There is an increasing demand for reducing the power consumption in the field of embedded-system development. A development methodology, which can change software's power consumption according to the power consumption of the hardware, can help fulfill this requirement. However, there will be a trade-off between the power consumption and service quality, which must be balanced for efficient operation. In this paper, we propose dynamic feature-oriented energy-aware adaptive modeling (DFEAM), which develops self-adaptive software through model-driven development for achieving a proper balance between the power consumption and quality of service (QoS). In this method, the application itself decides its behavior, according to the power-consumption situation, by linking the feature model describing the variability of the application with the description of its behavior, using the executable and translatable unified modeling language (xtUML). For achieving a satisfactory QoS for variations that are complex and dependent on variable points, a model is created to quantify the QoS values, which is then used as an index of comparison for finding the optimum variation. We conducted case studies on applications with multiple variable points, and evaluated them using the GQM model. The results of the evaluation showed that the adaptation incorporated provided the maximum software quality under the given power limitations, thus verifying the usefulness of the proposed DFEAM method.