Abstract:A novel data-driven stochastic robust optimization (DDSRO) framework is proposed for optimization under uncertainty leveraging labeled multi-class uncertainty data. Uncertainty data in large datasets are often collected from various conditions, which are encoded by class labels. Machine learning methods including Dirichlet process mixture model and maximum likelihood estimation are employed for uncertainty modeling. A DDSRO framework is further proposed based on the data-driven uncertainty model through a bi-level optimization structure. The outer optimization problem follows a two-stage stochastic programming approach to optimize the expected objective across different data classes; adaptive robust optimization is nested as the inner problem to ensure the robustness of the solution while maintaining computational tractability. A decomposition-based algorithm is further developed to solve the resulting multi-level optimization problem efficiently. Case studies on process network design and planning are presented to demonstrate the applicability of the proposed framework and algorithm.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to make optimal decisions using multi - class uncertain data with labels in the big data era. Specifically, the author proposes a data - driven stochastic robust optimization (DDSRO) framework based on machine learning to address the following challenges: 1. **Modeling of complex uncertain data**: In large - scale data sets, uncertain data are usually labeled, representing different categories or conditions (such as weather conditions). These data may have a complex distribution structure and are difficult to fit with common probability distributions. Therefore, effective machine - learning methods need to be developed to accurately extract useful information from these data. 2. **Integrating the advantages of different optimization methods**: Existing optimization methods such as stochastic programming and robust optimization each have their own advantages and disadvantages. Stochastic programming relies on accurate probability distribution information, while robust optimization does not require such information but may lead to conservative solutions. Therefore, a comprehensive method is needed to combine the advantages of both. 3. **Efficiently solving large - scale multi - level optimization problems**: Due to the introduction of multi - class uncertain data, the optimization problem becomes more complex, and traditional optimization solvers cannot directly handle it. Therefore, efficient algorithms need to be developed to solve such problems. ### Main contributions To address the above challenges, the paper makes the following innovations: - **Uncertainty modeling framework based on machine learning**: Through methods such as maximum - likelihood estimation and Dirichlet process mixture models, accurately extract information on probability distributions and uncertainty sets from multi - class uncertain data with labels. - **General data - driven decision - making framework**: Combine machine learning with mathematical programming, integrating the advantages of stochastic programming and robust optimization to form an organic overall framework. - **Efficient decomposition algorithm**: Develop a decomposition - based algorithm for efficiently solving the resulting multi - level mixed - integer linear programming (MILP) problems. ### Framework overview The main structure of the DDSRO framework is as follows: - **Outer optimization problem**: Adopt a two - stage stochastic programming method to optimize the expected target values of different data categories. - **Inner optimization problem**: Embed adaptive robust optimization (ARO) to ensure the robustness of the solution and maintain the solvability of the calculation. In this way, the DDSRO framework can provide high - quality and computationally feasible optimization solutions while making full use of various types of uncertain data.

Data-Driven Stochastic Robust Optimization: A General Computational Framework and Algorithm for Optimization under Uncertainty in the Big Data Era

Data-driven stochastic robust optimization: General computational framework and algorithm leveraging machine learning for optimization under uncertainty in the big data era

Improving Robust Decisions with Data

Multistage Robust Mixed-Integer Optimization under Endogenous Uncertainty

Data‐driven adaptive nested robust optimization: General modeling framework and efficient computational algorithm for decision making under uncertainty

Multistage Distributionally Robust Optimization for Integrated Production and Maintenance Scheduling

New Robust Optimization Approach Induced by Flexible Uncertainty Set: Optimization under Continuous Uncertainty

Optimization under uncertainty in the era of big data and deep learning: When machine learning meets mathematical programming

A Unified Framework for Adjustable Robust Optimization with Endogenous Uncertainty

Robust Optimization in High-Dimensional Data Space with Support Vector Clustering

A New Methodology for Robust Optimizations of Optimal Design Problems under Interval Uncertainty

Distributionally robust optimization for planning and scheduling under uncertainty

A Direct Coupled Solution Methodology for Efficient Robust Optimizations of Inverse Problems under Uncertainty

Non-Probabilistic Robust Equilibrium Optimization of Complex Uncertain Structures

Data-Driven Robust Optimization Under Correlated Uncertainty: A Case Study Of Production Scheduling In Ethylene Plant (Reprinted From Computers And Chemical Engineering, Vol 109, Pg 48-67, 2017)

Robust Optimization Using a Methodology Based on Cross Entropy Methods

A Robust Optimization Model For Project Portfolio Selection With Information Uncertainty

Multiple kernel learning-aided robust optimization: Learning algorithm, computational tractability, and usage in multi-stage decision-making

Effective Budget of Uncertainty for Classes of Robust Optimization

Data-Driven Robust Optimization Based on Position-Regulated Support Vector Clustering

Doubly Robust Data-Driven Distributionally Robust Optimization