基于 GA-BP 神经网络对不可避免漏失水量的确定
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">摘 要:</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">为了分析管网漏损影响要素和供水管网不可避免漏失水量之间存在的<span style="color: black;">繁杂</span>联系及对应关系,<span style="color: black;">创立</span>适用于我国不可避免漏失水量的确定<span style="color: black;">办法</span>。<span style="color: black;">经过</span>采集、筛选<span style="color: black;">实质</span>管网的漏损统计数据,对独立计量区域(DMA)展开夜间最小流量分析,并分析<span style="color: black;">科研</span>区域内管网的影响<span style="color: black;">原因</span>,将管龄、管材、用户数量及管网压力设定为<span style="color: black;">关联</span>影响参数。在此<span style="color: black;">基本</span>上,分别<span style="color: black;">创立</span>BP神经网络模型以及经过遗传算法优化后的BP神经网络模型(GA-BP神经网络),并对优化前后二者神经网络模型拟合结果对比分析。<span style="color: black;">最后</span>将优化后的模型应用于<span style="color: black;">区别</span>DMA进行漏损分析。结果<span style="color: black;">表示</span>,优化后的BP神经网络不仅训练效率更高,且误差平均降低了3.02%。将该模型应用于<span style="color: black;">实质</span>管网三个<span style="color: black;">区别</span>DMA时,<span style="color: black;">能够</span><span style="color: black;">发掘</span>DMA内均存在可控漏损水量,以经济可<span style="color: black;">行径</span>前提,分别可节约水量10 512 m3/a、12 702 m3/a、37 580.4 m3/a。<span style="color: black;">科研</span>成果可为DMA漏损<span style="color: black;">掌控</span><span style="color: black;">供给</span>参考。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">关键词:</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">BP神经网络;遗传算法;可行性<span style="color: black;">评定</span>;不可避免漏失;线性回归分析;人工智能算法;DMA;影响<span style="color: black;">原因</span>;</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">作者简介:</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">王俊岭(1973—),男,副教授,博士,<span style="color: black;">重点</span>从事节水技术、管网优化<span style="color: black;">掌控</span>等<span style="color: black;">科研</span>。E-mail:18813070523@163.com;</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">*胡颖梦(1996—),男,硕士<span style="color: black;">科研</span>生,<span style="color: black;">重点</span>从事管网漏损<span style="color: black;">掌控</span>等<span style="color: black;">科研</span>。E-mail:17858285940@163.com;</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">基金:</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">国家水体污染与治理科技重大专项课题(2017ZX07501-002-05);</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">引用:</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">王俊岭,胡颖梦,张昕喆,等. 基于 GA-BP 神经网络对不可避免漏失水量的确定[J]. 水利水电技术( 中英文) ,2021,52( 12) : 201-211. </p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">WANG Junling,HU Yingmeng,ZHANG Xinzhe,et al. GA-BP neural network-based determination of unavoidable annual real losses from water supply pipeline network[J]. Water Resources and Hydropower Engineering,2021,52( 12) : 201-211.</p>
<h1 style="color: black; text-align: left; margin-bottom: 10px;">0 引 言</h1>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">供水管网漏损是<span style="color: black;">困惑</span>所有供水企业的问题,漏损<span style="color: black;">能够</span>定义为在用户计量点之前,所有类型的泄漏、爆裂和溢出<span style="color: black;">导致</span>的水损失量。管网漏损<span style="color: black;">是由于</span>缺乏主动漏损<span style="color: black;">掌控</span>(ALC)、运行压力过大、操作和<span style="color: black;">守护</span>不良、振动和交通负荷以及管道腐蚀等<span style="color: black;">原因</span>综合<span style="color: black;">引起</span>。随着政府对管网漏损的加强<span style="color: black;">注重</span>,<span style="color: black;">显现</span>了越来越多先进的技术、<span style="color: black;">办法</span>与<span style="color: black;">办法</span>。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">然而在技术<span style="color: black;">持续</span>革新的<span style="color: black;">同期</span>,<span style="color: black;">显现</span>了技术<span style="color: black;">没法</span>得到有效应用等问题,其中关键一方面在于漏损<span style="color: black;">评估</span>,漏损<span style="color: black;">评估</span>是有效<span style="color: black;">掌控</span>漏损的<span style="color: black;">基本</span>。漏损<span style="color: black;">评估</span>中的关键点是漏损<span style="color: black;">评估</span>指标的<span style="color: black;">选取</span>,指标的<span style="color: black;">是不是</span>合适决定了漏损<span style="color: black;">状况</span>能否得到真实反映。1997年,国际水协 (IWA)工作小组提出了<span style="color: black;">基本</span><span style="color: black;">设备</span>渗漏指数(ILI),随着国内外<span style="color: black;">针对</span>漏损<span style="color: black;">科研</span>的深入,产销差率、泄漏性能指数(LPI)等指标<span style="color: black;">亦</span>被<span style="color: black;">持续</span>提出并<span style="color: black;">经过</span>实践应用纳入<span style="color: black;">评估</span>体系中去,GIOVANNA<span style="color: black;">创立</span>了泄漏性能指数(LPI),强调LPI值和漏损值之间的线性<span style="color: black;">关联</span>性。张现国在国内应用供水管网漏失指数(ILI)并进行了实例计算。代焕芳在分别分析了真实漏失量、产销差率、ILI等指标后,<span style="color: black;">意见</span>国内引入ILI概念来进行管网漏损<span style="color: black;">评估</span>。随后提出了真实漏失水量<span style="color: black;">评估</span>指标——供水管网背景漏失指数(BLI)。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">ILI是<span style="color: black;">日前</span>美国环境<span style="color: black;">守护</span>署(EPA)、美国水务协会(AWWA)和国际水协会(IWA)<span style="color: black;">一起</span>认可的衡量真实漏失的<span style="color: black;">靠谱</span>指标点之一。ILI定义为当前物理漏失水量(CARL)和系统内不可避免漏失水量(UARL)之间的无量纲比率,其中UARL(Unavoidable annual real losses)<span style="color: black;">寓意</span>着在当前背景<span style="color: black;">要求</span>下,无论是技术还是经济层面都<span style="color: black;">没法</span>避免的供水系统理论最小漏失。<span style="color: black;">因此呢</span><span style="color: black;">一般</span>将UARL视作基准值<span style="color: black;">评估</span>管网漏损<span style="color: black;">状况</span>,<span style="color: black;">知道</span>区域内漏损<span style="color: black;">状况</span>的可<span style="color: black;">提高</span>空间。不可避免漏失水量的计算<span style="color: black;">重点</span>分为实测与计算两种方式,实测法指在检漏水平高、计量准确且无偷盗水现象的DMA(District metering area, 即独立计量区域)中,<span style="color: black;">能够</span>将夜间最小流量视作夜间居民正常用水量与DMA不可避免漏失水量的和,<span style="color: black;">因此呢</span>得到公式UARL=QMNF-Q夜间居民。除此之外,国际水协提出了UARL经验计算公式,该公式系统性的将支干管长度、压力、连接点数等<span style="color: black;">原因</span>纳入<span style="color: black;">科研</span>范围,是<span style="color: black;">日前</span>为止国际通用最优的<span style="color: black;">办法</span>。董驹萍应用该公式与实测最小夜间流量数据进行对比,<span style="color: black;">发掘</span>二者的差异颇大。<span style="color: black;">一样</span>,现有的<span style="color: black;">科研</span><span style="color: black;">显示</span>,我国供水系统难以用IWA提出的经验公式照搬套用。分析<span style="color: black;">原由</span>,一是国内管网<span style="color: black;">状况</span>复杂,公式参数需要<span style="color: black;">调节</span>;二是公式中部分定义存在界限模糊问题,例如国内<span style="color: black;">无</span>对公式中“<span style="color: black;">评估</span>区域内连接点数”有<span style="color: black;">知道</span>的标准定义,即需要<span style="color: black;">按照</span>管网<span style="color: black;">区别</span><span style="color: black;">状况</span>重新定义<span style="color: black;">选择</span>。阅读国内现有的<span style="color: black;">科研</span>,<span style="color: black;">发掘</span>大多学者为<span style="color: black;">科研</span>出适应国内<span style="color: black;">状况</span>的公式,展开了一系列探讨对参数进行修正以及重新定义,<span style="color: black;">日前</span>来看,经验公式在国内应用存在<span style="color: black;">必定</span>的难度,国内<span style="color: black;">繁杂</span>的管网<span style="color: black;">状况</span><span style="color: black;">需求</span>该经验公式需要结合<span style="color: black;">更加多</span>的影响<span style="color: black;">原因</span>及其对应联系,或许能得到更为<span style="color: black;">恰当</span>的确定<span style="color: black;">办法</span>。调研漏损方面<span style="color: black;">文案</span>,<span style="color: black;">日前</span>国内还<span style="color: black;">无</span>学者对UARL进行建模模拟分析。<span style="color: black;">因此呢</span>,<span style="color: black;">思虑</span>提出一种新的方式,本文基于国内<span style="color: black;">状况</span>,<span style="color: black;">期盼</span><span style="color: black;">经过</span>建模对不可避免漏失水量进行模拟校核,试图得到更贴近现实的模拟值,支撑UARL的确定,从而为管网漏损<span style="color: black;">掌控</span><span style="color: black;">评估</span><span style="color: black;">供给</span>数据支撑。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">人工神经网络是当前最为流行的模拟预测<span style="color: black;">办法</span>,以结合数据分析与挖掘的方式,<span style="color: black;">经过</span><span style="color: black;">自己</span>的非线性处理,在内部进行各个节点的权、阈值的<span style="color: black;">调节</span>修正,从而输出结果。其中BP神经网络更受学者们的关注,BP神经网络无论是理论方面或是性能都<span style="color: black;">已然</span>是极为成熟的,尤其是针对难以<span style="color: black;">发掘</span>规律的样本库,<span style="color: black;">经过</span>特殊的网络构造以及强大的非线性映射的特点,恰好适用于挖掘UARL与多种漏损影响<span style="color: black;">原因</span>之间模糊的关系。BP神经网络在管网<span style="color: black;">评估</span>、漏损定位以及爆管预测等方面已有广泛应用,然而BP神经网络本身存在着局限性,<span style="color: black;">重点</span><span style="color: black;">原由</span>在于其初始权、阈值的随机<span style="color: black;">选择</span>,容易<span style="color: black;">引起</span>网络不收敛或是陷入局部极值点,从而影响模拟效果。为了弥补以上缺陷,万盛萍提出<span style="color: black;">能够</span>利用遗传算法的特性,先进行寻优<span style="color: black;">过程</span>,<span style="color: black;">这里</span><span style="color: black;">基本</span>上,利用BP网络<span style="color: black;">精细</span>求解,既<span style="color: black;">能够</span>有效规避BP神经网络的局限性,又<span style="color: black;">能够</span>进一步<span style="color: black;">提高</span>模拟精确性。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">本<span style="color: black;">科研</span><span style="color: black;">经过</span>采集多个DMA的<span style="color: black;">基本</span>数据,剔除无效数据,运用<span style="color: black;">持有</span>强大的数据处理能力与泛化能力的人工神经网络,构建了一种新型的不可避免漏失水量的确定<span style="color: black;">办法</span>——基于GA-BP神经网络的不可避免漏失水量模型,并<span style="color: black;">经过</span><span style="color: black;">实质</span>的管网DMA进行模型的应用,得出相应的漏损<span style="color: black;">评估</span>,给出改进<span style="color: black;">办法</span>,达到优化漏损<span style="color: black;">掌控</span><span style="color: black;">办法</span>、节约水资源的目的。期待能够将此模型广泛应用于我国供水管网漏损<span style="color: black;">评估</span>中去,能够为进一步<span style="color: black;">提高</span>DMA漏损<span style="color: black;">掌控</span><span style="color: black;">供给</span>参考。</p>
<h1 style="color: black; text-align: left; margin-bottom: 10px;">1 <span style="color: black;">科研</span>区域概况</h1>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">1.1 <span style="color: black;">科研</span>区域简介</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">为了探清国内不可避免漏失水量的应用前景,本<span style="color: black;">科研</span><span style="color: black;">经过</span>对<span style="color: black;">实质</span>DMA展开<span style="color: black;">科研</span>与分析,<span style="color: black;">经过</span>构建模型的方式对水量进行模拟,并以此<span style="color: black;">做为</span><span style="color: black;">指点</span>漏损<span style="color: black;">掌控</span>工作的依据。<span style="color: black;">科研</span>区域为我国南方X市某供水企业供水区域,共设有三座水厂,为该市生活、生产<span style="color: black;">供给</span><span style="color: black;">保证</span>,供水总用户超40万户,日供水能力超过60万m3,管网总长约1 500 km。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">该企业<span style="color: black;">按照</span>实地区域<span style="color: black;">状况</span>,<span style="color: black;">根据</span>分区规模、压力、分区完整性以及计量<span style="color: black;">选取</span>等原则进行了区域划分,在<span style="color: black;">每一个</span>区域进出管装上智能水表,形成独立计量区域DMA,并<span style="color: black;">创立</span>了国际上先进的GIS系统与SCADA系统为一体的智慧供水管理体系,以此进行实时数据监测。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">1.2 前期工作</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">针对</span>不可避免水量的计算,本<span style="color: black;">科研</span>基于<span style="color: black;">实质</span>DMA出发,<span style="color: black;">经过</span>夜间最小流量(Minimum night flow, MNF)展开分析,MNF<span style="color: black;">一般</span>由夜间合法用水量以及夜间漏损水量两部分<span style="color: black;">构成</span>。<span style="color: black;">因此呢</span>,对DMA内的漏点进行彻底的排查检修的<span style="color: black;">同期</span>,结合压力调控,使得管网压力满足最<span style="color: black;">有害</span>点压力,此时的MNF<span style="color: black;">能够</span>视为由夜间居民正常用水量与不可避免漏失水量<span style="color: black;">形成</span>。<span style="color: black;">经过</span>以下<span style="color: black;">过程</span>对数据进行确定:</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(1)对DMA展开全面的检漏修漏,<span style="color: black;">保证</span>修复一切可检测到的漏点。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(2)确定MNF计量区间,<span style="color: black;">按照</span>国内外<span style="color: black;">科研</span>,并结合当地居民生活习惯将计量区间定为01:30—03:30。<span style="color: black;">这里</span><span style="color: black;">基本</span>上,每间隔一分钟,进行一次流量数据采集,<span style="color: black;">这般</span><span style="color: black;">能够</span>在计量区间内的某一瞬间取到夜间最小流量,此时可视作不存在夜间居民正常用水量,将此<span style="color: black;">做为</span><span style="color: black;">科研</span>数据。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(3)为避免<span style="color: black;">显现</span>偶然性与极端性,连续取14 d流量并取均值,<span style="color: black;">做为</span><span style="color: black;">最后</span><span style="color: black;">科研</span>区域内的不可避免漏失水量。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">在<span style="color: black;">科研</span>区域内,<span style="color: black;">选择</span>流量数据连续且<span style="color: black;">恰当</span>的中型规模(用户数量1 000~3 000户)DMA,共36个DMA进行以上<span style="color: black;">过程</span>,进行<span style="color: black;">科研</span>模拟,以此验证模型的精确性。最后,利用剩余三个保持<span style="color: black;">状况</span>的DMA,<span style="color: black;">做为</span>模拟<span style="color: black;">科研</span>后的实例应用,代入模型后,<span style="color: black;">每一个</span>DMA<span style="color: black;">能够</span>得到模拟优化后的不可避免漏失水量(即当前管网<span style="color: black;">要求</span>下,可<span style="color: black;">掌控</span>达到的最低漏损状态),与DMA当前实测的不可避免漏失水量产生对比,从而<span style="color: black;">指点</span>漏损<span style="color: black;">掌控</span>工作。部分原始数据如<strong style="color: blue;">表1</strong>所列。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/b195fdda895f43c78ea70ca8508d71a9~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=13%2B9bWkF8AhpBV%2FkjaWsa8LYJvw%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">1.3 影响<span style="color: black;">原因</span></strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">选择</span>管网漏损影响<span style="color: black;">原因</span><span style="color: black;">做为</span>构建模型输入的参数,为得到与管网漏损联系最密切的参数变量,需要就管网漏损的影响<span style="color: black;">原因</span>展开分析讨论。本<span style="color: black;">科研</span>前期利用灰色<span style="color: black;">相关</span>度法对<span style="color: black;">科研</span>区域内漏损的影响<span style="color: black;">原因</span>进行<span style="color: black;">评估</span>分析,得到结论:管材与管龄是影响最为<span style="color: black;">明显</span>的两个影响<span style="color: black;">原因</span>。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">因此呢</span>,<span style="color: black;">首要</span>确定管龄、管材<span style="color: black;">做为</span>模型输入主变量。<span style="color: black;">这里</span><span style="color: black;">基本</span>上,<span style="color: black;">思虑</span>到用户数与管网供水量的密切<span style="color: black;">相关</span>,以及压力<span style="color: black;">针对</span>供水管网漏损的调节<span style="color: black;">掌控</span>效果,<span style="color: black;">另一</span><span style="color: black;">增多</span>两个变量:用户数N和管网平均压力P。模型的输出即设为不可避免漏失的水量。</p>
<h1 style="color: black; text-align: left; margin-bottom: 10px;">2 <span style="color: black;">科研</span><span style="color: black;">办法</span></h1>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">2.1 原理介绍</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">1986年,以RUMELHART为<span style="color: black;">表率</span>的<span style="color: black;">科研</span>者提出了一种<span style="color: black;">根据</span>误差逆向传播算法训练的神经网络—BP(back propagation)神经网络的概念,核心思想是正向输出与误差的逆向传播,<span style="color: black;">持续</span>迭代计算从而得到最优输出。BP神经网络分为输入、隐含、输出三个层次,如<strong style="color: blue;">图1</strong>所示,各个层次分别由多个节点<span style="color: black;">形成</span>,各个层次之间由<span style="color: black;">区别</span>的权值连接ωki、ωij。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/858e6cb7a7574bee8944d984cd93a512~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=2%2FZhhHMlMvZbNPFJr19D51pXNLc%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">图1 BP神经网络结构</p>
</div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">隐含层第i个神经元输出值</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/f2f3f652d57d426e853951d05f4f1791~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=JE1qLjdRAL0PgnmI8QMr3EmlMtA%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">正向传播输出层神经元j的输出值</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/4ad2d1532ad146afb644f565714b0e37~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=drEc1M9D2D2y5D%2F9TDxgZ3d6%2B7g%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">逆向传播误差定义的函数为</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/5f2d2bb72df6446498cabf0b59a63409~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=L3zcWr4WCOxPjgmE5Y%2FoewcZCc8%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">式中,n为样本数;ΔSv为样本值与模拟值的差。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">GA-BP神经网络模型的构建分为以下三块内容:(1)确定各项参数以及输入与输出量,<span style="color: black;">创立</span>结构;(2)<span style="color: black;">按照</span>遗传算法全局搜查的特性,主动寻优,优化初始权值和阈值;(3)将寻优后的结果重新赋予回到构建的BP神经网络中去,<span style="color: black;">起始</span>迭代计算、模拟输出,如<strong style="color: blue;">图2</strong>所示。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/684cd6df612441dc953769c2338be004~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=bN9QBsslAyk8%2F0FNPeqGdV6cyow%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">图2 GA-BP神经网络原理</p>
</div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">2.2 基于GA-BP神经网络不可避免漏失水量计算的可行性<span style="color: black;">评定</span></strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">2.2.1 </strong>强大的数据处理能力</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">不可避免漏失水量的确定需要基于<span style="color: black;">繁杂</span>多样的<span style="color: black;">基本</span>管网信息,GA-BP神经网络利用其特有的大规模并行、分布式储存和处理自适应性等技术,相<span style="color: black;">针对</span>传统耗时耗力且<span style="color: black;">繁杂</span>的模型,无需<span style="color: black;">太多</span>的设定与数理支撑,<span style="color: black;">能够</span>在给定的数据库中,找到潜藏的规律,并<span style="color: black;">创立</span><span style="color: black;">恰当</span>联系。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">2.2.2</strong> 从数据出发的客观性结论</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">对不可避免漏失水量进行建模预测,<span style="color: black;">首要</span>需要确定输入参数以及对应权值和阈值,GA-BP神经网络<span style="color: black;">经过</span>对样本数据进行训练,利用其全局搜索能力,求解最优初始权值与阈值反馈回神经网络中,结果源于数据,避免了人为主观<span style="color: black;">干涉</span>与随机初始值带来的误差,极大地<span style="color: black;">加强</span>了结论的客观性。GA-BP神经网络应用流程如<strong style="color: blue;">图3</strong>所示。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/081ad9954666412e9b7701a62693c622~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=8r5H2Nm31ofiLJQk9Vfnu0wu%2F%2Fc%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">图3 GA-BP神经网络应用流程</p>
</div>
<h1 style="color: black; text-align: left; margin-bottom: 10px;">3 模型对比及<span style="color: black;">实质</span>应用</h1>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">3.1 模型参数变量的确定</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">按照</span>对<span style="color: black;">科研</span>区域<span style="color: black;">状况</span>与影响<span style="color: black;">原因</span>分析,确定管龄、管材(分为铸铁管与非铸铁管)、用户数N和管网平均压力P共五个输入参数,不可避免漏失水量<span style="color: black;">做为</span>模型的<span style="color: black;">独一</span>输出。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">3.1.1</strong>确定神经网络层次结构与节点数</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">BP神经网络的性能优劣取决于网络结构中隐含层数和隐含层节点个数。最直观的影响即当隐含层节点数设置过少,会降低训练精确度;当取用的节点数量相应<span style="color: black;">增加</span>时,<span style="color: black;">能够</span>起到<span style="color: black;">提高</span>训练精确度的效果。但节点数的设置<span style="color: black;">亦</span>存在过犹不及的现象,若<span style="color: black;">选择</span>的节点数太多,会产生神经网络逼近能力过强的<span style="color: black;">状况</span>,而弱化了很重要的泛化能力。<span style="color: black;">因此呢</span>,节点数的设置存在一个适宜的区间,HO等、邵圆媛<span style="color: black;">发掘</span>,一个三层BP神经网络只需要一个隐含层,即<span style="color: black;">能够</span><span style="color: black;">经过</span>对该隐含层的节点数进行<span style="color: black;">调节</span>,从而映射任意连续函数,且满足精度<span style="color: black;">需要</span>。<span style="color: black;">因此呢</span>,本<span style="color: black;">实验</span>仅<span style="color: black;">选择</span>一个隐含层。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(1)输入层。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">本模型的输入参数分别为管龄、管材、用户数以及管网平均压力,其中<span style="color: black;">按照</span>区域管网特性,管材分为铸铁管与非铸铁管两个变量,则节点数定义为5。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(2)输出层。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">输出即不可避免漏失水量,故节点数定义为1。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(3)隐含层。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">迄今</span>为止,隐含层节点个数的确定还未产生<span style="color: black;">通常</span>规律性计算方式,<span style="color: black;">日前</span>大多<span style="color: black;">选取</span><span style="color: black;">经过</span>经验公式计算得到范围值如下</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/3187744dcd39419f8c0f99171d5bf42f~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=QGPle9%2FRqMeDs5OAakB9zcOI7Ak%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">式中,n为隐含层节点个数;m为输入层节点个数;a为输出层节点个数;l为1到10的随机整数。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">而后</span>进行试凑整法来计算隐含层节点个数,当均方误差(MSE)达到最小时,对应的节点个数即为最佳。计算得到隐含层节点数的范围为4到13。经过<span style="color: black;">持续</span>地往复训练,结果<span style="color: black;">表示</span>:当MSE达到最小值时,节点个数为10,即理想的节点数。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">3.1.2 </strong>设置神经网络其余参数</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">训练参数<span style="color: black;">详细</span>如<strong style="color: blue;">表2</strong>所列。当满足最高训练次数、最小误差值、梯度<span style="color: black;">体积</span><span style="color: black;">或</span>满足验证次数所设定的参数值的<span style="color: black;">状况</span>时,即<span style="color: black;">表率</span>完成训练,并自动结束训练。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/e92f44101c5d44fe8b7bd9f290bcf934~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=nScGt%2FFhcN46%2BDeLwKXKS85ZrdE%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">以MATLAB为平台,在所有的36组DMA数据中,随机抽取五组<span style="color: black;">做为</span>测试用的样本数据,其余的DMA数据则用于训练神经网络。为消除数量级、单位等影响,采用函数mapminmax对输入、输出数据进行归一化</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/4baa27d780b94b7bb4b358cd1edf7d2c~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=Y49wzoZxCce9ON5rqnPbkHN12To%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">式中,y为初始数据;ymin为最小值;ymax为最大值。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">3.2 设置遗传算法优化BP神经网络<span style="color: black;">掌控</span>参数</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(1)编码。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">随着<span style="color: black;">各样</span>算法的深入<span style="color: black;">研发</span><span style="color: black;">科研</span>,编码的方式多种多样,其中二进制编码<span style="color: black;">运用</span>频率较高,应用范围<span style="color: black;">亦</span>较广,但可能<span style="color: black;">引起</span>编码过长,从而会存在影响神经网络学习精度的可能性。本文<span style="color: black;">选取</span>实数编码,计算编码长度L,公式如下</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p9-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/84ce17a9976946fc96c75a01a3ea4ed6~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=uniQwjz44U%2FPY37rTrqkch0KQOM%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">式中,i为输入层节点数;k为隐含层节点数; j为输出层节点数。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(2)种群规模。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">种群规模的设置不存在严格的标准,在适宜区间范围内10~200之间<span style="color: black;">选取</span><span style="color: black;">就可</span>,本<span style="color: black;">科研</span><span style="color: black;">经过</span>对<span style="color: black;">区别</span>种群规模进行<span style="color: black;">海量</span>尝试,<span style="color: black;">最后</span><span style="color: black;">思虑</span>拟合效果,选定为10。<span style="color: black;">区别</span>的数据样本建模应当进行重新尝试与<span style="color: black;">选择</span>,在满足精确度前提<span style="color: black;">要求</span>下,种群规模应当<span style="color: black;">选择</span>足够大以避免结果陷入局部最优解。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(3)适应度。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">适应度<span style="color: black;">经过</span><span style="color: black;">实质</span>值与模拟值的误差绝对值之和来<span style="color: black;">暗示</span>。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(4)遗传操作。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">选取</span>操<span style="color: black;">做为</span>轮盘赌法(适应度比例法)。在交叉操作中,交叉概率取0.8(概率<span style="color: black;">体积</span>不仅与收敛速度有着紧密的<span style="color: black;">相关</span>性,<span style="color: black;">况且</span><span style="color: black;">能够</span>减小非最优解输出的可能性)。在变异操作中,本试验中取0.1,<span style="color: black;">处在</span><span style="color: black;">通常</span>经验值之间,避免过大或过小。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(5)终止进化代数<span style="color: black;">通常</span>在50~100之间<span style="color: black;">选择</span>,本<span style="color: black;">实验</span>选定为100。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">对经过遗传算法优化后的BP神经网络,进行样本数据训练,得到适应度变化曲线(见图4)。由<strong style="color: blue;">图4</strong>可知,在0~22代内,适应度随着迭代次数的<span style="color: black;">持续</span><span style="color: black;">增加</span>而<span style="color: black;">快速</span>下降,随后则趋于直线,说明在0~22代内,适应度在得到<span style="color: black;">连续</span>的优化,直至第22代,达到个体的最佳适应度,<span style="color: black;">亦</span>说明了本<span style="color: black;">实验</span>取用100代的进化代数是可行的。最后将22代的最优个体输出并<span style="color: black;">做为</span>GA-BP神经网络模型的初始权值和阈值。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/08831583878346b989c9c5333b2ddb59~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=kMTwz7%2FEv3dP40CgXIn8ML7EnRA%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">图4 适应度变化曲线</p>
</div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">3.3 应用对比</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">对二者神经网络模型同步开展训练,<strong style="color: blue;">图5</strong>为训练完成界面,并得到相对应的误差曲线变化图如<strong style="color: blue;">图6</strong>所示。<span style="color: black;">按照</span>表1,训练完成的标志即误差满足设定值,结合图6训练样本误差曲线图,<span style="color: black;">能够</span>看出BP神经网络经过训练后的MSE值会随之相应降低,尤其是前5次训练效果<span style="color: black;">明显</span>,之后均方误差缓慢降低,<span style="color: black;">最后</span>在41次训练后,误差值为9.675×10-5,完成训练。而GA-BP神经网络相比较,在效率上得到了<span style="color: black;">极重</span>的<span style="color: black;">提高</span>,仅仅<span style="color: black;">经过</span>12次训练,误差<span style="color: black;">已然</span>达到8.752×10-5,完成训练。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/c24288be14714c458be4f05d130def2e~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=e3zbULeUvdQJTsAFM8%2Bl2chgt0E%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">图5 训练结束示意</p>
</div>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/ae8057cdc7d14b15bf531abf5bdc7273~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=L1r%2Fun8RLoQXqdDk2AI6FJC8TpY%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">图6 训练误差曲线</p>
</div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">对GA-BP神经网络和BP神经网络的模拟效果进行对比分析。对5组随机样本数据,编号1—5,展开拟合效果分析,五组模拟结果曲线分别如图7所示。从<strong style="color: blue;">图7</strong>可知,对比<span style="color: black;">第1</span>组样本的模拟输出值,BP神经网络模拟值更为准确,但二者模拟值与实测值的误差均较小,剩余四组的结果均<span style="color: black;">显示</span>GA-BP神经网络的模拟值更加接近实测值,<span style="color: black;">能够</span>初步认为,GA-BP神经网络模拟值与<span style="color: black;">实质</span>值之间更加贴合。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/c249400c7f8c49c99b879ca04f144a0b~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=A3xXzHse9dC7Fh85f5jZBiAtyok%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">图7 模拟曲线对比</p>
</div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">3.4 模拟结果分析</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">从误差<span style="color: black;">方向</span>进行分析,GA-BP神经网络<span style="color: black;">针对</span>测试样本的数据模拟精确度上,显然有了<span style="color: black;">必定</span>的<span style="color: black;">提高</span>。平均误差在BP神经网络的<span style="color: black;">基本</span>上降低了3.02%,最大误差降低了5.16%,最小误差降低了1.07%,模型拟合的精度得到<span style="color: black;">明显</span><span style="color: black;">提高</span>;从神经网络训练效率<span style="color: black;">方向</span>,GA-BP神经网络中,综合遗传优化以及网络训练,对样本数据共计进行迭代运行34次,比BP神经网络的41次迭代训练更<span style="color: black;">有效</span>。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">为进一步对比模型效果,分别开展线性回归分析测试,如<strong style="color: blue;">图8</strong>所示。显然GA-BP网络的最佳拟合曲线与<span style="color: black;">目的</span>曲线更为贴近,BP神经网络和GA-BP神经网络测试样本数据的线性回归直线分别为Y=0.912 4X+0.311 6、Y=0.976 8X+0.050 2,GA-BP神经网络决定系数R2=0.993 4大于BP神经网络R2=0.960 6,说线性回归直线的拟合度更高,<span style="color: black;">亦</span><span style="color: black;">寓意</span>着模拟值更为精确。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/b09922ab14d94a9daf1f836bdf75fdf1~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=rqfqjk0peHUYFzXrmo4RD5Y4ieo%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">图8 线性回归对比</p>
</div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">综上,无论是从相对误差的<span style="color: black;">方向</span>还是从线性回归精确性方面,GA-BP神经网络模型都是一种更为<span style="color: black;">优良</span>的方式,<span style="color: black;">能够</span>更好的对不可漏失水量展开模拟。<span style="color: black;">经过</span><span style="color: black;">以上</span>一系列对比<span style="color: black;">科研</span>分析,得到以下结论:(1)BP神经网络满足了本<span style="color: black;">科研</span>的初步期待,确实<span style="color: black;">能够</span>满足对UARL的模拟,结果与实测值相当接近,<span style="color: black;">能够</span>为<span style="color: black;">实质</span>管网的<span style="color: black;">评估</span><span style="color: black;">供给</span>科学数据支撑。(2)遗传算法<span style="color: black;">针对</span>BP神经网络的训练速度,以及结果的输出上均有正向<span style="color: black;">提高</span>,<span style="color: black;">经过</span>GA-BP神经网络对UARL的确定是更优<span style="color: black;">选取</span>。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">3.5 <span style="color: black;">实质</span>应用</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">3.5.1</strong> 模型应用</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">将构建的模型应用于X市某DMA,并对其展开漏损<span style="color: black;">评估</span>,给出相应<span style="color: black;">掌控</span><span style="color: black;">办法</span>及<span style="color: black;">意见</span>。DMA<span style="color: black;">基本</span>信息如<strong style="color: blue;">表4</strong>所列。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/23dbb959ff5e45c59c1adbe1efeee083~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=gtMFAs3jRcgZkXH%2F7deFq1B5QXE%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">将<span style="color: black;">以上</span><span style="color: black;">基本</span>信息代入模型进行计算。<span style="color: black;">另外</span>,<span style="color: black;">因为</span>压力<span style="color: black;">掌控</span><span style="color: black;">做为</span>管网漏损<span style="color: black;">掌控</span><span style="color: black;">重点</span><span style="color: black;">办法</span>之一,为了探究其<span style="color: black;">针对</span>当前DMA区域的<span style="color: black;">掌控</span>效果,将<span style="color: black;">日前</span>供水压力降低0.01 MPa, <span style="color: black;">做为</span>压力调控<span style="color: black;">办法</span>后的<span style="color: black;">目的</span>压力值,再进行计算。模拟结果如<strong style="color: blue;">图9</strong>所示。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/4db391ddec4046b99c6276cb9e2f78da~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=OMxZq0ZXXz5o2Hy5Mo6yRh3Jacc%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">图9 模拟结果</p>
</div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">3.5.2 </strong>结果分析与<span style="color: black;">意见</span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">为了对比当前该区域的可节约水量(即<span style="color: black;">经过</span>检漏修漏、压力调控等<span style="color: black;">办法</span><span style="color: black;">掌控</span>漏损,以达到该区域内最低漏失水平,再用当前漏损水量与UARL值作差,得到可节约水量),将结果<span style="color: black;">经过</span>百分比进行展示如<strong style="color: blue;">图10</strong>所示。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/cd27cda38f894b4ba4fc15133b9c4ed5~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=cD8Bp%2B%2BLhV74aJy6rZTfZSpLhSg%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">图10 可节约水量占比</p>
</div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">按照</span>图10结果展开分析,并分别提出漏损<span style="color: black;">掌控</span><span style="color: black;">办法</span><span style="color: black;">意见</span>及预期效果。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">以DMA1为例进行计算,<span style="color: black;">按照</span>节1.1<span style="color: black;">说到</span>的前期工作,对DMA进行彻底的明、暗漏检修<span style="color: black;">办法</span>后,且在适当的压力下,<span style="color: black;">能够</span>认定此时该区域的漏失达到最理想状态,<span style="color: black;">因此呢</span>DMA1在当前压力下年度可节约水量=1.2×24×365=10 512 m3。同理DMA2、DMA3在当前压力下每年可分别节约水量23 827.2 m3、12 702 m3。而在目标压力下,DMA1、DMA2、DMA3分别可节约水量为12 614.4 m3、37 580.4 m3、14 804.4 m3。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">由图10可知,DMA1与DMA3二者<span style="color: black;">状况</span>类似,当前区域内漏损水量<span style="color: black;">显著</span>大于UARL,需要对<span style="color: black;">关联</span>DMA采用主动漏损<span style="color: black;">掌控</span><span style="color: black;">办法</span>,开展彻底的检漏并<span style="color: black;">即时</span>修复漏点,<span style="color: black;">同期</span>压力调控起到的控漏成效较低,<span style="color: black;">表率</span>此时管网运行压力相对合适,若<span style="color: black;">增多</span>压力<span style="color: black;">掌控</span>,此时控漏的经济效益很低<span style="color: black;">乃至</span><span style="color: black;">引起</span>负经济效益,<span style="color: black;">因此</span>对DMA1、DMA3均采取主动控漏的方式,<span style="color: black;">能够</span>分别达到降低24.19%、19.84%的漏损水量,预计可分别节水10 512 m3/a、12 702 m3/a。对应本<span style="color: black;">科研</span>章节1前期工作中的内容,重点需要<span style="color: black;">增多</span>这两个DMA巡检队伍检漏修漏强度,<span style="color: black;">按照</span>该区域的平均水价(当地水费分梯度计算,此处计算粗略取3元/t, 稍高于<span style="color: black;">第1</span>梯度)<span style="color: black;">能够</span>计算得到,每年DMA1、DMA3<span style="color: black;">能够</span><span style="color: black;">经过</span>主动检漏的方式(无需额外投入)分别节约31 536元/a、38 106元/a。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">在DMA2中,漏损水量<span style="color: black;">显著</span>大于UARL,<span style="color: black;">一样</span>需要主动控漏,<span style="color: black;">能够</span>降低15.91%的漏损水量。<span style="color: black;">同期</span>,该DMA在<span style="color: black;">目的</span>压力下,UARL值进一步降低,且降幅<span style="color: black;">很强</span>,说明压力调控在该DMA<span style="color: black;">能够</span>起到<span style="color: black;">明显</span>的控漏效果。<span style="color: black;">经过</span>主动控漏结合压力调控可将漏损水量进一步减少9.18%,达到综合降低25.09%的漏损水量。综上,需要<span style="color: black;">经过</span>结合主动控漏和压力调控<span style="color: black;">办法</span>,预计可节水37 580.4 m3/a。<span style="color: black;">针对</span>DMA2<span style="color: black;">来讲</span>,不仅需要加强检漏修漏<span style="color: black;">办法</span>,还需要对区域内的压力进行调节。<span style="color: black;">科研</span>区域内并未<span style="color: black;">显现</span>缺压<span style="color: black;">状况</span>,当前的供水压力<span style="color: black;">能够</span>满足用户正常<span style="color: black;">需要</span>,且有<span style="color: black;">必定</span>空间<span style="color: black;">能够</span>下调,<span style="color: black;">因此呢</span>仅需安装<span style="color: black;">必定</span>数量的减压阀<span style="color: black;">就可</span>进行初步的漏损<span style="color: black;">掌控</span>。若需进一步开展系统地分时段分区域调压,则需要<span style="color: black;">创立</span>压力反馈监测系统等,不适宜在单独DMA内进行,<span style="color: black;">意见</span>从<span style="color: black;">全部</span>大供水管网层面<span style="color: black;">创立</span>模型。综上,进行粗略的估计,DMA2<span style="color: black;">最少</span><span style="color: black;">能够</span>节约100 000元/a。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">经过</span>在X市进行验证,利用GA-BP神经网络建模的方式对优化后的DMA进行不可避免漏失水量预估,从而与实测不可避免漏失水量进行对标,据此<span style="color: black;">指点</span>漏损,是一种可行的<span style="color: black;">方法</span>,<span style="color: black;">能够</span>为漏损<span style="color: black;">评估</span><span style="color: black;">供给</span><span style="color: black;">必定</span>数据支撑,<span style="color: black;">亦</span><span style="color: black;">能够</span>为DMA的漏损<span style="color: black;">状况</span><span style="color: black;">供给</span>判断依据,估计区域内漏损可<span style="color: black;">提高</span>空间。且投资较小,成效立竿见影,适合在单独DMA区域内推广应用,用于管网漏损检验,<span style="color: black;">指点</span>管网漏损<span style="color: black;">掌控</span>工作。</p>
<h1 style="color: black; text-align: left; margin-bottom: 10px;">4 结论与展望</h1>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(1)本文<span style="color: black;">经过</span>结合遗传算法对BP神经网络进行优化,不仅<span style="color: black;">能够</span>有效<span style="color: black;">加强</span>模型训练效率,还在精度上有了<span style="color: black;">必定</span>的<span style="color: black;">提高</span>,平均误差在BP神经网络的<span style="color: black;">基本</span>上进一步降低了3.02%,证明GA-BP神经网络模型<span style="color: black;">能够</span>更好地模拟得出不可避免漏失水量。于是在管网<span style="color: black;">基本</span>信息<span style="color: black;">必定</span>的前提下,利用模型模拟出DMA可优化得到的不可避免漏失水量,与实测DMA的结果进行对比,来<span style="color: black;">评估</span>各DMA的漏损<span style="color: black;">状况</span>,并提出相应的漏损<span style="color: black;">掌控</span><span style="color: black;">办法</span>,结果<span style="color: black;">显示</span>对<span style="color: black;">区别</span>区域展开相应控漏<span style="color: black;">办法</span>,均有<span style="color: black;">区别</span>程度的节水成效,最大可节水37 580.4 m3/a。<span style="color: black;">期盼</span>本文对UARL的确定<span style="color: black;">办法</span><span style="color: black;">能够</span>为我国漏损<span style="color: black;">掌控</span>工作<span style="color: black;">供给</span>新的衡量方式。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">(2)本文<span style="color: black;">选择</span>了平均管龄、铸铁管管网长度、非铸铁管管网长度、用户数和管网平均压力五个参数<span style="color: black;">做为</span>模型的输入参数,从结果上看,较准确地得到了不可避免漏失水量模拟值,可以<span style="color: black;">做为</span>其他区域模型<span style="color: black;">创立</span>的参考。当然输入参数并不限于此,<span style="color: black;">区别</span>区域应当有<span style="color: black;">区别</span>的<span style="color: black;">科研</span>结论与输入参数,<span style="color: black;">必定</span>程度下,参数越多,反映的结论越接近<span style="color: black;">状况</span>,<span style="color: black;">因此呢</span>,<span style="color: black;">期盼</span><span style="color: black;">更加多</span>学者<span style="color: black;">能够</span>开展<span style="color: black;">更加多</span>元深层次的<span style="color: black;">科研</span>。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">供水管网漏损<span style="color: black;">掌控</span>是一个<span style="color: black;">连续</span><span style="color: black;">持续</span>的工作,管网漏损从起初的杂乱无序,到<span style="color: black;">此刻</span>的管理有方,从人工检漏到如今预警监测技术<span style="color: black;">持续</span>革新,漏损<span style="color: black;">掌控</span>工作逐步向大数据时代融合,<span style="color: black;">这里</span><span style="color: black;">基本</span>上进行数据的挖掘与分析,形成特定的结构与模型,从而辅佐漏损<span style="color: black;">掌控</span>工作的决策与展开<span style="color: black;">作为</span>当前乃至<span style="color: black;">将来</span>的一个方向,<span style="color: black;">因此呢</span>,<span style="color: black;">将来</span>学者应当投入更大的精力到数据化模型构建的<span style="color: black;">科研</span>中去,以更加客观的数据化结论<span style="color: black;">供给</span>科学的依据。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">水利水电技术(中英文)</strong></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">水利部《水利水电技术(中英文)》杂志是中国水利水电行业的综合性技术期刊(月刊),为全国中文核心期刊,面向国内外公<span style="color: black;">研发</span>行。本刊以介绍我国水资源的<span style="color: black;">研发</span>、利用、治理、配置、节约和<span style="color: black;">守护</span>,以及水利水电工程的勘测、设计、施工、运行管理和科学<span style="color: black;">科研</span>等方面的技术经验为主,<span style="color: black;">同期</span><span style="color: black;">亦</span><span style="color: black;">报告</span>国外的先进技术。期刊<span style="color: black;">重点</span>栏目有:水文水资源、水工建筑、工程施工、工程<span style="color: black;">基本</span>、水力学、机电技术、泥沙<span style="color: black;">科研</span>、水环境与水生态、运行管理、<span style="color: black;">实验</span><span style="color: black;">科研</span>、工程地质、金属结构、水利经济、水利规划、防汛抗旱、建设管理、新能源、城市水利、农村水利、水土保持、水库移民、水利现代化、国际水利等。</p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/pgc-image/1d240163fd1444a899a41dbaa3e126b0~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1725644746&x-signature=DywMCeBKqupKTEBftV%2BBMcwccME%3D" style="width: 50%; margin-bottom: 20px;"></div>
网站建设seio论坛http://www.fok120.com/ 太棒了、厉害、为你打call、点赞、非常精彩等。 你的话语如春风拂面,让我心生暖意。
页:
[1]