At a Glance
Industrial Goods and Manufacturing
Yingxi Qu created an electricity consumption prediction model that enables Covestro to more easily quantify and reduce its energy use.
Covestro wanted to enhance the design and use of its Utility Data Management System to support its energy management strategies. Focusing on two areas in particular, Covestro hired EDF Climate Corps fellow Yingxi Qu to establish a power prediction model for one of its Shanghai-based sites that would improve energy data accuracy and automation and create an electricity consumption prediction model that has a margin of error less than 5%.
Qu used the SPSS MODELER tool based on the cross-industry standard process for data mining concept. She divided the process into three steps: data filter, data partitioning (60% for training, 20% for testing, 20%for validation) and operation. Qu identified 17 types of models, including linear regression, neural networks, etc. After analyzing, she selected three as optimal models.
To calculate the error, Qu had to collect actual data for testing and verification. Having the data, she conducted python programming based on the screening results of the MODELER. This allows the model to be taken over easily, and for the results to be easily verified.
Qu’s model has the potential to save up to $200,000 per year, not to mention improve the life and reliability of the electricity grid. Based on the results of this model, Covestro has plans to build two systems for power forecasting: the power forecasting model system and the internal communication system.