Short-term power forecasting of fishing-solar complementary photovoltaic power station based on a data-driven model

A data-driven short-term power generation forecasting model has been proposed to address the problems of information redundancy and low forecasting accuracy for the previous model. Pearson correlation coefficient (PCC) was used to select the effective variables affecting photovoltaic (PV) power generation from the original data set. Radial basis function (RBF) neural network was used to train the existing data to fill the missing values, which ensured the authenticity and integrity of the data. Principal component analysis (PCA) was used to reduce the data dimension, and singular value decomposition (SVD) was used to reduce the matrix calculation of PCA. Then sine chaotic mapping was selected to optimize the sparrow search algorithm (SSA). Furthermore, the thresholds and weights of the back propagation (BP) neural network were further searched to construct the Sine-SSA-BP forecasting model. The original model was validated in a 100 MWp fishing-solar complementary PV power station with high relative humidity (RH). The results indicated this model had good adaptability to the high RH, especially in overcast and cloudy days. Under overcast, cloudy and sunny days, the mean error was reduced by 12.88%, 10.28% and 2.53% at maximum compared with BP, GA-BP and SSA-BP models respectively. The k-fold cross validation comprehensively and effectively verified the robustness and versatility of the prediction model. When k = 10, the minimum MAPE of the prediction model could reach 5.92 %, which effectively adapted to the continuous output prediction of PV power with strong volatility.