【作者】 杨淼；

【导师】 周广禄；

【作者基本信息】 哈尔滨工业大学 ， 计算机技术（专业学位）， 2021， 硕士

【摘要】 太阳能的广泛利用推动了光伏发电产业的发展,传统的电能采集方法易受人为因素影响,实时性差、可靠性低。另外,传统企业将物联网设备接入独立开发的系统或第三方平台进行功能显示和管理。随着设备类型的增加,系统的兼容性、可扩展性和稳定性都很差。同时,在物联网时代,数据也变得越来越重要。因此,建立以数据为中心、高可用性的光伏发电管理系统具有实际应用价值。针对上述问题,本文通过新型的光伏设备接入协议弥补原有固化协议的不足,有效保证系统安全性;利用多种传感器上报数据解析规则改善设备不兼容问题,保证系统的稳定性;改进Docker的调度算法提升集群的资源利用率。同时对光伏发电功率进行预测,及时对异常数据告警。本文针对以上现状进行研究,主要工作包括以下三个方面:(1)本文提出将卷积神经网络与长短期记忆网络相结合的混合神经网络架构应用于实际工业场景中,以提高光伏发电功率预测准确性。通过一维卷积预处理单变量的数据,并经过两层时间卷积运算后将其转变为多维数据,增强长短期记忆网络的预测能力。同时将单步预测扩展为多步预测策略,在实际工程应用中,仅利用系统中的光伏发电历史数据,即可进行短期、中期和长期的功率预测,证明该混合模型具有很好的鲁棒性。(2)本文将改进蚁群算法应用于Docker容器调度,解决资源利用率和负载不均衡问题。所提出的改进算法考虑了历史调度,从而增强调度决策。在相同的配置下,将该算法与基本蚁群算法(ACO)和先来先服务算法(FCFS)进行比较,实验结果表明,所提出的算法在响应时间和吞吐量方面具有优势,可以提高系统的整体性能。(3)本文设计并实现了一个数据可监控、可分析、可连通的光伏发电管理系统。采用前后端分离思想,针对传统物联网架构进行优化改进,引入容器化技术、使用消息队列异步消费等,搭建集数据采集模块、管理平台模块、光伏发电功率预测模块、数据存储模块和后台服务模块于一体的完整系统,通过功能性测试和非功能性测试验证了系统的实用性,并在系统正式上线投入使用中取得预期效果。更多还原

【Abstract】 The extensive use of solar energy has promoted the development of the photovoltaic power generation industry.The traditional electric energy collection methods are susceptible to human factors,and have poor real-time performance and low reliability.In addition,traditional enterprises connect their produced Io T devices to self-developed systems or third-party platforms for function display and management.As the number of types of devices increases,the compatibility,scalability,and stability of the system are poor.At the same time,in the era of the Internet of Things,the importance of data has become increasingly prominent.Therefore,the establishment of a data-centric,highly available photovoltaic power generation management system has practical application value.In response to the above problems,this article uses a new type of photovoltaic equipment access protocol to make up for the shortcomings of the original curing protocol,effectively ensuring system security;using multiple sensors to report data analysis rules to improve equipment incompatibility and ensure system stability;improve Docker Swarm The scheduling algorithm improves the resource utilization of the cluster.At the same time,it predicts the photovoltaic power generation and promptly warns of abnormal data.This article studies the above status quo,the main work includes the following three aspects:Firstly,this paper proposes to apply a hybrid neural network architecture combining convolutional neural network and long short-term memory network to actual industrial scenarios to improve the accuracy of photovoltaic power generation prediction.Preprocess single variable data through one-dimensional convolution,and transform it into multi-dimensional data after two-layer time convolution operation,which enhances the predictive ability of long-and short-term memory networks.At the same time,the single-step prediction is extended to a multi-step prediction strategy.In practical engineering applications,only the historical data of photovoltaic power generation in the system can be used to perform short-term,medium-term and long-term power prediction,which proves that the hybrid model is very robustness.Secondly,this article applies the improved ant colony algorithm to Docker Swarm container scheduling to solve the problem of resource utilization and load imbalance.The proposed improved algorithm considers historical scheduling,thereby enhancing scheduling decisions.Under the same configuration,the algorithm is compared with the basic ant colony algorithm(ACO)and the first-come-first-served algorithm(FCFS).The experimental results show that the proposed algorithm has advantages in response time and throughput,and can improve the system The overall performance.Finally,this paper designs and implements a photovoltaic power generation management system with data that can be monitored,analyzed,and connected.Adopt front-end and back-end separation ideas,optimize and improve the traditional Internet of Things architecture,introduce containerization technology,use message queue asynchronous consumption,etc.,build a collection of terminal equipment modules,access gateway modules,business and application support modules,application management platform modules and logs The complete system with the analysis module in one has verified the practicability of the system through functional tests and non-functional tests,and achieved the expected results when the system was officially put into use.更多还原