Deep Learning Approach Using K-Nearest Neighbors and Neural Network for Quality Assurance of Water Temperature Sensor in Automatic Weather Station

Water temperature is a crucial weather parameter observed by automatic weather stations, offering valuable insights into evaporation patterns, and playing a significant role in predicting rainfall. To ensure accurate and precise data from water temperature sensors, establishing a robust quality control process is essential. This research aims to achieve a unified quality control by employing deep learning with K-Nearest Neighbors (KNN) and neural networks and combined with step check and range check. The KNN method proves effective in imputing missing data, exhibiting a high correlation of0.992 and a low Root Mean Square Error (RMSE) of 0.78°C when compared to observed data. Furthermore, the neural network model, trained with 5, 10, and 30 previous data inputs, demonstrates remarkable accuracy, boasting an impressively low RMSE value of 0.088°C and a strong correlation of 0.99969. In the future, this study paves the way for future improvements in sensor performance and forecasting accuracy by exploring the integration of breakdown prediction mechanisms for water temperature sensors. Such an integration could bolster the system’s ability to detect potential issues and enhance its overall reliability.