Apogène | Olivier Juan

Apogène is EDF’s short-term unit commitment software, used to schedule a fleet of power plants over a short horizon (2-3 days ahead) to meet electricity demand at minimum cost, subject to complex operational constraints (ramp rates, minimum up/down times, fuel limits, reserve requirements). At EDF’s scale, even marginal efficiency gains translate to tens of millions of euros annually.

Today, Apogène is deployed and used at 3 different time horizons: Weekly (4 weeks), daily (2-3 days), and intra-day (24 hours). Apogène optimizes all of EDF’s production assets: 57 nuclear units, 4 Combined Cycle Gas plants, 13 Combustion Turbines, and 450 hydraulic generation units (lock-based and run-of-river). It also takes into account the different electricity markets. Recently, the group’s stationary batteries have also been optimized by Apogène.

My role

In 2012, I led a complete ground-up redesign of Apogène, driving the full software lifecycle:

Mathematical modeling of the unit commitment problem as a MILP
Algorithm design: branch-and-bound, cutting planes, decomposition methods, ADMM, Resource Constrained Shortest Path (RCSP), Min Cost Flow
Software architecture and C++ implementation
Multi-solver integration: CPLEX, GUROBI, SCIP, FICO Xpress, COIN-OR, HiGHS
Production infrastructure: HPC / Slurm / Singularity
CI/CD pipeline (GitLab), Docker, testing strategy, and release management
Technical leadership of a team of 10–25 engineers and researchers

Timeline

Year	Milestone
2008–2012	Early contributions to Apogène V1; parallel work on LNG scheduling, maintenance planning, emission optimization
2012	Led the ground-up redesign of the mathematical model and solver architecture
2018	V2 deployed in production — significant reduction in operating costs vs. V1
2020	V3 deployed — improved decomposition, RCSPP integration, Complete C++ migration.
2021	V4 deployed — improved performances, asymmetrical ancillary services integration
2022	IndusRO’2022 prize awarded by ROADEF for industrial impact
2024	Intraday Deployment - Adaptation for the intraday unit-commitment optimization.
2025	V5 deployed - improved decomposition algorithm, performance improvement for timestep refinement
2025	Weekly Deployment - Adaptation for the weekly unit-commitment optimization.
2025-	Batteries integration, ongoing integration of ML-based branching heuristics (RL, GNN)

Impact

Savings of several tens of millions of euros per major release, and a few tens of millions per year through intermediate improvements
4 production versions deployed over 10+ years
covers 3 different time horizons and rationalizes optimization toolchain into a unified platform
Team of 10–25 engineers and researchers led across the full lifecycle
Awarded the IndusRO’2022 prize by the French Operational Research Society (ROADEF)

Short video (2 min)

Full ROADEF presentation

Technical stack

Optimization & algorithms: MILP · Branch-and-bound · Cutting planes · Decomposition methods · ADMM · Resource Constrained Shortest Path Problem (RCSPP) · Min Cost Flow · Heuristics · Genetic Algorithm, Interior point method, Bundle method
Solvers: CPLEX · Gurobi · SCIP · FICO Xpress · COIN-OR · HiGHS
Languages: C++ · Python
Engineering & infrastructure: Conan · TBB · GitLab CI/CD · Docker · HPC / Slurm · Singularity
ML (ongoing integration): PyTorch · PyG · RLLib · Reinforcement learning · Graph Neural Networks

The ongoing integration of ML into Apogène’s solver loop is the subject of active research. Key results:

PlanB&B (AAAI 2026) (Strang et al., 2026) — model-based RL agent for branching in branch-and-bound, outperforming prior state-of-the-art RL methods on four standard MILP benchmarks.
BBMDP (NeurIPS 2025) (Strang et al., 2025) — principled MDP formulation for variable selection in B&B, enabling the use of standard RL algorithms for learning optimal branching policies.
FMSTS (CPAIOR 2020) (Etheve et al., 2020) — first use of RL to fully optimize a branching strategy from scratch.

References

2026

Proceedings

Planning in Branch-and-Bound: Model-Based Reinforcement Learning for Exact Combinatorial Optimization

Paul Strang, Zacharie Alès, Côme Bissuel, Olivier Juan, Safia Kedad-Sidhoum, and Emmanuel Rachelson

In Association for the Advancement of Artificial Intelligence (AAAI), 2026

Abs arXiv Code Website

Mixed-Integer Linear Programming (MILP) lies at the core of many real-world combinatorial optimization (CO) problems, traditionally solved by branch-and-bound (B&B). A key driver influencing B&B solvers efficiency is the variable selection heuristic that guides branching decisions. Looking to move beyond static, hand-crafted heuristics, recent work has explored adapting traditional reinforcement learning (RL) algorithms to the B&B setting, aiming to learn branching strategies tailored to specific MILP distributions. In parallel, RL agents have achieved remarkable success in board games, a very specific type of combinatorial problems, by leveraging environment simulators to plan via Monte Carlo Tree Search (MCTS). Building on these developments, we introduce Plan-and-Branch-and-Bound (PlanB&B), a model-based reinforcement learning (MBRL) agent that leverages a learned internal model of the B&B dynamics to discover improved branching strategies. Computational experiments empirically validate our approach, with our MBRL branching agent outperforming previous state-of-the-art RL methods across four standard MILP benchmarks.

2025

Proceedings

A Markov Decision Process for Variable Selection in Branch & Bound

Paul Strang, Zacharie Ales, Côme Bissuel, Olivier Juan, Safia Kedad-Sidhoum, and Emmanuel Rachelson

In Neural Information Processing Systems (NeurIPS), 2025

Abs arXiv Code Website

Mixed-Integer Linear Programming (MILP) is a powerful framework used to address a wide range of NP-hard combinatorial optimization problems, often solved by Branch and Bound (B&B). A key factor influencing the performance of B&B solvers is the variable selection heuristic governing branching decisions. Recent contributions have sought to adapt reinforcement learning (RL) algorithms to the B&B setting to learn optimal branching policies, through Markov Decision Processes (MDP) inspired formulations, and ad hoc convergence theorems and algorithms. In this work, we introduce BBMDP, a principled vanilla MDP formulation for variable selection in B&B, allowing to leverage a broad range of RL algorithms for the purpose of learning optimal B&B heuristics. Computational experiments validate our model empirically, as our branching agent outperforms prior state-of-the-art RL agents on four standard MILP benchmarks.

2020

Proceedings

Reinforcement learning for variable selection in a branch and bound algorithm

Marc Etheve, Zacharie Alès, Côme Bissuel, Olivier Juan, and Safia Kedad-Sidhoum

In International conference on integration of constraint programming, artificial intelligence, and operations research (CPAIOR), 2020

Abs DOI arXiv HAL Website

Mixed integer linear programs are commonly solved by Branch and Bound algorithms. A key factor of the efficiency of the most successful commercial solvers is their fine-tuned heuristics. In this paper, we leverage patterns in real-world instances to learn from scratch a new branching strategy optimised for a given problem and compare it with a commercial solver. We propose FMSTS, a novel Reinforcement Learning approach specifically designed for this task. The strength of our method lies in the consistency between a local value function and a global metric of interest. In addition, we provide insights for adapting known RL techniques to the Branch and Bound setting, and present a new neural network architecture inspired from the literature. To our knowledge, it is the first time Reinforcement Learning has been used to fully optimise the branching strategy. Computational experiments show that our method is appropriate and able to generalise well to new instances.