In any solar adsorption refrigeration system,there are three major components:a solar collector adsorbent bed,a condenser and an evaporator.All of those components operate at different temperature levels.A solar colle...In any solar adsorption refrigeration system,there are three major components:a solar collector adsorbent bed,a condenser and an evaporator.All of those components operate at different temperature levels.A solar collector with a tubular adsorbent configuration is proposed and numerically investigated.In this study,a nonlinear auto-regressive model with exogenous input is applied for the prediction of adsorbent bed temperature during the heating and desorption period.The developed neuronal model uses the MATLAB Network toolbox to obtain a better configuration network,applying multilayer feed-forward,the TANSIG transfer function,and the back-propagation learning algorithm.The input parameters are ambient temperature and the uncontrolled natural factor of solar radiation.The output network contains a variable representing the adsorbent bed temperature.The values obtained from the network model were compared with the experimental data,and the prediction performance of the network model was examined using various performance parameters.The mean square error(MSE)and the statistical coefficient of determination(R2)values are excellent numerical criteria for evaluating the performance of a prediction tool.A well-trained neural network model produces small MSE and higher R2 values.In the current study,the adsorbent bed temperature results obtained from a neural network with a two neuron in hidden layer and the number of the tapped time-delays d=9 provided a reasonable degree of accuracy:MSE=1.0121 and R2=0.99864 and the index of agreement was 0.9988.This network model,based on a high-performance algorithm,provided reliable and high-precision results concerning the predictable temperature of the adsorbent bed.展开更多
文摘In any solar adsorption refrigeration system,there are three major components:a solar collector adsorbent bed,a condenser and an evaporator.All of those components operate at different temperature levels.A solar collector with a tubular adsorbent configuration is proposed and numerically investigated.In this study,a nonlinear auto-regressive model with exogenous input is applied for the prediction of adsorbent bed temperature during the heating and desorption period.The developed neuronal model uses the MATLAB Network toolbox to obtain a better configuration network,applying multilayer feed-forward,the TANSIG transfer function,and the back-propagation learning algorithm.The input parameters are ambient temperature and the uncontrolled natural factor of solar radiation.The output network contains a variable representing the adsorbent bed temperature.The values obtained from the network model were compared with the experimental data,and the prediction performance of the network model was examined using various performance parameters.The mean square error(MSE)and the statistical coefficient of determination(R2)values are excellent numerical criteria for evaluating the performance of a prediction tool.A well-trained neural network model produces small MSE and higher R2 values.In the current study,the adsorbent bed temperature results obtained from a neural network with a two neuron in hidden layer and the number of the tapped time-delays d=9 provided a reasonable degree of accuracy:MSE=1.0121 and R2=0.99864 and the index of agreement was 0.9988.This network model,based on a high-performance algorithm,provided reliable and high-precision results concerning the predictable temperature of the adsorbent bed.
