This research aims to enhance the imputation accuracy of missing transport trip data, with the goal of reducing sensor installation costs and improving the downstream utility of time series data. Imputing missing values in multi-variate time series data presents significant challenges due to complex dependencies. State-of-the-art deep learning methods struggle to address compounding errors, irregular time intervals, and varying sequence lengths. To tackle these issues, we propose an enhanced version of SAlTS (Self-Attention-based imputation for Time Series) called Time-Aware SAlTS (TA-SAlTS). T A-SAlTS incorporates Time-Aware Positional Encoding, Self-Attention, and Padding-Aware Self-Attention techniques, enabling it to handle variable sequence lengths and integrate the intricate temporal aspect into the imputation process. Our results demonstrate that Time-Aware SAlTS outperforms conventional SAlTS methods and, in the best case, achieves a 9.5% reduction in Mean Relative Error and a 5.2% decrease in Mean Square Error, signifying substantial improvements in imputation accuracy.