Residential Multiple Energy Consumption Indexes Prediction Based On Partitioned Parallel Modeling

In modern society,residential energy consumption has accounted for a large proportion of the total social energy consumption.With the improvement of people's living standard,this proportion will increase further.Under such a background,Residential energy consumption prediction is particularly cri tical.The energy consumption prediction can guide the design of the house,thus improving the energy efficiency of the house.Meanwhile,it can help the renters or buyers to evaluate the comprehensive using cost of their target house.In the using phase of the residence,The main energy consumption comes from the HVAC(Heating Ventilation and Air Conditioning)system,water heater and lighting system.Traditional building energy consumption prediction based on physical modeling.Although it is well supported by the thermodynamic theory,the model calculation is too complex and the real-time performance is low.It is difficult to acquire the physical model's input parameters,which needs professional personnel to collect at home.If machine learning models based on data-driven approach can be used to predict the energy consumption of residences,the real-time performance of the prediction will be greatly improved compared with the physical modeling method.Then the data analysis algorithm is used to select the effective features of the model input data combined with different predictors to reduce the difficulty of data acquisition.In this way,the energy consumption prediction model can finally provide households with the prediction service directly online,which is of great significance for lowering the threshold and marginal cost of energy consumption predictionThe effective prediction range of the data-driven prediction model is limited by the size and distribution of model training data,and there are many factors and interactions that affect the main energy consumption system of the dwelling s.The existing single machine learning model cannot accurately predict m ultiple energy consumption indexes at the same time.Therefore,this paper proposes an integrated residential energy consumption prediction model based on SVR(Support Vector Regression)and RFR(Random Forest Regression),to predict annual heating,heating,and lighting costs for the dwelling.The system runs in a cloud computing cluster and provides services to the front end through the REST API interface.The training and validation of the model is based on the SAP EPC(Standard Assessment Procedure and Energy Performance Certificate)data set,which contains 870,000 residential energy consumption data covering 13 major UK regions.Considering that a large number of data sets are used to train SVR model,the essence is to solve complex quadratic optimization problems,which consumes a lot of computing time and memory space.Combined with the strong correlation between residential energy consumption performance and its geographical location distribution,this paper proposes a parallel partition modeling method to reduce the time cost of model training and improve the accuracy of model prediction.Based on the compatibility model algorithm,a two-layer load balancing framework is designed to realize the high availability,concurrency and scalability of the system.For the test of model algorithm,this paper firstly compares the difference of test results of different machine learning models based on different input data sets to illustrate the advantages of the integrated model.Then set up parallel partition and non-parallel partition control experiment to verify that parallel partition modeling method can effectively improve the accuracy of model prediction and theoretically accelerate the model training process.Finally,based on the real energy consumption data,the prediction accuracy of the model is verified.The results show that the integrated energy consumption prediction model proposed in this paper,explain variance score higher than 0.8 and time delay less than 0.7 second,and has practical value.For system testing,this paper tests the high availability of the interface through the postman and Apache Benchmark interface automation test tool.The test results of the interface prove that the system can run stably in the cloud computing cluster and provide prediction services to the front end...