Data-driven Building Energy Consumption Prediction And Its Application Research

Building energy consumption is one of the three major energy consumers in China.In order to alleviate the pressure of energy consumption,building energy conservation has important practical significance.The key to building energy efficiency is to predict building energy consumption.Building energy consumption prediction can analyze the energy saving potential of buildings and guide the use of energy in the future.At the same time,it can also improve the efficiency of building energy consumption equipment.In the total energy consumption of the building,the energy consumption of the air conditioning system accounts for a large proportion,so the energy saving of the air conditioner can greatly reduce the total energy consumption of the building.At present,the energy consumption caused by the failure of the air conditioning system accounts for 50% of the total energy consumption of the air conditioning system,and the air conditioning fault is often not found in time,so that the fault can be found in time to make the air conditioning system operate normally and stably,and achieve the purpose of air conditioning energy saving..This paper focuses on the prediction of building energy consumption and its application.At present,the common methods of energy consumption prediction are engineering method and artificial intelligence method.Artificial intelligence method is widely used,among which the most widely used support vector machine and artificial neural network.The main content is divided into two large blocks.The first block is to study the existing prediction methods,and to improve the predictions based on the shortcomings.The second block is the new algorithm proposed in the first block.Based on this,the new algorithm is applied to specific fields to verify the feasibility of the algorithm.The specific research contents are as follows:(1)As a classic data-driven method,support vector machine(SVM)is widely used in the field of building energy consumption prediction.In this paper,based on the parameter selection problem of support vector machine,the cross-validation,genetic algorithm(GA)and particle swarm optimization(PSO)method are used to establish the building energy consumption prediction model.The verification experiment was carried out using data from an elevator company and a weather website.The input variables were daily maximum solar irradiance,average wind speed and average humidity,and electric energy consumption data for the previous two days.The output variable was the current prediction accuracy and modeling time,and the prediction accuracy is much higher thanenergy consumption data.Three prediction models are constructed and the prediction results of the model are compared.The results show that the model established by PSO-SVM is superior to GA-SVM in prediction accuracy and modeling time,and the prediction accuracy is much higher than SVM.It embodies the superiority of the algorithm.(2)Due to the problem of slow optimization of PSO,this paper proposes an improved PSO algorithm(SFPSO),which is optimized based on the particle swarm optimization(SPSO)algorithm of inertia weight sinusoidal adjustment.The period of the sinusoidal function is adjusted to 1/5 of the original,which makes the range of the particle search more accurate and the optimization performance is better.The calculation results of the benchmark function also show that the SFPSO has better global search ability and faster convergence.In this paper,SFPSO algorithm is used to optimize the penalty factor C and kernel function parameter G of support vector machine SVM,and the building energy consumption prediction model based on SFPSO-SVM is constructed.The simulation results show that the prediction accuracy of the optimization algorithm is high,and the modeling time is close to a single SVM,which proves that the prediction efficiency of the model is also improved compared with PSO-SVM.(3)This paper applies the energy consumption prediction method to the field of fault detection of air-conditioning equipment.The research data used the operating energy consumption data of the variable air volume air conditioning system in an office area of the elevator company's office building in summer cooling conditions,including three groups,two of which are operational data in the non-faulty state for training.And verify the SFPSO-SVM model;the third set of data is the operational data set with faults,used to verify the applicability of SFPSO-SVM in the field of air conditioning fault detection.The average of the statistical sliding residual and the root mean square of the sliding residual are used as the evaluation indicators,and the fault prediction quantity combined with the two sliding indicators is used to detect the fault,and several different types are tested by analyzing the fault prediction amount.The fault detection capability of the algorithm proves that SFPSO-SVM has better model accuracy and fault detection capability.