Ultra-Short-Term Forecasting Method for Photovoltaic Power Based on Singular Spectrum Decomposition and Double Attention Mechanism

Xue Dong; Shengxiao Zhao; Yanyan Lu; Xiaofeng Chen; Yan Zhao; Lei Liu

doi:10.3788/LOP213335

Laser & Optoelectronics Progress, Volume. 60, Issue 5, 0525001(2023)

Ultra-Short-Term Forecasting Method for Photovoltaic Power Based on Singular Spectrum Decomposition and Double Attention Mechanism

Xue Dong^1,2,3, Shengxiao Zhao^1,2, Yanyan Lu^1,2, Xiaofeng Chen^1,2, Yan Zhao^1,2, and Lei Liu^3、*

Author Affiliations

¹Key Laboratory of Far-Shore Wind Power Technology of Zhejiang Province, Hangzhou 311122, Zhejiang, China

²Power China Huadong Engineering Corporation Limited, Hangzhou 311122, Zhejiang, China

³School of Engineering Science, University of Science and Technology of China, Hefei 230026, Anhui, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Structure of feature-temporal attention (FT-Attention)

Download full size

Fig. 2. Flow chart of ultra-short-term photovoltaic power forecasting method

Download full size

Fig. 3. SSD results of original signals

Download full size

Fig. 4. Forecasting results obtained by ablation models

Download full size

Fig. 5. Forecasting results obtained by different methods

Download full size

Fig. 6. Forecasting results under different weather types

Download full size

Table 1. Pearson coefficients between time-delayed features and photovoltaic power
View table
Table 1. Pearson coefficients between time-delayed features and photovoltaic power
Variable t-1 t-2 t-3 t-4 t-5 t-6 t-7
Historical power 0.982 0.961 0.933 0.899 0.858 0.812 0.761
Solar irradiance 0.969 0.947 0.920 0.887 0.847 0.802 0.752

Table 2. Performance comparison of ablation experiments
View table
Table 2. Performance comparison of ablation experiments
Model NMSE NRMSE MAPE
BiGRU 3.70% 5.80% 10.62%
SSD+BiGRU 3.01% 5.22% 9.23%
Proposed 2.57% 4.82% 8.14%

Table 3. Performance comparison with other methods
View table
Table 3. Performance comparison with other methods
Method NMAE NRMSE MAPE $R^{2}$ Pearson
SVR 4.67% 6.03% 10.68% 0.949 0.979
BPNN 4.15% 6.31% 12.81% 0.944 0.972
LSTM 4.06% 6.24% 11.68% 0.946 0.972
GRU 3.89% 6.03% 11.55% 0.949 0.974
Proposed 2.57% 4.82% 8.14% 0.968 0.984

Table 4. Performance comparison results in different seasons

View table

Table 4. Performance comparison results in different seasons

Season	Method	NMAE	NRMSE	MAPE	$R^{2}$	Pearson
Spring	SVR	4.91	6.24	11.13	0.948	0.978
	BPNN	4.37	6.48	12.79	0.943	0.972
	LSTM	4.32	6.45	12.06	0.944	0.972
	GRU	4.20	6.36	12.03	0.945	0.973
	Proposed	2.92	5.28	8.74	0.962	0.981
Summer	SVR	5.29	7.19	13.50	0.907	0.958
	BPNN	5.09	7.93	16.91	0.886	0.942
	LSTM	5.36	8.28	17.05	0.876	0.936
	GRU	4.98	7.83	16.05	0.889	0.944
	Proposed	3.68	6.43	12.58	0.925	0.962
Autumn	SVR	4.71	6.14	11.62	0.944	0.977
	BPNN	4.21	6.48	13.93	0.938	0.969
	LSTM	4.33	6.66	13.68	0.934	0.967
	GRU	4.14	6.42	13.34	0.939	0.970
	Proposed	2.83	5.17	9.90	0.960	0.980
Winter	SVR	4.41	5.21	10.52	0.946	0.985
	BPNN	3.32	5.14	15.52	0.948	0.974
	LSTM	3.22	4.82	12.78	0.954	0.977
	GRU	2.87	4.71	12.88	0.956	0.978
	Proposed	1.84	3.49	8.79	0.976	0.988

Table 5. Performance comparison under different weather types

View table

Table 5. Performance comparison under different weather types

Method	Sunny day			Rainy day			Abrupt change
Method	NMAE	NRMSE	MAPE	NMAE	NRMSE	MAPE	NMAE	NRMSE	MAPE
SVR	3.91%	5.03%	7.61%	4.67%	5.15%	19.31%	5.58%	7.51%	15.20%
BPNN	3.44%	5.38%	8.60%	2.61%	3.90%	34.36%	5.50%	8.26%	18.74%
LSTM	3.03%	4.73%	7.01%	3.01%	4.30%	31.05%	5.63%	8.47%	18.51%
GRU	2.99%	4.76%	6.95%	2.42%	3.81%	31.44%	5.33%	8.06%	17.80%
Proposed	1.59%	3.50%	3.81%	1.66%	2.78%	20.86%	3.95%	6.69%	14.18%

Tools

Get Citation

Copy Citation Text

Xue Dong, Shengxiao Zhao, Yanyan Lu, Xiaofeng Chen, Yan Zhao, Lei Liu. Ultra-Short-Term Forecasting Method for Photovoltaic Power Based on Singular Spectrum Decomposition and Double Attention Mechanism[J]. Laser & Optoelectronics Progress, 2023, 60(5): 0525001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites