Aiming at the problem that shortwave direction-finding errors exhibit intertwined characteristics of multi-periodicity, non-stationarity and long-range dependencies, which in turn limits the prediction accuracy, a shortwave direction-finding error prediction model based on time-series fusion network (TSF-Net) is proposed. Firstly, in view of the coexistence of random and consecutive missing values in the raw data, a hierarchical preprocessing strategy is designed, combining linear interpolation and periodic-similarity-based imputation to construct well-aligned and regular high-quality input sequences. Secondly, a parallel dual-stream feature extraction architecture is adopted, where the TimesNet branch targetedly extracts the multi-periodic evolution patterns of direction-finding errors, and the Transformer branch targeted captures the global long-range dependencies and non-stationary time-varying characteristics. Finally, an adaptive weighted fusion module is constructed to realize complementary fusion of dual-branch features through a static and dynamic dual-path weighting mechanism. Simulation results show that the proposed model achieves a 6.2% to 8.7% relative improvement in prediction accuracy compared with the second-best TimesNet baseline model.
ATHANASOPOULOSG, HYNDMANR J, KOURENTZESN, et al. Forecast reconciliation: a review[J]. Monash Econometrics and Business Statistics Working Papers, 2023:40(2). DOI:10.1016/j.ijforecast.2023.10.010 .
[3]
IKEDAS, ITOT, HASEBEK, et al. A comparative study on long-term cryptocurrency price prediction using LSTM, GRU, and Bi-LSTM[J]. Journal of Robotics, Networking and Artificial Life, 2025,11(1):49-53.
[4]
LIC, ZHENGX W, CHENC Q, et al. Tiny data is sufficient: a generalizable CNN architecture for temporal domain long sequence identification[J]. IEEE Transactions on Neural Networks and Learning Systems, 2025,36(7):12977-12990.
[5]
LIUS M. LTR-Net: a deep learning-based approach for financial data prediction and risk evaluation in enterprises[J]. PLoS One, 2025,20(8):No.e0328013.
[6]
ZHOUJ Q, YANGX H, RENZ. SCConv-denoising diffusion probabilistic model anomaly detection based on TimesNet[J]. Electronics, 2025,14(4):No.746.
[7]
ONUR CAKIROGLUM, KURBANH, BUXTONE, et al. A novel discrete time series representation with de bruijn graphs for enhanced forecasting using TimesNet[J]. IEEE Access, 2025,13:123182-123198.
[8]
LINR H, CHENS, WUB D, et al. WaveTimes: short-term load forecast based on wavelet decomposition and improved TimesNet[J]. Chinese Journal of Electronics, 2025,34(6):1911-1923.
[9]
TINGC C, WUK T, LINH C, et al. MixModel: a hybrid TimesNet-informer architecture with 11-dimensional time features for enhanced traffic flow forecasting[J]. Mathematics, 2025,13(19):No.3191.
[10]
BYUNG, YANGH, RAZAS M, et al. Generative spatiotemporal image exploitation for datacenter traffic prediction[J]. Computer Networks, 2024,254:No.110755.
[11]
MAHMOUDA, MOHAMMEDA. Leveraging hybrid deep learning models for enhanced multivariate time series forecasting[J]. Neural Processing Letters,2024,56(5):No.223.
[12]
SATHEA M, SUPRAJAR, THOMASA A. Enhancing agricultural sustainability: time series forecasting with ICEEMDAN-VMD-GRU for economic-resilience[J]. Results in Engineering, 2025,27:No.106423.
HASTIET, TIBSHIRANIR, FRIEDMANJ. The elements of statistical learning: data mining, inference,and prediction[M]. 2nd ed. New York,USA: Springer, 2009:1-745.