Search

Urban function evolution (UFE) has become more and more complex in emerging cities. However, insufficient theoretical support exists for the visual expression of the spatial correlation between UFE patterns. In order to fill this gap, we use the 2013 and 2022 Point-of-Interest (POI) data of Shenzhen city to implement the funtion2vec model based on the node2vec model and urban tree theory. In this model, we first divide UFE patterns into three categories: Function Replace (FR), Function Newly Added (FNA), and Function Vanishing (FV). Then, we calculate the correlation between those UFE patterns using their functional vectors, resulting in a graph structure representing the urban function evolution network (UFEN). Based on our case study, we obtained the following conclusions: (1) From 2013 to 2022, the UFE in Shenzhen was primarily dominated by FR (89.44%). (2) FV and FNA exhibit a long-tailed distribution, adhering to the 20–80 law. (3) Through the UFEN based on FR, healthcare services are well suited to form mutual complementarities with other functions; science, education, and cultural services demand a higher complementarity with other functions; administrative offices exhibit a strong diversity in their evolutionary patterns; and the integration of transportation hubs with other functions results in a significantly deviating urban function evolution from its original pattern. The above conclusions suggest that function2vec can well express UFE in emerging cities by adding spatial correlation in UFE. [ABSTRACT FROM AUTHOR]

Traffic volume forecast is the key to alleviating traffic congestion. However, the relationship between traffic data and outside factors makes the problem more complex. Existing traffic flow forecasting studies seldom consider the relationship between traffic data and outside factors. Therefore, we propose SGA-KGCN-LSTM to solve this problem. The model combines Savitzky-Golay (SG) filter, Knowledge Graph (KG), Graph Convolution Network (GCN), Long Short-Term Memory (LSTM), and Self-Attention Mechanism. Firstly, the SG filter can be employed to reduce the noise of traffic volume data. Then, aiming at the relationship between traffic data and outside factors, the KG theory is introduced, and the knowledge representation is applied to get the embedding of relevant knowledge. Secondly, new road features are obtained by integrating embedded information and traffic flow characteristics. GCN can be employed to acquire the spatial characteristic of traffic stream, and LSTM can be employed to extract the temporal characteristics of traffic stream. Finally, the input feature information is given enough weight by self-attention mechanism. The outcome is subsequently acquired by utilizing the fully connected layer to attain the ultimate consequence. The trajectory data of the Luohu taxi in Shenzhen are used in the experiment. The experimental consequence indicate that the SGA-KGCN-LSTM has high forecasting precision compared with the benchmark model and the ablation experiment. [ABSTRACT FROM AUTHOR]

Due to the interaction of many factors in the stock market, stock price prediction has always been a challenging problem in the field of machine learning. In particular, the mutation factors of the stock market often have a great impact on subsequent predictions. The existing prediction models seldom consider the impacts of other stocks in the stock market and mutation points on the prediction accuracy of target stocks. Therefore, this paper presents a new knowledge graph and deep learning method combined with a stock price prediction network focusing on related stocks and mutation points. First, the target stock price features are obtained through the ConvLSTM network. Second, the knowledge graph is used to mine the hidden relationships between stocks to find the stocks relevant to the target stock to obtain the market information vector and the market information features through the ConvLSTM network. Then, we find the mutation points according to the price change range, construct the mutation point distance weight matrix according to the distance from each trading day to the mutation points, and obtain the mutation point information features through the graph convolutional network (GCN). Finally, the features of market information, mutation point information and target stock price are fused to jointly predict the future stock price. The experimental results on the A share of Shenzhen from 2010 to 2019 show that the algorithm has good robustness and that the prediction accuracy is effectively improved. [ABSTRACT FROM AUTHOR]

Land subsidence is a geological disaster. It will lead to the decline of land elevation, resulting in the potential safety hazards of urban facilities. Thus, the prediction of land subsidence displacement is significant. Among the existing prediction methods, the methods based on the time-series prediction model only analyze the settlement series data without considering the settlement mechanism, so they are easy to apply. However, they less consider the influence of other factors on land subsidence. Besides, they independently input displacement time-series data from different monitoring points without considering their relationship. To solve these problems, we take the monitoring point as the entity and take the DEM, soil type, building height, and land subsidence displacement sequence at the corresponding position of the monitoring point as the attributes to construct the knowledge graph. And then, we propose a framework Graph-TAP for modeling temporal attribute prediction of the knowledge graph's entity. This framework learns the representation of events with the specific entity at first. Then it captures the temporal dependency between historical events using GRU. Finally, it predicts the entity's displacement attribute. We randomly selected 61 subsidence monitoring points in Shenzhen, China. We used the land subsidence displacement InSAR time-series data (12-day time interval) and other attribute data from June 22, 2015, to April 5, 2016, for model training, validation, and testing. The experimental results show that our method is better than the time-series prediction based on the LSTM model and the DARNET (a knowledge graph temporal attribute prediction framework). [ABSTRACT FROM AUTHOR]

• This paper models the metro system as knowledge graph for passenger flow prediction. • It combines traffic patterns and land-use features for knowledge graph construction. • It proposes a SARGCN model for spatiotemporal prediction on metro knowledge graphs. • It uses an attention mechanism to learn the correlation between inflow and outflow. • Validated on two metro datasets, it outperforms numerous advanced baselines. With the rapid development of intelligent operation and management in metro systems, accurate network-scale passenger flow prediction has become an essential component in real-time metro management. Although numerous novel methods have been applied in this field, critical barriers still exist in integrating travel behaviors and comprehensive spatiotemporal dependencies into prediction. This study constructs the metro system as a knowledge graph and proposes a split-attention relational graph convolutional network (SARGCN) to address these challenges. Breaking the limitations of physical metro networks, we develop a metro topological graph construction method based on the historical origin–destination (OD) matrix to involve travel behaviors. Then, we design a metro knowledge graph construction method to incorporate land-use features. To adapt prior knowledge of metro systems, we subsequently propose the SARGCN model for network-scale metro passenger flow prediction. This model integrates the relational graph convolutional network (R-GCN), split-attention mechanism, and long short-term memory (LSTM) to explore the spatiotemporal correlations and dependence between passenger inflow and outflow. According to the model validation conducted on the metro systems in Shenzhen and Hangzhou, China, the SARGCN model outperforms the advanced baselines. Furthermore, quantitative experiments also reveal the effectiveness of its component and the constructed metro knowledge graph. [ABSTRACT FROM AUTHOR]