The program created in this thesis is located in the middle of a complete workflow, so it can mainly focus on forecasting. As such, it receives the data after some amount of preprocessing, but this pretty much only means the data is formatted beforehand, but it’s still incomplete. For this reason, an important part of my thesis is improving the quality of the data, or finding an alternative way to forecast which can be used on incomplete data. I tried multiple models for forecasting, these are the following:
• A regressive model (Due to the weak performance of the method, I decided to make some improvements on its output data, there is a chapter dedicated to this, after which we will refer to the model and the applied improvements as “regressive model” together.)
• A neural network
• A decision tree
• And a random forest.
These are the models evaluated and compared in the thesis based on the average difference of their forecasted data and the expected forecast which is the actual consumption, and their runtime.