IAPE '19, Oxford, United Kingdom ISBN: 978-1-912532-05-6 DOI: http://dx.doi.org/10.17501........................................ Development of ANN algorithms for forecasting Fresnel thermal power production Denia Kolokotsa, Angeliki Mavrigiannaki, Nikos Kampelis Technical University of Crete Technical University of Crete Campus, Office K2.107 Kounoupidiana, Chania, Chania, 73100 Tel: 0030 28210 37808 dkolokotsa@enveng.tuc.gr FilippoParedes, Luca Venezia, Fabio Montagino IDEA srl C.da Molara, Z.I. III Fase, Termini Imerese, Italy, 90018 Tel: 0039 091580305 fparedes@consorzioarca.it ABSTRACT Renewable energy production technologies are indispensable to the development of a clean and energy efficient built environment. Being dependent on climatic parameters, renewable energy production may vary causing a mismatch between energy demand and available production. Therefore, forecasting of renewable energy production allows the design and implementation of management schedules depending on expected production, thus assisting towards a more efficient and secure operation. The application of artificial neural networks (ANN) has been proved effective towards this aim. The present work is investigating the application of ANN for production forecasting of a concentrated solar power (CSP) fresnel system. The fresnel system production is optimum under clear skies and peak direct solar radiation. The system under investigation is connected to a thermal storage and its production is used to cover the heating/cooling loads of a building. The prediction can be used for scheduling the times that the system focalizes and de-focalizes for optimizing energy production and storage. The present work is relevant to researchers, engineers and developers of renewable energy production technologies that work on optimization of energy production and management. This work can be extended to smart grid energy management. Keywords CSP, fresnel, ANN, forecasting. 1. INTRODUCTION The built environment constitutes the greatest component of energy consumption and one that is also dependent on non- renewable sources [1]. On the way to developing a clean and energy efficient built environment, building energy design as well as energy production and distribution to buildings has been put under new perspective where renewable energy sources (RES) can play a key role. The concepts of zero energy buildings as well as smart energy grids both assume RES integration [1], [2], [3], [4]. Renewables can be an unpredictable source of energy since their production can be affected by climatic variations. Specifically for the case smart grids, RES integration presents challenges that need to be overcome for achieving efficient energy management and stability of the energy grid [3], [5]. Forecasting of renewable energy production can be an invaluable tool towards efficient energy and cost efficient management [6]. Artificial intelligence models have been recognized in literature as a reliable forecasting method [7], [8]. Artificial Neural Networks (ANN) are artificial intelligence models that have been widely used for forecasting with high accuracy and have been extensively used for short-term prediction [7], [8], [9], [10]. In [9] short-term load forecasting is studied with the aim to control charge and discharge of an electrical storage that is connected to a smart-grid with high PV integration. The proposed methods achieve accurate predictions that support efficient operation of the storage and management of the smart-grid. Load forecasting with the aim to achieve cost optimum operation of a smart micro-grid through optimum management of an electric storage that is connected to the grid is studied in [11]. The forecasted demand and production values are used as inputs of a genetic algorithm that decides the best amanegement strategy. A feed forward artificial neural network is proposed in [10] for predicting the load demand of a power grid. The proposed algorithm is able to achieve a highly accurate prediction that can support energy and cost efficient management of the grid. In [12], an ANN has been developed for forecasting 24 hours ahead the excess power production from renewables in a micro- grid. The prediction can be used for charging a thermal storage and therefore allowing optimum utilization of the energy produced from the RES of the micro-grid. In the present work, 24h forecasting of the thermal power production of a fresnel system is investigated with the aim to achieve efficient management of the system and cover power production intermittencies through controlled charge and discharge of a thermal storage.