Accurate solar photovoltaic (PV) power forecasting is an essential tool for mitigating the negative effects caused by the uncertainty of PV output power in systems with high penetration levels of solar PV generation. Weather classification based modeling is an effective way to increase the accuracy of day-ahead short-term (DAST) solar PV power forecasting because PV output power is strongly dependent on the specific weather conditions in a given time period. However, the accuracy of daily weather classification relies on both the applied classifiers and the training data. This paper aims to reveal how these two factors impact the classification performance and to delineate the relation between classification accuracy and sample dataset scale. Two commonly used classification methods, K-nearest neighbors (KNN) and support vector machines (SVM) are applied to classify the daily local weather types for DAST solar PV power forecasting using the operation data from a grid-connected PV plant in Hohhot, Inner Mongolia, China. We assessed the performance of SVM and KNN approaches, and then investigated the influences of sample scale, the number of categories, and the data distribution in different categories on the daily weather classification results. The simulation results illustrate that SVM performs well with small sample scale, while KNN is more sensitive to the length of the training dataset and can achieve higher accuracy than SVM with sufficient samples.

Solar photovoltaic (PV) power forecasting has become an important issue with regard to the power grid in terms of the effective integration of large-scale PV plants. As the main influence factor of PV power generation, solar irradiance and its accurate forecasting are the prerequisite for solar PV power forecasting. However, previous forecasting approaches using manual feature extraction (MFE), traditional modeling and single deep learning (DL) models could not satisfy the performance requirements in partial scenarios with complex fluctuations. Therefore, an improved DL model based on wavelet decomposition (WD), the Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) is proposed for day-ahead solar irradiance forecasting. Given the high dependency of solar irradiance on weather status, the proposed model is individually established under four general weather type (i.e., sunny, cloudy, rainy and heavy rainy). For certain weather types, the raw solar irradiance sequence is decomposed into several subsequences via discrete wavelet transformation. Then each subsequence is fed into the CNN based local feature extractor to automatically learn the abstract feature representation from the raw subsequence data. Since the extracted features of each subsequence are also time series data, they are individually transported to LSTM to construct the subsequence forecasting model. In the end, the final solar irradiance forecasting results under certain weather types are obtained via the wavelet reconstruction of these forecasted subsequences. This case study further verifies the enhanced forecasting accuracy of our proposed method via a comparison with traditional and single DL models.

With the increase of solar photovoltaic (PV) permeability in power system, the impact of random fluctuation of PV power on the safe and stable operation of power grid becomes more and more serious. High-precision PV power forecasting can effectively promote the grid’s absorption of PV power generation. Cloud is the most important factor affecting the surface irradiance and PV power. For the ultra- short-term solar PV power forecast considering the influence of cloud movement, it is necessary to be able to obtain the surface irradiance according to the sky cloud observation data. Therefore, in order to accurately achieve the real-time mapping relationship between sky image and surface irradiance, a hybrid mapping model based on deep learning applied for solar PV power forecasting is proposed in this paper. First, the sky image data is clustered based on the feature extraction of convolutional autoencoder and k-means clustering algorithm after preprocess stage. Second, a hybrid mapping model based on deep learning methods are established for surface irradiance. Finally, the simulation results are compared and evaluated with different deep learning methods (CNN, LSTM and ANN). The results show that the proposed model in this paper has higher accuracy and can maintain robustness under different weather conditions.

Accurate short-term photovoltaic (PV) power forecasting is of great significance for the safe and stable operation of power system. Spatial information from neighboring PV sites contributes to improving forecasting performance. However, most of the current methods considering the spatial information of neighboring sites indiscriminately use all sites data for modeling, which will lead to information redundancy, resulting in low forecasting accuracy. Therefore, this paper proposes a short-term solar power forecasting method based on optimal graph structure considering surrounding spatio-temporal correlations. Firstly, the neighboring sites data is analyzed from the perspective of geographical and weather factors to select typical PV sites. Secondly, based on the complex network theory, a new index is proposed to evaluate the connectivity of the graph structure, which improves the predictive ability of the Graph Convolutional Network (GCN) model. Finally, considering numerical weather prediction (NWP) data, a hierarchical directed graph structure is constructed to indicate the unidirectional relationship between input samples, which is used as the input of GCN model to mine the spatio-temporal correlation around the targeted site. Through carrying out the case study, the proposed method shows excellent performance in improving accuracy of power forecasting compared with other benchmark methods.