The measurement concept provides a generic methodological framework to use mobile measurement technology for the energy system analysis of enterprises. Accordingly, it ensures the systematic and efficient collection of submetering data for subsequent data analyses. A detailed step-by-step process of data collection linked to the applied measurement concept is provided in [20]. The concept can be divided into two main sections, each of which has several sub-sections. Figure 1 illustrates the measurement concept with the main sections Measurement Preparation (1) and Measurement Operation (2). While step (1) covers preparatory and organizational measures related to the measurement, step (2) describes measures to perform short-term measurements with mobile measurement technology.

Figure 1.

Measurement concept to ensure an efficient and systematic use of mobile measurement technology for the energy system analysis (adopted and further developed from [20])

The mobile measurement technology is temporarily installed; that is, short-term measurements are conducted for up to three weeks. Due to this time limitation, it is important to consider operational processes during the measurement (e.g. high order volume leading to high consumption). In addition, data validation is necessary to ensure reliable data for the subsequent analysis process. As a further technical requirement, remote access to the submetering data ensures plausibility checks during the measurement and facilitates data export from a Structured Query Language (SQL)-server [21].

The applied mobile measurement technology was selected based on specific criteria (cf. detailed criteria evaluation in [15]). First, the more consumers are measured in parallel, the more effective the disaggregation of consumption. Therefore, a measurement case was selected that allows the parallel measurement of up to six three-phase consumers. Second, electrical power systems in enterprises are often divided into several physically separated sub-distributions. Accordingly, the measurement system consists of three measurement cases, raising the maximum number of parallel measurable consumers to 18. As a result, the measurement system is flexible in its range of applications – an important aspect when dealing with the heterogeneous application field of SMEs [16].

Three-phase electrical consumers can be measured inductively without interrupting the power supply system by installing Rogowski coils or split-core current transformers on each phase. The conductor current (primary side) induces a proportional current signal on the secondary side of the transducer [24]. This signal is then transferred to the measurement controller and converted into digital data. Additionally, the voltage signal is tapped from the enterprises’ electricity network. Thus, the average active power consumption P_avg in a time period T equals the product of voltage (U), current (I) and power factor φ (cosine of the phase angle between current and voltage) [25]:

$P_{\text{avg}}=\frac{1}{T}\times {\displaystyle \underset{0}{\overset{T}{\int }}u(t)\times i(t)dt=U\times I\times \text{cos }\text{}\text{(}\phi \text{)}}$ (1)

The total power consumption of a three-phase consumer can be calculated by adding the active power consumption of the three phases [26]. The measurement system records the data per second and stores it into databases on an SQL-server (assuming stable data reception). Aggregated data is also recorded to facilitate data access (e.g. 15-minute averages).

Concerning measurement priorities, Pareto principle is applied; that is, consumers responsible for 80 % of the total electricity consumption are identified [27]. The measurement concept thus follows a top-down approach. Starting at the low voltage main-distribution (LVMD), the load flows of connected sub-distributions (SDs) and consumption-groups (CGs) are measured. This allows to structure electrical power systems into specific levels as illustrated in Figure 2. Whereas data series of sites or areas usually represent a mix of different electrical consumers, CGs supply technical devices with identical characteristics or intended use. Thus, the most detailed information on consumption behavior can be obtained from a single measured consumer (e.g. a supply air motor in a ventilation system).

Figure 2.

Breakdown of company structure and corresponding levels of the electrical distribution

The measurement concept aims to disaggregate the electrical distribution of each level as well as possible. Therefore, both input load flows (e.g. the total load flow of a main distribution) and output load flows (e.g. several SDs or CGs) of each level are measured simultaneously. Taking these measurement principles into account, a prepared submetering dataset is formatted as shown in Table 1. Therein, time stamp describes the duration of the measurement from start date (t₁) to the end (t₂). For the application of the standardized data analyses to be developed, a measurement resolution of 15 minutes is set as standard. The other columns contain the active power data P of n measured main-distributions (MDs), sub-distributions (SDs) and consumption-groups (CGs). The prepared submetering dataset thus contains both input load flow and several output load flows recorded in parallel.

Due to the limited measurement period, the submetering data just represent a snapshot of the current energetic condition of the enterprise. Therefore, data validation is necessary before the data analysis process. This can be done either by comparing the submetering data with the utility’s registered metering data (historical data and data during measurement period), or by relating the sum of the output load flows to the input load flow at each measurement location in the hierarchical structure of the electrical distribution (cf. Figure 2). The developed data analyses show individual options for data validation in the results section.

Three main methodological steps are necessary to explore the research questions. Firstly, a series of short-term measurements carried out in enterprises serve as case studies. That is, mobile measurements in office buildings, metalworking industries and ventilation systems are included in this study. Secondly, based on the conducted case studies, submetering datasets are prepared. In a third step, standardized data analyses are developed. Finally, indicators and limitations related to the data analyses are explored and discussed. Figure 3 illustrates the method of the present research.

Figure 3.

Method to develop a systematic data analysis concept

The submetering datasets allow to disaggregate the total consumption (e.g. of a main distribution or sub-distribution) to individual sections (e.g. building blocks, electrical SDs or CGs). Thus, a basic understanding of the origin of energy consumption is created (cf. [18]). Furthermore, consumption disaggregation aims to identify significant energy users (SEUs), i.e. sections, areas or consumers that account for a large share of total consumption. These areas can then be targeted to identify inefficiencies and define appropriate measures.

Based on the measurement concept presented, consumption disaggregation can be implemented as follows. At a measured electrical main-distribution (MD) or sub-distribution (SD), the input load flow and several output load flows are measured in parallel. In a time period ∆T (e.g. usually a reference week from the measurement period), the energy consumption of the input load flow W_in is allocated to the output load flows W_{out n}. This can be done in absolute terms (kilowatt hour, kWh) or as a percentage of the input load flow, which is equal to 100%. The basic principle of consumption disaggregation is illustrated in Figure 4.

Figure 4.

Basic principle of consumption disaggregation

According to the top-down measurement principle, consumption disaggregation starts at the low voltage main-distribution (LVMD) (cf. Figure 2). From this level, further main-distributions, sub-distributions and consumption-groups are measured. Any further disaggregation of consumption usually requires additional measurements and time. Therefore, the cost-benefit ratio of each additional measurement must be considered when planning the measurement campaign.

Figure 5 visualizes exemplarily the results of consumption disaggregation for an office building using a Sankey diagram. Alternatively, pie charts can be used to illustrate the distribution of load flows. This case study covered the measurement of the building’s main-distribution and its disaggregation to connected SDs and CGs. Thus, several SDs, which supply different areas of the building, and CGs such as server, heating, refrigeration and elevator could be measured. The analysis covers a period of one reference week from the submetering dataset. Accordingly, the total consumption of the main-distribution in the reference week (18,743 kWh) is disaggregated to the recorded measurement points. By disaggregating to almost 100% of the total consumption, this example represents a best-case scenario for this type of analysis.

Figure 5.

Consumption disaggregation at the main distribution of an office building visualized in a Sankey diagram

To validate the measurement data, the power consumption of the measured input data series (reference week) was compared with the mean weekly power consumption of historical registered metering data. The results show a deviation of 1.7 % between the measured reference week and the historical mean weekly power consumption, indicating that the defined reference week represents a typical operating behavior of the measurement object. A comparison of the measured power consumption of the reference week with registered metering data during the measurement period shows a deviation of 3.4 %. Since this is within the measurement accuracy of the measurement system, it can be assumed that the measurement data is plausible to apply consumption disaggregation to the dataset.

Base load is defined as the permanent minimum load in electric power systems [15]. Thus, there’s a strong economic incentive to reduce base load consumption. Typical base load consumption occurs during off-peak hours, e.g. at night or weekends. However, this depends on the specific operations of the enterprise. As an initial step to reduce base load consumption, consumers responsible for base load consumption must be identified. Data collection and visualization with mobile measurement technology can provide detailed information on base load consumption [15]. Thus, base load disaggregation represents a standardized data analysis for short-term measurements.

In a similar approach to consumption disaggregation, the total base load of a measured level (e.g. MD or SD) is disaggregated (Figure 6). This can be done in a standardized way as follows. First, the average base load of the input load flow P_{base_input} and the output load flows P_{base_output} are read visually or calculated with a tool. A time range of a reference week or the entire measurement period can be used. Second, both absolute values and percentage shares of each sub-distribution or consumption-group can be visualized as part of the total base load of the input load flow. The individual values and shares can then be used to identify electrical consumers responsible for base load consumption.

Figure 6.

Basic principle of base load disaggregation

Figure 7 shows exemplarily the results of base load disaggregation in an office building using pie charts. The analysis showed that approx. 55% (26.64 kW) of the total base load can be allocated to the sub-distributions, each of which serves individual areas of the building. Workspaces and office equipment were thus identified as the consumers responsible for most of the total base load consumption. Further relevant base load consumers are the server-room and the refrigeration system responsible for the cooling of the server-room (cf. Figure 7). Base load disaggregation thus provides a first approach to identify relevant base load consumers. In a further step of analysis, load profile characterization can be used to analyse base load consumption in more detail (cf. section load profile characterization).

Figure 7.

Base load disaggregation at the main distribution of an office building visualized in pie charts

Validation of the measurement data with historical registered metering data from the measurement object indicates that the total base load over the measurement period corresponds to the typical operating conditions of the measurement object (deviation 4.1 %).

In an electric power system, the analysis and reduction of peak loads is particularly important in terms of energy cost savings [15]. In Germany, enterprises with an electricity consumption of more than 100,000 kWh per year are charged with a commodity price (€/kWh) and a capacity charge (€/kW). As the capacity charge depends upon the highest annual energy demand, peak loads have a direct impact on the enterprises’ energy costs. While shifting and reducing peak loads is common practice in enterprises with professional energy management, SMEs have not yet realised their full potential. Thus, the first step is to identify peak loads, i.e. to analyse the time period and frequency of peak loads [28]. In a second step, based on mobile measurement technology and submetering data, peak loads are disaggregated. This allows to identify consumers responsible for peak loads and implement appropriate measures to reduce them.

Figure 8 illustrates the basic principle of peak load disaggregation for submetering datasets. The first step is to determine the time t_peak at which the total peak load P_{peak_input} of the input load flow of a measured level (e.g. MD, SD) occurred during the short-term measurement. Then the active power of the output load flows P_{peak_n} can be disaggregated at time t_peak. To further investigate which sub-distribution or electrical consumer has an increased power consumption at time t_peak, the data immediately before and after the peak load should also be considered. In this way, it is possible to analyse the load gradient of each measuring point that may be responsible for the peak load.

Figure 8.

Basic principle of peak load disaggregation

The results of a peak load disaggregation are visualized in a bar chart based on a submetering dataset from a metalworking industry (Figure 9). The peak load (110 kW) of the submetering dataset was recorded at 10:15 am. The bar chart also shows the disaggregation of the loads before and after the occurrence of this peak. Further, the measured sub-distributions and consumption-groups are labelled. In this specific example, peak load can be attributed to an increased load of the compressor and several machines, whereas other consumers such as lighting or the lounge do not have a major impact on the peak load. Accordingly, the consumers responsible for the peak load during the measurement were identified.

Figure 9.

Peak load disaggregation at the main distribution of a metalworking industry visualized in a bar chart

Peak load disaggregation can thus provide a basis for further analysis, such as identifying the flexibility potential of consumers responsible for peak loads.

As just the highest annual peak load is charged in the electricity bill of the energy provider, a comparison between the highest annual peak load (e.g. from historical registered metering data) and the peak loads which occurred during the measurement period is necessary. In the example given in Figure 9, peak loads during the measurement period were about 18 % lower than the highest annual peak load. Nevertheless, peak loads that have occurred can be clearly attributed to electrical consumers, as the example in Figure 9 illustrates.

Due to their temporary usage, mobile measurements provide a snapshot of energy consumption in enterprises. However, previous analyses and case studies indicated that the operating behavior of electrical consumers can be interpreted based on submetering data from short-term measurements [15]. The more precisely the disaggregation of electricity consumption (cf. level structure, Figure 2) the more detailed the analysis of load characteristics [29]. Therefore, the analysis of load characteristics is particularly effective when consumption groups or even single consumers on machine level are measured.

Two steps are required to perform load profile characterization (cf. Figure 10). First, the output load flows of a measured level are clustered by day type and visualized in scatter plots. Clustering is necessary to analyse the operating behavior specifically for each day type cluster (e.g. Monday to Friday vs. Saturday and Sunday). In scatter plots, daily load profiles of each day type cluster are visualized as a set of points. This allows to observe the total dataset of each measured load flow in a single diagram. Subsequently, consumption behavior and operating conditions of the load flows can be analysed through visual observation of the scatter plots. In addition, box plots can illustrate the dispersion of the data at each discrete time value throughout the day. The interquartile range (IQR) can be used as a measure of dispersion of the data at each discrete time value. The IQR [kW] is calculated as the difference between the third quartile Q₃ and the first quartile Q₁ (2) [30]:

$IQR=Q_3-Q_1=Q_{0.75}-Q_{0.25}$ (2)

Thus, 50% of all data points for each discrete time value are always within the IQR. Consumption patterns and anomalies (e.g. outliers) can also be detected based on the IQR calculation. A common approach is to identify the data points outside 1.5 times the IQR as outliers [30]. As the whiskers of the box plot indicate the range Q₁  - 1.5 (IQR) and Q₃  + 1.5 (IQR), outliers can be detected by visual analysis of the scatter plots.

Figure 10.

Basic principle of load profile characterization

Furthermore, scatter plots allow the data series to be categorized by operating and non-operating time. Thus, the behavior of the electrical consumers can be interpreted individually for each timeline. The same applies to the calculation of IQR values as a measure of the average dispersion of the values in the on and off periods.

Statistical analysis of the measurement series has been identified to describe load characteristics based on calculated parameters. Similar to the approach described in Gontarz et al. (2015) [31], load characteristics of sub-distributions and consumption-groups can be described according to these parameters (e.g. constant load vs. controlled constant load vs. variable load). As the statistical analysis is beyond the scope of this paper, future research could address potential and limitations for short-term measurement data.

The results of load profile characterization are illustrated based on submetering data of a supply air motor and a compressor (Figure 11 and Figure 12). Therefore, both scatter plots contain all measurement data from the short-term measurements clustered by day type “operating days”.

Visual analysis of the scatter plots allows the following conclusions about consumer behavior and operating conditions. Daily operating time is clearly visible in the scatter plot of the supply air motor (06:30 am to 10:00 pm) (Figure 11). Box plots indicate that the power of the motor does not vary much over time. Thus, consumption behavior can be described as almost constant during the operating time. Visual analysis of the non-operating times shows that the supply air motor is in standby mode with virtually no power consumption.

Figure 11.

Load profile characterization of a supply air motor visualized in a scatter plot (18 operating days)

Daily operating time is also visible in the scatter plot of the compressor (06:00 am to 11:00 pm) (Figure 12). Box plots illustrate the cyclical cycling of the compressor during the operating and non-operating time series. Consumption behavior can be described as variable throughout the whole day. Visual analysis of the non-operating time series indicates a base load consumption of approx. 1-2 kW.

Figure 12.

Load profile characterization of a compressor visualized in a scatter plot (9 operating days)

The load characteristics of the supply air motor and compressor can also be compared by calculating the mean IQR values (〖IQR〗_avg) for operating and non-operating time series. The calculated values are shown in Table 2.

On the one hand, the 〖IQR〗_avg values indicate a higher dispersion in the operating time series than in the non-operating time series. Furthermore, the 〖IQR〗_avg values of the compressor are about three times higher than the supply air motor in both time series. The values shown in Table 2 can serve as indicators to evaluate load characteristics of measured sub-distributions, consumption-groups and single electrical consumers.

To sum up, visual observation allows to draw conclusions about the operating behavior of electrical consumers and sub-distributions. Results indicate that particularly cross-cutting technologies such as ventilation systems, compressed air systems or refrigeration systems can be systematically clustered according to their load characteristics. Assuming a data pool containing various load profiles for each cross-sectoral technology, characteristics and indicators for an efficient consumption behavior can be developed.

Further details of consumption behavior can be analysed by reducing the measurement resolution. However, the application of a more detailed load profile analysis also depends on the individual type of load or consumer measured. For instance, the cycle time of a compressor (compressor switches on and off) is often not visible in the 15-min average values. Thus, it is necessary to analyse the consumption behavior at a lower measurement resolution (e.g. 10-s).