Photovoltaic power level classification based on Res-MobileCom parallel network

doi:10.3969/j.issn.2097-0706.2026.01.001

Abstract

Abstract:

To enhance the accuracy of photovoltaic power level classification to meet industry requirements， a classification model based on a Res-MobileCom parallel network was proposed. In data preprocessing， denormalized bilinear interpolation was used to maximize the preservation of data features. Subsequently， through parallel training of simplified residual network（ResNet） and efficient convolutional neural networks for mobile vision（MobileNet）， their outputs were jointly fed into the channel estimation-signal detection network（ComNet） for further data feature extraction， ultimately obtaining the classification results. The experimental results demonstrated that compared to common deep learning models， the Res-MobileCom model retained the feature extraction capability and lightweight nature of ResNet and MobileNet， exhibiting good balance and generalization ability. By using the denormalized bilinear interpolation method and the ComNet for further data feature extraction， the model accuracy improved by more than 10 percentage points， providing a novel approach and idea for improving the accuracy of photovoltaic power level classification models. Future work will focus on stability optimization， cross-task validation， and engineering deployment.

Key words: photovoltaic power level classification, deep learning, residual network, mobile network, ComNet, lightweight, parallel network

CLC Number:

TK51：TP393

YIN Linfei, ZHOU Yanggang. Photovoltaic power level classification based on Res-MobileCom parallel network[J]. Integrated Intelligent Energy, 2026, 48(1): 1-12.

Figures/Tables 30

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Fig.6

Fig.7

Fig8

Fig.9

Fig.10

Fig.11

Fig.12

Table 1

Fig.13

Fig.14

Fig.15

Fig.16

Fig.17

Table 2

Table 3

Table 4

Table 5

Fig.18

Table 6

Table 7

Fig.19

Table 8

Table 9

Table 10

Ablation test results of Function module %

Function模块形式	测试集准确率
$1 0.1 + x$	51
$1 0.1 + x + R$	63

Table 10

Table 11

References 30

[1]	何兴, 王旭, 许野, 等. 光伏产业与环境支撑体系的耦合协调研究[J]. 太阳能学报, 2023, 44(9): 194-203. doi: 10.19912/j.0254-0096.tynxb.2022-0724
	HE Xing, WANG Xu, XU Ye, et al. Study on coupling coordination between photovoltaic industry and environmental support system[J]. Acta Energiae Solaris Sinica, 2023, 44(9): 194-203. doi: 10.19912/j.0254-0096.tynxb.2022-0724
[2]	叶幸, 邱辰, 艾琳, 等. “双碳”目标下可再生能源绿证与碳交易衔接机制的思考[J]. 综合智慧能源, 2025, 47(5): 12-20. doi: 10.3969/j.issn.2097-0706.2025.05.002
	YE Xing, QIU Chen, AI Lin, et al. Connection mechanism between green certificates and carbon trading mechanism under the "double carbon" target[J]. Integrated Intelligent Energy, 2025, 47(5): 12-20. doi: 10.3969/j.issn.2097-0706.2025.05.002
[3]	王剑斌, 傅金波, 陈博. 基于强化学习的多模型融合光伏发电功率预测方法[J]. 太阳能学报, 2024, 45(6): 382-388.
	WANG Jianbin, FU Jinbo, CHEN Bo. Multi-model fusion photovoltaic power generation prediction method based on reinforcement learning[J]. Acta Energiae Solaris Sinica, 2024, 45(6): 382-388.
[4]	江泽涛. 深度学习在光伏系统短期发电功率预测的应用研究[D]. 北京: 华北电力大学, 2024.
	JIANG Zetao. Research on the application of deep learning in short-term generation power prediction of photovoltaic system[D]. Beijing: North China Electric Power University, 2024.
[5]	张研, 景超, 王慧民, 等. 基于周期注意力机制的中长期光伏发电功率预测[J]. 太阳能学报, 2024, 45(10): 298-308.
	ZHANG Yan, JING Chao, WANG Huimin, et al. Medium and long term photovoltaic power prediction based on periodic attention mechanism[J]. Acta Energiae Solaris Sinica, 2024, 45(10): 298-308.
[6]	彭曙蓉, 陈慧霞, 孙万通, 等. 基于改进LSTM的光伏发电功率预测方法研究[J]. 太阳能学报, 2024, 45(11): 296-302.
	PENG Shurong, CHEN Huixia, SUN Wantong, et al. Research on photovoitaic power prediction method based on improved LSTM[J]. Acta Energiae Solaris Sinica, 2024, 45(11): 296-302.
[7]	温廷新, 郭晓赛. 基于特征组合的ECA-TCN光伏发电功率预测模型[J]. 太阳能学报, 2024, 45(12): 94-100.
	WEN Tingxin, GUO Xiaosai. ECA-TCN photovoltaic power prediction model based on feature combination[J]. Acta Energiae Solaris Sinica, 2024, 45(12): 94-100.
[8]	王玲芝, 李晨阳, 刘婧, 等. 基于GRO-SSA-LSTM的短期光伏发电功率预测[J]. 太阳能学报, 2025, 46(2): 401-409.
	WANG Lingzhi, LI Chenyang, LIU Jing, et al. Short term photovoltaic power prediction based on GRO-SSA-LSTM[J]. Acta Energiae Solaris Sinica, 2025, 46(2): 401-409.
[9]	冯建铭, 希望·阿不都瓦依提, 蔺红. 基于聚类SABO-VMD和组合神经网络的短期光伏发电功率预测[J]. 太阳能学报, 2025, 46(2): 357-366.
	FENG Jianming, ABUDUWAYITI Xiwang, LIN Hong. Short-term PV power forecasting based on clustering SABO-VMD and ensemble neural networks[J]. Acta Energiae Solaris Sinica, 2025, 46(2): 357-366.
[10]	李超, 涂腾, 彭勋辉, 等. 融合DT-BO-GRU的中长期光伏功率滚动预测模型[J]. 太阳能学报, 2025, 46(5): 275-284.
	LI Chao, TU Teng, PENG Xunhui, et al. Medium-and long-term rolling prediction model of pv power fused with DT-BO-GRU[J]. Acta Energiae Solaris Sinica, 2025, 46(5): 275-284.
[11]	YILDIZ C, ACIKGOZ H, KORKMAZ D, et al. An improved residual-based convolutional neural network for very short-term wind power forecasting[J]. Energy Conversion and Management, 2021, 228:113731. doi: 10.1016/j.enconman.2020.113731
[12]	ANG P Z, CHEN D, HOU Y S. Entropy method combined with extreme learning machine method for the short-term photovoltaic power generation forecasting[J]. Chaos Solitons & Fractals, 2015,89: 243-248.
[13]	DASH D R, DASH P K, RANJEETA B. Short term solar power forecasting using hybrid minimum variance expanded RVFLN and Sin-Cosine Levy Flight PSO algorithm[J]. Renewable Energy, 2021, 174: 513-537. doi: 10.1016/j.renene.2021.04.088
[14]	DUAN Z H, LU M, MA J, et al. QARV: Quantization-aware ResNet VAE for lossy image compression[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(1): 436-450. doi: 10.1109/TPAMI.2023.3322904
[15]	HAN Y M, CAO LIAN, GENG Z Q, et al. Novel economy and carbon emissions prediction model of different countries or regions in the world for energy optimization using improved residual neural network[J]. Science of the Total Environment, 2023, 860:160410. doi: 10.1016/j.scitotenv.2022.160410
[16]	MICHAEL R, SAMANTHA H, MEAGHAN S, et al. An AI based smart-phone system for asbestos identification[J]. Journal of Hazardous Materials, 2024,463:132853.
[17]	JIANG T Y, LI L Y, SAMALI B, et al. Lightweight object detection network for multi-damage recognition of concrete bridges in complex environments[J]. Computer-Aided Civil and Infrastructure Engineering, 2024, 39(23): 3646-3665. doi: 10.1111/mice.v39.23
[18]	QIAN S Y, NING C R, HU Y P. MobileNetV3 for image classification[C]// IEEE International Conference on Big Data, 2021: 490-497.
[19]	CUI Y, XIA J B, WANG Z T, et al. Lightweight spectral-spatial attention network for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60: 5510114.
[20]	PRASAD S B R, CHANDANA B S. MobileNetV3: A deep learning technique for human face expressions identification[J]. International Journal of Information Technology, 2023, 15(6): 3229-3243. doi: 10.1007/s41870-023-01380-x
[21]	XIAO P C, XU W G, ZHANG Y, et al. Research on scrap classification and rating method based on SE attention mechanism[J]. Chinese Journal of Engineering, 2023, 45(8):1342-1352.
[22]	LI Z Y, DENG X Y, ZHANG L M. et al. Dual-channel CNN structure based on swish activation function[J]. Computer & Digital Engineering, 2020, 48(6): 1413-1416.
[23]	JIANG A B, WANG W W. Research on optimization of ReLU activation function[J]. Transducer and Microsystem Technologies, 2018, 37(2):50-52.
[24]	HOWARD A, SANDLER M, CHU G, et al. Searching for MobileNetV3[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019: 1314-1324.
[25]	刘岱, 李晨辉. 基于Inception-BiLSTM的航空电缆电弧故障检测[J]. 科学技术与工程, 2025, 25(14):6100-6108.
	LIU Dai, LI Chenhui. Aviation cable arc fault detection based on Inception-BiLSTM[J]. Science Technology and Engineering, 2025, 25(14): 6100-6108.
[26]	毕贵红, 孔凡文, 黄泽, 等. 基于Inception和注意力机制的双分支日前电价预测[J]. 电力系统自动化, 2025, 49(5): 128-144.
	BI Guihong, KONG Fanwen, HUANG Ze, et al. Double-branch day-ahead electricity price forecasting based on Inception and attention mechanism[J]. Automation of Electric Power Systems, 2025, 49(5): 128-144.
[27]	MAN W Y, YANG G Q, FENG S R. Joint selfattention-SVM DDoS attack detection and defense mechanism based on self-attention mechanism and SVM classification for SDN networks[C]// IEICE Transactions on Fundamentals of Electronics Communications and Computer Sciences, 2023:881-889.
[28]	SUN N, XU W, LIU J X, et al. The multimodal scene recognition method based on self-attention and distillation[J]. IEEE Multimedia, 2024:31(4):25-36.
[29]	杨澜倩, 郭锦敏, 田慧丽, 等. 基于CNN-LSTM-Self attention的园区负荷多尺度预测研究[J]. 综合智慧能源, 2025, 47(2): 79-87. doi: 10.3969/j.issn.2097-0706.2025.02.008
	YANG Lanqian, GUO Jinmin, TIAN Huili, et al. Research on multi-scale load prediction in parks based on CNN-LSTM-Self attention[J]. Integrated Intelligent Energy, 2025, 47(2): 79-87. doi: 10.3969/j.issn.2097-0706.2025.02.008
[30]	FANTINI D, SILVA R, SIQUEIRA M, et al. Wind speed short-term prediction using recurrent neural network GRU model and stationary wavelet transform GRU hybrid model[J]. Energy Conversion and Management, 2024,308:118333.

功率等级	平均功率	功率等级	平均功率
f1	239.220	f6	3 750.000
f2	396.000	f7	2 000.000
f3	397.870	f8	500.000
f4	332.395	f9	6 000.000
f5	201.140

模型	Letterbox	双线性插值法	去归一化双线性插值法
InceptionV1	0.111 0	0.111 0	0.099 9
InceptionV2	0.111 1	0.111 0	0.113 7
Inception-ResNetV1	0.111 0	0.111 0	0.115 1
Inception-ResNetV2	0.111 0	0.111 0	0.113 7
ShuffleunitNet	0.111 1	0.111 0	0.119 3
SqueezeNet	0.111 0	0.111 0	0.099 9
DenseNet121	0.111 1	0.111 0	0.221 9
ResNet18	0.111 1	0.111 0	0.567 1
ResNet50	0.111 0	0.111 1	0.620 0
ResNet101	0.111 1	0.111 1	0.636 1
MobileNetV1	0.111 1	0.111 0	0.111 0
MobileNetV2	0.111 0	0.111 0	0.155 3
MobileNetV3	0.111 1	0.111 1	0.111 0

参数	数值
学习率	0.01
批次	32
输入通道数	1
输入数据尺寸	224×224
损失函数	交叉熵损失函数
优化器	随机梯度下降优化器

模型	准确率	精确率	召回率	F1分数
InceptionV1	0.099 9	0.010 0	0.099 9	0.018 1
InceptionV2	0.113 7	0.012 9	0.113 7	0.023 2
Inception-ResNetV1	0.115 1	0.125 2	0.115 1	0.032 4
Inception-ResNetV2	0.113 7	0.012 9	0.113 7	0.023 2
ShuffleunitNet	0.119 3	0.014 2	0.119 3	0.025 4
SqueezeNet	0.099 9	0.010 0	0.099 9	0.018 1
DenseNet121	0.221 9	0.062 7	0.221 9	0.096 7
ResNet18	0.567 1	0.564 9	0.567 1	0.549 2
ResNet50	0.620 0	0.616 7	0.620 0	0.597 6
ResNet101	0.636 1	0.634 2	0.636 1	0.599 6
MobileNetV1	0.111 0	0.012 3	0.111 0	0.022 2
MobileNetV2	0.155 3	0.085 4	0.155 3	0.107 5
MobileNetV3	0.111 0	0.012 3	0.111 0	0.022 2
Res-MobileComNet	0.726 8	0.734 4	0.726 8	0.721 3

模型	准确率
LightResNet	0.636 6
LightMobileNetV3large	0.632 5
Res-MobileCom	0.726 8