【作者】 张琴；

【导师】 汪雄海；

【作者基本信息】 浙江大学 ， 控制理论与控制工程， 2011， 博士

【摘要】 城市供水系统已逐步发展为广地域分布的联网多水源系统。虽然每个水源都有就近供水性质,但由于泵站所属用户用水特征不同,对泵站各时段供水要求存在明显差异。对系统水量的优化预测协调调度的目的是优化资源的合理利用,减少能量损耗又可以较好解决供水边缘地带供水质量,获取良好的社会和经济效益。城市供水系统表现为供水环境复杂、用户多样化,特别是受城市发展布局变化影响用水需求多变,对多水源供水系统的优化调度学术上可归为动态不确定环境下受多条件约束的多目标优化控制问题,该问题具有高度非线性和耦合混合变量的供水系统模型,要实现并满足实时调度需要找出足够精度的用水量预测方法和先进的优化调度策略。论文在寻求取得城市时用水量的时间序列基础上,通过推演算得延迟时间和嵌入维数,探索并应用相空间重构技术,借助混沌理论研究时用水量的混沌特性,证明了一般城市的时用水量具有典型的混沌特性,因而多水源供水系统属典型的混沌系统。通过现代先进理论工具,系统分析研究了某城市给定对象的沿程各大用户用水量观测序列的分形和链级混沌关联,并依据系统相图和最大Lyapunov指数变化研究比较漏损故障发生时的漏水时序的不同混沌特性。仿真结果表明,通过跟踪各混沌特征参数的演化趋势可识别与检测某些漏损故障,该结果可作实际检测验证参考,为及时查找及维护供水系统提供依据,有助于防止资源损耗。通过对不同长度历史训练数据下的各混沌方法预测效果作分析比较后,找出了适用于多水源供水系统时用水量短期预测的混沌预测工具。提出一种混沌横向分时段预测方法,以模式识别获得高关联度的横向时序为研究样本,重构横向分时段相空间并分析典型时段流量数据的混沌特性,通过在线最小二乘支持向量机做混沌预测工具求得各时段用水流量。为了能动态跟踪用水突变,提出纵向残差修正预测方法,实时采集纵向数据序列作残差计算,进而用灰色模型进行残差预测修正以提高实时预测精度。依据杭州市萧山某大用户时用水量实例,对一段时间的正常用水和突变用水做连续24小时以上时间的短期混沌预测,全面比较了相关类型方法的预测效果表明,混沌横向分时段预测方法能很好反映被调控对象沿程用户的用水特点,较好地达到了系统短期预测精度；在此基础上作纵向残差修正能很好跟踪实时用水变化情况,更易满足供水系统实时优化调度的需求。针对多水源供水系统用户用水的特征,借助日益成熟的计算机控制技术寻求现代理论调度控制算法,发现蚁群算法有其可取之处,但常规蚁群算法存在易陷入局部最优、易早熟和分散搜优收敛慢的缺点。采用聚度控制蚁群算法求取离散问题,引入黄金分割法则设置参数,根据蚂蚁聚集的信息权重调整决策概率,能较好克服上述缺点,获取了好的求解效果。并针对求解连续变量全局优化的难题,提出改进聚度控制的蚁群算法,在连续变量域采用实数编码,决策概率中引入区间聚度信息权重,并利用灵活转移步长和动态挥发系数对蚁群分布寻优进行控制,使其迭代前期保持解值多样性与后期又具有快速收敛的双重特性。仿真结果表明,该方法能较好平衡广范围搜索和信息素正反馈能力,有效降低了陷入局部最优的风险,在求解多维多极值连续函数的全局优化问题时具有明显优势。将这两种聚度控制蚁群算法应用于多水源供水系统常规优化调度,仿真论证了其有获得期望效果的潜力。文章最后针对给定的多水源供水系统约束条件众多耗时长且压力波动较大等问题做了更进一步的研究,提出一种基于增量模型控制压力恒定的优化调度新策略。由用户流量参数预测建立了增量模型,以分界监控点压力参数恒定为目标函数,并给出该策略：将机泵出口压力、机泵状态及变频机泵转速为优化决策变量,并采用混合聚度控制蚁群算法同时优化连续和离散变量做实时优化计算。算例仿真表明,在上述给定模型下,新算法可快速找到优化调度的最优解,能较好跟踪用水流量变化和保持监测控制点的实时压力恒定,提高了多水源的实时协调调度能力,有效降低了能耗。更多还原

【Abstract】 The urban supply system develops into a vast distributed large-scale system. For the significant different demands at different time due to the different characteristic of consumers, the multi-source water system needs to be scheduled properly. The rational forecast and scheduling of water demands can bring social and economic benefits by reducing energy consumption and providing high quality of marginal consumers. There are complex environments and different kinds of water consumers. Because the water demands are changed as the development of the city, it is a dynamic constrained multi-objective optimization problem. Since it is a mixed integer nonlinear programming problem, it is necessary to bring up a paccurate forecast method and an advanced scheduling strategy.By computing the delay time and embedded dimension, the phase space of urban hourly water consumption time series is reconstructed. Chaos theory is used to study the chaotic characters, and the hourly water consumption is turning out to be a typical chaotic system. The water time series of successive users are proved to be fractal and chaotic chaining relevant by advanced tools. On this basis, the different chaotic characters of booster and leakage are compared by phase diagram and the maximum lyapunov exponent variation. Test results show that the chaotic characters are changed immediately when the booster occurs, and slow leakage can be discovered after two hours. Thus this method supplies new judgments to effectively amend water system in time and decrease the loss of water resource.After the comparison of different chaos prediction methods by training different length history data, a new prediction method is proposed named chaotic horizontal period clustering, aiming at the high short-term prediction accuracy of hourly water consumption. The horizontal time series are determined as research samples by pattern recognition with high relevancy. After reconstructing the phase space of horizontal period clustering and analyzing chaotic characteristics of typical period data, chaotic prediction model is established. On line least square support vector machine (LS-SVM) is used to forecaste the period flows. Furthermore, to track water consumptions dynamically, vertical residuals are modified by grey model prediction (GMP) after collecting real-time data. The period historical data from Xiao Shan are supplied in the normal and abnormal case study to forecast the day-ahead hourly water consumption, and prediction accuracy with different methods are compared.A Cohesion Control Ant Colony algorithm (CCAC) is used to optimize the discrete problem, by using the golden section method to set parameter, and using gathered information weights to tuning decision policy. Then the CCAC is improved (ICCAC) with continuous coding to solve global optimization problem of continuous function. Section cohesion weight is brought into the decision policy to control ant distributions; flexible search steps are used to encourage local search after ICCAC locating the most promising direction to transfer; and dynamic evaporation factor is incorporated in pheromone updating. Experimental results show that ICCAC can keep good balance between wide exploration and pheromone exploitation, is an effective method to solve continuous global optimization problems. The algorithm mentioned above has shown an execellent performance in the short time scheduling problem of multi-source water system.This paper deals with the real-time scheduling problem of numerous constraints, long time-consuming, the highly variable nodal pressure and an optimal scheduling strategy based on increment model controlling the pressure constant is proposed. Firstly, the flow parameter increment model is established by predicting the nodal real-time water demands, and the variations of nodal pressures are obtained. Then the scheduling strategy is deduced from controlling the constant monitoring pressure parameters. At last, the parameters of pumps output pressures, pumps operation status and speed changes are regarded as decision variables to be regulated, and ICCAC is used to optimize the continuous and discrete variables at a time. The test results show that the new strategy can find the optimal schedule solution quickly, and well track the nodal stream variations to keep the constant monitoring nodal pressure, thus it strengthens the real-time regulation ability of multi-source water supply system, and can reduce the energy cost effectively.更多还原