https://github.com/soumilgit/ai_ia-2

This project uses Long Short Term Memory (LSTM) neural networks to forecast hourly energy consumption, enabling smarter grid management, demand planning, and improved efficiency through data-driven insights.
https://github.com/soumilgit/ai_ia-2

Last synced: 12 months ago
JSON representation

This project uses Long Short Term Memory (LSTM) neural networks to forecast hourly energy consumption, enabling smarter grid management, demand planning, and improved efficiency through data-driven insights.

• Host: GitHub
• URL: https://github.com/soumilgit/ai_ia-2
• Owner: Soumilgit
• Created: 2025-04-18T19:24:32.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2025-04-18T19:54:39.000Z (about 1 year ago)
• Last Synced: 2025-04-19T07:50:30.336Z (about 1 year ago)
• Language: Python
• Homepage:
• Size: 3.62 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
#  Forecasting Hourly Energy Consumption with LSTM Networks

This project focuses on predicting **hourly power usage** using Long Short-Term Memory (LSTM) neural networks to improve **energy grid efficiency** and enable proactive grid management. By leveraging the PJM Interconnection dataset, we aim to develop a robust deep learning model that captures complex time-series patterns in electricity consumption data.

---

##  Problem Statement

Accurate forecasting of hourly energy consumption is vital for efficient electricity distribution, cost savings, and sustainability. Traditional models often fail to capture the non-linear and dynamic nature of power usage patterns caused by:

- Time of day and day of week

- Weather conditions

- Public holidays

- Human activity trends

---

##  Objectives

- Preprocess PJM energy dataset (handle missing values, outliers, and scaling)

- Engineer temporal features (hour, day, weekend/weekday, holidays)

- Visualize consumption patterns and correlation insights

- Develop an optimized LSTM-based forecasting model

- Evaluate model performance using RMSE, MAE, and MAPE

- Compare results with baseline models (Persistence, ARIMA)

- Deploy an accurate and interpretable forecasting solution

---

##  Technologies Used

| Tool/Library     | Purpose                                  |

|------------------|-------------------------------------------|

| **Python**       | Core programming language                 |

| **Pandas**       | Data manipulation and preprocessing       |

| **NumPy**        | Numerical computations                    |

| **Matplotlib / Seaborn** | Data visualization              |

| **TensorFlow / Keras** | LSTM model building                |

| **scikit-learn** | Evaluation metrics and scaling            |

| **holidays**     | Generate national/public holiday features |

---

##  Dataset

**PJM Hourly Energy Consumption Dataset**  

- Covers power usage across 13 U.S. states and Washington, D.C.

- Hourly consumption records from **2002 to present**

- Publicly available and granular to regional levels

---

##  Implementation Overview

1. **Data Preprocessing**

   - Imputation for missing values

   - Outlier detection (IQR/Z-score)

   - Min-Max scaling for normalization

2. **Feature Engineering**

   - Time-based (hour, day, month, holidays)

   - Lagged and rolling statistical features

   - External signals (e.g., weather/temperature if available)

3. **Modeling with LSTM**

   - LSTM layers with dropout for generalization

   - Optimized using grid search for hyperparameters

   - Early stopping to avoid overfitting

4. **Evaluation**

   - Compared against ARIMA and Persistence models

   - Performance metrics: RMSE, MAE, MAPE

---

##  Results & Key Findings

- **15% improvement** in forecast accuracy vs ARIMA baseline

- Effective prediction of **peak demand periods**

- LSTM effectively learns **daily and weekly trends**

- Training and validation loss converged consistently

- Framework applicable for real-time grid forecasting

---

##  Future Scope

- Incorporate real-time **weather forecasts**

- Use advanced deep learning variants (e.g., Bi-LSTM, Transformers)

- Expand to **renewable energy forecasting**

- Deploy as a live dashboard or web app for grid operators

---

##  Contributors

- **Soumil Mukhopadhyay** (16010122257)  

- **Sharwar Patil** (16010122278)  

- **Shreyas Nair** (16010122274)  

- **Rohit Sharan** (16010122307)

---

##  Presentation

The full project presentation is available in the [`Presentation/`](./Presentation/) folder:  

📂 [`Forecasting-Hourly-Energy-Consumption-with-LSTM-Nw.pptx`](./Presentation/Forecasting-Hourly-Energy-Consumption-with-LSTM-Nw.pptx)

---

##  Summary

This project demonstrates how deep learning can significantly improve electricity consumption forecasting. LSTM networks offer a data-driven approach to predict usage trends, helping pave the way for **smart grids**, **load balancing**, and **energy sustainability**.

---