This paper focuses on the prediction of the coating amount before the construction of ship painting, i.e., the calculation of coating consumption quota. At present, each shipyard uses a larger coating loss coefficient to calculate the coating consumption quota; after the construction, there is often a lot of inventory left, which is not conducive to the scientific management of the ship coating process and the cost control of shipbuilding. Therefore, this paper proposes a coating consumption prediction method based on ensemble learning, using cosine similarity and gradient boosting regression tree hybrid algorithm (CS-GBRT) to calculate the coating loss coefficient under different working conditions. Cosine similarity is used to select similar data with less difference from the target to be predicted as the training set, and the loss function in GBRT is improved based on similarity weight to improve the prediction performance and calculation accuracy of GBRT. The coating data recorded by a shipyard from 2014 to 2019 are randomly selected to evaluate the prediction ability of the model established in the paper. The results show that when the proposed CS-GBRT algorithm is used to calculate the coating loss coefficient, the mean absolute error of training set and test set are both < 1.4, and the mean absolute error percentages are both < 4%. Compared with other research methods, the prediction accuracy is obviously improved, and the output feature importance is also consistent with the trend calculated by Spearman method, which proves the validity of the model again.

本文着重于船舶涂装前的涂料用量预测，即涂料消耗定额的计算。目前，每个造船厂使用较大的涂料损耗系数来计算涂料消耗量。施工结束后，往往会留下大量库存，不利于船舶涂装工艺的科学管理和造船成本控制。因此，本文提出了一种基于集合学习的涂料消耗量预测方法，利用余弦相似度和梯度提升回归树混合算法（CS-GBRT）来计算不同工况​​下的涂料损耗系数。余弦相似度用于选择与待预测目标作为训练集的差异较小的相似数据，基于相似度权重改进GBRT的损失函数，提高GBRT的预测性能和计算精度。从船厂2014年至2019年记录的涂层数据是随机选择的，以评估本文建立的模型的预测能力。结果表明，采用本文提出的CS-GBRT算法计算涂层损耗系数时，训练集和测试集的平均绝对误差均小于1.4，平均绝对误差百分率均小于4％。与其他研究方法相比，预测精度明显提高，输出特征重要性也与Spearman方法计算的趋势一致，再次证明了该模型的有效性。