Real-world time series often have multiple frequency components that are intertwined with each other, making accurate time series forecasting challenging. Decomposing the mixed frequency components into multiple single frequency components is a natural choice. However, the information density of patterns varies across different frequencies, and employing a uniform modeling approach for different frequency components can lead to inaccurate characterization. To address this challenges, inspired by the flexibility of the recent Kolmogorov-Arnold Network (KAN), we propose a KAN-based Frequency Decomposition Learning architecture (TimeKAN) to address the complex forecasting challenges caused by multiple frequency mixtures. Specifically, TimeKAN mainly consists of three components: Cascaded Frequency Decomposition (CFD) blocks, Multi-order KAN Representation Learning (M-KAN) blocks and Frequency Mixing blocks. CFD blocks adopt a bottom-up cascading approach to obtain series representations for each frequency band. Benefiting from the high flexibility of KAN, we design a novel M-KAN block to learn and represent specific temporal patterns within each frequency band. Finally, Frequency Mixing blocks is used to recombine the frequency bands into the original format. Extensive experimental results across multiple real-world time series datasets demonstrate that TimeKAN achieves state-of-the-art performance as an extremely lightweight architecture. Code is available at https://github.com/huangst21/TimeKAN.

Multi-source data enhances demand prediction performance by learning from multiple transport modes simultaneously. However, existing multimodal demand forecasting methods often require direct sharing of raw data, which can be infeasible or at least very difficult, due to privacy concerns or practical constraints posed by ownership of data from different institutions. This study proposes a Multimodal Transport Demand Forecasting model via Federated Learning (FL) to improve forecasting accuracy without the need for direct data sharing. In the model, a processing center is introduced to handle parameters of forecasting models trained by each private dataset, where the exact dataset does not have to be shared and sensitive private information cannot be identified. Specifically, a Fine-grained Graph Convolution Recurrent Network (F-GCRN) is designed to capture spatiotemporal correlations of each dataset, with stronger capabilities to handle dynamic latent dependencies among different demand patterns than existing multimodal demand forecasting models. The processing center distinguishes the importance of parameters sent by different modes based on the Attentive Federated Learning mechanism and returns the processing parameters to each institution. Each institution predicts the demand with the returned parameters. Evaluations on three real-world transport datasets demonstrate that the proposed model outperforms several baselines and state-of-the-art models. Overall, this study addresses the research gap of enhancing multimodal demand forecasting without the dependence on direct data sharing and illustrates that knowledge sharing via parameters sharing by FL can improve multimodal transport demand prediction.

Understanding and forecasting mobility patterns and travel demand are fundamental and critical to efficient transport infrastructure planning and service operation. However, most existing studies focused on deterministic demand estimation/prediction/analytics. Differently, this study provides confidence interval based demand forecasting, which can help transport planning and operation authorities to better accommodate demand uncertainty/variability. The proposed Origin-Destination (OD) demand prediction approach well captures and utilizes the correlations among spatial and temporal information. In particular, the proposed Probabilistic Graph Convolution Model (PGCM) consists of two components: (i) a prediction module based on Graph Convolution Network and combined with the gated mechanism to predict OD demand by utilizing spatio-temporal relations; (ii) a Bayesian-based approximation module to measure the confidence interval of demand prediction by evaluating the graph-based model uncertainty. We use a large-scale real-world public transit dataset from the Greater Sydney area to test and evaluate the proposed approach. The experimental results demonstrate that the proposed method is capable of capturing the spatial-temporal correlations for more robust demand prediction against several established tools in the literature.

Accurate demand forecasting of different public transport modes (e.g., buses and light rails) is essential for public service operation. However, the development level of various modes often varies significantly, which makes it hard to predict the demand of the modes with insufficient knowledge and sparse station distribution (i.e., station-sparse mode). Intuitively, different public transit modes may exhibit shared demand patterns temporally and spatially in a city. As such, we propose to enhance the demand prediction of station-sparse modes with the data from station-intensive mode and design a Memory-Augmented Multi-task Re current Network (MATURE) to derive the transferable demand patterns from each mode and boost the prediction of station-sparse modes through adapting the relevant patterns from the station-intensive mode. Specifically, MATURE comprises three components: 1) a memory-augmented recurrent network for strengthening the ability to capture the long-short term information and storing temporal knowledge of each transit mode; 2) a knowledge adaption module to adapt the relevant knowledge from a station-intensive source to station-sparse sources; 3) a multi-task learning framework to incorporate all the information and forecast the demand of multiple modes jointly. The experimental results on a real-world dataset covering four public transport modes demonstrate that our model can promote the demand forecasting performance for the station-sparse modes.

Modeling complex spatial and temporal correlations in the correlated time series data is indispensable for understanding the traffic dynamics and predicting the future status of an evolving traffic system. Recent works focus on designing complicated graph neural network architectures to capture shared patterns with the help of pre-defined graphs. In this paper, we argue that learning node-specific patterns is essential for traffic forecasting while pre-defined graph is avoidable. To this end, we propose two adaptive modules for enhancing Graph Convolutional Network (GCN) with new capabilities: 1) a Node Adaptive Parameter Learning (NAPL) module to capture node-specific patterns; 2) a Data Adaptive Graph Generation (DAGG) module to infer the inter-dependencies among different traffic series automatically. We further propose an Adaptive Graph Convolutional Recurrent Network (AGCRN) to capture fine-grained spatial and temporal correlations in traffic series automatically based on the two modules and recurrent networks. Our experiments on two real-world traffic datasets show AGCRN outperforms state-of-the-art by a significant margin without pre-defined graphs about spatial connections.