Renewable Energy Prediction and Market Penetration

The installation of utility-scale (centralized) and residential (distributed rooftop) solar photovoltaic (PV) has increased exponentially in the past decade. Predicting solar power output and understanding its impact to the national grid are extremely important in enabling a large percentage penetration of PV power into the energy sector. 

The intermittency and transitory nature of solar power, along with the heterogeneity of solar PV in terms of capacities, locations and configurations, demand a series of inter-dependent mechanisms to ensure its smooth integration: 1) (Cloud) tracking devices and (irradiation) sensors to monitor and collect spatiotemporal information on the solar/cloud system, 2) sophisticated models to forecast solar output, and, 3) cost-effective aggregation strategy for multiple renewable energy sources to minimize the negative effects of individual volatility and  spinning reserves.

Figure 1: Deep Learning Integrator (DLI)
 

This project will focus on advancing the state-of-the-art via cutting-edge research and development in machine learning that gathers insights from large volumes of spatial and temporal data taken from a network of pyranometers and geostationary satellite.

Deep convolution neural networks (CNN) will be applied to map cloud image features to solar irradiance predictions  and be integrated to a boosted forecast model for accurate solar prediction over time horizons ranging from minutes  to eight hours and to facilitate decision making in multiple applications, including solar power flicker mitigation, load following, safety reserve optimization, and unit commitment. 

Our project will uncover the data causality patterns as the following figures and use them to create a context for advanced machine learning and deep learning models.

Figure 2: Two Locations have inter-dependent irradiance and energy outputs in Long Island Solar Farm, the largest solar farm in the East Coast of United States.

Figure 3: Inter-correlated solar irradiance sensors have causal relationship and are determined by the movement speed and direction of clouds