Elsevier

Applied Soft Computing

Volume 71, October 2018, Pages 307-316
Applied Soft Computing

Multinodal load forecasting for distribution systems using a fuzzy-artmap neural network

https://doi.org/10.1016/j.asoc.2018.06.039Get rights and content

Highlights

  • A predictor system (multinodal forecasting) is proposed considering several points of the electrical network.

  • The processing time is equivalent to the processing required for global forecasting.

  • The proposed method is developed based on the use of the FANN and the global load participation factor concept.

  • The convergence is significantly faster than backpropagation neural networks (improved benchmark in precision).

Abstract

This work proposes a predictor system (multinodal forecasting) considering several points of an electrical network, such as substations, transformers, and feeders, based on an adaptive resonance theory (ART) neural network family. It is a problem similar to global forecasting, with the main difference being the strategy to align the input and output of the data with several parallel neural modules. Considering that multinodal prediction is more complex compared to global prediction, the multinodal prediction will use a fuzzy-ARTMAP neural network and a global load participation factor. The advantages of this approach are as follows: (1) the processing time is equivalent to the processing required for global forecasting (i.e., the additional time processing is quite low); and (2) Fuzzy-ARTMAP neural networks converge significantly faster than backpropagation neural networks (improved benchmark in precision). The preference for neural networks of the ART family is due to the characteristic stability and plasticity that these architectures have to provide results in a fast and precise way. To test the proposed forecast system, the results are presented for nine substations from the database of an electrical company.

Introduction

Precisely knowing the electrical load is a prior condition for implementing strategies to provide consumers with quality, cost-effective electrical services (e.g., voltage, frequency, and waveform shape). Most of the proposals found in the literature are for global load forecasting, which is the sum of all demanded loads. In this case, the predicted global load is divided considering the several busses of the system using some heuristic method. This division generally leads to certain errors that can compromise the quality of the required studies for a power system's operation (load flow analysis, voltage, and angle stability, among other important studies). Thus, this work proposes to develop a multinodal load forecasting method considering the load at several points of the electrical network, such as substations, transformers, and feeders, based on an adaptive resonance theory (ART) [1] family neural network. It is a similar problem to global load forecasting, with the main difference corresponding to the use of a strategy to prepare the input and output data with several parallel neural modules aggregated to a global predictor. Therefore, historical consumption data provided by the electrical distribution systems are used. In this system, the load behavior is expected to repeat with some uncertainty. This uncertainty can be mitigated by utilizing exogenous information from historical sources in the prediction system. Generally, this was implemented as in the classical example known as the Box–Jenkins approach [2,3]. Today, researchers prefer the use of intelligent techniques, e.g., artificial neural networks (ANNs) [4], fuzzy logic [5], etc. However, multinodal prediction is complex in comparison to global prediction. The multinodal prediction method developed in this work employs two important resources: the use of the fuzzy-ARTMAP neural network (FANN) proposed by Grossberg [1] and the use of a global load participation factor (GLPF) [6,7]. The advantages of this approach are: (1) the low processing time in comparison to global forecasting and (2) using a FANN results in solutions orders of magnitude faster than backpropagation (BP) [8], which is a major benchmark of precision for ANNs. Full convergence is not even assured with BP neural networks when using diversified data. The preference for the ART family of ANNs [1] is due to the characteristic stability and plasticity that these architectures have to provide results quickly and precisely.

It is emphasized that the stability analysis of neural networks is an important matter considering the need to assure convergence in the training phase. Reference [9] provides a study in this sense.

To test the proposed prediction system, the results are presented considering substations of a power system centralized dataset (CDS) [10].

The load forecasting addressed in this paper is for 24 h in advance with 1 h (or ½ h) discretization. However, other resolutions can be achieved with little modification.

Section snippets

Related works

Global load forecasting accounts for the majority of the publications available in the literature based on techniques such as ANN [11], fuzzy logic [[12], [13], [14]], genetic algorithms [13], classical procedures [[15], [16], [17]], fuzzy-ART&ARTMAP ANN [18,19], ARIMA [2,3,16], ANN based on Levenberg–Marquardt [20] training method, ANN by gradient descent learning [21], and load forecasting based on multiregression–decomposition model [22]. There are few publications dealing with multinodal

Proposed methodology

The multinodal load forecasting method proposed in this work is based on ANN concepts [4] and the GLPF [6,7]. The ANN used is the Fuzzy-ARTMAP [1], the implementation of which is presented in Appendix A.

The prediction system is composed of two important parts: (1) global load forecasting (GLF) and (2) multinodal load forecasting (MLF) (composed by n local prediction modules). Executing the GLF, the MLF is realized using several parallel modules, each predicting the load of a considered bus

Results and discussion

It is necessary to normalize the results (Eq. (A.1)) in consideration of the definitions of the input and output vectors of module GLF (Eqs. (1) and (2)) of the FANN processing. This is necessary to maintain every component of vectors a and bpertaining to the interval [0,1]. As part of input vector a corresponds to the binary codification vector E, the normalization can only be done in relation to the data provided on the global load (vector L) and temperature (vector T). For convenience, the

Conclusion

This paper presents a new prediction model for load prediction up to 24 h in advance, based on the use of a FANN. The load forecasting is constituted by global prediction, i.e., considering the sum of all loads on the system, as well as in each node (busses of the electrical system). It is achieved without additional computational processing time costs, as the modules are processed in parallel. A reduced quantity of data is used to execute the local node training, thus reducing the time for

Acknowledgment

The authors thank CAPES (Brazilian Research Foundation) and CNPq (National Council for Scientific and Technological Development) for financial support.

References (42)

  • L.A. Zadeh
    (1965)
  • A.B. Altran

    Intelligent System for Multinodal Load Forecasting of Electrical Power Systems, Ph.D. Thesis

    (1999)
  • K. Nose-Filho et al.

    Short-term multinodal load forecasting using a modified general regression neural network

    IEEE Trans. Power Deliv.

    (2011)
  • P.J. Werbos

    Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences, Ph.D. Thesis

    (1974)
  • H. Zhang et al.

    Comprehensive review of stability analysis of continuous-time recurrent neural networks

    IEEE Trans. Neural Netw. Learn. Syst.

    (2014)
  • CDS-Centralized Dataset

    Electricity Commission for the Load Dataset”

    (2010)
  • J.W. Taylor et al.

    Neural network load forecasting with weather ensemble predictions

    IEEE Trans. Power Syst.

    (August 2002)
  • S.C. Pandian et al.

    Fuzzy approach for short term load forecasting

    Electr. Power Syst. Res.

    (2006)
  • G.C. Liao et al.

    Application of a fuzzy neural network combined with a chaos genetic algorithm and simulated annealing to short-term load forecasting

    IEEE Trans. Evol. Comput.

    (2006)
  • G. Gross et al.

    Short-term load forecasting

    Proc. IEEE

    (1987)
  • I. Moghram et al.

    Analysis and evaluation of five short-term load forecasting techniques

    IEEE Trans. Power Syst.

    (1989)
  • Cited by (30)

    • Multi-node load forecasting based on multi-task learning with modal feature extraction

      2022, Engineering Applications of Artificial Intelligence
      Citation Excerpt :

      Instead, the deep neural network composed of TCN and GRU is embedded into the multi-task model, so our model has extraordinary nonlinear learning ability. ( 2) The model in Abreu et al. (2018), Nose-Filho et al. (2011b) and Abreu et al. (2019) is a single-task model that does not explore the coupling information between nodes well to decrease the prediction accuracy of nodes. Instead, the model proposed in this paper explores the coupling information between total load data and node load data using soft shared multi-task learning to improve the prediction accuracy of nodes.

    • Forecasting of energy demand in virtual power plants

      2022, Scheduling and Operation of Virtual Power Plants: Technical Challenges and Electricity Markets
    • Review of low voltage load forecasting: Methods, applications, and recommendations

      2021, Applied Energy
      Citation Excerpt :

      However, further details regarding hyperparameter tuning could have been provided. The work of Abreu et al. [92] combines ML with symbolic approaches by employing an ANN from fuzzy-adaptive resonance theory (which they refer to as fuzzy ARTMAP neural networks) for load forecasting, while considering different hierarchies of the distribution system. Their modelling comprises two components: global load forecasting (considering the sum of all loads on the system), and multimodal load forecasting (focusing on substations, transformers, and feeders).

    View all citing articles on Scopus
    View full text