Elsevier

Energy

Volume 155, 15 July 2018, Pages 705-720
Energy

Forecasting energy demand, wind generation and carbon dioxide emissions in Ireland using evolutionary neural networks

https://doi.org/10.1016/j.energy.2018.04.192Get rights and content

Highlights

  • Covariance Matrix Adaptation Evolutionary Strategy applied to energy forecasting.

  • Evolutionary Neural Networks applied to Ireland's energy sector.

  • Forecast wind power, carbon dioxide emissions and power demand in Ireland.

  • Covariance Matrix Adaptation Evolutionary Strategy gives highest forecast accuracy.

  • Evolved network maintains reasonable forecast accuracy when predicting 2.5 h ahead.

Abstract

The ability to accurately predict future power demands, power available from renewable resources and the environmental impact of power generation is vital to the energy sector for the purposes of planning, scheduling and policy making. Machine learning techniques, neural networks in particular, have proven to be very effective methods for addressing these challenging forecasting problems. This research utilizes the powerful evolutionary optimisation algorithm, covariance matrix adaptation evolutionary strategy, as a means of training neural networks to predict short term power demand, wind power generation and carbon dioxide intensity levels in Ireland over a two month period. The network is trained over one month and then tested over the following month. A neural network trained with covariance matrix adaptation evolutionary strategy performs very competitively when compared to other state of the art prediction methods when forecasting Ireland's energy needs, providing fast convergence, more accurate predictions and robust performance. The covariance matrix adaptation evolutionary strategy trained network also gives accurate predictions when predicting multiple time steps into the future.

Introduction

Worldwide there is an ever increasing demand for energy due to population growth, increased living standards and industrial development. This ever increasing appetite for energy gives rise to a number of problems, namely: 1) Producing the energy necessary to meet the power demand, 2) Reducing the harmful atmospheric pollutants that result from the power generation process, 3) Incorporating renewable energy sources into the power generation process. In each of these problems facing the energy sector, it is vital to develop accurate forecasting methods.

When generating power it is crucial to be able to accurately forecast energy demands in both the short term and long term. Long term energy forecasting enables policy makers, planners and engineers to prepare for future energy needs by building and developing infrastructure to generate power. Short term energy forecasting is critical to energy production as it is vital to know how much energy will be needed in the near future so that power generators can be scheduled to meet future energy needs.

Minimizing the carbon footprint of the power generation process is very important to the energy sector in recent years. It is well known that burning fossil fuels such as coal will produce harmful atmospheric pollutants such as Sulfur Dioxide (SO2), Nitrogen Oxide (NOx) and Carbon Dioxide (CO2). This is a problem as these chemical compounds directly contribute to global warming. In the 2015 Paris Climate Conference (COP21) nearly 200 countries worldwide agreed to cut green house gas emissions in the coming years. In order to reduce the production of these harmful atmospheric pollutants, it is vital to be able to predict how much of these pollutants will be produced.

Due to the harmful atmospheric effects of burning fossil fuels, many countries world wide have resorted to renewables as a source of energy, e.g. wind, wave and solar energy. The primary drawback with these renewable resources is that they do not provide consistent energy. The fact that renewables do not produce a consistent source of energy is a major hurdle that must be overcome. In order to incorporate these environmentally friendly resources into the power generation process, it is important to know in advance how much energy will be available from these sources. This further motivates the need for accurate forecasting techniques.

Recently neural networks have become a popular machine learning approach for forecasting problems. Neural networks are function approximators inspired by the brain. One of the main design considerations when implementing neural networks is how to optimise the network weights in order to produce the desired output for a given input. Traditionally these weights have been trained using the backpropagation algorithm. In recent years however there has been a large body of research conducted that focuses on the use of evolutionary algorithms to train these neural network weights. One of the most successful evolutionary algorithm is the Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES) algorithm [1]. This algorithm uses evolutionary principles such as selection and mutation and implements a convariance matrix to represent the dependencies between variables to evolve solutions to complex real valued optimisation problems. CMA-ES has never been applied to forecasting in the energy sector despite its impressive performance. It is hoped that the effectiveness of CMA-ES as an optimisation algorithm can benefit the energy sector by evolving more accurate neural network forecasting models.

The accuracy of a neural network trained using CMA-ES will be judged using Ireland's energy sector as a case study. Ireland is a moderately sized country. The population of the Republic of Ireland is approximately 4.75 million people [2]. The population of Northern Ireland is 1.85 million people [3]. This research will evaluate energy data for the entire island, the Republic and Northern Ireland. As Ireland is an island nation in the north Atlantic, a significant portion of its energy is generated from wind. In 2015, 22.8% of its energy is generated from wind farms, however the majority of its energy is generated from thermal power generators [4]. CMA-ES will be compared to a number of state of the art forecasting methods to predict: 1) Energy demand. 2) Wind power generation. 3) CO2 intensity. This will be done using data corresponding to a two month time period.

The research presented in this paper makes the following contributions:

  • 1.

    The application of CMA-ES to forecasting problems in the energy sector.

  • 2.

    Predicting Ireland's power demand, CO2 levels and wind power generation using evolutionary neural networks.

  • 3.

    To compare and contrast the effectiveness of the most well established evolutionary optimisation algorithms, i.e differential evolution, particle swarm optimisation and CMA-ES, in the context of energy forecasting.

  • 4.

    To evaluate how accurate the evolved networks can make predictions for multiple time steps into the future.

The outline of the paper is as follows. Sections 2 Energy forecasting, 3 Neural networks, 4 Swarm and evolutionary methods gives an overview of the relevant literature in energy forecasting, neural networks and evolutionary computing respectively. The experimental methods will be explained in Section 5. Section 6 will present the experimental results. Section 7 will give a discussion of the results. Finally, Section 8 will detail what conclusions can be made and also outline some potential avenues for future research.

Section snippets

Energy forecasting

This section will give an overview of energy demand forecasting, wind power generation forecasting and CO2 level forecasting. This will include outlining the importance and motivation of each forecasting problem along with the prominent research conducted on each problem.

Neural networks

Neural networks were first proposed in the 1960s. Juergen Schmidhuber provides a comprehensive overview of the history of neural networks [29]. Neural networks are computational models that are inspired by the network of neurons that biological brain is comprised of. The function of these networks is to read in an input signal and produce an output signal that corresponds to that input. This functionality is useful for a wide range of problems including: robotics, control, classification,

Swarm and evolutionary methods

The proposal of Genetic algorithms (GA) in the 1970s by John Holland was one of the first evolutionary algorithms proposed [43]. In the years since there have been a diverse range of proposed evolutionary algorithms. Some of the more popular and successful of these methods would include: Differential Evolution (DE), Particle Swarm Optimisation (PSO) and Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES). The advantage of evolutionary strategies such as those listed above is that they

Experimental methods

This section will outline the various experiments conducted and the implementation details of each algorithm. The experiments described in this research were implemented in Java.

Results

This section presents the results of each of the experiments outlined above followed by a discussion in order to gain insight and highlight their significance. The two tailed t-test with a significance level of α=0.05 was conducted to determine significant performance differences when comparing algorithms. All values were rounded to 4 significant figures.

Discussion

The results presented in the previous section demonstrate that evolving neural networks using CMA-ES is an effective approach to developing accurate forecasting models. In terms of accuracy, convergence and robustness, CMA-ES outperforms all other approaches evaluated for each of the three energy forecasting problems. CMA-ES provides the highest forecasting accuracy for the training data in all three forecasting problems. When evaluated using previously unseen test data, CMA-ES performs best on

Conclusion

The primary aim of this research was to investigate if a neural network train with CMA-ES is capable of accurately predicting Ireland's power demand, wind power generation and CO2 levels. The results obtained indicate that CMA-ES can in fact produce accurate predictions for each of these problems. Moreover CMA-ES performs very competitively when compared to other state of the art approaches. CMA-ES performs significantly better than other evolutionary methods, i.e. PSO and DE, on all problems

References (67)

  • L. Wu et al.

    Modelling and forecasting Co2 emissions in the brics (Brazil, Russia, India, China, and South Africa) countries using a novel multi-variable grey model

    Energy

    (2015)
  • Y. Bai et al.

    Air pollutants concentrations forecasting using back propagation neural network based on wavelet decomposition with meteorological conditions

    Atmos Pollut Res

    (2016)
  • J. Krzywanski et al.

    A generalized model of SO2 emissions from large-and small-scale cfb boilers by artificial neural network approach part 2. SO2 emissions from large-and pilot-scale cfb boilers in O2/N2, O2/CO2 and O2/rfg combustion atmospheres

    Fuel Process Technol

    (2015)
  • F. Biancofiore et al.

    Analysis of surface ozone using a recurrent neural network

    Sci Total Environ

    (2015)
  • J. Schmidhuber

    Deep learning in neural networks: an overview

    Neural Network

    (2015)
  • A. Azadeh et al.

    A simulated-based neural network algorithm for forecasting electrical energy consumption in Iran

    Energy Pol

    (2008)
  • L. Ekonomou

    Greek long-term energy consumption prediction using artificial neural networks

    Energy

    (2010)
  • Z.W. Geem et al.

    Energy demand estimation of South Korea using artificial neural network

    Energy Pol

    (2009)
  • B. Kermanshahi

    Recurrent neural network for forecasting next 10 years loads of nine Japanese utilities

    Neurocomputing

    (1998)
  • A. Kialashaki et al.

    Modeling of the energy demand of the residential sector in the United States using regression models and artificial neural networks

    Appl Energy

    (2013)
  • H. Pao

    Forecasting energy consumption in taiwan using hybrid nonlinear models

    Energy

    (2009)
  • H. Ye et al.

    Modeling energy-related CO2 emissions from office buildings using general regression neural network

    Resour Conserv Recycl

    (2018)
  • G. Osório et al.

    Short-term wind power forecasting using adaptive neuro-fuzzy inference system combined with evolutionary particle swarm optimization, wavelet transform and mutual information

    Renew Energy

    (2015)
  • G. Chang et al.

    An improved neural network-based approach for short-term wind speed and power forecast

    Renew Energy

    (2017)
  • J.H. Kämpf et al.

    A hybrid cma-es and hde optimisation algorithm with application to solar energy potential

    Appl Soft Comput

    (2009)
  • K. Mason et al.

    A meta optimisation analysis of particle swarm optimisation velocity update equations for watershed management learning

    Appl Soft Comput

    (2018)
  • S. Bahrami et al.

    Short term electric load forecasting by wavelet transform and grey model improved by pso (particle swarm optimization) algorithm

    Energy

    (2014)
  • A. Selakov et al.

    Hybrid pso–svm method for short-term load forecasting during periods with significant temperature variations in city of burbank

    Appl Soft Comput

    (2014)
  • Y. Yang et al.

    Modelling a combined method based on anfis and neural network improved by de algorithm: a case study for short-term electricity demand forecasting

    Appl Soft Comput

    (2016)
  • F. Kaytez et al.

    Forecasting electricity consumption: a comparison of regression analysis, neural networks and least squares support vector machines

    Int J Electr Power Energy Syst

    (2015)
  • K. Mason et al.

    A multi-objective neural network trained with differential evolution for dynamic economic emission dispatch

    Int J Electr Power Energy Syst

    (2018)
  • H.-T. Pao et al.

    Modeling and forecasting the co 2 emissions, energy consumption, and economic growth in Brazil

    Energy

    (2011)
  • N. Hansen et al.

    Completely derandomized self-adaptation in evolution strategies

    Evol Comput

    (2001)
  • Cited by (0)

    View full text