Abstarct: State estimation is a critical tool for static analysis of power systems, particularly in detecting cyberattacks. Traditional weighted least squares (WLS) estimators, while widely used, suffer from limitations such as computational time for large systems, susceptibility to divergence under certain conditions, and inadequate accuracy in specific scenarios. Conversely, data-driven methods, including neural networks, can lead to inaccurate decisions when network topology changes. This thesis proposes a novel hybrid approach that mitigates these drawbacks and enhances the reliability of cyberattack detection. The algorithm leverages the strengths of both WLS and neural networks, creating a robust and adaptive estimation frxamework. The proposed approach first trains a neural network on a set of pre-prepared network topologies. If the current network topology matches a pre-trained configuration, the neural network provides an initial assessment. However, if a topology mismatch occurs, the neural network requires retraining. To expedite this process, WLS serves as the primary estimator during retraining, providing a faster initial estimate. Once power flow calculations and network training are complete, a secondary estimation is performed using both WLS and the retrained neural network. If both algorithms detect a cyberattack, the final decision is made. If only one algorithm detects an attack, the other algorithm corrects the erroneous data, and the estimation and decision process is repeated. This iterative approach ensures accurate cyberattack detection by leveraging the complementary strengths of WLS and neural networks.