基于KD-LSTM的暖通空调能耗预测研究  

Research on Energy Consumption Prediction of HVAC Based on KD-LSTM

作　　者：向元柱 钟永彦[1] 陈娟[1] 丁士旵 XIANG Yuan-zhu;ZHONG Yong-yan;CHEN Juan;DING Shi-chan(College of Electrical Engineering,Nantong University,Nangtong Jiangsu 226019,China)

机构地区：[1]南通大学电气工程学院,江苏南通226019

出　　处：《计算机仿真》2025年第3期134-139,144,共7页Computer Simulation

基　　金：国家自然科学基金项目(62273188)。

摘　　要：针对暖通空调系统使用单一模型预测精度较低且长期能耗预测需要大量数据的问题,提出了一种基于KD-LSTM(K-means DBSCAN LSTM)的暖通空调短期能耗预测方法。该方法利用K-means和DBSCAN算法,对能耗数据进行聚类分析;并对每一类数据进行异常检测识别,采用KNN(K-nearest neighbors)方法修复异常数据;使用修复数据训练长短期记忆神经网络LSTM预测模型,实现了暖通空调系统能耗数据的短期预测。以某高校图书馆暖通空调系统为研究对象,使用KD-LSTM方法对三种工况下的能耗数据进行了预测,其均方根误差分别下降11.3855kWh、0.8484kWh、0.1505kWh,相关系数达到99.401%、98.267%、96.486%,验证了上述方法的有效性,可进一步优化图书馆暖通空调系统的能耗管控。In view of the low prediction accuracy of a single model for HVAC systems and the need for large amounts of data for long-term energy consumption prediction,a short-term energy consumption prediction method for HVAC systems based on KD-LSTM(K-means DBSCAN LSTM)was proposed.The method utilized K-means and DBSCAN algorithms for clustering analysis of energy consumption data.It performed anomaly detection and identification for each cluster and employs the K-nearest neighbors(KNN)method to repair the abnormal data.The repaired data was then used to train a long short-term memory neural network(LSTM)for short-term energy consumption prediction in HVAC systems.Taking a university library HVAC system as a case study,the KD-LSTM method was applied to predict energy consumption data under three operating conditions.The root mean square errors were reduced by 11.3855 kWh,0.8484 kWh,and 0.1505 kWh,respectively,and the correlation coefficients reached 99.401%,98.267%,and 96.486%.These results validate the effectiveness of the method and demonstrate its potential for optimizing energy consumption management in the library's HVAC system.

关 键 词：暖通空调 能耗预测 长短期记忆 

分 类 号：TU83[建筑科学—供热、供燃气、通风及空调工程] TP391.9[自动化与计算机技术—计算机应用技术]