How algorithms are boosting pumped storage hydropower

At first glance it may seem wasteful: using electricity to pump water uphill, only to release it later to generate electricity again — albeit with some energy loss. Yet with modern automation and smarter scheduling, pumped storage hydropower has become one of the most versatile assets in a renewable heavy energy system.

Vattenfall owns and operates six pumped hydropower plants in Germany. For many years the operating schedule was straightforward: pump at night when electricity prices were low and generate during the day when they were higher.

Today the system is far more complex. Coal and nuclear capacity has been phased out (nuclear entirely), replaced mainly by wind and solar. Output now varies with the weather, placing greater demands on pumped hydro stations.

A new plan every 15 minutes

Jörg Seidel, who heads the intraday trading and optimisation for Vattenfall’s continental operations, explains how this plays out:

“Today, and especially in summer, solar pushes prices very low around midday, sometimes the prices are even negative. This means we often have two cycles per day: we pump at night, generate in the morning, pump again at noon, and then we generate again in the evening. Adding to that, turbining is often partial, you may want 70 per cent output on a certain time of day, 80 percent the next moment, and so on."

“These small adjustments of the output happen constantly across all the assets and are done every 15 minutes, meaning 96 times per day. Doing that planning and calculation manually would be totally unmanageable.”

The solution is algorithms. They consider all relevant factors — market prices, weather forecasts, reservoir levels, the condition of pumps and turbines, and technical constraints — and recalculate optimal plans for every plant roughly every minute.

The first version, introduced in 2018, was simple: it could only buy or sell the output of a turbine already running. Later, pumps, start–stop decisions and mode switching were added. Today, nearly all operational decisions in the German hydropower portfolio are made autonomously, with the algorithms also placing bids for the next 15 minute market interval.

Huge provider of flexibility

Vattenfall’s German pumped storage hydropower stations form a powerful six pack of assets. Together they deliver 2,500 megawatts — comparable to two large nuclear plants — but with the ability to ramp output up or down within seconds to keep the grid balanced. They act like giant batteries, able to supply power for up to ten hours before needing to recharge.

“Pumped storage hydro is one of the biggest flexibility providers to the German electricity grid,“ says Brit Gericke, Head of Algorithms and Optimisation, who leads a team of 13 specialists responsible for designing the algorithms.

“With these algorithms, we can provide flexibility in every 15-minute period of the day, compared to only a few times per day when planning was made manually.”

The algos make it possible for the plants to deliver both active power and ancillary services — the supporting tools that help stabilise the grid — with far greater precision. The result: a more stable, better balanced power system.

And what about people — are these systems making operators redundant? Gericke disagrees:

“Automation does not remove the need for humans but changes their roles. Operators must understand what the algorithm is doing, why it makes certain decisions, and detect when something is wrong. The complexity has increased enormously and this level of automation is not possible without human oversight.”