Research On Tax Forecast Of Mineral Products Based On LSTM Neural Network

Taxation of mineral products is the main source of revenue for government taxes,and mineral resources are an important material basis for economic development.The prediction of tax revenue of mineral products can better control the risk of tax revenue,reasonably plan the mining of mineral products,and provide theoretical guidance for mining companies to formulate a more comprehensive production and sales plan.Accurate prediction of mineral taxation has important reference significance for the management of mineral taxation and customization of corresponding taxation measures.Analyzing the hidden relationship of historical tax data and using mathematical models to predict future tax revenue is the focus of tax forecast research.This paper proposes a tax forecasting model which combines wavelet transform with long short-term memory(LSTM)neural network.Data preprocessing combined with wavelet transform can remove noise from tax data and improve the generalization ability of the model.The LSTM neural network can better learn the correlation of historical tax data by adding hidden units and gated units,and further extract valid state information of input sequences.Moreover,the LSTM overcomes the long-term dependency problem of recurrent neural networks.The encoder-decoder structure of mineral product tax forecasting model constructed by LSTM neural network can further enhance feature extraction and enhance the model's ability to make multi-step predictions,which effectively reduces the model's error accumulation d ue to the increase in prediction days.The encoder-decoder structure uses the encoder to extract the context features in the data,and the decoder performs feature reconstruction and multi-step prediction.The experimental results show that in the medium and long-term prediction of mineral taxation,the prediction accuracy of the combined wavelet transform denoising with LSTM encoder-decoder model is 4.54 percentage points higher than that of the model without denoising;compared with the single-step prediction LSTM model based on sliding window,it is improved by 1.45 percentage points;compared with the grey model based on differencedifferential equation and autoregressive integrated moving average model(ARIMA)based on regression based,it is improved by 17.48 percentage points and 3.66 percentage points respectively.It also provides an important reference for the tax management and policy implementation of mineral products in practical applications,realizes the promotion of tax quality and the total tax revenue,and promotes the further optimization of tax services.