Research On The Key Technology Of Grey Forecast For Small Sample Time Series

Grey prediction method is one of the most important approaches to solve the predic-tion problem of time series with a few samples.Due to the large number of random events in the real world,the time series often show obvious nonlinearity and stochasticity.How-ever,the existing grey prediction models are difficult to solve all problems completely.So,grey prediction models still need to be improved and optimized continuously to improve their prediction abilitiese.This thesis mainly studies the improvement and optimization of the grey prediction model based on the grey action quantity optimization,fractional order accumulation operator,background value optimization and so on.Some grey prediction models are proposed and their parameters are optimized by using intelligent optimization algorithm.And the proposed grey models are applied to forecasting the short-term energy consumption.The main research contents of this theis are as follow:(1)According to the time-varying nature of grey action quantity of GM(1,1)model,the grey action quantity can be improved from static one to dynamic one.Then power-driven grey prediction model with exponential grey action quantity is proposed to effec-tively fit the dynamic characteristics of the sequence.The modeling method of the new model is systematically given.And the validation test of the new model is verified by comparative experiments.It can be drawn that the power-driven grey prediction model has better prediction performance.(2)By using the nonlinear feature of incomplete gamma function and the optimiza-tion of grey action quantity,the incomplete gamma grey prediction model is proposed to enhance the adaptive capacity.The parameter estimation and solution method of the new model are studied,and the nonlinear parameters of the model are optimized by Whale Op-timization Algorithm.The modeling and forecasting performance of the proposed model are verified through four validation experiments.It can be drawn that the incomplete gamma grey prediction model has better modelling and prediction performance.(3)Firstly,it is analyzed and proved that the grey action quantity of fractional or-der accumulation grey prediction model has time-varying property on the homogeneous exponential sequence.In other words,the grey action quantity changes with time.There-fore,the dynamic exponential grey action quantity is used to replace the static grey action quantity of the classical fractional order accumulation grey model.Then,the power-driven fractional order accumulation grey prediction model is proposed.After the advantages of the proposed model in modeling and prediction are verified on homogeneous exponential sequence and several kinds of non-homogeneous exponential sequences,it is applied to forecast the short-term wind power consumption.(4)Based on the time-varing nature of grey action quantity of fractional order ac-cumulation grey prediction model,a novel fractional order accumulation grey prediction model with parabolic grey action quantity is proposed to improve the performance of the classical model.And the the error of fractional order accumulation grey prediction model in the discretization process are analyzed.The method of trapezoid formula approximating the background value is replaced by Simpson's rule.And a novel fractional order accu-mulation grey prediction model based on Simpson's rule and optimization of parabolic grey action quantity is proposed.Finally,the two new models are used to forecast the global oil consumption in short term.And it can be drawn that the new models have better performance of modelling and forecasting.