Elsevier

Energy and Buildings

Volume 166, 1 May 2018, Pages 23-34
Energy and Buildings

Evaluation of a feedback control method for hydronic heating systems based on indoor temperature measurements

https://doi.org/10.1016/j.enbuild.2018.01.013Get rights and content

Highlights

  • The article presents a method for feedforward control of supply temperatures in hydronic heating systems based on outdoor temperatures and local feedback at room level using thermostatic valves enhanced by using a central feedback control of the supply temperature based on average indoor temperature.

  • The enhanced control method is evaluated primarily with respect to the resulting indoor temperatures compared to those created by the original control system.

  • The indoor temperatures became less dependent on the outdoor temperatures, indicating improved control of the hydronic heating system.

  • Correct commissioning and understanding of the limitations of the enhanced control are important factors for successful results.

Abstract

Indoor temperatures in apartment blocks are often indirectly controlled by the outdoor temperature using a feedforward control loop, in which the radiator supply temperature is a function of the outdoor temperature. However, this control principle cannot take into account heat gains or losses caused by tenants, electrical appliances, the sun, air leakage, etc., which may result in uneven indoor temperatures, overheating, airing and increased energy use. This can be partly addressed by using thermostatic valves on each radiator. A municipal housing company in Sweden that uses individual metering and billing (IBM) of space heating costs based on measurements of indoor temperatures in all rooms of each apartment has been studied. This article presents and evaluates a project in which these measurements were used for feedback control. The aim of the study was to evaluate the principle which is based on using the actual indoor temperatures. An existing feedforward control of the heating system with thermostatic valves was enhanced by a correction of the supply temperature. The magnitude of the correction was proportional to the difference between the actual mean indoor temperature of the apartments and the set-point temperature. The enhanced control resulted in more constant indoor temperatures, i.e. they were less dependent on the outdoor temperature. The results support the conclusion that the evaluated method would be promising to apply in multi-family buildings. The introduction of the enhanced control method provided valuable experience and awareness of influencing factors if it were to be implemented on a larger scale.

Introduction

To achieve the desired indoor climate, both in terms of thermal comfort and air quality, it is important not only to build and design an energy-efficient building but also to run the heating systems optimally. One important task is to control the indoor temperature and prevent overheating, which is also a prerequisite for good energy economy. Simulations performed by Peeters et al. [1] indicate potential savings up to 20% in Belgian gas-heated terrace houses due to better control of their heating system.

Today, indoor temperatures in most apartment buildings with hydronic heating systems are indirectly controlled by regulating the supply water temperature in the radiator system as a function of the outdoor temperature. This simple control principle has been used for over a century. In the first half of the 1900s, it was applied by the boiler attendant, who manually set the supply temperature according to the expected outside temperature over the next few hours. The same principle was automated in the middle of the 1900s, when district heating or oil-fired boilers were introduced, by measuring the outside temperature and regulating the supply temperature to match a calculated value. The principle is called feedforward or open loop control.

This control method means that the outdoor temperature is the only parameter that can influence the supply water temperature to the radiators. A disadvantage is that it does not take into account the factors and conditions inside the rooms that can affect the resulting indoor temperature. These factors, sometimes called “disturbances”, such as additional heat gains from the sun, tenants, electrical appliances and other internal heat sources, are not taken into account, neither are various cooling factors such as ventilation caused by wind, airing or natural ventilation. Consequently, a control method that only takes into account the outdoor temperature may, periodically, result in excessively high or low indoor temperatures. One of the main factors influencing the control choices made by municipal housing companies is the avoidance of tenant's complaints, mainly because of low temperatures. Appropriate settings that consistently provide indoor temperatures above the complaint threshold must be found. However, generous margins may result in negative impacts on energy use. Excessively high indoor temperatures imply that there is an unnecessary use of energy. Overheating may also lead to airing by the tenants to control the indoor temperature, which will also increase energy use. If airing is not carried out, excessive indoor temperatures result in poorer thermal comfort. It is, therefore, important to find more accurate control methods with respect to both indoor thermal comfort and energy use.

In buildings connected to a district heating system, the supply temperature to the heating systems can be controlled and based on the primary supply temperature in the district heating system instead of the outdoor temperature. This is possible because both temperatures can be described as a function of the outdoor temperature [2], [3], although this is not a straightforward way.

Thermostatic radiator valves (TRVs) can reduce overheating but only to a certain extent as there will be heat conduction between the radiator system and the thermostatic sensors, and when the upper temperature limit is set too high. The indoor temperature must rise to the temperature of the sensor element before the valve closes off the heat supply. Manufacturers specify closing times of around 20 min. The age and quality of TRVs are essential for their performance and ability to respond to changing conditions. Old TRVs tend to have longer time constants and limited rangeability [4]. Poorly managed commissioning of outdoor temperature compensators, affecting the control curve in the feedforward loop, has also led to lower thermal comfort and higher energy use [4], [5].

If the indoor temperature is measured, the factors inside the rooms, so-called disturbances, can be taken into account. This method of controlling the indoor temperature is sometimes used in single-family housing, for which a single temperature sensor can sometimes be sufficient. The method is called feedback or closed loop control. In apartment blocks, this principle is rarely used as it is costly to measure indoor temperatures in a large number of apartments. In practice, feedback control is normally a combination of both feedforward and feedback control in which the feedback control tries to eliminate the difference between the measured indoor temperature and the set point temperature.

If it were possible to measure all the different types of disturbances, better control could be achieved. Thomas et al. [6] have, through simulations, investigated P-control with and without feedforward for temperature control of a small single-room building with a short time constant. They concluded that the latter would be advantageous in the studied building if there were internal heat disturbances but point out that it would be more complicated for other types of disturbances and buildings with several temperature zones.

Generally, the greatest advantage of feedforward control is that rapidly varying disturbances can be eliminated by compensatory actions before the disturbance has time to influence the output signal. The disadvantages are that it demands having a model of the process and the disturbances must be possible to detect. The main disadvantage is slower control [7], [8].

Feedback control of the supply water temperature in hydronic heating systems in apartment buildings, based on indoor temperatures, is currently only applied in a limited number of properties. The indoor temperature is measured, with either one or a few sensors (reference sensors), placed in selected rooms in the buildings, or indirectly by measuring the exhaust air temperature.

In Sweden, the method employing reference sensors is used by two commercial companies, EnReduce Energy Control AB [9] and Kabona [10]. The control system provided by EnReduce works using a pure, closed control loop, i.e. there are no outdoor temperature sensors, only reference sensors in some of the south and north facing rooms. The Kabona system uses indoor temperature sensors in combination with weather forecast control whereby the signal from the usual outdoor temperature sensor is replaced by a signal based on the weather forecast and properties of the building. These properties include data regarding the building's dimensions, the building envelope, ventilation and heat capacity as well as operational data regarding internal heat generation and desired indoor temperature. Site location, orientation and surroundings, and factors such as wind conditions and shading, are also included. Simulations of weather forecast and feedback control for a two-zone model [11] showed that zone temperatures could be maintained within ±0.1 °C from the setpoint temperatures. Potential energy savings of 10%, in some cases over 20%, by using weather forecasts in the control algorithms, are reviewed by [12], [13].

An alternative method for measuring indoor temperatures is to do this indirectly by measuring the exhaust air temperatures for all the exhaust fans. This provides a weighted average of the temperatures in the kitchens, bathrooms and toilets, and may, therefore, differ slightly from the temperatures in an apartment's living rooms and bedrooms. The method was tested in one eight-storey building [14] and in three groups of houses [15], all in southern Sweden. The tests showed that it was possible to control the exhaust temperature in several test buildings with only minor deviations from set point. However, this also meant that the radiator temperatures fluctuated widely and were reduced considerably during days with large solar and internal heat gains. Despite having normal temperatures in their apartments, some tenants complained about having cold radiators. This problem was solved by limiting the lowest allowed supply temperature and allowing for small temperature differences above the set point. Controlling the indoor temperature via a central feedback loop has been further analyzed by measurements and has been documented in Hedin [16]. A thorough analysis of covariance showed that control of indoor temperature could eliminate variations in average temperature, but certainly not the individual variations, which could be relatively large. Something a central control cannot eliminate are the constantly occurring temperature differences between apartments. These can be addressed by balancing or, in some cases, by using individual room control with the help of TRVs (local feedback). The individual variations can be reduced if individual metering and billing (IMB) of heating costs is introduced, which would possibly increase the savings potentials for the central control of the heating system. However, reducing temperature differences between apartments might not be desirable as different tenants desire different temperatures. IMB can, instead, be used as a way of satisfying different needs [17]. However, due to poor or no thermal insulation between adjacent apartments, this could be difficult to achieve, i.e. a cold apartment would gain heat from warmer neighbours, which is a problem when allocating heating costs in buildings with IMB based on delivered energy [18].

Another way to estimate the average indoor temperature in a building is by measuring the supply and return temperatures in the heating system and the outdoor temperature. Examples which show that this is possible have been demonstrated in Hedin and Jensen [15] for a limited number of measured data. Though, there are too many parameters with varying uncertainties to be practically useful.

Jensen [19] has, with simulations of a simple single-room model verified by measurements, shown that feedforward control based on an outdoor temperature calculated by weighting prevailing outdoor temperature and an earlier outdoor temperature, e.g. 6 or 12 h backwards, with properly selected weighting, provide better results than traditional feedforward control. The same principle, but with an outdoor temperature calculated by weighting prevailing outdoor temperature and a later, i.e. a forecasted, outdoor temperature 6 or 12 h ahead, gives worse results than traditional feedforward control. A dynamic model with a filter that takes both the inertia of the building and the heating system into account, i.e. a change in outdoor temperature will not immediately result in a changing supply temperature in the heating system gave better result than both these.

Oldewurtel et al. [20] reviewed several works that investigated the use of weather predictions and predictions of other disturbances for building climate control. Their conclusion was that predictive control strategies are beneficial for energy-efficient building climate control. Their own work [20], [21] shows, by using simulations, that a stochastic model predictive control (MPC) for building climate control, which takes into account weather forecasts, performs better than persistence predictions, but is sensitive both to the quality of the model and the weather forecasts. In [22] the authors state that having a dynamic model of the building is crucial. This part is pointed out as being the most time-demanding in MPC [23], [24]. Several articles have been written as part of the OptiControl project (2007–2013), in which the main goal was to develop tools, methods and strategies for building control and especially for predictive control integrating weather and occupancy forecasts. Also worth mentioning is the large review of articles concerning advanced control systems for energy and comfort management in buildings performed by Dounis and Caraiscos [25].

In summary, it can be concluded that it is important, both for thermal comfort and energy use, to take into account factors that influence the indoor temperature, which feedforward control cannot do. Feedback control methods, which address these issues more precisely, are, however, neither widely spread nor scientifically investigated. The few studies that have been made were based on simulations or measurements from a limited number of real buildings. Therefore, there is an identified need to study feedback control methods and evaluate their performance with regard to the issues mentioned, i.e. indoor temperature and energy use with focus on performance in real buildings.

A municipal housing company in southern Sweden has implemented a feedback control method in a large number of buildings, of which 116 are included in this study. In these, the indoor temperature was measured in all living rooms and bedrooms, in total 3,248 rooms, every 15 min. This exceptionally large dataset provides a unique opportunity to examine the performance of a feedback control method in real apartment buildings.

This article describes and evaluates a feedforward, weather-compensating control system with TRVs which has been enhanced with feedback correction of the supply temperatures to the hydronic heating system, based on a building's interior temperatures, which, in this case, was the mean of all room temperatures in an apartment block. This study will contribute to filling the identified knowledge gap with regard to real conditions.

The aims of this study were to:

  • present a system for enhanced control of a hydronic heating system,

  • in a case study, investigate to what extent this control system makes the indoor temperature less dependent on the outdoor temperature by taking into account varying heat gains from occupants, lighting, electrical appliances and solar radiation, i.e. how a more uniform indoor climate and avoidance of overheating are achieved and

  • for the same case, investigate the effect of the system on energy use for space heating.

The evaluation of the feedback control system was carried out by comparing the indoor and outdoor temperatures before and after the system was implemented. This was carried out in test group A (main group), consisting of 10 sub-groups of housing estates comprising 116 separate residential buildings, Table 1. The buildings in this main group were equipped with IMB for space heating, based on measured room temperatures. Control of the supply water temperature for each estate during the first period was performed by a conventional feedforward control system based on outdoor temperature in connection augmented by local feedback control using thermostatic valves (TRVs). During the second period, the control system, i.e. feedforward plus TRVs, was enhanced by a central feedback control based on the mean indoor temperature of a building to increase or decrease the supply temperature. The local feedback control with TRVs remained in use as before and gave the tenants the opportunity to adjust the indoor temperature. The enhanced control is described in further detail below. Temperature measurements for January to March in 2010 and 2011 were analyzed. In order to take the thermal inertia of the buildings into account, daily mean temperatures, based on the 15 min values from all sensors, i.e. from all living rooms and all bedrooms in each building, were studied.

The intention was also to investigate how feedback control affected energy use. This was done by analyzing and comparing operating statistics before and after the system was implemented. The statistics, accessed from the property owner, included energy for heating and hot water, non-domestic electricity, and water use in 2009, 2010 and 2011. The energy meters registered the sum of space heating energy and energy for domestic hot water. The heating energy was weather-normalized with heating degree days for Lund, Sweden, according to Swedish Meteorological and Hydrological Institute (SMHI). Throughout the study, different types of purchased energy were analyzed. In this part of the study, in addition to the main group, there were two reference groups, each consisting of 10 housing estates with varying numbers of buildings, Table 1. As in group A, the estates in group C (reference group 2) were equipped with IMB of space heating based on measured room temperatures, while group B (reference group 1) did not have IMB. All data processing was performed using MATLAB software. The gain factor, g, used in group A took into account the factors affecting the indoor temperatures in the rooms and enhanced their influence. The larger the gain, the greater these factors affected the supply temperature.

All temperatures were accessed from an already existing system for IMB of heating costs, designed by the property owner (further described in [17], [26]). Temperatures in this system were measured every 15 min in all living rooms and bedrooms in all apartments, and outdoors on the northern façade of each building in group A. All indoor temperatures were registered 2.1 m above floor level. The data set for all 10 estates in group A comprised 20 million room temperatures for 2010 and 7.8 million for 2011. The maximum possible number for each three-month period was 28.1 million.

The measured temperatures were transmitted through the respective cable television networks via a device plugged into the outlet in each apartment. If this device was unplugged or the temperature sensors failed to register a room temperature, the IMB-system recorded a default reading of 21.0 °C, which resulted in a surplus of 21.0 °C values in the data set. To allow analysis of the temperatures, and to calculate true and fair daily mean temperatures, it was necessary to filter this surplus. For example, for estate 2 during the three-month period in 2010, the total number of 21 °C values was 117,000, compared to 19,000 occurrences of 20.9 °C and 47,000 of 21.1 °C. Day-by-day this was remedied by keeping as many 21 °C values as the average number of 20.9 °C and 21.1 °C values, resulting in 31,500 occurrences of the 21 °C values The temperature data for a single day was close to a normal distribution, with only little skewness and kurtosis, except for the peak at 21 °C. After removing the surplus values, the peak was eliminated. In 2011, the relative surplus was even higher and was handled in the same way. The figures mentioned here relate to estate 2. The situation was the same in all estates and was handled in the same way. A calculation, assuming a normal distribution, of how many true 21 °C values could be discarded by mistake or how many false values that could have been wrongly added, indicated ±5 per 1000, which would only affect the daily mean value by 0.01 °C.

For a more detailed presentation concerning the daily mean values and distribution of room temperature for the first half of 2010 in all buildings see [27]. In five estates, the indoor temperature distributions and variations at building, apartment and room level were presented in [17]. Temperature variations within apartments of different sizes and building categories were, for estates in group A, presented in [28].

Section snippets

2.1. General data

The 30 estates varied in age, size, standard of insulation, building styles, etc., and ranged from single multi-storey buildings to groups of terrace houses. An estate consisted of one or several buildings, with a maximum of 53. In most cases the buildings in an estate were connected to the same district heating substation, and had the same supply temperature. When feedback control was used the corrected supply temperature was based on the mean for all buildings in the estate except in estates

3.1. Conventional control

The control systems in groups B and C were conventional feedforward systems, i.e. the supply temperature to the hydronic radiator systems, Tsup, was a function of the actual outdoor temperature. This provided the temperature control curve for the building, represented by f(Te) in Equation (2). Group A also had the same type of control during the first period (2010). The feedforward systems in all buildings are supported by local feedback control using TRVs, which also provide the tenants with

4.1. Indoor temperature dependency on outdoor temperature

The enhanced control of the supply temperature was introduced in autumn 2010 which means that there was one heating period without and one with this control. To investigate how the control affected the indoor temperature, an analysis of the period from January to March both years was performed. The main purpose of the temperature measurements was, however, to provide data for IMB of space heating costs.

Due to interruptions and planned maintenance work, there were a number of data losses. The

5. Discussion

The transmission of measured temperatures was done through the cable television network. Unfortunately, there were data losses, mainly caused by interruptions in the data transmission. The data losses were for single readings up to failures lasting several days. Losses of single readings occurred in all the estates at the same time, which indicated problems with the data transfer. Losses over several days may have been caused for the same reason. In these cases, there were both short and long

6. Conclusions

The proposed feedback control resulted in less dependency on the outdoor temperature compared to the control method with feedforward and thermostatic radiator valves. The conclusion is drawn that the feedback control results in less outdoor temperature dependency and, based on that fact, that the enhanced control has, on the whole, been effective. The aim of the proposed control method, in terms of temperatures, was fulfilled. This implies that the method is worth considering when choosing

7. Future research

The feedback control system was implemented in buildings with single specific values of the factors in the proposed algorithm. Based on the promising results, one future research field would be to examine and improve the proposed algorithm and examine the effect other values would have in real buildings.

According to the sensitivity analyses, a gain larger than 10 will only marginally affect the ratio between k(g) and k(0), maybe even a slightly lower g could be sufficient, which could be tested.

References (30)

  • J. Široký et al.

    Experimental analysis of model predictive control for an energy efficient building heating system

    Appl. Energy

    (2011)
  • B. Thomas et al.

    Feed-forward in temperature control of buildings

    Energy Build.

    (2004)
  • T. Hägglund

    Praktisk Processreglering [Practical Process Control]

    (2008)
  • K.J. Åström

    Reglerteori. [Control Theory]

    (1976)
  • Cited by (0)

    View full text