In the era of big data and IoT, we are witnessing a massive data explosion from various online platforms. This presents both challenges and opportunities for managing and extracting potential knowledge from these enormous data resources. Among various data structures, time series is the most common, used widely in numerous applications. Multivariate time-series forecasting, a key time-series data analysis method, is considered challenging due to the need to efficiently manage dependencies between variables and the long length of input sequences. Despite advancements in deep learning forecasting techniques, applying a multi-channel learning approach to complex multivariate time-series data remains an open research problem. To address this, we propose a novel multi-channel transformer-based technique, MCT4AQF. The MCT4AQF model extends the series-aware time-series learning approach with a CNN-based multi-channel learning mechanism, effectively discerning and modeling the complex dependencies across input sequence variables. Experiments on real-world datasets about environmental states have validated MCT4AQF’s superior performance compared to state-of-the-art transformer-based forecasting models.