Abstract: Dam safety monitoring data exhibit diverse time series types. Inaccurate or untimely identification of time series types may lead to mismatches between anomaly detection models and warning criteria, thereby reducing the reliability and efficiency of anomaly identification. To improve the accuracy and efficiency of automatic time series type identification for dam safety monitoring data, this study addresses key challenges such as strong nonstationarity and blurred class boundaries in monitoring time series. An automatic time series type identification method integrating time-frequency analysis, feature learning, and parameter optimization, termed IWOA-STFT-SE-CNN, is proposed. In this method, the short-time Fourier transform (STFT) is employed to convert monitoring time series into multi-channel representations based on two-dimensional time-frequency spectrograms. A lightweight convolutional neural network (CNN) incorporating squeeze-and-excitation (SE) modules is constructed to extract discriminative features and perform classification. Furthermore, an improved whale optimization algorithm (IWOA) is introduced to adaptively optimize key hyperparameters of the network. Engineering validation and ablation experiments are conducted using monitoring data from multiple dams. The results demonstrate that the proposed method achieves an average classification accuracy and F1-score of 99.86% across six typical types of monitoring time series, significantly outperforming conventional end-to-end time-series models and CNN models without hyperparameter optimization. The collaborative effectiveness of the proposed functional modules is also verified. The proposed method provides reliable technical support for constructing an intelligent dam safety monitoring framework with dynamic matching among data types, identification models, and warning criteria.