Scheduling air conditioner testing tasks under time-of-use electricity tariff: A predict in and for optimization approach
Xinjun Lai,Ke Zhang,Zhantao Li,Ning Mao,Qingxin Chen,Sheng Zhang
DOI: https://doi.org/10.1016/j.cie.2022.108850
2022-12-04
Abstract:A prototype air conditioner should be tested under various conditions. However, creating such conditions consume a large amount of electricity. Particularly, the testing time durations and electricity costs of tasks are unknown and sequence-dependent: if the sequence changes, the duration and cost of each task may also significantly change. Therefore, it is difficult for air conditioner testing companies to schedule the testing tasks to benefit from the time-of-use (TOU) electricity tariff policies. To resolve this problem, we formulate a task scheduling problem under the TOU tariff, and we propose a "predict in and for optimization" solving method. (1) By embedding a forecasting method called random forest (RF) into a genetic algorithm (GA), an RF-GA hybrid algorithm is developed to solve the optimization problem where task duration and cost are sequence-dependent. Based on the characteristics of the scheduling problem, it is abstracted to a travelling salesman problem, so that a suitable GA operator is selected. (2) With the solutions generated from the RF-GA, a surrogate model is developed with a broad learning system, which is efficient in computation and satisfying in prediction. In addition, a smart system is proposed to reduce the extremely high dimensionality in the surrogate model. Numerical examples, as well as a real-world case with sensor data, are used to evaluate the framework. Results show our method can save up to approximately 41% on energy consumption, corresponding to approximately 1.7 million CNY savings on electricity costs each year for our studied company.
computer science, interdisciplinary applications,engineering, industrial