Correction Method Of Agricultural IoT Abnormal Data Based On Machine Learning

With the development of Io T information technology,a large amount of data has been generated and accumulated in the agricultural field,which providing a rich source for agricultural data processing.However,due to factors such as agricultural devices cost and complex environment,the massive data contains a large amount of low-quality data,which reduces the data availability.The primary problem to be solved in data processing is how to detect and utilize the abnormal data.In this paper,the environment of Houkui growth in Taiping was taken as the research object,and the correction method of abnormal data in tea plantation environment was studied.The correction method mainly combines the abnormal data detection method and the time series data prediction method,and updates the abnormal data detected by the abnormal data detection method through the prediction data obtained by the prediction method.The research contents include:Abnormal data detection method.According to the temporal and complex characteristics data of tea plantation environment,a method of abnormal data detection based on machine learning is proposed.In the proposed method,scale transformation and time slicing technology are used to preprocess the original data.The anomaly detection model is components of Convolution Neural Network,Gated Recurrent Unit and Support Vector Machine,which uses Convolution Neural Network to extract local features and Gated Recurrent Unit to extract time series features,to improve the nonlinear representation of the model.Then Support Vector Machine is used to classify the combined features.Finally,the simulation and comparative experiments are carried out by using the tea plantation environment data,and the experimental results are judged by the classification evaluation index and confusion matrix.The experimental results show that the accuracy of the CNN-GRU-SVM model proposed in this paper is 4%～14%higher than other detection models.In the classification evaluation index,the CNN-GRU-SVM model is more effective than other methods.Time series data prediction method.A time series data prediction method based on variant long short memory recurrent neural network is proposed.The CIFG network is adopted to constructs multiple time series data prediction models with multiple inputs and single outputs structure,which is increasing timeliness of time series data prediction,reducing model training time and improving model prediction accuracy,and the number of prediction models is mainly determined by the type of predicted data.The prediction model is composed of a double CIFG neural network layer and a dropout layer.The CIFG neural network layer is used to learn the time series characteristics of tea plantation environment data,and the dropout layer is used to improve the generalization ability of the model,simplify the learning parameters and increase the training efficiency.On this basis,the simulation and comparative experiments are carried out by using the data,the results show that the training time of CIFG network is 19%less than LSTM network.In the predictive evaluation index,the average MAE of each category of the CIFG prediction model is about0.0657,the average MSE is about 0.0138,and the average R~2 is about 0.9957.Abnormal data correction method.The abnormal data correction method proposed in this paper combines the abnormal data detection model and the time series data prediction model.Firstly,the abnormal data detection model is used to detect to determine input to which prediction model,and then the prediction data is updated with the normal data to output.Finally,carried out simulation experiments with the tea plantation environment data,the correction accuracy is proposed to judge the performance of the correction method.The results show that the average accuracy of the ten corrections of this correction method was90.65%.This method has good performance on abnormal data of tea plantation environment summarily.