项目名称: 基于数据挖掘的空调负荷预测方法研究
项目编号: No.51308560
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 建筑科学
项目作者: 何大四
作者单位: 中原工学院
项目金额: 24万元
中文摘要: 空调负荷预测是空调系统优化运行的基础。空调负荷预测过程存在要预测的数据个数多、采集到的样本数据含有一定的噪声、受诸多气象因素影响并具有随机性等特点。目前广泛使用的神经网络负荷预测方法存在过拟合、泛化能力差、抗干扰能力差及移植难等缺陷,使得已有负荷预测方法难以实用化。本研究结合空调系统负荷实测过程中容易产生大量连续的含有较大"噪声"的数据的特点,提出基于熔合机制和聚类特性的脏数据清洗方法,来消除历史数据中所包含的噪声数据、错误数据及缺失数据,以改善历史数据的质量。另外,通过构建新的启发式约简算法,利用粗糙集理论来合理确定预测模型的输入参数。从而能有效改善负荷预测中数据预处理和确定模型输入参数这两个环节的泛化能力,提高预测模型的适用性。研究成果将有力地促使空调负荷预测方法走向实用,也为建筑物的低碳运行提供了有力的技术支撑。
中文关键词: 数据挖掘;负荷预测;泛化能力;数据清洗;支持向量机
英文摘要: High Accuracy of the air conditioning load forecasting isa key for the optimal control of a HVAC system.Air conditioning load forecasting process of the data to predict the number of sample data collected with a certaindegree of noise by many meteorological factors and the randomness characteristics.Widely used neural network exists over-fitting and poor generalization ability, pooranti-interference ability , difficulty to transplant and other defects, wich make the load forecasting methods to be difficult for practical use. It is easy to produce a large number of successive data containing noise when measuring air-conditioning load. In this study, the dirty data cleaning method based on the fusion mechanism and clustering characteristic are puts forward to eliminate the noise data, erroneous data and missing data. The data cleaning method improve the quality of the historical data. With constructing a novel heuristic reduction algorithm, rough set theory is used to reasonably determine the input parameters of prediction model. Thus the generalization ability of .load prediction method will be effectively improved. The research result will strongly promote the air-conditioning load forecasting method to practically employ, and provides strong technical support for building low-carbon operation.
英文关键词: Data Mining;Load Prediction;Generalization Ability;Data Cleaning;Support Vector Machine