基于长短时记忆网络的光伏发电功率预测

Prediction of Photovoltaic Power Generation Based on Long Short-Term Memory Network

光伏发电系统的输出功率具有波动性和间歇性,其特性影响了电力系统安全、稳定与经济地运行,因此准确预测光伏发电系统的输出功率具有十分重要的意义。目前,光伏出力预测一般使用比较简单的网络,如BP神经网络和SVM等,并且大多数预测的时间级为小时级,而对于分钟级的预测具有一定的难度。光伏出力预测是一个回归问题,而长短时记忆(LSTM)在时间序列上具有良好的处理效果。本文研究影响光伏发电的因素,并从中选取主要因素作为特征,通过构建基于LSTM的深度学习模型来预测光伏发电功率。在不同天气情况下,光伏发电功率的波形具有不同的特征,因此对不同天气类型构建不同的LSTM预测模型。实测数据表明,不同天气类型的LSTM模型具有更忧的性能。

Bibliometrics is a quantitative analysis method to master the trend of discipline development. In traditional bibliometrics research, the research steps are manual and tedious. In order to improve it, an analysis system based on scrapy-redis, a distributed crawler framework, are designed and implemented to master the trend of discipline development. The system includes: 1.data processing layer.It is responsible for crawling literature information on web of science. The problem that spiders lose some web pages due to unstable network speed is solved,ensuring data integrity. Design an dynamic algorithm to calculate the discipline distribution according to references.2. web service layer. It is made by Django and used for showing the analysis result. User only needs to input the initial URL that user wants to crawl,then the user just need to wait for the analysis result and require it by web service. The system provides a more convenient, more rapid and high scalability approach to do bibliometrics research.