This paper presents an improved stator flux oriented (SFO) control system for a stand-alone induction generator. The induction generator supplies a variable resistive dc load. In order to provide an essentially constant terminal voltage, the product of the rotor speed and the stator flux reference should remain constant. However, in this case the control system is not able to function properly at different loads and dc-link voltages. In this paper, we introduce a new algorithm in which this product is constant at certain dc-load and dc-link voltage references. The dependence of the stator flux reference on the dc load and dc voltage reference is mapped using an artificial neural network (ANN). We also present an analysis of the efficiency of the SFO control system, as well as its performance during transients, over a wide range of both dc-link voltage references and loads. The validity of the proposed approach is verified by realistic simulation in a Matlab-Simulink environment.