Data Demand And Data Quality Enhancing Method Of Refrigeration And Heat Pump System Digital Twin Model

With the rapid development of big data,internet of things and artificial intelligence,the optimization method based on digital twin model is becoming the most potential approach to reduce the operation energy consumption of refrigeration,air-conditioning and heat pump systems.However,the energy-saving application of digital twin is based on huge amounts of actual data,and there are many data quality problems occurred in refrigeration and heat pump system operation data acquisition,such as missing data,abnormal data,error and so on.This is due to the limited sensor accuracy,the unstable connecting network and the working environment with much noise and interference.These data quality problems will deeply decrease the accuracy of digital twin models which are developed on the actual operation dataset.Thus,data quality problems have become the key difficulty when using digital twin model to reduce the operation energy consumption of refrigeration and heat pump systems.In order to improve the energy-saving application of digital twin model and intelligent operation optimization,this study focuses on the solutions of data quality problems occurred in the actual data acquisition of refrigeration and heat pump system.Data demand theory and data quality enhancing method based on soft sensing have been proposed to solve those data quality problems like “big but incomplete”and “big but inaccurate”.In the virtual case study and the actual case study,the improvement of data quality,digital twin model’s accuracy and the data quality enhancing effect on energy-saving operation have been verified,analyzed and discussed.The main content of this dissertation is as follow:1.Analyzing the properties and effects of common data quality problem in the actual data acquisition process of refrigeration and heat pump system.A virtual experimental bench for the simulation of data quality problems is set up based on mechanism simulation model and Monte Carlo method.The virtual data acquisition and the random data quality problem generating process can be simulated on the virtual experimental bench.The effect of data quality problems on data-driven modeling is analyzed based on three typical machine learning models: polynomial model,neural network and decision tree.The results indicate that all kinds of data quality problems in dataset can lead to the decline of prediction performance of machine learning model in varying degrees,and the effect is related to the type and severity of data quality problems and the model type.2.To made up for the lack of key variable collection in actual data acquisition,a data and sensor demand theory for refrigeration and heat pump system digital twin modeling is proposed.According to system structure and requirement of modeling,the data and sensor demand theory can provide the necessary number of different kinds of sensors by mechanism analysis.The theory can be in common use for different refrigeration and heat pump systems.An actual system can be divided into some modules and represented in an abstract way based on graph theory.Then the necessary number of different kinds of sensors can be analyzed in both system level and module level.The analysis results can support the design of data acquisition solution to avoid the deficiency of key variable measurement in refrigeration and heat pump actual data acquisition.3.To solve the common data quality problems in actual data acquisition of refrigeration and heat pump system,like missing data,abnormal data,lack of condition distribution,error and noise,the data quality enhancing method based on soft sensing is proposed.By mechanism modeling of local object in system and model uncertainty analysis,the soft sensing model can be set up and applied to recover missing data,detect and recover abnormal data,enhance the adequacy of condition distribution(with adequacy function)and calibrate data.By choosing an air source heat pump water heating system as the research object,performance of the data quality enhancing method is verified and analyzed in the virtual case study and the actual case study respectively.The results indicate that the data quality enhancing method can synthetically improve the data quality of actual dataset.And with the improvement of data quality,the prediction accuracy of digital twin models can be also improved.In the actual case study,the average prediction error of digital twin model can be decreased by about 70% while applying the data quality enhancing method,this indicates that the application of soft sensing data quality enhancing method can effectively improve the prediction performance of digital twin model.4.Based on the study of data quality enhancing method,the application of this method in actual system energy-saving operation optimization is further studied with a sample of air source heat pump water heating system.The operation strategy of actual system can be optimized by setting up and applying the operation optimization methodology system with four steps: operation data acquisition,data quality enhancing,digital twin modeling and model-based operation strategy optimizing.The operation strategy before and after optimization is tested on the actual system,and the energy-saving potential is analyzed by using the actual data and the verified digital twin model.The results indicate that,based on the digital twin model set up by quality enhancing data,the optimized operation strategy achieved 8.4% and 8.6%energy cost saving in the two test weeks,respectively.The comparison of operation performance with and without quality enhancing data shows that the data quality enhancing method has potential in actual energy-saving optimization and plays a key role in the successful application of digital twin model-based energy-saving operation optimization strategy.Finally,the author briefly expounds the shortcomings of this work and share suggestions on future investigations.