This talk presents weather and electric load forecasting using Python data science and machine learning tools (NumPy, Pandas, Scikit-learn). Using weather data from NOAA for the U.S., Florida, and Miami, we build a data pipeline for data exploration, summarization, visualization, feature engineering, regression and classification, and finally scoring and evaluation of predictive models.
This talk gives an introduction to weather and electric load forecasting using Python data science and machine learning tools (PyData). We begin by exploring, summarizing and visualizing weather data from NOAA for the U.S., Florida, Miami, and our neighbors to the south, the Caribbean and Latin America. We then move on to build weather forecasting models for multiple weather variables (focusing on temperature and wind), with a gentle introduction to feature engineering, regression and classification, and supervised and unsupervised machine learning. The models include predictive regression and classification models, clustering, and references to mode advanced models, such as deep learning. The talk presents feature engineering and how it may be used to address seasonality of the data to obtain accurate and efficient forecasting models. Data visualization, models and code are presented in Jupyter notebooks.
We will briefly introduce a related topic, electric load forecasting and its connection to weather. This talk is aimed at a novice to intermediate audience.