1. Independently implement heterogeneous map road fusion, automatically merge maps of the same region from two different sources, eliminating manual costs while updating map information in a timely manner.2. Perform feature embedding on the side features of link data to calculate scores; use a vectorized space Rtree to index and store geometry; Calculate spatial similarity and matching score for road links in each local mesh area separately; Perform weighted fusion to obtain the overall score, and then use subgraph isomorphism to achieve link matching; Finally, the topology structure between links is utilized for continuous segments similarity filtering, greatly improving precision (68.24->99.21) and recall (93.51->98.43)3. Apply this method to relevant businesses for online testing; in the situation of map diff discrimination, accuracy increased(51.3->78.3)

1. Responsible for the algorithm model architecture and updates of the heating project in Dongying District;2. Using future data such as weather forecasts to predict the size of gear adjustment for heating and cooling units; Real time prediction of electricity consumption using sensors at various points and hardware information of the unit.3. Realize intelligent regulation of the unit, saving 30% of electricity.