Traffic Flow Anomaly Detection and Analysis Method Combining Convolutional Neural Network and Long Short-Term Memory Network

Abstract

To solve the problems faced by traditional traffic flow anomaly detection research, such as insufficient modeling of spatiotemporal dependencies, low efficiency in large-scale data processing, and insufficient accuracy of anomaly detection in complex scenarios, this paper constructs a CNN-LSTM joint model. Convolutional neural networks (CNN) are used to extract spatial features in traffic flow data, and long short-term memory (LSTM) processes the temporal dependencies in traffic flow data. In this way, the model can simultaneously consider the spatiotemporal features of traffic flow and improve the accuracy and efficiency of anomaly detection. In addition, this paper also applies a dynamic threshold anomaly detection mechanism based on error calculation, which dynamically adjusts the anomaly detection threshold by combining the error distribution of historical data and calculating the error between the predicted value and the actual observed value. Experimental data shows that the CNN-LSTM model outperforms the K-means, support vector machine (SVM), and deep autoencoder (DAE) methods in accuracy, recall, precision, and F1-score, reaching 93.2%, 90.8%, 92.5%, and 91.6%, respectively, and the false positive rate is only 2.3%. The CNN-LSTM has the shortest response time, ranging from 0.11 seconds to 0.15 seconds, which verifies its effectiveness in improving the accuracy and real-time performance of anomaly detection.