Application of the PCA-DQN Algorithm in Optimizing Ice-Accretion and De-icing Strategies for Power Transmission Lines

With the increasing frequency of extreme weather events, ice accretion on power transmission lines has become a growing operational risk to the grid. To improve de-icing efficiency and reduce energy consumption, this paper combines Principal Component Analysis (PCA) with Deep Q-Network (DQN) to propose a dimension-reduction-accelerated de-icing strategy optimization algorithm. In the environmental simulation, we use multi-dimensional states—ice thickness, conductor temperature, and meteorological parameters—as inputs and apply PCA to retain the principal components that account for 95 % of the cumulative variance, thereby significantly reducing the state dimensionality. The DQN model is then trained on this reduced state space, employing an ε-greedy policy and soft updates of the target network. Experimental results show that, compared to the original DQN, the PCA-DQN achieves approximately a 30 % faster convergence rate under the same number of training episodes, reduces total de-icing energy consumption by about 15 %, and yields a more stable strategy, demonstrating strong potential for engineering applications.