Energy management engineering : a predictive energy management system incorporating an adaptive neural network for the direct heating of domestic and industrial fluid mediums.

摘要

The objective of this research project is to improve the control and provide a more cost-efficient operation in the direct heating of stored domestic or industrial fluid mediums; such to be achieved by means of an intelligent automated energy management system. For the residential customer this system concept applies to the hot water supply as stored in the familiar hot water cylinder; for the industrial or commercial customer the scope is considerably greater with larger quantities and varieties of fluid mediums. Both areas can obtain significant financial savings with improved energy management. Both consumers and power supply and distribution companies will benefit with increased utilisation of cheaper 'off-peak' electricity; reducing costs and spreading the system load demand. The project has focussed on domestic energy management with a definite view to the wider field of industrial applications. Domestic energy control methodology and equipment has not significantly altered for decades. However, computer hardware and software has since then flourished to an unprecedented proportion and has become relatively cheap and versatile; these factors pave the way for the application of computer technology in this area of great potential. The technology allows the implementation of a 'hot water energy management system', which makes a forecast of the hot water demand for the next 24 hours and proceeds to provide this demand in the most efficient manner possible. In the (near) future, the system, known as FEMS for Fluid Energy Management System, is able to take advantage and in fact will promote the use of a retail 'dynamic spot price tariff’. FEMS is a combination of hardware and software developed to replace the existing cylinder thermostat, take care of the necessary data-acquisition and control the cylinder's total energy instead of it's (single point) temperature. This provides, besides heating cost reduction, a greater accuracy, a degree of flexibility, improved feedback, legionella inhibition, and a diagnostic capability. To the domestic consumer the latter three items are of greatest relevance. The crux of the system lies in its predictive ability. Having explored the more conventional alternatives, a suitable solution was found in the utilisation of the Elman recurrent neural networks, which focus on the temporal characteristics of the hot water demand time series and are able to adapt to changing environments, coping with the presence of any non-linearity and noise in the data. Prior to developing FEMS a study was made of the basic fluid behaviour in medium and high pressure domestic hot water cylinders, an area not well-covered to date and of interest to engineers and manufacturers alike. For this step data acquisition equipment and software was purposely created. The control software plus equipment were combined into a fully automated test system with minimal operator input, allowing a large amount of data to be gathered over a period measured in months. A similar system was subsequently used to collect actual hot water demand data from a residential family, and in fact forms the basis for FEMS. Finally an enhanced version of FEMS is discussed and it is shown how the system is able to output multiple prediction and utilise varying tariff rates.