Non-technical losses represent a current problem faced by energy utilities in rural networks in Brazil. It represents a significant income decrease to the utilities that attend to rural regions. However, detecting non-technical losses at irrigation consumers is not a trivial task. The rural energy grids have long extensions, and the irrigating consumers are usually located in regions of difficult access, aspects that make inspections difficult. The objective of this paper is to present a methodology to detect non-technical losses in rural consumers with irrigation-based crops using fuzzy logic, artificial neural network and deep learning. The proposed methodology suggests proper indicators of non-technical losses and efficiency, that will assist utility company decision making. The results achieved showed a percentage less than 10%, which means a low risk of non-technical losses according to the limits defined, on three harvests of one rice crop in southern Brazil, from 2011 to 2014.