Negative effects of sufficiently small initial weights on back-propagation neural networks

In the training of feedforward neural networks,it is usually suggested that the initial weights should be small in magnitude in order to prevent premature saturation.The aim of this paper is to point out the other side of the story:In some cases,the gradient of the error functions is zero not only for infinitely large weights but also for zero weights.Slow convergence in the beginning of the training procedure is often the result of sufficiently small initial weights.Therefore,we suggest that,in these cases,the initial values of the weights should be neither too large,nor too small.For instance,a typical range of choices of the initial weights might be something like(0.4,0.1) ∪(0.1,0.4),rather than(0.1,0.1) as suggested by the usual strategy.Our theory that medium size weights should be used has also been extended to a few commonly used transfer functions and error functions.Numerical experiments are carried out to support our theoretical findings.     

Uniform modeling of parameter dependent nonlinear systems

This paper addresses the problem of approximating parameter dependent nonlinear systems in a unified framework. This modeling has been presented for the first time in the form of parameter dependent piecewise affine systems. In this model, the matrices and vectors defining piecewise affine systems are affine functions of parameters. Modeling of the system is done based on distinct spaces of state and parameter, and the operating regions are partitioned into the sections that we call ’multiplied simplices’. It is proven that this method of partitioning leads to less complexity of the approximated model compared with the few existing methods for modeling of parameter dependent nonlinear systems. It is also proven that the approximation is continuous for continuous functions and can be arbitrarily close to the original one. Next, the approximation error is calculated for a special class of parameter dependent nonlinear systems. For this class of systems, by solving an optimization problem, the operating regions can be partitioned into the minimum number of hyper-rectangles such that the modeling error does not exceed a specified value. This modeling method can be the first step towards analyzing the parameter dependent nonlinear systems with a uniform method.     

基于灰度共生矩阵的图像纹理特征提取分析

图像纹理作为一种重要的视觉手段,是图像中普遍存在而又难以描述的特征。目前常用的纹理特征提取的方法主要有统计方法、模型方法、信号处理方法和结构方法。灰度共生矩阵即为灰度级的空间相关矩阵,以其为基础的统计方法通过对矩阵统计量的求取较好地提取到了纹理特征,通过选取关键参数编程并进行仿真实现,分别求取了四个方向的灰度共生矩阵及其特征量来分析图像的纹理特征。     

储能设备获最大功率的改进型光伏充电系统设计

针对光伏充电系统中能量利用率不高的问题,给出了改进型光伏充电系统的硬件及软件系统的设计方法。该系统基于低功耗单片机MSP430F1611,利用扰动观察法,通过控制数字电位器来间接调节MPPT变换器的输出电压,从而光伏电池功率与MPPT变换器效率在比值最优的情况下,可使得储能设备获得最大功率。经过仿真验证,该方法可以达到预期的效果,储能设备在同等条件下能获得更多的电能。     

基于像素拓展的图形分割与合成的方法研究

提出了一种新的图像分割与合成方法。对于图像的分割,提出了新的分割算法,先是手工将一些明显属于物体和明显属于背景的像素分割出来,然后创建一个小矩形区域进行像素拓展迭代,迭代到图像边界为止。对于图像的合成,采用平均法,设计了新的算法,使物体的基本颜色保持不变的同时,降低了合成图像中从源图像分割的物体与新背景的差异,并使得其饱和度、色度、强度更加自然真实。     

基于跨层快速邻居感知的OLSR协议

提出了一种基于跨层快速邻居感知的OLSR协议——FS-OLSR,协议在采用hello消息进行链路感知和邻居探测的基础上,结合链路层反馈的信息对网络层邻居表进行更新,以增强节点的邻居感知能力;并将邻居节点的新鲜度作为MPR计算的依据,以提高拓扑消息中广播邻居节点的新鲜度,使之更加适合于快速移动场景。仿真结果表明,在快速移动场景中,FS-OLSR能够在增加少量路由开销的情况下,显著降低丢包率。     

基于空洞模型的地理位置路由改进算法研究

针对现有WSN中地理位置路由遇路由空洞后会产生冗余路由的问题,提出一种基于感知空洞形状的分段贪婪路由(Easy Modeling Greedy Routing,EMGR)算法。EMGR引入虚拟坐标概念,利用空洞边界探测包收集空洞边界节点信息,将其映射到虚拟坐标系中;根据空洞边界节点在虚拟坐标系的不同分布,定义凹、凸域来概括路由空洞形状;根据不同的空洞形状找到合适的中转节点,采用分段贪婪路由方式传输数据。仿真实验结果表明,相较于GPSR算法,EMGR算法产生的路由跳数减少了35%,网络能耗低,路径扩张率和网络时延小。     

数字磁通门传感器的自动相位对准

三轴数字磁通门传感器信号处理电路的延时(相位延迟)及各轴延时不一致,影响数字滤波器的输出性能。提出了基于加权最小二乘法的互功率谱相位法计算延时的方法,针对该数字滤波器的特性,选择传感器的不同谐波信号对比分析相位延迟计算的精度,并设计了传感器的自动相位对准方法。实验结果表明:该计算方法精确有效,自动相位对准使滤波器输出性能提高25.89%。     

基于间断性腐蚀法的铂热敏传感器的研制

针对铂湿法刻蚀工艺,提出了间断性腐蚀法,并与边腐蚀边搅拌法和静止法进行了对比,结果显示该方法腐蚀铂曲线光滑、效果好。之后,以普通硅片作为衬底材料,二氧化硅作为隔热层,采用上述微机电(MEMS)工艺加工制作了铂热敏传感器。经热敏性能测试,其电阻温度系数在20℃～80℃时为1571.2×10-6/℃;同一批次电阻阻值均匀性为0.35%;非线性度为0.57%;热时间常数为1.1μs;在0.5 h的稳定性测试中其变化幅值为0.005Ω,即精度为0.01%。该热敏传感器制作工艺简单,性能优异,可用于温度敏感、气体传感等热敏传感。     

连续体结构的参数化水平集统一优化

针对传统分步式结构优化设计的不足,提出一种同时进行结构拓扑、形状和尺寸统一优化的设计方法.首先采用水平集函数描述统一的结构优化模型和几何尺寸边界,通过引入紧支径向插值基函数将结构拓扑优化变量、形状优化变量和尺寸优化变量变换为基函数的扩展系数;然后取该扩展系数为设计变量,借助一种参数的变化表达3种优化要素对结构性能的影响,将复杂的多变量优化问题变换为相对简单的参数优化问题,有利于与相对成熟的优化算法相结合提高求解效率;进一步用R函数将其融合为一个整体,构造出统一优化模型,并用最优化准则法进行求解.最后通过数值案例证明了该方法的有效性和精确性.