Abnormal changes in groundwater levels serve as crucial indicators of seismic precursors. Prior to earthquakes, groundwater levels typically exhibit varying degrees of anomaly, manifesting as sudden rises or falls that may persist for extended periods. To accurately identify seismic precursor anomalies, we categorize groundwater level data into Seismic Active (SA) and Non-Seismic Active (non-SA) periods, providing a basis for segmenting the dataset. We propose a novel network architecture termed Residual Multi-scale TCN Sparse Expert Network-iTransformer (RMMoTCN-iTransformer). This model integrates the strengths of Residual Multi-scale TCN Sparse Expert Network (RMMoTCN) and iTransformer, enabling effective capture of multi-scale local features and global dependencies in groundwater levels. RMMoTCN achieves sTable training through its residual structure and sparse multi-scale TCN expert network, enabling flexible modeling of complex temporal features and learning trends across different time step scales. Concurrently, iTransformer enhances long-sequence prediction performance via an improved self-attention mechanism. Additionally, we incorporate a Wavelet Noise Reduction (WNR) method to further boost the model's robustness and prediction accuracy. Experimental results demonstrate the model's robust capabilities in groundwater level prediction and seismic precursor anomaly detection. To enhance anomaly detection accuracy, this study employs an exponentially weighted moving average (EWMA) control chart to precisely identify anomaly onset times. Earthquake validation confirms the model's effective identification of groundwater level anomalies under diverse geological conditions, providing sufficient advance warning time for earthquake preparedness. This validates its broad adaptability and practical utility. This research contributes scientific innovation and practical value to seismic precursor analysis, offering novel technical support and analytical methods for earthquake early warning technology development and disaster prevention efforts.