This study explores an intelligent automobile cockpit alarm system based on machine learning technology, aimed at addressing safety risk issues caused by numerous alarm sources. To tackle the deficiencies of redundancy and imprecise classification of alarm information in current systems, this paper proposes a hybrid selection strategy that combines manual experience filtering with an CNN (Artificial Neural Network) model. Specifically, this strategy integrates operational status information from various devices and performs effective classification, using manual experience to reduce suspected defect signals, while leveraging the CNN model for feature extraction and precise classification. Experimental results show that the CNN model achieved a classification accuracy of 89.07% on the test set, and after combining both methods, the selection accuracy for all original alarm signals reached an impressive 99.998%, significantly surpassing the selection accuracy of the existing VAS system (90%). This verifies the efficiency and excellence of the proposed method in alarm information filtering. Future research will focus on increasing the amount of training data, optimizing model parameters, and improving text pre-processing techniques, with the aim of further enhancing the overall system performance.