基于改进词嵌入表示和卷积神经网络的银行智能派单系统的研究

银行智能派单系统的实现和功能完善,对银行提升客户满意度、提高突发事件处理效率、降低人工处理成本等非常重要。针对现有的基于Word2vec和TextCNN模型的银行智能派单系统进行了改进,针对特征词权重表达性弱,特征词类别及位置区分性弱等问题,提出基于改进TF-IDF加权的Word2vec词嵌入表示和卷积神经网络结合的银行智能派单系统:首先利用Word2vec模型得到输入事件单的词嵌入向量;再针对经典TF-IDF方法不具备类别区分性、位置区分性,也没有考虑极端频率特征词代表性的情况,提出改进型TF-IDF算法,计算每个特征词的权重,得到基于改进TF-IDF加权的Word2vec词嵌入表示;最后在卷积神经网络模型中进行训练,通过迭代训练最终得到分类器,利用分类器可对输入事件单信息自动进行系统类别的判断。实验结果表明改进词嵌入表示的银行智能派单系统分类模型的宏查准率、宏查全率、准确率以及宏F1值都得到进一步的提高。     

氯碱企业快速熔断器运行状况调查报告

以2018年58家氯碱企业供电、整流技术与设备现状调查数据为依据,阐述了国内氯碱企业中快速熔断器的运行状况。调查数据中包括50家氯碱企业的快速熔断器的基本运行数据、投运时间及设备来源等。汇总了各企业在氯碱生产中使用快速熔断器时遇到的问题、发生的故障甚至酿成的生产事故及对氯碱生产造成的影响。并分析了事故发生的原因,提出了预防及解决措施,总结了事故处理经验,旨为氯碱生产提供参考依据。     

改进BP神经网络的再生粗骨料混凝土强度预测*

抗压强度作为评价再生混凝土主要性能指标之一,从四篇文献中搜集了43个抗压强度样本。基于改进的BP神经网络理论,建立了以再生粗骨料取代率、水灰比、龄期为输入,抗压强度为输出的神经网络模型,其结构形式为3-7-9-1。对该网络进行训练和学习,对并测试样本进行检测,结果表明,训练效率不仅大大提高,且而误差较小,并结合回归拟合系数,进一步表明该模型能够有效的的预测再生混凝土抗压强度值,能满足工程需要。     

某污水处理厂MBR膜系统改造与运行管理

某污水处理厂设计规模为10万t/d,采用AAO+MBR工艺,设计出水水质满足《城镇污水处理厂污染物排放标准》(GB 18918-2002)一级A标准。由于运行时间较长,MBR运行系统存在产能下降、膜丝断丝脱落、污堵和清洗频次以及强度增加等问题,严重影响正常生产运行。针对MBR系统运行中存在的问题,文中提出具体改造建议,并总结运行经验,进一步优化MBR膜系统的运行维护管理,为其他污水厂MBR膜系统运行问题提供解决思路和维护管理经验。     

Environmental economic dispatch in multi-area power system employing improved differential evolution with fuzzy selection

In a deregulated multi-area electrical power system the objective is to determine the most economical generation dispatch strategy that could satisfy the area load demands, the tie-line limits and other operating constraints. Usually, economic dispatch (ED) deals only with the cost minimization, but minimization of emission content has also become an equally important concern due to the mandatory requirement of pollution reduction for environmental protection. Environmental economic dispatch (EED) is a complex multi-objective optimization (MOO) problem with conflicting goals. Normally a fuzzy ranking is employed to rank the large number of Pareto solutions obtained after solving a MOO problem. But in this paper the preference of the decision maker (DM) is used to guide the search and to select the population for the next generation. An improved differential evolution (DE) method is proposed where the selection operation is modified to reduce the complexity of multi-attribute decision making with the help of a fuzzy framework. Solutions are assigned a fuzzy rank on the basis of their level of satisfaction for different objectives before the population selection and then the fuzzy rank is used to select and pass on better solutions to the next generation. A well distributed Pareto-front is obtained which presents a large number of alternate trade-off solutions for the power system operator. A momentum operation is also included to prevent stagnation and to create Pareto diversity. Studies are carried out on three test cases and results obtained are found to be better than some previous literature.     

大型球磨机订购（制造）设计联络中零部件结构优化的探讨

针对大型球磨机使用中存在的问题，对球磨机主要零部件结构缺陷进行分析并提出解决方案。通过设计联络，提前对球磨机主要零部件结构进行优化，消除设备缺陷。大型球磨机设计联络中对主要零部件结构进行优化，对提高球磨机的运转率有着事半功倍的作用。     

Development of a novel algorithm for optimal capacitor placement in distribution systems

Optimal capacitor placement in distribution systems solved by the hybrid method of CODEQ (called HCODEQ method) is proposed in this work. The concepts of chaotic search, opposition-based learning, and quantum mechanics are used in the CODEQ method to overcome the drawback of parameters selection in the differential evolution (DE). However, a larger population size must be used in the CODEQ method. That is a drawback for all evolutionary algorithms (EAs). To overcome this drawback, acceleration operation and migrating operation are embedded into the CODEQ method, i.e. HCODEQ method. The use of these two operations can increase the convergence speed without decreasing the diversity among individuals. One benchmark function and various-scale capacitor placement systems are used to compare the performance of the proposed method, CODEQ method, DE, simulated annealing (SA), and ant system (AS). Numerical results show that the performance of the HCODEQ method is better than the other methods.