Abstarct: Detection and classification of faults is an essential issue in transmission lines because a fast and accurate protection scheme for clearing faults can prevent damages to expensive power grid equipment and the spread of outages. As the main or backup protection, distance protection performs the fault detection task in transmission lines, the performance of which is improved by fast and accurate fault classification. However, this protection has difficulties in some situations, like the power swing, and the need to recognize this situation in its proper functioning is inevitable. So far, various methods have been proposed to improve the performance of disturbances detection and fault classification, which do not necessarily perform well in all possible conditions in the power system. In this thesis, a new method for detecting disturbances and a new method for fault classification are proposed, the input of which is calculated by applying the Teager-Kaiser energy operator to the voltage signal. The parameters of the proposed methods are independent of the system under study. The stable power swing, unstable power swing, and faults with or without the presence of power swing can be detected using the proposed method for disturbances detection with a good speed and accuracy. This method uses statistical indices of kurtosis, skewness, and coefficient of variation to define its proposed indices, including the data distribution shape index and data distribution dispersion index. Also, the proposed fault classification method is adaptive to the fault detection time that classifies the type of fault in the ⅕ cycle after detection. In this method, linear regression and variance are used to define the proposed adaptive fault classification index, and the support vector machine and feature selection method are used for the appropriate classification. The 9-bus WSCC and 39-bus New England test systems are used to evaluate the proposed methods. The results of various simulations show that the proposed method of disturbance detection has 100% accuracy with an average fault detection time of less than one millisecond under normal conditions, while the fault detection accuracy decreases slightly (about 0.66%) in the presence of noise. Also, according to the obtained results, various conditions such as noise, power swing, and fault detection time do not noticeably affect the performance of the proposed fault classification method, and the accuracy of the method in all conditions is more than 99%.