Abstract: Tool wear state is critical for quality stability and intelligent regulation in batch machining. However, most existing wear monitoring methods focus on short-term monitoring or single-step prediction, making it difficult to provide sufficient response time for regulatory compliance and tool changes. A model based on the whale optimization algorithm (WOA) for optimizing the long short-term memory network - iTransformer (longshort-term memory network-iTransformer, LSTM-iTransformer) has been proposed, which is used for multi-step prediction of tool wear. Firstly, the sensor signals during the turning process are collected and noise reduction processing is carried out on them. Secondly, the Spearman correlation coefficient is used to select the features that are most related to wear. Finally, the hyperparameters of the LSTM-iTransformer model are optimized using WOA, enabling it to fully utilize the memory capabilities of LSTM in handling time series data and the advantages of iTransformer (inverted transformer) in capturing long-range dependencies. Experimental results demonstrate that in the 20-step prediction of T3 data, the proposed model reduces the root mean square error (RMSE) by 43.7% and the mean absolute error (MAE) by 50.9% compared to the Transformer model. This significantly improves the prediction of future tool wear evolution and provides effective support for prospective health management in batch machining.