One of the most common—and most costly—problems in operating solar PV plants is production data loss. Sensors that fail, communications that drop, inverters down for maintenance. The result is always the same: incomplete data series that make both plant maintenance and forecasting how much energy will be generated more difficult.

At Smarkia, we've been researching this problem for some time, and the result of that work—a Deep Learning-based model—was published in Expert Systems With Applications, a leading international scientific journal. Here's a simple breakdown of what it's about.

What Deep Learning is, in a nutshell

Deep Learning is a branch of machine learning that uses multi-layered neural networks to identify complex patterns in data. It's the technology behind applications as different as image-based diagnostics and traffic forecasting, and it can also be applied to reconstructing incomplete energy data series.

Why solar production data gets lost

The causes vary: communication failures, inverter shutdowns, measurement noise, weather events, or sensor issues. According to various industry studies, series with more than 5% missing data already require some kind of treatment before they can be reliably analyzed, and in practice it's not unusual to find gaps of up to 40% in real solar production series.

This lack of data isn't a minor issue: it affects both plant operation and maintenance, making predictive maintenance harder, and the forecasts needed to properly integrate solar energy into the power grid.

What Smarkia's model brings to the table

The difference compared to previous solutions is that the model can be trained on a single production data series, even an incomplete one, without needing prior historical records or correlated external signals such as solar radiation or temperature. This makes it especially useful in facilities that don't have long historical series or their own weather sensors.

In the study's tests, the model maintains strong performance even with series that have up to 70% missing data, with the best results around 50% data loss, outperforming previous approaches.

Why this matters for solar plant operators

Beyond its academic component, this ability to reconstruct data in the face of sensor failures is now part of how Smarkia manages solar assets in production—something Verdantix has explicitly recognized in its Smart Innovators: Energy Management Software (2025) report, noting that providers like Smarkia apply deep learning at the edge of the grid to reconstruct missing solar generation data and maintain visibility into the facility even when sensors fail.

In practice, this translates into fewer gaps in the data, better generation forecasts, and less time wasted interpreting incomplete series, whether you're managing a solar plant in Spain or in the United States.

If you'd like to review the full technical details, the paper is publicly available in Expert Systems With Applications.