ISO-QLCNet: Optimized Q-Learning Based Convolutional Network for Harmonic Mitigation in Traction Network

Abstract

Specialized electrical infrastructure of Traction Power Supply Systems (TPSS) adopts the high voltage levels while varying load current dynamics, which is distinct from a general power system by handling the high power and train loads. However, transformer saturation, unbalanced loads, non-linear power electronic devices, and system resonance factors introduced the harmonics in the TPSS, which resulted in equipment damage issues and contributed to performance degradation. Consequently, the research work implements the Intelligent Swooping Optimization-based Q-Learning distributed one-dimensional Convolutional Network (ISO-QLCNet) model to perform the harmonic mitigation task in the TPSS. The QLCNet model predicts potential failures and learns the optimal control policies to mitigate the harmonics in the railway operations. Moreover, the developed model utilized the dynamic and non-linear loads to identify the complex harmonic patterns and indicate the power quality (PQ) issues to achieve the mitigation task effectively. Furthermore, the model parameters are optimized by the ISO algorithm, which helps to meet the maximum energy efficiency and lower total harmonic distortion (THD). Eventually, the ISO-QLCNet model exhibits the THD, train-1 voltage, train-1 current, switching output, train-2 current, and train-2 voltage are 0.62, 562.26 V, 8217.19 A, 5260.74, 961.54 A, and 632.49 V, respectively.