The paper introduces a novel model called Multi-Granularity Time Series Diffusion (MG-TSD) to improve the performance of time series forecasting using diffusion probabilistic models. The key idea is to leverage the inherent granularity levels within the data as targets at intermediate diffusion steps to guide the learning process, reducing instability and improving prediction accuracy. The model consists of three main components: a Multi-granularity Data Generator, a Temporal Process Module, and a Guided Diffusion Process Module. The Guided Diffusion Process Module uses a heuristic loss function that incorporates coarse-grained data to guide the denoising process, ensuring that the intermediate latent states preserve the underlying time series structure. Extensive experiments on real-world datasets demonstrate that the MG-TSD model outperforms existing methods in terms of CRPS$_{\text{sum}}$, NMAE$_{\text{sum}}$, and NRMSE$_{\text{sum}}$. The paper also includes an ablation study to validate the effectiveness of the share ratio selection and the impact of the number of granularity levels.The paper introduces a novel model called Multi-Granularity Time Series Diffusion (MG-TSD) to improve the performance of time series forecasting using diffusion probabilistic models. The key idea is to leverage the inherent granularity levels within the data as targets at intermediate diffusion steps to guide the learning process, reducing instability and improving prediction accuracy. The model consists of three main components: a Multi-granularity Data Generator, a Temporal Process Module, and a Guided Diffusion Process Module. The Guided Diffusion Process Module uses a heuristic loss function that incorporates coarse-grained data to guide the denoising process, ensuring that the intermediate latent states preserve the underlying time series structure. Extensive experiments on real-world datasets demonstrate that the MG-TSD model outperforms existing methods in terms of CRPS$_{\text{sum}}$, NMAE$_{\text{sum}}$, and NRMSE$_{\text{sum}}$. The paper also includes an ablation study to validate the effectiveness of the share ratio selection and the impact of the number of granularity levels.