基于时序序列的猪舍环境综合评价方法研究

Comprehensive Evaluation Method of Pig House Environment Based on Time Series Sequences

摘要:

在集约化养猪生产中，猪舍环境是影响猪健康水平的重要因素。然而，多环境因子联合精准调控是制约猪舍环境控制的共性难题。因此，本文利用自适应高斯滤波（Adaptive Gaussian filtering, AGF）算法结合长短时记忆神经网络(Long short term memory networks, LSTM)进行舍内环境因子预测，为优化舍内环境调控策略提供支撑；结合组合赋权方式，确定猪舍内环境评价指标权重，构建基于未确知测度法评价方法，为猪舍环境调控提供参考。以实测猪舍数据对本文所提出方法进行验证，结果表明：相比LSTM预测模型，应用AGF优化算法后的LSTM预测模型（LSTM-AGF），其氨气质量浓度、温度、相对湿度、二氧化碳质量浓度的预测性能R2分别提升0.33、0.03、0.05、0.12；提出的基于未确知测度法的预测评价方法敏感度SENS为0.215，比传统模糊综合评价方法高20.80%。因此，本文提出的环境质量评价方法可以为猪舍环境精准调控提供参考。

关键词:猪舍环境;评价方法;时序序列;未确知测度;深度学习

Abstract:

In intensive pig farming, the pig house environment is an important factor affecting the health of pigs. However, the joint precise control of multiple environmental factors has always been a common problem in pig house environment control. Therefore, the adaptive Gaussian filtering (AGF) algorithm combined with long short term memory networks (LSTM) was used to predict the environmental factors inside the pig house, providing support for optimizing the control strategy of the pig house environment. By combining the weighted method, the weights of the environmental evaluation indicators inside the pig house were determined, and an evaluation method based on the unknown measurement method was constructed to provide reference for pig house environment control. The proposed method was validated by using measured data from pig houses, and the results showed that compared with the LSTM prediction model, the LSTM prediction model with the AGF optimization algorithm (LSTM-AGF) improved the prediction performance (R2) of ammonia, temperature, relative humidity, and carbon dioxide concentration by 0.33, 0.03, 0.05 and 0.12, respectively. The proposed prediction evaluation method based on the unknown measurement method had a sensitivity (SENS) of 0.215, which was 20.8% higher than that of the traditional fuzzy comprehensive evaluation method. Therefore, the environmental quality evaluation method proposed can provide feasible reference for precise control of the pig house environment.

Key words:pig house environment;evaluation method;time series sequences;uncertain measurement;deep learning