Author(s)

XueYuan Zhu^*, JiaXin Peng

Affiliation(s)

School of Transportation, Changsha University of Science & Technology, Changsha 410114, Hunan, China.

Corresponding Author

XueYuan Zhu

ABSTRACT

In high-frequency data environments, traditional time-series forecasting methods generally face two major challenges. First, the structures of these models are too simple to capture both the long-term trends and short-term disturbances. Second, the forecasting granularity is too coarse to meet the refined requirements for real-time dynamic decision-making. To address these issues, this study proposes a dual-channel fusion forecasting framework, the Dual-Resolution Adaptive Forecasting Topology (DRAFT) architecture. The architecture comprises two modules: a trend modeling module and a disturbance modeling module. The modules are responsible for processing the linear trend components and nonlinear fluctuation signals in the time series data. They achieved adaptive integration of the prediction results using a lightweight fusion mechanism. Experiments on real-world datasets demonstrated that the DRAFT architecture significantly outperformed traditional single-model approaches in terms of metrics such as mean squared error (MSE) and mean absolute error (MAE), with error reductions exceeding 54.05% in certain scenarios. Furthermore, DRAFT possesses the capacity to refine the prediction output granularity to the 10-minute level, thereby providing more actionable prediction information for high-timeliness scenarios. This study establishes a new paradigm for the precise prediction of complex time-series data and provides theoretical and practical references for the construction of modular prediction systems.

KEYWORDS

Multi-granularity prediction; Time series modeling; Model fusion; Predictive granularity refinement

CITE THIS PAPER

XueYuan Zhu, JiaXin Peng. Multi-granularity time series forecasting methods based on dual-channel fusion. World Journal of Information Technology. 2025, 3(4): 69-79. DOI: https://doi.org/10.61784/wjit3058.

REFERENCES

[1] Spulbar C, Ene C C. Predictive Analytics in Finance Using the Arima Model. Application for Bucharest Stock Exchange Financial Companies Closing Prices. Studies in Business and Economics, 2024, 19(3): 30-49.

[2] Chen S, Lin R, Zeng W. Short-Term Load Forecasting Method Based on ARIMA and LSTM. 2022 IEEE 22nd International Conference on Communication Technology (ICCT), , Nanjing, China. 2022, 1913-1917. DOI: 10.1109/ICCT56141.2022.10073051.

[3] Sattarzadeh A R, Kutadinata R J, Pathirana P N, et al. A novel hybrid deep learning model with ARIMA Conv-LSTM networks and shuffle attention layer for short-term traffic flow prediction. Transportmetrica A: Transport Science, 2025, 21(1). DOI: https://doi.org/10.1080/23249935.2023.2236724.

[4] Wu D. Time Series hybrid Prediction Model Based on Deep Learning ARIMA-LSTM. 2024 IEEE 4th International Conference on Information Technology, Big Data and Artificial Intelligence (ICIBA), , Chongqing, China. 2024, 1656-1660. DOI: 10.1109/ICIBA62489.2024.10868045.

[5] Fattoh I E, Marwa E M I, Mousa F A. Unveiling market dynamics: a machine and deep learning approach to Egyptian stock prediction. Future Business Journal, 2025, 11(1): 1-12.

[6] Bichescu B, Polak G G. Time series modeling and forecasting by mathematical Programming. Computers & Operations Research, 2023, 151: 106079. DOI: https://doi.org/10.1016/j.cor.2022.106079.

[7] Li Y, Liu X, Li X, et al. Interruption time series analysis using autoregressive integrated moving average model: Evaluating the impact of COVID-19 on the epidemic trend of gonorrhea in China. BMC Public Health, 2023, 23(1): 2073. DOI: 10.1186/s12889-023-16953-5.

[8] Ma L, Zhao Y, Wang B, et al. A Multistep Sequence-to-Sequence Model With Attention LSTM Neural Networks for Industrial Soft Sensor Application. IEEE Sensors Journal, 2023, 23(10): 10801-10813.

[9] Yin H, Zhang X, Wang F, et al. Rainfall-runoff modeling using LSTM-based multi-state-vector sequence-to-sequence Model. Journal of Hydrology, 2021, 598: 126378.

[10] Lu W, Zhang Y, Li P, et al. Mul-DesLSTM: An integrative multi-time granularity deep learning prediction method for urban rail transit short-term passenger Flow. Engineering Applications of Artificial Intelligence, 2023, 125: 106741.

[11] He Z, Xu L, Yu J, et al. Dynamic Multi-fusion spatio-temporal graph neural network for multivariate time series Forecasting. Expert Systems with Applications, 2024, 241: 122729.