混响噪声下声源定位与跟踪的多特征自适应IMM粒子滤波算法EI北大核心CSCD

针对混响噪声下声源定位精度低和鲁棒性弱等问题,提出了多特征自适应IMM粒子滤波算法.该算法以麦克风接收信号的多特征作为观测信息,采用空时相关和迭代滤波建立了时延选择机制和波束输出能量优化机制,并在两者的基础上构建了似然函数以获得合理的声源位置信息.考虑到说话人运动的随机性,给出了自适应IMM算法,通过在线粒子集生成并将不同过程方差的模型进行交互来拟合说话人的不同运动模式,改善了说话人跟踪系统的稳健性.仿真和实测结果表明,所提算法利用了多特征定位信息的互补性,降低了观测误差不确定性对声源位置估计的影响,增强了随机运动声源跟踪系统的鲁棒性,提高了系统的定位精度.     

永磁同步电机模型预测转矩控制简化控制策略

传统永磁同步电机(PMSM)模型预测转矩控制(MPTC)遍历逆变器生成的全部7个电压矢量, 计算负担较大.当转矩误差较小时, 零电压矢量利用率较高, 则可当转矩误差位于阈值范围, 电机系统直接输出零电压矢量, 否则,依然遍历7个电压矢量, 并给出阈值确定方法. 基于上述策略, 本文增加了6个定子磁链扇区位置约束, 将转矩误差大于阈值时的备选电压矢量降至4个, 并增加磁链扇区数目至12个和磁链误差约束, 进一步减小备选电压矢量. 仿真结果表明, 提出的3种简化策略控制下, 永磁同步电机系统运行正常, 控制性能与传统模型预测转矩控制基本相当,平均开关频率分别降低至77.48%, 77.09%和76.12%, 平均遍历电压矢量个数分别降低至58.29%, 32.86%和29.14%.实时性实验结果表明运行时间分别减小至57.70%, 32.96%和29.48%.     

Combustion timing control of HCCI engine based on NNPC and Elman-NN model under complex conditions

Homogeneous Charge Compression Ignition (HCCI) combines the characteristics of gasoline engine and diesel engine with high thermal efficiency and low emissions. However, since there is no direct initiator of combustion, it is difficult to control the combustion timing in HCCI engines under complex working conditions. In this paper, Neural Network Predictive Control (NNPC) for combustion timing of the HCCI engine is designed and implemented. First, the black box model based on Elman neural network is designed and developed to estimate the combustion timing. The fuel equivalence ratio, intake valve closing timing, intake manifold temperature, intake manifold gas pressure, and engine speed are chosen as the system inputs. Then, a NNPC controller is designed to control combustion timing by controlling the intake valve closing timing. Simulation results show that the Elman neural network black box model is capable of estimating the HCCI engine combustion timing. In addition, regardless of whether the HCCI engine is in constant or complex condition, the designed NNPC controller is capable of keeping the combustion timing within the ideal range. In particular, under New European Driving Cycle (NEDC) working conditions, the maximum overshoot of the controller is 28.95% and the average error is 1.03 crank angle degree. It is concluded that the controller has good adaptability and robustness.     

Quasi-min-max MPC for visual servoing stabilization of omnidirectional wheeled mobile robots

This paper proposes a new visual servoing quasi-min-max MPC algorithm for stabilization control of an omnidirectional wheeled mobile robot subject to physical and visual constraints. The visual servoing dynamics of the robot are modeled as the state-dependent linear error system with nonlinear control inputs of rotation and deflection velocities of wheels. The state-dependent linear error system is covered as linear parameters-varying models which is used to design the visual servoing quasi-min-max MPC controller. The actual control inputs of the robot are then computed by the solution of an inverse algebraic equation of the MPC actions. The recursive feasibility and stability of the new visual servoing MPC are ensured by some LMIs conditions. The performance and practicability of the visual servoing MPC are verified by some simulation and experiment results.     

Sliding mode learning control for T-S fuzzy system and an application to hypersonic flight vehicle

Sliding mode-based learning control is presented for T-S fuzzy system. A T-S fuzzy model with both uncertainties and unmodeled dynamics is proposed firstly, in which the information of uncertainties and unmodeled dynamics are assumed to be unknown. Then, according to a given reference model, state-tracking error system is built. Respecting facts, the input matrices of the built T-S fuzzy model are different from each other. An extended state observer is built for estimating the unknown uncertainties and unmodeled dynamics, and a corresponding sliding surface is proposed. A learning controller is then presented for the closed loop system. Moreover, a numerical simulation result on hypersonic flight vehicles is considered to testify the controller's availability.     

Model-based periodic event-triggered stabilization for continuous-time stochastic nonlinear systems

In this paper, we investigate a model-based periodic event-triggered control framework for continuous-time stochastic nonlinear systems. In this framework, an auxiliary approximate discrete-time model of stochastic nonlinear systems is constructed in the controller module, which is utilized not only to design a discrete-time controller but also as a state predictor within trigger intervals. This discrete controller design approach, the strategy of state prediction, and the periodic detection strategy for the trigger rule not only provide a manner of more direct and easier implementation on the digital platform but also effectively reduce the communication load while a satisfactory control performance is maintained. Additionally, the mean-square exponentially stabilization for continuous-time stochastic nonlinear systems is achieved, in which a guideline for determining the maximum admissible sampling period is provided and the periodic event trigger rule is designed. The final numerical simulation also supports the effectiveness of our proposed framework.     

Wavelet CMAC for MIMO uncertain nonlinear systems using a modified social ski-driver algorithm

This study uses a Mexican hat wavelet membership function for a cerebellar model articulation controller (CMAC) to develop a more efficient adaptive controller for multiple input multiple output (MIMO) uncertain nonlinear systems. The main controller is called the adaptive Mexican hat wavelet CMAC (MWCMAC), and an auxiliary controller is used to remove the residual error. For the MWCMAC, the online learning laws are derived from the gradient descent method. In addition, the learning rate values are very important and have a great impact on the performance of the control system; however, they are difficult to choose accurately. Therefore, a modified social ski driver (SSD) algorithm is proposed to find optimal learning rates for the control parameters. Finally, a magnetic ball levitation system and a nine-link biped robot are used to illustrate the effectiveness of the proposed SSD-based MWCMAC control system. The comparisons with other existing control algorithms have shown the superiority of the proposed control system.     

研究型大学本硕博贯穿式培养探索与实践

为顺应高等教育对高素质人才培养的迫切需求,从学生的核心诉求出发,探索和实践了本科生-硕士生-博士生贯穿式培养模式,提出了卓有成效的具体措施,包括改进现有培养方案、将人才培养与高水平科学研究相融合、建设人才梯队,有效提高了本科阶段与研究生阶段培养的连贯性,初步形成了本硕博贯穿式人才培养体系,为高等学校人才培养提供借鉴和参考。     

BP神经网络结合变量选择方法在牛奶蛋白质含量检测中的应用北大核心CSCD

牛奶中的蛋白质含量会影响牛奶的品质,利用高光谱图像的光谱特征信息研究对牛奶蛋白质含量预测的可行性。本文提出一种基于竞争性自适应重加权算法(competitive adaptive reweighted sampling, CARS)和连续投影算法(successive projections algorithm, SPA)结合多层前馈神经网络(back propagation, BP)的预测建模方法,实验以含有不同浓度蛋白质的牛奶为对象,利用可见光/近红外高光谱成像系统共采集到5种牛奶共计250组高光谱数据,通过实验对比选择采用标准化方法对获取到的吸收光谱预处理,然后采用CARS结合SPA筛选特征波长,得到18个特征波长,建立CARS-SPA-BP模型,经过试验,CARS-SPA-BP模型的训练集决定系数和测试集决定系数R;和R;分别达到0.971和0.968,训练集均方根误差(root mean square error of calibration,RMSEC)和测试集均方根误差(root mean square error of prediction,RMSEP)达到了0.033和0.034。研究发现,采用CARS结合SPA筛选的牛奶特征波长建立的多层前馈神经网络模型,其模型预测结果与全波长建模相比并没有明显降低,因此将CARS结合SPA用于波长筛选并且结合BP神经网络基本可以完成对牛奶蛋白质含量的预测。为验证CARS-SPA-BP模型的预测能力,在相同数据环境下,使用较为传统的偏最小二乘回归(partial least squares regression, PLSR)进行建模,实验结果表明,CARS-SPA-BP相较于PLSR,R;和RMSEP均有明显提升。研究表明,CARS-SPA-BP可充分利用牛奶光谱特征信息实现较高精度的牛奶蛋白质含量检测。     

基于自注意力对抗的深度子空间聚类

子空间聚类(Subspace clustering)是一种当前较为流行的基于谱聚类的高维数据聚类框架.近年来,由于深度神经网络能够有效地挖掘出数据深层特征,其研究倍受各国学者的关注.深度子空间聚类旨在通过深度网络学习原始数据的低维特征表示,计算出数据集的相似度矩阵,然后利用谱聚类获得数据的最终聚类结果.然而,现实数据存在维度过高、数据结构复杂等问题,如何获得更鲁棒的数据表示,改善聚类性能,仍是一个挑战.因此,本文提出基于自注意力对抗的深度子空间聚类算法(SAADSC).利用自注意力对抗网络在自动编码器的特征学习中施加一个先验分布约束,引导所学习的特征表示更具有鲁棒性,从而提高聚类精度.通过在多个数据集上的实验,结果表明本文算法在精确率(ACC)、标准互信息(NMI)等指标上都优于目前最好的方法.