Sales forecasting plays an important role in decision-making in various industries. Especially, apparel products are highly sensitive to seasons and trends, making accurate sales forecasting essential. While many studies have applied data-driven models to forecast apparel sales, many have overlooked key characteristics of apparel sales data, such as patterns and distortions. In this study, we propose a transformer-based long-term time series forecasting model using timestamps and denoised input with DILATE (PatchTSDD). The proposed method is designed to capture the unique characteristics of apparel sales. By using denoising autoencoder, the model smooths input data and effectively reflects underlying sales trends. Additionally, the inclusion of time stamp information enables the model to capture global temporal patterns. To address sudden fluctuations in sales, the DILATE loss function is applied. We present experimental results using actual sales data from a leading domestic fashion retail company. The results demonstrate that our method outperforms traditional time series forecasting methods, with each component of the model contributes to the overall performance improvement.