Method for the selection of inputs and structure of feedforward neural networks

Saxén, Henrik; Pettersson, Frank

doi:10.1016/j.compchemeng.2006.01.007

Cited by 54 publications

(41 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…2, Table 2). This selection helped to prune the network by avoiding insignificant input data (Gunaratnam et al, 2003;Saxén and Pettersson, 2006) and to avoid cross-correlations (to some extent) between input variables, adding little extra information to the network. We have tested three approaches by using three different sets of input data: the most simple approach included solar radiation as a time of day indicator and the three seasonal fuzzy sets ("seasonal", Table 2), a second approach in which solar radiation has been removed and replaced by the time of day fuzzy sets ("diurnal", Table 2), while a third, a more thermo-hydrological approach is tested by integrating the lagged effect of precipitation and WTD which was applied to four out of six data sets ("lagged", Table 2).…”

Section: Discussionmentioning

confidence: 99%

“…There is currently no consensus in the scientific community on the number of neurons that should be used (Svozil et al, 1997;Saxén and Pettersson, 2006;Stathakis, 2009) when applying neural networks to data series. In order to apply the appropriate number of neurons, 25 repetitions were run for a selection of neurons (1-12) to help in choosing the appropriate number of neurons (Järvi et al, 2012) to be applied within the hidden layer of our networks.…”

Section: Applying Artificial Neural Network To Datamentioning

confidence: 99%

“…Another issue with neural networks is the possibility of under-or over-fitting of networks (Hansen and Salamon, 1990;Jain et al, 1996;Svozil et al, 1997). This can happen when data used to train the network is not representative enough for the entire observation span, if the number of hidden layers or neurons is not correct, if the global minimum is overshot or when the network learns the training pattern well but is underperforming during its validation (poor generalisation) (Jain et al, 1996;Gardner and Dorling, 1998;Nguyen and Chan, 2004;Wang et al, 2005;Saxén and Pettersson, 2006;Stathakis, 2009). To prevent this from happening, one can remove redundant input data (Gunaratnam et al, 2003;Saxén and Pettersson, 2006), reduce or increase the number of neurons in the network and use the appropriate generalisation such as early stopping (Hansen and Salamon, 1990;Amari et al, 1997;Svozil et al, 1997;Wang et al, 2005) or another stopping criteria (Günther and Fritsch, 2010;Fritsch and Günther, 2012).…”

Section: Artificial Neural Networkmentioning

confidence: 99%

“…The application of neural networks (Jain et al, 1996;Svozil et al, 1997;Elizondo and Góngora, 2005;Saxén and Pettersson, 2006) for data recovery and gap-filling (Aubinet et al, 2000;Gorban et al, 2002;Papale and Valentini, 2003;Ooba et al, 2006;Moffat et al, 2007 andSchmidt et al, 2008) has proven to be a very reliable tool in several scientific disciplines Dorling, 1998, 1999;Lek and Guégan, 1999;Lee and Jeng, 2002;Toptygin et al, 2005). In atmospheric sciences (Gardner and Dorling, 1998;Toptygin et al, 2005;Chattopadhyay G. and Chattopadhyay S., 2008), applying neural networks in forecasting has become a standard application.…”

Section: S Dengel Et Al: Testing the Applicability Of Neural Networmentioning

confidence: 99%

“…2). The information flows unidirectionally, from the input (covariates) layer to the output (response) layer, through the hidden layer(s) (Jain et al, 1996;Svozil et al, 1997;Elizondo and Góngora, 2005;Saxén and Pettersson, 2006).…”

Section: Artificial Neural Networkmentioning

confidence: 99%

See 4 more Smart Citations

Testing the applicability of neural networks as a gap-filling method using CH<sub>4</sub> flux data from high latitude wetlands

et al. 2013

View full text Add to dashboard Cite

Abstract. Since the advancement in CH 4 gas analyser technology and its applicability to eddy covariance flux measurements, monitoring of CH 4 emissions is becoming more widespread. In order to accurately determine the greenhouse gas balance, high quality gap-free data is required. Currently there is still no consensus on CH 4 gap-filling methods, and methods applied are still study-dependent and often carried out on low resolution, daily data.In the current study, we applied artificial neural networks to six distinctively different CH 4 time series from high latitudes, explain the method and test its functionality. We discuss the applicability of neural networks in CH 4 flux studies, the advantages and disadvantages of this method, and what information we were able to extract from such models.Three different approaches were tested by including drivers such as air and soil temperature, barometric air pressure, solar radiation, wind direction (indicator of source location) and in addition the lagged effect of water table depth and precipitation. In keeping with the principle of parsimony, we included up to five of these variables traditionally measured at CH 4 flux measurement sites. Fuzzy sets were included representing the seasonal change and time of day. High Pearson correlation coefficients (r) of up to 0.97 achieved in the final analysis are indicative for the high performance of neural networks and their applicability as a gapfilling method for CH 4 flux data time series. This novel approach which we show to be appropriate for CH 4 fluxes is a step towards standardising CH 4 gap-filling protocols.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Applying Artificial Neural Network To Datamentioning

confidence: 99%

Section: Artificial Neural Networkmentioning

confidence: 99%

Section: S Dengel Et Al: Testing the Applicability Of Neural Networmentioning

confidence: 99%

Section: Artificial Neural Networkmentioning

confidence: 99%

See 3 more Smart Citations

Testing the applicability of neural networks as a gap-filling method using CH<sub>4</sub> flux data from high latitude wetlands

et al. 2013

View full text Add to dashboard Cite

show abstract

Extraction of knowledge from high strength steel data using soft computing techniques—an overview

Datta

2009

Statistical Analysis

View full text Add to dashboard Cite

Soft computing techniques like artificial neural network, fuzzy logic and genetic algorithm are used to extract knowledge from experimentally developed data on the mechanical properties of thermomechanically processed high-strength steel. Though these techniques, in most case, are used for developing models or optimizing a system, here the additional factor of gathering better understanding of the steel system, under investigation, has also been targeted. The extracted information has been validated by the existing concepts of physical metallurgy of steel. It is seen that these tools have the capability to confirm some of the hypotheses generated through experimentation and could easily be utilized for designing the steel with superior and/or tailor-made properties.

show abstract

How multi‐objective genetic algorithms handle lack of data, sparse data and excess data: evaluation of some recent case studies in the materials domain

ChakrabortiN.¹

2009

Statistical Analysis

View full text Add to dashboard Cite

An overview of multi-objective optimization and the associated concept of Pareto-optimality are elucidated in detail, keeping the biologically inspired genetic algorithms in perspective. The effective role of the genetic algorithms in handling three different kinds of data driven models where the decision has to be made from (i) no data (ii) excess data or (iii) sparse data are elaborated through three materials engineering applications, where other strategies like inverse modeling, neural network and data mining have worked in tandem with the multi-objective genetic algorithms. 

show abstract

Method for the selection of inputs and structure of feedforward neural networks

Cited by 54 publications

References 13 publications

Testing the applicability of neural networks as a gap-filling method using CH<sub>4</sub> flux data from high latitude wetlands

Testing the applicability of neural networks as a gap-filling method using CH<sub>4</sub> flux data from high latitude wetlands

Extraction of knowledge from high strength steel data using soft computing techniques—an overview

How multi‐objective genetic algorithms handle lack of data, sparse data and excess data: evaluation of some recent case studies in the materials domain

Contact Info

Product

Resources

About

Method for the selection of inputs and structure of feedforward neural networks

Cited by 54 publications

References 13 publications

Testing the applicability of neural networks as a gap-filling method using CH&lt;sub&gt;4&lt;/sub&gt; flux data from high latitude wetlands

Testing the applicability of neural networks as a gap-filling method using CH&lt;sub&gt;4&lt;/sub&gt; flux data from high latitude wetlands

Extraction of knowledge from high strength steel data using soft computing techniques—an overview

How multi‐objective genetic algorithms handle lack of data, sparse data and excess data: evaluation of some recent case studies in the materials domain

Contact Info

Product

Resources

About

Testing the applicability of neural networks as a gap-filling method using CH<sub>4</sub> flux data from high latitude wetlands

Testing the applicability of neural networks as a gap-filling method using CH<sub>4</sub> flux data from high latitude wetlands