Korea Power Exchange plans AI-EMS to manage renewables variability
The Korea Power Exchange (KPX) is developing an AI-based next-generation power system operation system (AI-EMS) to expand AI's role in grid management, from demand forecasting to operational alternatives. This initiative, discussed at an October 22 forum, aims to address renewable energy variability by rapidly analyzing grid status and supporting controller decisions.
Our take
Will Korea's AI grid operation system change how battery storage and VPPs are dispatched?
Korea's AI grid operation system is unlikely to change battery storage or VPP dispatch before 2030, since it is an operator-support research programme and no dispatch or settlement rule change has been announced.
The system is built to support grid operators, with high-risk controls kept under their approval, so it changes how decisions are prepared, not the rules battery storage and VPPs are dispatched and paid under. What could matter later is how far KPX automates fast controls, which it has not set.
What the source says
- The AI-EMS development plan connects prediction, planning, interpretation, and control functions to comprehensively present operational alternatives for the power system.
- A core challenge for AI-EMS is cross-validation with existing interpretation technologies and setting control authority based on risk, as incorrect judgments could lead to blackouts or equipment damage.
- Related research for the AI-EMS is planned from April 2026 until December 2030.
- Kim Tae-eon Kim Tae-eon, KPX AI Innovation Office Gyeongin IT Development Center Director, stated that the initial goal for AI-EMS is 'partial autonomy,' where low-risk functions are automated and high-risk controls require operator approval.
- Kim Tae-eon Kim Tae-eon also stated that the system aims to provide the best solutions through comprehensive evaluation, rather than a simple sum of unit functions, by validating results in a digital twin before providing them to operators.
- Park Rae-jun Park Rae-jun, KEPRI Power System Research Institute Senior Researcher, explained that AI surrogate models aim to quickly estimate results under new conditions by learning from existing simulation data, allowing for broader examination of operating conditions.
- Anish Gaikwad Anish Gaikwad, EPRI Deputy Director, emphasized the 'human in the loop' development principle for the Raphson tool, clarifying it reduces simulation time but does not make decisions for transmission planners.
- The AI-EMS will combine AI with existing interpretation technologies, where AI provides an initial solution for generation planning, and an existing optimization engine refines it to reduce calculation time.
- A Small Language Model (SLM) is envisioned as an interface to connect operators with analysis functions, calling necessary analysis and explaining results based on past operational history.
- AI surrogate models showed errors in low-voltage areas during pre-validation, attributed to a lack of training data for those specific conditions.
For investors — VPP operators and aggregators should base Korea plans on current dispatch rules and watch whether KPX's control-authority framework, once set, covers aggregated resources.
Merchant storage investors and optimisers should assign no revenue to this programme, which names no new service or payment, and treat it as a research effort running to 2030.
Watch: KPX's announcement of the final framework for AI-EMS control authority and risk allocation.
Weekly digest
What changed in Korea's power market this week — one email, every week.
Email only, used solely to send this digest (delivered via Resend). Unconfirmed sign-ups are deleted after 30 days. Free.