Implementation of Multimodal Information Fusion Methods for Intelligent Converter Stations

This paper proposes a multimodal information fusion device designed for use in converter station scenarios, aimed at improving the efficiency of information fusion technologies. The device performs effective partitioning and parallel processing of multimodal fusion tasks based on the spatial physical location of the demand terminal and the heterogeneous characteristics of communication and computing power across network nodes. This approach enables efficient multimodal information fusion tailored to the specific spatial location of demand terminals. By establishing a standardized set of operations for representation, translation, and alignment, the device constructs a stream computing graph to handle complex multimodal information fusion tasks. During this process, it gathers spatial information from different demand sources, while taking into account factors such as node computing power, I/O, and energy supply, ensuring rational partitioning and mapping of the computational flow. This method significantly reduces the computational intensity of multimodal data fusion and enhances the efficiency of distributed multimodal information fusion in intelligent sensing scenarios at converter stations.