The integration of distributed energy resources (DERs) into smart grids relies heavily on power electronic inverters. However, with the increasing number of DERs and advancements in communication and computation technologies, these power electronic inverters have become vulnerable to cyber-attacks, making the cyber-physical security of power electronics-based smart grids (PESGs) a significant challenge. This dissertation proposes a set of innovative cyber-physical security solutions to address this challenge. Firstly, a comprehensive cyber-attack assessment method is developed that considers the security and stability of both the grid and the inverter. This method includes impedance-based cyber-attack modeling for PESGs and stability-based and security-based evaluation criteria. Secondly, a model-based approach is developed to detect spoofed sensor data (SSD) in grid-connected inverters. This approach uses an adaptive cumulative sum chart to meet the detection requirement of IEEE 1547 Standard. Thirdly, a machine-learning approach is developed for detecting cyber-attacks in PV farms using only point of common coupling (PCC) sensors to reduce the complexity of input data. A real-time simulation, detection, and visualization framework is designed to demonstrate the feasibility of this solution in a real-world application. Fourthly, a hybrid cyber-attack detection method is proposed to address the cyber-physical security of PV farms, which integrates model-based and data-driven methods by fusing detection scores at both device and system levels. Fifthly, a physics-data-based identification method is proposed to identify various attacks in PV farms using harmonic state space (HSS) models. The method utilizes the calculated residual generated by the HSS model and a clustering approach to investigate attack propagation and precisely locate attack sources within a PV farm. Furthermore, to address cyber-physical security in microgrids, a novel model-based cyber-attack detection and identification method is developed, and an optimal parameter verification method is proposed to estimate attack gains for stealthy attack identification. Finally, a robust model predictive control (RMPC) approach for virtual synchronous generators (VSGs) is developed to counteract frequency deviations caused by cyber-attacks, ensuring the overall frequency stability of the microgrid.