曾志雄, 罗毅智, 余乔东, 等. 基于时间序列和多元模型的集约化猪舍温度预测[J]. 华南农业大学学报, 2021, 42(3): 111-118. DOI: 10.7671/j.issn.1001-411X.202010008
    引用本文: 曾志雄, 罗毅智, 余乔东, 等. 基于时间序列和多元模型的集约化猪舍温度预测[J]. 华南农业大学学报, 2021, 42(3): 111-118. DOI: 10.7671/j.issn.1001-411X.202010008
    ZENG Zhixiong, LUO Yizhi, YU Qiaodong, et al. Temperature prediction of intensive pig house based on time series and multivariate models[J]. Journal of South China Agricultural University, 2021, 42(3): 111-118. DOI: 10.7671/j.issn.1001-411X.202010008
    Citation: ZENG Zhixiong, LUO Yizhi, YU Qiaodong, et al. Temperature prediction of intensive pig house based on time series and multivariate models[J]. Journal of South China Agricultural University, 2021, 42(3): 111-118. DOI: 10.7671/j.issn.1001-411X.202010008

    基于时间序列和多元模型的集约化猪舍温度预测

    Temperature prediction of intensive pig house based on time series and multivariate models

    • 摘要:
      目的  从挖掘猪舍历史环境参数数据时序信息角度出发,提出基于时间序列模型和多元模型序列的猪舍温度预测模型。
      方法  采取缺失部分环境因子统计预测,评估猪舍环境中相对湿度、二氧化碳浓度、氧气浓度等环境因子对温度预测的影响程度。针对猪舍温度时间序列进行数据预处理,滤除错误值和缺失值,采用时间序列模型构建基于门控循环单元网络(Gated recurrent unit,GRU)的猪舍温度预测模型,采用多元模型建立基于梯度提升决策树(Extreme gradient boosting,XGBoost)缺失值重要程度的猪舍温度预测模型。将该预测模型用于预测广东省某集约化猪场母猪分娩舍温度,并与循环神经网络(Recurrent neural network, RNN)模型、反向神经网络(Back propagation neural network, BPNN)模型进行对比试验。
      结果  对比温度预测值与实测值发现,基于GRU模型对应的猪舍温度均方根误差和平均绝对误差分别为0.25和0.19 ℃,平均绝对百分比误差为0.65%;基于XGBoost多元模型的猪舍温度均方根误差和平均绝对误差分别为1.21和0.71 ℃,平均绝对百分比误差为2.50%。在时间序列的温度预测模型中,GRU模型表现出更优的预测效果;在多元模型的温度预测中,XGBoost模型的预测效果更优。
      结论  本研究使用的GRU模型在时间维度上对母猪分娩舍温度的变化起到了预警作用,确定了各种环境参数对温度的影响程度,为养殖环境的精细调控提供了参考。

       

      Abstract:
      Objective  On account of history environmental factors such as temperature and humidity in farrowing houses, based on the data of a farrowing house of a modern pig farm in Guangdong Province, the temperature prediction model of pig house was studied based on time series and multivariate models.
      Method  The effects of relative humidity, carbon dioxide concentration, oxygen concentration and other environmental factors on temperature prediction in pigsty was evaluated by statistical prediction of missing part environmental factors. Data preprocessing was carried out for the time series of piggery temperature to filter out the error value and missing value. The time series model was used to construct the temperature prediction model of piggery based on gated recurrent unit (GRU). Based on extreme gradient boosting (XGBoost), the temperature prediction model of missing value importance degree in pigsty was established using multivariate model. The prediction model was applied to predict the temperature of a farrowing house of an intensive pig farm in Guangdong Province, and compared with the recurrent neural network (RNN) model and back propagation neural network (BPNN) model.
      Result  Comparing the predicted value with the measured value, root mean square error (RMSE), mean absolute error (MAE) and mean absolute percent error (MAPE) of the GRU model were 0.25 ℃, 0.19 ℃ and 0.65% respectively, while RMSE, MAE and MAPE of the XGBoost multicomponent model were 1.21 ℃, 0.71 ℃ and 2.50% respectively. For temperature prediction based on time series models, GRU model showed better prediction effect. The XGBoost model was better in temperature prediction for multivariate model.
      Conclusion  The CRU model used in this study plays an early warning role on the temperature change of the farrowing house in time dimension, and the influence degree of various environmental parameters on temperature is also determined, which provides a reference for the fine regulation of the breeding environment.

       

    /

    返回文章
    返回