CONVERGENCE OF ONLINE GRADIENT METHOD WITH A PENALTY TERM FOR FEEDFORWARD NEURAL NETWORKS WITH STOCHASTIC INPUTS

Online gradient algorithm has been widely used as a learning algorithm for feedforward neural network training. In this paper, we prove a weak convergence theorem of an online gradient algorithm with a penalty term, assuming that the training examples are input in a stochastic way. The monotonicity of the error function in the iteration and the boundedness of the weight are both guaranteed. We also present a numerical experiment to support our results.

Angle estimation in bistatic MIMO radar using improved reduced dimension Capon algorithm

This paper discusses the problem of direction of departure （DOD） and direction of arrival （DOA） estimation for a bistatic multiple input multiple output （MIMO） radar, and proposes an improved reduced-dimension Capon algorithm therein. Compared with the reduced-dimension Capon algorithm which requires pair matching between the two-dimensional angle estimation, the pro- posed algorithm can obtain automatically paired DOD and DOA estimation without debasing the performance of angle estimation in bistatic MIMO radar. Furthermore, the proposed algorithm has a lower complexity than the reduced-dimension Capon algorithm, and it is suitable for non-uniform linear arrays. The complexity of the proposed algorithm is analyzed and the Cramer-Rao bound （CRB） is also derived. Simulation results verify the usefulness of the proposed algorithm.

A Multiple Model Approach to Modeling Based on LPF Algorithm

Input-output data fitting methods are often used for unknown-structure nonlinear system modeling. Based on model-on-demand tactics, a multiple model approach to modeling for nonlinear systems is presented. The basic idea is to find out, from vast historical system input-output data sets, some data sets matching with the current working point, then to develop a local model using Local Polynomial Fitting (LPF) algorithm. With the change of working points, multiple local models are built, which realize the exact modeling for the global system. By comparing to other methods, the simulation results show good performance for its simple, effective and reliable estimation.

Functional-type Single-input-rule-modules Connected Neural Fuzzy System for Wind Speed Prediction

Wind is one kind of clean and free renewable energy sources. Wind speed plays a pivotal role in the wind power output. However, due to the random and unstable nature of the wind, accurate prediction of wind speed is a particularly challenging task. This paper presents a novel neural fuzzy method for the hourly wind speed prediction. Firstly, a neural structure is proposed for the functional-type single-input-rule-modules(FSIRMs) connected fuzzy inference system(FIS) to combine the merits of both the FSIRMs connected FIS and the neural network. Then, in order to achieve both the smallest training errors and the smallest parameters, a least square method based parameter learning algorithm is presented for the proposed FSIRMs connected neural fuzzy system(FSIRMNFS). Further,the proposed FSIRMNFS and its parameter learning algorithm are applied to the hourly wind speed prediction. Experiments and comparisons are also made to show the effectiveness and advantages of the proposed approach. Experimental results verified that our study has presented an effective approach for the hourly wind speed prediction. The proposed approach can also be used for the prediction of wind direction, wind power and some other prediction applications in the research field of renewable energy.

基于改进粒子群算法的机械臂振动抑制研究

SCARA机械臂在工作过程中和工作终止后都会出现振动问题,为了降低机械臂的残余振动,选取粒子群作为寻最优抑振轨迹的优化算法,针对粒子群算法在轨迹寻优时收敛性差以及出现停滞现象问题,提出采用非线性递减惯性权重和粒子种群跳跃的方法对粒子群算法进行改进。利用机械臂振动抑制实验平台,使用改进的粒子群算法进行抑振轨迹优化并进行实验研究,验证结果可行且有效,得到抑振指标最小的最优抑振轨迹。

Performance Improvement of Multi-User Multiple-Input Multiple-Output Protocol for WLAN

The increase in the number of devices with a massive revolution in mobile technology leads to increase the capacity of the wireless communications net-works. Multi-user Multiple-Input Multiple-Output is an advanced procedure of Multiple-Input Multiple-Output, which improves the performance of Wireless Local Area Networks. Moreover, Multi-user Multiple-Input Multiple-Output leads the Wireless Local Area Networks toward covering more areas. Due to the growth of the number of clients and requirements, researchers try to improve the performance of the Medium Access Control protocol of Multi-user Multiple-Input Multiple-Output technology to serve the user better, by supporting different data sizes, and reducing the waiting time to be able to transmit data quickly. In this paper, we propose a Clustering Multi-user Multiple-Input Multiple-Output protocol, which is an improved Medium Access Control protocol for Multi-user Multiple-Input Multiple-Out-put based on MIMOMate clustering technique and Padovan Backoff Algorithm. Utilizing MIMOMMate focuses on the signal power which only serves the user in that cluster, minimizes the energy consumption and increases the capacity. The implementation of Clustering Multi-user Multiple-Input Multiple-Output performs on the Network Simulator (NS2.34) platform. The results show that Clustering Multi-user Multiple-Input Multiple-Output protocol improves the throughput by 89.8%, and reduces the latency of wireless communication by 43.9% in scenarios with contention. As a result, the overall performances of the network are improved.

MIMO雷达分组正交波形集优化设计方法

在现代电子对抗中,数字射频存储(DRFM)设备能够快速截获机载脉冲多普勒雷达信号,能够实现对多输入多输出(MIMO)雷达的干扰。MIMO雷达可基于多组相互正交的波形集来对抗DRFM干扰。同时,为最大化MIMO雷达波形分集增益,每个脉冲内发射的波形也需要正交。为了平衡组内和组间的正交性,文中建立了一种分组正交波形集优化模型,其目标函数为组内和组间相关函数性能评估指标值的加权和;为了求解该优化问题,提出了一种分组正交波形集设计方法。所提方法将原优化问题简化为p-范数优化问题,基于MM算法导出了最小化目标函数的迭代求解表达式。仿真结果表明,所提方法可通过改变权重系数来灵活平衡MIMO雷达的干扰抑制性能和距离压缩性能。

Design and FPGA-Implementation of Minimum PED Based K-Best Algorithm in MIMO Detector

Minimum Partial Euclidean Distance (MPED) based K-best algorithm is proposed to detect the best signal for MIMO (Multiple Input Multiple Output) detector. It is based on Breadth-first search method. The proposed algorithm is independent of the number of transmitting/receiving antennas and constellation size. It provides a high throughput and reduced Bit Error Rate (BER) with the performance close to Maximum Likelihood Detection (MLD) method. The main innovations are the nodes that are expanded and visited based on MPED algorithm and it keeps track of finally selecting the best candidates at each cycle. It allows its complexity to scale linearly with the modulation order. Using Quadrature Amplitude Modulation (QAM) the complex domain input signals are modulated and are converted into wavelet packets and these packets are transmitted using Additive White Gaussian Noise (AWGN) channel. Then from the number of received signals the best signal is detected using MPED based K-best algorithm. It provides the exact best node solution with reduced complexity. The pipelined VLSI architecture is the best suited for implementation because the expansion and sorting cores are data driven. The proposed method is implemented targeting Xilinx Virtex 5 device for a 4 × 4, 64-QAM system and it achieves throughput of 1.1 Gbps. The results of resource utilization are tabulated and compared with the existing algorithms.

基于延迟线组合调相的高能效自适应混合预编码方案

针对毫米波大规模多输入多输出系统采用混合预编码技术所带来的高功耗影响,基于延迟线组合调相器件设计了一种间接全连接型混合预编码系统结构,并提出一种自适应连接的混合预编码方案,可大幅减少与天线直连的调相器数量从而降低硬件造成的功耗损失。方案将编码矩阵的求解分为数字编码矩阵、模拟编码矩阵和开关组成的连接交换矩阵3个部分进行交替优化;提出一种基于互补正交投影矩阵的算法来解决模拟编码矩阵的优化;利用功率约束条件和最小二乘算法简化数字编码矩阵的求解;针对连接交换矩阵的离散组合优化问题,利用低复杂度的交叉熵算法进行优化。仿真结果表明,所提方案可以保证系统性能在接近纯数字预编码方案的同时有效提高系统的能量效率。

Soft-output stack algorithm with lattice-reduction for MIMO detection

A computationally efficient soft-output detector with lattice-reduction （LR） for the multiple-input multiple-output （MIMO） systems is proposed. In the proposed scheme, the sorted QR de- composition is applied on the lattice-reduced equivalent channel to obtain the tree structure. With the aid of the boundary control, the stack algorithm searches a small part of the whole search tree to generate a handful of candidate lists in the reduced lattice. The proposed soft-output algorithm achieves near-optimal perfor- mance in a coded MIMO system and the associated computational complexity is substantially lower than that of previously proposed methods.

鲁棒参数自适应微振动控制算法

递推最小二乘(Recursive Least Squares,RLS)算法因其简单、快速的特点,在微振动自适应控制领域被广泛应用。由于微振动主动控制系统中扰动环境的特殊性及复杂性,需要重点考虑微振动控制中所采用的参数自适应算法在参数估计过程中的鲁棒性。针对多输入多输出(Multiple Input Multiple Output,MIMO)微振动主动控制系统,基于无限冲激响应(Infinite Impulse Response,IIR)滤波器,提出一种结合死区和归一化的MIMO鲁棒参数自适应算法,并给出其详细的算法推导与收敛性分析。在此基础上,通过构建三自由度微振动主动振动控制实验系统,针对单频窄带扰动、双频窄带扰动展开了对比实验分析,相关的实验结果验证了所提出鲁棒参数自适应算法的可行性和鲁棒性。

不同输入变量对光伏功率异常数据修复的影响分析

光伏电站及分布式光伏设备输出功率数据记录因量测设备异常、通信故障、信号干扰等因素会出现异常,影响电网决策。因此,本文提出基于遗传算法优化初值的反向传播神经网络,利用GA-BP神经网络进行异常数据修复,建立线性内插法数据修复模型作为对照组,研究了以数值气象记录(辐照强度、气温、相对湿度、风速及风向)、天气类型、邻近相似电站功率等参数的不同组合作为神经网络的输入变量对修复效果的影响。实例分析表明,采用全部的输入变量可取得较好的修复效果。

An MMSE Decoding Algorithm without Matrix Inversion in QSTBC

The matrix inversion operation is needed in the MMSE decoding algorithm of orthogonal space-time block coding （OSTBC） proposed by Papadias and Foschini. In this paper, an minimum mean square error （MMSE） decoding algorithm without matrix inversion is proposed, by which the computational complexity can be reduced directly but the decoding performance is not affected.

华氏经济优化新理论的实证案例

本文是我们研究小组在文献[1,2]中关于华罗庚经济最优化理论新探索的继续,研究目前最新的2017年中国投入产出模型的稳定性分析、产品排序与分类、预测与调整和结构优化等问题;并将2007年、2012年、2017年三个年度投入产出模型的产品排序与分类进行对比分析.所得结果展示出三个年度、跨越15年的产品等级排序和分类的令人吃惊的相似性,再次显示出前文理论的可靠性.

面向模分复用系统的遗传-MIMO均衡参数优化技术

在模分复用系统的数字信号处理单元中,多输入多输出(Multi-Input Multi-Output,MIMO)均衡技术可用来补偿由各类模式相关噪声引起的信号误码率(Bit Error Rate,BER)劣化问题.而MIMO均衡算法的工作性能严重依赖于步长因子μ以及抽头数K,因此在固化均衡器之前,确定MIMO均衡算法中μ-K参数组合的最优值至关重要.为提高参数优化效率,提出了一种基于遗传算法(Genetic Algorithm,GA)的MIMO均衡参数优化方案,即遗传-MIMO(GAMIMO),在保证最小BER输出的同时降低参数优化过程所需的计算开销.为验证GA-MIMO的工作性能,构建了基于10 km六模光纤的点对点通信实验系统,使用新方案补偿并行通信的六路数据,并与最速下降法和迭代算法进行性能比较.实验结果表明,所提GA方案可实现MIMO均衡中最高99.98%的最优μ-K参数命中率,且GA-MIMO算法的全局搜索性使其相比于最速下降法和迭代算法可最多分别节省86.14%和90.3%的均衡算法调用次数,有效降低了确定最优μ-K参数组合时的计算开销.

基于聚类算法的叠加星座图可见光通信系统

简述了利用叠加星座图调制的方法来解决可见光多输入多输出通信系统容易受高信道相干性影响的问题,同时探究了机器学习中的聚类算法,以降低系统的误码率,提高通信传输的可靠性。研究结果表明,当星座图因受到非线性效应影响而发生旋转时,与不加入任何算法的系统相比,采用了K-means和模糊C均值算法的通信系统的误码率有了大幅度的下降,且不会受到旋转角度的影响。此外,对比分析2种聚类算法发现,K-means的原理和运算过程相对而言更加简便,可以更好地提高数据传输速率,提供更可靠安全的通信传输。

基于双深度输入凸神经网络多模型的中间点过热度预测控制

新能源大量并网,超临界火电机组参与调峰容易造成中间点过热度较大波动,从而导致过热蒸汽超温等问题。为较好控制中间点过热度达到稳定,提出了一种基于双深度输入凸神经网络多模型(muti-DDICNN model)的中间点过热度预测方法,分别训练了不同预测步长下子模型,构建了中间点过热度状态预测网络(SPNN)和误差预测网络(EPNN)。利用此预测网络凸性质,设计了一种基于双深度输入凸神经网络多模型预测控制器(DDICNN-MPC),将控制问题转化为凸优化问题,求取控制矩阵对目标函数的雅可比矩阵,采用梯度下降法计算控制矩阵最优解。仿真结果表明,DDICNN-MPC能快速平稳地跟踪中间点过热度设定值,且稳态误差较小,具有较好的调节能力。

一种机载分布式MIMO雷达节点位置与路径分步优化管控方法

机载分布式多输入多输出(Multiple-Input Multiple-Output,MIMO)雷达系统是基于机载分布式平台,采用多个雷达节点同时发射、同时接收信号的方式,协同处理多雷达回波以提升信噪比,进而提高雷达系统对目标区域的监视性能。系统资源管控能够显著提升节点位置、飞行路径等资源的利用率,增强探测目标能力,是机载分布式MIMO雷达系统的关键技术之一。本文研究了一种机载分布式MIMO雷达节点位置与路径分步优化管控方法。首先,根据雷达系统的探测需求、运动学约束、雷达节点位置等因素,建立了机载分布式MIMO雷达节点位置与路径优化模型。其次,利用粒子群算法(Particle Swarm Optimization,PSO)对机载分布式MIMO雷达节点位置进行优化求解得到各雷达节点最佳布站位置。随后,考虑机载分布式多节点不同路径匹配准则,包括航迹总和最短、最长航迹最短以及航迹残差最小准则,建立了多机协同逐帧路径优化模型,通过遗传算法(Genetic Algorithms,GA)进行逐帧求解,得到不同节点的优化飞行路径。仿真结果表明,相比常规布站方法,所提布站优化方法具有更好的区域监视性能;相比于直线飞行方案,所提路径优化方法所得飞行方案在逐帧区域监视性能上更优。

基于加权循环算法的高分辨成像波形设计

为实现感兴趣区域目标的高分辨成像,MIMO-SAS声呐(Multiple Input Multiple output SyntheticAper-tureSonar,MIMO-SAS)一般采用长的发射信号对目标进行照射,但发射信号过长会导致近距离目标和远距离目标回波产生混叠,从而引起距离模糊,影响目标成像质量。为了抑制距离向的模糊,通常的做法是降低脉冲重复率,然而脉冲重复率过低会导致MIMO-SAS方位向欠采样,使回波信号频谱相互混叠,声呐成像时会在方位向出现珊瓣,很难把目标回波有效分离出来。针对这个问题,提出一种基于加权循环算法(Muti-sequenceWeightedCyclic Algorithms,Multi-WeCAN)的阵列发射波形设计方法,通过计算机仿真发现,相比传统的发射波形,在不改变脉冲重复率的情况下,通过Multi-WeCAN波形设计方法能够有效抑制回波混叠,提高MIMO-SAS目标的成像质量。

基于PSO-NARX网络的司机驾驶行为分析方法

舒适性、准时性、节能性等是衡量高速铁路自动驾驶水平的重要指标,通过不断学习优秀司机的驾驶行为,可以优化列车自动驾驶性能,促进高速铁路自动驾驶技术的发展。基于现场列车运行数据,提出一种带有外部输入的非线性自回归(NARX)网络的列车司机驾驶行为分析方法。该方法构建了具有时序特征的NARX网络模型,并选取多项影响司机决策的参数作为输入,利用粒子群优化算法(PSO)确定网络的权重和阈值,对下一时刻列车运行情况进行预测。仿真结果表明:本文提出的PSO-NARX网络分析模型的预测效果优于前馈型神经网络(BP)、PSO-BP、NARX,相比于BP算法,迭代步数降低了373步,误差降低了8382%,相关系数达到了90117%。通过此预测,可以优化列车的自动驾驶设备性能指标,保障列车准时的同时,提高了乘客乘坐的舒适性。