Photovoltaic power prediction based on secondary modal decomposition, kernel principal component analysis, and bidirectional long short-term memory neural network

Due to the unpredictability and complexity of natural conditions, the output of photovoltaic power generation has apparent nonlinearity and fluctuation, which makes it challenging to construct a prediction model. By considering five environmental factors, the improved entirely ensemble empirical mode decomposition is used to decompose the sequence of environmental factors. Secondly, the kurtosis of each ICIMFn component of the primary decomposition is calculated, and the improved complete set empirical mode decomposition is used further to show the multi-scale characteristics of the time series. Then, the kernel principal component analysis was used to reduce the dimensionality of the high-dimensional component data to reduce redundant information and retain the main features. Subsequently, the bidirectional long short-term memory neural network was used to dynamically model the multivariate feature sequence to predict the photovoltaic power generation. By verifying three months from the measured data of a power station, the results show that the model is better than the traditional prediction model and has higher prediction accuracy.