After another break, through personal problems, I have caught up with Mike again. He seems reasonably happy with the dashboard currently but is concerned about the accuracy. This matches my own concern so I am glad to be working on the data side of things again.
One of the things that I have been wanting to do for a while and is now done is automate the download of data from the Keele DEOPS system using the API. This took quite a bit of experimentation and communication with Siemens, who run the DEOPS system, and finally found there was a configuration setting at their end. This means I can now daily download data from Keele rather than manually having to log into the DEOPS system and download a CSV which will contribute to greater real-time accuracy.
I had a meeting with someone else who works for Mike who has successfully split the build and predict parts of the model. We spent time looking over her code and mine and came to the conclusion that while it was the right choice for her architecture (where that data that makes the base model is unlikely to change), for my use-case where solar panels degrade with time, and environmental issues could affect the production of solar power, that utilising the real-time instant ML functionality was more appropriate. I have discussed this with Mike, and he is in agreement.