White space is where the patents aren't, and the most interesting gaps sit between two clusters that are each filling fast but haven't merged. In H02J — the grid-management class — two such clusters are visible right now, and the space between them is conspicuously thin.
Cluster one: grid-forming converter control. Patents like Dynapower's US10153688B2 (multiple converters starting in grid-forming mode, CPC H02J 3/381) and ABB's US10044192B2 (microgrid resynchronization, H02J 3/383) define how inverter-based resources establish and coordinate a grid reference. This is classical control theory applied to power electronics — deterministic methods, explicitly designed loops.
Cluster two: learning-agent grid control. Grants like the Qatar Foundation's US12651906B2 (deep-RL demand response, H02J 3/17) and Banpu's US12646943B2 (ML-optimized VPP controller, H02J 3/004) put reinforcement-learning agents in charge of power decisions — but at the edge: loads, EV fleets, demand response. The learning is applied to scheduling and dispatch, not to the fast grid-forming control loop.
Here's the white space: the explicit combination — a reinforcement-learning agent performing grid-forming control, setting voltage and frequency reference adaptively rather than with fixed loops — is largely absent from the granted record. The two clusters share a CPC neighborhood (H02J 3) and an obvious technical adjacency, yet the patents that bridge them are scarce. Learning is at the edge; grid-forming is at the core; the bridge is open.
Why flag a gap rather than a grant? Because white space is where the next filings, and the next disputes, will land. The qualifier this column always attaches: absence in the record can mean nobody's solved it, the work is still pre-grant, or the combination sits in publications we haven't surveyed — a gap is a hypothesis, not a proof. But the structural logic is strong: as inverter-based resources take over grid-forming duty and learning controllers prove out at the edge, the incentive to claim a learning grid-forming controller grows. This desk will be watching H02J 3/381 for the first grants that cite the RL cluster.