CASIA OpenIR  > 毕业生  > 博士学位论文
融合多源数据知识的复杂工业过程优化决策方法研究
刘承宝
Subtype博士
Thesis Advisor谭杰 ; 王学雷
2019-05-28
Degree Grantor中国科学院大学
Place of Conferral北京
Degree Discipline控制理论与控制工程
Keyword多源数据融合 知识获取 生成式对抗网络 智能优化决策 锂离子电池单体电芯一致性分选
Abstract

智能制造以企业高效化、绿色化与安全化为目标,以实现制造流程的智能优化决策与加工装备(过程)智能自主控制为特征,是我国制造业转型升级的必由之路。智能化作为智能制造的核心特征,体现于复杂工业过程全生命周期活动的诸多方面,如生产计划与调度、运行指标优化、异常诊断与分析等优化决策问题,它们涉及制造过程全生命周期活动中的人、设备、物料、工艺、环境等多个要素,具有多冲突目标、多冲突约束、多尺度动态优化等困难。目前,复杂工业过程优化决策主要依靠知识型工作者凭经验来完成,但其人工决策随意性大,难以及时捕捉生产过程全生命周期活动中发生的频繁或剧烈变化,凭借人工经验知识不能及时准确地做出决策反应,导致产品的质量差、成本高和资源消耗大等问题。

本文针对以上问题,围绕复杂工业过程中涉及全生命周期活动的优化决策问题,以实现工业大数据环境下多源数据知识高效获取、语义知识自优化及复杂工业过程智能优化决策为目标,开展工业大数据环境下知识获取、知识学习、知识决策方法的研究。本文的主要研究工作和贡献如下:

(1)提出了融合多源数据语义与特征的知识获取方法,能够有效获取复杂工业过程海量数据背后隐含的创新知识。本文给出了新的知识表示形式,提出了两阶段时间序列聚类的语义提取方法和基于卷积自编码器的特征提取方法,分别对复杂工业过程多源数据进行语义提取和特征提取,并融合含有语义信息的多源数据特征,结合其决策语义,获取知识元,构建了复杂工业过程语义知识库,为下一步研究语义知识自优化奠定了基础。

(2)提出了基于生成式对抗网络的语义知识自优化方法,能够高效预测少数类知识元样本的概率分布,生成更具多样性且准确的少数类知识元,从而解决了由于复杂工业过程中挖掘的高价值知识稀少而导致语义知识库中语义知识不均衡的问题,为建立基于语义知识的复杂工业过程智能优化决策模型提供了可靠保证。

(3)提出了基于语义知识的复杂工业过程智能优化决策方法。本文建立了以产品质量、效率、成本、能耗等综合指标为优化目标的全局优化模型,对贯穿于复杂工业过程全生命周期活动中的优化决策问题展开描述,并建立相应的度量描述模型。针对复杂工业过程产生的海量数据背后隐含的语义知识与优化决策问题之间存在的映射缺失问题,在语义知识库的基础上,构建了基于深度神经网络的优化决策知识推理模型,从而实现复杂工业过程智能优化决策。

(4)集成以上三个关键技术,以新能源汽车动力锂离子电池PACK中单体电芯的一致性分选优化决策问题为背景,设计了单动态特性曲线的单体电芯一致性分选优化决策方案和多动态特性曲线的单体电芯一致性分选优化决策方案,对本文所提方法的有效性进行了应用验证。本文开发了基于工业大数据平台的单体电芯一致性分选优化决策系统,并将其应用于实际生产过程。应用验证结果表明,单动态特性曲线的单体电芯一致性分选优化决策方案相比于目前实际生产中使用的传统分选方法,其分选的单体电芯不一致率平均降低了91.08%;多动态特性曲线的单体电芯一致性分选优化决策方案相比于单动态特性曲线的单体电芯一致性分选优化决策方案,其分选的单体电芯不一致率进一步平均降低了93.74%。综上所述,本文提出的方法能够有效筛选电化学性能不一致的单体电芯,从而提高动力锂离子电池PACK的一致性,改善PACK的耐久性、可靠性和安全性。

Other Abstract

Intelligent manufacturing, which is aimed at high-efficiency, green and safety of enterprises, and is characterized by intelligent optimization decision-making of manufacturing processes and intelligent autonomous control of machining equipment or processes, is the necessary way to realize the transformation and upgrade of Chinese manufacturing industry. As the core feature of intelligent manufacturing, intelligentization is reflected in many aspects of the life cycle activities in complex industrial process, such as optimization decision-making problems liking production planning and scheduling, operational index optimization, and abnormal diagnosis and analysis. In particular, the optimization decision-making problems are involved in a lot of factors in the life cycle activities of manufacturing processes, including a large number of operators, various production equipment and materials, complex technological rules, and uncertain production environment, thereby causing some difficulties in multiple conflict objectives, multiple conflict constraints and multi-scale dynamic optimization for solving them. At present, the optimization decision-making in complex industrial process are addressed by the knowledge workers with their experiences. However, because artificial decision has greater randomness and cannot capture in time the frequent or drastic changes in the life cycle activities of manufacturing processes, it is difficult to make timely and accurate decision-making response with experiential knowledge, resulting in poor product quality, high cost and high resource consumption and other issues.

Subject to the above problems, this paper focuses on the optimization decision-making problems involving the life cycle activities in complex industrial process and aims at realizing the efficient acquisition of multi-source data knowledge in industrial big data environment, semantic knowledge self-optimization and intelligent optimization decision-making of complex industrial process. The research on knowledge acquisition, knowledge learning and knowledge decision-making in a big data environment has been carried out. The detailed work and contributions of this paper have been summarized as follows:

(1) A knowledge acquisition approach based on fusing semantics and features of multi-source data is proposed, which mines effectively knowledge from massive data generated in the complex industrial processes. This paper defines a new form of knowledge representation and proposes a semantic extraction method based on the two-step time series clustering (SE-TTSC) and a feature extraction method based on the convolutional auto-encoder (FE-CAE). For multi-source data in complex industrial process, the paper extracts their semantics and features by SE-TTSC and FE-CAE, respectively. The features with semantics in different data sources are fused to obtain a fusion feature, combining its decision semantics to obtain a knowledge element. Therefore, a knowledge base consisting of many knowledge elements is built, which lays the foundation for the next research on the semantic knowledge base self-optimization.

(2) A semantic knowledge self-optimization approach based on the conditional generative adversarial networks is proposed, which predicts efficiently probability distribution of minority class knowledge elements, thereby generating the more diverse and accurate minority class knowledge elements. The proposed method addresses an unbalanced semantic knowledge problem caused by lack of high value knowledge mining from complex industrial process, thereby providing a reliable guarantee for establishing an intelligent optimization decision-making model based on semantic knowledge for complex industrial process.

(3) An intelligent optimization decision-making approach based on semantic knowledge for complex industrial process is proposed. This paper takes comprehensive indicators including product quality, efficiency, cost and energy consumption as different optimization goals to establish their global optimization description models and analyzes the optimization decision-making problems embedded in the life cycle activities of complex industrial process to establish their metric description models. For the lack of mapping function between optimization decision-making problems and the semantic knowledge mining from massive data generated by complex industrial process, the paper proposes the knowledge reasoning model based on the deep neural networks (DNNs) method for optimization decision-making problems, thereby realizing intelligent optimization decision-making for complex industrial process.

(4) Based on the integration of the above three key technologies, the effectiveness of the proposed approach is verified by the optimization decision-making problem for consistent lithium-ion cell screening in the new energy vehicle power lithium-ion battery PACK with multiple lithium-ion cells configured in series, parallel, and series–parallel. This paper proposes two application verification methods: the optimization decision-making method for consistent lithium-ion cell screening based on single dynamic characteristic curves of lithium-ion cells (ODM-CLCS-SDCC) and the optimization decision-making method for consistent lithium-ion cell screening based on multiple dynamic characteristic curves of lithium-ion cells (ODM-CLCS-MDCC). An optimization decision-making system for consistent lithium-ion cell screening based on the industrial big data platform was developed and applied to the actual production process. Application verification results show that the inconsistency rate of the screened cells using the ODM-CLCS-SDCC drops by an average of 91.08%, compared with the traditional screening method that is currently used in industrial processes; and compared with the ODM-CLCS-SDCC, the inconsistency rate of the screened cells using the ODM-CLCS-MDCC drops by an average of 93.74%. To summarize, the proposed approach can effectively screen inconsistent lithium-ion cells in electrochemical characteristics, thereby improving the consistency of lithium-ion battery PACKs and their durability, reliability and safety.

Pages152
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/23870
Collection毕业生_博士学位论文
综合信息系统研究中心
Recommended Citation
GB/T 7714
刘承宝. 融合多源数据知识的复杂工业过程优化决策方法研究[D]. 北京. 中国科学院大学,2019.
Files in This Item:
File Name/Size DocType Version Access License
融合多源数据知识的复杂工业过程优化决策方(60574KB)学位论文 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[刘承宝]'s Articles
Baidu academic
Similar articles in Baidu academic
[刘承宝]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[刘承宝]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.