INFISYS, INC. — Department of Energy STTR Phase I: 13b

Description

The immense growth in weather and river sensor data availability along with increases in computational power and algorithmic sophistication have enabled the widespread application of physical, machine learning, and artificial intelligence algorithms to river forecasting. The forecasts produced by widely varying model approaches are inherently limited in different ways, such as by the number and quality of training observations, producing biased predictions that achieve optimal skill only over specific ranges of weather and river conditions. Thus, despite the growing availability and new sources of forecasts, hydropower owners cannot fully leverage the information because they lack an effective mechanism to determine the “best” forecast for given conditions or objectives. Tools are needed to consolidate forecast information, demonstrate skill, and extract optimal forecast guidance from the range of possibilities. Such tools will increase uptake and beneficial use of the growing wealth of forecast information, and lead to improved operating decisions, optimal power generation, and enhanced system value. This research proposes to improve the value of forecasts by developing a broadly accessible verification framework and applying machine learning techniques to identify the ‘best’ forecast from an ensemble of sources given the current weather, river and operating conditions. In Phase I, we will develop a proof-of-concept framework consisting of three primary components: 1) a data platform based on a unified data management paradigm, 2) a forecast verification system, and 3) AI/ML-based techniques for conditional forecast selection. The data platform and verification framework will build upon an existing foundation, adding enhancements to the architecture to support pervasive accessibility and commercialization potential, such as greater data diversity, cloud-based access, and user self-service interfaces. Phase I work will largely focus on development of the third component, namely AI/ML-based techniques to select the ‘best’ forecast for the current conditions from a set of available forecasts or hybrid of those forecasts. If the proposed research is successful and carried into full build-out in Phases II and III, anticipated public benefits, stemming from the increased uptake and improved use of forecasts by the hydropower sector, include greater reliability of electricity, reduced flood impacts, and enhanced recreation, environmental quality and public safety.