Demand forecasting in the context of production planning system
Tereshchenko, Aleksandr (2018)
Tereshchenko, Aleksandr
Metropolia Ammattikorkeakoulu
2018
All rights reserved
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2018112919067
https://urn.fi/URN:NBN:fi:amk-2018112919067
Tiivistelmä
The thesis presents the research on the demand forecasting solution within the context of the production planning system. Since any production planning depends primarily on successful demand predictions, developing an accurate forecasting model is crucial for the success of the final solution. In turn, the production planning system will resolve multiple problems of the client company including the unjustified overproduction, the inefficiency of the production process and enable the personnel currently performing the planning manually to concentrate on other activities.
The theoretical background related to the forecasting solutions studied over the course of the research is described. Besides, the theory related to model ensembling is presented. Furthermore, six forecasting models were developed and described in the thesis. Among the models, the one- dimensional convolutional neural network is presented as an alternative to more traditional forecasting solutions. Finally, the benchmark, which uses the actual historical data obtained from the client, is described, and the techniques for model evaluation are discussed.
The benchmark findings demonstrate promising results for the majority of the products examined. Besides, it is pointed out that averaging the forecasts of multiple models resulted in more accurate predictions in the 4 cases out of the 9 analyzed. At the end of the thesis, the possible improvements are discussed, which will further increase the models' accuracies, thus enabling even more efficient operation of the production planning system.
The theoretical background related to the forecasting solutions studied over the course of the research is described. Besides, the theory related to model ensembling is presented. Furthermore, six forecasting models were developed and described in the thesis. Among the models, the one- dimensional convolutional neural network is presented as an alternative to more traditional forecasting solutions. Finally, the benchmark, which uses the actual historical data obtained from the client, is described, and the techniques for model evaluation are discussed.
The benchmark findings demonstrate promising results for the majority of the products examined. Besides, it is pointed out that averaging the forecasts of multiple models resulted in more accurate predictions in the 4 cases out of the 9 analyzed. At the end of the thesis, the possible improvements are discussed, which will further increase the models' accuracies, thus enabling even more efficient operation of the production planning system.