Echo State Network (ESN) is an effective replacement for recurrent neural networks as a reservoir computing model. Similar to deep neural networks, adding hierarchical structure to an ESN is an effective way to improve the network efficiency. The paper examines the influence of the reservoir hierarchical structure on the network efficiency depending on the number of subreservoirs and the number of nodes in the reservoir, and also considers three main hyperparameters that influence the reservoir efficiency: leakage rate, spectral radius and input scale factor. To test the network efficiency, Mackey-Glass time series and power transformer data were selected as experimental data. Experimental results show that the reservoir hierarchical structure improves the prediction accuracy and reduces the computation time for the ESN model.