Forecasting regional electricity load based on recurrent support vector machines with genetic algorithms

Pai, Ping-Feng; Hong, Wei‐Chiang

doi:10.1016/j.epsr.2005.01.006

Cited by 348 publications

(135 citation statements)

References 13 publications

Supporting

Mentioning

131

Contrasting

Unclassified

Order By: Relevance

“…e.g. Ying & Pan, 2008;Yun, Quan, Caixin, Shaolan, Yumin, & Yang, 2008) or other methods, such as neural networks (Hsu & Chen, 2003) or support vector machines (Pai & Hong, 2005) for STLF purposes, mostly on a more aggregated level.…”

Section: Short-term Load Forecasting Approachmentioning

confidence: 99%

An efficient two-stage algorithm for decentralized scheduling of micro-CHP units

Jochem

Schönfelder

Fïchtner

2015

European Journal of Operational Research

View full text Add to dashboard Cite

In this paper we present an efficient two-stage hierarchical decomposition algorithm aiming at determining economically improved operation schedules for residential proton exchange membrane fuel cell microcombined heat and power (PEMFC micro-CHP) units and optimizing local charging of electric vehicles (EV) in the same household. Based on an individual short-term load forecasting (STLF) approach (imperfect forecast) for households implemented as an adaptive network-based fuzzy inference system (ANFIS), a mixed-integer linear program (MILP) and a two-stage greedy algorithm are used for determining optimized schedules based on a rolling-window approach. The results of the case study performed for eight variants in exemplary German households reveal that with both the MILP and the algorithmic approach, significant economic savings can be achieved compared to the standard heat-led strategy. Compared to the MILP, however, the two-stage algorithm has the additional advantage of a reduced computing time of only about 151 . Deviations from the MILP solutions are mostly smaller than 3 percent regarding the annual supply costs. Moreover, the comparison between the use of perfect and imperfect demand forecasts quantifies additional average losses due to forecasting errors of 2 percent and 3.3 percent at the maximum. Altogether, the algorithmic approach Postprint of the article "An efficient two-stage algorithm for decentralized scheduling of micro-CHP units" In European Journal of Operational Research, Volume 245, Issue 3, Pages 862-874, https://doi.org/10.1016/j.ejor.2015 3 seems to be convincing for real applications in households due to its good results, high reliability, easy implementation, and short computing times. The combination of a micro-CHP unit and an EV is highly synergetic.

show abstract

Section: Short-term Load Forecasting Approachmentioning

confidence: 99%

An efficient two-stage algorithm for decentralized scheduling of micro-CHP units

Jochem

Schönfelder

Fïchtner

2015

European Journal of Operational Research

View full text Add to dashboard Cite

show abstract

“…When the quantity of data is larger, it is required for large computing resources solving process of GASVM, which influenced the speed of assessment. Processing and obtain the index of comprehensive land utilization, can improve the accuracy and efficiency of land utilization evaluation by RR -GASVM [16].…”

Section: Resultsmentioning

confidence: 99%

Evaluation of Comprehensive Utilization of Land of Jiangzhehu in China Based on RR-GASVM

Chen¹

2014

TOCSJ

View full text Add to dashboard Cite

This paper established evaluation index system of comprehensive utilization of land from the three aspects of economic benefits, ecological benefits and social benefits. Firstly, ridge regression (RR) is used to the indicator index system for screening, and then genetic algorithm and support vector machine (that is GASVM) is trained to the transformed samples. Moreover, the comprehensive evaluation results between the GASVM and ridge regression and genetic algorithm and support vector machine(that is RR-GASVM) is compared, which results by MATLAB show that using RR-GASVM is more accurate and higher efficient for the comprehensive utilization of land evaluation than that of GASVM. The combination of ridge regression and genetic algorithm and support vector machine can improve the decision ability of SVM, so that the decision has higher adaptability, which is the development direction of future research. At last, based on the results of this study, some suggestions are put forward accordingly. The area of Jiangsu, Zhejiang and Shanghai (that is Jiangzhehu) should consolidate the existing achievements and strengthen the innovation of the land management system on the basis of utilizing land comprehensively, realize the intensive, large-scale operation and speed up agricultural modernization development by accelerating rural land circulation, and improve the efficiency of land utilization in order to save the land of urban construction. It is clear that people should insist the principle of paying equal attention to development and consolidation, and strengthen the cultivated land's ability to resist natural disasters in order to improve the ecological environment at the same time that improved the land utilization.

show abstract

“…A recent survey on Time Series prediction using Support Vector Machines [7] showed that most of the algorithms in a survey were implemented in financial market prediction and electric utility forecasting. Pai and Hong [16] proposed Recurrent Support Vector Machine that used Genetic Algorithm for optimization of SVM. Li et al [17] introduced combination of "similar day method" with Support Vector Regression (SVR) that gives lower error as "similar day method" positively influences the estimation of output.…”

Section: Related Workmentioning

confidence: 99%

Supplier Short Term Load Forecasting Using Support Vector Regression and Exogenous Input

Matijaš¹,

Vukičević²,

Krajcar³

2011

Journal of Electrical Engineering

View full text Add to dashboard Cite

In power systems, task of load forecasting is important for keeping equilibrium between production and consumption. With liberalization of electricity markets, task of load forecasting changed because each market participant has to forecast their own load. Consumption of end-consumers is stochastic in nature. Due to competition, suppliers are not in a position to transfer their costs to end-consumers; therefore it is essential to keep forecasting error as low as possible. Numerous papers are investigating load forecasting from the perspective of the grid or production planning. We research forecasting models from the perspective of a supplier. In this paper, we investigate different combinations of exogenous input on the simulated supplier loads and show that using points of delivery as a feature for Support Vector Regression leads to lower forecasting error, while adding customer number in different datasets does the opposite.

show abstract

Forecasting regional electricity load based on recurrent support vector machines with genetic algorithms

Cited by 348 publications

References 13 publications

An efficient two-stage algorithm for decentralized scheduling of micro-CHP units

An efficient two-stage algorithm for decentralized scheduling of micro-CHP units

Evaluation of Comprehensive Utilization of Land of Jiangzhehu in China Based on RR-GASVM

Supplier Short Term Load Forecasting Using Support Vector Regression and Exogenous Input

Contact Info

Product

Resources

About