In recent years, frequent cyber-attacks on Cyber-Physical Power Systems (CPPS) worldwide have caused significant economic losses and severely threatened social stability. These cyber-attacks inevitably affect the network traffic of CPPS, making anomaly detection in network traffic an effective means of detecting and defending against such attacks. However, existing anomaly detection methods for CPPS network traffic suffer from insufficient feature extraction and inadequate detection accuracy. To address these issues, this paper proposes a deep learning-based approach for anomaly detection in CPPS network traffic. This approach comprises a feature extraction algorithm, a feature reduction algorithm, and a deep learning-based classification algorithm. The feature extraction algorithm can capture multi-dimensional features of network traffic, while the feature reduction algorithm can streamline and remove redundant features, thereby improving data processing efficiency. The deep learning-based classification algorithm integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory networks (LSTM) to fully extract traffic features and enhance detection accuracy. This study evaluates the proposed approach using real attack datasets, demonstrating its effectiveness in detecting network attacks.