Lensless microscopy is an emerging technology that does not rely on the lens to focus and record the measured value. In conventional ptychography, the incremental gradient methods are widely used because of the superior computing efficiency and flexibility. However, because there are usually various noises collected by the sensor, which causes the reconstruction result to degrade sharply. The key point is how to control and design the step size. Therefore, we propose a loss correction algorithm with adaptive step size for lensless ptychography via multi-angle LED illumination. We add a loss correction term and adjust the step size according to the value of this term, which can effectively improve the convergence performance of the algorithm and can significantly improve the algorithm's anti-noise ability. Comparing with classical methods, both simulations and experiments prove that our method is better in efficiency and quality. Meanwhile, the results indicate the initial step size can be appropriately increased to improve the convergence speed without reducing the reconstruction quality.