The proliferation of photovoltaic (PV) has been increased in distribution systems worldwide. The intermittent PV generation can cause diverse operational problems in the grid, especially voltage deviations and violations. As a result, voltage assessment in distribution systems interconnected with PV, which has a heavy computational burden, is privileged to assist power utilities in decision-making. In this paper, a low-computational and accurate voltage assessment approach with PV considering fine-resolution simulations (i.e. time-step of 1 second) is proposed. Specifically, the proposed approach can rapidly compute the voltage deviation in the whole distribution system and terminal voltages of PV units based on a data-driven model. This model is built using machine learning considering various scenarios of PV and load profiles. The proposed approach has the following features: 1) its computational burden is very low compared to the widely used iterative based methods, 2) it can handle the full data with the finest available resolution, yielding accurate voltage assessment. The proposed method has been applied for voltage assessment considering daily and annual simulations of different distribution systems interconnected with PV units. The simulation results manifest the high accuracy and computational speed of the proposed approach, especially for fine-resolution simulations.