| As an essential part of educational buildings,university libraries are characterized by high pedestrian flow,high mobility,high frequency of use,and long opening hours,which makes the energy demand of university libraries high and the phenomenon of serious energy waste exists at the same time.By analyzing and modeling the historical energy consumption data of libraries,we can predict the future energy demand,identify and improve the wasteful problems in energy consumption,and formulate corresponding energy-saving strategies,which are important to achieve the goal of "double carbon" in the building field.In this paper,a university library in Xi’an is taken as the research object,and the research is conducted from two aspects of energy consumption prediction and energy saving diagnosis.The main researches are as follows:(1)Analysis of the energy consumption situation of a university library in Xi’an.The library energy consumption was analyzed by itemization based on data from the energy consumption monitoring platform to gain insight into the consumption of different energy consumption components.The collected energy consumption data,weather factor data,period data,and personnel behavior data were analyzed using Pearson correlation analysis to select appropriate features for the establishment of time series prediction models.(2)Establishment of energy consumption prediction model for a university library in Xi’an.A multi-step ahead time series LSTM-BP hybrid neural network prediction method is proposed,and a total of 18 sets of experiments are set up based on sub-item energy consumption and total energy consumption.The results show that the prediction performance of the multi-input multi-output model is better than that of the recursive model and the direct model in terms of single-item energy consumption prediction,with the average absolute error reduced by 0.96-3.46 and the root mean square error reduced by 1.33-4.34;the prediction accuracy of different models with one step ahead is better than that with multiple steps ahead,with the average absolute error reduced by 0.07-1.57 and the root mean square error reduced by 0.11-1.9.As for the prediction of energy consumption by different subsections,the prediction performance of lighting and electrical outlets and HVAC subsections is better than that of total energy consumption,with the average absolute error reduced by 1.72-7.95 and the root mean square error reduced by 2.45-10.63.(3)Diagnostic study of energy consumption and energy efficiency in a university library in Xi’an.Based on the time series prediction model,a top-down multi-level energy consumption energy-saving diagnosis method from building level to the system level and from day-by-day diagnosis to time-by-time diagnosis was proposed.The results show that different sub-energy consumptions are divided into different levels,and the more complex the building operation characteristics are,the more the number of levels are divided.The feasibility of the method to separate inefficient systems is verified,which helps to discover the reasons for the existence of energy consumption anomalies in buildings and propose corresponding energy-saving strategies.In summary,this paper proposes a multi-step ahead time series prediction model to establish a top-down multi-level energy consumption energy efficiency assessment and diagnosis method.It can not only solve the problems of feature extraction in the time dimension and the strong time dependence of building energy consumption but also extend the application of energy consumption prediction,simplify the energy-saving diagnosis process,and enhance the practicality. |
