Traffic prediction is a popular research topic in the field of Intelligent Transportation System (ITS), as it can allocate resources more reasonably, relieve traffic congestion, and improve road traffic efficiency. Graph neural networks are widely used in traffic prediction because they are good at dealing with complex nonlinear structures. Existing traffic prediction studies use distance-based graphs to represent spatial relationships, which ignores the deep connections between non-adjacent spatio-temporal information. The use of a simple approach to fuse spatio-temporal information is not conducive to obtaining long-term deep spatio-temporal dependencies. Therefore, we propose a new deep learning model Multi-Stage Spatio-Temporal Fusion Diffusion Graph Convolutional Network (MFDGCN). It generates multiple static and dynamic spatio-temporal association graphs to enhance features and adopts the multi-stage hybrid spatio-temporal fusion method. This promotes the effective fusion of a spatio-temporal multimodal and uses the diffuse convolution method to model the graph structure and time series in traffic prediction, respectively. The model can better predict both long and short-term traffic simultaneously. We evaluated MFDGCN using real road network traffic data and it shows good performance.

Multi-step traffic forecasting has always been extremely challenging due to constantly changing traffic conditions. Advanced Graph Convolutional Networks (GCNs) are widely used to extract spatial information from traffic networks. Existing GCNs for traffic forecasting are usually shallow networks that only aggregate two- or three-order node neighbor information. Because of aggregating deeper neighborhood information, an over-smoothing phenomenon occurs, thus leading to the degradation of model forecast performance. In addition, most existing traffic forecasting graph networks are based on fixed nodes and therefore need more flexibility. Based on the current problem, we propose Dynamic Adaptive Deeper Spatio-Temporal Graph Convolutional Networks (ADSTGCN), a new traffic forecasting model. The model addresses over-smoothing due to network deepening by using dynamic hidden layer connections and adaptively adjusting the hidden layer weights to reduce model degradation. Furthermore, the model can adaptively learn the spatial dependencies in the traffic graph by building the parameter-sharing adaptive matrix, and it can also adaptively adjust the network structure to discover the unknown dynamic changes in the traffic network. We evaluated ADSTGCN using real-world traffic data from the highway and urban road networks, and it shows good performance.

Recently, with the rise of deep learning, text classification techniques have developed rapidly. However, the existing work usually takes the entire text as the modeling object and pays less attention to the hierarchical structure within the text, ignoring the internal connection between the upper and lower sentences. To address these issues, this paper proposes a Bert-based hierarchical graph attention network model (BHGAttN) based on a large-scale pretrained model and graph attention network to model the hierarchical relationship of texts. During modeling, the semantic features are enhanced by the output of the intermediate layer of BERT, and the multilevel hierarchical graph network corresponding to each layer of BERT is constructed by using the dependencies between the whole sentence and the subsentence. This model pays attention to the layer-by-layer semantic information and the hierarchical relationship within the text. The experimental results show that the BHGAttN model exhibits significant competitive advantages compared with the current state-of-the-art baseline models.

Exploring the variation characteristics of actual evapotranspiration (ETa) and its response to climate change in the arid inland region of China is of great significance for strengthening regional water resources management and maintaining ecological environment security and stability. Taking the Dulan River Basin as the research area, based on the meteorological data from the Wulan Station and hydrological data from the Shanggaba Station from 1981 to 2020, the variation characteristics of ETa at the annual scale were analyzed. The ETa estimation model and joint distribution model of P and potential evapotranspiration (ET₀) was constructed based on climate factors, and the uncertainty of ETa response to climate change was explored with the water balance method, multiple linear regression, marginal distribution function, Copula function, and Monte Carlo algorithm. The results showed that the multi-year mean value of ETa in the study area was 261.6 mm, and the interannual process showed an insignificant upward trend, and had no abrupt change during the period. There were two obvious main cycles, which were 19-year periodic changes on the 30-year time scale and 6-year periodic changes on the 9-year time scale. The ETa estimation model based on precipitation (P) and ET₀ had good simulation accuracy. The optimal marginal distributions of P and ET₀ were Pearson-III (P-III) distribution and Generalized Extreme Value (GEV) distribution, respectively. The Copula joint distribution probability density of P and ET₀ was a symmetric saddle-shaped distribution. ETa showed an inverted ‘S’ distribution with the change in joint guarantee rate of P and ET₀, ranging from 116.9 mm to 498.6 mm. ETa was an interval range under a certain joint guarantee rate. The research results can provide support for the assessment of ETa, and help to further understand the driving mechanism of climate change on ETa in the arid inland region of China.