Delft University of Technology
Load estimation for microgrid planning based on a self-organizing map methodology
Llanos, Jacqueline; Morales, Raúl; Núñez, Alfredo; Sáez, Doris; Lacalle, Matías; Marín, Luis Gabriel;
Hernández, Roberto; Lanas, Fernando
DOI
10.1016/j.asoc.2016.12.054
Publication date
2017
Document Version
Accepted author manuscript
Published in
Applied Soft Computing
Citation (APA)
Llanos, J., Morales, R., Núñez, A., Sáez, D., Lacalle, M., Marín, L. G., Hernández, R., & Lanas, F. (2017).
Load estimation for microgrid planning based on a self-organizing map methodology.
Applied Soft
Computing
,
53
, 323-335. https://doi.org/10.1016/j.asoc.2016.12.054
Important note
To cite this publication, please use the final published version (if applicable).
Please check the document version above.
Copyright
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent
of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Takedown policy
Please contact us and provide details if you believe this document breaches copyrights.
We will remove access to the work immediately and investigate your claim.
This work is downloaded from Delft University of Technology.
For technical reasons the number of authors shown on this cover page is limited to a maximum of 10.
1
Preprint submitted to Applied Soft Computing December, 2016
Load Estimation for Microgrid Planning based on a Self-1
Organizing Map Methodology 2
Jacqueline Llanos
1
, Raúl Morales
1
, Alfredo Núñez
2
, Doris Sáez
1
, Matías Lacalle
1
, Luis 3
Gabriel Marín
1
, Roberto Hernández
3
, and Fernando Lanas
1
4
1
Electrical Engineering Department, University of Chile, Santiago, Chile 5
2
Section of Railway Engineering, Delft University of Technology, Delft, the Netherlands 6
3
Energy Center, Faculty of Physical and Mathematical Sciences, University of Chile, Santiago, Chile 7
8
Abstract— This study presents a novel load estimation method for isolated communities that do not receive energy or only 9
receive it for a limited time each day. These profiles have been used to determine the installed capacity of generating units for 10
microgrid electrification projects. The social characteristics and lifestyles of isolated communities differ from those in urban 11
areas; therefore, the load profiles of microgrids are sensitive to minor variations in generation and/or consumption. The 12
proposed methodology for obtaining the residential profiles is based on clustering algorithms such as k-means, a self-organizing 13
map (SOM) or others. In this work, SOM clustering is considered because it allows a better interpretation of results that can be 14
contrasted with social aspects. The proposed methodology includes the following components. First, the inputs are processed 15
based on surveys of residents that live in each socio-economic level of housing and the community. Second, family types are 16
clustered using an SOM, from which relevant information is derived that distinguishes one family from another. Third, the load 17
profiles of each cluster are selected from a database. Additionally, social aspects and relevant energy supply information from 18
communities with similar characteristics are used to generate the required database. The SOM for the clustering of families of 19
the community with available energy measurements is used as an initial guess for the clustering of the families in the community 20
with unknown energy measurements. 21
The methodology is applied and tested in the community of El Romeral, Chile, where a microgrid will be installed. The SOM 22
technique compares favorably with a benchmark method that uses the average load profile of a community; furthermore, the 23
SOM clustering algorithm for the methodology is favorably compared with the k-means algorithm because the results obtained 24
by SOM are consistent with the social aspects. 25
26
Keywords: planning, microgrid, Self-Organizing Map (SOM), load profile. 27
J. Llanos, R. Morales, A. Núñez, D. Sáez, M. Lacalle, L.G. Marin, R. Hernandez, and F. Lanas, “Load estimation for microgrid planning
based on a self-organizing map methodology”. Applied Soft Computing, Volume 53, April 2017, Pages: 323-335. DOI: 10.1016/j.asoc.2016.12.054
2
Preprint submitted to Applied Soft Computing December, 2016
I. INTRODUCTION 28
When designing and developing renewable energy projects that can provide power to an area, information should be obtained 29
on the available energy resources and required power supply. Because of the uncertainties surrounding the availability of 30
resources and their consumption, ensuring a sufficient capacity and availability of electricity to supply peak demand and daily 31
energy consumption levels must be prioritized [1]. 32
The planning and operation of traditional low-voltage electrical networks require the use of load models. Most power 33
companies implement systems that can automatically read electricity consumption (AMR, automatic meter reading) and 34
determine consumption profiles. Records of these measurements have been used to determine electricity consumption classes 35
and behavioral patterns of energy consumers and provide significant improvements in electricity demand forecasting. However, 36
there are a number of nAMR customers (users without automated meters) for whom the consumption profile is not known [2]. 37
This study focuses on consumers that live in isolated communities without an energy supply or only a partial supply and for 38
whom historical records of total consumption or housing are not available to use as references when measuring microgrid 39
generating units and increasing the efficiency of providing electricity to these areas. 40
Residential demand accounts for most of the system load in isolated electrical systems. These loads are currently modeled 41
with generalized profiles defined by statistical distributions, such as load profiles based on Gaussian functions that capture 42
residential customer behaviors, which have been traditionally assumed to be homogeneous [3]. In small systems, simply turning 43
on and off several appliances may generate significant disturbances to the overall power consumption profile, and several 44
projects have focused on residential demand and proposed algorithms that track the behavior of small loads indicative of 45
changes in the profile through the use of Bayesian change points to identify loads that may appear unpredictable [4]. 46
Demand profiles have traditionally been generated according to consumption measurements, although techniques have been 47
used to identify characteristic patterns of electrical appliance use, particularly in Canadian households [5]. Similarly, the energy 48
consumption profiles for one or more families can be generated by combining the electricity demand of each appliance with a 49
probabilistic approach [6]. In Dickert and Schegner [7], a load curve model based on a probabilistic time series was presented 50
along with measurements of different types of apparatuses used to determine individual load curves and analyze the sequence 51
and timing of operations to generate probabilistic profiles for each appliance according to the appliance power, use frequency, 52
ignition time, and operation times to obtain a load curve per customer or group. 53
Surveys are also useful tools for generating electrical profiles in domestic buildings as shown by work recently conducted in 54
the UK and reported in [8]. Simulation profiles have also achieved a good approximation of electrical energy usage based on 55
measurements at a substation [9]. Estimating electricity demand is an insufficient approach in cogeneration systems; thus, per-56
hour thermal profiles must be generated to optimize electricity usage [10]. 57
J. Llanos, R. Morales, A. Núñez, D. Sáez, M. Lacalle, L.G. Marin, R. Hernandez, and F. Lanas, “Load estimation for microgrid planning
based on a self-organizing map methodology”. Applied Soft Computing, Volume 53, April 2017, Pages: 323-335. DOI: 10.1016/j.asoc.2016.12.054
3
Preprint submitted to Applied Soft Computing December, 2016
Generating load profiles without measurements is a more difficult task, and limited developments have been achieved in this 58
area. In [2], the authors proposed a method for generating TLPs (typical load profiles) for smart grids by using AMR customer 59
data to analyze loads and generate a virtual load profile (VLP) for nAMR customers, with the data subsequently clustered and 60
classified. 61
A number of studies have considered stages of classification for load profiles generated in their models, and these stages 62
include clustering residential customers according to their appliances, identifying customer groups based on the number of 63
residents [7], and classifying users according to the type of electronics they own and times at which the electronics are operated 64
[6]. In Kim et al. [2], the authors evaluated several classification techniques for classifying AMR user profiles, such as k-means 65
and fuzzy c-means, which were later used to generate nAMR user profiles. 66
Self-organizing map (SOM) methodologies have been proposed in several studies. For short-term load forecasting, [11] 67
proposed the use of an input data classifier based on Kohonen neural networks. In Valero et al. [12], two methods were proposed 68
for short-term load forecasting using SOMs for classifying and memorizing historical data. According to [13], one of the major 69
advantages of using SOM for short-term load demand forecasting is its ability to display an intuitive visualization to compare 70
similar data. In [14], SOMs were used to automatically classify electricity customers based on their domestic energy 71
consumption demand patterns using a measurement database. In [15], SOMs were used for segmentation and demand pattern 72
classification for electrical customers. For short-term load forecasting, a neural model containing up to two hierarchical SOMs 73
was proposed in [16]. In [17] SOMs were used to cluster the data and support vector machine (SVM) to fit the testing data for 74
predicting the daily peak load for mid-term load forecasting purposes. In [18] a soft computing system was proposed for day-75
ahead electricity price based on SOMs, SVM and particle swarm optimization (PSO), improving the forecasting accuracy. In 76
[19], the authors presented three of the most used clustering methods, k-means, k-medoid and SOMs, for clustering domestic 77
electricity load profile using smart metering data, in Ireland. SOM proved to be the most suitable and was therefore used to 78
segment the data; a Davies-Bouldin (DB) validity index was used to identify the most suitable clustering method and an 79
appropriate number of clusters. 80
However, these methods are applicable to only traditional power systems. In the case of microgrids, generating profile results 81
is more difficult because of the high variation and uncertainty of load behavior with regard to domestic energy consumption. A 82
load profile generation method for isolated microgrid projects was presented in [20], in which the information is obtained from a 83
socio-economic survey that is conducted in a community, and an SOM classification stage is used to generate a characteristic 84
load profile for each class. However, the load profiles are based on limited measurements from other grid-connected 85
communities, whose load behavior could differ from that of an isolated community, such as not including electricity 86
consumption measurements. 87
J. Llanos, R. Morales, A. Núñez, D. Sáez, M. Lacalle, L.G. Marin, R. Hernandez, and F. Lanas, “Load estimation for microgrid planning
based on a self-organizing map methodology”. Applied Soft Computing, Volume 53, April 2017, Pages: 323-335. DOI: 10.1016/j.asoc.2016.12.054
4
Preprint submitted to Applied Soft Computing December, 2016
In this paper, an SOM algorithm is used as a clustering method for generating both the clusters and a representative for each 88
cluster. Several clustering techniques exist; a notable example is the k-means technique, in which each cluster is represented by 89
the most centrally located object in the cluster [21]. The k-means algorithm has also been used for power systems applications 90
such as in [22] for identifying similar types of profiles of a practical system for demand variation analysis and energy loss 91
estimation, and in [23] for classifying and recognizing the voltage sag from the measured historical data of a large-scale grid in 92
China. 93
According to [21], there are three main approaches for clustering times series: raw-data-based, feature-based, and model-94
based. Among them, SOM and k-means are raw-data-based methods that allow the user to work with raw data directly. In this 95
paper, a methodology for obtaining the residential profiles is proposed considering clustering algorithms such as k-means, 96
SOMs, or others. In particular, SOM clustering is considered because it allows a better interpretation of results that can be 97
contrasted with social aspects. Thus, the implementation of an SOM is described for estimating load profiles that can be used to 98
plan microgrids according to the unit sizes. The proposed methodology includes the socio-economic characteristics of the grid 99
users (by surveys) as well as the effect of the consumption behavior of the entire community. In addition, this methodology 100
allows new load profiles with the features of each family to be added. The load profiles are obtained from another community 101
with similar characteristics and with available energy supply measurements. The characteristics of this community are clustering 102
by another SOM, which is used to estimate the electrical demand of the families in the community that lacks measurements. In 103
this study, SOMs are employed because they are suitable for representing the utilized surveys. Using SOMs, the survey 104
properties are visualized and analyzed. They correctly represent the similarities among families, which corresponds with 105
expectations from reality and a practical point of view. Thus, a SOM enables a suitable interpretation of the inputs and results 106
and identifies the similarities and differences in the prototypes. Unlike the K-means, the SOMs do not require the cluster 107
number. 108
The proposed methodology is applied to the community of El Romeral, which is located 21 kilometers from La Serena, Chile. 109
This area lacks basic electricity services and potable drinking water [24], and a microgrid is currently being planned for the 110
energy supply; therefore, an estimate of the load profile is required. The required profiles of a similar community are collected 111
from measurements of Huatacondo village, which is located 230 kilometers from Iquique, Chile, and has a microgrid that 112
operates in standalone mode [25].The remainder of this paper is organized as follows. Section II describes the proposed load 113
estimation method that is based on an SOM, Section III provides the case study of the El Romeral community, and Section IV 114
presents the conclusions and suggestions for further research. 115
116
J. Llanos, R. Morales, A. Núñez, D. Sáez, M. Lacalle, L.G. Marin, R. Hernandez, and F. Lanas, “Load estimation for microgrid planning
based on a self-organizing map methodology”. Applied Soft Computing, Volume 53, April 2017, Pages: 323-335. DOI: 10.1016/j.asoc.2016.12.054
