Consumption patterns in low voltage grid : An investigation into typical customers

Sammanfattning: The climate crisis is more prevalent today than ever before. Many see a clean, carbonfree, electrification as a must if we are to turn this challenge around. Because more andmore of society runs on electricity there has never been a bigger need for an efficientpower grid.Huge amount of data are gathered from the grid every day and this data holds a lotof information that could help to optimise the power grid. This thesis aimed at investigating how consumption data could be used to understand consumption behaviourswithin the power grid and to try and capture these behaviours with typical customers.Consumption data comes in the form of a time series. In every Swedish house therewill soon be a electricity meter that measures the consumption of the house each hour.This creates a time series of consumption with 24 data points each day. One common,typical behaviour, within the power grid is a peak in consumption during the morninghours when everyone is preparing for work and one in the evening when customers startto arrive home and starts to prepare dinner.The data used in this thesis were based on a geographically restricted area, aroundUppsala, with customers in separate houses and a fuse size of 16 A, 20 A or 25 A. Theanalysed data set held 13997 of Vattenfall Eldistribution customers.One group of customers that are particularly interesting are those that owns an electricvehicle (EV). This is because the charging of the vehicle can be done at power muchlarger than what a normal separate house draws from the grid. In an attempt to gatherinformation about EV owners charging habits, a questionnaire were created and postedto different Swedish EV forums. In total 44 EV owners answered the questionnaire.The method applied in this thesis works by grouping customers based on attributescalculated from the time series of the customers. The attributes calculated within thisthesis were degree of utilization, utilization time, night timeshare of consumption, correlation with aggregated load, correlation with air temperature and load factor. Eachattribute are calculated for winter and summer separately. The k-Mean algorithm isutilised to cluster the customers into ten cluster based on the attributes. The time series of each customer is then scaled down using their daily max and min consumption.Then the correlation between three different percentiles of the time series data of eachcluster are calculated. If the correlation factor is high enough the cluster is said to holdcustomer that consumes according to the pattern described by the 57.5th percentile.Next the correlation between the representative curves of each cluster is calculated toensure that only unique patterns are saved as a typical customer. If the representativecustomer of two clusters show high correlation, they are combined into a new largergroup.iA set of typical customers based on degree of utilization, load factor, correlation andall three combined were created. The results indicate that customers with a low degreeof utilization and a high load factor have a flatter consumption profile. Customers thathave high correlation with aggregated load should be prioritised since they contributemore to the aggregated peak load.Using information from Trafikverket, 68 customers were identified within the original data set as being EV owners. This made it possible to analyse the time series tofind information about their charging behaviour. To find charging events, consecutivehours were analysed to find hours with a difference in consumption of at least 3.7 kWh.Of the 68 analysed, 39 showed more than 20 charging events during 2020. The mostpopular starting time for charging were 16 o’clock and the most common duration werebetween one and four hours.The method used in this thesis creates typical customers that illustrates a commonconsumption profile of customers within a group. If a typical customer that illustratesconsumption values are desired the customers has to be grouped based on some attributethat describes the amount of electricity that they consume. Creating the typical customers with a smaller sample of customer from the group creates a typical curve thatexhibits more of the randomness that characterises the real consumption profiles.