Hydraulic optimization to obtain the location and size of transient protection devises simultaneously is very important in the design of water supply systems and maintenance cost reduction. To prevent water column separation, when source pump rundown occurs after power failure, an air chamber and one-way surge tanks are included and optimized using the fuzzy inference system and the genetic algorithm method. Optimization of initial air volume in the air chamber, when the minimum pressures along the pipeline are acceptable, is maintained using the fuzzy inference system. It’s shown that the design of air chamber can be done by the fuzzy inference system, and designer’s idea can be reflected by the base rule of fuzzy inference system. Static and dynamic behavior proposed by housner, are considered to obtain the optimal ratio of height to diameter in the surge tanks. Smaller ratio results in more stability, but the best choice depends on many conditions. Economics, hydraulics, structural and aesthetic as well as constraints on height, diameter and location are major considerations. Free board and bathtub vortex criteria are also considered. Optimization of one-way surge tanks in the system combined with an air chamber has done with the combination of genetic algorithm and fuzzy inference system. In this method, each chromosome evaluated using a fuzzy decision defined for the objective function after transient analysis. It is shown that determination of the fitness value of each chromosome can be done simply, precisely, and traarently by the fuzzy inference system. In this way, a generalized fitness function that doesn’t change with varying the dimensional characteristics, time and location are obtained. It was observed that certain volume of water is required to handle effectively the water hammer conditions. The surge tanks closer to pump station should have a higher height. Fulfillment of the minimum pressure constraint and critical submergence show the effectiveness of the fuzzy-genetic method. Several refinements are considered to increase the effectiveness of the algorithm. The value of the crossover rate is adjusted dynamically. With normalization of fitness value in the range of [0,1], local optimum solutions are avoided.
