A new Swedish study shows how adaptive batteries powered by artificial intelligence can help solar greenhouses reduce electricity peaks and lower operating costs. The approach combines short-term forecasting, reinforcement learning and battery storage optimisation, offering potential applications well beyond agriculture.
Adaptive batteries for solar greenhouses
The combination of photovoltaics and energy storage is becoming increasingly important for controlled-environment agriculture. Researchers at Karlstad University have now developed a new intelligent energy management strategy using adaptive batteries to optimise electricity consumption in solar-powered greenhouses.
The project focuses on one of the most energy-intensive agricultural environments. Modern greenhouses rely on automated climate-control systems to regulate temperature, humidity, lighting and CO2 concentration. Depending on climate conditions and technology, greenhouse energy consumption can range from 140 GJ/ha to more than 7,000 GJ/ha annually, with electricity accounting for roughly half of total energy use.
Solar power already offers an attractive pathway for self-consumption, especially with the rise of semitransparent photovoltaic modules and suspended solar panels. However, fluctuating weather conditions and variable solar irradiance limit the effectiveness of standard PV integration. This is where adaptive battery systems become critical.
How the adaptive battery system works
The Swedish researchers designed an AI-based control architecture built around three core technologies:
- Variational Mode Decomposition (VMD)
- Long Short-Term Memory forecasting (LSTM)
- Deep Reinforcement Learning (DRL)
The system was connected to a 7.2 kWh battery paired with a bidirectional inverter and tested inside an experimental greenhouse equipped with environmental sensors, climate-control systems and a 7.1 kWp photovoltaic installation.
The researchers first used VMD to filter raw electricity consumption data, isolating the main energy demand patterns from sudden spikes caused by heating systems. The LSTM neural networks then predicted both electricity demand and solar generation 30 minutes in advance.
At the same time, the reinforcement learning agent continuously learned how to optimise battery charging and discharging cycles based on:
- battery state of charge (SoC)
- electricity prices
- short-term energy forecasts
Unlike traditional fixed-threshold systems, the adaptive model dynamically reacts to changing electricity prices and demand conditions.
Adaptive batteries reduce electricity peaks
The results showed that the adaptive battery strategy significantly outperformed conventional control systems.
Compared with a scenario without storage, the intelligent system reduced variable electricity costs by:
- 2.2% in February 2023
- 2.7% in March 2023
More importantly, the system sharply reduced peak electricity demand:
- 24% lower peak consumption in February
- 19% lower peak consumption in March
The study also found that total electricity costs were lower than in all comparison scenarios.
According to Jorge Solis, “The results demonstrate that the optimised energy management system could be applied on a broader scale, for example in manufacturing industries, where both production costs and carbon footprint reductions are required.”
Potential applications beyond agriculture
Although the research focused on solar greenhouses, the broader implications may be even more relevant for industry and commercial buildings. As electricity markets become increasingly volatile and peak demand charges rise across Europe, adaptive battery systems could become a strategic tool for:
- industrial energy optimisation
- peak shaving
- self-consumption enhancement
- carbon footprint reduction
- grid flexibility services
The study, titled Intelligent Control Strategy of a Battery Energy Storage for a Climate-Controlled Greenhouse with a High Proportion of Local Renewable Energy, was published in IEEJ Transactions on Electrical and Electronic Engineering in 2025.
