Consultancy Service Example
Extracting the maximum information from experimental data

When realistic data is available from a reasonably designed and executed experiment or monitoring exercise, the process of data analysis is one of extracting the maximum value from that data within the bounds of the validity, accuracy and relevance of the data. The challenge is recognising and overcoming the limitations of many data sources.

Unfortunately it is not possible to show analyses performed for clients due to confidentiality, so to illustrate this service area an example analysis is shown where the maximum benefit is extracted from a sparse data set to identify the benefit of programmable radiator valves replacing simple thermostatic radiator valves and the benefits of lighting a wood burning stove providing simple room heat output. The analysis example is summarised below and a short report can be provided on request - click on the Redford-Tech logo above to go to your email with the correct address.

Identifying the fuel saving benefits of programmable radiator
valves and wood burning stoves in a domestic property

- Identify the savings in fossil fuel burning (oil in this case) resulting from installing programmable radiator valves on some radiators in the home
- Wood Burning

Identify the savings in fossil fuel burning resulting from lighting a wood burning stove during cold winter days (room heat output only)

Electronic TRV


The raw data - temperature and oil consumption over three years
Oil consumption measured by periodic tank dipping

Data Sources
- Readings of oil consumption by periodic 'dipping' of the tank with additional notes identifying stove lighting
- Daily average temperatures derived from a nearby, publically available weather station (West Cheshire College, Chester, UK)

- Oil measurements by dipping are prone to reading error, ice forming in the base of the tank and temperature related volume change
- Weather station data sometimes contains 'holes' and other irregularities such as varying measurement periods

Analysis Process
- Data 'cleansing', repair, alignment, averaging, characterisation, etc. - the most time consuming and frustrating element of the task!
- Choosing an appropriate model for the underlying process - in this case a degree-day type model based on consumption being proportional to heat loss which is proportional to temperature difference between inside and outside the home but also taking account of constant daily demand for hot water
- Segregation of data to identify different events - when fire is lit, before and after radiator valves fitted
- Choosing appropriate statistical tools to gain the maximum information from the available data - minimum-variance fit to segregated data
- Verification that the model(s) reproduce sensible consumption data - generate estimates from different weather records to make sure they give a consumption result close to the recorded data

Sinusoidal fit

Initial fit of sinusoidal characteristics to the temperature and oil data, showing as expected that the oil consumption is 180 degrees out of phase with the temperature data - i.e. the colder it is the more oil is consumed. Red and blue sinusoids show the typical maximum and minimum temperature ranges, which show how cold 2010/11 winter was. Note that minimum temperatures (on average) are slightly later than the winter solstice, demonstrating the effect of ground mass on external temperature.

Weighted fit

Weighted minimum-variance fit to segregated data to identify effects
of programmable radiator valves and stove lighting

The above chart shows the final outcome which extracts the maximum value from the available and relatively sparse data to predict average daily oil consumption (Litres/day) against average daily (24hr) temperature. This is a degree-day type model where it is assumed that little or no heating is required when the average external daily temperature is at or above 15.5C.

This chart shows the weighted values as solid lines and unweighted as dotted lines. The white line is the simple annual model over all available data without segregation by events. The blue shows the characteristic before the Programmable Valves were fitted and when the stove was not lit. The yellow characteristic shows the best fit after programmable radiator valves were fitted but when the stove was not lit. All three of these characteristics have a constant hot water demand (horizontal part) of 1.64 Litres/day.

The red characteristic clearly shows that less oil is consumed when the fire is lit (with simple radiator valves) and the change from the blue consumption characteristic should take place when the average daily temperature is ~5.8C (9.7 degree-days).

Programmable Radiator Valves:
- Installing programmable radiator valves to only the three largest radiators in the home gave approximately 20% oil consumption savings in this example
- 300 pa savings in oil consumption in this example, giving a payback of less than six months
- Carbon dioxide savings of approximately 1.3 tonnes pa in this example
- Savings achieved by zone/time control and more accurate and responsive control than can be achieved by TRVs - greater savings may be achieved if more programmable valves are used in a home with more occupancy

Wood Burning Stove:
- Typical savings of 10% oil consumption by lighting a wood burning stove (room heat only) on the coldest days of the year, more during colder winters (~15% 2010/11)
- 150 pa typical oil consumption savings in this example
- Stove lighting benefits could be considerably extended with Redford-Tech's FHT Stove technology - see