In this study, we developed and assessed the potential of dynamic control strategies for a domestic scale 1-kW solar thermal power system based on a non-recuperated organic Rankine cycle (ORC) engine coupled to a solar energy system. Such solar-driven systems suffer from part-load performance deterioration due to diurnal and inter-seasonal fluctuations in solar irradiance and ambient temperature. Real-time control strategies for adjusting the operating parameters of these systems have shown great potential to optimise their transient response to time-varying conditions, thus allowing significant gains in the power output delivered by the system. Dynamic model predictive control strategies rely on the development of computationally efficient, fast-solving models. In contrast, traditional physics-based dynamic process models are often too complex to be used for real-time controls. Machine learning techniques (MLTs), especially deep learning artificial neural networks (ANN), have been applied successfully for controlling and optimising nonlinear dynamic systems. In this study, the solar system was controlled using a fuzzy logic controller with optimised decision parameters for maximum solar energy absorption. For the sake of obtaining the optimal ORC thermal efficiency at any instantaneous time, particularly during part-load operation, the first-law ORC model was first replaced by a fast-solving feedforward network model, which was then integrated with a multi-objective genetic algorithm, such that the optimal ORC operating parameters can be obtained. Despite the fact that the feedforward network model was trained using steady-state ORC performance data, it showed comparable results compared with the first-principle model in the dynamic context, with a mean absolute error of 3.3 percent for power prediction and 0.186 percentage points for efficiency prediction.