基于神经网络猪舍氨气浓度预测方法研究

Study on prediction method ammonia concentrations in pig house using Neural Network

猪舍内氨气浓度对猪生长发育影响较大,建立准确氨气浓度预测模型尤为必要。目前已有针对猪舍内氨气浓度预测研究,但氨气浓度受猪舍内多种环境因素影响,缺少准确预测模型。为此本研究从实测猪舍内环境数据(包括氨气浓度、温度、湿度、活动量、通风)中随机选取1 537组数据,使用L-M算法优化BP神经网络、线性神经网络和Elman神经网络预测猪舍内氨气浓度。结果表明,基于L-M算法优化BP神经网络建立5-9-9-1四层结构预测模型经290步后达目标误差,预测值和真实值最大绝对误差仅为0.1720,与Elman神经网络和线性神经网络预测方法相比可提高猪舍氨气浓度预测准确性和及时性,为猪舍环境预警提供支持。

It was necessary to establish prediction models for ammonia concentrations in pig buildings because high concentrations of ammonia can affect pigs＇ health and growth. There are few prediction models for ammonia concentrations, especially those using the neural network for ammonia concentration inside pig buildings because changes of ammonia concentrations are dynamic and nonlinear, and are affected by other environmental factors. In this paper, 1 537 sets of environment data including ammonia concentration, temperature, humidity, activity amount and ventilation were selected randomly and used for ammonia concentration prediction in three types of models including optimized L-M BP neural network, linear neural network and Elman neural network. The results showed that a 4-layer structure of 5-9-9-1 built with BP neural network based on optimized L-M algorithm was the best prediction model, which was set up after 290 steps to achieve the target error. The maximum absolute error between the real and estimated values was 0.1720. Compared with linear neural network and Elman neural network, it can improve the accuracy and timeliness of ammonia concentration prediction and provides support for early warning in pig buildings.