Building hourly thermal load prediction using an indexed ARX model

Yun, Kyungtae; Luck, Rogelio; Mago, Pedro J.; Cho, Heejin

doi:10.1016/j.enbuild.2012.08.007

Cited by 169 publications

(82 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Examples of these methods are the parametric methods (statistical), nonparametric methods (artificial intelligence), and hybrid methods (two or more methods combined). Parametric methods encompass time-series techniques, autoregressive moving average, linear regression, and the general exponential method [4,5]. Though these models give a reasonably accurate forecast result, they possess major drawbacks such as needing complex computational effort and responding improperly to the nonlinear behaviors of a load demand and other weather-related factors [6].…”

Section: Department Of Electrical and Electronics Engineering Univermentioning

confidence: 99%

Modified meta heuristics and improved backpropagation neural network‐based electrical load demand prediction technique for smart grid

Islam

Baharudin

Nallagownden

2017

IEEJ Transactions Elec Engng

View full text Add to dashboard Cite

For the assessment of 1‐h‐ahead electrical energy demand, this paper presents an improved backpropagation neural network that has been integrated with a simulated annealing algorithm and a chaos search genetic algorithm. A self‐adaptive learning rate and modified momentum factor are suggested to enhance the performance of the traditionally used backpropagation algorithm. For the combination scheme, the initial parameters of the improved backpropagation neural network have been modified through the utilization of the global search capability of the genetic algorithm. This has been enhanced by cat chaotic mapping to increase the genetic algorithm's optimization ability. The strong local search feature of the simulated annealing algorithm has been used to further enhance the solution set that was created by the optimized genetic algorithm. The model is tested for small and large‐sized grid data, integrated with renewable sources. The performance of the model is also verified for fluctuating load demand conditions. Furthermore, the proposed model performance is tested for all the four seasons of the year to validate its efficacy during seasonal variations. The results of the proposed technique in all these scenarios show higher prediction accuracy and fast convergence than the existing methods. The acceptable precision of 1‐h‐ahead load forecast and its adaptation in different load demand conditions determine the usefulness of the proposed model in the modern deregulated power industry. In particular, the model can be effectively implemented for the enhancement of demand response and other dynamic features of the smart grid. © 2017 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

show abstract

Section: Department Of Electrical and Electronics Engineering Univermentioning

confidence: 99%

Modified meta heuristics and improved backpropagation neural network‐based electrical load demand prediction technique for smart grid

Islam

Baharudin

Nallagownden

2017

IEEJ Transactions Elec Engng

View full text Add to dashboard Cite

show abstract

“…Wood et al [1] and Michopoulos et al [32] determine the thermal load using the Energy Plus software (U.S. Department of Energy) [33] for their investigation of the energy, environmental and financial benefits of housing space heating with biomass. Other authors develop load prediction methods based on regression analysis [34] or autoregressive with exogenous time and temperature indexed model [35]. Powell et al [36] find nonlinear autoregressive models with exogenous inputs as the best methodology between several ones based on artificial neural networks.…”

Section: Introductionmentioning

confidence: 99%

Size optimization of a biomass-fired cogeneration plant CHP/CCHP (Combined heat and power/Combined heat, cooling and power) based on Organic Rankine Cycle for a district network in Spain

Uris

Linares

Arenas

2015

Energy

View full text Add to dashboard Cite

“…The objective function (Eq. (14a)) is quadratic and all constraints are linear, therefore the problem (14) can be solved as a strictly convex quadratic program (QP). Such problems can be efficiently solved even for larger values of the prediction horizon, hence exploiting the full potential of the predictive control.…”

Section: Controllers Tuningmentioning

confidence: 99%

“…In the problem formulation (14), each input and each state is considered as an optimization variable. However, the computation cost to solve a linear-quadratic control problem is O N 3 (n x + n u ) 3 , with N the control horizon, n x the number of states and n u the number of inputs [37].…”

Section: State Condensingmentioning

confidence: 99%

“…Low order linear models are usually preferred due to their computational tractability [10]. Therefore, simplified models need to be generated by means of grey-box [11,3,12,13] or black-box system identification such as auto regressive [14], subspace [4] and artificial neural network methods [15] or by simplified white-box modeling [16,17,18,19,20,21]. While black-box identification has the advantage that no prior knowledge of the system is required and that it can deal more efficiently with large sets of data, its prediction performance for longer time horizons (e.g., more than 12 h) is not sufficiently accurate [11].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Impact of the controller model complexity on model predictive control performance for buildings

Picard

Drgoňa

Kvasnica

et al. 2017

Energy and Buildings

View full text Add to dashboard Cite

Model predictive control (MPC) for heating, ventilation and air conditioning (HVAC) in buildings requires accurate controller models of the building envelope and its HVAC systems. Controller models are typically obtained by means of black-or grey-box system identification or using a white-box modelling approach. However, the necessary level of model complexity used by each method in order to obtain good MPC performance remains a priori unknown and no systematic method or examples showing the optimal complexity is available. This paper systematically investigates the required controller model complexity necessary to obtain optimal control performance for a given building. First, a 6-room house is modeled in detail using building energy simulation software. The building model is then linearised to obtain a linear time invariant (LTI) state-space model (SSM) and the upper bound of the control performance is computed using an MPC with the SSM both as controller and as plant model. The accuracy of the SSM (containing more than 250 states) is then artificially decreased by reducing its number of states to different orders ranging from 4 to 100 using balanced truncation model order reduction technique. The performances of MPCs using these controller models are then compared with the upper bound for both a standard MPC formulation (S-MPC) and an offset-free formulation (OSF-MPC) and with the performance of a rule-based-controller (RBC). The procedure is repeated for the same house model with a higher level of insulation and for a lighter weight construction. This paper shows that the controller model should contain a minimum of states to model each zone separately, and that the walls and floors separating the zones should also have enough states to act as a low pass filter with correct cut-off frequency. The minimum number of states further increases with the building mass content. In the case of the investigated 6-room house, the thermal comfort achieved by MPC using a controller model with a minimum of 30 states instead of 20 states was improved with a factor 2 to 6 without significant increase of the energy use, showing that good MPC performances require controller models with a significantly higher number of states than the order used by most of the black-and grey-box system identification techniques. The minimum required number of states might be chosen lower when OSF-MPC is used instead of conventional MPC. However, OSF-MPC might significantly increase the energy use when poor controller models (high model mismatch) are used. Furthermore, if the controller model is an LTI model, this paper shows that the CPU time necessary to solve the MPC optimization problem becomes independent of the number of states of the controller model when a dense approach is used. The controller model can thus be as complex as necessary to produce accurate predictions without increasing the computation time of the optimization.

show abstract

Building hourly thermal load prediction using an indexed ARX model

Cited by 169 publications

References 11 publications

Modified meta heuristics and improved backpropagation neural network‐based electrical load demand prediction technique for smart grid

Modified meta heuristics and improved backpropagation neural network‐based electrical load demand prediction technique for smart grid

Size optimization of a biomass-fired cogeneration plant CHP/CCHP (Combined heat and power/Combined heat, cooling and power) based on Organic Rankine Cycle for a district network in Spain

Impact of the controller model complexity on model predictive control performance for buildings

Contact Info

Product

Resources

About