efficiency through intelligent pruning of truly infeasible regions.
Solution Approach 2:
The unit models and oracle components serve as intermediaries that accurately assess feasibility before search space reduction. These intermediaries ensure that only genuinely infeasible regions are pruned, while preserving all potentially optimal feasible regions, thus maintaining solution optimality while achieving computational efficiency through targeted search space reduction.
Data Source
AI summary
Real-time dynamic process modeling in an online model-based process control computing environment. A solver system utilizes an oracle to implement adaptive algorithms for a mixed integer nonlinear programming (MINLP) solver and a nonlinear programming (NLP) solver.