A smart grid (SG) is an interconnection of an information network, a communication network, and an electrical grid. The rapid development of SG technology has given rise to complex cyber-physical systems. This complexity expands the cyber-attack surface of SGs and increases their vulnerability to cyber-physical attacks. Protecting vital components and subsystems of energy networks, as well as communication, from hostile and external attacks, is the main goal of SG security. This work presents a hybrid method using an Enhanced Crow Search (ECS) algorithm in combination with a dense-gated U-Net for cyber-attack detection and classification in smart grids. The suggested method improves the precision and effectiveness of cyber-threat identification in smart grid framework by utilizing the advantages of both deep learning and nature-inspired optimization. This manuscript presents a method to detect false data injection attack (FDIA) in SGs by optimizing dense-gated U-Net using an Enhanced Crow Search (ECS) algorithm. This method uses analysis to identify any suspicious values and detect any measurement value that is unfamiliar. The ECS optimizes the dense-gated U-Net’s complex hyper-parameter. Different state-of-the-art deep learning techniques are used for comparisons.