Multi-step Ultra-short-term Wind Speed Prediction Based On Integrated Multi-model Fusion

With the continuous progress of science and technology,the demand for energy has become one of the most important issues in today’s social development.Excessive exploitation or waste of energy will cause irreversible damage to the environment and ecology.Therefore,it is necessary to seek clean,green and renewable energy.As a kind of clean energy,wind energy has many advantages.Its reproducibility and accessibility have become an important source of energy acquisition.Wind power makes wind power industry technology increasingly mature.However,the wind speed is nonlinear,random and unstable,and there are many difficulties in integrating wind power into the power grid.The prediction of ultra-short-term wind speed will directly affect the generating efficiency of the generator set.It has a certain influence on the maintenance,security evading and dispatching efficiency of power equipment.Ultra-short-term wind speed forecasts can be made at very short intervals,usually a few minutes to 30 minutes.The prediction of ultra-short-term wind speed is characterized by high variation rate,randomness and instability.Therefore,accurate prediction of ultra-short-term wind speed is of great significance to the power generation efficiency of wind farms and has a good application prospect in the field of meteorology.This thesis proposes a new decomposition and integration prediction model for ultra-short-term wind speed time series prediction.Firstly,the empirical mode decomposition algorithm is used to decompose the ultra-shortterm wind speed.Ultra-short-term wind speed data generally have high fluctuation time series data.In this thesis,the original ultra-short-term wind speed time series data with high fluctuation are decomposed into several stable time series data,and the decomposed data are more conducive to data prediction by prediction model.The combined prediction model solves the problem of data defect and unpredictability by empirical mode decomposition algorithm.Secondly,the decomposed components are predicted by a reinforced long and short-term memory neural network.Long and short-term memory neural network is reinforced,that is,on the basis of the traditional long and short-term memory neural network model,the “Peephole Connections” is added.The reinforced long and short-term memory neural network is more stable than the traditional long and short-term memory neural network,which has certain positive significance for regression prediction.The reinforced model can be more stable and reliable when dealing with large amounts of data.Thirdly,the hyperparameters of the neural network are optimized by the improved sparrow search algorithm.The improved sparrow search algorithm can obtain faster convergence speed,seek better extremum,and combine Cauchy mutation and reverse learning strategy to disturb the optimal solution position,and improve the ability to resist local extremum.The prediction accuracy of the reinforced long and short-term memory neural network model is greatly improved by optimizing the four hyperparameters of the reinforced long and short-term memory neural network model with the improved sparrow algorithm.Finally,the prediction values of the reinforced long and short-term memory neural network models are superimposed to obtain the final prediction values.Taking actual ultra-short-term wind speed data as the research object,R Square,root mean square error,relative root mean square error,modified root mean square error,mean absolute error,mean absolute percentile error,Theil inequality coefficient,the index of agreement,Pearson’s test,prediction error distribution box-plot and Taylor diagram are used to judge the performance index of the prediction model.In the case study,the performance of the prediction model designed in this thesis is compared with standard long and short-term memory neural network,empirical mode decomposition algorithm and standard long and short-term memory neural network,empirical mode decomposition algorithm and standard sparrow search algorithm and reinforced long and short-term memory neural network,and other latest models.The results show that the ultrashort-term wind speed prediction model proposed in this thesis has the best performance and accuracy.