关键词: 分布式光伏, 智能运维, 知识库, 知识图谱, 自然语言处理

Abstract: Aiming at the lack of domain-specific knowledge base in many distributed photovoltaic maintenance systems,a framework regarding how to construct a knowledge base intelligently for smart maintenance of distributed photovoltaic was proposed,which included two aspects that were entity extraction and entity relation extraction.To relax the potential susceptibility from sentence patterns,an improved TextRank algorithm based on word embedding was applied for entity extraction.To improve the unfitness between the classifier layer and data deficiency,a Piecewise Convolutional Neural Network (PCNN) model for multi-labeling was leveraged to extract entity relations.Categories of experiments showed that the improved approaches could help to provide a higher degree of automation and reduce labor costs with better accuracy.

Key words: distributed photovoltaic, smart maintenance, knowledge base, knowledge graph, natural language processing