选择性集成算法分类与比较

选择性集成是当前机器学习领域的研究热点之一。由于选择性集成属于NP"难"问题,人们多利用启发式方法将选择性集成转化为其他问题来求得近似最优解,因为各种算法的出发点和描述角度各不相同,现有的大量选择性集成算法显得繁杂而没有规律。为便于研究人员迅速了解和应用本领域的最新进展,本文根据选择过程中核心策略的特征将选择性集成算法分为四类,即迭代优化法、排名法、分簇法、模式挖掘法;然后利用UCI数据库的20个常用数据集,从预测性能、选择时间、结果集成分类器大小三个方面对这些典型算法进行了实验比较;最后总结了各类方法的优缺点,并展望了选择性集成的未来研究重点。

一种BP神经网络改进算法的研究

为提高神经网络传统BP算法的训练速度,以3层神经网络为例,通过对权值的分析与优化,推导出改良的BP算法——双权值迭代优化法,并对该算法与传统算法进行了比较,通过比较发现,新算法在保证精度的前提下可节省训练时间,同时对该算法特点进行了总结。

基于网络谐波电压综合补偿的配电网并联型有源电力滤波器优化配置策略研究

并联型有源电力滤波器(shunt active power filter,SAPF)多配置于非线性负载的并网点并对本地负载进行谐波电流补偿。随着配电网中非线性负载的广泛接入,导致基于本地补偿的配置方式不再具有经济、高效的优点,为了解决此问题,出现了以配电网整体多节点的电能质量为优化目标的综合补偿配置方案。基于此,本文提出了一种综合考虑配电网各节点电压谐波含量、SAPF总安装台数及安装容量的SAPF配置算法,结合多目标粒子群及单目标迭代算法确定在网络各节点谐波电压畸变率满足标准时SAPF的数量、配置节点及最优输出电流。最后,通过仿真搭建IEEE18节点标准模型,验证了所提出的算法在多台SAPF配置问题中的有效性。

基于共享阵列的集成雷达和通信波形设计

为了减少频谱和硬件资源对雷达发射过程的消耗,文中提出了三种用共享阵列发射雷达和通信波形的方法。利用最小范数优化法合成不同方向的雷达和通信波形,利用此方法形成的发射波束图具有低旁瓣电平和较高的峰均功率比。因此,在最小范数优化的基础上提出迭代优化法来设计恒模发射波形。为了进一步降低由设计的等模波形形成的发射波束图的旁瓣电平,对每个发射波形的幅度进行迭代优化,最终得到幅度加权迭代优化法。数值结果表明:所设计的发射波形能同时完成雷达和通信功能,并能形成低旁瓣电平的发射波束图。

通信感知一体化混合波束赋形技术

为解决无线通信与感知的性能日益强大而带来的频谱资源紧缺的问题,通信感知一体化(Integrated Sensing and Communication,ISAC)技术逐渐开始受到重视。在目前最有发展潜力的毫米波(millimeter Wave,mmWave)多输入输出(Multiple Input Multiple Output,MIMO)混合波束赋形系统基础上,提出了一种通信感知一体化的波束赋形算法。使用均方误差(Mean Square Error,MSE)衡量该系统的通信和雷达的性能,通过引入权重因子将通信与雷达的性能指标综合考虑,得到通感一体化波束赋形系统的最优解。针对求解过程中的非凸优化问题,提出了基于坐标迭代的交替优化算法对问题进行求解。针对不同权重因子,对通信的频谱效率和雷达的波束方向图进行了仿真,仿真结果验证了所提方案可以实现通信感知一体化系统下通信与感知性能的折中。

An analytical method to calculate station evacuation capacity

The major objective of this work was to calculate evacuation capacity and solve the optimal routing problem in a given station topology from a network optimization perspective where station facilities were modelled as open finite queueing networks with a multi-objective set of performance measures. The optimal routing problem was determined so that the number of evacuation passengers was maximized while the service level was higher than a certain criterion. An analytical technique for modelling open finite queueing networks, called the iteration generalized expansion method(IGEM), was utilized to calculate the desired outputs. A differential evolution algorithm was presented for determining the optimal routes. As demonstrated, the design methodology which combines the optimization and analytical queueing network models provides a very effective procedure for simultaneously determining the service level and the maximum number of evacuation passengers in the best evacuation routes.

An Improved Control Vector Iteration Approach for Nonlinear Dynamic Optimization (Ⅰ) Problems Without Path Constraints

This study proposes an efficient indirect approach for general nonlinear dynamic optimization problems without path constraints. The approach incorporates the virtues both from indirect and direct methods: it solves the optimality conditions like the traditional indirect methods do, but uses a discretization technique inspired from direct methods. Compared with other indirect approaches, the proposed approach has two main advantages: (1) the discretized optimization problem only employs unconstrained nonlinear programming (NLP) algorithms such as BFGS (Broyden-Fletcher-Goldfarb-Shanno), rather than constrained NLP algorithms, therefore the computational efficiency is increased; (2) the relationship between the number of the discretized time intervals and the integration error of the four-step Adams predictor-corrector algorithm is established, thus the minimal number of time intervals that under desired integration tolerance can be estimated. The classic batch reactor problem is tested and compared in detail with literature reports, and the results reveal the effectiveness of the proposed approach. Dealing with path constraints requires extra techniques, and will be studied in the second paper.

Iteration Linear Programming Method and Livestock Ration Optimization

Iterative linear programming methods are proposed for optimum balanced animal diet in this paper. According to ＂wooden bucket theory＂ of the nutritional balance, each nutrient in the feeding standard has equal importance. It＇s unreasonable to use common goal programming to attach different weighted value to different nutritional parameters. This paper introduces an effective algorithm to deal with this kind of problem. When the permitting cost of livestock ration is given, we can design a ration formula with linear program-this is the first round. Then, according to the differences between the permitting cost and the formula cost gained in the first round, adjust the feeding standard and the feeding raw materials, and conduct the second round of linear programming for ration formula. If there is still a very big difference between the formula cost and the permitting cost, the third round will be taken, and so on. In this iteration course the formula cost gradually approaches the permitting cost. It is the key that the feeding standard and feeding raw materials are modified in each round. This method ensured the nutritive equilibrium with the formulation of least-cost ration. This is an especially important method when the primary goal of the optimization tool is to improve economic and nutritive efficiency.

Research on Optimizing Computer Network Structure based on Genetic Algorithm and Modified Convex Optimization Theory

In this paper, we report in-depth analysis and research on the optimizing computer network structure based on genetic algorithm and modified convex optimization theory. Machine learning method has been widely used in the background and one of its core problems is to solve the optimization problem. Unlike traditional batch algorithm, stochastic gradient descent algorithm in each iteration calculation, the optimization of a single sample point only losses could greatly reduce the memory overhead. The experiment illustrates the feasibility of our proposed approach.

A NEWε-GENERALIZED PROJECTION METHOD OF STRONGLY SUB-FEASIBLE DIRECTIONS FOR INEQUALITY CONSTRAINED OPTIMIZATION

In this paper, the nonlinear optimization problems with inequality constraints are discussed. Combining the ideas of the strongly sub-feasible directions method and the s-generalized projection technique, a new algorithm starting with an arbitrary initial iteration point for the discussed problems is presented. At each iteration, the search direction is generated by a new s-generalized projection explicit formula, and the step length is yielded by a new Armijo line search. Under some necessary assumptions, not only the algorithm possesses global and strong convergence, but also the iterative points always get into the feasible set after finite iterations. Finally, some preliminary numerical results are reported.

A GENERAL ITERATIVE METHOD OF FIXED POINTS FOR EQUILIBRIUM PROBLEMS AND OPTIMIZATION PROBLEMS

The purpose of this paper is to present a general iterative scheme as below：{F（un,y）＋1/rn（y-un,un-xn）≥0,y∈C,xn＋1=（I-αnA）Sun＋αnγf（xn）and to prove that, if {an} and {rn} satisfy appropriate conditions, then iteration sequences {xn} and {un} converge strongly to a common element of the set of solutions of an equilibrium problem and the set of fixed points of a nonexpansive mapping and the set of solution of a variational inequality, too. Furthermore, by using the above result, we can also obtain an iterative algorithm for solution of an optimization problem min h（x）, where h（x） is a convex and lower semicontinuous functional defined on a closed convex subset C of a Hilbert space H. The results presented in this paper extend, generalize and improve the results of Combettes and Hirstoaga, Wittmann, S.Takahashi, Giuseppe Marino, Hong-Kun Xu, and some others.

ＩＴＥＲＡＴＩＶＥＡＬＧＯＲＩＴＨＭＳＦＯＲＤＡＴＡ ＡＳＳＩＭＩＬＡＴＩＯＮＰＲＯＢＬＥＭＳ

Iterative algorithms for solving the data assimilation problems are considered，based on the main and adjoint equations.Spectral properties of the control operators of the problem are studied, the iterative algorithms are justified.