【作者】 卢胜利；

【导师】 曹家麟；

【作者基本信息】 上海大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 实现引黄灌渠水流量连续自动测量和远程数据传输是实行可交易水权制度和建设数字灌区的重要技术基础之一。引黄灌渠斗口特指引黄干渠通向支渠的放水闸口(简称斗口),按照引黄灌区现行水资源管理办法,灌渠斗口水流量是水资源管理部门与用水单位结算水费的法定依据。利用闸口水工建筑物(灌渠斗口)量水不需要修筑专门的量水设施,可以避免量水设施带来的水头损失等弊端,是最为经济和广泛采用的灌渠量水方法。由于引黄灌渠泥砂含量大、淤积严重、漂浮物多、水流形态变化无常,一般只能通过观测干渠(闸前)和支渠(闸后)若干点水位以及启闸高度,依据闸口水工建筑物的类别和水流形态,选用基于水力学分析的量水公式和现场率定的综合流量系数来估算水流量,具有软测量的典型特征。本文基于国家自然科学基金项目“引黄灌渠斗口流量自动测量方法与装置研究”(60165001),主要开展了以下研究工作:(1)从实现可交易水权制度和引黄灌区数字化的目标出发,设计了基于短消息服务的引黄灌渠水流量远程监测网络总体方案,并研制了作为该网络重要现场单元的灌渠斗口水流量自动测量装置(ZL200420050359.1)以及与之配套的两种水位测量仪器(智能浮子水位计、智能电子水尺)和闸门开启度(位置)测量机构。为深入研究灌渠斗口水流量软测量的相关问题并对各种软测量模型进行训练和检验,还专门设计并搭建了引黄灌渠斗口缩尺水工模型。(2)按照软测量原理和工程开发规范,提出引黄灌渠斗口水流量软测量技术所包含的辅助变量选择、数据预处理、建立软测量模型和软模型在线校正四个方面的基本任务。首先将各种流态对应的量水公式转化为传统机理软测量模型,并由此确定了引黄灌渠斗口水流量软测量的辅助变量。在对各点水位信号特征分析的基础上,提出基于综合支持度的加权平均算法,以有效滤除水位测量数据中的异常值和随机干扰;还针对引黄灌渠斗口的特殊条件,提出了滚动式软测量模型在线校正策略。(3)为使基于传统水力学机理的软测量技术转化为可以实际应用的专家系统软件,设计并实现了基于产生式规则的引黄灌渠斗口水流量软测量专家系统。所提出的综合流量系数细分率定法并通过知识库扩充予以实现,使传统机理模型的估计精度和适应性得到一定程度的改善。(4)建立了引黄灌渠斗口水流量BP网络软测量模型,并采用水工实验数据对BP网络模型进行训练和检验,证明人工神经网络软测量模型在改善测量精度方面的成效。鉴于BP网络存在学习速度慢、易陷入局部极小以及泛化能力性能差等缺陷,建立了引黄灌渠斗口水流量RBF网络软测量模型,并采用自适应遗传算法(AGA)对RBF网络软测量模型参数进行优化。通过仿真实验证明RBF网络模型比BP网络模型更加适用。为适应嵌入式环境和水流量实时估计的要求,又建立了基于CMAC网络和水力学机理的混合软测量模型。从适应引黄灌渠斗口水流形态多样化以及训练样本稀疏性出发,还提出“分而治之”策略,建立了适应流态变化的软测量混合专家网络,使全流态复杂系统的软测量建模问题转化为针对单一流态简单系统的软测量建模问题。(5)针对实际中训练样本稀疏的问题,建立了基于支持向量回归(SVR)的软测量模型,并采用基于粒子群优化算法(PSO)对SVR软测量模型参数进行优化。相同的水工实验数据对采用不同核函数和损失函数的SVR软测量模型进行的仿真训练和性能检验结果表明,SVR软测量模型能够适应样本稀疏条件,比BP和RBF网络软测量模型估算精度高,而且泛化能力可从本质上得到改善。(6)现场量水实践和水工实验均表明,目前采用的量水公式在邻近流态边界处误差明显增大,由于实际中缺乏流态边界的训练样本,基于机器学习的软测量模型的泛化能力也会明显降低。在对引黄灌渠斗口水流态边界的模糊特性进行分析的基础上,提出流态区域参考模型以及综合流量系数模糊推理和数据融合相结合的软测量方案。在改善流态模糊边界水流量估计精度和模型适用性方面取得了一定的成效。本文的主要贡献与创新点主要体现在以下五个方面:其一,开发了适合引黄灌渠条件的灌渠斗口水流量自动测量装置,为软测量技术实现搭建了较为理想的硬件平台;其二,建立了基于水力学机理和综合流量系数细分率定的软测量专家系统,既为设计引黄灌渠斗口水流量软测量程序制定了一种规范,也为改善水流量估计精度提供了简便有效的方法;其三,选择BP、RBF、CMAC、SVR、AGA和PSO等多种软计算方法,建立了适应嵌入式环境和水流量实时估计以及训练样本稀缺的软测量模型,使软测量模型的估计精度以及适用性和可靠性得到全面改善。第四,提出并实际建立了基于模糊推理技术的混合软测量模型,有效提高了水力学机理模型的适用性和精确度;其五,提出由确定性区域和模糊边界构成的流态区域参考模型,采用基于距离测度的多模型数据融合技术,在提高流态模糊边界水流量估计精度和软测量技术适用性方面进行了有价值的探索。更多还原

【Abstract】 The continuous and automatic measurement of water discharge at lateral gate of the irrigation channel in the Yellow River and achievement of data remote transmission is one of important technical basis of implementing tradable water rights institution and digital irrigation area. The lateral gate of irrigation channel is specific reference to floodgate hydraulic building at main channel to branch canal in the Yellow River,the water discharge at lateral gate of the irrigation channel is legal base of water fee settlement between water management department and water users. Amongst all measurement methods for water discharge at lateral gate of irrigation channel, the method with floodgate hydraulic structure (the lateral gate) is the most economic one and has been used widely, because special discharge measuring facilities no longer been needed and hereby the flood-peak loss from the discharge measuring facilities can be avoided. Owing to many factors such as high content of mud and sand, heavy siltation, floating-debris, unsteady flow pattern and so on, the discharge at lateral gate of irrigation channel in the Yellow River is usually estimated by the water level at main channel (front of the floodgate), the water level at several points of branch canal (back of the floodgate) and the lift-height of floodgate. That is, the water discharge is calculated by the hydraulic measuring formula witch correspond with type of the hydraulic structure and present flow pattern as well as comprehensive discharge-coefficient from the in-situ calibration. This procedure has typical characteristics of soft-sensing.This dissertation is based on the fund project of national natural science (No.60165001)—The reseach on method and device for automatic measurement of water discharge at lateral gate of irrigation channel in the Yellow River. The following research works have been done under this fund project:(1) In order to realize tradable water rights institution and digital irrigation area in the Yellow River Basin, a overall scheme of remote discharge monitoring network based on SMS for irrigation channel in the Yellow River is designed, as important field unit, a kind of automatic measuring device for discharge at lateral gate of irrigation channel(ZL200420050359.1) is developed, including tow kinds of water level gauge (intelligent floater water lever sensor and Intelligent electric water-level rule) and measuring mechanism for the open position of floodgate matched with this device. For deep research on relevant problems about soft-sensing for discharge at lateral gate of irrigation channel as well as training and testing various soft-sensing models, a special reduced scale hydraulic model is designed and built.(2) Based on soft-sensing principle and engineering development specification, four technical items involved in soft-sensing for water discharge at lateral gate of the irrigation channel in the Yellow River are presented, they are secondary variables selecting, data preprocessing, soft-sensing model building and on-line calibration respectively. Firstly, water measuring formulas corresponding to specific flow patterns are involved into traditional mechanistic soft-sensing models and relevant secondary variables are determined. Based on the characteristic analysis of each water level signals, a comprehensive support degree based weighted average algorithm is proposed to effectively filter outliers and from random disturbance in water level measuring data. In view of special condition of lateral gate of the irrigation channel in the Yellow River, a rolling type on-line calibrating strategy for the soft-sensing models is also proposed.(3) In order to change soft-sensing technique based on traditional hydraulics mechanism into practical software, the soft-sensing expert system for discharge at lateral gate of the irrigation channel in the Yellow River is designed and implemented. The meticulous calculation of overall discharge-coefficient proposed and achieved by knowledge base expansion improves estimating precision and adaptability of traditional mechanistic models to some extent.(4) BP soft-sensing model for discharge at lateral gate of the irrigation channel in the Yellow River is firstly built, trained and tested by the hydraulic experimental data, the effect of ANN in improvement of estimating precision is proved. In view of disadvantage of BP model such as slower learning speed, easily getting into local minimum and worse generalization performance, corresponding RBF soft-sensing model is built and parameters are optimized by adaptive genetic algorithm(AGA) . The simulation results show that RBF soft-sensing model is more applicable than RBF one. In order to adapt embedded environment and discharge real-time estimation, a kind of hybrid soft-sensing model based on CMAC and hydraulics mechanism is built. Similarly, to meet the variety of flow patterns and the sparseness of training samples at lateral gate of the irrigation channel in the Yellow River, the strategy of dividing and ruling is proposed yet, hybrid soft-sensing expert network suitable to flow pattern variation is built and soft-sensing molding for all flow patterns is changed to soft-sensing molding only for single flow pattern.(5) Aiming at sparseness of training sample in engineering practice, a soft-sensing model base on Support Vector Regression(SVR) is built and PSO based optimizing method for soft-sensing model parameters is adopted. Training and testing for SVR base soft-sensing model under same experimental data and different kernel function and loss function, the results show that SVR base soft-sensing model is suitable to sparseness of training sample and estimating precision is higher than BP and the RBF soft-sensing model, Moreover, generalization can be improved essentially.(6) The discharge measuring practice on spot and hydraulic experiments indicate that estimating error of discharge measuring formulas recent adopted will increase evidently when approaching the boundaries of two or more than two flow patterns. Owing to lack of training samples at the flow pattern boundaries, the generalization of various soft-sensing model based on machine learning will also decrease significantly. Based on the analysis of fuzzy characteristics of plow pattern boundaries at lateral gate of the irrigation channel in the Yellow River, flower pattern regional reference model and soft-sensing scheme combined overall discharge coefficients fuzzy inference system (FIS) and data fusion are put forward, this soft-sensing scheme has achieved some effects in improvement of estimating precision and applicability of soft-sensing technique for fuzzy flow pattern boundaries.Main research achievements and innovative points of this dissertation are included in following five aspects: Firstly, development of automatic measuring device for discharge at lateral gate of irrigation channel witch is suitable to the irrigation channel in the Yellow River Basin and ideal hardware platform is established for implement of soft-sensing technique; Second, foundation of soft-sensing expert system based hydraulics mechanism and meticulous calculation of overall discharge-coefficients, both a kind of regulations for soft-sensing programming is established and a simple effective method for improving discharge estimating is provided; Third, various soft computing method such as BP, RBF, CMAC, AGA, SVR and PSO are selected to build the soft-sensing models witch are suitable for embedded environment, real-time discharge estimation and training sample scarcity, the estimating precision as well as applicability and reliability of the soft-sensing models are fully improved; Fourth, a hybrid soft-sensing model based on fuzzy inference technique is proposed and established, the estimating precision and applicability of the hydraulics mechanism model are improved effectively; Fifth, a kind of flower pattern regional reference model made of definite regions and fuzzy boundaries is proposed and distance measure based multi-model data fusion technique is adopted, valuable exploration has been done in improvement of discharge estimating precision and soft-sensing technique applicability at flow pattern fuzzy boundaries.更多还原