Data Reconstruction for faulty Sensor of chiller Based on Improved Neural Network Algorithm
Konferenz: MEMAT 2022 - 2nd International Conference on Mechanical Engineering, Intelligent Manufacturing and Automation Technology
07.01.2022 - 09.01.2022 in Guilin, China
Tagungsband: MEMAT 2022
Seiten: 7Sprache: EnglischTyp: PDF
Autoren:
Zhang, Zimo; He, Yan; Zhou, Chuanhui (School of Urban Construction, Wuhan University of Science and Technology, Wuhan, China)
Hu, Yunpeng (Department of Building Environment and Energy Engineering, Wuhan Business University, Wuhan, China)
Inhalt:
Most chiller sensors would have measurement faults after a long-time running, so that the maintenance would often consume a lot because it is hard difficult to identify the faulty characteristics. It is very urgent to find a convenient and effective method to reconstruct the measured data for the faulty sensor. The reconstructed measurement data can ensure the stable control and energy-saving operation for the whole central Air Condition system. In this paper, the optimized PSO-BPNN (Particle Swarm Optimization and Back Propagation Neural Network) and GA-PSO-BPNN (Genetic Algorithm, Particle Swarm Optimization and Back Propagation Neural Network) strategies were presented based-on the general BPNN (Back Propagation Neural Network) algorithm. In addition, a comparative analysis is made from three aspects: the goodness of model fitting, the accuracy of algorithm prediction and the accuracy of reconstructed data. The results based on the experimental data of ASHRAE 1043-RP show that GA-PSO-BP algorithm has the better model fitting effect, higher reconstruction accuracy and less error.