Melody Search Algorithm Using Online Evolving Artificial Neural Network Coupled with Singular Spectrum Analysis for Multireservoir System Management

This paper introduces a novel application of online evolving neural networks (ENNs) trained by melody search algorithm (MSA) and singular spectrum analysis (SSA) preprocessing technique in finding optimum operating policies of Tehran water resources system with multiple reservoirs and multiple purposes. Harmony search (HS) algorithm is also used for comparison purposes. In addition, in order to better evaluate the proposed method, different combinations of artificial neural network (ANN) models are developed. It was noticed that the application of SSA preprocessing technique shows substantial improvement in the ANN model performance. In fact, application of singular spectrum analysis on the input data eliminates the irrelevant features and inverts the remaining components to yield a filtered time series. Moreover, all the results indicate the superiority of MSA over HS algorithm in all combinations of ANN models. It was also noticed that as the model becomes more complex, the MSA shows more improved and distinct results as compared to HS algorithm. Thus, it can be stated that the melody search is more efficient and a suitable surrogate for the harmony search algorithm in pursuing the objectives in complicated problems.