Overview of the project results

The underlying cause of the issues raised by the UTwente study is a disconnection between idealised type tests for electricity meters which consist of simple single swept tones and the real-world signals present in the grid. These grid signals are induced by the power electronics used in increasing volumes of household appliances which give rise to highly impulsive current waveforms that are rich in harmonics and modulated in amplitude and phase. Meters are currently tested with single tones and whilst this tests the frequency response and issues related to aliasing, the dynamic range of meter electronics is not adequately tested.

Of the meters tested in the initial UTwente study, those that recorded the most startling errors employed Rogowski coils as current transducers (CT). These instruments respond to the differential of current (dI/dt) and their signal processing interface must therefore integrate the CT output to give a current response. As the dI/dt action is an amplifier of spikes, noise and signal edges, the CT output is likely to saturate the signal processing integrator when exposed to impulsive signals. This condition will never be picked up using simple swept sinewaves, as it is the edges typical in real appliance currents that will induce the saturation.

This project has been collecting and analysing real-word waveforms and developing the means to test meters using these waveforms as draft type-tests for the future meter testing. This will comprise of several advances, namely: digitising real world waveforms at Metered Supply Points (MSP), developing advanced digital signal processing (DSP) algorithms to accurately characterise complex, non-stationary waveforms, performing accurate synthesis of complex, non-stationary current signal for use in meter type-testing, and finally establishing benchmark electricity meters and new type tests for meters.

Sample digitisers and transformers for capturing real world waveforms at metered supply points (MSP):

Currently, smart meters are tested using simple signals which contain a fixed amount of interference at the 5th harmonic and with a recently introduced frequency sweep up to 150 kHz. However, this is not representative of household appliances currents which are highly complex and subject to switching and variation. To realistically test meters, the waveforms at the MSPs will be accurately captured using high bandwidth sampling instruments and wide-band transducers. These waveforms will augment the present idealised tests.

A digitiser has been selected with sufficient bandwidth, resolution and accuracy to capture the sort of current waveforms that are thought to cause EMI issues with static electricity meters. It is a commercial device which has been tested, characterised and multiple units have been purchased by project partners so that they can be deployed in several on‑site measurements at MSPs.  Current probes or transducers with enough bandwidth and range to be capable of measuring the high dI/dt currents have been selected and have been assessed and characterised. Associated software to capture, analyse and visualise the digitised waveform data has been written and tested.

A selection of mass-market electrical appliances have been identified and purchased on the basis of their potential to cause fast changing current waveforms.  Such waveforms are caused by power electronics such as those used in motor speed controls or power convertors.  Examples of selected appliances include smart fridge, vacuum cleaner, food blender, solar invertor, speed controlled pump, electric drill, direct drive washing machine, smart TV, a selection of energy saving lamps.

Measurements to digitise the current waveforms from these appliances in the laboratory have been made on this collection of appliance and the digitised waveform data has been stored in data files.

Contacts with utilities, industrial companies and authorities have been made to organise on-site measurements at MSPs to capture the current waveforms that meters are exposed to in domestic and industrial premises around Europe. Data from these on-site measurements of the type which is thought to cause meter errors is being stored for later analysis.

Algorithms to trigger capture and visualise events in the presence of highly implosive signals, and for the parsimonious specification of waveforms:

Capturing sporadic events of interest at MSPs requires the collection and the time-consuming analysis of very large data sets. This makes site measurements difficult, as the investigator is not sure whether any signals of interest have been captured until off-line analysis has been completed. To avoid this, new triggering algorithms based on various methods (rate of change of current, Gaussian filtered dI/dt, wavelets with four different mother functions with various taps) have been developed and have been incorporated into software to detect, capture and visualise each event as used in the above site measurements at MSPs.

Fourier transform methods are normally used to decompose waveforms into their different frequency components, however these transforms give errors when the waveforms are subject to the sporadic switching and variation as seen at real MSPs. Building on the triggering algorithms described above, new transforms based on wavelets and time-frequency distributions are being developed and tested building on work in other fields of applied mathematics.

Feature extraction methods as used in pattern recognition and machine learning are also being investigated as a means of finding the waveform features that cause meter errors.  Alternative metrics are being used to categorise the waveforms using energy and entropy and find correlations of waveform features to meter errors.  To perform well, these techniques require larger data sets which will be provided by the on-site measurements.

These new methods will lead to the efficient, yet accurate representation of long captured time series, suitable for accurate regeneration and inclusion in normative standards.

New testbeds for MID meter type approval:

Currently testbeds sweep a single frequency tone to 150 kHz to test a meter response. New testbeds are being developed, capable of accurately and repeatedly reproducing fast switching real world signals with a target uncertainty of 0.1 %. A specification for a new test bed has been agreed building on the normative specification published in IEC61000-4-19.

Where possible new testbeds will be achieved by modifying existing IEC61000-4-19 testbeds which mix interference with the mains frequency waveform in a so-called “split-signal” method, however signal alignment is critical to avoid distortion of fast edges. To do this, a system is being developed to generate a fast edge current waveforms which will be injected into the circuit and superimposed on the sinewave current that is normally used in this test bed. The generated signals will have a fixed and controllable alignment with existing sinewave.

An arbitrary waveform method is being developed as an alternative testbed and compared to the split-signal method. The digitised waveforms captured from appliances and real MSPs are loaded into the arbitrary waveform generator and can be replayed to produce voltage signals that are applied to current (transconductance) amplifiers.  Current amplifiers with sufficient rise time to produce the required dI/dt levels have been investigated and tested resulting in the selection of a suitable amplifier. An optical isolation amplifier is being developed which will provide the necessary electrical separation of various parts of the test circuit, essential to protect personnel and equipment. The arbitrary waveform method also allows for the possibility to make small changes to waveforms known to cause meter errors to investigate whether the errors are improved or worsened by certain changes to the waveform features.  This waveform editing capabilities are closely linked with the waveform transform work described above.

Schematic of the testbed, with the electricity meter under test which is read out by a PC using an optical pulse detector, a synchronized dual arbitrary waveform generator in combination with voltage and transconductance amplifiers (labelled V and I) and an isolation transformer to generate the test signals, and a current shunt, voltage divider and dual digitizer to measure the test signals.

These testbeds will form the basis for future testbeds to be used in MID meter approvals. The testbeds will then be used to test a representative selection of EU smart meters using regenerated versions of the captured waveforms. The testing will lead to new normative procedures for the future routine type-approval testing of meters in the presence of realistic interference. The results will reveal the extent of metering errors and determine the scope and degree of the standardisation response.

Testing electricity meters and specifying “Benchmark” meters to settle customer billing disputes:

Having tested the European installed meters with realistic waveforms, a comparison table of meters will result from an ensemble of electricity meters from around Europe. This will reveal the extent of the problem and will be an important input to the decision whether to update the MID related normative standards.

Work in in progress to develop a benchmark meter that will be used by electricity supply utilities to settle any customer disputes related to the billing accuracy of their installed electricity meter.  The benchmark meter must be portable and small so that it can be installed in confined spaces, connected in parallel with the customers meter and left in-situ for a prolonged period.

Several benchmark meters approaches have been investigated and a solution based on a commercial power quality analyser suitably modified for this application has been selected.  This selection was made based on a combination of accuracy, utility and being of compact size for meter enclosures.  Work to modify the bench mark instrument including the provision of a safety enclosure with suitable safety connection terminals, has been completed.