Characterization of mirror soiling in CSP applications

Daily reflectance measurements on mirrors used in CSP applications were obtained over a period of 4 months to monitor soiling as a function of meteorological and environmental parameters. The mirrors were placed at 45° increments from face down to face up, and the parameters monitored included temperature, relative humidity, wind speed, rainfall and particulate matter. In order to determine a relationship between the input parameters and the soiling, both multiple linear regression and artificial neural network models were employed. The feed-forward back-propagation neural network with two layers and 16 neurons per layer is the configuration that reaches the best predictive results without overfitting the data, reaching a correlation coefficient of 0.84, compared to a maximum correlation coefficient of 0.60 with the multiple linear regression allowing for quadratic interaction of terms. The model is used to predict the number of annual cleaning operations required for a CSP plant placed in a location with a similar climactic profile, in order to estimate the cleaning contribution to the annual operation and maintenance cost.