Terra Daily — May 6, 2026

Research Worth Reading

• High-Fidelity Full-Sky Video Prediction for Photovoltaic Ramp Event Forecasting — Introduces PhyDiffNet, a generative framework that predicts future sky video to feed a ramp-aware PV output model for ultra-short-term solar ramp forecasting. Directly relevant if you work with time-series prediction or video understanding — the hard part is modeling fast cloud transitions that destabilize grid frequency.
• Online Energy Management for Bidirectional EV Charging with Rooftop PV: An Aging-Aware MPC Approach — Builds an online model predictive control loop for vehicle-home-grid systems that explicitly models battery degradation. If you’ve tuned MPC controllers or worked on optimization under uncertainty, the formulation here — balancing V2G/V2H revenue against cycle-life cost — is a concrete systems problem.
• Towards accurate extreme event likelihoods from diffusion model climate emulators — Evaluates the cBottle diffusion model for generating physically consistent atmospheric states and improving extreme-event probability estimates in climate scenarios. Useful if you’re familiar with score-based generative models and want to understand how diffusion is being applied to conditional climate emulation rather than just image generation.
• Event-Based Dynamic Programming for Pumped-Storage Hydropower Scheduling — Reformulates pumped-storage scheduling as an event-based DP problem with mode-specific transitions, handling reservoir dynamics, ramp limits, and start-up costs. A clear example of how discrete-event modeling can outperform flat time-step approaches when the state space is sparse.

Open Source Projects

• Offshore wind firm that took Trump payout hits a milestone in Europe — Ocean Winds has commissioned a 30 MW floating wind array off southern France, one of the larger early-stage floating turbine deployments. Worth watching if you’re tracking floating foundation design, mooring system engineering, or how European permitting pathways differ from U.S. political constraints.

Today’s Synthesis

The PhyDiffNet paper shows how generating near-future sky video can improve ultra-short-term solar ramp forecasts — useful at the distribution level where sudden cloud cover shifts destabilize local grid frequency. Pair that with the aging-aware MPC for bidirectional EV charging with rooftop PV, and you have a concrete stack: predict the next few minutes of solar output variability, then run an optimization loop that decides when to charge, discharge, or feed the grid while accounting for battery wear. Ocean Winds’ 30 MW floating array in France is a reminder that the same variability problem scales up — offshore wind output also swings with wind shifts, and better short-horizon forecasts (whether from sky video or diffusion-based atmospheric emulators) directly improve dispatch decisions across storage, EVs, and grid assets. If you’re coming from ML or control systems, the open question is: how do you fuse these forecasts into a single real-time optimizer that handles multiple heterogeneous generation sources? That’s the systems integration problem worth prototyping with open datasets like NSRDB solar imagery and ENTSO-E wind profiles.