In today’s Internet environment, cybersecurity has become a critical issue. However, existing intrusion detection systems still face numerous challenges, including low detection accuracy, high false alarm rates, and ineffective handling of complex attack patterns. To address these issues, this paper proposes a network intrusion detection model based on Convolutional Neural Network (CNN) and Bidirectional Gated Recurrent Unit (BiGRU) (NIDS-CNNBiGRU). This model first extracts local features of network traffic data using CNN convolutional and pooling layers. Subsequently, BiGRU layers are employed to capture temporal information and contextual relationships in the traffic data, processing input sequences in both forward and backward directions to significantly enhance the recognition of long-term dependencies. To mitigate overfitting, Dropout layers are incorporated into the model to reduce complexity and enhance generalization capability. Finally, global max-pooling is applied to compress the feature sets of each channel into single values for final intrusion detection. Our experiments, conducted on the publicly available KDD Cup99 and NSL_KDD datasets, demonstrate that the NIDS-CNNBiGRU-based intrusion detection model excels in accuracy and robustness. Compared to traditional methods, our model not only achieves superior performance in detecting malicious network activities but also significantly reduces false alarm rates.