For the increasing travel demands and public transport problems, dynamically adjusting timetable or bus scheduling is necessary based on accurate real-time passenger flow forecasting. In order to get more accurate passenger flow in future, this paper proposes a novel hierarchical hybrid model based on time series model, deep belief networks (DBNs), and improved incremental extreme learning machine (Im-ELM) to forecast short-term passenger flow. The proposed model is named HTSDBNE with two modelling steps. First, referring the idea of parallelization, the hybrid model, constructed by time series model, DBN, and Im-ELM, is used to forecast short-term passenger flow in different time scales hierarchically and parallel. Second, Im-ELM is utilized to analyse the relationship of forecasting results from the first step, and the weighted outputs of Im-ELM are as the final forecasting results. Comparing with single forecasting models and typical hybrid forecasting models, the testing results indicate that HTSDBNE has better performances. The mean absolute percent error of prediction results is around 10% and fully meets the application requirements of bus operation enterprise.

For resolving or alleviating the transportation problems, it is necessary to efficiently manage the public transportation and provide public transport services with high quality and advocate green travel, which rely on accurate traffic data. In order to obtain more accurate bus speed in the future, this paper proposed a novel dynamic hierarchical spatial-temporal network model based on Grey Relation Analysis (EGRA), the convolutional neural network (CNN), and the gated recurrent unit (GRU). The proposed model is named the DHSTN; it exploited EGRA to analyze and choose the suitable candidate line sections with high impacts on the target section and, then, construct a multilayer structure based on the CNN, GRU, and attention mechanism to analyze and capture the spatial and temporal dependency, and finally, the extreme learning machine (ELM) is exploited for the fusion of the long-term and short-term dependency to predict the bus speed variation in the next time interval. Comparative experiments indicate that the DHSTN has better performances, the mean absolute error is around 2.6, and it meets the real requirements.

An extreme learning machine (ELM) is an innovative algorithm for the single hidden layer feed-forward neural networks and, essentially, only exists to find the optimal output weight so as to minimize output error based on the least squares regression from the hidden layer to the output layer. With a focus on the output weight, we introduce the orthogonal constraint into the output weight matrix, and propose a novel orthogonal extreme learning machine (NOELM) based on the idea of optimization column by column whose main characteristic is that the optimization of complex output weight matrix is decomposed into optimizing the single column vector of the matrix. The complex orthogonal procrustes problem is transformed into simple least squares regression with an orthogonal constraint, which can preserve more information from ELM feature space to output subspace, these make NOELM more regression analysis and discrimination ability. Experiments show that NOELM has better performance in training time, testing time and accuracy than ELM and OELM.

Constructing a trajectory index can efficiently improve the performances of trajectory data processing, provide basic supports for trajectory data mining. With the constantly growing of trajectory data scale and increasing demands for trajectory retrieval efficiency and accuracy, the indexing methods have become more and more crucial. The indexing method faces significant challenges in terms of spatiotemporal trajectory locality, imbalanced trajectory distribution and low trajectory data value density. To address these, we proposed an indexing method based on large-scale real-time trajectory data, it extends the vertical storage mode of HBase, designs the core index, and optimizes the design of the row key, refines the data retrieval process and provides specific mappings for each independent part of the dataset. Besides, it designs the primary index, implements a dynamic indexing mechanism, dynamically load relevant index based on query strategies to flexibly meet the complex query requirements. Comparative experiments demonstrate that the proposed index method is superior in range retrievals and trajectory retrievals, the responding speed is faster.

The changes of urban traffic flow is affected by many factors, and during predicting the traffic flow, it should consider each factor. The paper considers the characteristic between time and traffic flow, and introduces the time factor as a separated input parameter. Besides, some unexpected situations have large impacts on the predicting results, so, the unexpected factor is introduced as an improved factor to the input samples. Based on these, a novel prediction model is proposed, called at-ELM, which determines the hidden node number by using PCA, and reduces the impacts of the hidden node numbers on the predicting results. After performance tests and analysis, the novel model is a little improved in prediction accuracy and completely acceptable.

Accurate bus arrival time prediction (BATP) is crucial for improving public transportation reliability and passenger travel experience. However, bus operational data exhibit complex spatial dependencies among stops and strong temporal dynamics induced by periodic traffic patterns, posing significant challenges to accurate prediction. To address these issues, this paper proposes a novel bus arrival time prediction framework based on graph neural networks and hybrid time-frequency attention, termed GNN-HTFA. First, a heterogeneous graph is constructed to explicitly model spatial interactions between buses and stops. A graph attention network is then employed to learn node representations from the vehicle-stop graph via an adaptive attention mechanism, enabling dwell time prediction at bus stops. For inter-stop travel time prediction, a hybrid attention mechanism integrating time-domain and frequency-domain attention is introduced to jointly capture short-term dynamics and long-term periodic patterns. A Transformer-based model is further adopted to estimate inter-stop travel times. Experimental results on real-world bus trajectory datasets demonstrate that GNN-HTFA consistently outperforms state-of-the-art baselines in prediction accuracy. Ablation studies further verify the effectiveness of each module, indicating strong robustness under complex traffic conditions.