非平稳信号的递推最小二乘盲分离

针对非平稳信号盲分离问题提出了一种基于递推最小二乘(RLS)算法的非平稳信号盲分离新方法.首先引入遗忘因子对常规代价函数进行指数加权修正,得到一种新的具有递归结构的代价函数;然后利用RLS算法最小化代价函数,推导最优分离矩阵的自适应更新算法,逐步实现信号分离.该算法避免了最小二乘类算法关于学习速率选择困难的缺点,具有收...     

非负矩阵分解的分层最小二乘快速算法研究

非负矩阵分解是对于代价函数近似非线性优化问题,考虑均方误差值作为代价函数,通过对分层交替非负最小二乘算法的迭代运算量进行分析,对运算耗费大的矩阵运算提出利用限制更新的方法对分层交替非负最小二乘算法进行修改,达到加速收敛的目的。通过仿真,与原倍乘更新算法、投射梯度算法比较,验证算法的有效性和稳定性和高效性。     

基于PSO的反对称矩阵反问题的最小二乘解

针对在反对称矩阵反问题的最小二乘解求解计算中,难以从问题的原始形式出发,构造出高效的迭代算法的计算难题,提出一种基于PSO算法的反对称矩阵反问题的最小二乘解的计算算法.该算法采用以带约束条件的反问题矩阵范数作为粒子群优化算法的适应度函数,建立起最小二乘解的计算模型.算例仿真结果显示,该算法是一种高效实用的求解算法.     

基于稳健联合分块对角化的卷积盲分离

针对卷积盲分离问题,提出一种新的矩阵联合分块对角化(Joint block diagonalization, JBD)算法. 现有的迭代非正交联合分块对角化算法都存在不收敛的情况,本文利用分离矩阵的特殊结构确保其可逆性,使得算法的迭代过程稳定. 在已知矩阵分块结构的条件下,首先,将卷积盲分离模型写成瞬时形式,并说明其满足联合分块对角化结构; 然后,提出联合分块对角化的代价函数,依据代价函数的最小化等价于矩阵中每个分块的范数最小化, 将整个分离矩阵的迭代更新转化成每个分块的迭代更新;最后,利用最小化条件得到迭代算法. 实数和复数两种情况下的算法都进行了推导.基本实验验证了新算法在不同条件下的性能; 仿真实验中对在时域和频域都重叠的信号的卷积混合进行盲分离,实验结果验证了新算法具有更好的分离性能和更稳定的分离能力.     

非正交联合对角化的罚函数粒子群算法

针对多个矩阵近似联合对角化盲分离问题,提出一种新的非正交近似联合对角化算法.首先采用罚函数法将联合对角化的非线性约束优化模型转化为无约束优化模型;其次将粒子群优化算法引入无约束优化模型中实现目标函数的最优化,从而完成矩阵组的联合对角化.分析了惩罚因子的更新策略及算法的收敛性能,并设计仿真实验进行对比分析以检验算法解决实际盲分离问题的能力.     

随机布署平面传感器阵列实现色噪声下运动节点跟踪算法

针对色噪声下的空间运动节点跟踪问题,提出一种新的辅助变量子空间跟踪算法.该算法采用秩1更新,构造两个无约束代价函数,求其递归最小二乘解跟踪信号子空间.得到信号子空间后,构造矩阵方程,通过最小二乘解实现空间运动节点的方向跟踪,所得空间二维角度能自动配对.在任意的平面传感器阵列下,所提算法和著名EIV-PAST算法均具有较好的节点跟踪性能,跟踪的信号子空间夹角和信号子空间误差较小,所提算法的正交性能显著优于EIV-PAST算法.     

线性矩阵方程异类约束最小二乘解的迭代算法

多矩阵变量线性矩阵方程(LME)约束解的计算问题在参数识别、结构设计、振动理论、自动控制理论等领域都有广泛应用。本文借鉴求线性矩阵方程(LME)同类约束最小二乘解的迭代算法,通过构造等价的线性矩阵方程组,建立了求多矩阵变量LME的一种异类约束最小二乘解的迭代算法,并证明了该算法的收敛性。在不考虑舍入误差的情况下,利用该算法不仅可在有限步计算后得到LME的一组异类约束最小二乘解,而且选取特殊初始矩阵时,可求得LME的极小范数异类约束最小二乘解。另外,还可求得指定矩阵在该LME的异类约束最小二乘解集合中的最佳逼近解。算例表明,该算法是有效的。     

非持续激励条件下系统辨识递推最小二乘最小范数算法

系统辨识中广泛应用的最小二乘算法需要输入向量序列满足持续激励性条件(PE条件); 但在大多情况下这是难以满足的. 本文提出了一种不依赖于PE条件的递推最小二乘、最小范数辨识算法. 首先分析了最小二乘算法解空间的结构, 并运用罚函数方法, 将参数辨识问题转化为无约束优化问题. 然后, 提出了将步长、罚因子等过程控制参数统一的迭代－递推形式的辨识算法, 证明了算法在给定的控制参数约束下收敛于唯一的最小二乘、最小范数解向量. 仿真实验表明在非PE条件下算法的有效性.     

递归最小二乘循环神经网络

针对循环神经网络(Recurrent neural networks, RNNs)一阶优化算法学习效率不高和二阶优化算法时空开销过大, 提出一种新的迷你批递归最小二乘优化算法. 所提算法采用非激活线性输出误差替代传统的激活输出误差反向传播, 并结合加权线性最小二乘目标函数关于隐藏层线性输出的等效梯度, 逐层导出RNNs参数的迷你批递归最小二乘解. 相较随机梯度下降算法, 所提算法只在RNNs的隐藏层和输出层分别增加了一个协方差矩阵, 其时间复杂度和空间复杂度仅为随机梯度下降算法的3倍左右. 此外, 本文还就所提算法的遗忘因子自适应问题和过拟合问题分别给出一种解决办法. 仿真结果表明, 无论是对序列数据的分类问题还是预测问题, 所提算法的收敛速度要优于现有主流一阶优化算法, 而且在超参数的设置上具有较好的鲁棒性.     

主子阵约束下矩阵方程AX=B的对称最小二乘解

本文主要讨论主子阵约束下矩阵方程AX=B的对称最小二乘解．基于投影定理,巧妙的把最小二乘问题转化为等式问题求解,并利用奇异值分解的方法,给出了该对称最小二乘解的一般表达式．此外,文章还考虑了此对称最小二乘解集合对任一给定矩阵的最佳逼近问题,得到了最佳逼近解,并给出了相应的算法步骤和数值例子．     

Fast nonunitary joint block diagonalization with degenerate solution elimination for convolutive blind source separation

This paper addresses the problem of joint block diagonalization (JBD) of a set of given matrices. As is known that the nonunitary JBD algorithm has some advantages over the existing orthogonal one for convolutive blind source separation (CBSS). However, the nonunitary JBD algorithm is prone to converge to some unexpected degenerate solutions (singular or ill-conditioned solutions). Especially for the matrices of large dimension or the case that the number of the diagonal blocks is relatively large, the performances of the nonunitary JBD algorithm degrade more severely. To eliminate the degenerate solutions, we optimize a penalty term based weighted least-squares criterion and thus develop a fast efficient algorithm. The performance of the proposed algorithm is evaluated by computer simulations and compared with the existing state-of-the-art nonunitary JBD algorithm. The simulation results demonstrate the robustness and performance improvement of the proposed algorithm.     

具有非奇异约束的线性卷积混合信号盲分离联合对角化方法

联合对角化能够成功解决盲分离问题, 但在求解时会得到非期望的奇异解, 从而无法完全分离出源信号. 鉴于此, 提出一种用于线性卷积混合盲分离的联合对角化方法, 将卷积混合模型变换为瞬时模型, 并对变换后的模型应用联合对角化求取分离矩阵. 在求解过程中, 引入约束条件对解的范围进行限定, 避免了奇异解的出现. 仿真结果表明, 所提出的方法能够成功实现卷积混合信号盲分离.

     

Nonorthogonal Approximate Joint Diagonalization With Well-Conditioned Diagonalizers

To make the results reasonable, existing joint diagonalization algorithms have imposed a variety of constraints on diagonalizers. Actually, those constraints can be imposed uniformly by minimizing the condition number of diagonalizers. Motivated by this, the approximate joint diagonalization problem is reviewed as a multiobjective optimization problem for the first time. Based on this, a new algorithm for nonorthogonal joint diagonalization is developed. The new algorithm yields diagonalizers which not only minimize the diagonalization error but also have as small condition numbers as possible. Meanwhile, degenerate solutions are avoided strictly. Besides, the new algorithm imposes few restrictions on the target set of matrices to be diagonalized, which makes it widely applicable. Primary results on convergence are presented and we also show that, for exactly jointly diagonalizable sets, no local minima exist and the solutions are unique under mild conditions. Extensive numerical simulations illustrate the performance of the algorithm and provide comparison with other leading diagonalization methods. The practical use of our algorithm is shown for blind source separation (BSS) problems, especially when ill-conditioned mixing matrices are involved.      

Computation of system balancing transformations and other applications of simultaneous diagonalization algorithms

An algorithm is presented in this paper for computing state-space balancing transformations directly from a state-space realization. The algorithm requires no "squaring up" or unnecessary matrix products. Various algorithmic aspects are discussed in detail. A key feature of the algorithm is the determination of a contragredient transformation through computing the singular value decomposition of a certain product of matrices without explicitly forming the product. Other contragredient transformation applications are also described. It is further shown that a similar approach may be taken, involving the generalized singular value decomposition, to the classical simultaneous diagonalization problem. These SVD-based simultaneous diagonalization algorithms provide a computational alternative to existing methods for solving certain classes of symmetric positive definite generalized eigenvalue problems.     

两正定矩阵联合对角化盲分离算法

针对具有时间结构的盲分离问题,提出了一种基于两正定矩阵精确联合对角化的盲分离算法。利用多个不同时延统计量构造了两个正定矩阵,以提取出数据的时间结构;再利用所提算法联合对角化构造的两个正定矩阵,得到分离矩阵,进而估计出源信号。所提算法克服了已有算法因采用多个矩阵联合对角化导致的计算量大和采用单个矩阵导致的分离精度低的缺点。计算机仿真结果表明了在有或无噪声情况下,所提算法性能均优于其他对比算法。     

A modified trust region algorithm for nonlinear equations with new updating rule of trust region radius

In this article, we present a trust region algorithm for the nonlinear equations with a new updating rule of the trust region radius, which takes some function of the residual. We show that under the local error bound condition which is weaker than the non-singularity, the new algorithm converges quadratically to some solution of the nonlinear equations. Numerical results show that the new algorithm performs very well for some singular nonlinear equations.     

Consensus in Noncooperative Dynamic Games: A Multiretailer Inventory Application

We focus on Nash equilibria and Pareto optimal Nash equilibria for a finite horizon noncooperative dynamic game with a special structure of the stage cost. We study the existence of these solutions by proving that the game is a potential game. For the single-stage version of the game, we characterize the aforementioned solutions and derive a consensus protocol that makes the players converge to the unique Pareto optimal Nash equilibrium. Such an equilibrium guarantees the interests of the players and is also social optimal in the set of Nash equilibria. For the multistage version of the game, we present an algorithm that converges to Nash equilibria, unfortunately, not necessarily Pareto optimal. The algorithm returns a sequence of joint decisions, each one obtained from the previous one by an unilateral improvement on the part of a single player. We also specialize the game to a multiretailer inventory system.     

Prox-regularization of the dual method of centers for generalized fractional programs

We present in this paper a prox-dual regularization algorithm for solving generalized fractional programming problems. The algorithm combines the dual method of centres for generalized fractional programs and the proximal point algorithm and can handle nondifferentiable convex problems with possibly unbounded feasible constraints set. The proposed procedure generates two sequences of dual and primal values that approximate the optimal value of the considered problem respectively from below and from above at each step. It also generates a sequence of dual solutions that converges to a solution of the dual problem, and a sequence of primal solutions whose every accumulation point is a solution of the primal problem. For a class of problems, including linear fractional programs, the algorithm converges linearly.     

Identification of Hammerstein nonlinear ARMAX systems

Two identification algorithms, an iterative least-squares and a recursive least-squares, are developed for Hammerstein nonlinear systems with memoryless nonlinear blocks and linear dynamical blocks described by ARMAX/CARMA models. The basic idea is to replace unmeasurable noise terms in the information vectors by their estimates, and to compute the noise estimates based on the obtained parameter estimates. Convergence properties of the recursive algorithm in the stochastic framework show that the parameter estimation error consistently converges to zero under the generalized persistent excitation condition. The simulation results validate the algorithms proposed.