Predicting solar energy production on micro scale with Watson

Predicting the future isn’t easy, but new technologies such as Watson and the availability of accurate data are a great help to do so. I mentioned in previous blog posts already my progress in connecting the grid inverter from the solar panels with Watson IoT. I also tapped into the weather API to know the historical weather observations and the weather forecast. In total I collect following four data sources:

- The solar energy production is continuously measured, but I derive hourly data of the amount of kwh produced.

- From the weather API I receive the weather observations historical data (hourly data) for the GPS coordinates of the solar panels.

- From the weather API I request the next day weather forecast (hourly data) for the GPS coordinates of the solar panels.

- From the internet I connected to a service that provides the daily sunset and sunrise data for the GPS coordinates of the solar panels.

I did a lot of cleaning on the measurement data and finally consolidated the measurement data hour by hour. I used the historical data to create a predictive model. I could have calculated a theoretical model based on latitude, longitude, azimuth, temperature, wind speed, etc. but the list of parameters would probably be endless. Also, the objective of the test was to validate the use of predictive modelling and working with real data. The first phase uses the data prior to September 2017 to build the model. Thereafter I used that model daily to predict the next day energy production. The graphs below depict the measured data versus the predicted data. The first graph cover the entire month of September, the second graph is a snapshot of four days. It’s quite remarkable to see the same shapes of the curves.
NodeRed data collection
NodeRed data collection
The Pearson correlation factor is 95,50 %
If you want to browse the graph, feel free and go to
You might ask why this is important? Accurate prediction of energy production enables optimization in multiple ways; the manageable (postponeable) part of our energy consumption can be aligned with the forecasted energy production. This will bring

- Reduce grid usage (and grid charges).

- Optimize charging of local batteries.

- - Reduce losses.

- ....