Volume 30 (2020): Edizione 1 (March 2020)

Dynamical reconstruction of unknown time-varying controls from inexact measurements of the state function is investigated for a semilinear parabolic equation with memory. This system includes as particular cases the Schlögl model and the FitzHugh–Nagumo equations. A numerical method is suggested that is based on techniques of feedback control. An error analysis is performed. Numerical examples confirm the theoretical predictions.

An information based method for solving stochastic control problems with partial observation is proposed. First, information-theoretic lower bounds of the cost function are analysed. It is shown, under rather weak assumptions, that reduction in the expected cost with closed-loop control compared with the best open-loop strategy is upper bounded by a non-decreasing function of mutual information between control variables and the state trajectory. On the basis of this result, an information based control (IBC) method is developed. The main idea of IBC consists in replacing the original control task by a sequence of control problems that are relatively easy to solve and such that information about the system state is actively generated. Two examples of the IBC operation are given. It is shown that the method is able to find an optimal solution without using dynamic programming at least in these examples. Hence the computational complexity of IBC is substantially smaller than that of dynamic programming, which is the main advantage of the proposed method.

Classical model predictive control (MPC) algorithms need very long horizons when the controlled process has complex dynamics. In particular, the control horizon, which determines the number of decision variables optimised on-line at each sampling instant, is crucial since it significantly affects computational complexity. This work discusses a nonlinear MPC algorithm with on-line trajectory linearisation, which makes it possible to formulate a quadratic optimisation problem, as well as parameterisation using Laguerre functions, which reduces the number of decision variables. Simulation results of classical (not parameterised) MPC algorithms and some strategies with parameterisation are thoroughly compared. It is shown that for a benchmark system the MPC algorithm with on-line linearisation and parameterisation gives very good quality of control, comparable with that possible in classical MPC with long horizons and nonlinear optimisation.

A major goal in modern flight control systems is the need for improving reliability. This work presents a health-aware and fault-tolerant control approach for an octorotor UAV that allows distributing the control effort among the available actuators based on their health information. However, it is worth mentioning that, in the case of actuator fault occurrence, a reliability improvement can come into conflict with UAV controllability. Therefore, system reliability sensitivity is redefined and modified to prevent uncontrollable situations during the UAV’s mission. The priority given to each actuator is related to its importance in system reliability. Moreover, the proposed approach can reconfigure the controller to compensate actuator faults and improve the overall system reliability or delay maintenance tasks.

An active sensor fault tolerant controller for nonlinear systems represented by a decoupled multimodel is proposed. Active fault tolerant control requires accurate fault estimation. Thus, to estimate both state variables and sensor faults, a discrete unknown input multiobserver, based on an augmented state multimodel, is designed. The multiobserver gains are computed by solving linear matrix inequalities with equality constraints. A multicontrol strategy is proposed for the compensation of the sensor fault and recovering the desired performances. This strategy integrates a bank of controllers, corresponding to a set of partial models, to generate a set of control laws compensating the fault effect. Then, a switching strategy between the generated local control laws is established in order to apply the most suitable control law that tolerates the fault and maintains good closed loop performances. The effectiveness of the proposed strategy is proven through a numerical example and also through a real time application on a chemical reactor. The obtained results confirm satisfactory closed loop performance in terms of trajectory tracking and fault tolerance.

The paper concerns the properties of linear dynamical systems described by linear differential equations, excited by the Dirac delta function. A differential equation of the form a_nx(ⁿ)(t)+···+a₁x′ (t)+a₀x(t)= b_mu(^m)(t)+···+b₁u′ (t)+b₀u(t) is considered with a_i, b_j > 0. In the paper we assume that the polynomials M_n(s)= a_nsⁿ + ··· + a₁s + a₀ and L_m(s)= b_ms^m + ··· + b₁s + b₀ partly interlace. The solution of the above equation is denoted by x(t, L_m,M_n). It is proved that the function x(t, L_m,M_n) is nonnegative for t ∈ (0, ∞), and does not have more than one local extremum in the interval (0, ∞) (Theorems 1, 3 and 4). Besides, certain relationships are proved which occur between local extrema of the function x(t, L_m,M_n), depending on the degree of the polynomial M_n(s) or L_m(s) (Theorems 5 and 6).

A class of Clifford-valued high-order Hopfield neural networks (HHNNs) with state-dependent and leakage delays is considered. First, by using a continuation theorem of coincidence degree theory and the Wirtinger inequality, we obtain the existence of anti-periodic solutions of the networks considered. Then, by using the proof by contradiction, we obtain the global exponential stability of the anti-periodic solutions. Finally, two numerical examples are given to illustrate the feasibility of our results.

An insufficient number or lack of training samples is a bottleneck in traditional machine learning and object recognition. Recently, unsupervised domain adaptation has been proposed and then widely applied for cross-domain object recognition, which can utilize the labeled samples from a source domain to improve the classification performance in a target domain where no labeled sample is available. The two domains have the same feature and label spaces but different distributions. Most existing approaches aim to learn new representations of samples in source and target domains by reducing the distribution discrepancy between domains while maximizing the covariance of all samples. However, they ignore subspace discrimination, which is essential for classification. Recently, some approaches have incorporated discriminative information of source samples, but the learned space tends to be overfitted on these samples, because they do not consider the structure information of target samples. Therefore, we propose a feature reduction approach to learn robust transfer features for reducing the distribution discrepancy between domains and preserving discriminative information of the source domain and the local structure of the target domain. Experimental results on several well-known cross-domain datasets show that the proposed method outperforms state-of-the-art techniques in most cases.

Motivated by ideas from two-step models and combining second-order TV regularization in the LLT model, we propose a coupling model for MR image reconstruction. By applying the variables splitting technique, the split Bregman iterative scheme, and the alternating minimization method twice, we can divide the proposed model into several subproblems only related to second-order PDEs so as to avoid solving a fourth-order PDE. The solution of every subproblem is based on generalized shrinkage formulas, the shrink operator or the diagonalization technique of the Fourier transform, and hence can be obtained very easily. By means of the Barzilai–Borwein step size selection scheme, an ADMM type algorithm is proposed to solve the equations underlying the proposed model. The results of numerical implementation demonstrate the feasibility and effectiveness of the proposed model and algorithm.

We propose a skeletonization algorithm that is based on an iterative points contraction. We make an observation that the local center that is obtained via optimizing the sum of the distance to k nearest neighbors possesses good properties of robustness to noise and incomplete data. Based on such an observation, we devise a skeletonization algorithm that mainly consists of two stages: points contraction and skeleton nodes connection. Extensive experiments show that our method can work on raw scans of real-world objects and exhibits better robustness than the previous results in terms of extracting topology-preserving curve skeletons.

Data clustering is one of the most popular methods of data mining and cluster analysis. The goal of clustering algorithms is to partition a data set into a specific number of clusters for compressing or summarizing original values. There are a variety of clustering algorithms available in the related literature. However, the research on the clustering of data parametrized by unit quaternions, which are commonly used to represent 3D rotations, is limited. In this paper we present a quaternion clustering methodology including an algorithm proposal for quaternion based k-means along with quaternion clustering quality measures provided by an enhancement of known indices and an automated procedure of optimal cluster number selection. The validity of the proposed framework has been tested in experiments performed on generated and real data, including human gait sequences recorded using a motion capture technique.

In this work a new algorithm for quaternion-based spatial rotation is presented which reduces the number of underlying real multiplications. The performing of a quaternion-based rotation using a rotation matrix takes 15 ordinary multiplications, 6 trivial multiplications by 2 (left-shifts), 21 additions, and 4 squarings of real numbers, while the proposed algorithm can compute the same result in only 14 real multiplications (or multipliers—in a hardware implementation case), 43 additions, 4 right-shifts (multiplications by 1/4), and 3 left-shifts (multiplications by 2).

The biclustering of two-dimensional homogeneous data consists in finding a subset of rows and a subset of columns whose intersection provides a set of cells whose values fulfil a specified condition. Usually it is defined as equality or comparability. One of the presented approaches is based on the model of Boolean reasoning, in which finding biclusters in binary or discrete data comes down to the problem of finding prime implicants of some Boolean function. Due to the high computational complexity of this task, the application of some heuristics should be considered. In the paper, a modification of the well-known Johnson strategy for prime implicant approximation induction is presented, which is necessary for the biclustering problem. The new method is applied to artificial and biomedical datasets.

Given an undirected connected graph G = (V, E), a subset of vertices S is a maximum 2-packing set if the number of edges in the shortest path between any pair of vertices in S is at least 3 and S has the maximum cardinality. In this paper, we present a genetic algorithm for the maximum 2-packing set problem on arbitrary graphs, which is an NP-hard problem. To the best of our knowledge, this work is a pioneering effort to tackle this problem for arbitrary graphs. For comparison, we extended and outperformed a well-known genetic algorithm originally designed for the maximum independent set problem. We also compared our genetic algorithm with a polynomial-time one for the maximum 2-packing set problem on cactus graphs. Empirical results show that our genetic algorithm is capable of finding 2-packing sets with a cardinality relatively close (or equal) to that of the maximum 2-packing sets. Moreover, the cardinality of the 2-packing sets found by our genetic algorithm increases linearly with the number of vertices and with a larger population and a larger number of generations. Furthermore, we provide a theoretical proof demonstrating that our genetic algorithm increases the fitness for each candidate solution when certain conditions are met.

The classic interval has precise borders A=[a_,a¯]A = \left[ {\underline{a},\bar a} \right]. Therefore, it can be called a type 1 interval. Because of great practical importance of such interval data, several versions of type 1 interval arithmetic have been created. However, sometimes precise borders a_\underline{a} and ā of intervals cannot be determined in practice. If the borders are uncertain, then we have to do with type 2 intervals. A type 2 interval can be denoted as AT2=[[a_L,a_R],[a¯L,a¯R]]{A_{T2}} = \left[ {\left[ {{\underline{a}_L},{\underline{a}_R}} \right],\left[ {{{\bar a}_L},{{\bar a}_R}} \right]} \right]. The paper presents multidimensional decomposition RDM type 2 interval arithmetic (D-RDM-T2-I arithmetic), where RDM means relative-distance measure. The decomposition approach considerably simplifies calculations and is transparent for users. Apart from this arithmetic, examples of its applications are also presented. To the authors’ best knowledge, no papers on this arithmetic exist. D-RDM-T2-I arithmetic is necessary to create type 2 fuzzy arithmetic based on horizontal μ-cuts, which the authors aim to do.