The topic of this project is: "Long-Term Time-Series forecasting, using model identification theory".
Starting from a dataset, which concerns two years gas consumption trend in Italy in function of two parameters (day of the year, day of the week), we want to identify the model which represents our data in the better way. Afterwards, we created a function that predicts the gas consumption value, using the best identified model, once you insert a tuple composed by the two parameters mentioned before.
In order to identify the best model that represents our time series process we used: polynomial regression, neural networks and finally harmonic regression.
If you're interested in how we solve model identification, comparison between models, gas consumption prediction, and many other things ( using Matlab ), watch the "live_script.mlx" code. You can find it at this folder.
It is also obtainable, at the same link, another file including solely the prediction function, using the best model (for more details, see the "prediz.m" script).
At this link you can find the project presentation, in different extensions (pdf or pptx).
We suggest you downloading .pptx file, because 3D animation videos we putted in our presentation are not available if you download the .pdf version.