In the last fifty years, natural systems’ models have done tremendous progress in accurately reproducing a large variety of physical processes both in space and time. Conversely, despite human footprint is increasingly recognized as a major driver of undergoing global change, human behaviours and their interactions with natural processes are still described in an overly simplified in most of the models used to make long term projections and policy support. We can distinguish two behavioral modeling approaches: descriptive models, which derive behavioral rules specifying human actions in response to external stimuli, and normative models, which assume fully rational behaviors and provide optimal decisions that maximize a given utility function. In this talk we contribute advanced machine learning methods to cope with two major challenges of state-of-the-art behavioral models. First, we advance normative models to describe human behaviors in multipurpose water systems by identifying the operators’ preference in terms of tradeoff among multiple competing objectives and the dynamic evolution of this tradeoff driven by extreme drought or flood events. We map the tradeoff selection onto a newly developed negotiation protocol, where multiple virtual agents independently optimize different operating objectives and periodically negotiate a compromise solution. Second, we contribute a new approach based on eigenbehavior analysis to mine typical behavioral profiles from historical observational data across a large set of water operators in California. The extracted profiles are validated a posteriori with respect to specific features of the reservoirs (location, elevation, and capacity).

Data-driven behavioral models of water reservoir operators

M. Giuliani;A. Castelletti;
2018-01-01

Abstract

In the last fifty years, natural systems’ models have done tremendous progress in accurately reproducing a large variety of physical processes both in space and time. Conversely, despite human footprint is increasingly recognized as a major driver of undergoing global change, human behaviours and their interactions with natural processes are still described in an overly simplified in most of the models used to make long term projections and policy support. We can distinguish two behavioral modeling approaches: descriptive models, which derive behavioral rules specifying human actions in response to external stimuli, and normative models, which assume fully rational behaviors and provide optimal decisions that maximize a given utility function. In this talk we contribute advanced machine learning methods to cope with two major challenges of state-of-the-art behavioral models. First, we advance normative models to describe human behaviors in multipurpose water systems by identifying the operators’ preference in terms of tradeoff among multiple competing objectives and the dynamic evolution of this tradeoff driven by extreme drought or flood events. We map the tradeoff selection onto a newly developed negotiation protocol, where multiple virtual agents independently optimize different operating objectives and periodically negotiate a compromise solution. Second, we contribute a new approach based on eigenbehavior analysis to mine typical behavioral profiles from historical observational data across a large set of water operators in California. The extracted profiles are validated a posteriori with respect to specific features of the reservoirs (location, elevation, and capacity).
2018
978-84-09-02938-9
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11311/1071922
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact