GRU-ARIMA与多元回归集成模型的应用——以五粮液为例的销售额预测研究

Application of GRU-ARIMA and Multivariate Regression Integrated Model—A Study on Sales Forecasting of Wuliangye

摘要: 本研究以五粮液企业销售额为研究对象，深入探讨了影响其销售额的关键因素及GRU与时间序列集成模型的实证分析，旨在构建精准可靠的预测模型，帮助企业更好把握市场动态，优化销售策略。首先运用多元回归分析对五粮液销售额的影响因素进行系统分析。通过逐步回归筛选变量建立方程，发现显著变量归属净利润系数2.288和营业净利润率系数94.192，资产负债率系数0.021，反映了销售额与企业的成长能力和盈利能力呈正相关关系、与偿债能力呈负相关，还揭示了企业的经营能力和抗风险能力与以上能力存在共同作用，说明若企业若想提升销售额，则应提升企业本身的成长潜力以及盈利能力并降低企业负债。然后对五粮液2008年第四季度至2023年第三季度五粮液历史销售额数据的训练和验证。在模型预测准确性评估方面，采用了MAE、RMSE以及MASE等指标，与自适应ARIMA模型和决策树回归模型进行对比，发现集成模型的误差指标均最小，且拟合优度高达0.997，在预测五粮液销售额方面表现出色。

Abstract: Taking the sales of Wuliangye as the object of research, this study discusses in depth the key factors affecting its sales and the empirical analysis of GRU and time series integration model, aiming at constructing an accurate and reliable forecasting model to help the enterprise better grasp the market dynamics and optimize its sales strategy. Firstly, multiple regression analysis is applied to systematically analyze the influencing factors of Wuliangye’s sales. Through stepwise regression screening variables to establish the equation, found that the significant variables attributable to the net profit coefficient of 2.288 and operating net 222 profit margin coefficient of 94.192, the coefficient of the balance sheet ratio of 0.021, reflecting the sales and the enterprise’s growth and profitability are positively correlated with the negative correlation with the ability to pay off the debt, but also revealed that the enterprise’s operating ability and risk-resistant ability and the above ability to play a joint role in the above ability, that if the It also reveals that the operation ability and risk resistance ability of the enterprise work together with the above abilities, suggesting that if the enterprise wants to increase sales, it should improve its own growth potential and profitability and reduce its debts. Then the historical sales data of Wuliangye from the fourth quarter of 2008 to the third quarter of 2023 are trained and validated. In terms of model prediction accuracy assessment, indicators such as MAE, RMSE and MASE were used to compare with the adaptive ARIMA model and the decision tree regression model, and it was found that the integrated model had the smallest error indicators and the goodness of fit was as high as 0.997, which was excellent in predicting the sales of Wuliangye.