In this paper, a hybrid neural network approach to solve mixed integer quadratic bilevel programming problems is proposed. Bilevel programming problems arise when one optimization problem, the upper problem, is constrained by another optimization, the lower problem. The mixed integer quadratic bilevel programming problem is transformed into a double-layered neural network. The combination of a genetic algorithm (GA) and a meta-controlled Boltzmann machine (BM) enables us to formulate a hybrid neural network approach to solving bilevel programming problems. The GA is used to generate the feasible partial solutions of the upper level and to provide the parameters for the lower level. The meta-controlled BM is employed to cope with the lower level problem. The lower level solution is transmitted to the upper level. This procedure enables us to obtain the whole upper level solution. The iterative processes can converge on the complete solution of this problem to generate an optimal one. The proposed method leads the mixed integer quadratic bilevel programming problem to a global optimal solution. Finally, a numerical example is used to illustrate the application of the method in a power system environment, which shows that the algorithm is feasible and advantageous.
In this paper, we revisit the energy-based swing-up control problem for a rotational pendulum. Different from the existing energy-based control solution, first, we present a necessary and sufficient condition such that the control law has no singularities for all states of the rotational pendulum.Next, we carry out a global motion analysis of the pendulum, and we remove the previous required constraint on the initial state of the pendulum and the control parameters for preventing the pendulum getting stuck at the downward equilibrium point by revealing that the point is a saddle. Specifically, we show that the Jacobian matrix evaluated at the point has two and two eigenvalues in the open left- and right-half planes, respectively. We prove that the pendulum will eventually be swung up into the basin of attraction of any (locally) stabilizing controller for all initial conditions with the exception of a set of Lebesgue measure zero. Finally, we validate the presented theoretical results via numerical simulation. Our simulation results show that the swing-up control can be achieved quickly under the improved conditions on the control parameters.
This paper presents a new Hidden Markov Model (HMM) for the online signature verification of oriental characters such as Japanese and Chinese. These oriental characters usually consist of many individual strokes such as dots and straight lines. Taking into account of this characteristic, a new HMM is proposed, which is composed of many sub-models each of which corresponds to an individual stroke. In addition, the ‘pen-up’ state which represents the movement between strokes is explicitly introduced. Then, a parameter re-estimation scheme for this special class of HMM is derived exploiting the structure of the proposed HMM. Thanks to the structured learning mechanism, the proposed HMM not only can drastically reduce the computational time necessary for the learning process but also shows higher recognition performance for the rejection of the skilled forgery. Finally, the usefulness of the proposed scheme is demonstrated by comparing it with conventional models.
We propose a novel speaker recognition approach using a speaker-independent universal acoustic model (UAM) for sensornet applications. In sensornet applications such as “Business Microscope”, interactions among knowledge workers in an organization can be visualized by sensing face-to-face communication using wearable sensor nodes. In conventional studies, speakers are detected by comparing energy of input speech signals among the nodes. However, there are often synchronization errors among the nodes which degrade the speaker recognition performance. By focusing on property of the speaker's acoustic channel, UAM can provide robustness against the synchronization error. The overall speaker recognition accuracy is improved by combining UAM with the energy-based approach. For 0.1s speech inputs and 4 subjects, speaker recognition accuracy of 94% is achieved at the synchronization error less than 100ms.
Sampled-data consensus control for nonlinear multi-agent systems of strict-feedback form is considered. By using a change of state variables and an input transformation, the discrete-time double-integrator dynamics is derived from the Euler approximate model of each agent and discrete-time consensus control laws are designed. Then by applying the nonlinear sampled-data control theory, it is shown that the designed control laws achieve sampled-data consensus for nonlinear multi-agent systems in the continuous-time semiglobally practically uniformly asymptotically stable (SPUAS) sense. As an application of the proposed design method, sampled-data consensus control for fully-actuated ships is considered.
This paper is concerned with an approach for a nonlinear optimal control of polynomial systems. The Hamilton-Jacobi-Bellman (HJB) equation is relaxed into HJB inequalities. Both an upper bound and a lower bound on the cost function, as well as a suboptimal controller, can be computed from solutions of the resulting inequalities. Solving the HJB inequalities can be cast as state-dependent matrix inequalities (SDMIs), whose derivation is based on representation of the given polynomial system in a linear-like form. The resulting SDMI for the upper-bound computation is nonconvex in the decision variables, and hence an iterative procedure is proposed to deal with the non-convexity. On the other hand, the resulting SDMI for the lower-bound computation can be written as a state-dependent linear matrix inequality, which is a convex optimization problem solvable by existing numerical tools. Numerical examples are provided to illustrate the proposed approach.
A consensus control framework for configuration of two underactuated planar rigid bodies is developed. Specifically, we propose a series of control laws that achieve asymptotic consensus between the underactuated planar rigid bodies that possess small-time local controllability. The results are predicated on the characterizations of the approximate solution and the inversion algorithm for underactuated systems. Finally, we present a numerical example to show the utility of the proposed approach.
This paper presents a design method of an MIMO integral preceded by proportional-derivative (I-PD) controller based on an integral-type optimal servomechanism. The proposed method consists of two steps. First, a given plant is represented in a specific state-space form, and then an integral-type optimal servo controller is designed. Although the resultant controller does not always become a typical I-PD one, when the order of a given MIMO plant is equal to or less than twice the number of the outputs, the resultant control law is equivalent to an I-PD one. Moreover, the proposed I-PD controller design can be extended to a model-following type by adding a reference model and a feedforward compensator for a desirable output response. Controller design examples and numerical simulation studies are carried out in order to demonstrate that the proposed design method has sufficient effectiveness.
With the participation of 12 volunteers, the off-line application of independent component analysis for automatic artefacts removal based on power spectral density is investigated. By using the “range” values of the power spectra of the independent components within the frequency range of 2 to 8 Hz along with the integral values of the independent components in the range of 8 to 30 Hz, artificial independent components are automatically marked and removed. The artefact-free electroencephalographic signal is further classified using the method of common spatial pattern. It is found that the modification of the conventional common spatial pattern can result in a higher imagery task classification.
In this paper, we investigate a predictive control problem with information structured constraints motivated by control of micro grid. A system with information structures is defined as a system in which each subsystem collects spatio-temporally different information. For the system, we consider a predictive control law and the finite time constrained optimization problem to be solved online is reduced to a deterministic convex programming problem. Then, we reduce a control problem for a simple micro grid system to the framework and the effectiveness of the proposed control and estimation law is demonstrated through a numerical simulation.