Authors

Abstract

In winter, the poor ventilation conditions in broiler houses may lead to high ammonia concentration, which affects the health of yellow-feather broilers or even causes the death of many broilers. This research used a machine learning model to predict the ammonia concentration in a broiler house during winter. After analysis, it was found that the ammonia generation in the broiler house was a gradual accumulation featured by non-linear data. After the broilers entered the broiler house for several days, and the ammonia concentration reached a certain value, a ventilation system was used for regulating the concentration. Firstly, the back-propagation (BP) neural network model and gated recurrent unit (GRU) model were used for predicting the ammonia concentration, respectively. Then, ensemble empirical mode decomposition (EEMD) was performed on the time series data of ammonia concentration in the broiler house. After that, the EEMD-GRU prediction model has been established for the intrinsic mode function (IMF) components and the temperature and humidity data in the broiler house. Finally, all component results were summarized to obtain the final prediction result. A comparison was conducted among the prediction results obtained by the above three models. The results show that the root mean square errors of the above three models are 6.2 ppm, 4.4 ppm, and 2.4 ppm, respectively, and the average absolute errors were 4.9 ppm, 2.8 ppm, and 1.6 ppm, respectively. It could be seen that the EEMD-GRU model had higher accuracy in predicting the ammonia concentration in the broiler house. The EEMD-GRU model can effectively predict the ammonia concentration in broiler houses, facilitating the feedback to the central system for timely adjustment.

Abstract in Chinese

针对冬季鸡舍通风限制，舍内氨气浓度高，轻则影响黄羽鸡健康，重则导致大量鸡死亡的问题，本文采用机器学习模型对冬季黄羽鸡舍内的氨气浓度进行预测。本文分析发现鸡舍内氨气生成是一个逐渐累积的过程，数据具有非线性，在鸡只进入鸡舍氨气浓度达到一定值后，采用通风系统进行调控。本文首先选择BP神经网络模型和GRU模型进行预测，再对鸡舍氨气浓度时间序列数据进行集合经验模态分解，并分别对分解得到的IMF分量和鸡舍内对应时间的温度湿度数据建立EEMD-GRU预测模型，最后对每个分量结果求和得到最终的预测结果。通过三种模型的预测结果对比，得到预测结果的均方根误差分别为5.957，4.681，2.491，平均绝对误差分别为4.106，3.126，1.812。由此可见基于EEMD-GRU模型的鸡舍氨气浓度预测精度更高，更准确，在鸡舍中应用可以有效预测未来的氨气浓度，反馈给系统及时调控。