1、附录 A: 外文文献及译文 第一部分：原文 Artificial intelligence in long term electric load forecasting K. Metaxiotis, A. Kagiannas, D. Askounis, J. Psarras Abstract: Intelligent solutions, based on artificial intelligence (AI) technologies, to solve complicated practicalproblems in various sectors are becoming more a
2、nd more widespread nowadays. AI-based systems arebeing developed and deployedworldwide in myriad applications, mainly because of their symbolic reasoning,flexibility and explanation capabilities.This paper provides an overview for the researcher of AI technologies, as well as their current use in th
3、efield of long term electric load forecasting (LTELF). The history of AI in LTELF is outlined, leading to adiscussion of the various approaches as well as the current research directions. The paper concludes bysharing thoughts and estimations on AI future prospects in this area. This review reveals
4、that although stillregarded as a novel methodology, AI technologies are shown to have matured to the point of offering realpractical benefits in many of their applications. Keywords: Artificial intelligence； Electric load forecasting； Energy 1. Introduction In the past two decades. AI has been defin
5、ed as the study of how to make computers do thingsthat, at the moment, people do better. AI provides powerful and flexible means for obtaining solutions to a variety ofproblems that often cannot be solved by other, more traditional and orthodox methods. This review bears witness to the application o
6、f AI technologies in the field of long term electric loadforecasting (LTELF). Certainly, this is not the first paper to review the application of AI basedsystems in energy related problems with varying success. In general, AI developments in the fieldof energy have been reviewed by several authors f
7、rom various points of view. Taylor and Lubkemanreviewed the applications of knowledge based programming to powerengineering problems, describing prototype projectsdeveloped at North Carolina State University,while the survey of Zhang et al. concerned the use of ES technology in electric powersystems
8、. Ypsilantis and Yee presented a review of ESs for SCADA based power applicationsand Lubarskii et al. discussed the use of ESs for power networks. Since that time, several othersurvey papers have been written invariousenergy related areas. However, this paper has a different focus. Writing a fully c
9、omprehensive survey of AI applicationsin energy systems is objectively impracticable. For this reason, our paper aims to create a large knowledgebase for the researcher, introducing him/her to the specific area of AI applications in LTELF andindicating other fields fertile for research. 2. AI applic
10、ations in long term electric load forecasting 2.1Expert systems ESsare one of the most commercially successful branches of AI. Welbankdefines an ES asfollows:An expert system is a program, which has a wide base of knowledge in a restricted domain,and uses complex inferential reasoning to perform tas
11、ks, which ahuman expert could do. In other words, an ES is a computer system containing a well organised body of knowledge,which emulates expert problem solving skills in a bounded domain ofexpertise. The systemis able to achieve expert levels of problem solving performance, which would normally bea
12、chieved by a skilled human, when confronted with significant problems in the domain. The first works in ES application in LTELF were implemented by Rahman and Bhatnagarand Jabbour et al. The objective of these approaches was to use the knowledge, experienceand analytical thinking of experimental sys
13、tem operators. Park et al. made a further step by usingfuzzy logic in an ES for a LTELF problem. In 1990, Ho et al. presented the use of aknowledge based ES in long term load forecasting of a Taiwan power system, while in 1993,Rahman and Hazim tried to generalize his first work. Markovic and Fraissl
14、er proposedan ES approach (based on Prolog) for long term load forecasting by plausibility checking ofannounced demand. In 1995, Kim et al. implemented a long term load forecaster by using ANNs and afuzzy ES, while later, Mori and Kobayashi presented an optimal fuzzy inference approach forthe LTELF
15、problem. Ranaweera et al. proposed a fuzzy logic ES model for the LTELFproblem, which used fuzzy rules to incorporate historical weather and load data. These fuzzy ruleswere obtained from historical data using a learning type algorithm. A back propagation neural network with the output provided by a
16、 rule based ESwas designedby Chiu et al. for the LTELF problem. To demonstrate that the inclusion of the predictionfrom a rule based ES of a power system would improve the predictive capability of the network,load forecasting was performed on the Taiwan power system. The evaluation of the systemshow
17、ed that the inclusion of the rule based ES prediction significantly improved the neural networks prediction of power load. 2.2Artificial neural networks ANNs are an information processing technique based on the way biological nervous systems,such as the brain, process information. The fundamental co
18、ncept of ANNs is the structure of theinformation processing system. Composed of a large number of highly interconnected processingunits (“neurons”) connected into networks, a neural network system uses the human-like techniqueof learning by example to resolve problems. Every neuron applies an input,
19、 activationand an output function to its net input to calculate itsoutput. The neural network is configured for a specific application, such as data classification orpattern recognition, through a learning process called “training”. The first researchers who introduced the ANN application in LTELF w
20、ere Lee et al., whoproposed an innovative ANN methodology for the LTELF problem. Park et al. proposed theuse of a multilayer network with three layers, i.e. one input, one hidden and one output. Thetraining of the network was performed through a simple back-propagation algorithm. Using loadand weath
21、er information, the system produced three different forecast variables. Lee et al. treated electric load demands as a non-stationary timeseries, and they modeled the load profile by a recurrent neural network. In 1992,Peng et al. presented a search procedure for selecting the training cases for ANNs
22、to recognize the relationship between weather changes and load shape, while Ho et al. implementeda multilayer neural network with an adaptive learningalgorithm. Chen et al. proposed an ANN for weather sensitive long term load forecasting, while analternative technique using a recurrent high order ne
23、ural network was considered by Kariniotakiset al. Papalexopoulos et al. proposed the inclusion of additional input variables, suchas a seasonal factor and a cooling/heating degree into a single neural network. Czernichow et al. used a fully connected recurrent network for load forecasting in whichth
24、e learning database consisted of 70,000 patterns with a high degree of diversity. Mandal et al.applied neural networks for LTELF in which the inputs consisted of the past load data only,and no weather variables were used, while Sforna and Proverbio investigated the application of ANNs in LTELF, thro
25、ugh a research project at ENEL, and confirmed their positivecontribution. In1997, Kiartzis et al. presented the Bayesian combined predictor, aprobabilisticallymotivated predictor for LTELF based on the combination of an ANN predictor and two linearregression predictors. The method was applied to LTE
26、LF for the Greek Public Power Corporationdispatching center of the island of Crete. Ramanathan et al. made several comparisonsof statistical, time series and ANN methods for the LTELF. In 1998, Sforna reported the implementation of a software tool, called NEUFOR, based onANN technology and specifica
27、lly designed to meet the operational needs of utility power systemdispatchers regarding online operation, while Papadakis et al. continued to improve theirprevious work. The same goes for Drezga and Rahman. The development of improved neuralnetwork based LTELF models for the power system of the Gree
28、k island of Crete, as well as radialbasis function networks and fuzzy neural type networks, were proposed and discussed by Kodogiannisand Anagnostakis in 1999. In the years 2000 and 2001, several researchers dealt withthe application of ANN to the LTELF problem, with varying success. 3. Conclusions
29、Electricity long term load forecasting is important for the power industry, especially in thederegulated electricity market. Proper demand forecasts help the market participants to maximizetheir profits and/or reduce their possible losses by preparing an appropriate bidding strategy.Traditional stat
30、istics based linear regression methods need modification to capture the more andmore non-linearities in demand signals under the market conditions. What emerges from this discussion is that AI based systems are becoming more and morecommon decision making tools in LTELF. AI methods for forecasting have shown an ability togive better performance in dealing with the non-linearities and
