Smart Greenhouse Early Warning System Based On Improved LSHiForest

Smart greenhouse is a representative application of Internet of Things technology in agricultural production field,which can monitor the temperature,humidity,lighting and other information in real time through various sensors and detect these information to find outliers supported by warning system.In this way,anomaly information is reported to the user.Furthermore,rolling shutters,lights,irrigation and other equipment in the greenhouse could be controlled automatically and the crops in the greenhouse could grow in a suitable environment.As a result,the yield and quality of crops could be enhanced and the revenue also increases.In this process,detecting the data collected by the sensor for warning system is a crucial issue.However,traditional anomaly detection algorithms originally designed for anomaly detection in static data have not properly considered the inherent characteristics of data stream produced by wireless sensor such as infiniteness,correlations and concept drift,which may pose a considerable challenge on anomaly detection based on data stream as follows:(1)data stream usually generates quickly which means that it is infinite.So any traditional off-line anomaly detection algorithm that attempts to store the whole dataset or to scan the dataset multiple times for detection accuracy will run out of memory space.(2)there exist correlations among different data streams,which traditional anomaly detection algorithms hardly consider.In this situation,detection accuracy of traditional anomaly detection algorithms may drop.(3)the data distribution in data stream may change over time which means that the detection model established by initial historical data may no longer fit the current concept and need to be updated to adapt to concept drift.Thus,traditional anomaly detection algorithms with no model update will be unable to predict new data distribution correctly.Besides,detection efficiency is also one of the most important issues to be considered.To address the above challenges,our paper improves traditional anomaly detection algorithm while considering the features of data stream and applies the improved algorithm to smart greenhouse warning system to achieve accurate and efficient anomaly detection.The specific research work is as follows:(1)LSHiForest is utilized to solve the problem of low efficiency in traditional anomaly detection,which takes advantage of hash functions to map data points.In this way,outliers are faster and easier to be distinguished and time cost is reduced.The detection efficiency has been enhanced.(2)For three challenges faced by traditional anomaly detection algorithms when processing data streams,a novel data stream anomaly detection method DLSHiForest is proposed based on LSHiForest,which combines sliding window and model update technique to detect anomalies in data stream.The method employs sliding window to store real-time data points to handle the infiniteness of data stream.In consideration of correlations between data streams,the method exploits hash functions to hash data points.To address the problem of concept drift,the method update anomaly detection model regularly.Comprehensive experiments are executed using real-world agricultural greenhouse dataset to verify the validity of above method,which compares different evaluation metrics by designing comparative experiments.Experimental results indicate that our method is feasible and effective.(3)An accurate and efficient smart greenhouse warning system is designed and implemented combined with the above improved algorithm.On the one hand,drawing lessons from the existing warning system,the system is designed based on the B/S framework pattern and three-tier architecture design.On the other hand,using Java programming language,the system is implemented in IDEA and Web Storm development tools.