Power Prediction of Photovoltaic Generation Based on Improved Temporal Convolutional Network

Guilan Li; Jie Yang; Manguo Zhou

doi:10.3788/LOP202259.0820001

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 8 >Page 0820001 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 8, 0820001 (2022)

Power Prediction of Photovoltaic Generation Based on Improved Temporal Convolutional Network

Guilan Li^1、2, Jie Yang^1、2、*, and Manguo Zhou¹

Author Affiliations

¹School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou , Jiangxi 341000, China

²Jiangxi Key Laboratory of Magnetic Levitation Technology, Ganzhou , Jiangxi 341000, China

show less

DOI: 10.3788/LOP202259.0820001 Cite this Article Set citation alerts

Guilan Li, Jie Yang, Manguo Zhou. Power Prediction of Photovoltaic Generation Based on Improved Temporal Convolutional Network[J]. Laser & Optoelectronics Progress, 2022, 59(8): 0820001 Copy Citation Text

show less

Fig. 1. TCN structure. (a) ResNet unit; (b) causal convolution; (c) dilated convolution

Download full size

Fig. 2. Improved TCN structure

Download full size

Fig. 3. Distance correlation coefficient

Download full size

Fig. 4. Maximum mutual information coefficient

Download full size

Fig. 5. Pearson correlation coefficient

Download full size

Fig. 6. Iterative loss process of different neural network models

Download full size

Fig. 7. Model prediction under different weather. (a) Cloudy; (b) sunny; (c) sunny to cloudy; (d) light rain to cloudy

Download full size

Fig. 8. Error comparison of classical algorithms

Download full size

Parameter	Temperature	Module temperature	Solar irradiance	Voltage	Current	Cloud cover	Humidity	Wind speed	Solar azimuth angle	Incidence angle
PCC	0.3925	0.4610	0.4726	0.4725	0.4725	0.4592	-0.0488	-0.0158	0.1593	-0.2899
DCC	0.6042	0.6687	0.6515	0.6513	0.6571	0.3115	0.6415	0.4527	0.4342	0.4537
MIC	0.5304	0.8579	0.9915	0.9843	0.9894	0.8941	0.9745	0.1477	0.9204	0.9742

Table 1. Analysis of the correlation between photovoltaic power generation and other characteristics

Weather	Model	MAE	RMSE	MAPE	r²
Cloudy	LSTM	384.335	642.671	20.192	0.905
	GRU	345.204	600.387	18.907	0.917
	1D-CNN	457.634	595.407	6.113	0.918
	TCN	290.931	431.454	9.643	0.957
	Proposed	200.885	310.366	4.318	0.979
Sunny	LSTM	302.201	497.337	17.201	0.943
	GRU	277.025	478.866	13.126	0.947
	1D-CNN	314.923	472.902	11.843	0.948
	TCN	232.694	351.677	5.683	0.971
	Proposed	192.175	326.938	3.751	0.976
Sunny to cloudy	LSTM	423.528	714.623	23.112	0.883
	GRU	416.409	642.681	17.058	0.905
	1D-CNN	432.407	714.425	19.201	0.883
	TCN	360.379	602.137	19.091	0.917
	Proposed	338.439	535.395	14.918	0.934
Light rain to cloudy	LSTM	372.363	613.397	20.194	0.913
	GRU	366.87	574.832	15.933	0.924
	1D-CNN	584.657	945.496	27.932	0.795
	TCN	334.873	516.787	10.691	0.939
	Proposed	298.706	479.917	11.999	0.947

Table 2. Model indicators under different weather conditions

Model	MAE	RMSE	MAPE	r²	Time /s
LSTM	301.361	618.115	14.952	0.868	150
LSTM-R	180.943	292.308	3.935	0.979	79
GRU	298.494	595.763	14.751	0.876	114.4
GRU-R	178.457	300.641	3.099	0.98	72.4
1D-CNN	325.815	614.318	16.334	0.866	77.2
1D-CNN-R	265.65	375.654	5.413	0.967	55
TCN	306.436	591.91	14.131	0.878	47.6
TCN-R	164.049	287.174	2.505	0.981	34
Proposed	298.865	585.988	13.985	0.880	42.4
Proposed-R	160.496	277.396	2.402	0.982	30

Table 3. Comparison of different models before and after fusion

Guilan Li, Jie Yang, Manguo Zhou. Power Prediction of Photovoltaic Generation Based on Improved Temporal Convolutional Network[J]. Laser & Optoelectronics Progress, 2022, 59(8): 0820001

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information