Ultra-short-term Power Prediction Method And Application Of Photovoltaic Power Generation System

Nowadays,both companies and people around the world have an increasing demand for energy,and traditional mineral energy is becoming scarce.Since renewable energy has the characteristics of being inexhaustible,clean and pollution-free,everyone is developing various types of renewable energy to replace traditional energy,with the aim of establishing a green and sustainable energy system.However,at this stage,the technology of solar power generation is not yet perfect,which leads to great randomness and volatility in solar power generation.These characteristics can also destabilise the power grid when connected to it,causing difficulties in the operation and scheduling of power plants.In addition,the connection of solar power to the grid can cause disturbances to the entire grid.The ultra-short term prediction of solar power generation can provide powerful data support to the dispatching department,make better overall arrangements for power generation,and reduce the interference caused by solar grid connection.The main works and innovations of this paper are as follows:1.Considering the uneven performance of ultra-short term power prediction models in terms of prediction accuracy,model training complexity,and other performance indicators,a method is proposed to objectively select models through experiments using different prediction models on the same common data set.Through comparative analysis of the prediction levels of different models in practical applications,the experimental results show that among all models,the Bi-LSTM model predicts the best outcome evaluation indicators with the best prediction results,eliminating the effects of prediction scenarios and data heterogeneity,and the conclusions drawn are more objective.2.The problems of low prediction accuracy and poor performance of ultra-short term power prediction models are analysed,and it is concluded that due to the mutual coupling between different dimensions of data,there is a lot of redundant information,which affects the performance of ultra-short term power prediction.In response to this conclusion,an improved correlation coefficient(ACC)is proposed to analyse the correlation between data features,select appropriate features with high correlation,and perform dimensionality reduction of data features.And combined with the power prediction model Bi-LSTM to form a new ultra-short term power prediction model ACC-Bi-LSTM for ultra-short term power prediction.By evaluating and analysing the prediction results of the public dataset and comparing them with the prediction results of the unreduced model,the final result shows that this model has the best prediction effect and can achieve the best results among all evaluation indicators.3.Targeting the low accuracy of ultra-short term power prediction results for photovoltaic applications,the feature dimensionality reduction based ultra-short term prediction method ACC-Bi-LSTM was experimentally verified on a real photovoltaic device.After comparing the results predicted by ACC-BP,ACC-RNN,ACC-LSTM under different models,as well as the prediction model PCA-Bi-LSTM constructed by the unreduced Bi-LSTM model and PCA dimensionality reduction method,ACC-Bi-LSTM has the best evaluation index for the predicted results,verifying the effectiveness of this method in photovoltaic-oriented ultra-short term power prediction.