Articles
MUSHROOM TUNNEL CLIMATE: NEURAL NETWORKS COMPARED
Article number
456_57
Pages
477 – 484
Language
Abstract
This paper examines the application of artificial Neural Networks (NN) to the modelling of climate in mushroom tunnels.
The work is motivated by the difficulty encountered in the development of a structurally accurate first-principles model, compounded by the difficulty in choosing or estimating a set of model parameters.
The standard tunnel architecture is such that the climate contained within has several distinguishing characteristics, which is what makes a first-principles modelling approach particularly difficult.
Both Multi-Layer Perceptron (MLP) and Radial Basis Function (RBF), neural networks are employed for modelling.
Comparison of the performance of the networks is based on the Mean Squared Error (MSE) criterion.
Qualitatively the models achieved reasonable results, but did not meet the error criterion desired.
However, as the controlled variables examined in the models are enslaved by the airflow, the model of the airflow itself is of paramount concern.
So, the development of a NN airflow model in itself is a very positive result.
However, a successful modelling outcome is contingent on the incorporation of important structural information to determine appropriate model inputs, combined with a complete set of data for all the causal inputs.
The paper concludes that significant benefit is obtained from the employment of artificial neural networks as modelling tools.
The work is motivated by the difficulty encountered in the development of a structurally accurate first-principles model, compounded by the difficulty in choosing or estimating a set of model parameters.
The standard tunnel architecture is such that the climate contained within has several distinguishing characteristics, which is what makes a first-principles modelling approach particularly difficult.
Both Multi-Layer Perceptron (MLP) and Radial Basis Function (RBF), neural networks are employed for modelling.
Comparison of the performance of the networks is based on the Mean Squared Error (MSE) criterion.
Qualitatively the models achieved reasonable results, but did not meet the error criterion desired.
However, as the controlled variables examined in the models are enslaved by the airflow, the model of the airflow itself is of paramount concern.
So, the development of a NN airflow model in itself is a very positive result.
However, a successful modelling outcome is contingent on the incorporation of important structural information to determine appropriate model inputs, combined with a complete set of data for all the causal inputs.
The paper concludes that significant benefit is obtained from the employment of artificial neural networks as modelling tools.
Publication
Authors
D.P. Martin, J.V. Ringwood, J. Grant
Keywords
Environmental Control, Model, Characterisation, NN, MLP, RBF
Online Articles (65)