【作者】 赵晓东；

【导师】 张宏建； 周洪亮；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2014， 博士

【摘要】 饮用水源地(地表水)的水体富营养化是影响水源地水质的主要原因。国内外针对水体富营养化研究多针对藻类“水华”进行预测。近年研究表明,藻类“水华”并非藻类生物量的短时爆发,而是藻体上浮和聚集的一种物理迁移过程,且在“水华”发生之前,水体藻类浓度已远超过水质管理标准。可见,与“罕见”的“水华”事件预测相比,短时连续预测对饮用水源地安全管理更具实用性。对此,本文提出了一种短时预测方法,实现对河流的藻类叶绿素a连续预测。其预测原理是：以短期藻类生长相似与基于事例推理(Case-based Reasoning,简称CBR)方法的假设一致性为基础,将按拉格朗日法划分的流体单元作为藻体生长单元,根据CBR方法的四阶段(“4R”)要求构建相似要素,包括：相似因子、相似系数、相似判据、相似误差以及多元相似综合指数。根据相似预测原理,在同期历史数据中寻找与当前藻类生长相似的历史时段,并将该历史时段的延伸部分作为当前预测结果。本文依托于近期历史数据,根据拉朗格日法构建了河流藻类叶绿素a单日12时预测机理模型,采用L-M率定方法确定最优参数组合和率定时段,获得最优参数率定值,将该模型在预测时段内的计算结果作为预测值。本文采用非线性系统混沌分析理论分析了时叶绿素a观测序列,对主要混沌特征量,包括：嵌入维数、时间延迟、关联维数以及最大Lyapunov指数进行了分析。采用C-C方法同时估计嵌入维数和时间延迟,采用G-P算法估计关联维数,采用Rosenstein方法估计最大Lyapunov指数,并根据最大Lypunov指数对时间序列的最大可预测时长进行估计。根据相似预测原理分别对德国易北河5月1日至8月9日期间(2000、2001年)未来24-72小时的时叶绿素a浓度序列进行预测。在通过数据预处理和权重设置提高预测精度的基础上,采用了叶绿素a浓度多阈值分段评价原则,进行了三日河流营养化等级时变化预测,准确率达85%。国内外目前类似的研究仅采用叶绿素a日均浓度且以单阈值(“水华”预测)作为评判标准,其预测准确率约为83%,预测时长为2.5-3天。通过对比表明该预测模型在营养化等级预估精度及实时性上均具有优势。传统生长机理模型采用的先验参数无法反映藻类在时间和空间上生长异质性的特点,因此,本文采用变量参数建模思想构建了河流藻类生长机理预测模型,并采用易北河2000年5月-8月间的时观测序列进行了单日12时预测验证。验证结果表明,通过选择合理的率定参数组合(率定最优的5参数组合)和率定时段(7天),其预测效果要远好于机理模型在先验参数条件下的预测结果,在高精度预测方面(相对误差小于士10%)甚至优于采用相似原理进行的连续预测结果。该方法为采用机理模型进行河流短期藻类预测提供了一种新的解决方案。河流藻类生长的高度非线性和藻类数据的采样稀疏使藻类实时预测难度增加,其可预测时长问题也鲜有研究报道。本文采用了混沌分析理论对易北河1997-2001年的3-9月的时叶绿素a观测序列进行了分析。分析结果显示河流藻类叶绿素a时间序列具有低维混沌特性,关联维数D=2.75-4.02。采用同样混沌分析方法证实了易北河同时段的径流量序列也具有混沌特征(λ1=0.0125),该结论与国内一些针对河流径流量的研究结果相近。但目前针对河流藻类叶绿素a序列的混沌特性研究尚无相关报道。本文对藻类叶绿素a浓度和径流量的最大预测时间进行了估计,各年的时叶绿素a序列的最长预测时间变化范围为8.01-18.94天,平均为13.98天(约2周),与当前天气预报的最长预报时长相当,而径流量的最长预测时间估计约为80天。气候因素的混沌特性对藻类生长表现出的混沌特征的影响可能要大于径流量等水文因素的影响。该分析结果可为河流藻类浓度可预测能力问题提供新的研究思路。本文的研究成果,对于拓宽预测理论的研究方法,提高藻类预测精度,揭示藻类生长的真实规律,提升饮用水源地水质管理水平具有重要的现实意义。更多还原

【Abstract】 Eutrophication occurred in the drinking water source areas (surface water) often results in the water quality deterioration. Recent studies showed that algal blooms are the process of algae physical migration by floating and aggregating to the surface of water body, which are not caused by the short-time outbreak of algae biomass. The algal concentration has far exceeded the limits of water quality standards even before the occurrence of algal bloom. It shows more practical for drinking water security to make the short-scale continuous prediction of algal concentration, rather than to predict such’rare events’.In the thesis, a method of short-term prediction was proposed to make continuous prediction of chlorophyll a concentrations in river. The prediction theory is based on the consistency between the similarity of the algal short-term growth and the assumptions of the method of Case-based Reasoning (CBR). According to’four-REs’in the method of CBR, the similar conditions including similar factors, similarity coefficient, similarity criteria, similarity error, and multi-criterion synthetic similitude index, were established for the element of fluid meshed as the unit of algal growth according to Lagrangian method. The historical period in which the algal growth is most similar to the current algal growth, was firstly determined by those similar conditions, then the values of chlorophyll a in the backward extension of the historical period were selected as the predictive values. Additionally, a growth model based on the recent historical data was built to predict chlorophyll a concentration on12o’clock next day according to Lagrangian method,. The optimal combination of calibrated parameters and the optimal period of calibration data were both determined by the L-M method. The results from the model with the optimal calibration parameters input in the forecast period were uses as the predicted values. Finally, the chaos of the time series of hourly chlorophyll a observations was analized, the main chaotic charateristeics, which included embedding dimension, time delay, correlation dimension, and the largest Lyapunov exponent, were estimated. The classical methods estimating such characteristics included the C-C method for embedding dimension and time delay, the G-P algorithm for correlation dimension, and the Rosenstein method for the largest Lyapunov exponent. The maximum prediction time was finally estimated by the largest Lyapunov exponent.The series of the chlorophyll a concentrations, which were observed between May1st and August9th on2000and2001in Elbe River, were made the next24-72hours prediction according to the similar prediction principle. After the prediction accuracy was improved by data preprocessing and weight setting, hourly variation of trophic level for the next three days were predicted by multi-threshold sectional evaluations of chlorophyll a concentration. The forecasting accuracy was up to85%. Only the prediction of daily mean values of chlorophyll a concentrations according to a single threshold (often called’algae bloom’prediction) were focused in recent studies, which were able to give a precision of83%and2.5-3days of predictable time. The results indicated that the predict model proposed in the thesis have advantage in the prediction accuracy of trophic level and the real-time performance of prediction.The mechanic model of algae growth based on the variable parameters was verified by predicting the noontime data for the next day observed between May and August in2000in Elbe River, which gave more accurate description of algal growth heterogeneity in space and time than only piror paprameters used. The results obtained by selecting the reasonable parameters combinations (five parameters) and the period of calibration data(seven days), showed much better than those obtained on the condition of the prior parameters, even had more days with the high prediction precision (the relative error less than±10%) compared to the results by CBR at the same predictive time. The method provides a new solution for shore-term algal prediction in river when using mechanic model.Highly nonlinear growth of algae and sprase sampling of algae data in make the real-time preiditon of alage more difficult. The estimation of predictablity has few reports at present. The theory of chaos was used to analyze the observed sequence of chlorophyll a from March to September between1997and2001in Elbe River. The results showed that the time series of chlorophyll a had low-dimensional chaos with low correlation dimension (D=2.75-4.02). It also confirmed that, the runoff sequence of Elbe River in the same period was chaotic judged by the largest Lyapunov exponent larger than zero (λ1=0.0125). The results were similar to those from domestic research about the chaos of river runoff. However, there are no relevant reports about chaotic characteristics of chlorophyll a sequence in river currently. In the thesis, the maximum predictive times for the sequence of hourly chlorophyll a concentration in each year were estimated respectively, which changed from8.01to18.94days. The average value was13.98days (about two weeks) just close to the current biggest day-to-day weather forecast time. A much larger value for the runoff series in the same period was estimated to be80days. The results indicates that, compared with the weather factor greatly influencing the chaotic characteristics of chlorophyll a, the runoff factor is clearly weaker. The results of analysis are expected to provide a new thought for the researches of predictablity of algae in river.In conclusion, the achievements of the thesis are expected to help broaden the research method of prediction theory, improve the prediction accuracy of algae, and reveal the real growth rhythm of algae. The jobs in the thesis show the important practical significance for improving the level of water quality management of portable water source district.更多还原