Atmospheric Turbidity Forecasting using Side-by-side ANFIS

In a context of sustainable development, enthusiasm for CSP technologies is increasing. In addition, the CSPIMP (Concentrated Solar Power efficiency IMProvement) European project has been recently initiated to achieve a better competitiveness of the CSP plants. Its main objective is to develop a new procedure to improve the steam turbine start up cycles, maintenance activities and advanced plant control schemes. A challenge in the project is to forecast the solar resource with the aim of improving the management of CSP plants. A key parameter when trying to estimate or forecast solar radiation is atmospheric turbidity. Indeed, the Direct Normal Irradiance (DNI) under clear sky conditions can be expressed as a function of extraterrestrial irradiation, altitude and atmospheric turbidity. So, this paper focuses on forecasting atmospheric turbidity at different time horizons (up to 3 hours) using side-by-side Adaptive Network-based Fuzzy Inference Systems (ANFIS). First, a Multi-Resolution Analysis (MRA) based on the discrete wavelet transform allowed clear sky DNI values to be extracted from the NREL database. In addition, a Principal Component Analysis (PCA) has been considered in order to develop the forecasting model using uncorrelated input variables and reduce its complexity (and, as a consequence, computation time). Finally, the results we obtained about atmospheric turbidity forecasting are satisfactory and validate the proposed approach.