A Neural Network Structure for System Identification

Haesloop, Dan; Holt, Bradley R.

doi:10.23919/acc.1990.4791170

Cited by 42 publications

(12 citation statements)

References 1 publication

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Three models were created based on these data sets: A continuous linear model with first-order elements, a nonlinear ANN-ARMA model of the type found in the literature Bhat etal. 1990;Donat et al 1990;Haesloop and Holt 1990;Jones et al 1989;Jordan and Jacobs 1990;Narenda and Parthasarathy 1990;Pineda 1989;Waibel 1989), and a nonlinear ANN-CT model of the architecture described above. For each of the network models, results are averaged over 10 runs each.…”

Section: ___mentioning

confidence: 98%

“…The method of Haesloop and Holt uses linear information in the form of direct links from the input layer to the output layer, in addition to the standard pathways for activation through the hidden layers (Haesloop and Holt 1990). In this case, the output of the network would be calculated as the sum of the linear model and the nonlinear network, which is trained to account for only the residual nonlinear behavior .…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Neural Network Process Models Based on Linear Model Structures

Scott

Ray

1994

Neural Computation

View full text Add to dashboard Cite

The KBANN (Knowledge-Based Artificial Neural Networks) approach uses neural networks to refine knowledge that can be written in the form of simple propositional rules. This idea is extended by presenting the MANNIDENT (Multivariable Artificial Neural Network Identification) algorithm by which the mathematical equations of linear dynamic process models determine the topology and initial weights of a network, which is further trained using backpropagation. This method is applied to the task of modeling a nonisothermal chemical reactor in which a first-order exothermic reaction is occurring. This method produces statistically significant gains in accuracy over both a standard neural network approach and a linear model. Furthermore, using the approximate linear model to initialize the weights of the network produces statistically less variation in model fidelity. By structuring the neural network according to the approximate linear model, the model can be readily interpreted.

show abstract

Section: ___mentioning

confidence: 98%

Section: Related Workmentioning

confidence: 99%

Neural Network Process Models Based on Linear Model Structures

Scott

Ray

1994

Neural Computation

View full text Add to dashboard Cite

show abstract

“…By using the chain rule recursively [14]- [17], the parameters are adjusted in the point n of training samples. …”

Section: B the Training Of The Hf Elman Wnn Modelmentioning

confidence: 99%

A soft-sensing method of gas/liquid mass flow- rate based on hybrid feedback (HF) Elman wavelet neural network

Han

Dong

2010

2010 IEEE Instrumentation &Amp; Measurement Technology Conference Proceedings

View full text Add to dashboard Cite

In industry, gas/liquid two-phase flow in horizontal pipe is very common in many industry processes, and their measurements are of significance. In our works, a series of gas/water two-phase flow experiments in a horizontal pipe were conducted in which the V-cone differential pressure meter was used, and a soft-sensing method was brought forward in which the hybrid feedback (HF) Elman wavelet neural network was developed to achieve the gas/water mass flow-rate. Experimental results showed that the soft-sensing method combined V-cone with the HF Elman wavelet network could satisfy the demand of gas/liquid mass flow-rate measurement and the error of the mass flow-rate was small. The test results show the adaptability of the HF Elman wavelet network is better. The method is attractive for possible applications in measurement.Keywords-hybrid feedback Elman wavelet neural network; differential pressure; gas/liquid two-phase flow; mass flow-rate; Vcone

show abstract

“…By using the direct term, the SRWNN has advantages of a direct linear feedthrough network such as the initialization of network parameters based on process knowledge and enhanced extrapolation outside of examples of the learning data sets [9]. W is the weighting vector of SRWNN represented by:…”

Section: Description Of the Srwnn Structurementioning

confidence: 99%