Time Series Analysis & Forecasting of Energy Usage Using RNN (PyTorch)

This project focuses on analyzing and forecasting the daily peak power demand of Maharashtra from 01-Apr-2024 to 21-Sep-2025. The dataset is sourced from the official Government of India open data portal:

Dataset: Daily Peak Demand (MW) – Power

Source: https://www.data.gov.in/resource/daily-peak-demand-mw-power-01-apr-2024-21-sep-2025

Using PyTorch, a Recurrent Neural Network (RNN) model is built, trained, evaluated, and used to forecast future peak energy demand.

Project Workflow

1. Exploratory Data Analysis (EDA)

Loaded the dataset and converted date columns into proper datetime format.

Visualized the demand trend over the full time period.

Checked for:

Seasonality patterns

Cyclic behavior

Trend shifts

Anomalies or missing values

2. Data Preprocessing

Handled missing and inconsistent values (if any).

Scaled the demand values using MinMaxScaler to ensure stable neural-network training.

Created input sequences and labels for supervised RNN training.

Split the dataset into training and testing sets.

3. Modeling with PyTorch

A Recurrent Neural Network (RNN) is built from scratch using PyTorch, with:

RNN Layer (with chosen hidden size and number of layers)

Fully Connected Layer for output

Adam Optimizer for training

Mean Squared Error (MSE) as the loss function

4. Training

Trained the RNN to learn temporal dependencies in the energy-usage data.

Visualized training loss to ensure model convergence.

5. Evaluation & Forecasting

Evaluated the model on the test set.

Compared actual vs. predicted values visually.

Forecasted future peak power demand using the trained model.

Running the Notebook in Google Colab

You can run this project in Google Colab without any setup on your local machine.

Steps to Run:

Open Google Colab

Upload the Notebook

Upload the Dataset

Install Required Libraries (if not available):

!pip install torch pandas numpy matplotlib scikit-learn

Select GPU and run cells in order.