Comparative Time Series Forecasting for Electricity Consumption

Overview

This repository hosts a comparative analysis of five distinct methodologies for time series forecasting, applied to historical monthly electricity consumption data (data/27.csv). The project's objective is to evaluate model stability and predictive accuracy to produce a robust 7-year (84-month) out-of-sample forecast.

Repository Structure

The code and data are organized into the following directories:

• code/: Contains all Python source code (.py files) for data utilities, modeling, and reporting.
• results/: Reserved for all generated output files (forecast plots and CSVs).
• EDA/: Contains additional plots for time series analysis.

Models and Techniques Implemented

The project compares model performance using Un-scaled Root Mean Squared Error (RMSE) on a fixed 24-month test set.

1. Classical Statistical Models:
   • sarima.py: Seasonal AutoRegressive Integrated Moving Average (SARIMA) for mean forecasting.
   • garch.py: Generalized AutoRegressive Conditional Heteroskedasticity (GARCH) for volatility modeling of SARIMA residuals.
2. Statistical Smoothing:
   • ets.py: Exponential Smoothing (Holt-Winters) for recursive forecasting.
3. Decomposition:
   • FBprophet.py: Prophet model for robust trend and seasonality capture.
4. Deep Learning:
   • lstm.py: Long Short-Term Memory (LSTM) network optimized for capturing non-linear seasonality.

Setup and Execution

Prerequisites

Python 3.8+ and the libraries listed in requirements.txt.

Installation

pip install -r requirements.txt