Your system is not well equilibrated, how many frames did you used to calculate this binding free energy?
Plus, MM/PBSA is more like qualitative method than quantitative. TO get a better results, you need to focus on the trajectory where the RMSD fluctuations are minimal. Moreover, you can trim/exclude the initial frames where the structure is undergoing higher fluctuation, it can also improve the result.
why method did you use? 1-trajectory or 3-trajectory? In 3-trajectory protocol, the standard deviation of energy is pretty high because the internal energy component
You need to run longer simulation to get any meaningful results as 20ns is not sufficient. At least 100ns * 3 replicates can provide you reasonable results.