Abstract: In largescale modern intelligent livestock and poultry breeding, due to the uneven distribution of temperature in livestock and poultry houses and the low efficiency of sensor data collection, it is impossible to comprehensively, accurately and timely reflect the change of temperature in livestock and poultry houses. To improve the performance of temperature monitoring system in livestock breeding, a realtime fusion strategy of layered wireless sensor network (WSN) was proposed in this paper. The WSN designed by this strategy is divided into two layers. Firstly, the temperature data collected by the bottom sensor is preprocessed by an improved unscented Kalman filter (IUKF). Then, the fusion center uses the improved dung beetle algorithm to optimize the nuclear Extreme Learning machine (IDBOKELM) for realtime fusion of the preprocessed temperature data. The experimental results show that the improved unscented Kalman filter in data preprocessing can effectively suppress noise interference in livestock and poultry houses, overcome abnormal and divergent phenomena in collected data. In terms of multisensor data fusion, the IDBOKELM algorithm established in this article has an accuracy of 99.15% in the training set and 98.12% in the test set, respectively. Compared to the original algorithm, the accuracy is improved by 6.98%, and the data fusion time is 3.36 s, ensuring the efficiency and accuracy of temperature monitoring in poultry houses while reducing computational time.

Key words: livestock premises, multisensor data fusion, environmental monitoring, IUKF

摘要： 在现代化大型智能畜禽养殖中，由于畜禽舍内温度分布不均匀、传感器采集数据效率低下等原因，无法全面、准确及时反映畜禽舍内温度变化情况。为提高畜禽养殖温度监测系统的性能，提出一种分层无线传感器网络(WSN)实时融合策略。该策略设计的无线传感器网络分为两层，首先将底层传感器采集的温度数据利用改进的无迹卡尔曼滤波器(IUKF)进行预处理，然后融合中心利用改进蜣螂算法优化核极限学习机(IDBOKELM)对预处理后的数据进行实时融合。试验结果表明，在数据预处理方面改进的无迹卡尔曼滤波器能够有效抑制畜禽舍内噪声干扰，克服采集数据出现异常和发散现象；在多传感器数据融合方面本文建立的IDBOKELM算法其训练集与测试集准确率分别是99.15%和98.12%，相较于原始算法准确率提升6.98%，数据融合用时3.36 s，保证禽畜舍内温度监测的效率和准确性，同时减少运算时间。

关键词: 畜禽舍, 多传感器数据融合, 环境监测, 改进无迹卡尔曼滤波