Short-term photovoltaic power generation prediction based on VMD-IGWO-LSTM

In order to improve the prediction accuracy of photovoltaic power generation, a short-term photovoltaic power generation prediction method based on variational mode decomposition (VMD) and improved grey wolf optimization algorithm (IGWO) to optimize long short term memory (LSTM) neural network is proposed. The multi-dimensional photovoltaic feature data is decomposed into several intrinsic modes and residual components of different frequencies by VMD algorithm to reduce the non-stationarity of the original sequence; IGWO is used to globally optimize the hyperparameters of LSTM neural network, and the IGWO-LSTM combination model under different modal sequence components is established; the trained combination model is used to perform multi-dimensional prediction of the modal feature components of each decomposed subsequence, and the prediction results of each modal component are summed and reconstructed as the final prediction result. The actual data of a photovoltaic power generation are used for experimental analysis. The simulation results show that the constructed VMD-IGWO-LSTM combination model has better prediction effect than the conventional short-term photovoltaic power generation prediction model, which verifies that the method has higher prediction accuracy.

I. Introduction

Solar energy has abundant resource advantages and can be used sustainably. Photovoltaic power generation has strong randomness, volatility and uncertainty. After a high proportion of photovoltaic power is connected to the grid, the difficulty and complexity of grid dispatching increases. Accurate photovoltaic power generation prediction is of great significance to improving photovoltaic access to the grid, formulating reasonable dispatching plans and achieving safe and stable operation of the power system ^[1]. Reference [2] combines the autoregressive moving average model (ARMA) and the artificial neural network (ANN) algorithm, but does not consider the impact of meteorological factors on prediction accuracy. Reference [3] uses the improved gray wolf algorithm to dynamically optimize the initial weights and thresholds of the extreme learning machine (kernel extreme learning machine, KELM), thereby improving the load prediction accuracy. Reference [4] improves the BP neural network and uses the similarity algorithm to extract the output features of different weather types. However, due to the large time interval between the similar day and the predicted day, the prediction results are not ideal. Reference [5] uses the