Example 1:
As a specific example, assume that the HVAC unit has, as an efficiency measure, a kW/ton rating of 1.33 at an input power of 40 kW, thereby providing a nominal heat transfer in the form of 30 tons of refrigeration capacity. However, the HVAC unit has a kW/ton rating of 1.50 at an input power of 60 kW, thereby providing a nominal heat transfer in the form of 40 tons of refrigeration capacity. Therefore, the HVAC unit has a higher efficiency at 40 kW input power than at 60 kW input power. In this way, the HVAC unit will have an efficiency profile over a range of input powers. Assume 1.33 is the peak efficiency, with the efficiency decreasing on either side of 40 kW of input power, and a target heat transfer of 30 tons delivered over the desired adjustment time achieves the desired indoor temperature by the end of the desired adjustment time. Then to achieve a maximum efficiency measure over the desired adjustment time, the input power profile supplied to the HVAC unit would comprise supplying an input power of 40 kW over the whole desired adjustment time.
Example 2:
However, now assume a target heat transfer of 40 tons delivered over the desired adjustment time is required to achieve the desired indoor temperature by the end of the desired adjustment time, and the efficiency of the HVAC unit continues to decrease at input powers above 60 kW. Then to achieve a maximum efficiency measure over the desired adjustment time, the input power profile supplied to the HVAC unit would comprise supplying an input power of 60 kW over the whole desired adjustment time.
Example 3:
Now assume that the HVAC unit has a kW/ton rating of 1.20 at an input power of 60 kW, thereby providing a nominal heat transfer in the form of 50 tons of refrigeration capacity, and a kW/ton rating of 1.33 at an input power of 40 kW, thereby providing a nominal heat transfer in the form of 30 tons of refrigeration capacity. Also, assume 1.20 is the peak efficiency, with the efficiency decreasing on either side of 60 kW of input power. Finally, assume a target heat transfer of 30 tons delivered over the desired adjustment time achieves the desired indoor temperature by the end of the desired adjustment time. Then to achieve a maximum efficiency measure over the desired adjustment time, the input power profile supplied to the HVAC unit would comprise supplying an input power of 60 kW over the final 60% of the desired adjustment time, that is, 60% of 50 tons, in order to provide the required 30 tons.
Alternatively or additionally, the input power profile can correspond to a minimum power cost over the desired adjustment time calculated on the basis of the input power and one or more power cost rates applicable over the desired adjustment time. This allows for load shifting or load sharing in order to take advantage of non-peak or discounted electricity cost rates.
As a specific example, assume the same parameters in Example 3 above. Also, assume that the desired adjustment time is 1 hour, and 60 kW is the maximum input power. Finally, assume that the power cost rate in the first 24 minutes of the 1 hour is 1 cent/kWh, and 50 cents/kWh in the remaining 36 minutes. Then to achieve a minimum power cost over the 1 hour desired adjustment time, the input power profile supplied to the HVAC unit would comprise supplying an input power of 40 kW over the whole 1 hour. This would result in a total power cost of $12.16 (24 minutes at 40 kW costing 1 cent/kWh plus 36 minutes at 40 kW costing 50 cents/kWh) over the 1 hour desired adjustment time. In particular, this would be cheaper than if the input power profile comprised of supplying an input power of 60 kW over the final 60% (36 minutes) of the 1 hour desired adjustment time, which would incur a total power cost of $18 (36 minutes at 60 kW costing 50 cents/kWh).
Fig. 6 is a flow diagram showing an algorithm used by the controller 1 to compare the different input powers (i.e. power consumptions) corresponding to respective efficiency measures, which in this case are in the form of coefficients of performance (COPs). Each COP is derived from respective target heat transfers calculated from the thermal model 8, predicted ambient outdoor temperature of the outdoor space 4, and respective ambient indoor temperatures of the indoor space 3. The respective target heat transfers are matched with nominal heat transfers that correspond to particular input powers being delivered to the HVAC unit 2, thereby providing the different input powers and corresponding COPs being compared by the controller 1. The input powers can be compared in this way to determine a minimum power cost over the desired adjustment time in order to implement load shifting or load sharing as described above.
In one embodiment, the desired indoor temperature is any temperature within a desired indoor temperature range. In practice, the desired indoor temperature range can be a comfort range, comfort band, comfort zone, or any other range acceptable to a user. This of course provides for further possible input power profiles and corresponding target heat transfers in order to achieve maximum efficiencies and/or minimum power costs over desired adjustment times.
In one embodiment, the desired adjustment time is an estimated arrival time of a user for arriving at the indoor space 3 based on a current position of the user, a current speed of the user, and a current distance between the current position and the indoor space. In particular, the estimated arrival time is the current distance divided by the current speed. The current distance can be the shortest or most likely path distance determined by a GPS system.
In another embodiment, the desired adjustment time is an estimated arrival time of a user for arriving at the indoor space selected from a database of one or more estimated arrival times each corresponding to a respective predetermined location, the estimated arrival time selected based on the predetermined location closest to a current position of the user. The database can be populated by the user or can be one which is pre-prepared by another. For example, if the indoor space 3 is the home of a user, the database can include estimated arrival times from a place of work to home, from a train station to home, and from a bus stop to home.
In another embodiment, the desired adjustment time is an estimated arrival time of a user for arriving at the indoor space selected from a database of one or more estimated arrival times each corresponding to a respective geofence, the estimated arrival time selected based on a current position of the user reaching a said geofence. The geofence can be defined as a circle with a predetermined radius and centred on the indoor space 3. The geofence can also be defined as an irregular perimeter around the indoor space 3 whereby each point on the perimeter represents a location from which a user travels to the indoor space 3 in the same estimated arrival time.
A GPS unit or system in a user device 15 can be used to determine one or more of: the current position, the current speed, and the current distance. Also, the estimated arrival time can be updated as the user travels. The user device 15 can take the form of a tablet computer, a mobile phone, a desktop computer, or other suitable device.
The operating parameters of the controller 1, including the desired indoor temperature, desired indoor temperature range, desired adjustment time, whether the HVAC unit 2 is to be operated to achieve a maximum efficiency measure or a minimum power cost, and whether the desired adjustment time is to be determined based on the estimated arrival time of a user, can all be set, updated, modified, or otherwise controlled, by a user via, for example, a device such as the user device 15.
As noted above, one or more input powers are deliverable to the HVAC unit 2, with each input power corresponding to a respective nominal heat transfer provided by the HVAC unit. In one embodiment, one or more of the respective nominal heat transfers are predetermined from manufacturer’s data. For example, rated efficiencies can be provided by manufacturers for HVAC units in an initial factory condition. The rated efficiencies relate to the ratio between the input power and a respective nominal heat transfer provided by the HVAC unit. There can be a single average rated efficiency, a single maximum rated efficiency, a table listing a plurality of rated efficiencies corresponding to respective input powers, or a graph relating rated efficiencies to respective input powers. For air conditioners, the input power is typically the power delivered to the compressor motor of the air conditioner.
In another embodiment, one or more of the respective nominal heat transfers are calculated from manufacturer’s data using a mathematical model. For example, where a table listing a plurality of rated efficiencies corresponding to respective input powers is provided, rated efficiencies can be interpolated for input powers between listed input powers. Alternatively, a formula modelling the relationship between rated efficiencies and respective input powers can be provided.
In another embodiment, one or more of the respective nominal heat transfers are calculated from one or more outputs from the HVAC unit 2. This can include the flow rate and temperature of the air supplied by the HVAC unit.
Alternatively, combinations of the above methods for determining nominal heat transfers can be used.
As mentioned above, the rated efficiencies or nominal heat transfers predetermined or calculated from manufacturer’s data are based on the initial factory condition of the HVAC unit 2. Over time, the performance and therefore efficiency of the HVAC unit will inevitably deteriorate. Accordingly, over time, the actual heat transfer provided by an HVAC unit 2 at a particular input power will decrease from the nominal heat transfer. The controller 1 includes a monitoring function to take this deterioration into account so that rectification or maintenance of the HVAC unit can be performed, or nominal heat transfer values can be updated to reflect actual heat transfer values.
In particular, the estimation module 7 calculates the target heat transfer as the heat transfer load provided over a monitoring time, thereby defining an actual heat transfer, and the control unit monitors for a difference between the nominal heat transfer and the actual heat transfer over the monitoring time, the nominal heat transfer corresponding to the input power over the monitoring time. Thus, the controller 1 uses the same model-based approach described above to calculate a target heat transfer for achieving a future desired ambient indoor temperature in order to review the performance of the HVAC unit 2 in the past over a monitoring time.
The monitoring function, and all the associated parameters, such as the monitoring time, can be controlled by a user via, for example, a device such as the user device 15. This includes initiating the monitoring function at a particular time or setting the monitoring function to activate at predefined times or intervals.
Fig. 7 is a flow diagram showing an algorithm used by the controller 1 in one embodiment to perform this monitoring function described above. In particular, the box titled “Model System” calculates the actual heat transfer based on a time series of input variables, and the box titled “Factory Defined System” calculates the nominal heat transfer based on the same time series of input variables. Both the actual heat transfer and the nominal heat transfer correspond to a common input power or input power profile, and therefore, respective efficiency measures (in the form of COPs in this embodiment) can be calculated for the actual and nominal heat transfers. If the difference between these COPs exceed a predefined threshold then the HVAC unit 2 is either underperforming or overperforming and remedial action can be carried out.
In one embodiment, the remedial action can include rectification or maintenance of the HVAC unit 2 in an attempt to bring the performance back to or towards the initial factory condition. The control unit 9 can send an alert to a user device when the difference between the COPs, or other efficiency measures, reaches the predefined or predetermined threshold.
Alternatively or additionally, the remedial action can include updating the nominal heat transfer to match the actual heat transfer at the corresponding input power. In particular, the control unit 9 updates the nominal heat transfer corresponding to a respective input power to match the actual heat transfer. This will allow the control unit 9 to control the HVAC unit 9 more accurately over time.
In this way, the flow diagram of Fig. 7 also represents a “fault detection” algorithm.
The external heat transfer is calculated in accordance with an ambient outdoor temperature of the outdoor space and actual variations to the ambient outdoor temperature over the monitoring time. The calculated external heat transfer also takes into account one or both of direct solar radiation and defuse solar radiation. Further, the calculated external heat transfer takes into account actual variations in one or both of direct solar radiation and defuse solar radiation over the monitoring time.
In calculating the heat transfer load, the heat transfer load can further comprise an internal heat transfer between the indoor space and objects within the indoor space. The objects can comprise one or more of: a person; and an electrical appliance. For example, an estimate of the heat produced by office equipment such as computers and monitors and passing into the indoor space 3 can be taken into account.
The thermal model 8 can be based on an evolutionary algorithm or a genetic algorithm. The thermal model 8 can be a black-box model or a grey-box model. A grey-box model can incorporate building physics and advanced model training techniques, in order to provide more detailed analysis of the thermal response of indoor spaces. The developed thermal model is used to characterize the thermal response of indoor spaces, considering both envelope and indoor thermal mass, such as furniture, bedding and carpets, and to estimate the indoor air temperature and energy consumption of an AC under realistic weather conditions. With the addition of the real-time power measurement of an individual AC unit, the health of the AC unit can be assessed by comparing the actual power consumption with the predicted power consumption as discussed above.
One particular embodiment comprises a self-learning thermal model 8 and the associated algorithm for training the thermal model. In residential homes, each room is usually served by individual air conditioners (ACs). Therefore, they can be considered as thermally isolated zones and then can be modelled independently. Based on the thermal model, the thermal solution developed is able to perform simplified thermal analysis to identify the possible saving opportunities and to empower the users to improve energy efficiency through implementing smart control. Meanwhile, it is also possible to generate automatic control strategies/measures in each individual zone to control and maintain the indoor thermal environment in accordance with any desired thermal comfort requirements or standards such as the thermal comfort requirements defined in ASHRAE 55.
In particular, a thermal resistance and thermal capacitance model (RC model) can be used to model components of the building envelope 5. In one version, a second order transfer function established from an assumed 3R2C, i.e. 3 thermal resistances and 2 thermal capacitances, thermal network model is used to predict the target heat transfer. All the parameters of the 3R2C model for external walls, roof, internal walls, etc., whose values are assumed within certain ranges, are identified by a non-linear regression algorithm to minimize errors between the predicted and actual cooling demand.
Figs. 3 and 4 are schematic diagrams of RC models that can be used in the present invention. Fig. 5 is a flow diagram showing the steps of a genetic algorithm (GA) used to determine the values of variables of a thermal model, and in this particular case an RC model, in accordance with an embodiment of the present invention. In the specific example shown in Fig. 5, the GA shown is used to determine the values of variablesRext,1,Rext,2,Rext,3,Rim,Cext,1,Cext,2, andCim of the RC model shown in Fig. 4.Rext,1,Rext,2, andRext,3 represent the 3 thermal resistances of an external wall. Temperature nodeText,1 is betweenRext,1 andRext,2, and temperature nodeText,2 is betweenRext,2 andRext,3. Thermal capacitanceCext,1 of the external wall is connected toText,1, and thermal capacitanceCext,2 is connected toText,2.Rim is a thermal resistance of an internal mass, andCim is a thermal capacitance of the internal mass, both of which are connected to temperature nodeTim.
The simplified RC thermal network models shown in Figs. 3 and 4 are lumped grey-box models, developed to reflect the thermal status and thermal response of both a building envelope and an internal thermal mass. It is probably not feasible to obtain the physical properties of neither the building envelope nor the indoor thermal mass when the developed building thermal model is embedded in a real-world controller and implemented in real flats with unknown conditions. Therefore, the developed thermal models adopt self-learning and adaptive methods to estimate the thermal characteristics. For on-situ measurement and application, the developed models require less training and calibration efforts with a short range, e.g. 2 weeks, of historic operation data. The required computational costs and memory demands are also not significant.
The simplified 3R2C model shown in Fig. 3 is in an electrical analogue pattern with resistance (R, m2K/W) and capacity (C, J/(m2K)).Rext,1,Rext,2,Rext,3,Cext,1, andCext,2 are assumed to consist of the thermal characteristics of the building envelope including walls and roofs. The heat transfer, i.e. heat gain from external sources, through walls, roofs and windows can be characterized accordingly.
The physical interpretation of these parameters is dependent on how the building envelope is divided into entities:Rext,1 is the resistance between an internal surface and the node “point 1” of an equivalent unit area inside of the envelope.Rext,2 is the resistance between an external surface and the node “point 2” of an equivalent unit area inside of the envelope. The sum ofRext,1,Rext,2, andRext,3 is the total heat transfer resistance of the whole envelope, which includes both convection and conduction resistance.Text,1 andText,2 are the temperatures of two “points” or nodes of an equivalent unit area inside of the envelope.
The developed lumped grey-box model is described by the following differential equations, which represent the heat dynamic and energy balance in building.
Where,Qβ is heat gains from infiltration (W).Qin is sensible heat gain from indoor heat resources (W), e.g. human, equipment and lighting.Qdem is the cooling demand (W) supplied by an air-conditioner (AC) and it is zero when no AC is working.Tsol is solar-air temperature (°C), which is determined by the following equation:
where,Tout is outdoor dry bulb temperature (°C).I is global solar radiation (W/m2),αwall is wall absorption coefficient andαout is convective heat transfer coefficient of envelop external surface (W/m2·K).
As described above, a genetic algorithm (GA) can be used to determine the best values for the variables of the thermal RC model in this non-linear optimization process. Other conventional optimization methods have to start from initial guesses of the optimal variables and their convergence speed, which are affected by the initial guess in most cases, while genetic algorithm (GA) is a better optimization method, especially when the optimal problem is not perfectly smooth and uni-modal. It can quickly find a sufficiently approximate solution, e.g. near optimal control, and can be applied when a task does not require an “absolute” optimal results. The algorithm was usually used to search for global optimal solutions in air-conditioning research fields. In the present invention, a GA method is utilized to search for optimal parameters of the 3R2C model to minimize the errors between measured and predicted values. The following introduces the implementation details of the GA method.
The equations (1) to (4) above are used to compare with the measured building indoor dry bulb temperature. The optimized parameters are the resistances (R) and capacitances (C) of the developed 3R2C model that give the best fit with the measured data. The objective functionJ of such optimization employs the integrated root mean square error (RMSE) defined in equation (6) below:
where,Tin,act andTin,pre are the actual building indoor dry bulb temperature and predicted temperature respectively.Rext,1,Rext,2,Rext,3,Cext,1, andCext,2 are the parameters required to be determined. This is a typical non-linear optimization problem. The GA is employed to search for the optimal values of RC model as illustrated in next sub-section.
In one embodiment, the actual building indoor temperature is measured by the controller 1 and saved in a notebook computer. To calculate/predict the indoor temperature, the outdoor dry bulb temperature and solar radiation, occupancy and internal gains are used as inputs to the developed RC model.
Generally, a GA is an advanced search and optimization technique. It was developed to imitate the principle of natural genetic evolution. One of the main advantages of a GA is that it is generally robust in finding global solutions, particularly in multi-model and multi-objective optimization problems. Extensive research on the theoretical fundamentals and applications of GA is still being carried out to achieve better computation efficiency and improved robustness.
Fig. 5 schematically shows a flow chart of a GA estimator developed for parameter identification of the RC model in accordance with the present invention. It starts with an initial estimation of the individual capacitances and resistances with reasonable values. The part enclosed by the box represents the procedures of a GA run. Multiple runs are allowed. Equation (7) below represents the fitness functionf, which is the reciprocal of the objective function, i.e. equation (6) above.
In the GA, the parameters constitute the chromosome of an individual, and the assumed ranges of these parameters are the search spaces for these parameters. Initializing these parameters produces the initial population to start a GA run. Termination of a GA run is decided if the number of the current generation is equal to a predefined maximum. As least two runs of the GA process are necessary when running the GA estimator. The criterion to stop the GA estimator is based on the comparison the best fitness values of two consecutive runs. If the relative difference between the two maximum fitness (df) is less than a threshold value, the GA estimator is stopped.
The above RC thermal model was validated as follows. One master bedroom in a residential home was selected as the test bedroom for the validation of the RC model. The test period was over nine days from 17:30 hrs on the first day to 17:40hrs on the ninth day, in which the data from 17:30 hrs on the first day to 4:30 hrs on the fifth day were used as training data and the data from 4:40 hrs on the fifth day to 17:40 hrs on the ninth day were used as validation data. The results are shown in Fig. 8.Tin,act is the actual indoor temperature andTin,RC is the resultant indoor air temperature from the RC model. The identified parameters are:Rext,1 = 0.129 m2K/W,Rext,2 = 0.1376 m2K/W,Rext,3 = 0.0830 m2K/W,Cext,1 = 481,559 J/m2K,Cext,2 = 18441 J/m2K.
The RC model can be improved to provide better results. In particular, the maximum temperature error is around 1.5 degrees while the time delay is too long, around 4 hours. One possible reason is that the thermal resistance and capacitance of the internal thermal mass, including floors, partitions, furniture and the like, is not considered. This internal mass absorbs radiant heat through the windows and from indoor heat sources, such as occupants, lighting, machines and the like, and then releases the heat gradually to the air space. For refining the model, anotherR and anotherC, which represent the heat gain from building internal mass, can be added, as shown in Fig. 4. This improved RC model can be referred to more specifically as 3R2C+1R1C.
The improved RC model is described by the following differential equations:
where,Cim is the total thermal capacity of internal thermal mass, including floors, partitions, furniture and the like.Rim is the thermal resistance of internal thermal mass. It is worth noticing thatR andC also reflect the physical characteristics of the building envelope, and they are all assumed to be time invariant. Meanwhile, the RC models can predict reliably the performance of the global building system.
The improved RC model in Fig. 4 was also validated. The test room was still the same master bedroom for the validation of this improved RC model. The whole test period, the selected period for training and validation were also the same as for the model in Fig. 3 above. The results are shown in Fig. 9. The identified parameters are:Rext,1 = 0.186 m2K/W,Rext,2 = 0.247 m2K/W,Rext,3 = 0.169 m2K/W,Rim = 0.159 m2K/W,Cext,1 = 448,881 J/m2K,Cext,2 = 46,999 J/m2K,Cim = 33,7326 J/m2K.
As shown in Fig. 9, it is obvious that the simulated building indoor air temperature is more close to the actual indoor air temperature, which shows that the accuracy of the model has been enhanced significantly.
All the mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) of the outputs in 3R2C and 3R2C+1R1C models are listed in Table 1 below. The accuracy of the improved model is acceptable.
Another dormitory bedroom was selected for further validation of the developed RC model. The test period was over 11 days from 00:00 hrs on the first day to 23:50 on the eleventh day, in which the data from 00:00 hrs on the first day to 23:50 hrs on the sixth day were used as training data and the data from 00:00 hrs on the seventh day to 23:50 hrs on the eleventh day were used as validation data. The results are shown in Fig. 10. The identified parameters are:Rext,1 = 0.053 m2K/W,Rext,2 = 0.078 m2K/W,Rext,3 = 0.178 m2K/W,Rim = 0.026 m2K/W,Cext,1 = 286,208 J/m2K,Cext,2 = 311,695 J/m2K,Cim = 642,127 J/m2K.
As shown in Fig. 10, the prediction of the building indoor temperature has satisfactory performance with high accuracy. It can be proved that the values of accuracy indices: the mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) are 0.241°C, 0.88% and 0.35°C respectively.
It is appreciated that the aforesaid embodiments are only exemplary embodiments adopted to describe the principles of the present invention, and the present invention is not merely limited thereto. Various variants and modifications can be made by those of ordinary skill in the art without departing from the spirit and essence of the present invention, and these variants and modifications are also covered within the scope of the present invention. Accordingly, although the invention has been described with reference to specific examples, it is appreciated by those skilled in the art that the invention can be embodied in many other forms. It is also appreciated by those skilled in the art that the features of the various examples described can be combined in other combinations.