| Due to the increasing shortage of traditional fossil energy and the environmental damage caused by the massive use of fossil energy,the use of clean energy is being vigorously developed at home and abroad,and the proportion of PV(photovoltaic)power generation is increasing.A large amount of PV power generation generates electricity into the power grid,and the impact on grid stability is huge.Therefore,PV power forecasting has become an important research topic.Based on the research of PV power fluctuation,this paper analyzed and verified the periodic characteristics of PV power itself,realized the effective identification of PV anomaly data and the effective filling of missing data,and builds a short-term prediction model of PV power.Finally,A quality assessment method for PV power prediction was proposed.The periodic characteristics of PV power were verified by spectral analysis of the PV power sequence.The decomposition and reconstruction of the PV power sequence were realized by using the Fourier series,and the PV power was decomposed into periodic components,high frequency residual components and low frequency residual components.The individual components were analyzed as a basis for subsequent research.There are many types of PV power data,including irradiance,component temperature,ambient temperature,air pressure,wind speed,power,etc.In the process of data acquisition,PV power plants have a certain proportion of abnormal data in PV power data due to human factors or communication failures.these abnormal data affects the related research of PV power.In this paper,the factors affecting PV power were fully analyzed,and it was found that the incident angle of PV had a great influence on the distribution of irradiance-power scatter.The power data for different illumination characteristics were classified.Under a certain degree of confidence,the Copula function of the irradiance and temperature probability distribution function was fitted,and the conditional probability distribution of PV power generation under each irradiance was calculated.Combined with the abnormal data discriminant criteria,the PV power generation data was effectively identified.Similarly,a certain proportion of PV power missing data appears in the data acquisition process,and another source of missing data is abnormal data identification.In this paper,the normal random entropy of the traditional cloud model was changed to the random entropy based on Copula theory,and an improved cloud model was constructed.The conditional interpolation complement model was established by referring to the length and the volatility of the missing segments of the PV power,and the effective filling of the PV power missing data was realized.The PV power was decomposed into periodic components and residual components,the periodic components had strict regularity,so only the short-term prediction of the remaining components was required.The coupling relationship between the residual component of PV power and various influencing factors was analyzed by Pearson correlation coefficient method.The weather was divided into three generalized weather types.Under different weather types,the local sensitive hash algorithm was used to realize the proximity search between multidimensional data.The short-term prediction result of PV power was obtained by combining the retrieved value with the periodic component.The validity and practicability of the proposed method were verified by an example.Finally,the types of prediction errors in the PV power prediction mode were analyzed,and the influence of each component of PV power on the prediction accuracy and the predictability and unpredictability of the PV power sequence itself were studied.We defined different unpredictable components of PV power,solved the standard deviation of modeling error,and determined the minimum value of the standard deviation of PV power modeling error.The results showed that the regularity of PV power is different in different regions. |
PDF Full Text Request