Customer grouping and load profiling

In the long-term forecasting, the customer groups have to be valid and the AMR data correct. However, the customer grouping phase can be problematic; a significant challenge is to define a suitable number of customer groups. In addition, customers have to be classified reliably, because the existing customer groups in the DSO’s data systems are not necessarily valid. More information of customers can be obtained from spatial combination of the DSO’s external and internal data. Typically, the same types of customers have similar load curves, but there may be large variations in loads.

Furthermore, there may be significant differences between customer groups. There are many possible ways to classify customers. Previously, the customer classification has been made based on national-level models in Finland. These SLY load profiles include numerous customer groups and precise classifications (SLY, 1992). Figure 5.6 shows a typical example of the customer classification in Finland.

Figure 5.6. Example of customer grouping. Reproduced from (SLY, 1992).

However, the DSOs may also have classified their customers by themselves. In the customer grouping, the type, characteristics, and number of customers in the research area should be determined. Further, the amount of initial data has an impact on the number of customer groups. At least, the customers should be grouped into the following main categories: residential, agriculture, service sector, and industrial customers. Further, the residential customers should be divided into detached and terraced houses, apartments, and holiday homes. In addition, detached houses with direct electric and non-electric heating should be analysed separately. Consumption in agricultural, industrial, and service groups may vary markedly. Again, different building types and buildings with electric heating systems should be analysed separately. It is reasonable to study specific customers with significantly diverging electricity consumption patterns separately.

Customer grouping plays a crucial role in the forecasting process. The main benefit of the customer classification is that a similar analysis can be made for the same type of customers. For instance, customers living in detached houses with electric space heating can be modelled in a similar way in the whole case area. The categories have to be analysed separately, because the consumption patterns differ significantly from each other. Furthermore, the loads within the customer groups may vary considerably.

Basically, the number of customer groups should not be too large, because the number of analyses will increase accordingly. In (Lakervi and Holmes, 2003) it is stated that the

5.2 Present load analysis 113

within reasonable limits. Nowadays, however, the computational capacity is higher, and thus, also the number of customer classes can be larger. An essential decision in the case area is to determine the suitable number of customer groups.

In general, customer grouping can be managed by applying load modelling methods.

There are several options to group the customers, for instance the DSO’s present customer grouping supplemented with AMR data or a clustering method. These methods are presented in Section 4.1. Basically, there are two alternatives to carry out customer grouping: a load profile updating method or a clustering method. In the load profile updating, predefined customer groups are used. In the clustering approach, the new load profiles are produced with the new customer groups, where predefined customer groups can be used as a starting point for clustering. By combining clustering and AMR data it is possible to produce load profiles for each customer group. The normalized AMR data have to adequately represent the customer’s loads. Futher, exceptions or events that do not fit into the load profiles have to be removed. Exceptional values can be eliminated for instance by using representative type weeks as discussed in section 4.1.3. The customers’

AMR data should be scaled to 1 so that customers of the same type are classified into the same category. Customer information like annual energy consumption, location, and other additional information can be maintained despite the clustering process. This provides an opportunity to scale the customer profiles based on annual energy consumption, when the load profile uses the characteristic hourly based load profile for a year, and the annual electricity end-use is the same as in the initial stage. After clustering the customer data, it may be difficult to specify different customer types. Here, cross-checking with the original customer type classification can be applied to compare the customer types. The clustering results and the predefined load models may contradict each other. However, the customer groups and customer data can be updated by clustering, and better information of the loads in the case area can be obtained. If exceptions or other unusual phenomena occur in this phase, the required changes are made, and possibly, the clustering phase has to be repeated. Finally, the load profiles and customer groups for load forecasting are achieved.

Clustering benefits are illustrated in Figure 5.7. The figure shows the clustering results of the predefined detached house customers with direct electric heating. The clustering results show that predefined customer groups may include totally different types of customers. The customers that do not belong to the predefined customer group have to be removed from that customer group, and transferred into a customer group that fits best.

After the clustering results, customers have been classified into selected groups, and a load curve has been obtained for each customer group on an hourly basis for a year.

Figure 5.7. Clustering results of the predefined electric heating customers in the case area.

Eventually, the customer grouping produces results that can be applied easily and diversely to the forecasting and modelling of the present loads for long-term load forecasting purposes. In the forecasting process, clustered and normalized load profiles of customers can be applied. Clustered profiles are reasonable a choice for small-scale customers such as residential customers. For large-scale customers, normalized individual load profiles can be applied, as described in section 4.1.3.

In spite of the fact that clustered curves may produce excellent results, they may involve issues that do not always represent all customers very well. In the clustering, the customer’s original load profile (normalized profile) is changed over to a clustered profile. Thus, the customer’s new profile pattern and peak power can be different from the normalized profile. However, in network loads, energy and peak power are almost the same at the primary substation and secondary substation level. On the whole, this means that there have to be enough clustered profiles if the profiles better represent the original load profile, and the clustered profiles are accurate enough.