⚡ Power Output Prediction using Artificial Neural Networks (ANN)

Predicting the electrical power output of a Combined Cycle Power Plant using Deep Learning
_{Built with TensorFlow • Keras • NumPy • Scikit-Learn • Matplotlib}

---

---

---

📑 Table of Contents

• Project Overview
• Motivation
• Dataset
• Model Architecture
• Results & Evaluation
• Getting Started
• Example Usage
• Repository Structure
• Future Improvements
• References
• License

---

📘 Project Overview

This project demonstrates a complete machine learning regression pipeline that predicts power plant electrical output (MW) using environmental and operational parameters.
The solution is implemented with a fully connected Artificial Neural Network (ANN), trained on the Combined Cycle Power Plant (CCPP) dataset from the UCI Machine Learning Repository.

The notebook walks through all steps — data preprocessing → model training → evaluation → visualization — in a clear, educational manner.

---

🎯 Motivation

Accurate prediction of power output helps:

• Optimize power generation and resource allocation
• Support grid management and forecasting systems
• Serve as a case study for regression using neural networks in industrial contexts

---

🧠 Dataset

• Source: UCI Machine Learning Repository — CCPP Dataset
• Samples: 9,568
• Features:
  • Ambient Temperature (AT, °C)
  • Ambient Pressure (AP, mbar)
  • Relative Humidity (RH, %)
  • Exhaust Vacuum (V, cm Hg)
• Target: Electrical energy output (PE, MW)

Preprocessing includes:

• Missing value handling
• Feature scaling / normalization
• Train-validation-test split

---

🧩 Model Architecture

Layer	Units	Activation	Description
Input	4	—	Input features (AT, AP, RH, V)
Dense	128	ReLU	Hidden layer 1
Dense	64	ReLU	Hidden layer 2
Output	1	Linear	Predicted Power Output

Training Details:

• Optimizer: Adam
• Loss: Mean Squared Error (MSE)
• Metrics: R², MAE, RMSE
• Callbacks: Early Stopping, Learning Rate Scheduler

---

📊 Results & Evaluation

Metric	Train	Validation
MSE	0.0032	0.0037
R²	0.94	0.93

Visual Insights:

• ✅ Training vs Validation Loss Curve

• ✅ Predicted vs Actual Scatter Plot

• ✅ Residual Distribution Plot

---

🚀 Getting Started

🧰 Requirements

Dependency	Version
Python	3.8+
TensorFlow	2.x
NumPy	≥1.22
Scikit-learn	≥1.0
Matplotlib	≥3.5

Install everything with:

pip install -r requirements.txt

▶️ Running the Project

Run the notebook interactively:

jupyter notebook power_output_prediction_ann.ipynb

---

💡 Example Usage

from predictor import PowerPredictor

# Load trained model
model = PowerPredictor.load("models/ann_model.pkl")

# Predict power output for new sample
X_new = [[25.0, 1015.2, 60.3, 40.1]]
y_pred = model.predict(X_new)

print(f"⚡ Predicted Power Output: {y_pred[0]:.2f} MW")

---

📁 Repository Structure

Path	Description
.gitignore	Specifies files and folders to exclude from Git tracking
Folds5x2_pp.xlsx	Dataset used for training and evaluation
LICENSE	MIT License for open-source sharing
README.md	Project overview, documentation, and visual insights
requirements.txt	Python dependencies for reproducibility
power_output_prediction_ann.ipynb	Jupyter notebook containing full model workflow and evaluation
assets/	Folder containing result plots and project thumbnail
├── power_output_thumbnail.png	Thumbnail image summarizing the project
├── results_training_validation_loss.png	Line plot of training vs validation loss
├── results_actual_vs_predicted.png	Scatter plot comparing actual vs predicted outputs
└── results_residuals_distribution.png	Histogram of residuals distribution

---

🌱 Future Improvements

• 🔍 Hyperparameter optimization (Grid / Bayesian search)
• 🧮 Model benchmarking (ANN vs RF vs XGBoost)
• 🧠 Cross-validation and uncertainty quantification
• ⚙️ Deployment as REST API (FastAPI / Flask)
• 📈 Integration with MLflow or Weights & Biases

---

📚 References

• Dataset: UCI Machine Learning Repository – Combined Cycle Power Plant Dataset
• Frameworks: TensorFlow, Keras, Scikit-Learn
• Paper: Pınar Tüfekci (2014) — Prediction of full load electrical power output using machine learning methods

---

📄 License

This project is licensed under the MIT License.
See the LICENSE file for details.

---

👤 Author

Arian Jr.
📧 My Email
🔗 My GitHub

---

Made with ❤️ by ArianJr

_{⭐ If you found this project useful, please consider giving it a star! It helps others discover it and supports my work.}

---