snehitha-tadapaneni/Hourly-Power-Consumption-Forecasting-Morocco-

Time Series Analysis for Forecasting # Hourly Power Consumption Forecasting – Morocco ## Project Overview This project focuses on forecasting **hourly electricity consumption in Morocco** using classical and stochastic time series models. The goal is to understand consumption patterns, compare multiple forecasting approaches, and identify the most accurate model for short-term demand prediction. The analysis uses hourly data from **January 1, 2017 to December 30, 2017 (8,736 observations)** and evaluates deterministic models, exponential smoothing, regression-based approaches, and stochastic time series models. 📄 All methodology, diagnostics, and results are documented in the final project report. --- ## Course Information - **Course:** Time Series Analysis for Forecasting (DNSC 6319) - **Institution:** The George Washington University - **Instructor:** Prof. Refik Soyer --- ## Contributors - Deepankar Makwana - **Snehitha Tadapaneni** - Carissa Paul - Vansh Kumar --- ## 📊 Dataset Description - **Target Variable:** Total hourly power consumption - **Time Period:** Jan 1, 2017 – Dec 30, 2017 - **Frequency:** Hourly - **Total Observations:** 8,736 ### Exogenous Variables - Temperature (°C) - Humidity (%) - Wind Speed (m/s) - General Diffuse Flow - Diffuse Flow ### Train–Test Split - **Training set:** First 7,500 observations - **Test set:** Final 1,236 observations --- ## 🔍 Exploratory Data Analysis - Strong **daily seasonality (24-hour cycle)** and long-term trends - Non-stationarity confirmed via ACF and PACF analysis - Hourly, daily, and monthly boxplots reveal structured consumption behavior - Additive decomposition separates trend, seasonal, and residual components --- ## Models Implemented ### 1️. Deterministic Time Series Models - **Seasonal Dummy + Trend Model** - Hour-of-day dummy variables with segmented time trends - Captures daily usage patterns effectively - **Cyclical Trend Model** - Harmonic terms derived from periodogram analysis - Models dominant frequencies in the data …