Premium
Fuzzy Learning Decomposition for the Scheduling of Hydroelectric Power Systems
Author(s) -
Saad M.,
Bigras P.,
Turgeon A.,
Duquette R.
Publication year - 1996
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1029/95wr02971
Subject(s) - multivariable calculus , artificial neural network , principal component analysis , fuzzy logic , mathematical optimization , cluster analysis , computer science , nonlinear system , piecewise , scheduling (production processes) , artificial intelligence , mathematics , engineering , control engineering , mathematical analysis , physics , quantum mechanics
This paper presents a nonlinear multivariable fitting model to decompose the optimal policies obtained by dynamic programming of a unique aggregated reservoir. The nonlinear functions are generated using radial basis functions (RBF) neural networks. In this method the potential energy of all the reservoirs in the hydropower system is added to form one equivalent reservoir. The operating policy of the equivalent reservoir is determined by stochastic dynamic programming, and finally the operating rules of each reservoir are determined using RBF neural networks. To improve the multivariable representation of the data, a series of piecewise RBF neural networks is determined using clustering analysis. A fuzzy clustering approach is used to determine the RBF's parameters. This approach has the advantages of being fast and simple to implement with well‐established convergence properties. It also has a good representation of the covariance matrix, since all the data belong to all the classes at the same time with different membership grades. A comparison with the back propagation learning and principal components techniques is also reported for Québec's La Grande River installations. As a result, the proposed approach gives satisfactory operating rules compared with principal component analysis, and the CPU time is reduced by a factor of 15 to 20 compared with the back propagation technique.