Enhanced power control model based on hybrid prediction and preprocessing/post-processing

Abstract

Since last decade, energy management and conservation in residential buildings received a great attraction of the researchers. A number of methods exist in the literature for energy conservation, but the trade-off between occupant comfort level and energy consumption is still a major challenge and needs more attention. Particle swarm optimization (PSO) and genetic algorithm (GA) based power control methodologies have been proposed previously. These techniques achieved good performance up-to some extent, but still there is room for improvements. In this paper, an enhanced optimized power control and hybrid prediction model based on preprocessing/post-processing, GA and hybrid prediction algorithms for occupants comfort index, energy saving and energy consumption prediction is proposed. Main focus is given to increase user’s comfort index and minimize energy consumption using GA based optimized and hybrid predicted systems with preprocessing and post-processing of data. Proposed method provides energy efficient environment by reducing energy consumption and improving occupants comfort index as compared to previous GA based power prediction model. The proposed system is also compared with individual Kalman filter ARIMA model prediction. The comparative results show the efficiency of the proposed model in decreasing the predicted power consumption and enhancing the occupants comfort index.

Keywords

Energy efficiency genetic algorithms comfort index fuzzy logic hybrid parallel prediction preprocessing and post-processing

Notations

Temperature

Illumination

Air-quality

USP:

User set parameters

AP:

Adjusted power

CP:

Consumed power

SCP:

Smooth consumed power

RP:

Required power

Ω:

No of generations

η:

Few successive generations

β₁, β₂, β₃:

greaterthan User defined factors [0, 1]

e_T:

Error difference in temperature

c_eT:

Change in error difference in temperature

e_L:

Error difference in illumination

e_A:

Error difference in air-quality

T_set, L_set, A_set:

User set parameters

P_(k):

Total power

Pavailable_(K):

Total energy source (external and internal power sources)

P_max(k):

Maximum power provided by the external or internal power sources

Consume power

PCP:

Predicted consumed power

Time

The weight factor

Operation power of ventilator

HS:

High Small

BS:

Basic Small

OK:

Zero (No change)

SH:

Simple High

High

Little:

Small

LH:

Little High

MH:

Medium High

1 Introduction

Energy resources are vital for economic development of every country. Also energy management plays an important role in future energy efficient building environment. Due to immense developments and advancements in electrical home appliances, electrical cars, smart phones and many other electrical devices, the utilization of energy increases day by day while its sources of generations are less and expensive as well. As far as energy consumption in building environment is concerns, the occupants of the building wants to spend less money on energy and gets more benefits by reduction of energy consumption without compromising comfort index. So energy consumption reduction and occupants comfort index are two basic design objectives in upcoming energy efficient buildings. The requirement of minimum energy consumption without compromising users comfort index is an interesting issue to the research community to cope with. This leads to the win-win situation between energy consumption and user comfort index [1 –4]. To address this win-win situation, an intelligent power control model is desperately required to maintain both energy consumption and occupants’ comfort index concurrently.

In future energy efficient residential buildings, the important three factors which satisfy occupant’s quality of lives are thermal comfort, visual comfort and air-quality [5]. Temperature indicates the thermal comfort of the occupant’s in a residential building. The heating or cooling system is used to preserve the temperature in building’s comfortable area. The illumination level is used to indicate the visual comfort of the occupants in a residential building [6]. The electrical lighting system is used to indicate the visual comfort. CO₂ concentration is used as an index to measure the air-quality in the building. Ventilation system is utilized to keep low CO₂ concentration [7]. The combination of these three parameters serves as occupant’s comfort index in the residential building. These three parameters are considered to evaluate the occupant’s comfort index and energy savings.

In the literature immense works have been done in the area of energy conservation and some well-known energy management systems have been proposed. Techniques based on traditional control systems have been proposed in earlier studies [8 –10]. These traditional controllers consist of Proportional Integral Derivative (PID) controllers, optimal controller and adaptive controller respectively. Designers used PID controllers in order to overcome the overshoot of temperature. These traditional controllers have some disadvantages, like they need a model of the building, they are not user friendly (user set parameters are not considered) and there are many difficulties in monitoring and controlling the parameters caused by nonlinear features. An optimized fuzzy controller applied for the control of environmental parameters at the building zone level has been proposed in [11]. In this method the occupants’ preferences are monitored via a smart card unit. Applications based on predictive control techniques proposed in [12, 13]. In these predictive control approaches weather predictions has been applied to heating, ventilating and air-conditioning system.

Optimal control strategies for variable air volume and air-conditioning system is proposed in [14]. The control strategies included a base control strategy of fixed temperature set point and two advanced strategies for insuring comfort and indoor air-quality (IAQ). The first advanced control adjusts the fresh air-supply rate and the supply air-temperature to maintain the temperature set point in each zone while assuring indoor air-quality. The second strategy controls the fresh air-rate and the supply air-temperature to maintain an acceptable thermal comfort and IAQ in each zone. The optimization problem for each control strategy was formulated based on the cost of energy consumption and constrained by system and thermal space transient models. An agent based control system with information fusion has been proposed in [15]. The agent based control system achieved high level of user comfort index with minimum energy consumption. Previously it is investigated that building, social and personal factors can influence one’s perceived health and comfort. No doubt perceived comfort in office buildings is powerfully influenced by several personal, social and building factors. The relationship among these factors are complex, so to get a better understanding of the relationships among these factors a proposal has been proposed in [16]. An indoor environmental quality comfort classification indexes suitable for both single environment and whole building is proposed in [17]. The approach allows evaluation of both energy consumption and polluting impacts and takes into account comfort conditions of indoor environment and outdoor climate. Another method called artificial neural network controller (ANNC) for energy conservation and control has been devised in [18]. ANNC system proposed a backup unit based on hydrogen energy utilization, composed of a combination of an electrolyzer and a fuel cell (FC) system. ANNC provides better energy efficiency and effectively operate a wind, solar, and hydrogen energy-based hybrid renewable stand-alone systems. GA has been proposed for energy conservation in many ways, like GA adopted for heating, ventilation and air-conditioning (HVAC) control problems [19]. GA based method also applied to the control problems of energy systems consisting of fuel cells, thermal storage, and heat pumps [20].

The energy prediction techniques can be classified into two classes. One is time series data prediction techniques and the other kind of energy prediction techniques is artificial neural network (ANN). The ANN’s technique that keeps excellent strength and error tolerance is an effective way to solve the complex nonlinear problems. ANN’s has received attention of researchers due to its clear model and good performance in solving nonlinear problems, but it is hard to establish a model for each building. Previously ANN has been applied to predict building power consumptions [21 –26]. These works frequently apply neural network model, which contains many parameters. These parameters are continuously judged by experience and the model become difficult to be established [27]. So it is difficult to create a model using ANN. Moreover, it has been perceived that while the neural network (NN) gives small error during training the patterns, the error for testing patterns is usually of a larger order [28], in other words, when this technique is applied to practical system, the prediction accuracy is not good. Moreover, this algorithm is required to transform the characters of all the problems into numbers and change all the inferences into numerical calculation. Nevertheless, it definitely cause the loss of information which degrade the prediction accuracy. Although ANN based price forecasting techniques can also be used for energy prediction, but its disadvantages reported above for price forecasting restrict its further applications for energy consumption prediction.

Stationary time series prototypes such as autoregressive (AR) [29], Dynamic Regression (DR), Transfer Function (TF) [30], non-stationary time series models like Autoregressive Integrated Moving Average (ARIMA) [52] have been devised to forecast electricity price in the recent time. These techniques can also be applied to energy consumption predictions. Some methods combinations ANN and autoregressive model to predict the thermal behavior of office building [31]. A technique based on autoregressive with exogenous input is proposed to predict 1 hour ahead building energy load [32]. In most modest energy markets the series of energy describes the following features: high frequency, non-constant mean and variance, daily and weekly, monthly, seasonality, calendar effect on weekend and public holidays, high volatility and high percentage of unusual energy usage. It is not easy to predict energy accurately; therefore, it has to require special dealing in case of estimating energy changes.

In this paper, an enhanced optimized energy efficient model for user’s comforts index and energy saving based on hybrid prediction and preprocessing/post-processing is proposed. The proposed method addressed both energy efficiency and occupants comfort index simultaneously. For prediction of energy consumption, a hybrid energy consumption prediction method based on Kalman filter and ARIMA model is considered. Hybrid prediction method predict the consumed power (CP) using both Kalman filter and ARIMA model and then best predicted value is selected from the two predictions based on the minimum power consumption. All the work discussed above except [2 , 14] either addressed users comfort index in the building or consumed minimum energy. GA based optimized system described in [3] improved occupants comfort index as compared to PSO based optimized system [2]. After achieving improvements in GA based system, an enhanced version of GA based optimization system is proposed. The proposed system achieved more prominent improvements as compared to previous GA based optimized system. The proposed model not only consumed minimum power but also provides better occupants comfort index as compared to PSO and GA based optimized system with no hybrid prediction [2, 3].

2 Proposed enhanced optimized energy management model

2.1 System diagram

In this section proposed enhanced optimized energy management model in building environment is described. Figure 1 shows optimized system diagram for the energy management in building environment. The environmental parameters are passed to smoothing component for preprocessing. After smoothing, the environmental parameters and user set parameters are passed to GA optimizer to get optimal parameters. The user set parameters are optimized to minimize the error difference between user set parameters and environmental parameters. Then optimized parameters are used to calculate the occupant’s comfort index. Three controllers based on fuzzy logic are used to control temperature, illumination and air-quality. Coordinator agent adjusted the power according to the required power from the fuzzy controllers and available power from the external power grid or internal local power sources. Coordinator agent performs the function of coordination among the three fuzzy controllers based on the required power and available power. The consumed power is passed to Kalman filter and ARIMA model to predict power consumption in parallel passion. Initially, the consumed power (CP) is predicted using both Kalman filter and ARIMA model and then best predicted value is selected from the two predictions based on the minimum power consumption. At the end again smoothing is applied as a post-processing. Each component of the proposed system model is described below one by one.

2.1.1 Preprocessing (Smoothing)

The sensing data is checked against the noise of outlier, zero cell data, standard form and normalization. If the data is found to be noisy then simply re-moves the outlier’s data, zero cell data and bring the data into a standard form. When the data becomes processed and in the standard form, it is then illegible to be input to the optimization component. Smoothing is applied here to environmental data as a preprocessing mechanism. The aim behind application of smoothing to environmental data is to reduce and bring the data in to smoother form. The data points which are higher or lower than the adjacent data points are decreases or increases to become smooth.

2.1.2 Optimization algorithm using GA

GA [33] steps for parameters optimizations and comfort index are:

Initial random population

Calculate fitness function for user comfort using Equation (1)

Select best individuals using any of three selection criteria (Rank, Roulette wheel or Tournament selection), in this paper rank based selection

Perform ‘two point’ crossover of the selected individuals

After crossover, we get off-springs

Calculate comfort for the off-springs

Combining populations of step (3) and (5)

If mutation criteria met, then perform mutation

Repeat above eight steps until required number of iterations

Then after arrival of termination criteria select best fitted chromosome.

These parameters are selected after running the algorithm for λ times to get optimal parameters. GA stops either when the maximum number of generation’s Ω met, or no significant change is observed in the fitness for η (few successive) generations. The maximum population size selected is 100. Two points crossover is performed with the probability of 0.9 and mutation rate of 0.1. GA parameters (population size, selection of parents, crossover rate and mutation rate) have been set after running GA for number times.

Rank based selection of parents is performed to give chance to the most of the chromosomes as compared to the tournament based selection method. Tournament based selection has a tendency to give little chance to the bad chromosomes. The probability of crossover (0.9) is set to involve both good and bad chromosomes in creating off-springs. Mutation rate of 0.1 is selected to involve randomness with a very little probability. The experimentations are performed using Intel(R) Core(TM)i3-2130 3.40 GHz with 8GB RAM. The C # 2012 is used for the simulation. When GA evaluation process finishes, best fitted chromosome is to be selected to get optimal parameters and comfort index.

2.1.3 Occupants comfort index

The comfort index can be calculated by using Equation (1) [1].

$\begin{matrix} comfort & = & β_{1} [1 - (e_{T} / T_{set})^{2}] \\ + β_{2} [1 - (e_{L} / L_{set})^{2}] \\ + β_{3} [1 - (e_{A} / A_{set})^{2}] \end{matrix}$ (1)

Where “comfort” is the overall comfort level of the user and is ranged between [1]. β₁, β₂, β₃ are the user defined factors which solve any possible conflict between the three comfort factors (temperature, illumination and air-quality). e_T, is the error difference between optimal parameter of GA(temperature in this case) and actual sensor temperature. e_L, is the error difference between optimal parameter of GA (illumination in this case) and actual sensor illumination. e_A, is the error difference between optimal parameter of GA (air-quality in this case) and actual sensor air-quality. T_set, L_set, A_set are the user set parameters of temperature, illumination and air-quality.

2.1.4 Coordinator agent

The coordinator agent received the optimized required building power from fuzzy controller. It adjusted the building power on the basis of available power and optimized required power to fulfill the occupants comfort index. The adjusted building power is compared with the required power to get the actual consume power. The consumed power is input to the Kalman filter and ARIMA model to predict consume power twice and in parallel passion. The PCP is given to the actuators for usage. The PCP is the power to be consumed in the building.

In Equations (2– 4) [3], P (k) is the total power, which is the sum of power demands from temperature, illumination and air-quality. P_available, is the total energy source (outside grid-power or internal local power source). P_max (k) is the maximum input power either from the power grid or from the local micro sources to the building. $P_{T (k + 1)} = P_{T (k)}$ (2) $P_{L (k + 1)} = P_{L (k)}$ (3) $P_{A (k + 1)} = P_{A (k)}$ (4) $P_{T (k)} + P_{L (k)} + P_{A (k)} = P_{available (k)}$ (5) $P_{available} \leq P_{max}$ (6)

2.1.5 Fuzzy logic based controllers

The idea of Fuzzy Logic (FL) was first introduced by L. A. Zadeh a professor in the University of California at Berkley [34].

The real parameters, optimal parameters and rate-of-change in these parameters are passed as input to fuzzy logic controller. Fuzzy controllers provide output results based on the membership functions. The output of the fuzzy controller(s) is the required power to control temperature, illumination and air-quality inside building. This optimized required power is input to the coordinator agent.

The input to the fuzzy controller for temperature is the error difference between optimal parameters of GA and real environmental parameters after smoothing along with the rate of change of temperature. For efficient control, both error difference e_T and change in error ce_T (difference between current and previous error) is used. Fuzzy controller rules for temperature control and input/output membership functions for temperature, illumination and air-quality control are described in our previous work [2]. Table 1 shows fuzzy controller rules for illumination control. The input to the fuzzy controller for illumination is the error difference between optimal parameter of GA and real environmental illumination parameter after preprocessing (smoothing).

When the input error is HS, the required optimal output power is OLittle. For error MS, the required optimal output power is OMS, for BS the required optimal power is OBS, for OK the required optimal power is OOK, for SH the optimal output power is OSH and for H the required power is OH. Table 2 shows the fuzzy logic controller rules for air-quality control. The input to the fuzzy logic controller for air-quality is the error difference between optimized air-quality parameter of GA and real environmental air-quality parameter after preprocessing (smoothing). If the input error is ‘Little’ the required optimized output power is OFF. For OK the required optimal power is ON, for LH the required optimized output power is OL, for MH the required output power is OLH and for HIGH the required power is OHIGH. The rules (relation between variables) in Tables 1 and 2 are selected based on the error difference. If the error difference decreases then required power is also decreases and vice versa. For-example if the error difference is HS (High Small) then required output power is OLittle (Very Small) and vice versa.

In Equations (8– 10) [4] γ_T, γ_L and γ_A are the temperature, illumination and air-quality increment relationship with consumed power P in time k respectively. θ is the weight factor to balance the relationship. The value of θ is between [1] and “d” is the basic operation power of ventilator. $γ_{T} = θ * P_{T} / K$ (7) $γ_{L} = θ * P_{L} / K$ (8) $γ_{A} = θ * P_{A} / K * d$ (9)

2.1.6 Kalman filter

In this section Kalman filter is discussed briefly. Kalman filter is one of the optimal estimators that can be used for time series data. It collects information’s of interest from indirect, inaccurate and uncertain observations. Naturally, Kalman filter is a recursive technique so that new measurements can be processed as they are available to use. The Kalman filter addresses the general problem of trying to predict the state x ∈ Rⁿ of a discrete-time controlled process that is administrated by the standard linear stochastic difference equation.

The matrix A in the difference Equation (10) relates the state at the previous time step t - 1 to the state at the current step t in the absence of process noise. In practical the value of A might change with each time step t, but here it is assumed as constant. The matrix P relates the optional control input to the state x. The matrix H in the measurement Equation (11) relates the state to the measurement z_t. In practical value of H might change with each time step t, but here it is assumed as constant. The process noise covariance Q and measurement noise covariance R matrices are change with each time step, however here Q and R assumed to be as constant. These noises are random normal distribution functions without considering mean value. The random normal distribution functions for each of the variables m and n are shown in Equations (12) and (13). The random variables m_t and n_t represent the process and measurement noise respectively. $X_{t} = {Ax}_{(t - 1)} + P + m_{(t - 1)}$ (10) $Z_{t} = {Hx}_{(t)} + n_{t}$ (11) $m \sim N (0, Q)$ (12) $n \sim N (0, R)$ (13)

2.1.7 ARIMA model

ARIMA (0, 1, 1) model is one of the variation of ARIMA (p, d, q) model. The ARIMA (p, d, q) model is the common class of model for predicting a time series data which can be stationaries by using some sort of transformations like differencing and logging. In fact, one of the easiest way to think about ARIMA models is as fine-tuned types of random-walk and random-trend models, the fine-tuning contains of adding lags of the differenced series and lags of the forecast errors to the prediction equation, as required to eliminate any last traces of autocorrelation from the forecast errors. In ARIMA (p, d, q) model p is the number of autoregressive terms, d is the number of non-seasonal differences, and q is the number of lagged predict errors in the prediction equation. $Y_{(t)} = μ + Y_{(t - 1)} - β * e_{(t - 1)}$ (14) $e_{(t - 1)} = Y_{(t)} - Y_{(t - 1)}$ (15) $β = 1 - α$ (16)

Where Y_(t) is the prediction, β is the coefficient of the lagged predicted error, e_(t-1) denotes the error at time period t - 1, α value varies between [1] and μ is the trajectory of the long-term prediction.

2.1.8 Post-processing (Smoothing)

Smoothing is also applied to predicted power consumption data as a post-processing mechanism. The basic aim behind application of smoothing to predicted power consumption data is to reduce noise and bring the predicted power consumption data in to smoother form. The major benefit of smoothing the predicted power consumption data is to provide smooth consumption of power in each time stamp.

2.1.9 Switching controller

Switching controller manages the available power sources. For example when the external power source unable to provide enough power to the building or its price is high, then it switch to the internal local power sources and vice versa.

2.1.10 Building actuators

Building actuators are the devices which actually use the power inside the building. The common actuators are AC used for cooling, heater for heating the residential building and fridge to provide cooling for eatable things and oven used for heating purpose. Sensors devices are used to get updated environmental information’s regarding temperature, illumination and air-quality level.

3 Simulation and results detail

Matlab/Simulink used for input/output membership functions construction. While actual simulation carried out in C# 2012.

User preference set parameters range was T_set = [66, 78] (Kelvin), L_set = [720, 880] (lux) and A_set = [700, 880] (ppm).

Figure 2 shows the comparisons of power consumption. X-axis shows the time in minutes while Y-axis shows the predicted power consumption in kilowatts and comfort index between 0.0 and 1.0 is the minimum and maximum comfort index respectively. From the results of Fig. 2a it can be observed that in case of power consumption for temperature, proposed system consumed less power as compared to GA based system. So when environmental disruption occurs, proposed system consumed less power as compared to previous GA based system. Less power consumption is guaranteed by the decision of coordinator agent on the basis of optimized parameters. For illumination as shown in Fig. 2b, proposed system confirmed minimum power consumption as compared to previous GA based predicted system. Figure 3c shows the results for the air-quality control. Here also proposed system consumed little power as compared to its counterpart GA based predicted system. Figure 3d shows the total predicted power consumption in case of proposed system and GA based predicted system. The total predicted power consumed by the proposed system is much less than its counterpart GA based prediction system with no hybrid prediction and preprocessing/post-processing involved.

The power disturbance first time arises at 82 min. At that time comfort level of proposed system goes down to 0.970 similar to its counterpart GA based predicted system with no hybrid prediction and preprocessing/post-processing. When second time power disturbance occurs, GA based predicted system degraded immediately goes down as compared to proposed system. At time 115 min proposed system degraded to 0.978 as compared to 0.967 of GA based prediction model with no preprocessing/post-processing at time 113 min. Similarly in all cases of degradation proposed system provides improved comfort index as compared to GA based predicted system where no hybrid prediction and preprocessing involved.

So whenever there is an environmental disturbance, proposed system provides better comfort index as compared to its counterpart GA based predicted system.

Figure 4 shows the results of user comfort index in case of proposed system and GA based prediction system. From Fig. 4, it is clear that proposed system provides enhanced and better comfort index as compared to previous GA based prediction system [3]. Although in proposed system, less power is consumed as compare to its counterpart GA based prediction system, but still proposed system achieved improved and better comfort index as compared to GA based predicted system. Figures 5 and 6 shows comparison of predicted power consumption for individual algorithm i.e. ARIMA model, proposed system and Kalman filter. The comparison is also shown in Table 3. From the results of Fig. 5a it can be perceived that in case of power consumption for temperature, proposed hybrid model consumed less power as compared to ARIMA model and Kalman filter model, while ARIMA model consumed less power as compared to Kalman filter. So when environmental disruption occurs, proposed system consumed less power as compared to ARIMA model and Kalman filter. Less power consumption is guaranteed by picking the best prediction value for hybrid prediction with respect to the minimum power consumption. For illumination as shown in Fig. 5b, proposed system confirmed minimum power consumption as compared to both ARIMA model and Kalman filter, while ARIMA model consumed less power than Kalman filter. Figure 6c describes the results for the air-quality power consumption. Here also proposed system consumed minimum power as compared to ARIMA model and Kalman filter, while ARIMA model consumed less power as compared to Kalman filter. Figure 6d shows the total predicted power consumption in case of ARIMA, proposed system and Kalman filter based prediction. Total predicted power consumed by the proposed system is much less as compared to ARIMA and Kalman filter based prediction, while ARIMA model itself consumed less power as compared to Kalman filter.

So the bottom line is proposed system outperforms individual prediction based on Kalman filter and ARIMA model.

4 Conclusions

In this paper, an enhanced optimized power control model for occupants comfort index, energy efficiency and energy prediction using GA, parallel hybrid prediction and preprocessing/post-processing has been presented. Both issues energy efficiency and occupants comfort index are addressed. To make sure users interaction with the system, users set parameters are considered in deciding the occupants comfort index. The emphasis of proposed work is given to increase occupants comfort index, minimize energy consumption and comparison of proposed system with GA based prediction system, ARIMA model prediction and Kalman filter prediction.

Proposed system performs well as compared to individual ARIMA model and Kalman filter prediction for each of the temperature, illumination, air-quality and total power consumption. The bottom line can be drawn based on the results shown in Table 3. The performance of the proposed hybrid energy prediction system is improved as compared to individual systems based on Kalman filter and ARIMA model. Besides this, proposed model provides not only improved and better comfort index but also consumed less power as compared to previous GA based system [3]. So using GA based enhanced optimized power control model for users comfort index and energy savings, building environment can be made user friendly. The proposed GA based enhanced hybrid prediction model can be integrated with SCADA software of buildings for real-world applications.

Footnotes

Acknowledgments

This work was partly supported by Institute for Information & communications Technology Promotion (IITP) grant funded by the Korea government (MSIP) (No. 10043907, Development of high performance IoT device and Open Platform with Intelligent Software). This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (2013011117), and this research was supported by the MSIP (Ministry of Science, ICT and Future Planning), Korea, under the ITRC (Information Technology Research Center) support program (IITP-2015-H8501-15-1017) supervised by the IITP (Institute for Information & communications Technology Promotion).

References

Ali

and Kim

D.H.

, Optimized power control and comfort management in building environment, pp, FTRA-AIM Advanced IT, Engineering and Management Conference (2013), 145–146.

Ali

and Kim

D.H.

, Effective and comfortable power control model using Kalman filter for building energy management, Journal of Wireless Personal Communication73 (2013), 1439–1453.

Ali

and Kim

D.H.

, Optimized power control methodology using genetic algorithm, Journal of Wireless Personal Communication83 (2015), 493–505.

Wang

, Yang

and Wang

, Multi-agent control system with intelligent optimization for smart and energy-efficient buildings, pp, In Proceedings of the 36th Annual Conference of the IEEE Industrial Electronics Society (2010), 1144–1149.

Dounis

A.I.

and Caraiscos

, Advanced control systems engineering for energy and comfort management in a building environment— a review, Renewable and Sustainable Energy Reviews13 (2009), 1246–1261.

Wang

, Yang

and Wang

, Multi-agent intelligent controller design for smart and sustainable buildings, pp, In Proceedings of 4th Annual IEEE International Systems Conference (2010), 277–282.

Emmerich

S.J.

and Persily

A.K.

, State-of-the-art review of CO₂ demand controlled ventilation technology and application, National Institute of Standards and Technology, Technology Administration, US, Department of Commerce, 2001.

Levermore

G.J.

, Building Energy Management Systems: An Application to Heating, Natural Ventilation, Lighting and Occupant Satisfaction, 2nd ed., London, E & FN SPON, 1992.

Bernard

, Guerrier

and Rasset-Louerant

M.M.

, Optimal building energy management, Part II: Control, ASME Journal of Solar Energy Engineering114 (1982), 13–22.

10.

Curtis

P.S.

, Shavit

and Kreider

, Neural networks applied to buildings— a tutorial and case studies in prediction and adaptive control, ASHRAE Transactions102 (1996), 732–737.

11.

Kolokotsa

, Stavrakakis

G.S.

, Kalaitzakis

and Agoris

, Genetic algorithms optimized fuzzy controller for the indoor environmental management in buildings implemented using PLC and local operating networks, Engineering Application of Artificial Intelligence15 (2002), 417–428.

12.

Kusiak

, Li

and Zhang

, A data-driven approach for steam load prediction in buildings, Applied Energy87 (2010), 925–933.

13.

Siroky

, Oldewurtel

, Cigler

and Privara

, Experimental analysis of model predictive control for an energy efficient building heating system, Applied Energy88 (2011), 3079–3087.

14.

Mossolly

, Ghali

and Ghaddar

, Optimal control strategy for a multi-zone air conditioning system using a genetic algorithm, Energy34 (2009), 58–66.

15.

Wang

, Wang

, Dounis

A.I.

and Yang

, Multi-agent control system with information fusion based control model for smart buildings, Applied Energy99 (2012), 247–254.

16.

Bluyssen

P.M.

, Aries

and Dommelen

P.V.

, Comfort of workers in office buildings: The European HOPE project, Building and Environment46 (2011), 280–288.

17.

Marino

, Nucara

and Pietrafesa

, Proposal of comfort classification indexes suitable for both single environments and whole buildings, Building and Environment57 (2012), 58–67.

18.

Yumurtaci

, Role of energy management in hybrid renewable energy systems: Case study-based analysis considering varying seasonal conditions, Turkish Journal of Electrical Engineering & Computer Sciences21 (2013), 1077–1091.

19.

Huang

and Lam

H.N.

, Using genetic algorithms to optimize controller parameters for HVAC systems, Energy and Buildings26 (1997), 277–282.

20.

Obara

and Kudo

, Multiple-purpose operational planning of fuel cell and heat pump compound system using genetic algorithm, Transaction of the Society of Heating, Air-Conditioning and Sanitary Engineers of Japan9 (2003), 65–75.

21.

Radisa

Z.J.

, Aleksandra

A.S.

and Branislav

D.Z.

, Ensemble of various neural networks for prediction of heating energy consumption, Energy and Buildings94 (2015), 189–199.

22.

Kangji

, Chenglei

, Guohai

and Wenping

, Building’s electricity consumption prediction using optimized artificial neural networks and principal component analysis, Energy and Buildings108 (2015), 106–113.

23.

Betul

B.E.

and Aksoy

U.T.

, Prediction of building energy consumption by using artificial neural networks, Energy and Buildings40 (2009), 356–362.

24.

Wong

S.L.

, Wan

K.K.W.

and Lam

T.N.T.

, Artificial neural networks for energy analysis of office buildings with daylighting, Energy and Buildings87 (2010), 551–557.

25.

Melek

, Energy-savings predictions for building-equipment retrofits, Energy and Buildings40 (2008), 2111–2120.

26.

Sandels

C.J.

, Widen

L.N.

and Andersson

, Day-ahead predictions of electricity consumption in a Swedish office building from weather, occupancy, and temporal data, Energy and Buildings108 (2015), 279–290.

27.

Lapedes

and Farber

, Nonlinear signal processing using neural networks, prediction and system modeling. Technical report LA-VR87-2662, Los Alamos, New Mexico: Los Alamos National Laboratory, (1987).

28.

Van

E.R.J.

, The application of neural network in the forecasting of share prices, Finance and Technology Publishing (1996).

29.

Rodriguez

C.P.

and Anders

G.J.

, Energy price forecasting in the Ontario competitive power system market, IEEE Trans Power Syst19 (2004), 366–374.

30.

, Liu

C.C.

, Mattson

and Lawarree

, Day-ahead electricity price forecasting in a grid environment, IEEE Trans Power Syst22 (2007), 266–274.

31.

Mustafaraj

, Lowry

and Chen

, Prediction of room temperature and relative humidity by autoregressive linear and nonlinear neural network models for an open office, Energy and Buildings43 (2011), 1452–1460.

32.

Kyungtae

, Rogelio

, Pedro

J.M.

and Heejin

, Building hourly thermal load prediction using an indexed ARX model, Energy and Buildings54 (2012), 225–233.

33.

Holland

J.H.

, Adaptation in natural and artificial systems, Arbor

Ann

, MI, The University of Michigan Press, 1975.

34.

Zadeh

L.A.

, Fuzzy algorithms, Information and Control12 (1968), 94–102.