Prediction of Dissolved Oxygen Concentration for Shrimp Farming Using Quadratic Regression and Artifcial Neural Network Kasorn Galajit NECTEC National Science and Technology Development Agency Pathum Thani, Thailand kasorn.galajit@nectec.or.th Pitisit Dillon Faculty of Applied Science King Mongkut’s University of Technology North Bangkok Bangkok, Thailand s5804021620048@email.kmutnb.ac.th Suradej Duangpummet NECTEC National Science and Technology Development Agency Pathum Thani, Thailand suradej.duangpummet@nectec.or.th Jakkaphob Intha NECTEC National Science and Technology Development Agency Pathum Thani, Thailand jakkaphob.intha@nectec.or.th Prachumpong Dangsakul NECTEC National Science and Technology Development Agency Pathum Thani, Thailand prachumpong.dangsakul@nectec.or.th Khongpan Rungprateepthaworn NECTEC National Science and Technology Development Agency Pathum Thani, Thailand khongpan.rungprateepthaworn@nectec.or.th Rachaporn Keinprasit NECTEC National Science and Technology Development Agency Pathum Thani, Thailand rachaporn.keinprasit@nectec.or.th Jessada Karnjana NECTEC National Science and Technology Development Agency Pathum Thani, Thailand jessada.karnjana@nectec.or.th Abstract—In aquaculture, one of the most critical factors for sustaining life under the water is the dissolved oxygen (DO) since it afects not only the animal survival rate but also the growth rate. Therefore, in smart aquafarming, the DO content should be monitored thoroughly. As a consequence, in practice, many DO sensors are installed in systems, and they contribute markedly to the system cost. This work aims to reduce the cost by replacing some DO sensors with a model that can describe the dynamics of DO content in a specifc controlled environment. Thus, we propose two predictive models: one based on the quadratic regression and another based on an artifcial neural network. Experimental results show that, under the limitation of the number of data used in the model construction, both models perform equally. Also, both prediction ftted more to observed data when the DO level is low. This fnding supports the practical model usage since in practice we more concern with the efciency of the model in the case of low DO concentration. Index Terms—dissolved oxygen model, shrimp farming, quadratic regression, neural network I. Introduction During 2002 and 2006, industrial shrimp farming in Thailand shifted from domesticating the native black tiger shrimp (Penaeus monodon) to domesticating the Pacifc whiteleg shrimp (Litopenaeus vannamei). From then on, the whiteleg shrimp (both frozen and processed) has become one of the primary export products of Thailand [1], [2]. When the demand increases due to the world population explosion [3], the aquaculture sector adopts the information, communication, and embedding technologies for intensive farming. In order to operate the intensive farming productively, automatic control systems are unavoidable [4]–[7] because environmental parameters need to be carefully controlled to meet the requirement for sustaining life under the water, as well as to prevent losses due to infectious diseases [8], [9]. For shrimp farming, one of the most critical water- quality parameters is the dissolved oxygen (DO) since it afects the survival rate, growth rate, and, consequently, the gross yield [10], [11]. Therefore, the DO concentration is thoroughly monitored in smart aquaculture. Hence, in practice, many DO sensors are deployed in the automatic control system [12]. These sensors, especially ones with the high accuracy, are expensive. As it goes without saying, they contribute markedly to the implementation cost of the system.