Interval Prediction Of Photovoltaic Output Based On Similar Day Clustering

In recent years,clean energy such as solar energy has developed rapidly and become the main force of energy construction gradually.The prediction of photovoltaic output is of great significance for reducing the risk of grid connection,making and timely adjusting the power dispatching plan and reducing the phenomenon of "light abandonment".However,due to the influence of solar radiation,weather,temperature,humidity,wind speed and other factors,photovoltaic output has strong randomness and volatility.The prediction accuracy of existing prediction methods is poor,and it is difficult for this methods to reflect the uncertain information in the photovoltaic output process.To solve this problem,this paper proposes an interval prediction method of photovoltaic output based on similar day clustering.Firstly,this paper analyzes the correlation between each factor and the photovoltaic output and removes the factors that have a weak influence on the photovoltaic output,proposes a similar day clustering method based on fuzzy c-means and Frechet distance,which divides the sample set into three similar days: Sunny similar day,cloudy similar day,rainy similar day,to reduce the impact of its volatility on photovoltaic output interval prediction.Then the interval prediction methods for different similar day scenarios are constructed: 1)For sunny and cloudy similar days,considering its strong regularity,an interval prediction model based on NSGA-Ⅱ-ADLSSVM was constructed: First,a variable interval coefficient interval estimation method is proposed for the construction of interval prediction sample sets.Then,a dual support vector machine model is established to predict the upper and lower limits of the interval.Finally,NSGA-II is selected to optimize the model parameters.2)For rainy similar days,considering its poor regularity,an AVMD-SDBN-AKDE interval prediction method is proposed: First,an adaptive variational modal decomposition method is proposed,and for the frequency difference and variation coefficient of each modal component after signal decomposition,the number of decomposition layers is adaptively selected to improve its regularity.Then,based on the sample entropy method,the decomposed modal components are divided into high-frequency and low-frequency sets,and a point prediction model based on a deep confidence network is constructed for the high-frequency and low-frequency components to predict the photovoltaic output point.Aiming at the prediction error,an adaptive kernel density estimation model is constructed to obtain the confidence interval of the error under different confidence levels.Combined with the photovoltaic output prediction result and the error interval,a photovoltaic output interval prediction model under a rainy and similar day is finally obtained.The method proposed in this paper is tested and verified by the practical data of an industrial area in Shaanxi Province.The results show that the method proposed in this paper can achieve better prediction results for different similar days,which can provide a reference for grid dispatching and safe and stable operation.