ISSN:0020-7179    

DOI:10.1080/00207179108934213

出版商:Taylor & Francis Group    

年代:1991

数据来源: Taylor

摘要:

The matrix sign function is the basis of a parallel algorithm for solving the generalized algebraic Riccati equation. Three forms of the algorithm were implemented and tested on a distributed memory hypercube multiprocessor. Performance results indicate that the method is an excellent means of solving large-scale problems on a parallel computer.

ISSN:0020-7179    

DOI:10.1080/00207179108934214

出版商:Taylor & Francis Group    

年代:1991

数据来源: Taylor

摘要:

A multidimensional binary partition (MBP) is a data structure determined by a set of points inn-dimensional space. On certain parallel architectures, this data structure can be easily distributed across the processing nodes of the machine and can provide a natural technique for load balancing and partitioning of application problems that depend on a distribution of dynamically changing points in multidimensional space. This paper describes parallel algorithms for generating and using MBPs on a hypercube parallel machine. It is also shown how these distributed data structures allow efficient parallel searches of the data set. The performance of an implementation of these algorithms on an NCUBE hypercube is presented.

ISSN:0020-7179    

DOI:10.1080/00207179108934215

出版商:Taylor & Francis Group    

年代:1991

数据来源: Taylor

摘要:

Fast self-tuning discrete-time control algorithms based on ‘recursive least squares’ parameter estimation and ‘generalized minimum-variance’ (GMV) control design are presented both in lattice and transversal forms. This derivation relies both on the linear prediction interpretation of GMV control and on the embedding of the pole-zero predictive model (for the system output) into a multichannel all-pole model for the joint system input-output process. The proposed multichannel algorithms are simplified via a modular decomposition principle which converts all matrix recursions into coupled scalar recursions. The resulting control algorithms require 0(n) computations per time instant, wherendenotes the controller order. Furthermore, we present parallel implementations of such fast (self-tuning controllers’ (STCs) in 0(n) processor arrays. Several structures, obtained for different implementations of the controller and of the identifier, are devised to this end but, as far as we are concerned with an ‘ideal’ STC (i.e. one which generates the control signal at each time instant with the best and most recent currently available parameter estimates), none of them can lead to a pipelined STC. We show that a pipelined STC can be obtained by implementing the controller in transversal form and introducing an 0(n) ‘lag’ in the transfer of parameters from the identifier to the controller. This amounts to using delayed parameter estimates in the control computation which, for slowly time-varying systems, could be a viable solution in order to increase the sampling rate.

ISSN:0020-7179    

DOI:10.1080/00207179108934216

出版商:Taylor & Francis Group    

年代:1991

数据来源: Taylor

摘要:

We have obtained a parallel algorithm for the Chandrasekhar filter for estimation in constant parameter state space models. The parallel Chandrasekhar filter can be expressed in terms of three distinct elementary operations which can be performed in parallel. The parallel Chandrasekhar filter is mapped systematically on rectangularly connected systolic arrays consisting of three types of simple processors each of which performs the elementary operations.

ISSN:0020-7179    

DOI:10.1080/00207179108934217

出版商:Taylor & Francis Group    

年代:1991

数据来源: Taylor

摘要:

This paper reports further work on the application of systolic architectures to systems where high speed real-time feedback operations are required. In particular, transputer operated array processors which exploit the IMS/A 100 parallel processing device are used to develop flexible (general purpose) systolic realizations of adaptive/self-tuning control schemes. An important feature of this approach is the fact that the identification (and hence the controller coefficient updating) and the control signal computation phases operate asynchronously. Hence, by running the latter at a higher rate than the former, it is possible to deal effectively with cases where the system dynamics undergo ‘rapid changes’. During each updating cycle the controller is, in effect, based on a fixed control law and multiple control signals are provided.

ISSN:0020-7179    

DOI:10.1080/00207179108934218

出版商:Taylor & Francis Group    

年代:1991

数据来源: Taylor

摘要:

Possible approaches for parallelism extraction in digital controller algorithms targeted at medium- to fine-grain parallel architectures are investigated. The methods use information on the node computation precedence of realization structures expressed in the factored state variable description (FSVD). Three architectures are considered, namely a multiprocessor system, a systolic array and PACE, a novel architecture which can support structured and non-structured algorithms within a regular processor array. It is shown that this facility makes PACE particularly promising for the implementation of controller structures expressed in terms of the FSVD.

ISSN:0020-7179    

DOI:10.1080/00207179108934219

出版商:Taylor & Francis Group    

年代:1991

数据来源: Taylor

摘要:

Neural networks can be used to solve highly nonlinear control problems. This paper shows how a neural network can learn of its own accord to control a non-linear dynamic system. An emulator, a multilayered neural network, learns to identify the system's dynamic characteristics. The controller, another multilayered neural network, next learns to control the emulator. The self-trained controller is then used to control the actual dynamic system. The learning process continues as the emulator and controller improve and track the physical process. An example is given to illustrate these ideas. The ‘truck backer-upper’, a neural network controller steering a trailer truck while backing up to a loading dock, is demonstrated. The controller is able to guide the truck to the dock from almost any initial position. The technique explored here should be applicable to a wide variety of non-linear control problems.

ISSN:0020-7179    

DOI:10.1080/00207179108934220

出版商:Taylor & Francis Group    

年代:1991

数据来源: Taylor

摘要:

Neural networks hold great promise for application in the general area of process control. This paper focuses on using a back propagation network in an optimization based model predictive control scheme. Since an analytical expression for the gradient of the neural net model can be derived and this expression can be calculated in parallel, extremely fast computation times are possible. The control approach is illustrated on a pH CSTR example.

ISSN:0020-7179    

DOI:10.1080/00207179108934221

出版商:Taylor & Francis Group    

年代:1991

数据来源: Taylor

摘要:

A ‘neural-like’ parallel system is described for the real-time implementation of the Hough transform. It is applied to simple object recognition in computer vision.

ISSN:0020-7179    

DOI:10.1080/00207179108934222

出版商:Taylor & Francis Group    

年代:1991

数据来源: Taylor