ARCHIVES

Energy and environmental systems are full of decisions that must be tuned under uncertainty: how much power to draw from a solar panel, how to bill and balance consumption on a smart grid, how to share spectrum in a wireless network and how to read a biosensor in the field. Computing itself is part of the problem, since data centres, web services and smartphones draw a growing share of electricity. Many of these decisions have no closed-form solution because the underlying models are nonlinear, noisy or only partly known. Metaheuristic search, fuzzy reasoning, nonlinear analysis and data-driven forecasting are now used to find good settings, but studies are spread across power engineering, networking and environmental science. This chapter reviews optimization and learning methods for five groups of problems: renewable generation and grid interfacing, smart metering and energy forecasting, resource allocation and energy-aware computing in communication and IoT systems, environmental and agricultural monitoring, and the approximation of nonlinear physical models. The review finds that hybrid metaheuristics, such as grey wolf optimization with opposition-based learning, improve tracking and control in renewable systems, that tree ensembles and regression models are standard for consumption forecasting, that fuzzy optimization handles uncertainty in network resource allocation, that signal-aware transfer scheduling and green design guidelines reduce the energy used by smartphones and web services, and that neural approximation can stand in for expensive solvers of nonlinear equations. Few studies, however, compare methods on shared benchmarks or report computation cost. Future work should use common test systems, report run time next to solution quality and link optimization with forecasting in one loop.