Abstract:In order to effectively improve the accuracy of the thermal error prediction model based on nonlinear time series, the cluster number of fuzzy C mean clustering was determined by statistical test, and the key temperature measurement points were sorted and selected by combining the grey correlation analysis with different dimensionless processing. The thermal error prediction model based on NARX neural network was established; by setting the ranges of the input delay order, output delay order and the number of hidden neurons, the mind evolutionary algorithm was used to optimize the input and output delay order and the number of hidden neurons. Compared with the NARX neural network prediction model with randomly selected parameters, the prediction accuracy of the model is improved by 36.98%.