Delayed Investment Decisions in Renewable Energy under Uncertainty: A Deep Learning–Based Approach

We investigate a stochastic control problem for renewable energy capacity installation under uncertainty and implementation delay. Investment decisions are irreversible and subject to time-to-build constraints such as construction, regulatory approval, and grid integration. Electricity demand uncertainty and renewable intermittency are modeled through jump-driven stochastic dynamics, capturing both continuous fluctuations and rare extreme events. The introduction of delay induces path dependence and leads to a non-Markovian control problem. To address this challenge, we propose a deep learning-based global control framework that directly approximates optimal feedback policies from simulated trajectories. Unlike dynamic programming or BSDE-based methods, the approach avoids value function approximation and remains tractable in high-dimensional and delayed settings. Numerical experiments show that delay significantly alters optimal investment timing and induces smoother, anticipative strategies.