Abstract: Traditional BP neural network thermal error modeling is highly dependent on the initial value of the network and easily falls into the local optimal solution, resulting in the high sensitivity but insufficient robustness of the prediction model. This paper proposes to optimize the threshold weights of the BP neural network by using the whale optimization algorithm (WOA), which somehow remedies the fore-mentioned problem of BP neural network thermal error modeling. Take a Vcenter-55 three-axis vertical machining center for example. Six thermal error experiments are conducted in six months. Through fuzzy clustering and grey correlation, two temperature-sensitive points are identified, and then the Z-axis thermal error is taken as an example to build the WOA-BP neural network prediction model. A comparative analysis shows the prediction model improves the robustness of prediction accuracy by 3.35 μm on average compared with the traditional BP model, demonstrating its application values in engineering.