19 Polish Technical Review No. 1/2020 YIELD FORECASTING Jan BARWICKI 1) , Marek HRYNIEWICZ, Anna GRZYBEK 1) Institute of Technology and Life Sciences, Warsaw Branch, Department of Rural Technical Infrastructure Systems E-mail of the corresponding author: jbarwicki@.gmail.com Jan Barwicki ORCID: 0000-0002-5437-5284 Marek Hryniewicz ORCID: 0000-0001-9926-699X ResearcherID: N-1547-2016 Scopus Author ID: 57210414113 YIELD FORECASTING USING ARTIFICIAL INTELLIGENCE PROGNOZOWANIE PLONÓW PRZY UŻYCIU SZTUCZNEJ INTELIGENCJI Summary: The article reviews and analyzes literature for application of artifcial intelligence in forecasting of crop yield. Yield forecasting models were based on neural networks, fuzzy logic or hybrid solutions. When designing new yield forecasting models, analyzes of the main factors of components that are important for yield forecasting should be performed. This is to eliminate unnecessary or negligible factors for forecasting. It is also important to review the databases that will be used for forecasting. The data with unusual numerical results that differ signifcantly from reality should be deleted. This will improve the quality of the databases and, as a result, will give better forecasting results. In more complex cases, it would be recommended to create hybrid solutions combining neural networks and fuzzy logic to combine the advantages of both solutions. Keywords: artifcial intelligence, hybrid solutions, fuzzy logic, neural networks, yield Streszczenie: W artykule wykonano przegląd i analizę literatury dla zastosowań sztucznej inteligencji przy prognozowaniu plonów. Modele prognozowania plonów były oparte o sieci neuronowe, logikę rozmytą lub rozwiązania hybrydowe. Przy projektowaniu nowych modeli prognozowania plonów należy przeprowadzić analizy głównych składowych czynników, które są istotne dla prognozowania plonu. Ma to na celu eliminację czynników zbędnych lub mało znaczących dla prognozowania. Istotne jest również dokonanie przeglądu baz danych, które zostaną wykorzystane do prognozowania. Słowa kluczowe: sztuczna inteligencja, rozwiązania hybrydowe, logika rozmyta, sieci neuronowe, wydajność Introduction Jayram and Marad [2012] found that accurate forecasting of crop yield is of increasing importance in the developed and the developing countries and everywhere where agricultural production is carried out. Reliable forecasts are expected due to the cost-effectiveness of the agricultural production and a high involvement of mechanical equipment. Sawasawa [2003] showed that knowing the size of crop yield before plants harvesting is important for decision-makers and politicians, especially in the regions with major climate changes and a capricious weather. It allows them to make a decision about buying cereals in case of their shortage or selling in case of their excess. This is related to food security, the risk of which can also be assessed using fuzzy logic belonging to one of the methods used in artifcial intelligence [Kadir and Inni, 2013]. Plant production can take place in the feld or in greenhouses and plastic tunnels. Field production is closely related to the weather [Baruth and Inni, 2008]. In greenhouses, on the other hand, production is rather independent on the weather conditions [Qaddoum i inni, 2013]. Yields in plant production depend on several overlapping factors that usually change relative to each other at the same time, in a non-linear way [Boniecki & Niżewski, 2010]. This makes it diffcult to predict yields using traditional me- thods. Boniecki and Niżewski [2010] proposed the use of procedures based on the artifcial intelligence methods. Artifcial intelligence appeared as a new feld of knowledge along with the development of science. According to Kwatera [2016], artifcial intelligence is a relatively new interdisciplinary feld of science, a subject of great expectations and lively debates. In a theoretical sense, it combines the issues in the feld of computer science, psychology, anthropology, mathematics, electronics, neurophysiology and philosophy. This serves to solve problems based on natural cognitive processes of man [Bartman 2017]. Objective, scope and methodology of work The aim of the study was to review the artifcial intelligence methods used to forecast crop yields. The scope of the work was to specify the methods and factors used in these methods that DOI: 10.15199/180.2020.1.3