Research On Improved Algorithm Of Coal Taxation Forecast

The coal industry is a pillar industry in national economy of China,and it has a very important strategic position in the national economy.Accurate coal tax forecasting is important for both tax authorities and coal enterprises,on the one hand,it can provide policymakers with theory basis to make more scientific guidelines on coal taxation and to make reasonable guidance to tax authorities about the next step of work;on the other hand,it will help coal enterprises make plan for the material,the amount of coal mining and sales in advance.So accurate coal taxation forecast is of great importance.The prediction algorithm will directly affect the accuracy of the prediction results.This paper aimed to introduce the common forecasting models into the field of coal tax forecast with high precision and high stability,and the author made some improvements to the algorithm of some models.At the same time,the simulations of the improved algorithms were realized in Matlab,and the simulation results were compared and analyzed.First of all,this paper introduced two commonly used prediction models,namely,traditional GM(1,1)prediction model and standard BP neural model,and then applied them to coal revenue forecasting respectively and analyzed the scope,advantages and disadvantages of the two models.Secondly,in order to improve the prediction precision of traditional GM(1,1)model,the metabolism GM(1,1)model had been proposed for coal tax forecasting,and the simulation results show that the improved GM(1,1)model has higher accuracy than the standard GM(1,1)model.Thirdly,it was introduced that to improve the convergence speed of BP network by change training functions,and experiments were conducted to verify it.Although the BP neural network can fit all kinds of curves in theory,it also has some disadvantages,such as easily falling into local extremum,slow convergence speed,and so on.The characteristics of coal taxation determine that it needs higher predicted precision,so finally,according to the shortcomings of BP neural network,the genetic algorithm was used to optimize BP neural network' s initial weights and thresholds to improve the accuracy of the coal taxation prediction.The simulation results show that the adaptability and convergence of the GA-BP neural network model are better than BP neural network model.In this paper,the feasibility and stability of the three improved prediction models were verified by a large number of coal tax forecasting experiments,and the prediction results of the three models were compared and analyzed.Simulation experiments show that the grey forecasting model and neural network model have similar average relative errors which satisfy the standard of engineering prediction error(less than 10%)when doing short-term prediction for coal tax,but the algorithm complexity of the grey prediction is much less than the neural network,so grey forecasting model is a better choice for short-term and real time prediction;when forecasting a medium to long-term coal tax,the prediction accuracy form high to low in order is:GA-BP neural network?BP neural network and metabolism GM(1,1)model,and their average relative errors are 3.1%,4.6%and 5%.In general,GA-BP neural network prediction model has smaller forecasting error and higher accuracy,so it is more suitable for coal tax forecast.Finally,the three kinds of improved forecasting algorithms were applied to the coal tax specialization management system to realize the coal tax forecasting module.