基于RF-LSTM的鸡舍恶臭气体预测研究

以鸡舍氨气为研究对象，对鸡舍氨气预测模型进行了研究.首先，利用随机森林算法（RF）对影响鸡舍氨气浓度的环境变量进行重要性排序，选取温度、湿度、光照、气象温度、降雨量作为模型的输入变量；在此基础上，构建了基于长短时记忆神经网络（LSTM）的鸡舍氨气浓度预测模型，并将提出的预测模型应用于江苏省宜兴市某养鸡场的氨气浓度预测中，并与LSTM模型、RF-Elman模型和RF-BP模型进行了对比实验，结果表明，基于RF-LSTM模型的预测效果最好，其平均绝对误差（MAE）、平均绝对百分误差（MAPE）和均方根误差（RMSE）分别为0.9183、4.9637%和1.4262；同时，为了验证该模型的性能，本文还实现了不同时间尺度的鸡舍氨气浓度预测，提前2h、3h、4h、5h氨气预测的平均绝对误差（MAE）分别为1.6218、2.1991、2.8553和3.0677.本文提出的预测模型提高了鸡舍氨气浓度的预测精度，可为减少鸡舍恶臭气体排放提供科学依据.

Prediction of stench gas in chicken house based on RF-LSTM

In this paper, the concentration of stench gas in chicken house was studied in order to construct prediction model. First, random forest (RF) algorithm was used to rank the importance of environmental variables that affect chicken house ammonia gas concentration, temperature, humidity, light, meteorological temperature and rainfall were selected as input variables of the model; Based on this, a model for predicting ammonia concentration in chicken houses based on long-term and short-term memory neural network (LSTM) was constructed. The prediction model proposed in this paper was applied to the ammonia concentration prediction of a chicken farm in Yixing experimental base of Jiangsu Province. The results were compared with LSTM model, RF-Elman model and RF-BP model. The results showed that the prediction effect based on RF-LSTM model was the best. The average absolute error (MAE), average absolute percentage error (MAPE) and root mean square error (RMSE) were 0.9183, 4.9637% and 1.4262, respectively. At the same time, in order to validate the performance of the model, the ammonia concentration prediction in chicken houses at different time scales was also realized. The average absolute errors (MAE) of ammonia prediction in advance of 2hours, 3hours, 4hours and 5hours were 1.6218, 2.1991, 2.8553 and 3.0677, respectively. The prediction model proposed in this paper improves the prediction accuracy of ammonia concentration in chicken houses, and provides scientific basis for reducing the odor emissions from chicken houses.