The Shared Bicycle Demand Prediction Algorithm Based on Improved Graph Convolutional Network with Non-Negative Matrix Factorization

In response to the tidal effect and the complexity of demand prediction in shared bicycle systems, this paper proposes a prediction model based on an improved Graph Convolutional Network (GCN) with Non-Negative Matrix Factorization (NMF) for shared bicycle demand forecasting. The model utilizes NMF to extract latent temporal and spatial patterns, enhancing its ability to capture implicit relationships between data features, thus improving prediction accuracy and reducing errors. Experimental results demonstrate that the NMF-GCN model achieves lower prediction errors compared to the LR, MLP, and STGCN models, with a mean absolute error of only 2.0074, which is a reduction of 5.99% compared to the STGCN model. This highlights the effectiveness of matrix factorization in feature learning.