arXiv subjects

Ynte Vanderhoydonc

Publications and source records attributed to Ynte Vanderhoydonc.

4 recordsLinked to original sources

arXiv2025

A Hybrid Inductive-Transductive Network for Traffic Flow Imputation on Unsampled Locations

Accurately imputing traffic flow at unsensed locations is difficult: loop detectors provide precise but sparse measurements, speed from probe vehicles is widely available yet only weakly correlated with flow, and nearby links often exhibit strong heterophily in the scale of traffic flow (e.g., ramps vs. mainline), which breaks standard GNN assumptions. We propose HINT, a Hybrid INductive-Transductive Network, and an INDU-TRANSDUCTIVE training strategy that treats speed as a transductive, network-wide signal while learning flow inductively to generalize to unseen locations. HINT couples (i) an inductive spatial transformer that learns similarity-driven, long-range interactions from node features with (ii) a diffusion GCN conditioned by FiLM on rich static context (OSM-derived attributes and traffic simulation), and (iii) a node-wise calibration layer that corrects scale biases per segment. Training uses masked reconstruction with epoch-wise node sampling, hard-node mining to emphasize difficult sensors, and noise injection on visible flows to prevent identity mapping, while graph structure is built from driving distances. Across three real-world datasets, MOW (Antwerp, Belgium), UTD19-Torino, and UTD19-Essen, HINT consistently surpasses state-of-the-art inductive baselines. Relative to KITS, HINT reduces MAE on MOW by $\approx42$% with basic simulation and $\approx50$% with calibrated simulation; on Torino by $\approx22$%, and on Essen by $\approx12$%. Even without simulation, HINT remains superior on MOW and Torino, while simulation is crucial on Essen. These results show that combining inductive flow imputation with transductive speed, traffic simulations and external geospatial improves accuracy for the task described above.

cs.LG↗

arXiv2014

Constrained Runs algorithm as a lifting operator for the Boltzmann equation

Lifting operators play an important role in starting a kinetic Boltzmann model from given macroscopic information. The macroscopic variables need to be mapped to the distribution functions, mesoscopic variables of the Boltzmann model. A well-known numerical method for the initialization of Boltzmann models is the Constrained Runs algorithm. This algorithm is used in literature for the initialization of lattice Boltzmann models, special discretizations of the Boltzmann equation. It is based on the attraction of the dynamics toward the slow manifold and uses lattice Boltzmann steps to converge to the desired dynamics on the slow manifold. We focus on applying the Constrained Runs algorithm to map density, average flow velocity, and temperature, the macroscopic variables, to distribution functions. Furthermore, we do not consider only lattice Boltzmann models. We want to perform the algorithm for different discretizations of the Boltzmann equation and consider a standard finite volume discretization.

cs.CE↗

arXiv2014

Initialization of lattice Boltzmann models with the help of the numerical Chapman-Enskog expansion

We extend the applicability of the numerical Chapman-Enskog expansion as a lifting operator for lattice Boltzmann models to map density and momentum to distribution functions. In earlier work [Vanderhoydonc et al. Multiscale Model. Simul. 10(3): 766-791, 2012] such an expansion was constructed in the context of lifting only the zeroth order velocity moment, namely the density. A lifting operator is necessary to convert information from the macroscopic to the mesoscopic scale. This operator is used for the initialization of lattice Boltzmann models. Given only density and momentum, the goal is to initialize the distribution functions of lattice Boltzmann models. For this initialization, the numerical Chapman-Enskog expansion is used in this paper.

cs.CE↗

arXiv2012

Numerical extraction of a macroscopic pde and a lifting operator from a lattice Boltzmann model

Lifting operators play an important role in starting a lattice Boltzmann model from a given initial density. The density, a macroscopic variable, needs to be mapped to the distribution functions, mesoscopic variables, of the lattice Boltzmann model. Several methods proposed as lifting operators have been tested and discussed in the literature. The most famous methods are an analytically found lifting operator, like the Chapman-Enskog expansion, and a numerical method, like the Constrained Runs algorithm, to arrive at an implicit expression for the unknown distribution functions with the help of the density. This paper proposes a lifting operator that alleviates several drawbacks of these existing methods. In particular, we focus on the computational expense and the analytical work that needs to be done. The proposed lifting operator, a numerical Chapman-Enskog expansion, obtains the coefficients of the Chapman-Enskog expansion numerically. Another important feature of the use of lifting operators is found in hybrid models. There the lattice Boltzmann model is spatially coupled with a model based on a more macroscopic description, for example an advection-diffusion-reaction equation. In one part of the domain, the lattice Boltzmann model is used, while in another part, the more macroscopic model. Such a hybrid coupling results in missing data at the interfaces between the different models. A lifting operator is then an important tool since the lattice Boltzmann model is typically described by more variables than a model based on a macroscopic partial differential equation.

cs.CE↗