Skip-based combined prediction method for distributed photovoltaic power generation
Distributed photovoltaics (PV) power generation forecasting plays an important role in ensuring the safety of power grid operation and nearby consumption. In order to enhance the accuracy of distributed PV power generation forecasting, we propose a meteorological feature extraction method and futher...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Science Press (China Science Publishing & Media Ltd.)
2024-05-01
|
| Series: | Shenzhen Daxue xuebao. Ligong ban |
| Subjects: | |
| Online Access: | https://journal.szu.edu.cn/en/#/digest?ArticleID=2617 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Distributed photovoltaics (PV) power generation forecasting plays an important role in ensuring the safety of power grid operation and nearby consumption. In order to enhance the accuracy of distributed PV power generation forecasting, we propose a meteorological feature extraction method and futher design a PV power generation forecasting model based on skip-connect models fusion. In the feature extraction, we use statistical analysis, features cross-correlation, periodicity information, approximate entropy, and the temperature of PV panels to achieve deep feature extraction of time, weather, and power generation data, enriching the model inputs. In model construction, we propose a multi-layer model fusion method based on residual connections. Firstly, we introduce a k-nearest neighbor (kNN)-based softmax regression prediction model. Secondly, we design a three-layer model structure with multiple prediction models fused through residual connections and multi-layer stacking, continuously impg the prediction accuracy of PV power generation forecasting. Based on the real data of electric power companies, we compare the proposed method with others, such as random forest (RF), TabNet and extreme gradient boosting (XGBoost) for photovoltaic power generation prediction. The results show that the proposed model can reduce the root mean square error, mean absolute error, mean squared error, and mean absolute percentage error by 0.109 7, 0.059 1, 0.050 7, and 0.036 8 respectively, and improve the goodness of fit by 0.080 4. The feature extraction method based on multi-meteorological factors and the photovoltaic power generation prediction model based on residual connections for multi-model fusion effectively improve the accuracy and stability of distributed PV power generation forecasting. |
|---|---|
| ISSN: | 1000-2618 |