Abstract: To address the issue of missing correlation features caused by traditional single-point time series prediction models failing to capture spatial relationships among devices in digital twin hydraulic monitoring perception networks, as well as the uncertainty issues arising from the strong subjectivity in model structure and parameter design, this paper proposes a multi-parameter time series prediction model for monitoring perception networks based on Bayesian Optimization & Hyperband (BOHB), self-learning graph structures, and Bidirectional Long Short-Term Memory (BiLSTM) networks. First, a self-learning graph structure is generated to extract spatial features of the perception network using graph neural networks. Second, Bidirectional Long Short-Term Memory networks are employed to extract temporal features. Further, the BOHB method is utilized to optimize hyperparameters and improve prediction accuracy. Finally, proactive predictions of future states of the monitoring perception network are conducted. It has been verified that compared with various prediction models, the proposed model has optimization rates of more than 4.35%, 33.14%, 20.47%, 9.09% and 15.03% in R2, RMSE, MAE, MAPE and RMSRE, with higher accuracy and stronger generalization ability, and has significant performance advantages.