This document summarizes a research paper on using artificial neural networks (ANN) to predict the performance of solar air heaters. Solar air heaters convert solar energy to thermal energy and have various applications. ANN is a machine learning technique that can be used to optimize, simulate, and predict non-linear systems like solar air heaters. The summary discusses common ANN models like multi-layer perceptrons that are used for performance prediction. Data collection, preprocessing, and dividing the data into training and testing sets are important steps. ANN techniques can successfully predict thermal and exergetic performance of different solar air heater designs. Proper selection of input parameters and training data is key to accurate predictions.
1. AN ASSIGNMENT ON PERFORMANCE
ANANLYSIS OF SOLAR AIR HEATER USING
ANN TECHNIQUE
NATIONAL INSTITUTE OE TECHNOLOGY
JAMSEDPUR
Submitted by :-
Rajbala Purnima Priya
rajbalapurnimapriya@gmail.com
3. SAH are advantageousbecause of lowfabricationcost,low operationandinstallationcost( as compared
to other solar energy techniques ) , high efficiency and many others.
Artificial Neural Network (ANN)
Artificial NeuralNetworkisatechnique usedfor optimization , simulation, clustering of information or
data , pattern detection , perdition, and to solve the non- linear functions. Artificial Neural Network
(ANN) is the best and most commonly used technique in the field of Articial Intelligence (AI) . The
operation of ANN is similar to human’s brain. To increase the performance of SAH many experimental
examination are used but they are very expensive and time consuming. ANN is soft computing
technique whichisfast andaccurate andit doesnotrequire anyprogramming code.InthispaperANN is
used as performance predictor tool for Solar air heaters.
The working of ANN is done in two ways: (i) Learning the data set (ii)storing the data set in weights
There are various model in ANN to solve the problems . They are : (i) Multi -Layer Perceptran (MLP)
(ii) Generalized Regression Neural Network (GRNN) (iii) Radial Basis Function (RBF) Neural Network
Multi-Layer Perceptran
In general the multi-layer perceptron (MLP) model is most commonly used neural model, which
structuredwiththree differentlayerssuchasinputlayer,output layer and hidden layer. Figure 3 shows
the general structure of MLP feed forward neural networks. The product of each input signals (ai) and
weights (wij) are passed through the summing junction and added with bias (bj), which is
expressed by
X=(∑ 𝒘𝒊𝒋 𝒂𝒊𝒏
𝒊=𝟏 ) + bj
For the outputgenerationthe sumXis goesto transferfunctionasshowninbelow figure
FIGURE 2 : Types of SAH
4. The transferfunctionsare usedinthe hiddenandoutputlayer.The mostcommonlyusedfunctionsare
logsig,tansigandpurelin.The logsigandtansigare nonlinearactivationfunctionswhichiscalledas
sigmoidal function.The logsigtransferfunction outputsare liesbetween0and 1 whichisexpressedas
followingfunction:
F(X) = 𝟏
𝟏+𝑬𝑿𝑷(−𝑿)
If negative valuesare foundininputandoutputlayers,then tansig transferfunctionispreferredanditis
givenby
F(X)=
𝑬𝑿𝑷( 𝑿)−𝑬𝑿𝑷(−𝑿)
𝑬𝑿𝑷( 𝑿)+𝑬𝑿𝑷(−𝑿)
Basic steps for prediction using ANN Technique
Step1 : Selectthe variables
Step2 : Collectionof data
Step3 : Data pre processing
Step4 : Divide the datasetsinTraining,testingandvalidation
Step5 : Model developmentbyselecting:numberof hiddenlayer,numberof neurons
and transferfunction
Step6 : Train the model
Step7 : Checkthe model performance byerroranalysis
Step8 : Save the outputdata
RESULT
In this paper, a comprehensive review has been carried out to predict the performances of solar air
heatersusingartificial neural network. Different types of solar air heaters using absorber plate such as
artificial roughness, extended surfaces, and porous/packed bed were used by researchers in the
experimentsfordatacollectionwhichisusedinANN modelingandtraining.The thermal performances,
exergeticperformancesandheattransfercharacteristicswere predictedsuccessfully byvarioustypes of
ANN models andvariouslearningalgorithms .Foraccurate predictionusingANN technique,selection of
suitable inputparametersandtrainingdata sets is most important. Generally, in input layer: operating
parameters, system parameters and metrological parameters are used as input parameters, and in
output layer the estimated/ calculated parameters are used as output parameters. In middle layer
(hiddenlayer) optimal numbers of neurons are taken up for accurate predictions. Due to the ability to
solve the complicated problems where conventional methods are not used, now-a-days the ANN
applications are more popular for optimization and prediction of thermal systems especially in solar
energy systems.
The main conclusions of the present review can be summarized as follows:
1. Mostly MLP model is used for performance predictions of different types of solar air heaters.
FIGURE 3: Basic design of Neural Structure
5. 2. The present literature review reveals that the ANN technique is better than the conventional
methods.
3. Selection of input parameters and hidden neurons plays an important role in ANN prediction.
4. The performance of ANN models highly depends on the selection of training
data sets.