Energy Consumption Optimization Strategy Based On Air Conditioning System Data Prediction And The Impact Analysis Of Uncertainness On The Strategy

To achieve the carbon peak goal,China formulates the carbon peak action plan and encourages the development and application of various low-carbon technologies.Data-driven study in the heating ventilation and air conditioning(HVAC)system is a research hotspot.However,most of the studies concentrate on improving prediction performance or fault detection and neglect the application of the predicting results and the impact of uncertainness on predicting results.Based on machine learning theory,particle swarm optimization(PSO)theory,and real options analysis(ROA)theory,this study builds up a comprehensive framework including data monitoring,data application,and the impact of uncertainness on loss mitigation strategy investment analysis.The main contents of this paper are as follows:Firstly,the meta-learning prediction model is built based on an artificial neural network(ANN)recommendation system,and the optimal data monitoring scheme is determined by multi-objective decision-making algorithms(MODMA)and user preferences.The inputs of the meta-learning model are meta-features(e.g.,the length and variance of data),and the output of the meta-learning model is the best prediction performance model in the model pool(e.g.,long short-term memory network).The meta-learning method could predict the best prediction performance model in advance for a certain dataset.In the multi-objective decision-making process,the objective properties include user preference for data monitoring system,and the subjective properties include prediction accuracy and the prediction time consumption et al.The results show that the quality of data affects the prediction performance most and even traditional time series predicting model can obtain high accuracy in most cases.The recommendation success rate predicted by the meta-learning model remains above 99%,but the success rate is impacted by the complexity of the input dataset.Mixed datasets(e.g.,a dataset with COP and cooling load data)obtain worse success rates than pure datasets.The option of“a prediction system combined with a monitoring system”remains the best choice for users,and this option remains superior when the length of data is changing.Secondly,the K-means and Affinity propagation methods are utilized to make the clustering analysis of the thermal zone’s energy consumption,then the bottom-up cross-scale load prediction model from thermal zone scale to building scale is built based on the clustering results.The key idea of the cross-scale model is to use the similarity of the load characteristics in one cluster and utilize a load of the cluster centroids to predict the other cluster members’load.Summarizing the prediction results of all clusters,the cross-scale prediction accuracy on the building scale is obtained.The results show that the cross-scale predicting accuracy is 0.21for normalized root mean square error,and the accuracy is improved to 0.15 for cumulative load within two hours.At the same time,the number of monitoring points is reduced by five times by the cross-scale load prediction model.Regression results show that the relationship between cross-scale prediction accuracy and the ratio of each cluster’s sample size is fitted well by a quadratic polynomial.Logistic regression proves that the maximum mean accumulative load,mean accumulative load at start and end time of the HVAC system,the“non-equilibrium factor”of the west to the south wall,and exterior window area determine the cluster belongings.Next,a new HVAC energy consumption optimization strategy is built based on the traditional optimal chiller loading(OCL)model.Not only the power consumption for chillers but also for air conditioning terminals are included in the objective function,and the constraints are set based on the prediction results of chillers.To let the load supply of chillers matches terminals faster,a penalty coefficient is calculated based on energy flexibility to amender the terminals’energy consumption.The results show that the energy-saving ratio of the proposed strategy is between 6%～7.5%,and the penalty coefficient is more sensitive than the cooling temperature set point to the air conditioning terminals’energy saving.Moreover,the energy-saving ratio of air conditioning terminals is larger than that of chiller units.The discounted payback value is 5.8(year)and the NPV value is 58.88 ten thousand yuan in 300 hours for the proposed optimization strategy,which is substantial under low additional investment.Finally,the impact of climate uncertainness and its propagation in the proposed energy optimization strategy is analyzed.The analysis is focused on the impact of climate change uncertainty on building energy consumption and investment strategy for loss mitigation projects.The future weather files under different radiative forcing are generated by typical meteorological year weather files and the Morphing procedure to reflect the climate change uncertainty.Moreover,future building energy simulations are made based on future weather files and three mitigation strategies:horizontal shading,PV panels,and“horizontal shading+PV panels”.Individual and sequential investment analyses are made based on ROA theory to optimize investment strategy.The results show that the energy consumption loss due to climate change is about 12615 k W·h per year on average and grows at the rate of 4.15%.The impact of climate change uncertainty on building energy is 0.04%～0.34%relative to the total building energy consumption,and radical mitigation strategies amplify this impact.If only one project is allowed,the shading strategy is the best choice,if two projects were allowed,the PV panel should be invested first,and the shading strategy should be adopted later.The climate change uncertainty defers the optimal investment time,but not enough to change the optimal mitigation options for individual investments,and the optimal sequence of investments in sequential investments.Sensitivity analysis shows that the increasing discount rate and cost growth rate advance the optimal investment time,and discount rate and option premium have an obvious exponential relationship.A“marginal effect”occurs in project return with the increase of forcing,at the same time,the benefit of using ROA theory occurs the“increase first then decrease”trend.The amount of“ROA using benefit”change due to unit dry bulb temperature is 3.01×10~6 CNY to 5.7 7×10~6 CNY.