Abstract:

The capture of spatial computational features and the prediction of their intensity are pivotal challenges in the field of high-performance geocomputation and are crucial under the paradigm of intelligent computing. Traditional expert knowledge-driven research paradigms are limited in scalability, modeling complexity, and accuracy, leading to load imbalance and resource wastage. Addressing these issues, this paper introduces an intelligent parallel geocomputation engine framework, synergizing meta-intelligence, perceptual intelligence, and cognitive intelligence in geocomputation. At the meta-intelligence level, a universal feature representation space for spatial computational domains is constructed, considering morphological structures, set quantities, spatial distributions, and topological relationships. The perceptual intelligence level incorporates customized machine learning and deep learning models to enable automatic perception of computational intensity in geospatial domains. At the cognitive intelligence level, an adaptive dynamic scheduling strategy is proposed, combining high-intensity task prioritization and task stealing. The efficacy and efficiency of the intelligent parallel geocomputation engine are validated through two cases, including polygon intersection and viewshed analysis, demonstrating a more than 20-fold improvement in load balancing performance.

Key words: geocomputation, intelligent computing, spatial computational features, spatial computational intensity, load balance