Optical Wireless Power Transmission (OWPT) system, using beam shaping, is a promising technology that involves a transmitter side using a laser and a receiver side using solar cells to transfer power wirelessly over long distances. The system faces a challenge of potential mismatch between both sides, which can lead to low efficiency of power transfer. The objective of this research is to improve the mismatch between the beam shape of a semiconductor laser light source and the shape of a photodetector solar cell. The shape of a typical solar cell is rectangular, which does not match the circular beam shape of the light source, resulting in lower power conversion efficiency. To address this, methods have been proposed to beam shape into a rectangle using Fresnel lenses or rectangular core optical fibers. However, the distance from the solar cell and the rotation of the solar cell itself change the size of the beam, resulting in a trapezoidal image as seen from the light source. In this study, we report the results of an evaluation of the change in efficiency with respect to changes in the object, using a spatial light modulator (SLM) that can arbitrarily shape the light beam. Specifically, we evaluate how adjusting the shape of a circular beam onto a solar cell target using the SLM leads to a decrease in photocurrent, and how recognizing an object's beam shape in real-time to control a camera using machine learning can predict object shape for error suppression compared to linear prediction.