Research On Photovoltaic Power Generation Power Prediction Based On MSSA-BiLSTM-AT

The share of photovoltaic power generation in the total global power generation is increasing year by year.Photovoltaic power generation power is susceptible to factors such as solar radiation intensity,atmospheric temperature and atmospheric humidity,and has the drawbacks of intermittency and fluctuation.Accurate prediction of PV power generation can effectively improve the proportion of PV consumption,which is important to ensure the stable operation of power system.PV power prediction is divided into physical prediction methods and artificial intelligence prediction methods.Physical methods have the characteristics of simple methods and strong applications,but it is difficult to achieve accurate scientific prediction.Artificial intelligence methods predict with the help of algorithms such as BP neural networks,support vector machines and long short-term memory(LSTM)neural networks,which effectively improve the accuracy and stability of PV power prediction.In this paper,a PV power prediction model based on multi-strategy sparrow search algorithm(MSSA)optimized bi-directional long and short term memory(BiLSTM)neural network and attention mechanism(Attention)is constructed,and the main work is as follows:(1)The basic principles of photovoltaic power generation and the main components of photovoltaic power generation systems are introduced.The factors affecting the power of photovoltaic power generation are normalized and analyzed,and the solar radiation intensity,atmospheric temperature and atmospheric humidity are selected as the input variables of the photovoltaic power prediction model.(2)The sparrow search algorithm(SSA)suffers from the problems of insufficient population diversity,easy to fall into local optimum and low pioneering,etc.,and the algorithm optimization effect is improved by various strategies.We introduce good point set and backward learning mechanism to improve the initial population uniformity and diversity;optimize the discoverer and scout-warning search mechanism to compensate for the lack of convergence of the algorithm toward the origin;introduce the Corsi perturbation to improve the ability of the algorithm to jump out of the local optimum and enhance the pioneering and robustness of the algorithm.The effectiveness and superiority of the improved strategy are verified through multidimensional experiments with 12 benchmark test functions,and the significant advantages of the multi-strategy sparrow search algorithm are demonstrated with the help of time complexity analysis and Wilcoxon rank sum test.The optimized scheduling model of island microgrid based on MSSA is constructed,and the optimized solution by MSSA improves the ratio of each renewable energy consumption in island microgrid and the stability of microgrid operation,which verifies the feasibility and superiority of MSSA.(3)To address the problem of low accuracy of BiLSTM PV power prediction,a PV power prediction model based on MSSA-BiLSTM-AT is constructed.The historical data are pre-processed by anomalous data correction,data complementation and data normalization,and mean absolute percentage error,root mean square error,mean absolute error and coefficient of determination are used as the result evaluation indexes.The attention mechanism is introduced in BiLSTM to help BiLSTM learn the key information in the sequence better,and the optimal hyperparameters of BiLSTM fusion Attention model(BiLSTM-AT)are solved by MSSA to construct the MSSA-BiLSTM-AT based PV power prediction model.The experimental results show that the prediction accuracy of the MSSA-BiLSTM-AT-based PV power prediction model is significantly improved compared with other models,which proves the feasibility of the MSSA-optimized BiLSTM-AT model and also verifies the feasibility of the MSSA-BiLSTM-AT model.BiLSTM-AT model is applied to PV power prediction.In summary,this paper introduces the PV power generation principle and system components from the basic principles of PV power generation,and determines the input variables of the PV power prediction model.The sparrow search algorithm is improved by strategies such as good point set,backward learning and Corsi perturbation to enhance the pioneering and optimization effect of the algorithm itself.The MSSA-based island microgrid optimization scheduling model is constructed to improve the stability of island microgrid operation and the proportion of each renewable energy consumption.The MSSA-BiLSTM-AT based PV power prediction model is constructed,which improves the BiLSTM parameter search effect and PV power prediction accuracy with certain superiority and practicality.