A Neural Network Model for Digitizing Enterprise Carbon Assets Based on Multimodal Knowledge Mapping.

Abstract

In this paper, a multimodal knowledge mapping approach is used to digitize enterprise carbon assets, and a corresponding neural network model is designed for use in the practical process. Rich textual entity labels associated with images are obtained using an entity annotation system. A topology-based data fusion method is also designed based on the hierarchical relationship between WordNet and DBpedia to fuse the knowledge obtained from image visualization and text description mining. Existing neural network-based entity linking methods ignore the semantic gap between the context of sequential entity denotative items and the context of graph-structured entities, thus affecting the accuracy of entity linking. It is observed that the importance of words in the context of entity denotative items is different, and the importance of content in the entity context is also different. To solve the above problems, this paper proposes an entity linking method that combines a common attention mechanism with a graph convolutional neural network. Secondly, based on the basic theory of value assessment, the characteristics of classical asset valuation methods and their inapplicability to the valuation of carbon assets are analyzed, and thus the real option valuation method and its two classical models are introduced; after demonstrating the real option characteristics of carbon assets of power enterprise projects, a real option model-based carbon asset valuation model for power enterprise projects is constructed and its applicability is verified with case studies. Through analyzing the current situation and problems of carbon asset valuation work in power enterprises, targeted practical suggestions are put forward to further strengthen and enhance the carbon asset valuation work in power enterprises in the future.