基于LSTM神经网络的采煤机自适应截割研究.pdf

全日制全日制硕士硕士学位论文学位论文 基于 LSTM 神经网络的采煤机自适应截割研 究 Research on Self-adaption Cutting of Shearer Based on LSTM Neural Network 作者姓名 南鹏飞 导师姓名 陈伟华 副教授 学科专业 控制理论与控制工程 研究方向 工业过程综合自动化 完成日期 2020 年 6 月 11 日 辽宁工程技术大学 Liaoning Technical University 万方数据 基 于 L S T M 神 经 网 络 的 采 煤 机 自 适 应 截 割 研 究 南 鹏 飞 辽 宁 工 程 技 术 大 学 万方数据 万方数据 中图分类号 TD421.6 学校代码 10147 UDC 621.3 密 级 公 开 辽宁工程技术大学 全日制全日制硕士学位论文硕士学位论文 基于 LSTM 神经网络的采煤机自适应截割研究 Research on Self-adaption Cutting of Shearer Based on LSTM Neural Network 作者姓名 南鹏飞 学 号 471720652 导师姓名 陈伟华 （副教授） 副导师姓名 申请学位 工学硕士 培养单位 电气与控制工程学院 学科专业 控制理论与控制工程 研究方向 工业过程综合自动化 二○二○年六月 万方数据 致致 谢谢 光阴似箭，时间飞逝，转眼到了毕业的时间，在研究生学习和生活即将结束之际，向 七年里一直指导和鼓励我的老师、同学、朋友和亲人致以最崇高的敬意和最真挚的谢意， 同时也感谢我的母校辽宁工程技术大学对我的培养。 回首完成研究生学业的三年时光里，首先要感谢的是我的导师陈伟华老师和闫孝姮老 师。无论是生活还是学业，两位老师无数次在我犹豫、迷茫和信心不足的时候给予我关心 和鼓励，给了我一次又一次坚持下去的勇气和力量。本论文的研究工作也是在陈老师和闫 老师的悉心关怀和精心指导下完成的。在我学业和论文工作中无不倾注着导师辛勤的汗水 和心血。陈老师谦虚认真的处世之道和勤奋扎实的工作态度给我留下了深刻的印象，是我 学习的榜样。而闫老师严谨、忘我的工作精神以及不断进取、精益求精的治学态度同样不 断激励着我前进。在此谨向两位老师表示衷心的感谢 感谢同实验室的朱正印师兄以及姜兆迪、邢鹏亮、周明亮等各位师弟师妹，感谢他们 在完成论文过程中给予的积极帮助。感谢胡善华、周奇、孔繁会和吕培贤同学在学校生活 中给我的帮助，感谢他们的陪伴。我将会永远铭记这段难忘的回忆。 感谢阜新工大中远电器有限公司的周程勇先生、周程鹏先生等人，感谢他们在我实习 期间给予的帮助和指导。 最后，感谢百忙之中的评审专家和前辈们，感谢你们的专业指导和建议。 万方数据 I 摘摘 要要 记忆截割对于提高采煤机自动化水平，促进综采工作面“无人化、少人化”发展具有重 要意义。但传统采煤机记忆截割技术存在截割精度低，需要采煤机司机频繁手动调整截割 滚筒高度的问题。为了提高记忆截割的截割精度与截割效率，本文在前人的研究基础上， 将深度学习与记忆截割技术相结合，并对此进行研究。 本文对采煤机记忆截割原理进行了研究后，建立坐标系对采煤机进行定位定姿分析并 获取采煤机的截割轨迹。 并提出一种基于深度长短时记忆 （Long Short-term Memory， LSTM） 神经网络的采煤机截割轨迹预测模型，利用 MATLAB 软件对深度 LSTM 神经网络模型的 预测效果进行了分析验证。此外由于采煤机在工作中需要进行连续多刀的截割作业，所以 需要神经网络连续进行多步预测。为了适应工作需求，提升模型多步预测能力，提出一种 MSLSTM 神经网络模型，减缓了预测步数的增加时，深度 LSTM 神经网络误差增大问题 的发生。最后，利用迁移学习相关知识，对小规模截割数据集进行了截割轨迹预测建模和 研究。针对采煤机截割轨迹预测任务，对应用不同学习策略的模型训练效果进行了比较， 并对迁移学习策略的有效性进行了对比分析。通过对深度学习和记忆截割的研究，为其今 后的发展和应用提供了有效参考。 该论文有图 29 幅，表 13 个，参考文献 49 篇。 关键词关键词采煤机；记忆截割；深度学习；LSTM；迁移学习 万方数据 II Abstract Memory cutting is of great significance to improve the automation of Shearer. It is also beneficial to the development of reducing workers in the working face. But the traditional shearer memory cutting has the problem of low cutting accuracy and requiring the shearer driver to adjust the cutting drum height manually and frequently.In order to improve the accuracy and efficiency of memory cutting, this thesis combines deep learning with memory cutting technology, and studies it on the basis of previous studies. In this thesis, after the study of the memory cutting principle of the shearer, the coordinate system is established to analyze the Shearer to get the cutting track of the drum.Based on the Long Short-term Memory LSTM neural network, a cutting trajectory prediction model of shearer is proposed. The prediction effect of deep LSTM neural network model is analyzed and verified by MATLAB software.In addition, because the shearer needs to carry out continuous cutting in the work, it needs neural network to predict the cutting trajectory continuously.In order to satisfy the work requirements and improve the multi-step prediction ability of the model, MSLSTM neural network model is proposed. MSLSTM neural network mitigate the problem that the error of deep LSTM neural network increases with the increase of prediction stepsFinally, the model and research of cutting path prediction based on small-scale dataset are carried out by using transfer learning.The training effects of different learning strategies are compared, and the effectiveness of transfer learning strategies is analyzed. Through the research of deep learning and memory cutting, this thesis provides an effective reference for the future development and application of memory cutting. Keywords Shearer; memory cutting; deep learning; LSTM; transfer learning 万方数据 III 目目 录 摘摘 要要 ............................................................................................................................................. I 目目 录录 .......................................................................................................................................... III 表清单表清单 ......................................................................................................................................... VII 图清单图清单 ....................................................................................................................................... VIII 变量注释表变量注释表 .................................................................................................................................... X 1 绪论绪论 ............................................................................................................................................. 1 1.1 选题背景及研究意义 .............................................................................................................. 1 1.2 采煤机自适应控制的国内外研究现状与趋势 ...................................................................... 3 1.3 深度学习发展历程 .................................................................................................................. 7 1.4 本文的研究内容 ...................................................................................................................... 9 2 采煤机工作原理与姿态分析采煤机工作原理与姿态分析 ................................................................................................... 11 2.1 电牵引双滚筒采煤机的结构原理和工作过程 .................................................................... 11 2.2 记忆截割原理 ........................................................................................................................ 13 2.3 信息采样方法 ........................................................................................................................ 14 2.4 本章小结 ................................................................................................................................ 21 3 基于基于 LSTM 的截割轨迹预测建模与分析的截割轨迹预测建模与分析 ............................................................................. 22 3.1 深度学习的基本原理 ............................................................................................................ 22 3.2 深度学习模型的构建 ............................................................................................................ 23 3.3 数据说明与数据预处理 ........................................................................................................ 25 3.4 模型训练方法设计 ................................................................................................................ 26 3.5 模型参数的选取 .................................................................................................................... 28 3.6 本章小结 ................................................................................................................................ 31 4 基于基于 MSLSTM 的截割轨迹预测模型改进与仿真分析的截割轨迹预测模型改进与仿真分析 ....................................................... 32 4.1 深度 LSTM 神经网络与不同算法的对比与误差分析 ....................................................... 32 4.2 LSTM 神经网络的分析与改进 ............................................................................................. 33 4.3 MSLSTM 神经网络的模型构建与对比 ............................................................................... 35 万方数据 IV 4.4 MSLSTM 神经网络与不同算法的对比与误差分析 ........................................................... 37 4.5 本章小结 ................................................................................................................................ 38 5 基于迁移学习的截割轨迹预测模型研究基于迁移学习的截割轨迹预测模型研究 ............................................................................... 40 5.1 迁移学习的基本原理 ............................................................................................................ 40 5.2 迁移学习策略选择 ................................................................................................................ 41 5.3 迁移学习神经网络模型构建 ................................................................................................ 42 5.4 数据说明 ................................................................................................................................ 43 5.5 不同迁移学习策略效果对比 ................................................................................................ 44 5.6 迁移学习效果验证 ................................................................................................................ 44 5.7 本章小结 ................................................................................................................................ 47 6 结论与展望结论与展望 ............................................................................................................................... 48 6.1 结 论 ...................................................................................................................................... 48 6.2 展 望 ...................................................................................................................................... 48 参考文献参考文献 ....................................................................................................................................... 49 作者简历作者简历 ....................................................................................................................................... 51 学位论文原创性声明学位论文原创性声明 ................................................................................................................... 52 学位论文数据集学位论文数据集 ........................................................................................................................... 53 万方数据 V Contents Abstract .......................................................................................................................................... I Contents ....................................................................................................................................... III List of Figures .......................................................................................................................... VIII List of Tables .............................................................................................................................. VII List of Variables ....................................................................................................................... VIII 1 Introduction ................................................................................................................................ 1 1.1 Background and significance of the topic ................................................................................. 1 1.2 Research status and trend of adaptive control of shearer at home and abroad .......................... 3 1.3 Development process of deep learning ..................................................................................... 7 1.4 Technical Route ......................................................................................................................... 9 2 Working principle and attitude analysis of shearer .............................................................. 11 2.1 Structure principle and working process of electric traction double drum shearer ................. 11 2.2 Principle of memory cutting .................................................................................................... 13 2.3 Ination sampling s ................................................................................................ 14 2.4 Summary ................................................................................................................................. 21 3 Modeling and analysis of cutting trajectory prediction based on LSTM ........................... 22 3.1 Fundamentals of deep learning ............................................................................................... 22 3.2 Construction of deep learning model ...................................................................................... 23 3.3 Data description and data preprocessing ................................................................................. 25 3.4 Model training design ................................................................................................. 25 3.5 Selection of model parameters ................................................................................................ 28 3.6 Summary ................................................................................................................................. 31 4 Improvement and simulation analysis of cutting trajectory prediction model based on MSLSTM ...................................................................................................................................... 32 4.1 Comparison and error analysis of depth LSTM neural network and different algorithms ..... 32 4.2 Analysis and improvement of LSTM neural network ............................................................. 33 4.3 Model construction and training based on mslstm neural network ......................................... 35 万方数据 VI 4.4 Comparison and error analysis between MSLSTM neural network and different algorithms 37 4.5 Summary ................................................................................................................................. 38 5 Prediction model of cutting path based on transfer learning ............................................... 40 5.1 Basic principles of transfer learning ........................................................................................ 40 5.2 Construction of neural network model for transfer learning ................................................... 41 5.3 Data description ....................................................................................................................... 42 5.4 Transfer learning strategy selection ......................................................................................... 43 5.5 Comparison of the effect of different transfer learning strategies ........................................... 44 5.6 Verification of transfer learning effect .................................................................................... 44 5.7 Summary ................................................................................................................................. 47 6 Conclusion and prospect .......................................................................................................... 48 6.1 Conclusion ............................................................................................................................... 48 6.2 Outlook .................................................................................................................................... 48 References .......................................................................................................................