Prediction of thermal storage loads using a neural network

Frank J. Ferrano, Kau Fui Vincent Wong

Research output: Contribution to journalConference articlepeer-review

25 Scopus citations


The objective of the present work is to develop a neural network computer program to predict the next-day cooling load and use this prediction in conjunction with a real-time expert system to simulate management of a cold thermal storage system. The next-day cooling load prediction allows the ice thermal storage system to maximize off-peak utility rates and minimize mechanical system operation. The management system is designed to be used by mechanical engineers in the field of air-conditioning control and maintenance. The computer program is written with the aid of commercially available neural network computer software. The temperature data base includes hourly ambient temperature values for a typical year in Miami, Florida. The technique used to predict the required ice production of the thermal storage system is conducted by training a neural network with the use of definition, fact, and training network files. Once the network is trained, any temperature pattern for a 24-hour period can be used to calculate the required next-day cooling load. Neural network training included unique temperature patterns for 'cold' and 'warm' weather fronts, as well as seasonally adjusted 'normal' temperature patterns. The resulting cold thermal storage prediction is then entered into a real-time expert system for the control of the overnight chiller operation and ice storage production.

Original languageEnglish (US)
Pages (from-to)723-726
Number of pages4
JournalASHRAE Transactions
Issue numberpt 2
StatePublished - Dec 1 1990
Event1990 Annual Meeting of the American Society of Heating, Refrigerating and Air-Conditioning Engineers, Technical and Symposium Papers - St. Louis, MO, USA
Duration: Jun 10 1990Jun 13 1990

ASJC Scopus subject areas

  • Building and Construction
  • Mechanical Engineering


Dive into the research topics of 'Prediction of thermal storage loads using a neural network'. Together they form a unique fingerprint.

Cite this