In recent years, time series analysis has made significant progress in key areas such as anomaly detection and classification. Among these tasks, multivariate time series forecasting has attracted increasing attention due to its wide applications in weather prediction, traffic flow analysis, and industrial maintenance. The primary challenge lies in effectively capturing complex temporal dependencies and inter-variable correlations. Temporal patterns can be intricate, while variable relationships are dynamic—some providing valuable information, others introducing noise. To tackle these issues, we propose TVCorNet, a Time-Variable Correlation Learning Enhancement Network that incorporates a Time-Lag Learning Module and a Variable-Biased Correlation module. The Time Lag module introduces multiple lag steps and multi-scale feature extractors to capture latent temporal structures. The Correlation module employs a channel masking mechanism that leverages both time- and frequency-domain cues to assess inter-variable dependencies, selectively attending to highly correlated variables. Experiments on eight real-world datasets demonstrate that TVCorNet consistently outperforms existing models in both accuracy and robustness, highlighting its effectiveness and practical value in complex multivariate forecasting tasks.