Literature research (II): quantitative evaluation of building energy efficiency performance based on short-term energy prediction

Quantitative evaluation of the building energy performance based on short-term energy predictions

This is a ENERGY2021 An article in 1996 , Impact factors of the Journal 6.082.

Abstract

This paper presents a building energy evaluation method based on short-term energy prediction . First , be based on RNN, use MIMO（ Multiple inputs, multiple outputs ） Strategy , Established 24 Hourly building energy consumption prediction model ; secondly , A quantitative energy evaluation strategy is proposed , be based on 1-D k-means Clustering to quantify prediction gaps ; then , Yes 5 A case study of a real building , To verify the reliability of the proposed method . Besides , This paper also analyzes the absolute percentage error of each time step （APE） The change of , Deeply understand the energy-saving performance of buildings , And according to the building details 、 Separate energy performance , Build a customized energy quantification system for it . Building energy consumption is achieved by APEs Mark as multiple levels to quantify .

brief introduction

The author thinks that , It is very important to make a comprehensive evaluation of building energy efficiency performance , Objectively evaluate the energy performance before and after building renovation , It is conducive to the selection and deployment of appropriate energy-saving technologies .
Building benchmark is an effective evaluation method , However , Due to the uncertain climate 、 Habitation behavior 、 Plug in load equipment, etc , The dynamic energy performance of a single building is difficult to be reflected by the monthly average and annual average , It is necessary to shorten the time span , Such as by day or by hour . At this point , Building energy forecasting may be an ideal solution .
There has been a lot of research on energy consumption prediction , But they are applied to different advanced energy management systems , Few studies have applied them to building energy consumption evaluation .
Based on the above questions , This paper has done the following research ：

• A quantitative energy evaluation method is proposed , Used to evaluate the short-term performance of a single building ( That is, every hour ) Energy use .

• A method of energy evaluation using short-term energy prediction is proposed , And the energy quantification system is established by quantifying the energy prediction gap .

• A customized energy quantification system has been established for each building .
  say concretely , The main contents of this paper are as follows ：

• adopt MIMO Strategy , be based on RNN A multi-step ahead short-term building energy consumption prediction model is developed , And visualize and analyze the gap between predicted and actual energy use .

• be based on 1-D k-means clustering algorithm , A quantitative energy evaluation strategy is proposed to quantify the gap , And customized a variety of energy quantification systems for different buildings .

• Experiment on five buildings , Verify the reliability of the proposed quantitative energy assessment method , And the characteristics of their energy utilization are deeply analyzed .

The basic theory

This part introduces the theoretical basis of this paper , involves ：

1. RNN and LSTM
2. Multi-step ahead energy predictions, It mainly includes recursive strategy 、 Direct strategy and MIMO Strategy , In this paper MIMO Strategy , Pictured 5 Shown
   
3. Predict performance evaluation indicators ：AE、APE、CV-RMSE、MAPE
   
   
4. 1-D k-means clustering
   This paper does not use the traditional clustering method , They chose to Ckmeans.1d.dp: Optimal k-means Clustering in One Dimension by Dynamic Programming An optimized one-dimensional clustering algorithm .

Method

The quantitative building energy consumption evaluation method proposed in this paper is shown in the figure below , There are three stages .

Short term energy forecasts

1. Model inputs and outputs

In this paper MIMO Strategies for energy forecasting , Model input 168h（1 Zhou ） The data of , forecast 24h The data of . in addition , The input variables for each time step can be divided into two parts , Part is historical measurement , Including power consumption 、 temperature 、 Relative humidity ; The second part is the information about time , Such as hours 、 Day and month , Namely 24、7、12 Discrete variables of level , Turn them into one-hot form , Then we get a 24 Columns of the matrix , alike , Output is the future 24 Hour prediction , Should be 24 Dimensional .
In this paper LSTM combination MIMO As a model for short-term energy prediction .

2. Quantitative energy assessment strategy

By comparing the predicted value with the actual value , The established short-term energy prediction model is called the benchmark of energy evaluation , Use models to predict , If the predicted value is exactly the same as the real value , It indicates that the current energy consumption is reasonable . actually , No matter how you optimize it , The predicted value of the model can never be exactly equal to the real value , But the gap between them should be kept within a certain range .
The difference between the predicted result and the real value is grouped into different clusters by cluster analysis , To quantitatively assess these gaps , Pictured 7 Shown , according to APE The clustering results are divided into levels with different dimensions , Establish energy quantification system .