PSO-SVM模型的构建与应用

为了使支持向量机(SVM)获得更好的分类效果,针对人为选择参数的随机性,提出了利用粒子群算法(PSO)进行参数自动选取的优化方法,构建了PSO-SVM模型.在个人信用评估中,通过对粒子适应度函数的设置来控制造成较大损失的第二类误判,应用结果表明:模型在训练和测试样本中的分类精度可以达到95%,第二类误判率分别仅为0.78%和2.02%.利用PSO对SVM中的参数进行优化,可以避免人为选择的随机性,并且在解决分类问题中表现出较好的稳健性.     

基于多特征融合的运动物体识别方法研究

为了实现运动目标的实时检测,本文基于多特征融合,提出了一种运动物体分级识别方法.首先通过对运动目标的分割,分析得到各个物体的形状矩特征、面积大小及运动速度.对上述特征,采用支持向量机方法对其矩特征进行快速的一级分类,然后对一级分类结果及各物体的面积、速度等信源根据其对分类的重要程度,采用模糊积分技术进行多特征融合分类.实验表明,本方法具有较强的学习能力及较高的识别精度,可有效地应用于道路监控等系统.     

Simulation and optimization of SOFC-BCHP system

As the prime motor of dispersed energy system, the high-temperature solid oxide fuel cells (SOFC) are high efficient with large heat recovery. This study presents a simulation of SOFC building-based cooling, heat and power (BCHP) system, which can meet basic requirements in power and heating (cooling) of the designated customers. The peak power load can be met by power grid, while the peak heating (cooling) load requirement can be met by backup equipments. In order to solve the economic dispatch problem of the energy system, a restricted nonlinear optimization model has been developed. The production costs can be minimized via both the equality constraints of customer’s heat and power demands, and other inequality constrains of equipments’ capacities. The sequential quadratic programming method has been used to search the solution. The study indicates that the model can be used to optimize the system’s capacities and run strategy. An office building case has been computed, and it is indicated that the model can be served in design and optimization of SOFC-BCHP system.     

网络环境下大学英语自主学习模式的研究

随着社会的飞速发展和要求,培养学生在网络环境下的英语自主学习能力成为大学英语教学的重点之一。在网络环境下,学生可以根据各自的需要,通过选择网络上的有效信息进行学习和交流,并在虚拟的交际环境中灵活、自主地运用语言,达到构建自主学习能力和提高学习效率的目的。     

海洋平台灰预测和支持向量机的逆控制

导管架式海洋平台在随机波浪等外载荷作用下极易产生有害振动,且其动力响应具有极强的非线性和时变性,采用被动控制方法和基于精确数学模型的主动控制方法控制海洋平台的有害振动很难达到理想的控制效果.结合灰预测和支持向量机的优点,提出了基于灰预测和支持向量机的自适应逆控制方法,并将其与动态刚度阵法相结合应用于导管架式海洋平台的振动主动控制中.数值算例分析表明,该控制方法可有效地控制波浪和风载荷作用下引起导管架式平台的有害振动,并能解决由于控制信号传输等原因引起的时滞问题.     

可信性空间上基于复模糊变量的学习理论的关键定理

给出了可信性测度空间上复模糊变量定义及一些性质;给出了复模糊变量的经验风险泛函、期望风险泛函以及经验风险最小化原则严格一致收敛的定义。在此基础上给出并证明了基于复模糊变量的学习理论的关键定理,为系统建立可信性空间上的复统计学习理论奠定了理论基础。     

网络课程设计中学习者学习动力的调查与建议

网络教育中有关学习者学习动力的研究日益受到重视,通过对现有网络课程设计中学习者学习动力设计状况的调查可发现其中存在的问题,分析这些问题与不足并针对性地提出改进建议,以期对网络课程的完善提供有益参考。     

窄带滤光片光学极值法直接监控的光学设计

窄带滤光片是应用于光谱学、激光、天文物理学、通信等各个领域的光学组件,精确控制每一层膜厚是制备滤光片的困难所在.采用光学极值法直接监控工件,可以同时制备多件滤光片.设计了监控光路和信号获得电路,通过探测到的测试信号,监控薄膜膜厚,实现了直接监控.     

Experimental study on temperature stress calculation and temperature control optimization of concrete based on early age parameters

Temperature control curve is the key to achieving temperature control and crack prevention of high concrete dam during construction,and its rationality depends on the accurate measurement of temperature stress.With the simulation testing machine for the temperature stress,in the present study,we carried out the deformation process tests of concrete under three temperature curves：convex,straight and concave.Besides,we not only measured the early-age elastic modulus,creep parameters and stress process,but also proposed the preferred type.The results show that at early age,higher temperature always leads to greater elastic modulus and smaller creep.However,the traditional indoor experiments have underestimated the elastic modulus and creep development at early age,which makes the calculated value of temperature stress too small,thus increasing the cracking risk.In this study,the stress values of the three curves calculated based on the strain and early-age parameters are in good agreement with the temperature stress measured by the temperature stress testing machine,which verifies the method accuracy.When the temperature changes along the concave curve,the law of stress development is in consistent with that of strength.Under this condition,the stress fluctuation is small and the crack prevention safety of the concave type is higher,so the concave type is better.The test results provide a reliable basis and support for temperature control curve design and optimization of concrete dams.     

基于迁移学习的卷积神经网络通道剪枝

卷积神经网络在计算机视觉等多个领域应用广泛,然而其模型参数量众多、计算开销庞大,导致许多边缘设备无法满足其存储与计算资源要求。针对其边缘部署困难,提出使用迁移学习策略改进基于BN层缩放因子通道剪枝方法的稀疏化过程。本文对比不同层级迁移方案对稀疏化效果与通道剪枝选取容限的影响;并基于网络结构搜索观点设计实验,探究其精度保持极限与迭代结构的收敛性。实验结果表明,对比原模型,采用迁移学习的通道剪枝算法,在精度损失不超过0.10的前提下,参数量减少89.1%,模型存储大小压缩89.3%;对比原剪枝方法,将剪枝阈值从0.85提升到0.97,进一步减少参数42.6%。实验证明,引入迁移策略更易实现充分的稀疏化,提高通道剪枝阈值选取容限,实现更高压缩率;并在迭代剪枝的网络结构搜索过程中,提供更高效的搜索起点,利于快速迭代趋近至搜索空间的一个网络结构局部最优解。