Abstract: The year-end stage of a multi-year regulating reservoir is a significant control variable for coordinating current and future benefits. To optimize this control stage in the case of potential uncertainty in future reservoir inflows, this paper constructs a general framework for an expert system. By analyzing its causes and mechanism, we develop a method for generating inference data sources and implementing a procedure of reservoir stage reasoning in an interface machine. Application in a case study of the Tankeng reservoir shows that the year-end stage solution of the expert system is close to the optimal solution with only a small difference, and that other indicators, such as power supply stability and water resource utilization rate, are also at high levels. Thus, our method and expert system are practical and useful in the planning of annual power generation of multi-year regulating reservoirs.

Key words: hydroelectric power, multi-year regulating reservoir, year-end water level, expert system, interface engine