Highly accurate short-term prediction of distributed photovoltaic power generation is one of the keys to the new power system scheduling plan and the implementation of new energy consumption. However

distributed photovoltaic has the characteristics of complex operating conditions

wide range of points

strong randomness and insufficient data collection frequency

etc. Traditional data-driven methods are faced with problems such as poor data sample quality and difficult optimization of model parameters. This paper proposes a Kernel based extreme learning machine based on weighted mean of vectors algorithm (INFO) short-term forecast method of distributed photovoltaic power generation by machine

KELM. To solve the problem that the extreme learning machine prediction model needs more hidden layer neurons when processing complex predictions

resulting in a very complex network structure

kernel function is introduced into the model

which can enhance the stability of the model and have strong generalization ability

realize large-scale photoelectric data processing and maintain high real-time performance. In view of the difficulty of tuning the model parameters of the nuclear extreme learning machine

INFO is used to optimize the KELM parameters and reduce the error of the prediction results. Finally

seasonal data with obvious characteristics and three weather types

sunny

cloudy and rainy

are analyzed. Compared with KELM model

RF model

SVM model

GA-BP model

CNN-LSTM model and GA-KELM model

the results show that: the comprehensive determination coefficient of INFO-KELM PV power short-term prediction is 0.988

which is 8.51% and 6.14% higher than that of GA-BP and GA-KELM respectively. This method not only realizes accurate prediction

but also ensures high real-time performance.