Short-Term Load Forecasting Based on Frequency Domain Decomposition and Deep Learning

Table 1

The forecasting error calculation results of four methods.

Time sampling point (15 min)	HFDA-LSTM		EMD-LSTM		LSTM		RNN
	MAPE (%)	RMSE (MW)	MAPE (%)	RMSE (MW)	MAPE (%)	RMSE (MW)	MAPE (%)	RMSE (MW)
1–96	0.95	7.38	2.23	14.93	2.36	21.36	3.01	22.23
97–192	0.92	6.27	2.09	13.72	2.55	22.31	3.24	23.18
193–288	0.82	5.68	2.26	14.02	2.13	20.06	2.56	19.73
289–384	0.80	5.88	2.24	15.12	2.40	21.47	2.85	22.97
385–480	0.88	6.08	2.06	13.59	2.11	19.98	2.78	21.56
481–576	1.55	8.47	1.77	10.99	3.77	29.66	3.22	21.37
577–672	1.35	7.81	2.43	13.87	3.56	28.30	3.54	24.89
1–672	1.04	6.87	2.15	13.81	2.70	23.31	3.03	22.33