| With the increase proportion of photovoltaic(PV)installed capacity in China,for the sake of random and intermittent properties of new energy resources,the difficulties of new energy resources absorption and dissipation are getting more and more serious.As one of the most economical and efficient solutions to the above-mentioned problems,PV power generation forecasting plays a vital role in dispatching operation,power flow optimization,equipment maintenance,etc.Affected by different power generation plans and decision-making objects in different time scales,PV power generation data with different time scales are required.Therefore,it is necessary to improve the accuracy of PV power generation forecasts at various time scales,in order to maintain safety of the power grid.For the minute-level PV power generation forecasting,it can not only revise the real-time scheduling plan,but also have a guiding impact on the trading of the electricity spot market.In view of this,this paper conducts a systematic study on the minute-level PV power generation forecasting to provide key technologies which can improve the PV power generation capacity and PV power penetration rate.For the prediction of ultra-short-term minute-level PV power generation power,cloud motion plays a key role for the volatility and intermittent power output of solar power stations.Therefore,the study of cloud movement displacement calculation is of great significance for ultra-short-term minute-scale photovoltaic power prediction.In this area,Fourier phase correlation theory(FPCT)is widely used in image registration due to its fast calculation speed.However,in the actual simulation process,the FPCT algorithm can easily form wrong displacement vector results due to noise problems.Therefore,in this paper,an image-phase-shift-invariance-based(IPSI)multi-transform-fusion method(MTF)is proposed to improve the robustness and accuracy of the sky image cloud motion displacement vector calculation.First,various transforms are verified the IPSI characteristics,such as Wavelet Transform(WT),Affine Transform(AT)and Convolution Transform(CT).Then select one of the three transforms to a pair of adjacent sky images by changing the transform parameters.Then various of results can be acquired by using FPCT.Finally,apply Gaussian distribution curve to fit the result point distribution,then the coordinates of peak point can be extracted as final cloud motion displacement result.4 different weather categories are selected for testing the performance of proposed MTF method.The experimental results show that the performance of MTF method is better than traditional FPCT,optical flow,AT and other IPSI family algorithms.MTF method can reduce the error rate of cloud motion displacement effectively,which shows its high accuracy and robustness.After calculating the cloud displacement vector,the relationship between cloud color characteristics and irradiance is analyzed.Combining with distance characteristics,“ground-based sky image-irradiance” mapping model can be established by using BPNN/SVM algorithm.After inputting the cloud color and distance features corresponding to each pixel of the real-time image into the model,a surface irradiance distribution map within a few square kilometers can be obtained.After that,predict the sun-covered area in next 10 min and extract the corresponding cloud features in each minute,then put them into the mapping model to obtain the irradiance prediction results.With other meteorological factors as inputs,minute-level prediction of PV power generation can be achieved.In this paper,3 different cloud conditions: blocking cloud,thin cloud,and thick cloud are applied to verify the forecasting accuracy.The results show that in spite of under cloudy circumstance,the surface irradiance is fluctuant drastically,compared with traditional pure data-driven forecasting model,the proposed model with cloud features still shows higher accuracy,and further enriches the technical methods of irradiance and PV power generation forecasting. |
