GDP prediction method based on N-BEATS

Abstract

The invention discloses a GDP prediction method based on N-BEATS. The method comprises the following steps: downloading an original GDP data set on a data platform; performing data preprocessing on the original GDP data set to obtain a preprocessed GDP data set; performing feature selection on the preprocessed GDP data set, and extracting data features of the preprocessed GDP data set; an N-BEATS model is established; an N-BEATS model is trained through the preprocessed GDP data set, and a trained N-BEATS model is obtained; and on the basis of the original GDP data set, predicting a to-be-predicted GDP by using the trained N-BEATS model. The method is easier to train, the prediction accuracy is also improved, explainable output can be provided, the first knowledge is used as little as possible, the data volume of a training set is greatly reduced, the training time is shortened, and the prediction efficiency is superior to that of LSTM and SVM models.

Description

Technical field [0001] The present invention relates to the technical field of computer prediction model, and more particularly to a GDP prediction method based on N-Beats. Background technique [0002] GDP (Gross Domstic Product, GDP), refers to the ultimate results produced by all of the residents in a certain period of time during a certain period of time. GDP is not only a core indicator for accounting national economy, but also one of the indicators that measure the good or short-lived level of this country or region in this period. At the same time, the government and businesses can refer to GDP growth in macro regulation and adjustment of financing decisions. In many research areas of macroeconomics, it is one of the important research parts for GDP. Nowadays, with the fact that domestic and foreign economic situation is more complex, under the role of structural reform of my country's supply side, economic growth is affected by economic structural characteristics, while i...

AI Technical Summary

Problems solved by technology

[0005] Compared with nonlinear models, linear models such as Bayesian vector autoregressive model (BVAR) have relatively unstable prediction curves when predicting GDP, and the predicted value of the BVAR model generally has a better fit with the actual value of GDP. Low, and the prediction curve calculated by the BAVR model is relatively rough in the prediction of local extreme points, resulting in the phenomenon of amplifying the short-term trend, resulting in a large degree of error in the prediction results

Another linear model, the ARIMA model, requires the time series data to be sufficiently stable during data processing, or the time series data after differential differentiation is sufficiently stable. However, GDP is affected by various complex factors in actual situations and cannot always maintain data stability. Therefore, when using the ARIMA model to predict GDP, the prediction results will be inaccurate due to the defects of the model itself, and in essence, the model can only capture linear relationships and cannot capture nonlinear relationships, which brings limitations to GDP forecasting

[0006] For nonlinear models, the traditional ANN model cannot reflect the time series relationship between samples during data processing, but adding the time series relationship between samples when analyzing and predicting GDP will be of great help to the prediction results, so ANN The model is not as good as the LSTM model at predicting GDP

However, the LSTM model depends largely on its step-by-step forecast when predicting time series, so the results produced by using the LSTM model for long-term GDP forecasting may be invalid. In addition, as a neural network, the LSTM model is in Prediction requires a large amount of data to correctly train the model, and the training time is relatively long, so the model is less efficient in predicting GDP, and the LSTM model is prone to overfitting due to the increase in the number of layers in the process of predicting GDP The phenomenon

As a new type of small-sample learning method, SVM can achieve efficient transduction reasoning from training samples to prediction samples due to its characteristics different from existing statistical methods, which can greatly simplify the problem, but the essence of SVM is to use Quadratic programming is used to solve the support vector, so the calculation of the matrix of order m is involved. When the number of samples m is very large, it needs to consume a lot of internal storage and calculation time of the computer. Therefore, the use of the SVM model needs to rely on a large number of feature engineering. In Forecasting GDP requires a lot of computing time

Images