A new open and cheap self-scalable open power meter design suitable

for the NILM

C. Rodriguez-Navarro 1, A. Alcayde 1, V. Isanbaev1, L. Castro-Santos 2, A. Filgueira-Vizoso3, F.G. Montoya 1.

1 Universidad de Almería, Escuela Superior de Ingeniería, La Cañada de San Urbano, 04120, Almería, Spain; email: [email protected],
[email protected], [email protected], [email protected]

2
Universidade da Coruña, Campus Industrial de Ferrol, Departamento de Enxeñaría Naval e Industrial, Escola Politécnica de Enxeñaría de Ferrol,
Esteiro, 15471 Ferrol, Spain; email: [email protected]

3 Universidade da Coruña, Campus Industrial de Ferrol, Departamento de Química, Escola Politécnica de Enxeñaría de Ferrol, Esteiro, 15471 Ferrol,
Spain; email: [email protected]

Abstract. In our world, it has never been more important to optimise the We also have the NILMTK toolkit, which implements two of the
use of energy in such a complex socio-political context, which is clearly algorithms known and studied in the world of disaggregation, such as
dependent on fossil fuels. With the idea that nothing can be optimised that the combinatorial algorithm (CO) and the hidden model algorithm of
cannot be measured, precisely why it is important to use accurate, intelligent Markoff (FHMM). [1], which are the perfect tools to test the reliability
energy meters, which provide us with these measurements of electricity and usefulness of the measures taken with the OMPM.
consumption, to optimise their use using the NILM (Non-Intrusive Load
Monitoring) technique.

Our aim is therefore to make it possible for engineers and researchers to 2. Related work
carry out basic electrical measurements in a reproducible and extensible way, There are commercial and research meters that can be used together
which we believe we have achieved through a new open hardware solution that
or separately to measure electricity consumption or power quality (or
we have called OMPM (Open Multi Power Meter), based on a single
microcontroller with its associated hardware, whose firmware has been
devices that integrate both functions). For example, some power
developed in open source, and several standard measurement modules quality analysers are used by engineers, electricians, maintenance
interconnected by a bus, which we believe offers us an easily implementable technicians and facilities to record power quality, perform diagnostic
device, all without losing the reliability and precision that some non-professional work on electrical systems or equipment, identify energy waste in
devices suffer from. facilities, or detect and prevent energy problems before they occur.
It is precisely the characteristics of this new solution Open Multi Power
Meter (OMPM), based on the new measurements obtained using and adapting
Unfortunately, all these devices are expensive and difficult for non-
the Non-Intrusive Load Monitoring Tool Kit (NILMTK), that allow us to create a qualified users, as they are typically used to perform advanced energy
new DS with which to compare the results obtained by applying this toolkit with savings audits, repairs, network analysis, etc. On the other hand, as for
those obtained by applying it to other public dataset using commercial monitors the software of these devices, it is usually proprietary, although the
or even those based on open hardware (as in the case of the OZM). The results growing trend is that it is constituted on open-source resources, which
obtained will undoubtedly be surprising, given the simplicity and flexibility of this today is evident not only for energy measurement, but for a series of
innovative approach.
areas such as the NILM.
[2] among others.

There are several open-source Arduino-based energy monitoring

1. Introduction
developments:
It is not easy for the interested researcher to find smart meters with
open hardware that can be easily implemented at a reasonable cost
to collect electrical measurements, to optimizing electricity - OpenEnergy Monitor: is a home energy monitoring system
consumption, look for anomalies, analyse patterns, study for analysing real-time energy usage and daily energy
improvements, etc. consumption. This device supports measurements of
active power, RMS voltage and RMS current at a high
To try to fill this gap in electrical measurement equipment, a new low- sampling rate and has an HTML5 interface. It also has signal
cost, easily replicable, and highly scalable hardware has been zero crossing detection, WIFI and Ethernet support, and an
developed, based on a single microcontroller, and a SD reader and API. However, it does not consider the measurement of
several measurement modules on a RS485 bus, thus offering good reactive power, power factor or cosine of fi.
accuracy and great scalability. - Arduino Energy Monitor: an open-source project that uses
an Arduino board, a non-invasive current sensor and
displays the measurements on this LCD screen or on a web
 page. The Arduino can display real-time consumption enough limitations (either commercial options or DIY alternatives).
information, store data, and communicate with other While the OZM is a great high accuracy analyser meter, it is not
devices. Its open-source nature allows for customization designed to be easily scalable and expandable, so the OMPM tries to
and adaptability. It provides an affordable and flexible meet this need, especially for use in the NILM field.
solution for tracking and analysing energy consumption in
various applications, such as home energy monitoring and 3. Methodology
energy efficiency projects.
- Emontx: an open-source energy monitoring system In this paper we present a new hardware designed under the
designed to measure and record electricity consumption in paradigm of open hardware, offering a completely scalable solution
real time. It is used in energy efficiency, renewable energy thanks to the fortunate implementation of an RS485 bus that allows
and building monitoring projects. The hardware includes a us to add as many low-cost measurement modules as required.
device that connects directly to electrical circuits and uses
The new hardware for acquisition and recording of electrical
current and voltage sensors to measure the energy
measurements is based on the following components.
consumed. The data is transmitted via radio frequency (RF)
or wires to a receiver that sends it to a computer or cloud - ESP32 node MCU.
platform for visualisation and analysis. The software - 6 x PZEM004.
associated with Emontx allows the system to be configured - SD card reader.
and calibrated, as well as to visualise the collected data. It - SD card.
also offers logging and long-term data storage functions, - 6 x diode Schottky.
allowing detailed monitoring of energy consumption and - I2C screen of 16x2 (optional).
detection of usage patterns. - A power supply of 5V/300mA.
- Others: A simple switch, a resistor, one box to enclosure,
By the other hand there is other interesting commercial projects etc.
based in Arduino that are not open source like these: On the other hand, with the idea of automating the measurements, a
simple sequencer circuit has been chosen to allow consideration of all
- IotaWatt: an IoT device based on an ESP32 that monitors combinations of on and off applications.
energy consumption in real time, recording data and
transmitting it to the cloud for analysis. It also measures This completely optional hardware consists of the following
energy generated by renewable sources and adapts to components.
different monitoring needs. It aims to improve energy
- Arduino One.
efficiency.
- Module of six opto-isolated relays compatible with
- Smapee: It is a commercial energy monitor that offers a Arduino.
variety of devices to measure and monitor electrical energy - 12V 1Amp power supply.
consumption. It provides a user-friendly interface and
provides detailed information about real-time energy Next, we will describe the measurement system based on the PZEM-
consumption. It also offers logging and analysis capabilities 004 modules, the data acquisition system and finally the sequencer
through its online platform. system.

A. Module PZEM004
There are also other platforms based on other boards, such as
The PZEM004 is an extremely popular low-cost real-time power
Raspberry Pi, where proposals have emerged such as Wattson, which
consumption monitoring module. This is the most important part of
use a non-invasive current sensor and an oled screen, emonPi, a the proposed solution, since it allows us to obtain the five basic
Raspberry Pi-based device for energy monitoring and data logging, electrical characteristics of one circuit, such as RMS voltage, RMS
providing real-time consumption information and online access for current, active power, frequency, and power factor.
analysis, or RPICT, a Raspberry Pi-based hardware project for energy
monitoring that use current transformers to measure and monitor The module is self-powered, with optocoupled outputs for serial
electrical energy consumption offering a cost-effective and communication at TTL level, using Rogowsky coils for current
customizable solution for real-time energy monitoring and analysis. measurement.

In the following figure we can see the block diagram:

Likewise, there are solutions based on the STM32F4 microcontroller
such as Smappee, the latter with the ability to detect some types of
electrical appliances and analyse consumption. Of note is the
OpenZMeter (OZM) [3], an electrical energy meter (with single-phase
and three-phase versions) developed by the Universities of Granada
and Almería, which is also an energy quality analyser. This device is
both open source and open hardware with IoT capabilities that not
only allow us to measure a wide range of electrical variables at a high
sampling frequency of 15625 Hz (voltage, intensity, active power,
reactive power, total harmonic distortion or THD, power factor and
harmonics of both intensity and voltage and power up to order 50),
but also allows us to record and process all these measurements. Figure 1-Block diagram PZEM 004.

If you doubt before such a wide range of options surprises the The core of the module is the Vango Tec 9881, an ARM microcontroller
wonderful difference between some energy monitors present on the Cutx -M0 32-bit, 32kb flash memory and 8kb RAM, designed for
market perform extremely basic functions and estimate the energy applications of control and monitoring of electrical energy with
with a large error, because in general we could say that they have functions of protection against over-current, over-voltage and short-
circuit. It also has communication interfaces such as UART, SPI and I2C, As already mentioned, the assembly is completely scalable according
allowing integration with other electronic devices. to needs thanks to the implementation of an RS485 bus, so that all the
measurements of voltage, current, power, frequency, and power
Below is the internal schematic of this versatile module: factor obtained by each module are transmitted to the main controller
via the RX (GPIO 16) and TX lines(GPIO17). All measurements are
acquired at a frequency above approximately 10Hz.

This design also includes a 2x16 LCD display connected to the

microcontroller via I2C. The wiring is as follows:

• SDA : GPIO 13
• SCL : GPIO 14

The whole assembly is supplied with +5 V DC directly from the

controller's own USB bus, since the consumption of the Rx/Tx part of
each PZEM004 module is exceptionally low, as it is only needed to
power the optocouplers of the transmission part of each module.

Finally, as a choice, a small switch connected to GPIO15 has been

installed to activate the recording of the measurements on the SD
Figure 2 Wiring diagram module PZEM004.
card.

Below we present the diagram of the final circuit that has been
As for the accuracy of each PZEM-004 module, they have a built-in
implemented (the screen of 16x2 not is mandatory but it is
calibration function that allows the offset and gain of the module to
convenient).
be adjusted to ensure that it supplies correct readings.

These are the electrical specifications for measurements with the

PZEM-004 T-100A:

- Voltage: 80~260V, Resolution: 0.1V, accuracy: 0.5%.

- Current: Measuring range: 0~100A, initial measuring
current: 0.024, resolution: 0.001, accuracy: 0.5%.
- Active Power: Measuring range: 0~23kW, initial power:
0.4W, resolution: 0.1W, Display format: <1000W, displays
one decimal, such as 999.9W and ≥ 1000W, shows only
integer, like 1000W, measurement accuracy: 0.5%.
- Power factor: Measurement range: 0.00~1.00, resolution:
0.01, measurement accuracy: 1%.
- Frequency: Measuring range: 45Hz~65Hz, resolution:
0.1Hz and accuracy: 0.5%.
- Active power: Measuring range: 0~9999.99kWh,
resolution: 1Wh, measurement accuracy: 0.5%, display
format: <10kWh, the display unit is Wh (1kWh=1000Wh),
such as 9999Wh and ≥ 10kWh, the display unit is kWh,
such as: 9999.99kWh.

B. Measurement Module
Figure 4-Wiring diagram of the solution of OMPM.

The OMPM uses an ESP32 Node MCU microcontroller to collect and

acquire measurements from the measurement modules. An SD card
adapter using the MISO/MOSI, CS and SCK lines of the controller and As reflected in the diagram above, the RS485 Bus is implemented in
an I2C display (optional) using the SCL/SDA lines for control are non-standard form with the help of Schottky diodes (to block reverse
connected to the ESP32. current that can generate interference) in all transmission lines of
each PZEM004 module, and a common 10K resistor connected.
As you can imagine, the SD card is where we store the measurements
between the positive and that line (to limit the current flowing
in a controlled way, using a file in CSV format for each meter. The SD
through the diodes avoiding a cut or circuit and also to help keep the
card IS attached to SPI bus as follows:
voltage of the RX line thus easing the detection of the signal on the TX
• CS: GPIO 5. line).
• MOSI: GPIO 23. As for the firmware of the ESP32, in order to use each PZEM004
• MISO: GPIO 19. module, we must first program a unique address for each module
• SCK: GPIO 18. individually, so that each one is identified univocally.
In the assembly presented, we have 6 PZEM 004 modules with their These are the addresses of the individual setter counters:
respective Rogowsky coils, which will be used to record the intensity
measurements for 6 electrical devices. The voltage measurements are • 0x110: aggregate consumption.
taken by parallel wiring, which also feeds the 6 measurement modules • 0x120: plug 1.
themselves. • 0x130: plug 2.
 • 0x140: plug 3. solution based on a simple Arduino Uno board connected by 6 digital
• 0x150: plug 4. bits to a relay board with coupled opto inputs supplied the whole
• 0x160: plug 5. assembly with a 12V DC voltage.

As for the acquisition firmware developed for the microcontroller, we

briefly discuss its operation.

First, we initialize the SD card, capture the current date and time by
STP using a network connection and at once we create 6 files for each
application differentiated by the counter number concatenated with
the first capture date.

Once the measurement files are created, we add the headers of the
fields in the first line in NILMTK format: "timestamp,VLN,A,W,F,PF",
which corresponds, respectively "timestamp" to the value of the
timestamp in 13-digit format , "VLN " to the value of the RMS voltage,
"A" at RMS current value, "W" at actual power, "F" at frequency, and
"PF" at power factor. The format is therefore 6 fields separated by
commas, which will help us to find the 5 measurements offered by
each PZEM 004 module, along with the 13-digit timestamp mark.
Figure 6-Picture of the sequencer in action.

The body of the main program periodically takes all the readings of
Obviously, from the relay board, the normally open contacts of each
each meter, making sure before that each meter is active and has its
output relay are connected in parallel to each switch of each of the 5
measurements accessible (otherwise the request would be the next).
applications used in this assembly, leaving their manual switches in
Logically, each group of measurements is recorded in its
the open position.
corresponding file together with the value of the corresponding
timestamp.

The program continuously checks the total number of measurements

D. Desegregation with NILMTK
taken for each meter to ensure that all meters are accessible and
operational. For the disaggregation process using OMPM, we will use the NILMTK
toolkit [4], which we can see in the following figure.
Below is the photo of the final circuit working showing the measures
in the lcd display, the ESP32 Node MCU capturing the measures in a
SD card ( blue LED lighting) and the pzem004 modules working (red
LEDs blinking):

Figure 3-Implementing NILMTK in OMPM.

Figure 5-Photograph of the final assembly.

In this experiment, the target devices connected to each PZEM004

Based on the DSUAL converter (in turn based on the IAWE converter),
module, whose measurements are entered into NILMTK for analysis,
given that in the PZEM modules we do not have the apparent power
are as follows.
or reactive power, (so we have only 5 measurements), a new
1. Meter for measuring total consumption (Aggregation). converter called UALM2 has been created for generate a new dataset
2. Fan. with the measurements of the OMPM consistent in 6 files where each
3. Laptop computer. file contains the fields that make up the timestamp value, together
4. Light bulb. with the 5 fundamental electrical measurements: RMS voltage value,
5. LED light. RMS current value, real power, frequency and power factor.
6. Electric welder.
Once the new dataset has been generated, we can graphically display
the active power, power factor, voltage and current for all
applications, such as the aggregate meter:
C. Sequencer module
In order to simplify the measurements that simulate the real behavior
of an installation using the measurement module presented, we use a
 Figure 7-Key measures for the aggregate.

Figure 10-Both percent disaggregated energy.

It is also remarkably interesting to see how all the measures are

recorded as we can see in the following graph: 4. Results
As a result of the execution of the two algorithms considered, we
obtain a first estimate of what is the best combination, which in
principle tells us that the sampling time to be chosen will be 60 ".

Figure 8-Consumption of the applicative plus the aggregate. Table 1-First estimations.

We can even see the periods when a measurement is recorded at a Once we have an idea of what could be the rightest combination for
defined threshold (including peaks). the disaggregation of energy consumption, using many more sampling
periods, we run the two CO and FHMM algorithms with the three
different filling methods, obtaining the following results:

Figure 9-Registry of activity.

Table 2 Final Results.

In the earlier table, we can see that the best sampling time is 60
The data set is divided into three groups: Training, Validation and seconds for all combinations, being the best combination of the CO
Testing. algorithm with the mean filling method, which confirms that the
system could be interesting from the point of view of NILM.
Then we execute the two algorithms implemented by NILMTK (CO and
FHMM) [3], with the different methods of filling (mean, median and If we compare them with the results obtained by applying NILMTK to
first ) and with different sampling periods (between 1 second and 30 the IAWE dataset, we see that the best result is obtained for a
minutes), to obtain which is the optimal combination, for from there sampling time much greater than 10 minutes.
to generate the disaggregation model and calculate all the metrics
available in NILMTK as F1,EAE and MNEAP.
 Table 6-Results of F1-score metrics for DEPS.

B. EAE Metric

In terms of EAE (absolute energy error), excellent values are

obtained, with practically zero results for all appliances (i.e.,
Table 3-Final Results for Dataset IAWE. there are hardly any absolute differences between the real
energy and that estimated by the model).
We see how the most efficient algorithm for IAWE Dataset the
Combinatorial Optimization (CO) is using the Mean method as filling
method and a period of sampling of 10 minutes, compared to only the
60 seconds needed with the OMPM .

NILMTK has the calculation of evaluation metrics through the use of Table 7 Metrics of EAE with OMPM.

the MeterGroup, for the validation of the results through the

With respect to the results obtained with DEPS, we can see no
validation set. It is necessary to perform, on the models obtained,
differences as they are equally excellent.
different metrics such as FEAC, F1, EAE, MNEAP and RMSE, which
would give us an output somewhat similar to Table 4.

Table 8- Metrics of EAE with DEPS.

C. MNEAP Metrics

The MNEAP metric also gives us particularly good values being the
best without a doubt for the LED lamp followed by the halogen lamp
Table 4 Main metrics obtains for applications.
and the fan. For the laptop or the fryer, they are not so excellent, but
they also get incredibly good values (the arithmetic mean obtained is
0.724).

As we can see, the F1 metric (both expressed as a percentage of so

much per one) presents a percentage of good assertiveness close to
50% in the worst case for the laptop or the fryer, obtaining particularly
good values for the two lights or the fan.
Table 9-Metrics of MNEAP with OMPM.
With regard to the EAE (absolute energy error), excellent values are
obtained, since the results are practically zero for all appliances (i.e., With respect to the results obtained with DEPS, we can see that the
there are hardly any absolute differences between the real energy and results are remarkably similar to the results obtained with OMPM, in
that estimated by the model). fact, the arithmetic mean obtained is terribly similar (0.73).

Similarly, the MNEAP metric also gives exceptionally satisfactory

results.

Finally, the RMSE (root mean square error) gives interesting results, Table 10-Metrics of EAE with DEPS.
with the LED lamp being the best performer, followed by the fan, the
computer, the deep-fat fryer and finally the incandescent lamp.
D. Metric RMSE

A. F1 Metrics On the RMSE metric this time the data are incredibly good for all
applications especially if we compare them with those obtained with
The F1 metric gives particularly reliable results for the fan, the halogen the DEPS.
lamp, and the LED lamp. Perhaps the less good data are the results for
the laptop or, very curiously, for the deep fryer.

Table 11- Metrics of RMSE with OMPM.

It is notorious the significant difference obtained in this metric for

Table 5 Results of F1-score metric for OMPM.
DEPS, since exceedingly high values are obtained except for the values
If we compare them with the results obtained for DEPS [5] except for obtained for the rack (being excellent for this but bad for the other
HVAC_1 and HVAC_2, which have excellent values, the rest of the appliances).
results are even worse than those obtained with the OMPM.
 Table 12-Metrics of RMSE with DEPS.

Index matching

As far as the index correspondence is concerned, we see that for

sampling times less than 1 minute, they behave similarly for the
metrics F1-score (for higher time values, the FHMM algorithm would
certainly be better).

For metrics EAE, the error is zero for all the cases and for metric
MNEAP, the error is greater when we exceed the sampling time of 1
minute, increasing for the combinatorial model and decreasing for
FHMM.

Finally, with respect to the RMSE, even with sampling times of 10

minutes, it stays at low values, without highlighting any particular
algorithm, sampling time or filling method. Figure 13-OMPM Results.

In this case, the comparison is again with DEPS dataset, whose best
performance was with the FHMM algorithm.

Figure 12-OMPM index correspondence.

Figure 14- DEPS Results.

Since the best behavior offered is the combinatorial algorithm, we
now present the different sampling methods for the different
appliances for the dataset OMPM.
The results shown again show that although F1 excels in some
devices with DEPS, the result is terrible in others (HVAC_4), so we
could say that for OMPM the results are more homogeneous.

Regarding the EAE metric, the results obtained are excellent

(identical to those obtained with OMPM) and regarding MNEAP for
DEPS, the excellent result for the rack application stands out, but the
rest are similar. Finally, for the RMSE metric, as already mentioned,
the worst results of OMPM compared to DEPS stand out strongly.

E. Results of the best model

To finish for CO median method, we can see the metrics for OMPM.
 treatment of NILMTK in its separate phases (validation, training, and
metrics).

In particular, if we compare the results obtained by applying NILMTK

metrics to the new dataset produced with OMPM with those obtained
by applying NILMK metrics to the dataset of DEPS (produced with
professional hardware), the differences are very pronounced in
aspects such as the sampling time, which is considerably lower in the
dataset of OMPM, and especially in the RMSE metric, where there is a
difference of the order of 200% compared to the results obtained with
the dataset of DEPS.

With the good results obtained by applying NILMTK metrics to OMPM

data, a new avenue for researchers to generate their own datasets and
to continue improving NILM is certainly opened, since the proposed
solution is fully scalable at low cost, thanks to the fortunate
implementation of a rs485 bus that allows us with a single
Figure 15-Results of the best model with OMPM.
microcontroller to use a large number of channels obtaining all the
Here too, the results of best model with the dataset of DEPS. fundamental electrical measurements with a more than acceptable
accuracy. Also, as this solution is scalable, having been successfully
evaluated with 6 modules, which is more or less the number of circuits
in a house, it could be extended to an even greater number of modules
in the future.

Finally, to evaluate the new hardware, applications with low power

consumption were used (to make the disaggregation more complex),
still obtaining particularly satisfactory results in the metrics, which
makes us think that this new solution is interesting for the
continuation of the research with the NILM. In future work it would be
interesting to further improve the accuracy of the measurement
modules, which we believe could be achieved by improving the power
supply of the measurement modules, since currently each module is
powered directly from the mains voltage by a simple RC circuit, a
rectifier diode, and a Zener diode, with a U3 regulator (it is a 7133) at
the output. We believe this simple solution could be improved by
feeding the regulator from an isolated independent source such as an
Table 13-Results of the best model with DEPS. R05P125.

The data shown again shows that although the metrics F1-score stands REFERENCES
out in some devices with the dataset of DEPS, the result is not so good
in some others (the HVAC_4 and lights_2), so we could say that for the [1] Y. Karen, V. Nick, and P. Thibaut, “Energy Disaggregation
dataset of OMPM the results are more homogeneous. from Non-Intrusive Load Monitoring Combinatorial
Optimization, Factorial Hidden Markov Model,
Too for the metric EAE, the results are again excellent for both Convolutional Neural Network,” AM 207, 2016.
datasets (DEPS and OMPM), and in the other hand the metric MNEAP
for dataset of DEPS, the excellent result for the rack application stands [2] S. Biansoongnern and B. Plangklang, Nonintrusive load
out, but for the rest of appliances are both similar. monitoring (NILM) using an Artificial Neural Network in
embedded system with low sampling rate; Nonintrusive
Finally, for the RMSE metric as we had already seen, the differences load monitoring (NILM) using an Artificial Neural Network
are very remarkable, strongly highlighting the superior results in embedded system with low sampling rate. 2016. doi:
obtained with the dataset of OMPM compared to those obtained with 10.1109/ECTICon.2016.7561398.
dataset of DEPS, which as we see are not so good.
[3] V. Eduardo, A. Alfredo, M. Francisco G., B. Raul, and A.-C.
5. Conclusions Francisco M., “OpenZmeter: An efficient low-cost energy
smart meter and power quality analyzer,” Sustainability
(Switzerland), vol. 10, no. 11, Nov. 2018, doi:
This work presents a new hardware-based solution that is scalable, 10.3390/su10114038.
accessible, and replicable at an extremely low cost, with the high
precision typical of a professional solution. We also use open [4] B. Nipun, K. Jack, P. Oliver, D. Haimonti, and K. William,
software, both in the firmware of the microcontroller and in the “NILMTK: An Open Source Toolkit for Non-intrusive Load
processing software based on the NILMTK toolkit, which has been Monitoring,” Apr. 2014, doi: 10.1145/2602044.2602051.
adapted to support a new set of data obtained with the new
hardware. [5] “AriasSilva/DEPS_NILM_Dataset: DEPS: Dataset de la
Escuela Politécnica Superior.”
It is therefore worth highlighting the creation of a new converter, https://github.com/AriasSilva/DEPS_NILM_Dataset
based on the OMPM measurement files, which generates the new (accessed Jun. 03, 2022).
dataset with 13-digit timestamp support, allowing us to follow the