基于多变量神经网络模型的菜品销量预测

Forecast of Dish Sales Based on Multivariate Neural Network Model

随着互联网技术的发展,人们进入了数字化和智能化的“互联网共享”时代。餐饮企业越来越重视利用数据指引企业理性发展,而餐饮业菜品库存过多或过少会直接影响企业的成本与净利润,因此能够精准预测菜品销量有利于降低餐饮企业的生产成本和提高净利润。为了减少采购菜品的浪费和保持菜品的新鲜度,提出了多变量神经网络模型,并利用该模型预测陕西省某餐饮企业近两年的销量数据。结果表明,多变量长短时记忆神经网络模型(Multi-variable Long Short-Term Memory,Multi-LSTM)的预测精度明显优于季节性差分自回归滑动平均模型(Seasonal Autoregressive Integrated Moving Average,SARIMA)和时间序列与神经网络组合模型,且略优于单变量长短时记忆神经网络模型(Single-variable Long Short-Term Memory,Single-LSTM)。

With the development of internet technology,people have entered the era of digital and intelligent“internet sharing”.Catering enterprises are paying more and more attention to using data to guide the rational development of enterprises.However,too much or too little food inventory in the catering industry will directly affect the cost and net profit of enterprises.Therefore,accurate prediction of food sales is conducive to reducing the production cost and improving the net profit of catering enterprises.In order to reduce the waste of purchasing dishes and maintain the freshness of dishes,a Multi-variable Neural Network Model is proposed and used to predict the sales data of a catering enterprise in Shaanxi in recent two years.The results show that the prediction accuracy of the Multi-variable Long Short-Term Memory(Multi-LSTM)model is significantly better than Seasonal Autoregressive Integrated Moving Average(SARIMA)and comparison of artificial neural network and time series model and slightly better than the Single-variable Long Short-Term Memory(Single-LSTM)model.