Automated Sensor Data Validation and Correction with Long Short-Term Memory Recurring Neural Network Model

Authors : Abraham Sudharson PONRAJ, Vigneswaran T, Christy Jackson J

Pages : 43-57

View : 47 | Download : 8

Publication Date : 2021-03-29

Article Type : Research Paper

Abstract :Proper irrigation planning by matching reference evapotranspiration insert ignore into journalissuearticles values(ETo); with active crop growth requirement leads to an improved water usage efficiency and thereby improving the crop yield. ETo is primarily influenced by the following weather parameters the air temperature, relative humidity, wind speed and solar radiation. To make the ETo estimation system fault tolerant it is important to validate the real time data from the weather station, since the sensors used in these weather stations are prone to error due to influence of various environmental factors. A Recurring Neural Network insert ignore into journalissuearticles values(RNN); based Data Validation and Correction insert ignore into journalissuearticles values(DVC); algorithm was proposed to identify the faulty data and to correct them. Long Short-Term Memory insert ignore into journalissuearticles values(LSTM); RNN model is used to forecast the weather data such as temperature, solar radiation, wind speed and relative humidity. It uses statistical significance test to identify faulty data and isolate them. Then the DVC approach corrects the faulty data by replacing them by LSTM forecasted data. The performance evaluation of this approach showed better forecasting ability when compared with Seasonal Autoregressive Integrated Moving Average insert ignore into journalissuearticles values(SARIMA); based DVC and thereby improving overall performance of the DVC approach.
Keywords : Data Validation, LSTM, SARIMA, Reference Evapotranspiration ETo, Weather Data

ORIGINAL ARTICLE URL