Power system faults in distribution networks can lead to severe service disruptions and compromise system reliability. This paper introduces a new fault classification approach using a Long Short-Term Memory (LSTM) neural network optimized by the Sea Lion Optimization (SLO) algorithm. The LSTM-SLO model autonomously tunes key hyperparameters, such as learning rate and hidden layer size, resulting in improved classification performance and faster convergence. Ten types of faults were simulated in MATLAB 2019b, including AG, BG, CG, AB, BC, AC, ABG, BCG, ACG, and ABC. Features were extracted using the Fast Fourier Transform (FFT) and processed in PyCharm 2022 for training and evaluation. The proposed classifier achieved an accuracy of 99.9%, an F1-score of 0.997, and an average response time of 0.42 s, outperforming traditional methods such as SVM, ANN, and CNN in both precision and speed. The standard deviation of classification accuracy across folds was found to be less than 1%. Unlike conventional approaches that require extensive manual tuning and struggle with fault overlap, the LSTM-SLO hybrid framework integrates deep learning with metaheuristic optimization to deliver a robust, intelligent, and scalable framework for fault classification in power distribution infrastructure.