Data analysis of a monitored building using machine learning and optimization of integrated photovoltaic panel, battery and electric vehicles in a Central European climatic condition. (1st October 2020)

• Record Type: Journal Article
• Authors: Rehman, Hassam ur; Korvola, Timo; Abdurafikov, Rinat; Laakko, Timo; Hasan, Ala; Reda, Francesco
• Is Part Of: Energy conversion and management. Volume 221(2020)
• Publication Date: 2020-10-01
• Subjects: Net-zero energy building -- Solar energy -- Energy storage -- Electric vehicle -- Experimental analysis -- Machine learning; Direct energy conversion -- Periodicals; Energy storage -- Periodicals; Energy transfer -- Periodicals; Énergie -- Conversion directe -- Périodiques; Direct energy conversion; Periodicals; 621.3105
• DOI: 10.1016/j.enconman.2020.113206
• Languages: English
• ISSNs: 0196-8904

Abstract

Highlights: Using machine learning to estimate EV charging and PV generation from measure data. Using measured building load, EV and PV profiles to simulate and optimize system. Cost saving is not large with EV charge optimization, grid side analysis is needed. Energy cost reduce by 31% if PV generation to demand ratio increase from 49% to 97% The OEFe increased to 68% from 45% and when PV generation to demand ratio is 97% Abstract: Climate change is the biggest challenge of the present era. Buildings being one of the largest consumers of energy can help address this issue in various ways. The objective of the present study is to simulate and optimize a photovoltaic-based energy system integrated with onsite battery and electric vehicles for a demo site in Belgium in the Central European climatic conditions. The energy system of the demo site consists of photovoltaic-thermal panels, providing electricity and heat to the building, a battery as well as several long- and short-term thermal storages to reduce electricity curtailment and import. In the present study, the focus is on the optimization of the electric vehicle and battery charging and the import from the electrical network. Monitored data from the demo site is processed using a machine learning method to identify overproduction curtailment and electric vehicle charging events and to generate profiles of the photovoltaic (PV) energy production, feasible time windows and energy requirements for electric vehicle charging and building's energy demand. These profiles and data are then used in an optimization method using linear and mixed integer programming to minimize the annual cost of the purchased electricity. Different case studies in terms of PV capacity, battery capacity and electric vehicle charging scenarios are investigated to compare against the performance of the real system. Various criteria, such as the curtailment of the PV production, imported energy cost, onsite energy fraction and matching indicators and lastly the net-zero energy balance of the building are considered to compare the results of the case studies. The study shows that machine learning can be used for the investigation of the monitored data, which can be later used for simulating and optimizing the case studies. The results shows that PV panels with 20 kWp (PV generation to demand ratio of 97%) and battery capacity of 16.1 kWh can result in annual energy cost saving of up to 807€ (31% reduction) with PV curtailment ratio of 34%, compared with the installed real demo building that has 10 kWp PV panels (PV generation to demand ratio of 49%) and battery capacity of 32.2 kWh. This would lead to an increase in the onsite-energy fraction indicator from 45% to 68%. Moreover, this combination can bring the building closer to the net-zero energy balance if the surplus electricity is exported rather than being curtailed. The optimization of the energy system in terms of equipment sizing and operation control is needed to increase the techno-economic performance of the whole system.