OPERATING RULES FOR MULTIRESERVOIR SYSTEMS

Multireservoir operating policies are usually defined by rules that sp ecify either individual reservoir desired (target) storage volumes or desired (target) releases based on the time of year and the existing t otal storage volume in all reservoirs. This paper focuses on the use o f genetic search algorithms to derive these multireservoir operating p olicies. The genetic algorithms use real-valued vectors containing inf ormation needed to define both system release and individual reservoir storage volume targets as functions of total storage in each of multi ple within-year periods. Elitism, arithmetic crossover, mutation, and ''en bloc'' replacement are used in the algorithms to generate success ive sets of possible operating policies. Each policy is then evaluated using simulation to compute a performance index for a given flow seri es. The better performing policies are then used as a basis for genera ting new sets of possible policies. The process of improved policy gen eration and evaluation is repeated until no further improvement in per formance is obtained. The proposed algorithm is applied to example res ervoir systems used for water supply and hydropower.