Research On Energy Consumption Prediction And Anomaly Detection Of Building Air-conditioning System

Energy consumption of building air conditioning system has been a hot and difficult subject in the field of building energy conservation.How to accurately and stably predict and detect energy consumption is a bottleneck problem to be solved,while the numerical simulation cannot achieve the desired effect.This dissertation,by using the technology of data mining to deeply investigate the characteristics of building air-conditioning system energy consumption,builds the energy consumption of building air conditioning system energy consumption prediction framework and anomaly detection framework to specific solutions to the key issues of building air conditioning system energy saving.The energy consumption prediction framework includes the short and medium and long-term energy consumption prediction models,while the anomaly detection framework includes the abnormal energy detection and anomaly grading model.According to the characteristics of energy consumption of air conditioning systems in different types of buildings,this dissertation takes office buildings,commercial buildings,and residential buildings as the research objects,and fully verifies the reliability of various models proposed in the two frames and a relatively complete energy consumption monitoring system of building airconditioning system is formed.The main research contents of this dissertation are as follows:Aiming at the stability of short-term energy consumption prediction of building air conditioning systems,this dissertation proposes an ensemble learning prediction strategy based on data decomposition.Firstly,the variational mode decomposition(VMD)algorithm is used to decompose the complex original energy consumption data set into the regular subsets,and then the short term energy consumption prediction model of each subset is established based on the machine learning method,finally,the results of each prediction model are combined through ensemble learning to obtain the prediction results of the original energy consumption data set.To verify the superiority of the proposed strategy,Empirical mode decomposition(EMD)algorithm is introduced as the reference model,and the generalization of each model is tested by Friedman test and Nemenyi test.The results show that the predictive performance and stability of the ensemble learning model based on VMD are better than that of the EMD based model and the single machine learning model.Aiming at the problem of low accuracy and lack of reliability of medium and long term energy consumption prediction results of building air conditioning system,the essential characteristics are deeply analyzed,the point prediction model is converted into the interval prediction model,and an interval energy consumption prediction model combining deep learning and penalty quantile regression is proposed.This dissertation adopts the long shortterm memory(LSTM)model to extend the prediction result of a one-time step to the long term prediction through the iterative method and then carries out the penalized quantile regression(PQR)to obtain the interval prediction result.The results show that the performance of LSTM is better than other common machine learning models in medium and long term energy consumption prediction.The interval prediction results of energy consumption can cover the variation range of actual energy consumption,which indicates that it has high prediction accuracy.Since it is difficult to identify the abnormal energy consumption of building air conditioning systems with limited information,this dissertation proposes an abnormal energy consumption detection model based on the machine learning method and statistical rules.Firstly,one time-step ahead energy consumption prediction model is established based on the characteristics of energy consumption change in the air-conditioning system.Then,the residual between the predicted value and the actual value of energy consumption is calculated.Based on the residual data set,an abnormal energy consumption detection model is established according to specific statistical rules.Since the energy consumption data set lacks a clear label for abnormal energy consumption,this dissertation analyzes the rationality and reliability of abnormal energy consumption detection results based on the conclusion of field investigation.To further verify the advantages of the proposed model,common anomaly detection methods,namely local outlier factor(LOF)and Z-score method,are introduced to compare with the model.The simulation model of energy consumption of the target building is established,and the reliability of the abnormal energy consumption detection method proposed in this dissertation is fully verified by adding abnormal energy consumption data related to human operation.Finally,given the general situation of operation monitoring of building air conditioning systems,this dissertation proposes an abnormal energy consumption detection and grading model under the background of only energy consumption information.Based on the results of the energy consumption prediction model,the quantile regression(QR)model is used to identify the abnormal energy consumption,and then divide the abnormal energy consumption is into different grades.The change of the energy consumption in a short time and the energy-using habits of the inhabitants are used as judgments on the rationality of detected abnormal energy consumption,and the range of different QR correspond to the different levels of anomalies.In addition,the standard residential building model is used to simulate the change of energy consumption under the condition of the dirty fault of the chiller,to verify the abnormal energy consumption detection and classification model proposed in this dissertation,the results show that the presented anomaly detection model can reliably detect the reasonable energy consumption system that exists in the abnormal values of energy consumption,and the proposed classification benchmark can be used to judge the severity of the abnormal energy consumption.This dissertation introduces the modal decomposition method to improve the prediction accuracy and stability of the short term energy consumption of building airconditioning system and adopts the interval prediction to improve the reliability of the prediction results of medium and long term energy consumption prediction.On this basis,this dissertation combines statistical rules and system operation characteristics to detect abnormal energy consumption,then puts forward the judgment model and standard of it.Aiming at the general problem of limited energy consumption information of airconditioning system,the grade division standard of abnormal energy consumption severity is further established,and the anomaly detection system of building air-conditioning systems is improved.