This paper deals with a residential hybrid thermal/electrical grid-connected home energy system incorporating real data for the load demand. A day-ahead scheduling (DAS) algorithm for dispatching different resources has been developed in previous studies to determine the optimal operation scheduling for the distributed energy resources at each time interval so that the operational cost of a smart house is minimized. However, demand of houses may be changed in each hour and cannot be exactly predicted one day ahead. System complexity caused by nonlinear dynamics of the fuel cell, as a combined heat and power device, and battery charging and discharging time make it difficult to find the optimal operating point of the system by using the optimization algorithms quickly in online applications. In this paper, the demand forecast error is studied and a near-optimal dispatch strategy by using artificial neural network (ANN) is proposed for the residential energy system when the demand changes are known one hour ahead with respect to the predicted day-ahead values. The day-ahead and hour-ahead optimizations are combined and ANN training inputs are adjusted according to the problem such that the economic dispatch of different energy resources can be achieved by the proposed method compared with previous studies. Using the model of the fuel cell extracted from experimental measurement and real data for the load demand makes the results more applicable in real residential energy systems. (C) 2016 Elsevier Ltd. All rights reserved.