Prediction model of continuous discharge coefficient from tank based on KPCA-DE-SVR

Abstract

The discharge of hazardous liquids from storage tanks poses a serious threat to the surrounding environment and humans in consideration of the potential risk of catastrophic fire and explosion. Hence, it is essential to precisely predict discharge coefficient of a continuous leakage to benefit risk assessment and management and accident prevention. This study proposed a prediction model using a hybrid KPCA-DE-SVR algorithm for the discharge coefficient for sustaining discharge (C_s). It was developed based on experimental data of a continuous discharge. The Kernel Principal Component Analysis (KPCA) algorithm was applied to reduce redundant variables to improve data quality; the Differential Evolution (DE) algorithm was employed to optimize the Support Vector Regression (SVR) model to improve the generalization ability of model; and the SVR algorithm was utilized for both training and testing in order to construct the prediction model of C_s. Compared with the prediction performance of four models (SVR, KPCA-SVR, KPCA-GA-SVR, and KPCA-DE-SVR), it was found that the KPCA-DE-SVR model had the highest prediction accuracy (MAE = 0.0211, RMSE = 0.0006, R² = 0.9649). This study provides an important technical insight for improving the prediction accuracy of C_s from a continuous discharge.