Model Predictive Control of a Wave Energy Converter Farm for Energy Optimization Using Each Devices Neighbors as Its Wave Forecast

Abstract

Optimal control of a wave energy converter (WEC) needs the wave excitation about one period ahead, which a single device cannot measure. We show in simulation that a WEC farm can supply this preview from the inertial measurement unit (IMU) and power take-off (PTO) force each device already carries. Heaving buoys are modeled in six degrees of freedom with the RM3 reference mooring resolved by a lumped-mass line model. A Fourier neural operator (FNO) recovers the incident elevation with a normalized error of 0.12 at a 2-s lag and the fairlead tension of the three mooring legs with 0.22. A least-squares fit to the estimated elevations of neighboring devices forecasts the elevation one peak period ahead with errors of 0.210.32 behind the front row, against 0.61.0 from a devices own record. A model-predictive controller driven by this forecast raises the electrical energy of a 12-WEC farmby 78% over passive damping tuned to the sea state, with the peak mooring tension unchanged in that sea; the front row, whose forecast is no better than its own record, gains 51%, nearly 60% of what the rows behind it gain. The heave PTO cannot reduce storm mooring loads, which surge and pitch drive. Fed with noisy true elevations, the farm forecast and an FNO tension model warn of all six exceedances of an illustrative limit in one storm record 2027 s ahead, with 14 of 25 alarms false. Over a year of measured sea states and wave directions at Humboldt Bay, with the estimator replaced by a noisy-truth stand-in and the mooring linearized, the controller yields 62% more annual energy for 12 WECs (64% for 50, modeled), which lowers the levelized cost of energy of the RM3 cost model by 3435% with the PTO cost doubled.