Development and Evaluation of Forecasting Methods for Production Planning and Control by Considering Prediction Models of Nonlinear Dynamics

为了实现有理有据的生产计划和控制，制造企业需要尽可能精确地预测未来的客户需求。因此，公司的盈利能力会受到积极的影响。为了获得较高的预测精度，首先要选择合适的预测方法。此外，该方法必须根据所考虑的需求演变进行调整。在文献中，存在着大量不同的预测方法。然而，在实际应用中，往往使用相对简单的方法，因为更复杂的方法需要大量的专家知识和大量的适应时间。因此，复杂预测方法在提高预测精度方面的潜力没有得到充分挖掘。针对这一点，提出的研究方法涉及一种技术的发展，以选择适当的预测方法。基于广泛构想的模拟研究，应构建一个包含建议的分配矩阵。一方面，该矩阵将包含各种预测方法对不同需求演变的适用性和准确性的信息。另一方面，它将包括适用的参数配置，以适应特殊的演变。因此，矩阵包含专家知识，用于在不同的生产物流应用场景中选择和调整合适的预测方法。为了建立分配矩阵，首先需要研究不同需求演化的动态特征。因此，将采用线性和非线性时间序列分析方法来确定特征模式和分类指标。利用这些结果，将完成对不同需求演变的分类。同时，预测方法将应用于基于模拟和实际案例的生产数据，并评估不同时间范围内的预测准确性。特别是，将采用非线性动力学方法，该方法可以对复杂的动态时间序列演化进行建模，从而具有预测客户需求的巨大潜力。最后，将分类需求演变的动态特征与不同预测方法的预测精度相结合。作为项目的主要结果，所开发的分配矩阵将为选择合适的预测方法提供建议，并根据不同的需求演变进行调整。

In order to achieve a well-founded production planning and control, manufacturing enterprises need to predict future customer demands as precisely as possible. As a result, the company´s profitability can be influenced positively. For the purpose of a prediction with high accuracy, firstly, an appropriate forecasting method has to be chosen. In addition, this method has to be adjusted to the considered demand evolution. In literature, there exists an abundance of different prediction methods. Nevertheless, in practical applications, often relatively simple methods are used because more complex methods require a high amount of expert knowledge and an enormous amount of time for adaptation. Hence, potentials of complex prediction methods to improve forecasting accuracy are not tapped sufficiently. In response to this, the proposed research approach deals with a development of a technique to choose appropriate prediction methods. Based on a broadly conceived simulation study, an assignment matrix with recommendations shall be constructed. On the one hand, this matrix will contain information about suitability and accuracy of various prediction methods for different demand evolutions. On the other hand, it will include applicable parameter configurations for the adjustment to special evolutions. Hence, the matrix comprises expert knowledge for choice and adjustment of appropriate prediction methods in different application scenarios of production logistics. In order to develop this assignment matrix, firstly dynamic characteristics of different demand evolutions shall be studied. On that account, methods of linear and nonlinear time series analysis will be applied to identify characteristic patterns and classification indexes. Using these results, a classification of different demand evolutions shall be accomplished. In parallel, prediction methods will be applied to simulation based as well as to real case production data and evaluated with regard to prediction accuracy over different time horizons. Particularly, methods of nonlinear dynamics will be adapted, which can model complex dynamic time series evolutions and thus have great potential to predict customer demands. Concluding, the dynamic characteristics of the classified demand evolutions will be combined with the prediction accuracy of different forecasting methods. As the main result of the project, the developed assignment matrix will provide recommendations to choose suitable prediction methods as well as adjustments to different demand evolutions.