Data fusion method of livestock and poultry breeding internet of things based on improved support function

Abstract: Currently, IoT (Internet of Things) technology has been widely applied in livestock breeding. For the characteristics of perception data collected from livestock breeding IoT (such as multi-source, heterogeneous, real-time, and so on), we propose a livestock breeding IoT data fusion model. Firstly, it's easy to produce abnormal data in livestock breeding IoT due to the harsh work environment, transmission network noise and other factors. In order to guarantee the accuracy of livestock breeding IoT data, we check the consistency of original data collected by livestock breeding IoT sensors through the regression prediction method based on sliding window. We determine the sliding window size, then estimate the sensor measurement value at a certain moment through the regression forecast method and calculate the prediction interval, next determine whether the actual measurement value is abnormal compared with prediction interval, and finally replace abnormal data with predictive value. Secondly, there is the problem of uneven distribution of environmental monitoring values in breeding room. In order to comprehensively analyze and evaluate the breeding environment and provide accurate basis for automatic control equipment, we propose a homogeneous data weighted fusion algorithm which fuses homogeneous data that are collected from the same type of sensors based on support function. The support function has been proposed to describe a certain support degree or proximity between 2 numbers, and used to calculate the weights of a set of data. The weighted algorithm in this paper improves support functions so as to improve the accuracy of data fusion. We calculate the support degree of a set of pig breeding environment perception data to get corresponding optimal weights through the improved support function and then the weighted fusion for the same type of livestock breeding IoT sensor set data. Finally, in view of the characteristics different with livestock breeding multi-source perception data storage format and sensor devices encoding, the encoding rules format and data organization for livestock breeding IoT are formulated, which can uniformly describe heterogeneous data in the process of livestock and poultry breeding. Such a standard can be converted into a standard data format, which can work as the basis of data analysis and data applications. Areal production data from pig breeding IoT in Tianjin Huikang breeding pig farm show that in the data consistency test phase, the abnormal sensory data detection rate is 96.67%, which ensures the accuracy of the data, and in homogeneous data fusion phase, the improved support function in this paper has the minimum fusion variance of 0.192 5 compared with other 2 kinds of support functions in the multi-source data fusion calculation, which improves the accuracy of homogeneous data fusion and provides the accurate and reasonable basis for the automatic control of equipment. And coding standard of livestock IoT environment factors and sensor equipment makes unified expression to the sensors, which are from different manufacturers and environmental data with varied representations. Livestock breeding IoT data fusion model proposed in this paper can effectively improve the accuracy of data fusion, and meet the requirements of livestock breeding IoT data analysis.