Generating action plans for poultry management using artificial neural networks

• Published in: Computers and electronics in agriculture Vol. 161; pp. 131 - 140
• Main Authors: Ribeiro, Richardson, Casanova, Dalcimar, Teixeira, Marcelo, Wirth, André, Gomes, Heitor M., Borges, André P., Enembreck, Fabrício
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.06.2019; Elsevier BV
• Subjects: Artificial neural networks; Automatic control; Controllers; Conversion ratio; Data processing; Decision making; Economic factors; Machine learning; Management; Model accuracy; Neural artificial networks; Perception; Poultry; Poultry farming; Poultry management; Reasoning; Sensing; Ventilation; Neural artificial networks; Sensing; Data processing; Poultry management; Machine learning
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/j.compag.2018.02.017

•The consumption of poultry meat has increased 17.96% over the last decade.•Sensor networks are used in agroindustry to collect large amounts of data, withing short periods of time.•Knowledge can be derived by applying machine learning algorithms over these data, generating usable models for decision making.•We generate action plans for poultry daily management using Artificial Neural Networks.•The result is a model that suggests actions plans to improve broilers production. The fundamental role for poultry farmers to be successful in their activities is to precisely increase, decrease, or maintain, in a short time span, factors that determine poultry growth, such as humidity, temperature, amount of feed ration, ventilation, and others. Although there are modern automatic control technologies supporting these aspects, systems are architected to react to environmental conditions based on predefined programmed control rules, without considering knowledge readings from historical data and, most importantly, the human specialist’s reasoning. In practice, when control actions diverge from the specialist’s opinion, signals of the automatic controller are immediately intercepted (via the system interface) to recalibrate them for a different control rule to be applied based on human perception. As the set of parameters tends to be large and they are frequently combined, whereas human perception tends to be limited, this intervention of automatic control tends to be an error-prone decision-making option. In this paper, we demonstrate that action plans for poultry management can be derived by systematically collecting data from the production environment. A sensor network is used to register poultry management data, which are then preprocessed using machine-learning techniques. To validate the obtained results, we compare them against action plans generated by a human specialist and baseline results. Analysis suggest that action plans derived from the proposed model follow, with acceptable accuracy, the control actions that should be taken by the controller when considering a knowledge-based perception that absorbs expert reasoning and best practices guidelines. The benefits of the proposed approach are discussed regarding economic factors such as average broiler weight and feed conversion ratio.