Elsevier

Energy

Volume 95, 15 January 2016, Pages 266-273
Energy

Prediction of heat load in district heating systems by Support Vector Machine with Firefly searching algorithm

https://doi.org/10.1016/j.energy.2015.11.079Get rights and content

Highlights

  • District heating systems for increase in fuel efficiency.

  • Control and prediction future improvement of district heating systems operation.

  • To predict the heat load for individual consumers in district heating systems.

  • A process which simulates the head load conditions.

  • Soft computing methodologies.

Abstract

District heating systems operation can be improved by control strategies. One of the options is the introduction of predictive control model. Predictive models of heat load can be applied to improve district heating system performances. In this article, short-term multistep-ahead predictive models of heat load for consumers connected to district heating system were developed using SVMs (Support Vector Machines) with FFA (Firefly Algorithm). Firefly algorithm was used to optimize SVM parameters. Seven SVM-FFA predictive models for different time horizons were developed. Obtained results of the SVM-FFA models were compared with GP (genetic programming), ANNs (artificial neural networks), and SVMs models with grid search algorithm. The experimental results show that the developed SVM-FFA models can be used with certainty for further work on formulating novel model predictive strategies in district heating systems.

Introduction

DHS (District heating system) enables utilization of waste or low-cost heat, which is the main precondition for DHS (district heating system) competitiveness when compared to onsite, individual, boilers [1]. Another precondition is high heat density. Although some previous studies showed that sparse DHSs (systems with low heat density) can be economic [2], [3], [4], [5], [6], their success highly correlates with energy taxation, which is country specific.

A possible solution for further increase of competitiveness of DHSs lies in the improvement of current control strategy [7], [8], [9]. Predictive heat load models of all consumers in the system, or at least the most influential ones, are indispensable as inputs for advanced model predictive control. Predictive models should provide several-hour-ahead forecasts of required heat, where the prediction horizon can be defined according to the endmost consumer in the network. Oktay et al. [10] developed a novel approach to predict the outdoor temperature fluctuations during daytime as a dimensionless temperature variation coefficient to simplify building energy calculations. Papakostas et al. [11] proposed a new procedure for the calculation of equivalent full-load hours of operation for heating and cooling systems, from hourly temperature bin data. Equivalent full hours used for a rough estimation of annual heating and cooling energy requirements in buildings. Korolijaa et al. [12] discussed the building characteristics needed to be considered for energy performance simulation, their values, and how they can be used in parametric studies. Seo et al. [13] developed a nine-node-based Lagrangian finite element model for estimating the heating and cooling demand of a residential building with a different envelope design, which can be useful for an architect or a construction manager in the early design phase. There have been many researches related to prediction using different algorithms [14], [15], [16], [17], [18], [19], [20], [21], [22], [23]. Utilizing the best possible methods using different sources of energies was focus of many researchers in the past [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [46], [47].

In this article, the heat load prediction models of consumers for different prediction horizons using the data acquired from one heating substation in DHS Novi Sad, Serbia is investigated. The proposed models were developed using a soft computing approach, namely SVMs (Support Vector Machines) FFA (Firefly Algorithm), which was used for optimal parameter selection.

SVMs (Support Vector Machines) were applied in different engineering disciplines [34], [35]. The prediction accuracy of an SVM model relies heavily on proper determination of model parameters [36], [37], [38]. Grid search [39] and gradient decent [40], [41] are among many searching algorithms for SVM parameters. Computational complexity is the major disadvantage of these algorithms, which restricted applicability to simple cases. Also convergence to local minima is another disadvantage of these methods. In the most cases the multiple local minima occurs and therefore evolutionary algorithms present goo approach because of their global solution to such problems.

In this study, short term predictive models of consumers' heat load were developed using the SVM method, whereas the parameters were selected using the Firefly algorithm. Results indicate that the proposed models can adequately predict the heat load for different prediction horizons. SVM-FFA results were also compared with the results of GP (genetic programming), ANNs (artificial neural networks), and SVMs with grid search algorithm.

The rest of this paper is structured as follows: materials and methods are presented in Section 2. Section 3, results and discussions are discussed. Finally, conclusion is drawn in Section 4.

Section snippets

System and data description

For model building data from DHS Novi Sad was used. Heat production in DHS Novi Sad is performed in six heat supply units (heat only boilers) and one CHP (combined head and power) unit “Novi Sad”. Heat from CHP is transferred to one main manifold where it is diverted towards the three supply units inside the station [42]. Data was collected during the 2009/10 heating season in one of the system's heating substations. Following five variables were continually measured and logged at

Results and discussion

Predictive models of heat load were developed for 1, 2, 3, 4, 5, 8, and 10 h ahead. Five models, for a prediction horizon from 1 to 5 h ahead, are intended for control purpose, while the models for 8 and 10 h ahead are intended for planning purpose.

Conclusion

Predictive models for consumers' heat load in DHSs were developed using the SVM-FFA method. The proposed SVM-FFA models were obtained with the combination of the two methods, SVM and FFA.

Data collected in heating substations in a DHS during the 2009/10 heating season were used for model building and testing. Seven predictive models for different prediction horizons (from one hour to ten hours ahead) were created using the SVM-FFA method. A comparison of SVM-FFA method with GP, ANNs, and SVMs

Acknowledgement

This project was supported by the High Impact Research Grant (UM.C/625/1/HIR/MOHE/FCSIT/15) and Fundamental Research Grant Scheme (FRGS) - FP071-2015A from the University of Malaya and the Ministry of Higher Education, Malaysia.

References (47)

  • K. Mohammadi et al.

    Economic feasibility of developing wind turbines in Aligoodarz, Iran

    Energy Convers Manag

    (2013)
  • K. Mohammadi et al.

    Assessment of solar and wind energy potentials for three free economic and industrial zones of Iran

    Energy

    (2014)
  • A. Mostafaeipour et al.

    Evaluation of wind energy potential as a power generation source for electricity production in Binalood, Iran

    Renew Energy

    (2013)
  • A. Mostafaeipour et al.

    Economic evaluation for cooling and ventilation of medicine storage warehouses utilizing wind catchers

    Renew Sustain Energy Rev

    (2014)
  • L. Ornella et al.

    Supervised machine learning and heterotic classification of maize (Zea mays L.) using molecular marker data

    Comput Electron Agric

    (2010)
  • P. Jain et al.

    Supervised machine learning algorithms for protein structure classification

    Comput Biol Chem

    (2009)
  • Y. Bao et al.

    A PSO and pattern search based memetic algorithm for SVMs parameters optimization

    Neurocomputing

    (2013)
  • F. Friedrichs et al.

    Evolutionary tuning of multiple SVM parameters

    Neurocomputing

    (2005)
  • A.C. Lorena et al.

    Evolutionary tuning of SVM parameter values in multiclass problems

    Neurocomputing

    (2008)
  • S. Frederiksen et al.

    District heating and cooling, Lund

    (2013)
  • C. Reihav et al.

    Profitability of sparse district heating

    Appl Energy

    (2008)
  • Population Division, Department of Economic and Social affairs, United Nations. World Urbanization Prospects: The 2009...
  • Population Division

    Department of economic and social affairs

    (2010)
  • Cited by (107)

    • Surrogate modeling for long-term and high-resolution prediction of building thermal load with a metric-optimized KNN algorithm

      2023, Energy and Built Environment
      Citation Excerpt :

      The main disadvantage of ANNs is that the algorithm requires a huge number of training data, while the shortage of recorded data directly lowers the accuracy of prediction. Al-Shammari et al. [22] developed a short-term prediction model for heating loads of a district heating system using SVM. In this study, obtained test results of the SVM model showed better prediction precision than the ANN model [22].

    View all citing articles on Scopus
    View full text