The demand for energy is becoming increasingly important, and who says strong demands for energy says rising CO 2 emissions. Everyone agrees that a great part of the energy consumed by industry and households can be saved. The energy savings can take many forms. In addition to the necessity to build equipments more and more energy efficient, it is also necessary to get a clear view of how the energy is used. This obviously involves the implementation of an energy flow measuring system for long lasting optimization solutions. It is precisely in this context that the project CHIC (Low cost industry utilities monitoring systems for energy savings), funded by the French National Research Agency (ANR), emerged. The objective of this project is to develop and test low-cost non-intrusive sensors to monitor and analyze the energy consumption of major flows used in the manufacturing sector (electricity, gas, compressed air). With such sensors, it should be possible to tool up a factory, equipment by equipment, which is not feasible with intrusive sensors. The ultimate goal is the long term consumption monitoring and the detection of the consumption deviations rather than a precise measurement. The measurement accuracy is fixed to 5%. These developments are based on the recent approaches in system identification and parametric estimation. This project, concretely, involves the design of new low-cost sensors in the following areas: current sensors, voltage, power, and gas flow, relying on the international ISO 50001 standard for Energy Management Systems. The work presented in this chapter focuses on the modeling of the gas flow supplied to a boiler in order to implement a soft sensor. This implementation requires the estimation of a mathematical model that expresses the flow rate from the control signal of the solenoid valve and the gas pressure and temperature measurements. Two types of models are studied: LPV (Linear Parameter Varying) model with pressure and temperature as scheduling variables and a non-parametric model based on Gaussian processes.