Development of Early-Warning Model for Intensive Pig Breeding

Following the rapid development of intensive pig breeding in China, the impact of environmental factors on the production and health of pigs has become increasingly apparent, and the monitoring of these environmental factors recognized as critical for improved breeding productivity. Based on the effects of environmental factors on pig growth, this paper established an early-warning model of the piggery environment. Using the model and the environmental factors, which were obtained in real time from a piggery, it was possible to obtain timely warning information, conducive to both creating an appropriate breeding environment for pigs and reducing the incidence of disease. In this article, we established the environmental early-warning indicators relating to pig breeding and then demonstrated the method based on single-factor and fuzzy comprehensive multi-factor models of the piggery environment. Finally, the two models were analyzed based on the experimental results, which showed that the fuzzy comprehensive early-warning model performed better than the single-factor model, and that it could be applied in an intensive farming environment to provide timely warning of environmental deterioration, to maintain the safety of the pig-breeding environment.


INTRODUCTION
In China environmental factors are becoming increasingly important to the health and productivity of pigs because of the rapid development of pig breeding. As a result, control of the piggery environment has become a critical factor in improving the efficiency of pig breeding. The environmental factors and other elements that directly or indirectly affect the growth, development, reproduction, and health of livestock include temperature, humidity, noise, illumination, and harmful gases [1][2][3]. For example, high temperature can cause a pig to reduce its nourishment intake and thus affect its growth. In this article, we analyzed the effect of each selected environmental factor and based on the results, built an environmental early-warning model. According to environmental data acquired in real time, warning information could be delivered by the early-warning model, making it conducive to both creating an appropriate breeding environment for pigs and improving the efficiency of pig breeding.
At present, research on the breeding of livestock and poultry is concentrated mainly on the influence of different environmental factors on the growth of the animal [2][3] and on monitoring the breeding environment [4][5][6][7]. Hansen et al. adopted a model-based control design method to develop a controller to regulate the airflow of the breeding environment to control its inner temperature and humidity [8]. Soldato et al. used robust nonlinear feedback control and feed forward control of the inner temperature and humidity to maximize livestock productivity [9]. However, in the field of pig breeding, little research has been reported on early-warning models, especially studies of early-warning models combining multiple environmental factors. At present, common early-warning models can be divided into single and multi-factor models. Single-factor models always incorporate the theory of "certainty" in the index model, and establish the worst factor for the warning of the piggery environment by comparing various factors with early-warning standards [10]. The common feature of multi-factor early-warning models is the inclusion of techniques such as artificial neural networks, the N.L. Nemerow Index method, and fuzzy comprehensive evaluation methods. Warnings achieved using artificial neural networks are better, but require additional training data [11]. The N.L. Nemerow Index method, despite considering the most significant factors, cannot reflect the relative importance of the various factors [11]. A fuzzy comprehensive early-warning model is a relatively mature method used widely in water-quality prediction that can produce realistic results. It describes the warning level by membership degree in fuzzy mathematics, and then determines the weight of each factor, evaluates it, and generates the warning [12][13][14].
In this article, we selected five environmental factors as early-warning indicators for the pig-breeding environment, and designed four grades of environmental warning: suitable, mild warning, moderate warning, and severe warning. Then, we demonstrated the method based on single-and fuzzy comprehensive multi-factor models for the piggery environment, and an experiment was designed to validate the model.

CONSTRUCTION OF EARLY-WARNING INDICATORS
The factors that affect the pig-breeding environment are complex and they can act collectively or individually to affect the animals in many ways. DTFs include temperature and humidity because values of these factors that are either too high or too low will affect the health of the pigs.
According to the environmental standards of pig breeding [15,16], the grading standards of environmental warnings are as shown in Tables 1 and   2.

Single-factor early-warning model
The single-factor early-warning model (SFM) is evaluated on single environmental factors of the piggery, and the worst factor taken as the indicator of the piggery environment.
The definition of the STF"s early-warning index i IS is as follows: where is the measured value of environmental factor i and is the standard value of environmental factor i; the value is exceeded when For DTFs, the early-warning index i ID is defined as follows: where is the measured value of environmental factor i, is the standard low threshold of factor i, and is the standard high threshold of factor i; the value is exceeded when ≠ 0 . In the FCM, a subjection degree matrix is initially established according to a set of subjection degrees for each environmental factor. Then, a comprehensive evaluation vector, which is the product of the subjection degree and weight of each environmental factor, is obtained. This vector reflects the subjection degree of the current environment belonging to each environmental early-warning grade [8].

Fuzzy comprehensive early-warning model
(1) The subjection degree calculation of STF This paper defined four warning grade standards: suitable, mild warning, moderate warning, and severe warning, which correspond to values of "j" = 0, 1, 2, and 3, respectively, and designate the subjection degree function for each grade for each factor.
In this paper, denotes the subjection degree of the STF "i" on the environmental standard grade "j".
The subjection degree function is constructed using the following trapezoid formula.
The subjection degree function of grade 0 for the STF is defined as follows: The subjection degree function of grade j (j = 1, 2) is defined as follows: The subjection degree function of grade 3 is defined as follows: where is the measured value of the STF I and is the maximal value of the warning grade j in the early-warning grading standard for the STF. In this paper, denotes the subjection degree of the DTF "i" on the environmental warning grade "j".
The DTF includes temperature and humidity, which correspond to values of "i" = 0 and 1, respectively.
The subjection degree function is constructed using the following trapezoid formula.
The subjection degree function of grade -3 for the DTF is defined as follows: The subjection degree function of grade j (j = -2, -1, 0, 1, 2) is defined as follows: The subjection degree function of grade 3 is defined as follows: where is the measured value of the DTF i and is the maximal value of the warning grade j in the early-warning grading standard for the DTF.
In the formula, means the i-th factor"s weight, is the measured value of the i-th factor, is the maximum of grade j of the warning grading standard, and is the minimum of grade j of the warning grading standard.
The weight of each factor is calculated as follows: where is the base weight of the i-th factor.
(4) The calculation of the early-warning grade Using the methods stated earlier, we can obtain a subjection degree matrix R that consists of a subjection degree for each grade of each factor, and the fuzzy comprehensive warning vector is the product of the weight vector W and the subjection degree matrix R.
Then, the fuzzy comprehensive warning vector B is calculated as follows: Finally, we calculated the grade of fuzzy comprehensive evaluation as follows: When the result matches: 0 ≤ < 1, the warning grade is 1; 1 ≤ < 2, the warning grade is 2; 2 ≤ < 3, the warning grade is 3; 3 ≤ , the warning grade is 4.

WARNING PROCESS
Based on the early-warning model stated earlier, this paper designed an environmental warning process, as shown in Figure 1. The detail of the warning process is as follows: (1) Search the breeding environmental standard from the breeding information library according to the pigs" breed and weight; (2) Every 10 minutes, data from the sensors are filtered using the Distribution method [17], and the average of these data calculated to evaluate the environment during that 10-minute interval using the early-warning model; (3) Check whether a warning is required according to the evaluation results. If a warning is not needed, wait another 10 minutes and return to step (2); otherwise, choose the appropriate warning from the warning method database according to the evaluated result.

EXPERIMENT AND ANALYSIS OF RESULTS
We installed a set of sensors of environmental parameters at an experimental station in Zhuozhou.
The parameters monitored included NH 3 , CO 2 , H 2 S, temperature, and humidity. We chose four periods for taking the experimental measurements, as shown in  Table 4.

CONCLUSIONS
The main contributions of this paper are as follows: (2) This paper compared a single-factor early-warning model with a fuzzy comprehensive early-warning model. The result of the experiment shows that when one or more factors exhibit significant deterioration, both models are able to describe the degree of deterioration of the environment accurately. However, when no single factor is seriously deteriorated, the performance of the fuzzy comprehensive early-warning model is superior to the single-factor early-warning model. ( 3) The warning model presented in this paper is shown to perform well and therefore it could act as a timely warning of environmental deterioration.