Advances in Predictive Modeling for Complex Systems

The field of predictive modeling is rapidly advancing, with a focus on developing innovative solutions for complex systems. Recent developments have highlighted the importance of integrating multiple tasks and data sources to improve forecasting accuracy and capture intricate temporal dependencies. Graph neural networks have emerged as a powerful tool for localized and high-resolution forecasting, with applications in temperature forecasting, ice sheet simulations, and solar photovoltaic system monitoring. Additionally, domain-decomposed graph neural networks have shown promise in surrogate modeling for large-scale simulations, enabling efficient and accurate predictions. Noteworthy papers include:

  • A Multi-Task Temporal Fusion Transformer framework for joint sales and inventory forecasting, which achieved significant reductions in sales and inventory errors.
  • A Graph Neural Network Approach for localized and high-resolution temperature forecasting, which demonstrated impressive accuracy in capturing temperature patterns.
  • BlendedNet++, a large-scale aerodynamic dataset and benchmark for blended wing body aircraft, which provides a unified forward and inverse protocol for reproducible research in field-level aerodynamics and inverse design.

Sources

Multi-Task Temporal Fusion Transformer for Joint Sales and Inventory Forecasting in Amazon E-Commerce Supply Chain

A Graph Neural Network Approach for Localized and High-Resolution Temperature Forecasting

Domain-Decomposed Graph Neural Network Surrogate Modeling for Ice Sheets

Off-grid solar energy storage system with lithium iron phosphate (LFP) batteries in high mountains: a case report of Tianchi Lodge in Taiwan

Temporal Graph Neural Networks for Early Anomaly Detection and Performance Prediction via PV System Monitoring Data

BlendedNet++: A Large-Scale Blended Wing Body Aerodynamics Dataset and Benchmark

Built with on top of