Reseach On Forcasting Model Of Energy Saving Retrofit Of Large Scale Building Based On Data Mining

With the continuous vision of China's new urbanization and the retrofit of existing urban areas and the improvement of building energy efficiency,Large-scale building energy-saving retrofit(LSBESR)has become an important part of eco-friendly low-carbon content and an important means of improving regional energy efficiency.However,the regional building group is a multi-format cluster.Due to the characteristics of multi-system coupling,multi-device integration,and multi-parameter aggregation,the regional architectural group becomes a highly nonlinear complex system.In the actual energy-saving renovation project,we are faced with more uncertain factors and complex components,which complicates the difficulty of the project.At the same time,the domestic and international LSBESR prediction theory system is not mature,and the relevant research lacks mature predictive mathematical models and Method tools;In addition,some urban energy informatization construction leads to the accumulation of building energy consumption data.Therefore,there is an urgent need to mine the intrinsic value of data in energy-saving renovation projects.In this thesis,the data mining aided LSBESR prediction model based on data mining is proposed and verified through the construction of the basic database.Meanwhile,the proposed prediction model gives the uncertainty energy parameter quantification and correction method.The predictive correction model of the willingness factor and the multi-objective decision optimization strategy solve many core problems,including: physical modeling of regional building group prediction model is complicated,model setting parameter uncertainty is difficult to quantify,the existing models lack technical support and it is difficult to obtain the correction of the bottom-up actual energy consumption data.In particular:(1)Aiming at the characteristics of multi-dimensional data of regional building group,we constructed a city-level and regional-level LSBESR database based on GIS technology,and studied the statistical rules of energy parameters of regional buildings.On this basis,the reverse modeling method is used to establish a regional building group energy consumption prediction model based on data mining.Based on the proposed prediction model,we integrate traditional modeling methods such as regression statistics and physical modeling,and construct macroscopic steady-state energy balance equations and typical building models as well as incorporate uncertainties.In the Matlab program,we used Monte Carlo simulation(MC)to simulate the sampled data.The proposed prediction model solves the defects such as the single building assumptions and the macro statistical characteristics of regional buildings,and the high cost of physical modeling of the existing building energy simulation software.The research case of a central city shows that the deviation between the simulated value and the actual observed annual energy consumption data is 2.14%,the monthly mean variance coefficient of variation is 6%,and the energy consumption normalization index variation interval is [0.87,1.13].The hourly,day-to-day,month-by-month,and item-by-item energy consumption data that the model can output can provide a basis for predicting energy-saving retrofits.(2)This thesis carried out simulation prediction and verification of LSBESR.Through the multi-variable function relationship of regional energy consumption,the top-down back-push calibration method is used to calculate the energy-saving potential.In addition,this thesis also establishes a global sensitive parameter analysis method and a single-parameter energysaving retrofit factor regression model to prove that the internal load factor has the most significant impact on energy consumption.By adjusting the energy parameters in the prediction model,we can simulate the energy saving probability distribution curve and the distribution interval of the normalized values under different scenarios.The research case shows that the average energy-saving rate is 10.8% in the conventional scenario,and the probability of more than 5.8% is 84%,while in the low-carbon scenario,the probability of more than 22.5% is 84%.Meanwhile,the model is validated by taking 15 retrofitted buildings as samples.The comparison results show that the deviation between simulated energy savings and actual observation is 14%.The above results can support the subsequent risk decision.(3)In this thesis,we studied the quantification methods of nine uncertain sources and uncertain parameters in the LSBESR model,and then proposed an energy correction model based on Bayesian theory.According to the random error theory,the difference between the model output value and the actual observation value can be expressed by a Gaussian conditional density function.Based on the above results,this thesis established a Bayesian model for identifiable and to-be-estimated parameters.The proposed model used Hamiltonian Markov chain Monte Carlo algorithm to take the prior information of energy parameters as the initial input condition,and obtained the posterior of uncertain energy parameters by using the observation value as well as the posterior joint conditional probability distribution.In addition,this thesis solved the uncertainty of energy parameters,the deviation of theoretical models from architectural prototypes and the transmission of observation errors.The result of the case study showed that the average heat transfer coefficient of the envelope structure and the posterior distribution probability of the performance coefficient of the refrigeration system are 19% and 12%,respectively,compared with the prior distribution probability of the above parameters,and the degree of certainty is 9.5%.After correcting the energy parameters with Bayesian energy model,the deviation between the simulated and observed values is reduced from 5.4% to 0.97%.Compared with the simulated values of prior information,the deviation of energy savings is 27.8%.The above results ensure that the model can significantly improve the predictive power and accuracy of LSBESR.(4)In this thesis,a fuzzy multi-attribute decision-making retrofit willing factor prediction correction model and multi-objective decision making optimization model were built.Firstly,by constructing the membership function of the willingn factor,the probability matrix of LSBSER was obtained,and the discrete MC simulation was used to predict LSBSER based on the willingness.Secondly,the incremental cost model of unit energy-saving benefit based on the whole life cycle and the quantitative method of risk random variable were constructed,and the risk decision control optimization model based on MC simulation was established.The case study showed that the energy saving correction curve describes a unimodal normal distribution.In contrast,the energy saving investment shows a bimodal Gaussian distribution with an expected energy saving adjustment factor of 0.8,and there exists a strong Logarithmic relationship between the willing factor and the energy saving expectation.At the same time,the multi-dimensional index threshold control(including energy saving,economy,risk,etc.)can realize the trade-off judgment and gradual optimization of the energy-saving measures combination strategy package.In summary,the prediction model proposed in this thesis further enhances the performance of the traditional LSBESR model,and provides theoretical guidance for the feasibility study,planning,prediction,decisionmaking and optimization of LSBSER.