MultiNet: Application of the Novel Hybrid Network Combining CNN and LSTM in Vehicle Trajectory Prediction

The time series prediction of vehicle trajectories is becoming increasingly crucial in numerous practical scenarios, such as facilitating corporate resource allocation, aiding government decision-making, and monitoring abnormal data. Historical time series data can outline the overall trend of the data, whether it is ascending, descending, or remaining stable. However, given that vehicle trajectory data is subject to a multitude of factors, it is challenging to comprehensively consider these factors and identify an appropriate model to accurately characterize them. This study introduces an innovative hybrid network architecture—multiNet, which is used to predict future values of time series. Driven by target data, multiNet does not rely on any prior knowledge. It employs Convolutional Neural Networks (CNNs) to extract local salient features from neighboring raw data and utilizes Long Short-Term Memory networks (LSTMs) to capture long-term memory information. Notably, multiNet is applied for the first time in the field of vehicle trajectory prediction to infer future trajectory points. Experimental results demonstrate that this method can effectively extract valuable information and significantly enhance prediction performance. In three vehicle trajectory datasets, multiNet achieved the best performance, outperforming other methods. These findings confirm the applicability of this hybrid neural network architecture in various vehicle trajectory time series prediction problems.