A signal processing framework based on dynamic neural networks with application to problems in adaptation, filtering, and classification

Feldkamp, L.A.; Puskorius, G.V.

doi:10.1109/5.726790

Cited by 108 publications

(58 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The results of our experiments show that the use of the stabilization matrix has correctly addressed this significant issue, and the concept has been used to attack the long-standing problem of efficiently training RNNs to solve the tracking problem. Our approach contrasts to conventional RNN research which tends to focus on the complementary problem known as "vanishing" gradients (Hochreiter and Schmidhuber, 1997), or superior optimizers (Feldkamp and Puskorius, 1998). We have described how the method we proposed can be used in conjunction with these more powerful optimizers.…”

Section: Discussionmentioning

confidence: 99%

“…Of course more sophisticated learning optimizers might be able to navigate a crinkly surface better than RPROP did (such as the multi-stream extended Kalman filter algorithm (Feldkamp and Puskorius, 1998), or conjugate gradient descent), but it would be expected that the stabilization-matrix method would assist these second-order algorithms to achieve better RNNs than they otherwise would.…”

Section: Using the Stabilization Matrixmentioning

confidence: 99%

See 1 more Smart Citation

An adaptive recurrent neural-network controller using a stabilization matrix and predictive inputs to solve a tracking problem under disturbances

Fairbank¹,

Li²,

Fu³

et al. 2014

Neural Networks

View full text Add to dashboard Cite

This is the accepted version of the paper.This version of the publication may differ from the final published version. Permanent repository link AbstractWe present a recurrent neural-network (RNN) controller designed to solve the tracking problem for control systems. We demonstrate that a major difficulty in training any RNN is the problem of exploding gradients, and we propose a solution to this in the case of tracking problems, by introducing a stabilization matrix and by using carefully constrained context units. This solution allows us to achieve consistently lower training errors, and hence allows us to more easily introduce adaptive capabilities. The resulting RNN is one that has been trained off-line to be rapidly adaptive to changing plant conditions and changing tracking targets.The case study we use is a renewable-energy generator application; that of producing an efficient controller for a three-phase grid-connected converter. The controller we produce can cope with random variation of system parameters and fluctuating grid voltages. It produces tracking control with almost instantaneous response to changing reference states, and virtually zero oscil-$ This work was supported in part by the U.S. National Science Foundation under Grant EECS 1059265/1102159, the Mary K. Finley Missouri Endowment, and the Missouri S&T Intelligent Systems Center.Email addresses: michael.fairbank@virgin.net (Michael Fairbank), sli@eng.ua.edu (Shuhui Li), xfu@crimson.ua.edu (Xingang Fu), E.Alonso@city.ac.uk (Eduardo Alonso), dwunsch@mst.edu (Donald Wunsch) Preprint submitted to Neural Networks September 23, 2013 lation. This compares very favorably to the classical proportional integrator (PI) controllers, which we show produce a much slower response and settling time. In addition, the RNN we propose exhibits better learning stability and convergence properties, and can exhibit faster adaptation, than is achievable with adaptive critic designs.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Using the Stabilization Matrixmentioning

confidence: 99%

An adaptive recurrent neural-network controller using a stabilization matrix and predictive inputs to solve a tracking problem under disturbances

Fairbank¹,

Li²,

Fu³

et al. 2014

Neural Networks

View full text Add to dashboard Cite

show abstract

“…The operations described by (13) and (14) can be carried out using simple linear perceptrons, connected recurrently in a network structure shown in Fig.1. Fig.1.…”

Section: Neural Network Technique For the Real-time Dynamic Error Cormentioning

confidence: 99%

“…Classical dynamic error correction algorithms are usually characterized by high complexity of numerical operations, in particular in the case of describing the transducer dynamics by means of higher order differential equations. ANN as "universal approximators" [12,13,14] have been widely used for transducer static error correction [15,16,17,18], in particular for transducer and measuring instrument calibration [19,20,21]. Nevertheless, in the field of realtime dynamic error correction, solutions using DSP [22,23,24], FPGA technique [25] and analog circuits [26,27] are dominant.…”

Section: Introductionmentioning

confidence: 99%

Neural Network Based Real-time Correction of Transducer Dynamic Errors

Roj¹

2013

Measurement Science Review

View full text Add to dashboard Cite

In order to carry out real-time dynamic error correction of transducers described by a linear differential equation, a novel recurrent neural network was developed. The network structure is based on solving this equation with respect to the input quantity when using the state variables. It is shown that such a real-time correction can be carried out using simple linear perceptrons. Due to the use of a neural technique, knowledge of the dynamic parameters of the transducer is not necessary. Theoretical considerations are illustrated by the results of simulation studies performed for the modeled second order transducer. The most important properties of the neural dynamic error correction, when emphasizing the fundamental advantages and disadvantages, are discussed.

show abstract

“…[3][4][5] Neural networks have also been widely applied to model other strongly nonlinear, complex industrial processes containing a large number of process variables. Applications range from modeling effluent concentration of wastewater treatment plants 6 and automotive exhaust catalyst operation 7 to modeling nitro-cellulose production 6 and polymer production 8 in chemical plants. Neural networks are prime candidates for modeling complex dynamical processes due to their ability to approximate large classes of nonlinear functions with sufficient accuracy.…”

Section: Introductionmentioning

confidence: 99%

An Intelligent Emissions Controller for Fuel Lean Gas Reburn in Coal-Fired Power Plants

Reifman

Feldman

Wei

et al. 2000

Journal of the Air & Waste Management Association

View full text Add to dashboard Cite

The application of artificial intelligence techniques for performance optimization of the fuel lean gas reburn (FLGR) system is investigated. A multilayer, feedforward artificial neural network is applied to model static nonlinear relationships between the distribution of injected natural gas into the upper region of the furnace of a coalfired boiler and the corresponding oxides of nitrogen (NO x ) emissions exiting the furnace. Based on this model, optimal distributions of injected gas are determined such that the largest NO x reduction is achieved for each value of total injected gas. This optimization is accomplished through the development of a new optimization method based on neural networks. This new optimal control algorithm, which can be used as an alternative generic tool for solving multidimensional nonlinear constrained optimization problems, is described and its results are successfully validated against an off-the-shelf tool for solving mathematical programming problems. Encouraging results obtained using plant data from one of Commonwealth Edison's coal-fired electric power plants demonstrate the feasibility of the overall approach.Preliminary results show that the use of this intelligent controller will also enable the determination of the IMPLICATIONS This article describes the application of artificial neural networks for optimizing the performance of a new gas reburn technology, the FLGR, for reducing NO x emissions from coal-fired boilers. Proof-of-concept results obtained with demonstration test data indicate the feasibility of the approach and its use to optimize the performance of other NO x control technologies, involving the injection of chemicals above the boiler primary combustion zone. Future research should consider a more comprehensive database that allows for modeling CO. This application of neural networks should prove valuable for those involved in the development of advanced NO x control technologies. most cost-effective operating conditions of the FLGR system by considering, along with the optimal distribution of the injected gas, the cost differential between natural gas and coal and the open-market price of NO x emission credits. Further study, however, is necessary, including the construction of a more comprehensive database, needed to develop high-fidelity process models and to add carbon monoxide (CO) emissions to the model of the gas reburn system.

show abstract

A signal processing framework based on dynamic neural networks with application to problems in adaptation, filtering, and classification

Cited by 108 publications

References 19 publications

An adaptive recurrent neural-network controller using a stabilization matrix and predictive inputs to solve a tracking problem under disturbances

An adaptive recurrent neural-network controller using a stabilization matrix and predictive inputs to solve a tracking problem under disturbances

Neural Network Based Real-time Correction of Transducer Dynamic Errors

An Intelligent Emissions Controller for Fuel Lean Gas Reburn in Coal-Fired Power Plants

Contact Info

Product

Resources

About