Font Size: a A A

Study On Load Prediction Model And Optimal Control Of Ice Storage System

Posted on:2003-10-27Degree:DoctorType:Dissertation
Country:ChinaCandidate:J WuFull Text:PDF
GTID:1102360095455024Subject:Engineering Thermal Physics
Abstract/Summary:PDF Full Text Request
Ice storage system revived in 1990s because it can move the power used from day to night. As the power price of day is higher than that of night, ice storage system can help to save power charge. Chiller priority is the most common control strategy for existing ice storage systems. But only optimal control can significantly reduce the operating cost to the minimum. The accuracy of the load prediction is a key for optimizing the system control. This thesis takes the ice storage system building in Construct Bank Hangzhou Mansion as research example, mainly deals with the artificial neural network model for load prediction, and compares the operating cost and COP of optimal control strategy with that of chiller priority strategy.Firstly, author sets up a Self Organization feature Map (SOM) model for identifying the cooling load. Applying SOM to identify cooling load has not been reported up to now. This thesis initially reveals the relationship of cooling load of different weekdays. Back Propagation (BP) algorithm is described in detail and various improved methods to BP algorithm are introduced. Author also discusses how to improve the generalization of neural network. The two methods for improving generalization, "regularization" and "early stopping" are presented. The data preprocessing techniques, which can improve the efficiency of network training, are also discussed.Temperature predicting ANN model for the next 24 hours is established. By this method, Mean absolute error (MAE) is reduced to 0.4512癈 from 0.6663癈 that is calculated by improved ASHRAE calculation method. Mean relative error (MRE) is reduced to 1.36% from 2.02%.Based on a unique day cooling load predicting ANN model, day cooling load predicting ANN model for workday and holiday is established respectively. Then, MRE reaches 3.21% for workday and 5.96% for holiday.A unique next 24 hours hourly cooling load prediction ANN model is established. According to the research results from SOM model, 8 sub neural network is adopted in inner and MAE of hourly cooling load prediction is reduced 80.64kWh. Expected error percentage (EEP) is reduced to 3.27%.Next 24 hours hourly cooling load prediction multi-output dynamic model is established and prediction accuracy is improved again. MAE of hourly load prediction reduced to 65.07kWh and EEP reduced to 2.60%. This kind of model has not been reported by literature.A cost-minimum model for ice storage system is established and numerical calculation is carried out. When cooling load is less than ice-melting ability, optimal control strategy is just ice priority, optimal control saves 24-45% operating cost compared with chiller-priority. When cooling load is more than ice-melting ability but still less than ice-melting ability plus half of chiller cooling ability, optimal control is to keep chiller load to half of chiller cooling ability and tune ice-melting to meet load, optimal control consumes 13-20% more power but saves 9.2-11% operating cost compared with chiller-priority. When cooling load is more than ice-melting ability plus half of chiller cooling ability, optimal control is just ice-priority again, optimal control is to keep ice-melting to maximum and tune chiller to meet load, optimal control consumes 0-13% more power but saves 0-9.2% operating cost compared with chiller-priority.
Keywords/Search Tags:Ice storage, Artificial neural network, Load prediction, Optimal control
PDF Full Text Request
Related items