Planning multilayer networks under deep uncertainties: an approach based on flexibility in engineering design

Multiperiod planning of multilayer transport networks under deep traffic uncertainties poses significant challenges for network operators. Conventional planning approaches rely on static forecasts and short-term optimizations that fail to capture the long-term economic trade-offs under highly uncertain demand evolutions. To solve this problem, this paper applies the concept of flexibility in engineering design to optimize flexible decision rules (FDRs) for adaptive capacity planning. Multiperiod planning is modeled as a policy-learning problem, where investment and capacity expansion decisions are dynamically adjusted based on realized traffic history. The FDRs are learned using deep reinforcement learning (DRL), enabling the optimization of long-term economic performance under stochastic demand scenarios. The approach is assessed for selected OTN-over-WDM network implementation variants using the expected net present value (ENPV) as a performance metric. Compared to conventional planning baselines, the learned FDRs improve the ENPV by a factor of 9.5% and 472.9% in the worst and best cases, respectively.