Tittle: Estimate Energy Consumption in London
Language: Python
Library: Tensorflow,keras,scikit-learning
Description:
The aim is to estimate energy consumption for buildings using various models of machine learning. As for now models that I already developed and the performance are good using R-squared such as GradientBoosting,Random Forest,SVR,MLR. In future, I plan to apply DNN. The input is 7 years of data set from 2017-2023 for domestic buildings where each residential has a yearly data with other categorical variables (1.59 million datasets) such as type of heating system,windows,floor and roof that may affect the consumption of energy in a building.
In contrast, I tried another time series models such as RNN, CNN with combination of DNN and LSTM where the input in univariate series the energy itself. The input data is 8 years 2016 until 2023 (2789 datasets) Here are my questions:
- If I want to use multivariate series I need to use continuous variables such as temperature and humidity along with the energy, right?
- The time series models only learn historic data to forecast future data , am I right?
- After go through short course Sequence,time series models (C4) , I am still not clear how to modify my window data set to forecast for the next 1 month for instance. Below is my current definition for window data set:
Split_Time = 2789
Window_Size = 64
Batch_Size = 32
Shuffle = 1000
however, my architecture of model is incompatible and return my mse and mae as nan. Please guide