Research On The Detection Of Abnormal Data And The Application Of Supplementary Value In Agricultural Internet Of Things Based On Trusted Assessment

Intelligent agriculture,which deeply combines Internet of Things technology with agriculture,has strong advantages such as scientific cultivation,precise control and green agriculture.Abnormal data may occur in the process of data acquisition by sensors in the Internet of Things,and data loss may also occur in the process of data transmission.If these data are used directly,it will cause serious errors in the follow-up system and in decisionmaking.In this paper,the collected agricultural data will be tested for anomalies,and the outliers and missing values in the transmission process will be supplemented.For the agricultural data collected by sensors,this paper uses the Mahalanobis-Taguchi System(MTS)for anomaly detection,it can better judge the abnormal data collected.The main research contents include :(1)Mahalanobis Distance(MD)is calculated for normal data and abnormal data respectively to construct the Mahalanobis Space(MS).If the Mahalanobis Distance of abnormal data is not far greater than the Mahalanobis Distance of normal data,the data needs to be re-selected to build the Mahalanobis Space.(2)In order to reduce the workload of later prediction calculation and analysis,orthogonal array and SNR are used to optimize the reference space.The threshold value is determined by the F maximum method.(3)Using the collected real agricultural data to conduct anomaly monitoring experiment verification,the experimental results show that the application of Mahalanobis-Taguchi System(MTS)in data anomaly detection has a correct rate of 80.13%.For the outliers and missing values in the data,this paper uses the combination of time attribute and trust attribute to evaluate the missing values.The main research contents are as follows:(1)ARIMA(Auto Regressive Integrated Moving Average Model)is used to estimate the missing value of time attribute.The order of Auto Regressive(AR)and Moving Average(MA)models can be determined by ACF(Autocorrelation Function)and PACF(Partial Autocorrelation Function)after the difference of sample data,so that time attribute of data can be predicted.(2)ARIMA valuation method may be affected by its own errors,which may lead to the valuation errors.This paper proposes a multiple linear regression model based on trust nodes to estimate the missing value of trust attribute.Using the information interaction between nodes as the basis,the direct trust value between adjacent nodes and the indirect trust value between non-adjacent nodes are calculated by using the number of successful interactions,thenumber of failures and the undetermined number of communications.The subjective logic is converted into the trust value,and the discount operator and fusion operator are used to synthesize the trust value.Then the trust node is selected through the trust value,thus the multiple linear regression is used to predict the missing value.(3)ARIMA-MLRTA algorithm is composed of ARIMA model of time attribute and multiple linear regression based on trust attribute by weight allocation method,which has higher accuracy for complement value.(4)Experimental results show that ARIMA-MLRTA algorithm has better prediction effect and higher prediction accuracy than LIN estimation algorithm and KMRA estimation algorithm.