1) As I cannot share all of the data I put a small portion of it in the Jupyter Notebook in my github.

Step 1: Set up the sensors at the incoming power node and in the rooms. It is a serial based sensor so, tie it to a software serial and push the data to cloud using an ESP32.

Step 2: Download the feed from the cloud platform to use the data to train your AI

Step 3: Clean the data by filling NAN's and other improper data. 

Step 4: Set all the elements in the data to float64 as the computations are mostly floating point and are very small so we need good precision.

Step 5 : Find the optimal values

Step 6 : Train your ANN

Step 7 : Deploy your application

Step 8 : After a month deploy a LSTM trained on the collected data for the entire duration and deploy your LSTM.

This is a brief description of the project.