多模型融合驱动的湘西州零食店营业额的预测与仿真

Prediction and Simulation of Xiangxi Prefecture Snack Store Revenue Driven by Multi-Model Integration

摘要: 近年来，中国经济持续增长，零食消费市场快速发展，消费者对零食的需求日益旺盛。在此背景下，本研究提出一种融合多种算法的建模策略，对零食店营业额进行系统仿真。首先，基于问卷调查选取12项关键指标，构建结构方程模型，识别出影响消费者购买意愿的三大核心因素，并建立购买意愿评价体系；其次，采用长短期记忆网络(LSTM)预测店铺人流量，结合粒子群算法(PSO)对模型参数进行优化，获得训练集R²为0.9067，验证集R²为0.9142；最后，整合目标店铺实际数据，运用Metropolis-Hastings (MH)算法与最大后验估计(MAP)方法对营业额进行仿真，并通过网格搜索法确定购买意愿临界点v，使模型验证的平均R²达到0.88。基于上述结果，本文总结了研究结论，并探讨了模型在实际运营中的潜在应用。

Abstract: In recent years, China’s economy has continued to grow, the snack consumption market has developed rapidly, and consumer demand for snacks has been increasingly strong. In this context, this study proposes a modeling strategy that integrates multiple algorithms to systematically simulate the revenue of snack shops. First, based on a questionnaire survey, 12 key indicators were selected to construct a structural equation model, identifying three core factors influencing consumer purchase intention and establishing a purchase intention evaluation system. Second, a long short-term memory network (LSTM) was used to predict store foot traffic, combined with a particle swarm optimization (PSO) algorithm to optimize model parameters, achieving R² 0.9067 for the training set and R² 0.9142 for the validation set. Finally, actual data from the target store were integrated, and the Metropolis-Hastings (MH) algorithm and maximum a posteriori estimation (MAP) method were used to simulate revenue, with the purchase intention threshold v determined through grid search, resulting in an average verification of R² 0.88 for the model. Based on the above results, this paper summarizes the research conclusions and explores the potential applications of the model in practical operations.