Prediction Ability Analysis of Common Machine Learning Algorithms on Flow Field

Abstract

In this work, the primary mission is to apply the machine learning algorithms into practical use, predicting the velocity value at different positions according to the previous velocity history. There is a large data set containing the velocity data of 12000-time steps. Using the nonlinear methods, multiple linear regression (MLR), k-nearest neighbor (KNN), support vector machine (SVM), and artificial neural networks (ANN), the predictions of u and v component of the flow in the next time step are established. The results of the prediction will be compared with the actual data which is at the next time point. Then, the mean error and standard deviation in each case are evaluated. The results illustrate that ANN method possesses the lowest error and narrow distributions, which shows good accuracy and stability, the KNN method and SVM method are weaker than the artificial neural networks, and the MLR algorithm is the most inaccurate.