Hydropower Scheduling Under Missing Power Load Data: Optimizing Energy Efficiency and Navigation Performance

Hydropower stations integrated into the grid system often suffer from imbalance in the dispatch of power generation and shipping benefits due to untimely acquisition of load data. In this study, a dispatch model integrating power load forecasting and multi-objective optimization is constructed and practically applied to Shatuo Hydropower Station in Guizhou Province, China. Compared with existing models, the model can accomplish the optimal dispatch of grid-connected hydropower stations in the absence of power load. The model first uses seven prediction models to predict the power load, and then uses the GA-NSGA-II algorithm (i.e., the improved elite non-dominated sorting genetic algorithm) to dispatch the discharged flow of Shatuo Hydropower Station. At the same time, we set the power load demand value in the power generation objective function as the load value obtained from the prediction and the real load value for experimental comparison, respectively. The results show that the prediction accuracy of the CNN-GRU model (i.e., Convolutional Neural Network-Gate Recursive Unit) is better than the other models, and the values of R-squared and RMSE are 0.991 and 0.026, respectively. The scheduling of the GA-NSGA-II algorithm results in Hypervolume metrics and Spacing metrics better than the NSGA-II algorithm. The difference between scheduling using predicted load values and scheduling using real load values is negligible within 5 (${m}^{3}/s$). The result of optimized scheduling enables Shatuo Hydropower Station to achieve double benefits of power generation and shipping. This method solves the problems of untimely load data collection and inadequate scheduling, and provides an effective way for hydropower station scheduling for power generation and navigation.