Real-time daily reservoir operation by mathematical programing

The use of mathematical programing models for real-time reservoir operation is investigated. Two measures of the quality of the operations designated by the math programing models are used: (1) the total value of actual penalties (surrogate for costs and benefits) incurred by the operation and (2) the closeness of the physical operations designated by the models to the physical operations of the theoretically best operations. Of the five different types of objective functions used in the models, two were clearly superior to the others. One of these requires trial and error fitting of a multiparameter objective function; the other does not. The model which does not require trial and error parameter fitting may be used to operate a reservoir system either when correct benefit and loss functions defining the effects of operation are available or when only descriptions (in the form of cumulative distribution functions) of desirable physical operations are available.