Optimizing predictive maintenance for communication power systems using a GA-PSO-CNN-LSTM hybrid approach

The reliability of communication power supplies is crucial for modern networks, yet traditional maintenance methods remain inefficient and costly. In this paper, we propose a hybrid CNN-LSTM model enhanced by an attention mechanism, which optimizes model hyperparameters efficiently through the fusion of the Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) algorithms. This model leverages the spatial feature extraction capability of CNNs and the time series modeling capability of LSTMs, leading to a significant improvement in prediction performance. Experimental results demonstrate that the optimized model outperforms traditional methods in key metrics, including prediction accuracy (96%), RMSE (0.60), and MAE (0.37), while also accelerating convergence speed and enhancing stability. The proposed method reduces the complexity of predictive maintenance for communication power supplies and offers a more efficient solution for resource-constrained environments. Future research could focus on combining lightweight optimization algorithms with novel architectures to further enhance real-time performance and generalization capabilities.