Machine Learning Based Energy Consumption Prediction Of Building HVAC System And Model Optimal Dispatch

Nowadays,under the increasingly serious influence of severe weather,the buildings cannot be ignored as major energy resource consumers.How to carry on the effective energy management and save energy resources,optimize the energy structure has received the widespread attention.And heating ventilation air conditioning system(HVACS),as one of the main energy consumption equipment in buildings.Energy consumption prediction,energy saving optimization for HVACS both will have an important realistic significance and practical guiding role to reduce the cost of building and the global energy consumption reductionIn this framework,a data-driven based method is used to predict a large office building energy consumption in the north of China,simulate its whole year energy consumption and optimization scheduling.At first,this thesis collected the building's energy consumption data,air conditioning energy consumption and data of operation since November 2016 to March 2017.Via artificial eliminate the abnormal data after the bad sample points,using principal component analysis algorithm to achieve a dimension reduction data set.After that,using an ensemble learning algorithm to build an energy prediction model according to the dimension reduction sample set.Then,refer to prediction result earn whole year energy demand of this office building via Energy plus.Finally,comprehensive analysis of the overall energy consumption,heating load and energy consumption of HVACS build an energy hub model with optimal dispatching.Adaboost-BP is an ensemble learning algorithm which can improve the prediction accuracy of BP neural network.Furthermore,it can revise shortage of BP neural network such as slow convergence and existence of local minimum.Besides,this ensemble learning algorithm has low requirements for the weak classifier and hardly needs to adjust its parameters,so it has a wide range of application and good robustness.The prediction accuracy of each month increased from 86%,89.01%,89.89% and 81.16% to 88.13%,90.31%,90.14%,89.16% and 85.91%,respectively.Especially for the algorithm model with poor self-classification effect,the improvement effect was more obvious.According to the physical characteristics and climatic characteristics of the office building,this thesis combines the data obtained from Energy consumption prediction model with the simulation method Energy Plus to obtain the reference significance of the annual Energy consumption of the building,including the electricity demand,heating demand,energy consumption of HVACS and other data.According to its energy consumption characteristics,an energy hub system with multiple optimization objectives is built and optimized in MATLAB environment.Finally,an energy hub system with flexible thermoelectric ratio of cogeneration(CHP)equipment is formed.The CHP ratio changes within a certain range in response to the ratio between the power consumption of HVACS and the overall thermal load of the building.Compared to energy hub with a stationary CHP system,this novel energy hub system can save 4.5% of the economic cost,2.9% of the natural gas,3.3% of the electricity input and 3.2% of the CO2 emission under the regional 2017 climate conditions.In addition,the energy hub system is conducive to realizing the cascade utilization of energy and has practical guiding significance for the absorption and application of renewable energy.In conclusion,this thesis present a data-driven based energy consumption prediction and optimization scheduling strategy,which has benefit to optimize structure of office building energy consumption,save energy resources and reduce greenhouse gas emissions.At the meantime,this strategy has good effect to reduce the peak load impact to main grid and also provides a guide idea for the design of the distributed energy network pattern.