Treffer: Prediction of optimal binding constraints of linear programming problem using deep learning

This thesis presents innovative deep learning models designed to enhance the efficiency of solving linear programming, transportation, and traveling salesman problems. The first model, LP-Net, predicts the binding optimal constraints in linear programming problems, achieving faster solution times compared to traditional solvers like CPLEX. The second model, Transport-Net, predicts optimal basic feasible arcs for transportation problems, demonstrating significant speedups and superior scalability over both CPLEX and Gurobi in large-scale instances. The third model, TSP-Net, addresses the traveling salesman problem by predicting an initial near-optimal tour, outperforming Gurobi and genetic algorithms in computational efficiency. Collectively, these deep learning approaches offer transformative potential for accelerating solutions to fundamental optimization problems.