Energy consumption forecasting

Overview

This project focuses on modeling and predicting energy consumption using a variety of machine-learning techniques. It involves loading, preprocessing, and exploring energy consumption data, engineering relevant features, building predictive models, and evaluating their performance. The project also implements a Flask API to serve the model evaluation results.

File Structure

├── app.py                       # Flask application for serving model results via API
├── data_preprocessing.py        # Data loading, preprocessing, feature engineering, and visualization
├── model_building.py            # Model training, evaluation (including LSTM), and time series cross-validation
├── energy_consumption_data.csv  # The dataset used for training and evaluation
├── requirements.txt             # List of Python dependencies for the project
└── README.md                    # This file (project overview and setup instructions)

Dependencies

The project relies on the following Python libraries:

• pandas: For data manipulation and analysis
• scikit-learn: For various machine learning models and evaluation metrics
• tensorflow/keras: For building and training the LSTM model
• Flask: For creating the web API
• numpy: For numerical computations

To install all required dependencies, it is recommended to use a virtual environment and run:

pip install -r requirements.txt

A sample requirements.txt is shown below:

flask==3.1.0
pandas==2.2.3
scikit-learn==1.6.1
tensorflow==2.18.0
numpy==2.02
seaborn==0.13.2
matplotlib==3.10.1

Usage

1. Clone the repository:

   git clone https://github.com/hydrogeohc/Optim_Energy
   cd Optim_Energy
   

2. Create a virtual environment:

   python3 -m venv venv  # create the virtual environment
   source venv/bin/activate   # activate on macOS/Linux
   # venv\Scripts\activate   # activate on Windows
   

3. Install dependencies:

   pip install --upgrade pip # upgrade pip
   pip install -r requirements.txt # Install the dependencies
   

4. Run the Flask API:

   python app.py
   

5. Access the API endpoints:

   • http://127.0.0.1:5000/ - Returns a welcome message.
   • http://127.0.0.1:5000/model_results - Retrieves model evaluation results in JSON format. Note: The AirPlay Receiver service on macOS may conflict with port 5000. If you encounter issues, try changing the Flask port to 5001 in app.py and accessing http://127.0.0.1:5001/model_results.

Models

The following machine learning models are implemented and evaluated in this project:

• Linear Regression
• Random Forest
• Gradient Boosting
• LSTM (Long Short-Term Memory)

Note: The traditional models (Linear Regression, Random Forest, and Gradient Boosting) are evaluated using Time Series Cross-Validation. However, the current implementation evaluates the LSTM using a single train/test split. The LSTM results are therefore not directly comparable to the other models.

Data

The energy_consumption_data.csv file contains the historical energy consumption data used to train and evaluate the models. The dataset includes features such as:

• timestamp: Date and time of the reading
• temperature: Temperature at the time of the reading
• humidity: Humidity at the time of the reading
• energy_consumption: The target variable, representing the energy consumed

The project also engineers new features such as:

• temp_humidity_interaction: Interaction term between temperature and humidity
• energy_consumption_lag1: Energy consumption from the previous time period (lag 1)
• energy_consumption_lag2: Energy consumption from two time periods prior (lag 2)
• hour, day_of_week, is_weekend: Temporal features derived from the timestamp.

Evaluation Metrics

The models are evaluated based on the following metrics:

• Mean Squared Error (MSE)
• Mean Absolute Error (MAE)
• Bias

Model Selection

Model Evaluation Results

The following table presents the model evaluation results, including Mean Squared Error (MSE), Mean Absolute Error (MAE), and Bias:

Model	MSE	MAE	Bias
Random Forest	14245.34	104.40	-8.63
Gradient Boosting	15696.48	106.75	-9.58
Linear Regression	16427.58	109.70	-12.03
LSTM	18726.25	109.30	-18.30

Base on the above results, the Random Forest seems to be the best with all the lower MSE, MAE, and closeer to zero bias.

• note:
• MSE might be the most important metric when the project wants to avoid large errors
• MAE might be preferred when the project want a model that's generally accurate
• Bias is important when the project need a model that doesn't systematically over- or under-predict.

Troubleshooting

• ModuleNotFoundError: Ensure that all Python dependencies are installed in your virtual environment. If you see the error, install the libraries that are missing.
• File Permissions: Make sure that the energy_consumption_data.csv file is readable by the Flask application.

License

MIT License