NiSaNo70/SDS-CP036-powercast

PowerCast is a time-series regression project aimed at predicting the power consumption in three urban zones of Tetouan City, Morocco, using environmental conditions such as temperature, humidity, wind speed, and solar radiation metrics. PowerCast ⚡: A Deep Learning App for Electricity Demand Forecasting Live Application Link: sds-cp036-powercast-nima.st… Overview PowerCast is an end-to-end data science project that forecasts short-term electricity demand for three different zones in Tetuan City, Morocco. The project demonstrates a complete machine learning pipeline, from initial data exploration and feature engineering to model training, optimization, and final deployment as an interactive web application. The core of the application is a Bidirectional LSTM (Long Short-Term Memory) model built with PyTorch, which is capable of learning complex temporal patterns from historical weather and power consumption data. Features Interactive Forecasting: Users can upload the historical dataset, select any date and time, and receive an on-demand forecast for the next 10 minutes. Multi-Zone Prediction: Separate deep learning models are trained and deployed for each of the three zones, allowing for specialized predictions. Time-Series Analysis: The project includes a deep dive into time-series feature engineering, using cyclical features to capture daily and weekly seasonality. Model Interpretability: Uses SHAP (SHapley Additive exPlanations) to understand which features (like temperature, humidity, or time of day) are driving the model's predictions. Tech Stack Modeling & Data Science: Python, PyTorch, Scikit-learn, Pandas, NumPy Web Application & Deployment: Streamlit, Streamlit Community Cloud Data Source: UCI Machine Learning Repository - Tetouan City Power Consumption