How AI Recovers Lost Data in Solar Plants (Up to 70%)
Anyone who operates a photovoltaic (PV) plant knows the problem: production data goes missing. Sensors fail, communications drop, inverters pause for maintenance, and you are left with incomplete data series that make both maintenance and generation forecasting harder, and more expensive.
We spent a long time on this problem at Smarkia, and the result, a deep-learning model for reconstructing lost solar production data, was published in Expert Systems With Applications, a peer-reviewed international journal. Here is what it does, in plain terms.
What deep learning is, in two sentences
Deep learning is a branch of machine learning that uses multi-layer neural networks to find complex patterns in data. It is the technology behind uses as different as medical imaging and traffic forecasting, and it applies just as well to reconstructing incomplete energy data series.
Why solar production data goes missing
The causes vary: communication failures, inverter shutdowns, measurement noise, weather events, sensor problems. Industry studies suggest that once a series is missing more than 5% of its data it already needs treatment before it can be analyzed reliably, and in real solar production series, losses of up to 40% are not unusual.
This is not a minor issue. Missing data undermines operations and maintenance, making predictive maintenance harder, and it weakens the forecasts needed to integrate solar into the grid.
What Smarkia's model brings
The difference from earlier approaches is that the model can be trained on a single production series, even an incomplete one, with no prior history and no correlated external signals such as irradiance or temperature. That makes it especially useful at sites with no long historical record and no weather sensors of their own.
In the study, the model held up well even with series missing up to 70% of their data, with the best results around 50% loss, outperforming previous methods.
Why this matters if you operate solar plants
Beyond the research, this ability to rebuild data when sensors fail is now part of how Smarkia manages solar assets in production, something Verdantix noted explicitly in its Smart Innovators: Energy Management Software (2025) report, describing how providers like Smarkia apply deep learning at the grid edge to reconstruct missing solar generation data and keep visibility of a site even when its sensors fail.
In practice that means fewer gaps in the data, better generation forecasts, and less time lost interpreting broken series, whether the plant is in Spain or the United States.
For the full technical detail, the paper is publicly available in Expert Systems With Applications.