陳彥匡徐雅甄侯明均2025-08-282025-08-282024-09-24U0061-1306202417043300https://www.airitilibrary.com/Publication/alDetailedMesh1?DocID=U0061-1306202417043300https://nutcir-lib.nutc.edu.tw/handle/123456789/1178一、研究目的：本研究旨在透過構建最佳預測迴歸模型，以精準預測冷藏飲品的銷售量，有助於統倉及連鎖販售商店減少庫存量與降低物流成本，提升整體供應鏈流程效率及增加企業營利，並滿足消費者對冷藏飲品的需求。 二、研究方法：本研究使用過去真實資料探討冷藏飲品需求預測模式，以某販售商店於全台5區統倉(北北基、桃竹苗、中彰投、雲嘉南、高屏)之銷售數據歷史資料來分析，透過利用不同預測方法，包含相關分析、灰色關聯分析、線性迴歸及類神經網路分析之比較來開發最佳預測迴歸模型，以預測最佳銷售量。 三、研究結果：實證結果顯示，相關分析及灰色關聯分析所挑選出來的因子對於實際銷售量皆有顯著影響，且以淡旺季的方式下去區分，能更清楚針對實際銷售量的準確性，在分析方法上，線性迴歸分析的預測效能比類神經網路分析更加顯著，訓練方式則以80%-20%訓練法及十折交叉驗證中對於資料分析預測更具有可靠性。 四、研究貢獻：本研究以預測未來的角度，建立最佳預測迴歸模型，為販售商店提供實質性的參考，並為相關文獻做出了貢獻。Purpose: This study aims to construct an optimal forecast regression model to accurately predict the sales volume of refrigerated beverages, which can help consolidated warehouses and chain stores to reduce inventory and logistics costs, enhance the overall supply chain process efficiency and increase the profitability of the enterprises, as well as satisfy the consumers' demand for refrigerated beverages. Method: In this study, the demand forecasting model of refrigerated beverages is investigated using real data from the past. The historical data of sales data from a sales store in five regions of Taiwan are analyzed, and an optimal forecast regression model is developed by comparing the different forecasting methods, including correlation analysis, grey correlation analysis, linear regression, and neural network analysis, to forecast the optimal sales volume. The best forecast regression model is developed by comparing different forecasting methods, correlation analysis, grey correlation analysis, linear regression and neural net-like analysis, to forecast the optimal sales volume. Findings: The empirical results show that the factors selected by correlation analysis and grey correlation analysis have a significant effect on the actual sales volume, and the distinction is made in the way of low and high seasons, so that the accuracy of the actual sales volume can be more clearly targeted. In terms of the analysis method, the predictive effectiveness of linear regression analysis is more significant than that of neural network-like analysis, and the training method is 80%-20% training method and ten-fold cross-validation on the data analysis. The training method is 80%-20% training method and ten-fold cross-validation for data analysis and prediction is more reliable. Contribution: This study models the optimal forecast regression in terms of predicting the future, provides a substantial reference for vending stores, and contributes to the relevant literature.zh冷藏飲品需求預測相關分析灰色關聯分析類神經網路多元迴歸模型refrigerated beveragesdemand forecastcorrelation analysisgrey relational analysisartificial neural networkmultiple regression modelmaster thesis