基于关联规则及组合模型的面料需求预测

由于服装的面料组成具有复杂性，企业在不同时间对不同规格型号面料需求量不一致，传统的人工预测及单维度智能预测模型难以解决问题。针对服装企业面料需求非确定性、预测难的痛点，提出基于关联规则及组合模型的面料需求预测方法。文章首先构建Apriori面料型号关联模型，挖掘多批多类面料间的型号关联规则；然后构建Prophet时间序列模型与长短期记忆神经网络(long short-term memory,LSTM)的组合预测模型Prophet-LSTM，结合其在解决面料需求预测问题上的优势；最后将挖掘出的高关联面料型号历史需求数据作为输入，采用量子粒子群算法(quantum particle swarm optimization,QPSO)优化组合模型权值系数，进行关联面料需求量预测。使用RMES(root mean squared error,RMSE)及MAE(mean absolute error,MAE)作为评价指标设计对比实验，实验结果表明：采用量子粒子群优化的QPSOProphet-LSTM面料需求预测模型RMES较Prophet降低5.464，较LSTM降低1.184；MAE较Prophet降低4.261，较LSTM降低0.819，需求预测精度更高，支持服装企业面料柔性生产。

Fabric Demand Prediction Based on Association Rule and Combination Model

Due to the complexity of the fabric composition of clothing, the demand of enterprises for fabrics of different specifications and models is inconsistent at different times, and the traditional manual prediction and one-dimensional intelligent prediction models are difficult to solve the problem. Aiming at the pain points of uncertain and difficult prediction of fabric demand in garment enterprises, a fabric demand prediction method based on association rules and combination model is proposed. Firstly, Apriori fabric type association model is constructed to mine the type association rules between multiple batches and categories of fabrics; Then, a combined prediction model of Prophet time series model and long short term memory neural network (LSTM) is constructed, which combines its advantages in solving the problem of fabric demand prediction; Finally, taking the historical demand data of highly correlated fabric models as input, the weight coefficients of the combined model are optimized by quantum particle swarm optimization (QPSO) to predict the demand of correlated fabrics. Using RMES (root mean squared error, RMSE) and MAE (mean absolute error, MAE) as evaluation indexes to design comparative experiments, the experimental results show that the qpsoprophet LSTM fabric demand prediction model using quantum particle swarm optimization RMES is 5.464 lower than prophet and 1.184 lower than LSTM; MAE is 4.261 lower than prophet and 0.819 lower than LSTM, and the demand prediction accuracy is higher, which supports the flexible production of fabric in garment enterprises.