ELECTRICAL-LOAD MONITORING"

CROSS-REFERENCE TO RELATED APPLICATIONS

This Patent application claims priority from Italian Patent Application No. 102018000009169 filed on 04/10/2018, the disclosure of which is incorporated by reference.

TECHNICAL FIELD

The present invention relates to a system applied to non-intrusive electrical-load monitoring (so-called NILM) , providing improved performance.

BACKGROUND ART

The recent increase in installations by electricity companies of smart meters, or consumption monitoring devices in residential homes and in commercial or industrial locations has led to an increased interest in monitoring electricity consumption, in order to provide a better service and obtain useful information on the use of appliances and consumer behaviour.

In particular, a non-invasive monitoring system for electrical loads (hereinafter referred to as NILM) determines which household appliances (or other types of electrical loads) are used by private or professional users and their individual energy consumption through the analysis of electrical quantities measured on the electricity supply network; the NILM system is considered a technological alternative to the connection of individual meters to each household appliance/electrical load.

Instead of monitoring each load with a specific meter, the NILM system allows to use only one general meter (so- called measuring device or "meter") that provides aggregated total consumption data, and to distinguish the switching-on of the various household appliances or electrical loads by recognizing their energy profile/pattern, namely, the electrical signal that such switching-on generates, within the aggregated total consumption data. The NILM system is therefore an infrastructure with low-cost installation and maintenance compared to the installation of smart meter (Smart Plug) networks.

In particular, the meter device is usually a commercial device, with suitable technical characteristics for measuring electrical data such as: active power, reactive power, absorbed current, mains voltage and/or other related quantities such as the "Power Factor" or harmonic distortion.

Such meter device performs measurements at predetermined time intervals, with a fixed period; the NILM system typically acquires all of the data measured by the meter device for subsequent processing.

The phase of learning the energy profiles of electric loads (which will then be used, in a known manner, for the disaggregation of the electrical load by the NILM system) can be carried out using various methodologies, more or less invasive for the user.

In particular, in a less invasive manner, standard load patterns/profiles deriving from sample acquisitions can be loaded (for example for the "Refrigerator", "Washing Machine", "Dishwasher" type , etc.); in ways that provide greater interaction with the user, the same user can activate each individual load alternately and start the individual profile acquisitions.

There are several advantages to using NILM systems, including :

- identification of active loads in an environment;

- detection of start-up transients, line or equipment failures ;

- the availability of surveys on both residential and commercial energy consumption;

- the availability of information on the amount of energy to be used in smart grids.

Currently, in NILM systems, sampling of the aggregated consumption data (for purposes of the processing required for load disaggregation) is uniform over time, i.e. the data are measured on the power line by the metering device at regular intervals over time and sent to the NILM system for processing . By way of a non-limiting example, one might consider as the metering device the electrical energy meter, or a specific metering device placed on the power line; this metering device performs measurements with a predetermined measuring frequency (fixed period) and the measured data are sent to the NILM system with a given frequency (which therefore corresponds to the sampling frequency) .

Typically, the sampling frequency is equal to the measuring frequency of the metering device. In some situations (e.g. stationary) such sampling frequency may be oversized, in other situations (e.g. in the presence of sudden variations), it may be limiting, as will be described below with examples.

In particular, in the case of high frequency sampling, the continuous acquisition of high frequency data requires a high computational capacity by the NILM system and long processing times.

In addition, a high sampling frequency results in high bandwidth occupancy on the transmission channel used for the transmission of the data measured by the metering device to the NILM system (in the case in which the same NILM system is arranged remotely from the metering device) .

On the other hand, the sampling frequency cannot be low or limited either, since, in the event of sudden changes or activations of significant loads, the metering device would not be able to provide updated data for the effective recognition of the load.

To date, the issues listed above are addressed in some known solutions by means of an optimization providing for the metering device to perform high-detail processing on the measured signal and send to the NILM system only an approximate version of this signal, in order to reduce the computational load required of the same NILM system.

These solutions, while reducing the computational load and also the bandwidth required for the transmission of the measured data, have the disadvantage of providing the NILM system with approximate information, which may lead to inaccuracies and possible errors in the load-disaggregation operations (or the impossibility of carrying out such disaggregation operations by the NILM system) .

DISCLOSURE OF INVENTION

The aim of the present invention is to provide a solution which overcomes the above-mentioned problems.

According to the present invention a system and method applied to the non-invasive monitoring of electrical loads are therefore provided as defined in the appended claims.

BRIEF DESCRIPTION OF THE DRAWINGS

For a better understanding of the present invention, preferred embodiments are described below, purely by way of non-limiting examples, with reference to the attached wherein:

- Figure 1 is a general block diagram of a system according to the present solution;

- Figures 2A-2B are more detailed block diagrams of a system according to a first embodiment of the present solution;

- Figures 3A-3B are more detailed block diagrams of a system according to a second embodiment of the present solution;

- Figures 4A-4B are more detailed block diagrams of a system according to a third embodiment of the present solution;

- Figures 5A-5B are more detailed block diagrams of a system according to a fourth embodiment of the present solution;

- Figures 6A-6B and 7A-7B show further embodiments of the present solution;

- Figures 8 and 9 show plots of electrical signals relating to detection algorithms implemented in the system, according to further aspects of the present solution;

- Figure 10 is a block diagram of a further embodiment of the system, according to a further aspect of the present solution; and

- Figure 11 shows plots of additional electrical signals relating to the system according to the present solution. BEST MODE FOR CARRYING OUT THE INVENTION

As will be discussed in detail below, one aspect of the present solution envisages the introduction of a new computing module which will be referred to as detection unit; this detection unit may reside in a load disaggregation unit (NILM unit), in a metering unit, or be independent. This detection unit implements suitable processing of the measurement data it receives from the metering unit and may additionally or alternatively carry out independent measurements of other quantities on the same power line with which the metering unit is associated, on which to base at least part of the processing.

The purpose of the aforementioned computing unit is to detect time intervals of significant variations in the aggregated consumption data (and/or additional electrical quantities associated with the power line) so that only or predominantly the data measured by the metering unit at these time intervals are sent to the NILM unit.

According to one embodiment, the computing unit optimizes the sampling frequency of the data to be sent to the NILM unit. In particular, the detection unit detects intervals of significant variations in the aggregated consumption signal provided by the metering unit (or other significant electrical signal) so as to send to the NILM unit the data measured by the metering unit only or predominantly in such intervals (suitably increasing the sampling frequency during such intervals) . In the case of steady or non-significant time intervals of the processed signal, the computing unit instead reduces the sampling frequency to a value suitable for the reduced signal- variability.

In a possible embodiment, the resulting system is thus configured to cause uneven sampling over time (i.e., with a variable sampling frequency) by the NILM unit of the aggregated consumption signal provided by the metering unit coupled to the power line.

By way of a non-limiting case, a variation (beyond a given threshold) in active power consumption or any other electrical quantity (e.g. a voltage or current variation), related to a change in the operating status of one or more electrical appliances or other electrical loads may be considered as a significant event. By way of a further non limiting case, a significant event can be determined by analyses applied to the measurement data of the aggregated consumption signal, such as spectral analyses and Fourier transform calculation for detection of frequency features.

As will be clarified below, the proposed solution, applied to NILM monitoring, improves its efficiency, resulting in lower communication bandwidth occupancy and/or a lower computational load, allowing to limit or reduce the sampling and communication of aggregated consumption data in case of inactivity of electrical loads, or in general during periods of stationary operation, while maintaining the original detail of the measured data at times of greater energy activity.

By way of a non-limiting case, the system may be considered in stationary operation if a profile of the aggregated consumption signal (or another processed signal) does not undergo significant variations over time.

Figure 1 schematically shows a system 1, applied to the non-invasive electrical load monitoring (NILM) , according to one aspect of the present solution, comprising:

a metering unit ("meter") 2, coupled to a power line 3 (by way of example only shown as a two-phase line) to which a building 4 is connected, which may be a household unit or a commercial or industrial building, inside which there are a plurality of electrical loads 5 (only one of such electrical loads 5 being shown by way of example) , the metering unit 2 being configured to measure one or more electrical quantities or parameters (e.g. relating to an active or reactive power, a voltage, a current, etc.) on the power line 3, associated with the consumption of the electrical loads 5, and to provide an associated aggregated consumption signal, denoted as S _c (such aggregated consumption signal S _c is considered to consist of one or more parameters of those available in the metering unit 2, which also implements a response system to data request, so-called "data polling") ;

a load disaggregation unit ("NILM" unit) 6, operatively coupled to the metering unit 2 and configured to process the aggregated consumption signal S _c by means of an electrical load disaggregation algorithm ("NILM" algorithm of a type known in itself, not described herein in detail; as will be discussed below, the NILM unit is configured to process aggregated consumption data that are sampled in a non-uniform manner over time, only or predominantly at time intervals where a significant variation of one or more electrical quantities measured on the power line is detected; and

a detection unit 8, configured to perform a detection algorithm ("AR" algorithm) to detect significant variations in quantities or electrical parameters on the power line 3 (associated with the electrical loads 5) , and operatively coupled to the load disaggregation unit 6, so that reception and/or processing of data by the load disaggregation unit 6 occurs only, or predominantly, at the time intervals in which the aforementioned significant variations occur.

As indicated above, in a possible embodiment, the detection unit 8 may be configured to cause a change in the sampling frequency f _c of the aggregated consumption signal S _c (i.e., the frequency of receiving and/or processing the aggregated consumption signal S _c by the load disaggregation unit 6) .

As will be described in detail, the detection unit 8 may be implemented locally with respect to the metering unit 2 and reside in the same metering unit 2 (which in this case is itself able to detect the aforementioned significant variations), or it can be implemented remotely with respect to the metering unit 2, thus constituting an external and distinct device with respect to the metering unit 2.

Similarly, the load disaggregation unit 6 may be implemented locally with respect to the metering unit 2 and reside in the same metering unit 2, or may be implemented remotely with respect to the metering unit 2, thereby constituting an external and distinct device with respect to the same metering unit 2, in direct wired or wireless connection with the metering unit 2, for example being network-connected (in "cloud") .

In addition, the detection unit 8 may be implemented jointly with the load disaggregation unit 6, in such case being part of a single device, which may again be implemented locally or remotely with respect to the metering unit 2.

Figures 2A-2B show in more detail a first embodiment of the present solution, wherein both the load disaggregation unit 6 and the detection unit 8 are implemented locally to the metering unit 2, in a single measuring and processing device, globally denoted by reference numeral 10.

In detail, the measuring and processing device 10 is coupled to the power line 3 to which the building 4 is connected (building inside which the electrical loads 5, the consumption of which has to be detected and which are coupled to the same power line 3, are located) .

The measuring and processing device 10 comprises:

the metering unit 2, of a known type, configured to measure one or more electrical quantities or parameters on the power line 3 associated with consumption by the electrical loads 5, so as to provide the aggregated consumption signal S _c ; and

a processing module 12, which jointly implements the load disaggregation unit 6 and the detection unit 8.

In particular, the processing module 12 is made in an electronic board 13, provided with a processor 13a (CPU - Central Processing Unit) , or similar digital computing unit, and a memory 13b, operatively coupled to the processor 13a and in which a first and a second computing algorithms are stored, respectively for the implementation of the functions of the load disaggregation unit 6 (NILM algorithm) and of the detection unit 8 (AR algorithm) .

In this embodiment, the measuring and processing device 10, which has available a high detail of the electrical signal under analysis (which is in fact measured with a high measurement frequency by the metering unit 2), may also detect significant events by means of the operation of the detection unit 8 (AR algorithm) and vary the frequency of receipt and processing of data (sampling frequency f _c ) by the load disaggregation unit 6 based on the detection of the same significant events.

In particular, the assessment of significant time intervals may be performed by the detection unit 8 on the aggregated consumption data and/or other electrical quantities associated with the power line 3, measured, in this case, by the metering unit 2.

This solution therefore reduces the amount of data to be processed in the NILM algorithm of the load disaggregation unit 6, which would instead be, in the case of uniform sampling, very high even at times when the system is in steady state operation.

Figures 3A-3B show a second embodiment of the present solution, wherein the detection unit 8 is again implemented locally to the metering unit 2, in the measuring and processing device 10, while the load disaggregation unit 6 is implemented remotely with respect to such measuring and processing device 10, in a distinct processing device 16.

Such processing device 16 is made in a respective electronic board 17, provided with a processor 17a (CPU - Central Processing Unit) , or similar digital computing unit, and a memory 17b, operatively coupled to the processor 17a and in which a suitable computing algorithm (the NILM algorithm) is stored, for the implementation of the functions of the load disaggregation unit 6.

In this case, each electronic board 13, 17 of the measuring and processing device 10 and of the distinct processing device 16 is further provided with a respective communication interface 18, being wired (Ethernet, serial or other) , wireless or internet-based (cloud) , for transmitting data through a communication channel 19, again of the wired, wireless or internet (cloud) type.

In this embodiment, in addition to the problems related to the computational load of the load disaggregation unit 6, those related to bandwidth occupancy in the communication between the measuring and processing device 10 and the processing device 16 (the latter dedicated to the NILM algorithm) are therefore added.

In the case of uniform sampling, a constant flow of data would be generated over time, resulting in wasted communication bandwidth during periods of stationary operation of the system; with the proposed solution, on the contrary, the transmission bandwidth occupancy on the communication channel 19 is optimised.

As in the embodiment discussed above with reference to

Figures 2A-2B, the assessment of significant time intervals can be implemented by the detection unit 8 both on the aggregated consumption data and on other electrical quantities measured by the metering unit 2.

By way of a non-limiting example, the reception and processing of the data by the load disaggregation unit 6 may only take place at the data-sampling instants (considering the sampling frequency f _c of the aggregated consumption signal S _c to be variable); i.e. data receipt and processing may take place more frequently at certain time intervals corresponding to the significant events detected by the detection unit 8.

Figures 4A-4B show a third embodiment of the present solution, wherein both the detection unit 8 and the load disaggregation unit 6 are implemented remotely with respect to the metering unit 2, being in fact implemented in the same distinct processing device 16.

The aforementioned processing device 16 is again provided with the communication interface 18, wired, wireless or internet-based, for transmitting data through the communication channel 19; in this case, the metering unit 2 is provided with a respective communication interface 18 for interfacing with the same communication channel 19.

The processing device 16 may in this case be provided with an acquisition module 22, configured to acquire one or more electrical quantities or parameters (e.g. an effective voltage signal V _rms , a current or a power) associated with the power line 3, which may be processed (possibly in combination with the aggregated consumption data) by the detection unit 8 for the implementation of the significant event detection algorithm.

The variation of the sampling frequency f _c is therefore made not only on the basis of the data acquired by the metering unit 2, but also on the basis of the processing of new data and signals acquired by the acquisition module 22 (the detection unit 8 is in this case a unit having its own autonomous capacity to acquire and process signals in order to discriminate the useful samples of the aggregated consumption signal S _c ) .

In this case, the detection unit 8 may have available an approximate (or processed) version of the aggregated consumption data (in a non-limiting case, e.g. relating only to the effective voltage V _rms ) , on the basis of which the analyses to detect significant variations can be performed.

Such variations, in a possible implementation, may be communicated to the metering unit 2, in order to request a greater detail of the analysed data (e.g. to cause an increase in the sampling frequency of the aggregated consumption signal S _c ) .

Figures 5A-5B show a fourth embodiment of the present solution, wherein both the detection unit 8 and the load disaggregation unit 6 are implemented remotely with respect to the metering unit 2, in such case being implemented in respective and distinct processing devices 16. The acquisition module 22 is in such case implemented by the processing device 16 in which the detection unit 8 is present .

Each processing device 16 is provided with a respective communication interface 18, so as to interface on the communication channel 19.

In particular, in this case there is a communication link between the detection unit 8 and the metering unit 2; and further a communication link between the same detection unit 8 and the load disaggregation unit 6.

Again, the detection unit 8 may have available an approximate (or processed) version of the aggregated consumption data, on the basis of which the analyses to detect significant changes are performed.

Such variations, in a possible implementation, may be communicated to the metering unit 2, in order to request a greater detail of the analysed data causing the increase in the sampling frequency of the aggregated consumption signal Sc

In the embodiment shown in Figures 6A-6B, wherein again both the detection unit 8 and the load disaggregation unit 6 are implemented remotely with respect to the metering unit 2, in respective distinct processing devices 16, communication links between the metering unit 2 and the detection unit 8 and also between the same metering unit 2 and the load disaggregation unit 6 are provided.

Also in this case, the detection unit 8 may thus have available an approximate (or processed) version of the aggregated consumption data, to be analysed for detection of the significant variations.

Such variations, in a possible implementation, may be communicated to the metering unit 2, in order to request the transmission to the load disaggregation unit 6 of a greater detail of the analysed data and the increase in the sampling frequency of the aggregated consumption signal S _c .

In the embodiment shown in Figures 7A-7B, again both the detection unit 8 and the load disaggregation unit 6 are implemented remotely with respect to the metering unit 2, in respective and distinct processing devices 16, and communication links between the load disaggregation unit 6 and the detection unit 8 and further between the same load disaggregation unit 6 and the metering unit 2 are provided.

Again, the detection unit 8 may have available an approximate (or processed) version of the aggregated consumption data, on the basis of which to perform analyses to detect significant changes.

The detection unit 8 may discriminate significant time intervals by analysing only the acquired quantities, in a non-limiting case for example the effective voltage V _rms . In this specific case, however, no comparison (or other type of joint processing) can be made with the data of the metering unit 2, as communication between such metering unit 2 and the processing unit 8 is not provided for in this embodiment.

The detection unit 8 is in any case configured to cause the variation of the data processing frequency by the load disaggregation unit 6, which receives the data from the metering unit 2.

It is noted that in all the discussed embodiments, the measurement frequency of the metering unit 2 does not change and only of the sampling frequency f _c of the data processed by the load disaggregation unit 6 changes.

Some examples of detection algorithms which may be implemented by the detection unit 8 to detect significant variations on the power line 3 (associated with variations in electrical loads 5) will now be described.

In a possible implementation, the detection algorithm may be self-adaptive or proportional, i.e. based on significant variations of one or more electrical signals X on the power line 3 (possibly of the same aggregated consumption signal S _c and/or additional electrical signals acquired by the metering unit 2 and/or the acquisition module 22, e.g. power, voltage, current signals, etc.) . In such case, the detection unit 8 may be configured to cause the variation of the sampling frequency f _c with which data are received and/or processed by the load disaggregation unit 6 according to the following expression:

/ _c (t) - / _c (t _o ) = K[C(ί _c ) - X(t ₀ )]

where ti and to are two subsequent time instants and K is a proportionality constant of a suitable value, fixed or proportional to the variation of the electrical signal X.

Based on the above expression, a proportional variation of the sampling frequency f _c corresponds to the variation of the electrical signal X. It is noted that the proportionality constant K may for example have a null value for signal variations below a certain threshold, so as not to have variations in the sampling frequency f _c from the initial value f _c (to) , assumed low, under stationary conditions of such electrical signal X.

In a different implementation, the detection unit 8 may implement an algorithmic approach by analysing the information content of the electrical signal X (possibly of the aggregated consumption signal S _c and/or other appropriate signal), in order to distinguish the times at which to increase or decrease the sampling frequency f _c .

The detection unit 8 may in this case implement appropriate rules to distinguish the amount of information content, for example to distinguish an electrical signal X with sudden, unpredictable time variations, with more information and to which a high sampling frequency f _c has to be associated (shown by way of example in Figure 8); from a substantially stationary or stable electrical signal X, with less information and therefore with which a reduced sampling frequency f _c has to be associated (shown by way of example in Figure 9) .

As a non-limiting case, algorithms based on spectral content analysis and detection of significant variations in the frequency domain can be considered; in this example, a periodic signal would not result in the transmission of new samples since the nature of the signal would not change.

By way of a further non-limiting case, another applicable method may envisage analysing user behaviour within the system, whether a domestic or industrial context, to discriminate repetitive trends which would allow sampling of only or mainly the significant data in the case of variations .

In any case, the aforesaid algorithmic approach envisages a higher level or higher abstraction processing compared to the analysis of the trace or trend of the raw signal only.

According to a further implementation of the present solution, as shown schematically in Figure 10, the detection unit 8 may cause the variation of the sampling frequency f at the request of the same load disaggregation unit 6.

The NILM algorithm implemented by the load disaggregation unit 6 may in fact request more data, i.e. a greater detail of the aggregated consumption signal S _c , in the case in which, a posteriori, a non-effective result is obtained (for example, with respect to the correct disaggregation of the electrical loads) .

A non-limiting example of this method envisages verifying that the disaggregation result is congruent with the nature of the household appliance (or in general the electrical load) to be observed. For example, if the duration of the profile measured, associated with a predetermined household appliance to be disaggregated, is different from that of the predetermined profiles, or if the profile itself undergoes time variations outside an expected range, it may be decided to request a greater detail so as to refine the disaggregation process.

In this case, the disaggregation unit 6 makes a request for additional data to the detection unit 8, which defines the type and quantity of additional detail requested of the metering unit 2.

The metering unit 2, which has a buffer memory available, indicated herein as 2a, with all the latest measurements stored (the measuring frequency is indeed high and generally much greater than the sampling frequency f _c ) , is controlled by the detection unit 8 to send the detail requested, i.e. of the data previously stored, and to increase the sampling frequency f _c of the aggregated consumption signal S _c . It should be emphasised again that the measurement frequency of the metering unit 2 is not modified according the present solution.

Figure 11 shows:

the aggregated consumption signal S _c generated on the basis of the signals acquired by the metering unit 2 (in the example, an active power signal); and

an effective voltage signal V _rms on the power line 3 (it should be noted that this is a non-limiting example of electric signal associated with the same power line 3) .

This effective voltage signal V _rms (or other significant electrical signal associated with the power line 3), as discussed above, can be acquired by the same metering unit 2, or by the acquisition module 22 of the processing unit 8.

As discussed above, the detection unit 8 can discriminate the significant time intervals by analysing a single quantity acquired, or by comparison (or other type of joint analysis) of several quantities, for example the aforesaid effective voltage signal V _rms (or other significant electrical signal associated with the power line 3) and the same aggregated consumption signal S _c .

In particular, the same Figure 11 shows, highlighted in the circles, two significant variation time intervals detected; in such time intervals, the sampling frequency f _c associated with the processing by the load disaggregation unit 6 is increased compared to the other time intervals where no significant variations are detected.

In the aforesaid Figure 11 the sampling instants of the aggregated consumption signal S _c are highlighted, the increase in sampling frequency f _c being particularly evident during the time intervals of significant variation detected by the detection unit 8.

The advantages of the present solution are evident from the above description.

In any case, it is once again underlined that the solution described allows a significant reduction of the computational load required of the load disaggregation unit 6 for the implementation of the NILM algorithm and moreover, in the case in which the load disaggregation unit 6 is situated remotely from the metering unit 2, a significant reduction of the bandwidth required for data communication.

The aforesaid reduction is particularly significant and effective in the case in which the system 1 is operating in a stationary or almost stationary situation.

The possibility, for the detection unit 8 , to process not only the aggregated consumption data, but also, alternatively or additionally, further electrical signals associated with the same power line 3, which can be acquired by the metering unit 2 and/or the detection unit 8 itself (in the case in which this is provided with its own acquisition module 22) is also particularly advantageous.

Lastly, it is clear that modifications and variants may be made to what is described and illustrated herein while remaining within the scope of the present invention, as defined by the appended claims.

In particular, it is noted that further and different algorithms may be implemented by the detection unit 8 to detect the significant changes in consumption on the power line 3, and that the same detection unit 8 may analyse different parameters and/or electrical quantities associated with the same power line 3.

In this regard, it is underlined that, if provided with the acquisition module 22, the detection unit 8 may possibly base the aforesaid detection of significant variations on the signals acquired by the same acquisition module 22, in an independent and unrelated manner with respect to the data acquired by the metering unit 2.

Moreover, the load disaggregation unit 6 may implement any NILM algorithm of a known type, to perform disaggregation of the electrical loads starting from the aggregated consumption signal S _c .