东江流域水利工程对流域地表水文过程影响模拟研究

东江流域为香港及广州等特大城市的重要水源地,新丰江、枫树坝与白盆珠三大水库对东江全流域水资源有控制性调蓄作用,加上人类活动影响,导致东江流域地表径流发生显著变异.研究分析东江流域上中下游水利工程建成前、后径流在时间与空间的变异特征,对水利工程与人类活动影响的东江流域水资源管理提供重要依据.基于此,选取多元线性回归、神经网络与支持向量机等模型模拟水利工程建成前的径流变化,研究认为:①神经网络为水利工程建成后径流变化模拟的最优模型;②水利工程建成后,流域径流变异特征分两阶段:1974-1982年流域水文变异不显著,水利工程影响不大;1983-2000 年径流变异显著,水利工程削峰填谷作用明显;③东江流域龙川、岭下、博罗站水文变化均受水利工程调蓄影响,但影响程度各异:博罗站水文变化受水利工程影响最大,岭下站次之,龙川站受影响最小.另外,在水利工程调蓄作用影响下,东江流域月径流变化幅度趋于平坦,在一定程度上降低了洪水风险及减少了极端气候条件对东江流域水资源供给造成的负面影响.     

基于PSO算法的BP神经网络对水体叶绿素a的预测

BP神经网络(Back Propagation Network)在水体富营养化评价及预测中已广泛应用,但传统BP算法的收敛速度慢并易陷入局部最优. 提出了一种基于微粒群(PSO)算法的BP神经网络模型,利用PSO对神经网络的权值进行修正,优化神经网络结构及算法全局收敛性. 选择最能代表明湖水质状况的5号采样点作为研究对象,把2009年4月—2010年3月的月样本插值为周样本,对明湖ρ(Chla)的短期变化趋势进行了预测,并用6号采样点数据来验证网络的泛化能力. 比较分析基于PSO算法的新模型与传统BP算法模型的预测精度表明,新模型有效克服了传统算法的缺点,提高了网络的预测能力和学习能力.      

发挥供销系统网络优势 延伸餐厨废弃物资源利用的思考

对处于国家中心城市地位的广州来说,随着城市人口、经济社会不断发展,尤其随着餐饮行业的蓬勃发展,日产餐厨废弃物近1万t,如不妥善合理处理,将对城市环境和城市管理造成极大压力。无论是焚烧、填埋、还是生化处理模式,餐厨废弃物资源化处理后产品的出路都没有得到有效解决。广州供销社系统在产业链后端延伸方面具有我国任何社会组织无法比拟的优势,是解决餐厨废弃物资源化产品深加工出路的重要力量。除了发挥供销社网络优势外,仍需要政府在政策、土地、资金等方面的大力支持,共同推进,才能实现餐厨废弃物"由餐桌中来,加工调制后再到农田中去,生产出来的有机食品再回到餐桌"的真正循环。     

不同管材对氯胺消毒副产物生成与水质生物稳定性的影响

模拟考察了上海某水厂出厂水在不同输配管网体系中水质的变化规律,研究了管材和氯胺投加量对输配过程水质变化的影响.结果表明,不同输配管网体系氯胺的衰减速率总体上符合如下规律:镀锌管>铜管>不锈钢管>聚乙烯管(PE管)>三型聚丙烯管(PPR管),季节因素对氯胺衰减速率影响较大,秋冬季节氯胺衰减的速率显著降低;消毒副产物三卤甲烷(THMs)和卤乙酸(HAAs)的生成规律为:PE管?PPR管>玻璃管>不锈钢管>铜管>镀锌管;五种管材中异氧菌总数(HPC)的生长数量为:镀锌管>>PE管>不锈钢管>PPR管>铜管;铜管和PE管中生物可同化有机碳(AOC)显著高于其余管材及出厂水.     

Simulating wildfire patterns using a small-world network model

This paper presents the development and validation results of a weighted small-world network model designed to simulate fire patterns in real heterogeneous landscapes. Fire spread is simulated on a gridded landscape, a mosaic in which each cell represents an area of the land surface. In this model, the interaction between burning and non-burning cells (here, due to flame radiation) may extend well beyond nearest neighbors, and depends on local conditions of wind, topography, and vegetation. An approach based on the coupling of the solid flame model with the Monte Carlo method is used to predict the radiative heat flux from the flame generated by the burning of each combustible cell to its neighbors. The weighting procedure takes into account latency (a combustible cell will only ignite when it has accumulated enough energy along time) and flaming persistence of burning cells. The model is applied to very different fire scenarios: a historical Mediterranean fire that occurred in southeastern France in 2005 and experimental fires conducted in arid savanna fuels in South Africa in 1992. Model results are found to be in agreement with real fire patterns, in terms both of rate of spread, and of the area and shape of the burn. This work also shows that the fractal properties of fire patterns predicted by the model are similar to those observed from satellite images of three other Mediterranean fire scars.     

Social network properties within a fish assemblage invaded by non-native sunbleak Leucaspius delineatus

Network theory is commonly used to reveal social interactions and the organisation of interconnected nodes—but has not yet been applied to animal invasions. Non-native species invasions are now considered one of the foremost threats to natural ecosystems and biodiversity. This is the first attempt to assess social network properties within a freshwater fish assemblage invaded by a non-native fish species. We show that invasive sunbleak Leucaspius delineatus is socially more strongly interconnected with native species than the native species with each other. The social networks also reveal characteristics of a ‘small world’ such as low clustering coefficients C and short path lengths L. The findings may indicate potential traits of successful invaders and the implication for the spread of pathogens between individuals within a group of animals that contain a non-native invasive species. The success of establishment and subsequent invasion may be highlighted not only by the capacity of the new species to adapt to the new environment, but also in its capacity to penetrate the social circle of the native community.     

基于RBF神经网络的鄱阳湖表层水体总悬浮颗粒物浓度遥感反演

为进一步提高湖泊总悬浮颗粒物浓度遥感反演的准确性,引进适应复杂非线性映射的RBF神经网络模型,以鄱阳湖通江湖体为例进行了实证分析,根据实测水体悬浮颗粒物浓度和MODIS遥感数据,对遥感数据进行预处理,建立了RBF神经网络悬浮颗粒物浓度反演模型,神经元个数为8个,误差性能目标值为0.001,对悬浮颗粒物浓度进行反演。研究结果表明,验证样本相关系数R2=0.956 8,均方根误差RMSE=0.54。利用神经网络模型反演水悬浮颗粒物浓度是有效的,其反演结果优于非线性回归模型的结果。     

用微生物群落评价水系的自组织人工神经网络方法

利用原生物群落在PFU上群集过程中的变化,可以评价水质和监测水污染,运用T．Kohonen自组织人工神经网络模型,根据NacArthur－Wilson平衡模型提出了3个功能参数和另外2个结构参数,对常德市16个站四季的水质进行了分析,建立了水质污染程度预测的计算机智能专家系统,预测成功率达100％。     

水利水电工程环境影响综合评价的人工神经网络专家系统

在分级加权评价模型层次结构体系基础上,以三峡工程为背景,应用人工神经网络技术,建立了一个水利水电工程环境影响综合评价的人工神经网络专家系统,该专家系统的知识获取和存贮方式与普通专家系统不同,具有较高的推理效率,较强的容错、自适应和自我更新能力。      

Prediction of high-density polyethylene pyrolysis using kinetic parameters based on thermogravimetric and artificial neural networks

● Reducting the sampling frequency can enhance the modelling process. ● The pyrolysis of HDPE was investigated at three different heating rates. ● The average E_a and k₀ were calculated by Friedman, KAS, FWO, and CR methods. ● ANN was employed to predict the HDPE weight loss with the optimal MSE and R². Pyrolysis is considered an attractive option and a promising way to dispose waste plastics. The thermogravimetric experiments of high-density polyethylene (HDPE) were conducted from 105 °C to 900 °C at different heating rates (10 °C/min, 20 °C/min, and 30 °C/min) to investigate their thermal pyrolysis behavior. We investigated four methods including three model-free methods and one model-fitting method to estimate dynamic parameters. Additionally, an artificial neural network model was developed by providing the heating rates and temperatures to predict the weight loss (wt.%) of HDPE, and optimized via assessing mean squared error and determination coefficient on the test set. The optimal MSE (2.6297 × 10⁻²) and R² value (R² > 0.999) were obtained. Activation energy and pre-exponential factor obtained from four different models achieves the acceptable value between experimental and predicted results. The relative error of the model increased from 2.4 % to 6.8 % when the sampling frequency changed from 50 s to 60 s, but showed no significant difference when the sampling frequency was below 50 s. This result provides a promising approach to simplify the further modelling work and to reduce the required data storage space. This study revealed the possibility of simulating the HDPE pyrolysis process via machine learning with no significant accuracy loss of the kinetic parameters. It is hoped that this work could potentially benefit to the development of pyrolysis process modelling of HDPE and the other plastics.