• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>IAEE International Conference;International Association for Energy Economics >FORECASTING ANNUAL ELECTRICITY CONSUMPTION USING ARTIFICIAL NEURAL NETWORKS: THE CASE STUDY OF THE ANDEAN COMMUNITY
【24h】

FORECASTING ANNUAL ELECTRICITY CONSUMPTION USING ARTIFICIAL NEURAL NETWORKS: THE CASE STUDY OF THE ANDEAN COMMUNITY

机译：利用人工神经网络预测年度用电量：以安第斯社区为例

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

OverviewBetween 1973 and 2015, the world total electricity consumption increased from 5117 TWh to 20201 TWh,reflecting an average annual growth rate of approximately 3.4% [1]. This significant change in electricityconsumption has been attributed to numerous factors, including high population growth, increased shares of accessto electricity, rising incomes in developing countries and better energy productivity [2]. Over the next 20 to 25years, the International Energy Agency (IEA) projects that electricity demand will rise by 60% and most of thegrowth, approximately 85%, will occur in developing countries [3]. Due to the rapid economic and structuraltransformation of developing countries, the modeling and prediction of long-term electricity consumption havebecome a key issue for policymakers, private investors, and related organizations. In the last two decades, theeconomic development of the member countries of the Andean Community of Nations, a trade bloc of fourcountries – Bolivia, Colombia, Ecuador, Peru, and the increase in electricity consumption have encouragedgovernments to invest in power capacity. Predicting long-term electricity consumption is by no means a trivialtask as it is dependent on a number of uncertainties such as economic, technological and even political instability.Moreover, reliable electricity consumption forecasts are indispensable for an optimal long-term energy planningof a national power generation system.A review of the literature shows that the vast majority of studies focus on developed economies and that most ofthose studies employ time series or econometric models. In recent years, Computational Intelligence (CI) methodsused for the estimation and forecasting of energy demand and electricity consumption have drawn considerableattention due to their advantage (e.g. capturing the nonlinear relationships of complex data) over traditionalforecasting techniques. Despite the extensive literature on forecasting electricity consumption, there is a lack ofstudies that explore the application of computational intelligence methods for the prediction of annual electricityconsumption in Latin American countries, and in particular, on member states of the Andean Community(Colombia, Ecuador, Peru, Bolivia).In this context, the main aim of this case study is to evaluate the predictive performance of time series andmultilayer perceptron (MLP) artificial neural network (ANN) models. The time series ANNs were constructed toforecast future values of selected socio-economic indicators while the MLP ANN models were used to forecastthe possible future annual electricity consumption for each member state of the Andean Community. Themethodology implemented takes into account historical socio-economic data as well as other variables that couldhave an effect on electricity consumption. The paper is structured as follows: Section 1 presents a brief technicaloverview of artificial intelligence and in particular artificial neural networks. Section 2 describes the methodologyapplied in the development of the ANN model. Section 3 presents the computational results while section 5presents a summary and conclusions. MethodsIn this study, we use an artificial neural network (ANN) method to forecast the annual electricity consumption.The ANN models are trained with previously observed socio-economic and energy statistical data. The data wereobtained primarily from governmental and non-governmental reports, the Latin American Energy Organization(OLADE) and the World Bank. The predicted values of the ANN model were compared with historical data andofficial forecasts of these Latin American countries. Fig. 2 presents a simplified workflow of the electricityconsumption forecasting method.Each of the ANN models was computed under various configurations of hidden layers and number of hiddenneurons. The predictive performance of each neural network model was investigated through well-known forecasterror measures: the mean absolute percentage error (MAPE), root mean square error (RMSE), and the coefficientof determination R2.ğ‘€ğ´ğ‘ƒğ¸ = (1/ğ‘Σ~N_t=1(|y_𑡠− ğ‘¦_ğ‘¡ |/ğ‘¦_ğ‘¡)). 100% (1)ğ‘…ğ‘€ğ‘†ğ¸ = 1/ğ‘Σ~N_t=1(y_𑡠− ğ‘¦_ğ‘¡)~2~(1/2) (2)ResultsThe results presented in this paper show that two different architectures of neural networks, time series ANNscombined with MLP neural networks, are capable of predicting long-term electricity consumption. Furthermore,the prediction performance obtained for each model shows that the approach used in this study could beimplemented as a possible mechanism for long-term energy planning of a national power generation system.ConclusionsUntil recently, the vast majority of studies focused on developed economies and most of those studies employ timeseries or econometric models. This paper aims to fill the gap in the literature on electricity consumption forecastingfor countries that integrate the Andean community using a Computational Intelligence method.
机译：概述 1973年至2015年期间，全球总用电量从5117 TWh增加到20201 TWh， 反映了大约3.4％的年平均增长率[1]。电力的重大变化 消费归因于众多因素，包括人口的快速增长，获取的份额增加 电力，发展中国家的收入增加和更高的能源生产率[2]。在接下来的20到25 十年来，国际能源署（IEA）预计电力需求将增长60％，其中大部分 发展中国家将实现约85％的增长[3]。由于经济和结构的飞速发展 发展中国家的转型，长期用电量的建模和预测 成为决策者，私人投资者和相关组织的关键问题。在过去的二十年中， 一个由四个贸易集团组成的安第斯国家共同体成员国的经济发展 个国家-玻利维亚，哥伦比亚，厄瓜多尔，秘鲁，以及用电量的增加鼓励了 政府投资于发电能力。预测长期用电量绝非易事 这项任务取决于许多不确定因素，例如经济，技术甚至政治不稳定。 此外，可靠的电力消耗预测对于最佳的长期能源规划必不可少 国家发电系统。 对文献的回顾表明，绝大多数研究集中在发达经济体，并且大多数研究 这些研究采用时间序列或计量经济模型。近年来，计算智能（CI）方法 用于能源需求和电力消耗的估算和预测 相对于传统的优势（例如，捕获复杂数据的非线性关系）具有优势 预测技术。尽管有大量的预测用电量的文献，但仍然缺乏 探索将计算智能方法应用于年用电量预测的研究 拉丁美洲国家，尤其是安第斯共同体成员国的消费 （哥伦比亚，厄瓜多尔，秘鲁，玻利维亚）。 在这种情况下，本案例研究的主要目的是评估时间序列的预测性能，并 多层感知器（MLP）人工神经网络（ANN）模型。时间序列人工神经网络构造为 预测选定的社会经济指标的未来价值，同时使用MLP ANN模型进行预测 安第斯共同体每个成员国未来可能的年度用电量。这 所采用的方法论考虑了历史的社会经济数据以及其他可能 对用电量有影响。该文件的结构如下：第1节简要介绍了技术 人工智能，尤其是人工神经网络概述。第2节介绍了方法 在ANN模型的开发中得到了应用。第3节介绍计算结果，而第5节介绍 提出了总结和结论。方法 在这项研究中，我们使用人工神经网络（ANN）方法来预测年度用电量。 使用先前观察到的社会经济和能源统计数据对ANN模型进行训练。数据是 主要从政府和非政府报告获得，拉丁美洲能源组织 （OLADE）和世界银行。将ANN模型的预测值与历史数据进行比较， 这些拉丁美洲国家的官方预测。图2展示了简化的电力工作流程 消费预测方法。 每个ANN模型都是在不同的隐藏层配置和隐藏数量下进行计算的 神经元。通过众所周知的预测研究了每个神经网络模型的预测性能 误差度量：平均绝对百分比误差（MAPE），均方根误差（RMSE）和系数 确定R2。 ğ’€ğ´ğ’ƒğ¸ =（1 / ğ’Î〜N_t = 1（| y_ğ’¡âˆ ğ’¦_ğ‘¡ | / ğ‘¦_ğ‘¡））。 100％（1） ğ’…ğ’€ğ’†ğ¸= 1 / ğ’Î ~~ N_t = 1（y_𑡠−ğ’¦_ğ’¡）〜2〜（1/2）（2） 结果 本文提出的结果表明，神经网络的两种不同架构，即时间序列人工神经网络 结合MLP神经网络，可以预测长期用电量。此外， 每个模型获得的预测性能表明，本研究中使用的方法可能是 作为国家发电系统长期能源规划的可能机制而实施。 结论 直到最近，绝大多数研究都集中在发达经济体上，其中大多数研究都花时间 系列或计量经济模型。本文旨在填补用电量预测文献中的空白 适用于使用计算智能方法整合安第斯社区的国家。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号