This presentation proposes a novel algorithm for solving the multi-year Transmission Network Expansion Planning (TNEP) problem, that combines the use of ML, together with Column Generation (CG) algorithm. Using the Column Generation technique, the TNEP problem is decomposed into a Master Problem, which is computationally inexpensive to solve, and several subproblems, one for each year, which are of MILP nature. To increase convergence of the CG algorithm, we solve a linearly relaxed version (LP) of the subproblems and trained a binary classifier using random forest for estimating the optimal values of the investment (binary) variables. Input data of the classifier are the results of the linearized version of the subproblems, which are characterized into key features with physical meaning. Then, the subproblems are optimized again, but fixing the values of the investment variables according to the output of the classifier, i.e., by resolving an LP formulation of the subproblems. This strategy allows solving the subproblems much faster than its MILP counterpart while providing, at the same time, a feasible solution, which is fed into the master problem. One key advantage of this strategy is that, upon convergence, the optimality of the solution is guaranteed. This proposal was tested in a case study based on the HRP 38-bus test system for different scenarios. The results showed that the binary classifier reached an accuracy above 96\% for estimating the value of the binary variables, thus providing good-quality solutions of the subproblems. This allowed, in turn, reduction in the computational time required for solving the TNEP problem in around 53\%, which shows the great benefits of the proposed approach.

In recent years, electric vehicles (EVs) have emerged as a preferred mode of transportation due to their advantages over traditional fuel-based vehicles, particularly concerning environmental impact. However, despite their benefits, EV adoption faces limitations. Among those limitations, this work focuses on driving range uncertainty stemming from various exogenous and endogenous factors impacting energy consumption. To address this challenge, this talk introduces a two-stage robust optimization framework tailored for the electric vehicle routing problem. The proposed approach proposes a column-and-constraint generation framework incorporating variable neighborhood search and alternating direction techniques, to tackle the proposed optimization model. Computational experiments demonstrate the economic efficiency and robustness of the proposed framework.

We study a general class of stochastic capacitated facility location problems arising in the design and reconfiguration of e-commerce supply chains. It considers probabilistic service level constraints to ensure the stochastic delivery times of customers' orders, defined as the sum of waiting time (including service time) at the processing facility plus shipping time to reach the customer location, is within a prescribed time limit with a probability greater or equal to a threshold value. We provide alternative polyhedral representations of these highly nonlinear and non-convex probabilistic constraints, and develop an exact branch-and-cut algorithm for solving the resulting reformulations. The proposed algorithm is enhanced with several algorithmic refinements to accelerate its convergence. We perform extensive computational experiments based on real location data from the United States to evaluate the performance of our algorithm and to provide insights on the optimal structure of solution networks. Results obtained on large-scale instances with up to 2,500 customers and 225 potential facilities under different service level scenarios confirm the effectiveness of the proposed algorithm.