Research On Energy Consumption Prediction Of Office Building By Neural Network

In the context of"new era,new Shandong,new kinetic energy",to further achieve sustainable development,to achieve the goal of energy saving and emission reduction,social energy consumption must be controlled and predicted.At this stage,in the total energy consumption of national society,the proportion of building energy consumption has accounted for 26.7%,and with the development of the economy and the acceleration of urbanization,building energy consumption will show a rapid growth trend.The share of energy consumption in office buildings has also gradually increased,and it has the characteristics of large volume,uneven distribution,and difficult control.Therefore,realizing the forecast of the annual energy consumption of office buildings can not only reduce the labor,material and financial costs of energy consumption research,but also be of great significance to accelerate the construction of low-energy or even ultra-low-energy office buildings.This paper establishes an office building model through DeST-c software,and considers the influenceofenergy consumption from three aspects:outdoor meteorological parameters,indoor thermal disturbance intensity and thermal performance parameter of building envelope structure.Setting the heat transfer coefficient of external walls,roofs and exterior windows,and thermal disturbance intensity of indoor personnel,lighting and equipment,40 sets of annual hourly energy consumption simulation calculations were carried out to obtain simulation results of hourly and hourly energy consumption of office buildings.The minimum value of total energy consumption of the building is 85.19kWh/m~2·a,and the maximum value is 92.67 kWh/m~2·a.The air conditioning energy consumption,lighting energy consumption and equipment energy consumption account for 24%,44%and 16%respectively.Subsequently,The neural network toolbox in MATLAB software is used to construct the Back-ProPagation neural network building energy consumption prediction model.According to the specific conditions of the network input layer,hidden layer and output layer,the network structure is determined to be 16-4-1.After that,the network epochs,expected precision goal,learning target lr and other attributes are set,running the network to get the best performance,and then 40 sets of energy consumption basic arrays calculated by DeST-c are input into the network for learning training,finally,the energy consumption prediction of 10 sets of typical arrays in the base array is implemented.After the energy consumption simulation calculation and prediction,the simulation results were analyzed separately.The analysis found that nearly 55%of the40 data samples simulated by DeST-c meet the energy consumption limit of the office building in Shandong Province;The predicted 10 sets of energy consumption values were analyzed with the original data.The minimum error was 0.4%,the maximum value was 4.4%,and the basic error value was 2%.It can be concluded that neural networks are a reliable tool for predicting the annual energy consumption of buildings.Finally,10 actual office buildings were selected,and the annual building energy consumption value of the office building and the thermal performance parameter values of the envelope structure were statistically summarized,and the energy consumption per unit building of each building was calculated.Furthermore,BP neural network program is used to predict energy consumption of 10 actual buildings,The predicted energy consumption values are 98.4512 kWh/(a·m~2),88.4231kWh/(a·m~2),100.4339 kWh/(a·m~2),90.4512 kWh/(a·m~2),78.9982 kWh/(a·m~2),88.4548kWh/(a·m~2),86.4379 kWh/(a·m~2),88.4336 kWh/(a·m~2),98.4113 kWh/(a·m~2),85.3907kWh/(a·m~2).The predicted value is compared with the measured value,the results show that the two are more consistent.The feasibility of BP neural network in building energy consumption prediction is further verified.This paper studies the energy consumption prediction of office buildings in Shandong Province based on neural network algorithm,the feasibility of this prediction algorithm is verified from both theoretical and practical aspects.This not only provides a novelmethod for studying the energy consumptionforecast of office buildings,but also laids the foundation for the revision of relevant standards for energy consumption in public buildings in the later period.