{"id":541427,"date":"2026-06-22T13:44:03","date_gmt":"2026-06-22T11:44:03","guid":{"rendered":"https:\/\/silicon-saxony.de\/fraunhofer-ipa-ai-based-electricity-price-forecast-reduces-costs-by-up-to-17-percent\/"},"modified":"2026-06-22T13:43:47","modified_gmt":"2026-06-22T11:43:47","slug":"fraunhofer-ipa-ai-based-electricity-price-forecast-reduces-costs-by-up-to-17-percent","status":"publish","type":"post","link":"https:\/\/silicon-saxony.de\/en\/fraunhofer-ipa-ai-based-electricity-price-forecast-reduces-costs-by-up-to-17-percent\/","title":{"rendered":"Fraunhofer IPA: AI-Based Electricity Price Forecast Reduces Costs by Up to 17 Percent"},"content":{"rendered":"
\n
\n

<\/p>\n

For industrial companies, the price of electricity is increasingly becoming an operational control variable. Companies that operate battery storage systems, charging infrastructure, heating and cooling systems, or other flexible consumers can, in principle, shift loads to times when electricity prices are low. In practice, however, this is challenging: The favorable time windows must be identified well in advance, technical limitations of the systems must be adhered to, and the decision must be reliable enough to avoid ultimately incurring higher rather than lower costs.<\/p>\n

This is precisely where researchers at Fraunhofer IPA come in. As part of the Kopernikus project \u201cSynErgie,\u201d they have developed an AI-based electricity price forecast and, through extensive validation, investigated how well it is suited for controlling flexible loads. A commercially available forecasting service for the next day\u2019s electricity prices served as a comparison. More than 20,000 time points were analyzed over the period from October 2025 to May 2026.<\/p>\n

\u201cFor companies, it\u2019s not the smoothest forecast curve that counts, but the operational benefits. If a flexible load, a storage system, or a charging infrastructure is running at the wrong time, it results in real additional costs. That\u2019s why we evaluate not only statistical forecasting errors but also whether a model reliably identifies the low-cost and high-cost time windows,\u201d says Can Kaymakci, head of the Data-Driven Energy System Optimization research team at the Fraunhofer Institute for Manufacturing Engineering and Automation IPA. \u201cIn the end, it comes down to one simple question: Is the system being operated during the right time window or not? That is precisely what determines whether dynamic electricity prices truly provide an economic advantage.\u201d<\/p>\n

Simulation Shows Concrete Cost Benefits<\/h4>\n

The researchers investigated the economic impact of these differences in a simulation. To do this, a flexible load was scheduled once a day during the forecasted cheapest time window. Various operating durations\u2014ranging from 15 minutes to several hours\u2014were considered, focusing exclusively on procurement costs on the electricity market.<\/p>\n

Across all scenarios considered, the Fraunhofer IPA\u2019s forecast resulted in lower costs. Depending on the duration of use, electricity procurement costs fell by 7.95 to 16.96 percent. For every megawatt of flexibly controllable capacity, the Fraunhofer IPA\u2019s electricity price forecast in this scenario thus resulted in approximately 3,000 euros less in additional annual costs attributable to forecasting errors. The simulation thus makes it clear: Forecast quality is not an abstract model value, but rather determines whether flexible loads can actually be operated more cost-effectively.<\/p>\n

\u201cPoor forecasts are not just a technical problem. They directly cost money,\u201d says Kaymakci. \u201cThis is especially true for companies that will rely more heavily in the future on dynamic electricity prices, flexible consumers, and automated energy management systems. Anyone who wants to tap into this potential needs not only data but also a robust decision-making framework.\u201d<\/p>\n

From Price Forecasting to Industrial Operation<\/h4>\n

The results build on earlier work by Fraunhofer IPA on forecasting industrial energy systems. That research had already shown that standardized AI models are often insufficient for industrial load forecasting because production processes, shift schedules, plant conditions, and external factors such as weather or calendar data must be specifically taken into account. The same applies to electricity price forecasting: the algorithm alone is not the real lever. Rather, a combination of forecasting, plant constraints, economic evaluation, and operational implementation is crucial.<\/p>\n

This opens up a concrete path to application for companies:<\/b><\/p>\n