Renmin Zhujiang (Jun 2025)
Medium- and Long-term Forecast of Inflow of Dahuofang Reservoir Based on CEEMDAN-LSTM Model
Abstract
As the flow process is subject to rain fall, temperature fluctuations, human activity intervention, and other multiple factors of complexity, its law of change shows significant randomness and uncertainty, greatly increasing the difficulty of medium- and long-term forecast and restricting its application effect in production practice. Therefore, how to break through the technical bottleneck of medium- and long-term flow forecast has become a key problem to be solved in the current hydrological science research. In view of this, with Dahuofang Reservoir as the research object, the monthly flow forecast of the reservoir was carried out by using gray correlation analysis and principal component analysis combined with the CEEMDAN-LSTM model. Monthly flow, rainfall and meteorological data from 1961 to 2008 were selected to calibrate the model parameters. The monthly flow data from 2009 to 2020 was used to validate the model. The indicators of determination coefficient, root mean square error and average relative error were applied to evaluate the forecast results. According to the results, when four principal components are selected by using principal component analysis to reduce the dimensionality of the forecast factor sets after the addition of the previous average temperature and maximum temperature data, the CEEMDAN-LSTM model can effectively improve the forecast accuracy and become the optimal model for the monthly flow forecast of Dahuofang Reservoir. Technical support is provided for the formulation of the future medium- and long-term dispatching plan of Dahuofang Reservoir.
