Dimension reduction of decision variables for multireservoir operation: A spectral optimization model

Abstract

Optimizing the operation of a multireservoir system is challenging due to the high dimension of the decision variables that lead to a large and complex search space. A spectral optimization model (SOM), which transforms the decision variables from time domain to frequency domain, is proposed to reduce the dimensionality. The SOM couples a spectral dimensionality-reduction method called KarhunenLoeve (KL) expansion within the routine of Nondominated Sorting Genetic Algorithm (NSGA-II). The KL expansion is used to represent the decision variables as a series of terms that are deterministic orthogonal functions with undetermined coefficients. The KL expansion can be truncated into fewer significant terms, and consequently, fewer coefficients by a predetermined number. During optimization, operators of the NSGA-II (e.g., crossover) are conducted only on the coefficients of the KL expansion rather than the large number of decision variables, significantly reducing the search space. The SOM is applied to the short-term operation of a 10-reservoir system in the Columbia River of the United States. Two scenarios are considered herein, the first with 140 decision variables and the second with 3360 decision variables. The hypervolume index is used to evaluate the optimization performance in terms of convergence and diversity. The evaluation of optimization performance is conducted for both conventional optimization model (i.e., NSGA-II without KL) and the SOM with different number of KL terms. The results show that the number of decision variables can be greatly reduced in the SOM to achieve a similar or better performance compared to the conventional optimization model. For the scenario with 140 decision variables, the optimal performance of the SOM model is found with six KL terms. For the scenario with 3360 decision variables, the optimal performance of the SOM model is obtained with 11 KL terms.

Application of NSGA-II and Improved Risk Decision Method for Integrated Water Resources Management of Malian River Basin

“…Subsequent research basically used model results to support decision/policy-making, optimize the allocation of water resources, and efficiently collect and utilize water resources [8][9][10][11][12][13][14][15][16][17][18][19]. In terms of the large-scale multi-objective model and the Non-dominated Sorting Genetic Algorithm-II (NSGA-II) method, most of the research is focused on water resource management [20][21][22][23][24][25][26][27][28][29][30][31][32][33][34]. Unfortunately, nearly all the risk decision analyses on water resource systems have primarily focused on flood control systems and reservoir operation [35].…”

Dynamic Management of Water Storage for Flood Control in a Wetland System: A Case Study in Texas

“…A Pareto front (or Pareto frontier) is a set of nondominated optimal solutions where no objective can be improved without sacrificing at least one other objective (e.g., [40]). On the other hand, a solution is referred to as dominated by a second solution if, and only if, the second solution is equally good or better than the first solution with respect to all objectives (e.g., [40]). Each point on a Pareto front is associated with a cost and the percentage of water released from the wetlands.…”

Aggregation Methods for Modelling Hydropower and Its Implications for a Highly Decarbonised Energy System in Europe

“…Depending on the area and focus of the analysis, a large number of hydro plants and reservoirs can severely amplify the market model complexity, e.g., in the Nordic and Alpine regions in Europe. This is particularly the case for cascaded multireservoir systems which are subject to uncertain natural inflows and require simultaneous operational decisions for each reservoir in the system [1].…”