Device and method for detecting changes in the operating mode and identifying electrical appliances

Abstract

The subject of the application is a device for detecting changes in the operating mode and identifying electrical receivers and a method for detecting changes in the operating mode and identifying electrical receivers, in particular with arrangements for learning and detecting changes in these states. The device for detecting changes in the operating mode and identifying electric appliances EAs in the power network includes a voltage track, which includes a voltage measurement module (22). The input of the voltage measurement module (22) is connected in parallel to the power network and the output via the phase detection module (10) to the sample selection module (9). The device includes a current track that includes a current measurement module (32) whose input is connected in series to the power network and the output to the sample selection module (9). The output of the sample selection module (9) is attached to the matrix forming module (11), wherein the sample matrix of the matrix forming module (11) comprises a predetermined number of last K periods of voltage signal and m samples of instantaneous current value for each of the last K periods. The output of the matrix forming module (11) through the filtration module (12) and through the potential event detection module (13) is attached to the change vector determination module (15). The potential event detection module (13) is adapted to transmit to the input of the change vector determination module (15) the sample matrix from the matrix forming module (11) and additional information about the period number in which the potential change in the EA operating mode occurred. The change vector determination module (15) is adapted to determine the change vectors, the change vectors being transmitted to the input of the decision-making module (18) through the output of the change vector determination module (15). In addition, the input of the decision-making module (18) is connected to the output of the EA pattern memory module (17). The output of the decision-making module (18) provides information about the detection of a change in operating mode or the identification of EAs.

Description

[0001] Device and method for detecting changes in the operating mode and identifying electrical appliances

[0002] The subject of the present invention is a device for detecting changes in the operating mode and identifying electrical appliances and a method for detecting changes in the operating mode and identifying electrical appliances, in particular with systems for learning and detecting changes in these states.

[0003] Nonintrusive load monitoring (NILM), otherwise known as nonintrusive appliance load monitoring (NIALM) devices supply information on the energy consumption of individual electrical appliances (EAs) in a given low-voltage (LV) network circuit. An important feature of the NILM systems is that the measurements are made only at one point of such a circuit, most often near an energy meter or a fuse box. The purpose of the NILM devices and systems is to provide end users with detailed information concerning the energy consumption of all or part of the electrical appliances (EAs) from the area of this user's network (an apartment, house or multi-family building). The NILM system enables users to learn about their energy consumption habits and thus increase their awareness in this area, which, according to research (International Energy Agency (IEA) World Energy Outlook (WEO), 2022), leads to a reduction in energy consumption.

[0004] The NILM systems and devices are designed to monitor only the main circuit that supplies electricity and voltage to the user's area, e.g. a house, building or other location, and based on changes in the signals occurring there, determine which EA inside that location is switched on and what operating status the EA is currently in. The main advantage of the NILM systems is that it is not necessary to install many measuring devices at each of the EAs (as it happens in the 1LM, intrusive load monitoring, class systems), which should ultimately lead to a reduction in the cost of manufacturing the energy monitoring system and at the same time significantly facilitate its installation in existing locations. As a result of the NILM system operation, the end user may receive a report which details the individual energy consumption for at least part of the EA in this user’s area.

[0005] Each NILM device features an acquisition block, which records current and voltage signal samples, and a sample processing block, in which the used algorithms determine the distribution of the circuit load on individual EA. NILM devices use data processing methods to extract the characteristic features of signals, based on which artificial intelligence algorithms analyse the current operating states of individual EA. The result of this analysis is information about which EAs are currently enabled and, if applicable, in which mode of operation they are currently in. This information makes it possible to estimate the electricity consumption of individual EA. A single EA is recognized based on the so-called signature (pattern), which can be, for example, a vector of numerical data, enabling the recognition of EA and their operating states. The devices developed so far using NILM methods do not allow for flawless identification of all EA used in the user area, such as a household, which was the motivation to develop own, effective and cheap methods for use in the NILM device.

[0006] One of the methods of overcoming the imperfections of the NILM methods known from the state of the art is the development of mathematical models (e.g. in the form of equations of states of the dynamic system), reflecting the physical properties of some EA, such as in the solution disclosed and published in the patent document EP2686937B1 and in other publications, e.g. [Wittmann, F.M.; Lopez, J.C.; Rider, M.J. Nonintrusive Load Monitoring Algorithm Using Mixed-Integer Linear Programming. IEEE Trans. Consum. Electron. 2018, 64, 180-187, doi: 10.1109 / TCE.2018.2843292], The methods presented in these disclosures have not yet been implemented, mainly because the models contained therein are too simplified and do not give satisfactory results in the identification of EA (especially under the conditions of their actual use). It should be noted here that creating a good and sufficient universal model for EA is an open and difficult task due to the complicated physical nature of most EA. On the other hand, the use of mathematical models, which are already complex, requires complicated computational methods and high computing power to handle these models when identifying EA. Creating an accurate NILM device using the EA mathematical model is difficult to implement, due to the high production costs.

[0007] Due to the aforementioned shortcomings of existing solutions for detecting potential events, it would be beneficial to find a practical solution to the problem of effective detection of potential EA changes that does not require a lot of hardware resources from the NILM device.

[0008] State-of-the-art EA identification systems and methods, using e.g. Hidden Markov Models or EMI frequency analysis, or the aforementioned mathematical models, are difficult to transfer to small and cheap devices that could be installed outside a specialized laboratory in real conditions (e.g. in a fuse box in an apartment). In order to improve the low efficiency of NILM systems, additional sensors are used in solutions known in the art, not related to electrical signals coming from the LV power supply network, such as in the invention disclosed and described under the following numbers: US11036189B2 and in scientific articles, e.g.: [Zhao, T.; Zhang, C.; Ujeed, T.; Ma, L. Online Methodology for Separating the Power Consumption of Lighting Sockets and Air-Conditioning in Public Buildings Based on an Outdoor Temperature Partition Model and Historical Energy Consumption Data. Appl. Sci. 2021, 11, 1-23, doi: 10.3390 / appl 1031031], Usually, these are sensors giving environmental information, e.g. about temperature, lighting, sound, vibrations, etc., but also data from other media, e.g. from a LAN network - an example may be patent publications No. US10175276B2 or US9699529B1. Information from such sensors is to support the decision-making systems of NILM devices in more accurately identifying changes in EA operating states. However, the use of many additional sensors is contrary to the basic advantage of NILM systems, which assumes the use of as few sensors as possible, unlike ILM systems. This problem, together with the dubious effectiveness of NILM systems supported by information from environmental sensors, means that such solutions are also not applicable in practice. The noticed problem of solutions known from the prior art in the field of NILM methods is that many of the disclosed EA identification methods have been developed only with the use of test sets, which often do not reflect the conditions of actual use of EA, e.g. at home, this results in two significant disadvantages of these solutions. The first of these is the aforementioned lack of immunity to interference existing in real LV power networks, coming from outside the monitored user area (apartment, house or multi-family building). The second disadvantage is insufficient accuracy in identifying low-power EA when these EA operate in the presence of other EA with simultaneously higher and variable power.

[0009] From the point of view of practical implementation of NILM devices, insufficient accuracy in detecting potential events often leads to numerous false reports of potential events related to the monitored EA, which must then be identified at a later stage of the device's operation. This phenomenon causes a manifold increase in the consumption of NILM device resources, which significantly increases the cost of practical implementation of this device. On the other hand, insufficient sensitivity in detecting potential events usually leads to these events being omitted at the identification stage, and thus to the NILM device overlooking certain changes in the monitored EA states. Most of the proposed practical solutions utilise the detection of changes in averaged power or current values, e.g. RMS, or changes in phase angles. The authors of this application noted that this approach is characterised by low effectiveness in detecting events originating from low-power EA on the one hand, and on the other hand, high susceptibility to interference penetrating the monitored power network circuit from external circuits (this phenomenon is particularly significant in networks with a configuration similar to those used in Europe). Additionally, the authors of this application noted that in NILM systems and devices with potential event detection using changes in average power or current values, there is often a phenomenon of masking events originating from EA with lower power by multi-state EA with higher power and variable power consumption.

[0010] There are a number of disclosed solutions of NILM methods, devices and systems identifying events in the LV power supply network associated with EA based on EA patterns consisting of a predetermined number of signal parameters associated with the power consumed. Devices of this type acquire power or energy signals from meters and sensors measuring the basic harmonic of the power network voltage and in the case of methods based on higher harmonic parameters, the devices use additional measurement modules measuring current or voltage signals usually up to the frequency of the seventh harmonic of the LV power network voltage signal. The parameters of signals that are used in these NILM methods for the identification of EA are usually related to the basic harmonic of the voltage or current in the LV network, such as in the publication of the international application No. WO2021176459, and in the methods described in scientific articles, e.g.: [Liu, B.; Luan, W.; Yu, Y. Dynamic Time Warping Based Non-Intrusive Load Transient Identification. Appl. Energy 2017, 195, 634-645, doi: 10. 1016 / j.apenergy.2017.03.010] or parameters related to the harmonics of these signals, as in the inventions published under the numbers: WO2016079229A1, and in the methods described in scientific articles, e.g.: [Agyeman, K.A.; Han, S.; Han, S. Real-Time Recognition Non-Intrusive Electrical Appliance Monitoring Algorithm for a Residential Building Energy Management System. Energies 2015, 8, 9029-9048, doi: 10.3390 / en8099029]„

[0011] In order to improve EA identification, many early NILM systems and methods used Hidden Markov Models (HMM), as in the invention disclosed and published under the following number: EP2499462B1 and in scientific articles, e.g.: [Kolter, J.Z.; Jaakkola, T. Approximate Inference in Additive Factorial HMMs with Application to Energy Disaggregation. J. Mach. Learn. Res. 2012, 22, 1472-1482], However, these models are computationally complicated both in the testing phase of the NILM system and in the EA identification phase. Moreover, such methods and systems typically do not yield levels of EA identification effectiveness high enough to justify high computational complexify. Additionally, models using HMM have a practical problem involving a significant delay in identification relative to the event associated with a change in EA status. Unfortunately, despite many years of research on a group of NILM methods using signals related to the first harmonic of the LV network voltage signal or its harmonics (low frequency LF methods), they have not been effectively implemented mainly due to the imperfection of identification under real conditions in which typical EA are used. Methods from the LF group are characterised by low resistance to interference from outside the monitored LV power supply circuit, low effectiveness in identifying EA operating states with low rated power, and low effectiveness in identifying EA operating states with variable power.

[0012] Most EA identification methods require prior detection of potential events related to these EA, understood as switching on, switching off or changing the operating status of these EA. The development of such methods of EA identification towards commercialization has been inhibited by the level of unwanted noise occurring in the electricity networks of modem buildings. In such conditions, attempts to detect potential events (e.g. changes in the state of devices) turn out to be much more difficult than in laboratory conditions, which significantly hinders the accuracy of such detection. Despite numerous publications on the algorithms used to identify energy consumption by individual EA, there is still little research showing the results of testing these algorithms in the real working environment of the NILM systems, i.e. in actual residential buildings. It should be noted that the aspect of potential events detection in the NILM methods is not exposed in scientific articles, as these usually focus only on the aspect of EA identification. The authors of scientific publications describe research on EA identification algorithms often carried out only with the use of publicly available data sets, e.g. those described in scientific articles.

[0013] Such “tagged” data sets, commonly used for the development and testing of EA identification methods, contain data collected from the monitored circuits, which already contain information from measurement systems about the moments of EA state changes. Although, in scientific research on EA identification algorithms, such collections are very convenient to use, such an approach is not sufficient in practice, and an effective event detection method is necessary for the practical application of EA identification methods in a real NILM system. A partial solution to the identified problems of identification effectiveness are the NILM methods that identify EA operating states based on changes in the frequency spectrum of voltage or current signals in the high frequency band (HF methods). Solutions disclosed e.g. in the publication of the US application No. US20130179124 and methods discussed in scientific articles are known, e.g. in: [Gupta, S.; Reynolds, M.S.; Patel, S.N. ElectriSense: Single-Point Sensing Using EMI for Electrical Event Detection and Classification in the Home. In Proceedings of the Proceedings of the 12th ACM international conference on Ubiquitous computing; ACM: New York, NY, USA, September 26, 2010; pp. 139-148], The methods discussed in these disclosures address some of the EA identification issues previously reported, but these methods result in an unacceptable price increase for the consumer market for equipment and systems that utilize such methods. This is due to the significant demand for computing power, the need to determine the signal transformers, and the increased requirements for the analog-to-digital converters used (high sampling frequency, high bit resolution), as well as often the significant requirements for telecommunications links connecting the elements of the NILM system.

[0014] The disadvantage of most of the disclosed NILM methods using changes in the high frequency range spectrum is their susceptibility to interference coming from outside the monitored LV power supply network circuit, as they use voltage measurements in the monitored circuit. This property is less visible in LV power supply networks, e.g. in the USA, due to a different distribution network architecture. In European networks, due to their architecture, in which areas of multiple users are powered within a single distribution network circuit, interference coming from outside the monitored power network circuit (user area) causes a significant reduction in the effectiveness of the NILM devices using changes in the voltage signal. Better properties are provided by methods using the spectrum of the current signal in steady-state EA operation and in non-steady-states, as well as additionally using an artificially introduced known pulse signal to the monitored circuit, such as in the solutions disclosed and published under the following numbers: US20140207398A1, US9104189B2 and discussed in the scientific article [Wojcik, A.; Bilski, P.; Lukaszewski, R.; Dowalla, K.; Kowalik, R. Identification of the State of Electrical Appliances with the Use of a Pulse Signal Generator. Energies 2021, 14, 673, doi: 10.3390 / enl4030673]. However, these methods are difficult and expensive to implement, as they require, for example, the installation of a pulse generator and high-end analogue and analogue -digital converters.

[0015] Both in LF methods using the parameters of harmonic signals of currents and voltages of the LV power supply network and in HF methods in the high frequency band, the EA often uses the FFT transform to calculate the parameters of the patterns, such as in the inventions disclosed and published under the following numbers: US20140200725A1 or in the methods described in scientific articles, e.g.: [Gupta, S.; Reynolds, M.S.; Patel, S.N. ElectriSense: Single-Point Sensing Using EMI for Electrical Event Detection and Classification in the Home. In Proceedings of the Proceedings of the 12th ACM international conference on Ubiquitous computing; ACM: New York, NY, USA, September 26, 2010; pp. 139-148], In other known solutions, the authors use the wavelet transform instead of the FFT transformation, such as disclosed in the publication of the following number: US8983670B2, and in a scientific article, e.g.: [ Su, Y.C.; Lian, K.L.; Chang, H.H. Feature Selection of Non-Intrusive Load Monitoring System Using STFT and Wavelet Transform. Proc. - 2011 8th IEA Int. Conf. E-bus. Eng. ICEBE 2011 2011, 293-298, doi: 10. 1109 / ICEBE.2011.49], The first method - the FFT method - completely blurs the information about time, when there are significant signals related to the change of the EA operating mode, relative to the basic component of the signal. Using appropriate wavelet transforms is a better solution. However, even in this case, the necessary information is not preserved in its entirety.

[0016] The NILM methods, using EA signatures, encountered another limitation on the way of commercialization, important for potential users of the NILM devices. All such methods require either long-term learning of the classifier used in them, e.g. in the form of a neural network, or a cumbersome learning process of the NILM device in interaction with its user. This problem results from the properties of the algorithms used, which require a number of examples to be given in the process of learning the classification of EA state change signatures. One of the disclosed approaches to training the NILM algorithms is the multiple manual activation and deactivation of EA by a user when the NILM device is in training mode, for example the solution disclosed in U.S. Patent Application Publication No. 20210158186A1. Solutions known from the prior art reveal methods of learning, i.e. determining patterns, e.g. in the form of EA signatures vectors. In these solutions, the signature values are specific parameters calculated on the basis of measured current or voltage waveforms or directly measured parameters of these signals such as: e.g. active power, reactive power, phase shift angle between the current and voltage signals, peak, average or effective values of these signals, etc. In the previously described aspect related to the effectiveness of EA identification methods, it was noted that the use of such signatures is not sufficient for effective identification of EA states. The known inventions do not describe methods of teaching the NILM devices that use vectors that are current signal samples or differences of such vectors as EA signature patterns, and such patterns are used in the identification method of the present invention. Other disclosed inventions in which different EA models are created describe a method for training these models. The EA models used require long-term training using a lot of data from various sources. Any suitable techniques, as described e.g. in US9443195, may be used to train the EA model. Training techniques may include, but are not limited to, neural network training, self-organizing maps, SVM (Support Vector Machines), decision trees, random forests, logistic regression, Bayesian models, linear and nonlinear regression, and Gaussian distribution mixture models; as e.g. described in the inventions: WO2011128883A2.

[0017] It should be noted that in the disclosed prior art solutions, the aspect of the NILM device training stage is often omitted, as e.g. in the publication of international application number WO2021176459A1. Similarly to the issue of detecting potential events, authors of other publications often do not address the problem of training the NILM system (understood as determining actual EA patterns), focusing instead on the scientific problem of EA identification, e.g. in: [S. Bao et al., "Feature Selection Method for Nonintrusive Load Monitoring With Balanced Redundancy and Relevancy," in IEA Transactions on Industry Applications, vol. 58, no. 1, pp. 163-172, Jan. -Feb. 2022, doi: 10.1109 / TIA.2021.3128469], In other studies, the authors of scientific articles determine EA patterns based on existing, described in detail (“tagged”) data sets, e.g. in: [Joao Gois, Lucas Pereira, A novel methodology for identifying appliance usage patterns in buildings based on auto-correlation and probability distribution analysis, Energy and Buildings, Volume 256, 2022, 111618, ISSN 0378-7788, https: / / doi.Org / 10.1016 / j.enbuild.2021.111618], This approach has no practical application in real implementations of the NILM devices and systems, which the user expects to add to the system patterns for a new EA, added to the monitored system after the installation of the NILM device in the operating environment of this device or the NILM system. ft would be advantageous to provide a practical solution to the problem of the NILM device learning, consisting in one interaction with the user for a given EA in the monitored electrical circuit for the entire period of this NILM device usage. It would be advantageous to adapt the developed effective learning method to the NILM device, which uses vectors as EA signature patterns, being current signal samples or differences of such vectors.

[0018] Many of the NILM methods and systems disclosed, e.g. using Markov models (HMM), such as in the published patent document number: EP2499462B1 or described in scientific articles, e.g.: and in scientific articles, e.g.: [Kolter, J.Z.; Jaakkola, T. Approximate Inference in Additive Factorial HMMs with Application to Energy Disaggregation. J. Mach. Learn. Res. 2012, 22, 1472-1482] or state models, as in the inventions disclosed and published under the following numbers: US9739813B2 and in scientific articles, e.g.: [Liu, Y.; Liu, C.; Shen, Y.; Zhao, X.; Gao, S.; Huang, X. Non-Intrusive Energy Estimation Using Random Forest Based Multi-Label Classification and Integer Linear Programming. Energy Reports 2021, 7, 283-291, doi: 10. 1016 / j.egyr.2021.08.045] cannot report real-time results to end users. This is because such approaches require long observation periods of full EA duty cycles in order to identify their states in the NILM device. Due to that fact, the results showing which EA were switched on and off and in which operating states they were, can only be reported after many hours. In other inventions, data from measurements of currents or voltages are often sent from time to time, e.g. every 30 seconds, as in the invention published under the following number: WO2021176459A1, e.g. to a central server or cloud services to identify EA based on the transmitted data. Only after sending this data, the system decides which of the EAs are in the currently defined operating states and then, with quite a long delay, reports the results to the user. Solutions with this disadvantage are not suitable for widespread implementation in practice. Therefore, solutions that operate in near real-time mode and show changes in EA operating states with the least possible delay will be preferred. The authors of this invention observed that only appropriate processing of the current signal in the time domain allows all relevant information contained in this signal, characteristic for the identification of EA operating states, to be preserved.

[0019] This application is a continuation of a partially Polish invention registered under patent number Pat. 243560 of 24.04.2020. The subject of this invention was a device for detecting changes in the operating mode and identifying electrical appliances and a method for detecting changes in the operating mode and identifying electrical appliances.

[0020] The device for detecting changes in the operating mode and identifying EAs in the power network, according to Pat. 243560, comprises a voltage track, which includes a low -pass filter, the input of which is connected in parallel to the power network and the output through a voltage measurement module and an analogue-to-digital converter module and through a phase detection module to a processing module, and a current track, which includes a current measurement module, the input of which is connected in series to the power network, and the output through an analogue-to-digital converter module and a buffer module to a processing module, at the same time, the output of the processing module through a cyclic buffer and a filtration module is connected to the input of a change vector determination module, the output of which is connected to the input of a discriminator module, in addition, the output of an EA pattern memory module is connected to the discriminator module.

[0021] Patent 243560 also discloses a method for detecting changes in the operating mode and identifying EA electrical appliances using measurements of instantaneous values of current signals and voltage phase detection, which includes: a) the step of measuring the current in the power network and the step of measuring the voltage, characterised in that b) after the current measurement step, the current signal is converted into its digital form, which is transmitted to the buffer module, and c) after the voltage measurement step, the voltage signal is converted into its digital form on the basis of which, using the phase detection module, the detection of a specific value of the voltage signal phase is carried out, and then d) by means of the processing module, appropriate sample values are selected from the buffer module, which are further formed into a matrix in which subsequent columns contain the values of current samples in subsequent voltage periods starting from the moment of reaching a specific phase value by the voltage signal, and then e) using the filtration module, the matrix is filtered by its rows to obtain a filtration sample matrix, based on which the vectors of changes are determined in the change vector determination module, subtracting the values of the sample matrix samples after filtration from the period before the change of the EA state from the values of the sample matrix samples after filtration from the period of the recorded change of state, when the EA that changed the state and will already be in the new steady state, and then f) the determined change vectors using the discriminator module are compared with the reference feature vectors of all EAs in a given electrical circuit placed in the pattern memory module, identifying the EA based on the most suitable reference pattern.

[0022] The present invention solves the problem of event detection for identification methods using the temporal form of a signal carrying electrical power information, preferably for a current signal. Solving this problem removes the main limitation of many known event-driven identification methods, which is the strong dependence of the entire system performance on the detection of an event. In such solutions, if a single event is omitted, the error tends to propagate in the system. The present invention makes it possible to increase the reliability of event detection in practical embodiments of the present invention, comprising a module adapted for effective event detection, with continuous analysis of current signal samples. It should be noted that the applied event detection method is described by preferably simple operations comprising subtracting and comparing values of a numerical vector. Thus, the computational complexity of event detection used in the present invention is preferably very low.

[0023] The present invention solves the problem of learning for identification methods using the temporal form of a signal carrying electrical power information, preferably for a current signal. According to the described invention, in the identification method used in it, there is no need to develop an EA model, therefore there is no need to train it. Preferably, the learning of the system consists in a single (but not necessarily limited to a single instance) monitoring and recording two numerical vectors of the values of the current signal samples measured before and after the change in the state of the monitored EA, and calculating and recording the differences in the values of these vectors as the pattern of this EA.

[0024] The apparatus for detecting changes in the operating mode or identifying EAs in the power network, according to the invention, comprises: a voltage track that includes a voltage measurement module whose input is connected in parallel to the power network and output through a phase detection module to the sample selection module; and a current track that includes a current measurement module, the input of which is connected in series to the power network, and the output to the sample selection module; the output of the sample selection module is attached to a matrix forming module, the sample matrix of the matrix forming module comprising a predetermined number of the last K periods of the voltage signal and m samples of the instantaneous current value for each of the last K periods; at the same time, the output of the matrix forming module through the filtration module is connected to the input of the change vector determination module, the change vector determination module being adapted to determine the change vectors, the change vectors being transmitted through the output of the change vector determination module to the input of the decision-making module, in addition, the output of the EA pattern memory module is attached to the input of the decision-making module; wherein the output of the decision module provides information about the detection of changes in operating mode or the identification of EAs.

[0025] The essence of the invention is that the device for detecting changes in the operating mode or identifying EAs in the power network is characterized in that the filtration module is attached to the change vector determination module through the potential event detection module, wherein the potential event detection module is adapted to transmit to the input of the change vector determination module of the sample array from the matrix forming module and additional information about the number of the period in which the potential change in the EA operating mode occurred.

[0026] It is preferable that the decision-making module includes a similarity determination module and a decision module, wherein data from the change vector determination module and data from the pattern memory module are fed to the input of the similarity determination module, the output of the similarity determination module is connected to the first input of the decision module, while the second input of the decision module is connected to the output of the pattern memory module.

[0027] Preferably, the similarity determination module in the decision-making module is adapted to compare the change vector values with the pattern definition vectors that are stored in the EA pattern memory module.

[0028] It is preferable that the decision module in the decision-making module is adapted to identify the EA and / or identify a change in the state of the EA based on data from the similarity determination module and the pattern memory module. It is preferable that a mode selection module is arranged between the potential event detection module and the change vector determination module, wherein one of the outputs of the operating mode selection module is attached to the decision module via the pattern vector determination module and the pattern memory module, and the other output of the operating mode selection module is attached to the decision module via the change vector determination module and via the similarity determination module, wherein when the operating mode of registration of the new EA pattern is active, the operating mode selection module is adapted to transmit information from the input to the pattern determination module, when the operating mode of registration of the new EA pattern is not active, the operating mode selection module is adapted to transmit data only to the change vector determination module. It is preferable that the phase detection module is adapted to determine the time at which the value of the voltage signal of the electrical network reaches a predefined value of the phase, preferably when the value of the voltage signal of the electrical network changes from a negative value to a positive value.

[0029] It is preferable that the sample selection module is adapted to transmit the values of the current samples to the matrix forming module, starting from the sample number determined by the phase detection module for each successive period of the voltage signal, wherein the number of samples transmitted results from the duration of one period of the voltage signal and the sampling frequency in the current measurement module.

[0030] It is preferable that the matrix forming module is adapted to store the values of the current samples for a certain number K of the last periods of the voltage signal that form the sample matrix that is fed to the input of the filtration module.

[0031] It is preferable that the filtration module is adapted to perform median filtration or averaging filtration or mode filtration.

[0032] It is preferable that the filtration module is adapted to perform filtration for a series of samples taken at equally distant moments in time from the phase determined by the phase detection module. Preferably, the voltage measurement module includes a voltage transformer to the secondary winding of which the low-pass filter is connected, and the filter output through the amplifier module is connected to the analogue -to-digital converter module, whose output is the output of the voltage measurement module, or the voltage measurement module includes a voltage transformer to which the amplifier module is connected to the secondary winding, which is connected to the analogue-to-digital converter module and further to the digital low -pass filter module, and the digital low -pass filter output is the output of the voltage measurement module.

[0033] Preferably, the current measurement module comprises a current transformer to the secondary winding of which an amplifier module is coupled that is connected to the analogue-to-digital converter module, preferably an anti-aliasing filter is placed between the secondary winding of the current transformer and the amplifier module, wherein the throughput band of the anti-aliasing filter is related to the sampling frequency of the converter in the analogue-to-digital converter module, or the current measurement module comprises a current transformer to the secondary winding of which the amplifier module is coupled, which is connected to the analogue-to-digital converter module, preferably the analogue-to-digital converter module is connected to the digital anti-aliasing filter module, wherein the output of the digital anti-aliasing filter or the analogue-to-digital converter is the output of the current measurement module.

[0034] According to the invention, the method for detecting a change in the operating mode and identifying electrical appliances of EA using measurements of instantaneous values of current signals and voltage phase detection includes: a current measurement step in a power supply network and a voltage measurement step, wherein, after the current measurement step, the current signal is converted into its digital form, which is transmitted to the sample selection module, and after the voltage measurement step, the voltage signal is converted into its digital form on the basis of which, using the phase detection module, the detection of a specific value of the voltage signal phase is carried out, and then at the stage of matrix formation with the help of the matrix formation module, appropriate sample values are selected from the sample selection module, which is further formed into a matrix in which subsequent m rows contain samples of instantaneous current values in k columns of subsequent voltage periods starting from the moment of reaching a specific phase value by the voltage signal, and then in the filtration stage, using the filtration module, the matrix is filtered by its rows to obtain a sample matrix after filtration, from which the vectors of changes are determined in the change vector determination module, covering the differences in the values of samples of the sample matrix after filtration from the period before the change in the state of the EA from the values of samples of the sample matrix after filtration from the period recorded after the analysed state change, when the EA has changed the state and will be in the new steady state, and an identification step in which the designated change vectors are compared by means of a decision module with the reference feature vectors of the EAs in a given electrical circuit placed in the pattern memory module, identifying the EA based on the most matching pattern

[0035] The essence of the invention is that the method of detecting changes in the operating mode and identifying electrical appliances of the EA using measurements of current signals instantaneous values and voltage phase detection is characterized in that after filtering the sample matrix in the filtration module, there is a stage of detecting potential events, wherein the detection consists in determining in the potential event detection module vectors of changes for at least some of the K voltage periods, which vectors allow to determine a proportionality coefficient proportional to the mean of the absolute values of the elements of these vectors, wherein the period k, in which the determined coefficient locally reaches the maximum value, is considered to be the period in which the potential change in the mode or state of the EA occurred, then information about the number of the period in which the potential change in the mode or state of the EA occurred is transmitted to the module for determining the vectors of changes. The change vectors determined in the change vector determination module include the differences in the values of the sample matrix samples after filtration for the period prior to the period in which a potential change in EA mode or status occurred, from the values of the filtered sample matrix for the period after the period in which the potential change in EA mode or status occurred, taking into account only those periods for which the proportionality coefficient reached its maximum value. Only the selected change vectors determined in the change vector determination module are further compared in the decision module and used to identify EA based on the most fitting pattern. ft is preferable that the determination of the proportionality coefficient in the potential event detection module includes the determination of change vectors comprising differences in the values of the sample matrix samples after filtration from the period k - o ’ before the change in the state of the EA from the values of the sample matrix samples after filtration from the period k + o ’ recorded after the analysed period k, whereby only selected periods before and after are analysed for a predefined at least one pair of offset values o ’ and o ”, preferably the offset o ’ and o ” corresponds to the EA category and the offset values o ’ and o ” may not be equal, the proportionality coefficient is determined on the basis of the determined difference of the values of the matrix samples for each predefined at least one pair of offsets o ’ and o ”, preferably the proportionality coefficient is the average of the absolute value of the determined difference of the sample values. ft is preferable that, by means of the phase detection module, a time moment is determined in which the voltage value of the electrical network changes from a negative to a positive value. ft is preferable that, by means of the matrix-forming module, the values of the current samples are transmitted from the sample selection module in the form of a sample matrix, wherein for each period k starting from the sample determined by the voltage phase detection module, a number m of samples corresponding to each of the voltage periods is transmitted, preferably the matrix-forming module comprises a matrix buffer at K of the last voltage periods and a sample matrix is transmitted to the output every each voltage period, which sample matrix is fed to the input of the filtration module. ft is preferable that filtration is carried out for a series of samples taken at moments equidistant in time from the phase determined by the phase detection module. It is preferable that all steps are carried out on signal samples in the time domain.

[0036] It is preferable that the sampling moments of the current signal are correlated in time with the moment of reaching a specific phase by the voltage signal.

[0037] It is preferable that the filtration is median filtration or averaging filtration or mode filtration.

[0038] It is preferable when the method comprises:

[0039] - determining similarities in the similarities determination module between all the reference change vectors taken from the pattern memory module, wherein the pattern memory module contains the reference change vectors characteristic of EA, and the set of change vectors determined in the determination module of change vectors, wherein the determined similarities are transmitted to the decision module;

[0040] - making decisions by the decision module based on the similarities determined in the similarity determination module and based on the data from the pattern memory module with a threshold value characteristic of each type of EA and a possible change in the state of this EA, the pattern of which is stored in the pattern memory module.

[0041] It is preferable that the decision making by the decision module includes: determining the maximum value of the similarity coefficient for the entire similarity vector, and comparing this maximum value with the threshold value taken from the pattern memory module, then making decisions by the decision module based on this comparison, if the maximum value of the similarity coefficient for the similarity vector exceeds the threshold value, then the decision module decides that the potential event in the period k concerns a change in the state of the specified EA, while if the maximum value of the similarity coefficient for the similarity vector does not exceed the threshold value, then the decision module decides that the previously detected potential event is not an event indicating a change in the state of any EA in the tested power supply circuit of the power network.

[0042] It is preferable that at the decision-making stage in the decision module, in a situation where the given decision selection criteria are insufficient, statistical algorithms are used to select the most likely EA from among the recorded patterns, preferably it is an algorithm of the decision tree or random forest, or K-nearest neighbours (KNN), or convolutional neural networks (CNN), or support vector machines (SVM), or deep neural networks (DNN), or Naive Bayes.

[0043] It is preferable that before the decision-making stage in the decision module (19), a step of adding patterns of changes to the pattern memory module (17) is performed, wherein the step of adding the patterns depends on the predefined and provided by the user data on the category of the device being added and the moment of its connection to the network, the steps of adding the patterns include: setting the operating mode selection module (14) to the pattern addition mode (output duplicate mode), which supplies the sample matrix to the pattern determination module (16), determining the reference change vectors, wherein the determination of the reference change vectors takes place similarly to the determination of the reference change vectors, wherein in the determination of the reference change vectors, predefined values of offsets o ’ and o ” are adopted, and a predefined threshold value characteristic for the given category of EA, then memorizing a reference change vector, along with corresponding o ’ and o ’ ’ offset values, and a predefined threshold characteristic of a given EA category in a pattern memory module (17), wherein after the addition of the reference change vector to the pattern memory module (16), the EA recognition, as defined in the method according to the invention, is started or a further device is added.

[0044] The present disclosure constitutes an improvement of an earlier invention (Pat. 234560), at least by changes in signal processing at the identification stage, but also by supplementing with new effective solutions concerning the stage of detection of potential events related to the monitored EA and by supplementing with solutions concerning the training mode of the device.

[0045] The present invention employs an identification method using, preferably only the temporal form of signals, preferably of medium frequency. This method is described in detail in the further part of this study. A characteristic feature of this identification method is that it preferably considers moments in the periods of the basic component, preferably according to the moment of the occurrence of a specific voltage level (phase of this voltage), in order to obtain information, preferably from a pseudo-periodic signal characteristic of the monitored EA. In addition, a characteristic feature of the identification method used in the present invention is a special method of filtration, preferably median filtration, of this pseudo - periodic signal. The filtration method used, further described in the detailed description of the invention, is different from the denoising filtration methods used in other inventions, which, although facilitating the detection of events, significantly reduce the effectiveness of EA identification, as they cause the loss of significant information contained in the current signals of the LV power network, such as in the inventions disclosed and published under the following numbers: US9104189B2, CN106022645B (WO2017211288A1, US20170351288A1).

[0046] Note that the EA identification method used in the present invention is described by simple operations comprising subtracting and comparing values of a numerical vector. Thus, the computational complexity of event detection used in the present invention is preferably very low. In addition, by using a simple measure of similarity to the EA classification (where scaling requires only the addition of new numerical vectors as patterns of new EA), the EA classification can be easily scaled to include new EA in the monitored LV supply network circuit. On the other hand, the classification methods used in other inventions, such as: LDA, SVM, and neural networks often require reclassification of the classifier or addition (e.g., manually) of new classifiers to the system. The proposed solution of the invention is thus scalable to larger environments with more EA or with EA having multiple modes of operation.

[0047] The approach used to identify EA is characterized by lower computational complexify and greater scalability compared to other inventions based on, for example, mathematical models, HMM or high- frequency analysis (EMI).

[0048] A beneficial effect of using the developed identification method in the device according to the invention is that a very good EA identification efficiency was obtained, using cheap means in the practical implementation of the NILM device.

[0049] In contrast to inventions employing identification methods using, e.g., HMM, or state models, the invention disclosed herein is preferably capable of producing EA classification results in near real-time, with a delay of about 1 second or less, when the actual EA is turned on.

[0050] In the following, an apparatus and method that implements both a measurement data processing method and an event detection method as well as a system learning method are described in detail.

[0051] The object of the present invention is presented in the embodiment in the drawing, in which fig. 1 shows a block diagram of the device according to embodiment of the invention.

[0052] During stable operation (i.e. in the steady state) of an electrical appliance (EA) or many such appliances, it was observed that the recorded instantaneous current values are similar in repetitive moments (tm) in relation to the time of reaching a specific phase by the LV mains base component signal (tok, where k are the numbers of successive periods of the mains voltage base component signal). When an additional appliance is switched on, an already switched-on appliance is switched off, or its state is changed, the instantaneous current values change in a manner characteristic of the type of appliance or its state. The concept of the device according to the present invention consists in identifying the operating state of the electric appliance on the basis of continuous measurements of the total current and voltage of the LV mains supply and in processing the results of these measurements. The essence of data processing used to identify working appliances consists in determining the difference in the instantaneous values of the current signal in properly selected periods of the LV mains voltage signal measured at moments tm.

[0053] The device according to an embodiment of the invention includes a voltage sensor module in the voltage path 22, which, for example, is a voltage transformer 2, the input of which is connected in parallel to the power network, and which realizes the transformation of the voltage signal of the LV network (low- voltage network) to levels acceptable by typical A / A and A / D processing modules. In an exemplary device, a voltage transformer is a transformer that lowers the USUppiy voltage level. At the same time, the voltage transformer TN implements galvanic separation of the measurement systems from the LV network. As a result, the voltage transformer module 2 converts the supply voltage signal USUppiy to the voltage signal ui(t).

[0054] The output of the voltage transformer 2 is connected via the filter module 4 and the amplifier module 6 and the analogue-to-digital converter module 8 to the phase detection module 10.

[0055] The filter module 4 converts the voltage signal ui(t), received from the voltage transformer 2, into a voltage signal ui ’ (t) by removing components with frequencies higher than the base voltage of the power network (in Europe 50 Hz) from the voltage signal ui(t). The low-pass filtration used in the filter module 4 is intended to facilitate the determination of the moment to. For example, a low-pass filter with a buffer band above 70 Hz was used for this purpose.

[0056] It is possible to implement the filter module 4 in a digital way. In an aspect of the embodiment, this module follows the analogue-to-digital processing module 8 in the voltage path of the inventive apparatus, and the amplifier module 6 is connected directly to the voltage transformer 2. Changing the analogue form to the digital form of the filter module 4 does not change the essence of operation of the device according to the invention.

[0057] The amplifier module 6 converts the voltage signal ui’(t) to the voltage signal ui”(t) in such a way that the signal ui”(t) meets the requirements of the analogue-to-digital processing module 8 in the voltage path of the device according to the embodiment. These requirements apply to the selected range of input voltages for the A / D converter used in this signal processing path and depend on the specific implementation of the device. For example, the voltage amplitude ui”(t) is 12V p-p.

[0058] In the voltage path of the device in the embodiment, the analogue-to-digital processing module 8 receives a voltage signal ui”(t), which in the analogue-to-digital processing module 8 is processed into a series of samples of this signal u(n). For example, a converter with a bit resolution of 12 bits and a sampling rate of Fs= 10 kHz was used.

[0059] After sampling the voltage signal ui”(t) in the analogue-to-digital converter module 8, the digital signal u(n) is subjected to phase detection in the phase detection module 10. The operation of the phase 10 detection module consists in determining the index «o of the sample of the u(n) series for the moment tok. The moment tok represents the time of reaching a specific phase by the base voltage signal of the LV network (where k is the numbers of consecutive periods of the base component of the mains voltage signal). The index «o is transmitted from the output of the phase detection module 10 to one of the inputs of the sample selection module 9 of current i(n) in the current path and allows for the appropriate selection of current samples i(n) in this module and further transformation of these samples in the manner described below.

[0060] The device according to an embodiment of the invention includes in the current path 32 a current measurement sensor module, which, for example, is a current transformer 1, the input of which is connected in a serial (current) manner to a phase conductor of the power network. The operation of the 1 Tpcurrent transformer consists in transforming the LV network current signal into a voltage signal at levels acceptable by typical A / A and A / D processing modules. The changes in the voltage signal U2(isuppiy(t)) at the output of the current transformer 1 reproduce in a proportional way the changes in the currents supplying the appliances (EA) connected to the measured LV supply network circuit. At the same time, the current transformer 1 implements galvanic separation of the measurement systems from the LV power supply network. The voltage signal U2(isuppiy(t)) is transmitted from the output of the current transformer 1 to further modules in the current path of the device according to the invention.

[0061] The output of the current transformer 1 is connected via the anti-aliasing filter module 3 and the amplifier module 5 and the analogue-to-digital converter module 7 to the sample selection module 9 of the current i(n).

[0062] The anti-aliasing filter module 3 is optional and is an aspect of the embodiment, and its absence does not significantly affect the performance of the entire device according to the invention. In the absence of the anti-aliasing filter module 3, the output of the current transformer 1 is connected via the amplifier module 5 and the analogue-to-digital converter module 7 to the sample selection module 9 of the current i(n).

[0063] The filter module 3 is a low-pass filtration module, acting as an anti-aliasing filter with a limit frequency F1Cdepending on the selected sampling frequency F1Sused in the analogue-to-digital converter module 7 of the current path.

[0064] The operation of the filter module 3 consists in converting the voltage signal U2(isuppiy(t)), received from the current transformer 1, into a voltage signal U2’(isuppiy(t)) by removing from the voltage signal U2(izas(t)) components with frequencies higher than the limit frequency F1C.

[0065] The operation of the amplifier module 5 consists in further converting the voltage signal U2’(isuppiy(t)) to the voltage signal U2”(isuppiy(t)) in such a way that the signal U2”(isuppiy(t)) meets the requirements of the analog-to-digital converter module 7 in the current path of the device according to the invention. These requirements apply to the selected range of input voltages for the A / D converter used in this signal processing path.

[0066] In the current path of the device according to the invention, the analogue-to-digital converter module 7 receives a voltage signal U2”(isuppiy(t)), which in the analogue-to-digital converter module 7 is converted into a series of samples of this signal i(n). For example, a converter with a bit resolution of 12 bits and a sampling rate of F1S= 10 kHz was used

[0067] The current sampling module 9 and(n) of the device in the embodiment at its inputs receives a plurality of samples i(n) from the output of the analogue-to-digital converter module 7 of the current measurement module 32 and the values of the indices «o of reaching a specific voltage signal phase from the phase detection module 10, which have been determined from the output of the voltage measurement module 22. The operation of the sample selection module 9 of the current i(n) consists in selecting samples from the series i(n) starting from the index no for each voltage period of the LV power supply network that has reached a specific phase - for example, the voltage signal changed the sign (the so-called zero transition).

[0068] The output of the sample selection module 9 of the device according to the invention is connected via the matrix forming module 11 and the filter module 12 to the potential event detection module 13.

[0069] The matrix forming module 11 receives at its input the selected samples i(n) from the output of the sample selection module (9). The matrix forming module 11, using, for example, a cyclic buffer, converts the values of the signal samples i(n) into a matrix Ip.

[0070] The individual rows of the matrix Ip correspond to instantaneous current values that occurred at the same time interval relative to the moment to in the periods k from 1 to K. Index k indicates the number of the basic component period of the voltage signal in the power network, and index m the number of the current sample in this period. A single period has M samples, where A / is a natural number, corresponding to the number of samples recorded during one period (20 ms) of the base voltage component of the power network (50 Hz), for example M = 200 samples. The individual columns of the Ip matrix thus contain samples that were recorded in the k-th period (basic component 50 Hz). The matrix Ip has Fs / 50 rows, where Fs is the applied sampling frequency in the analogue-to-digital converter module 7 in the current measurement module 32 of the device. The data in the form of a matrix Ip is transmitted from the input of the matrix forming module 11 to the input of the filter module 12. The filter module 12 subjects the rows of the matrix Ip to filtration, as a result of which matrix I is obtained. For example, a median filtration of the order of n = 15 was used. In other aspects of the embodiment, the filter module 12 implements averaging filtration or modal filtration. Median filtration of the order n consists in determining the moving median, for a window having a width n on the values of samples along the rows of the matrix Ip. Matrix Z, contains the filtered values of current signal samples of subsequent periods of the basic component of the voltage of the LV power supply network in its each subsequent column. In other aspects of the embodiment, the design of the matrices I and Ip may be different, e.g. rows may be interchanged with columns. In still other aspects of the embodiment, the matrix forming module 11 may convert the values of the signal samples i(n) to a corresponding series of value vectors of those samples. The person skilled in the art may arrange the scheduling of the values of the signal samples i(n) and their filtration in another manner, which does not change the essence of the invention.

[0071] The matrix I is transmitted from the output of the matrix forming module 11 to the input of the potential event detection module 13. The potential event detection module 13 first determines the vector of changesd / fe 0that have occurred in the current signal, and more specifically in the columns of the matrix Z, in a time corresponding to the assumed number of periods of the basic voltage component of the LV network. At the same time, the assumed number of periods o, is predefined and depends on the type of EA detected, for example, for inductive loads, the value of the parameter o takes the value o = 7. Possible offset parameter values can be determined empirically. For the value of the order of the median filter n = 15 and the distance of the inductive load used, it is preferable to adopt the offset o = 7, meaning that the vector of changes is Ik odetermined for the current signal samples taken in the period k-o and for the samples in the period Ho. in accordance with the formula W. 1. The change vectors Alk oare determined for all periods k from the period of+ to the last period for which i(n) current signal samples can be taken reduced by o +1 (available in matrix I).

[0072] The further operation of the potential event detection module 13 consists in the fact that for each designated vector of changesd / k 0, i.e. for each period k, the values of the average change in the period SZkare calculated in accordance with formula W.2. SZk= (mean(abs(AIk oy)) (W.2)

[0073] Then, the potential event detection module 13, for all periods k, from the interval previously specified), determines peak values from all values of average changes in the period SZk.

[0074] As a result of its operation, the potential event detection module 13 determines and transmits at its output the period numbers k, where period k is the period in which the determined coefficient of average changes in the period SZklocally reaches the maximum value for the determined o. These periods are considered to be periods in which a potential change in the mode or state of the EA for the selected o.

[0075] Additionally, the potential event detection module 13 passes the matrix I at its output in the identical form in which the matrix was passed to its input from the output of the filter module 12.

[0076] The operating mode selection module 14 at its input receives from the potential event detection module 13 data in the form of matrix I and period numbers k in which the state change event of one of the EA may have occurred. The operating mode selection module 14 determines the operation of further modules of the device according to the settings made by the user of this device - the device either registers a new appliance or recognizes the defined types of EA. Since the user decides on the operating mode, he can also set appropriate parameters such as offset o.

[0077] In the learning mode circuit, the device in the embodiment comprises modules for determining EA patterns and saving these patterns to the non-volatile memory of the device. In an aspect of the embodiment, this memory can be constructed in the form of a database.

[0078] In the learning mode circuit of the device according to the invention, the operating mode selection module 14 is connected to the pattern memory module 17 through the pattern vectoring module 16.

[0079] The pattern vectoring module 16 at its input receives from the output of the operating mode selection module 14 data in the form of matrix I and period numbers k in which the state change event of one of the EA may have occurred.

[0080] The standard vector determination module 16 takes the values of samples from matrix / from the period k" before the change in the state of the EA and the values of samples from matrix / from the period k" recorded after the analysed change in the state, when the EA that changed state is already in its new steady state.

[0081] Assays

[0082] • The m-th value of the sample in the k-th period is determined as im,k- Then the vector of the samples corresponding to the k-th period of the current signal can be determined by the formula (W.3). ffc

[0083] • As X the number of EA is denoted in the monitored LV supply network circuit that can be detected using the device according to the invention.

[0084] • Asx, where: x G N A x G < 1, X > the sequential number of each EA is denoted in the LV supply network circuit to be monitored.

[0085] • The number of EA operating states with ordinal number x is determined as sx.

[0086] • As a c <1,5X> the number of the previous state of operation of the EA with ordinal number x is determined.

[0087] • The new EA operating state number with ordinal number x is denoted as b e <1,5X>.

[0088] • An appropriate number of periods of a signal with a frequency of 50 Hz before the change of the EA state with ordinal number x from the state with the number a to another state with the number b is determined as o’(X,a,b) i.e. then the period k,=k-o ^a,b).

[0089] • An appropriate number of periods of 50 Hz signal is determined as o”(X,a,b) after the change of the state of the device x from the state with the number a to another state with the number b, i.e. then the period L”=L+o”(X,a,b). The values of o’(X,a,b) and o”(X,a,b) are characteristic of EA with ordinal number x. The values of o’(x,a,b) and o”(X,a,b) should be selected in such a way that if b is the period occurring before the change of state a to b of the EA with the ordinal number x, then the period k” is the period after the transition of the EA with the ordinal number x into the steady state b. The values that can take o’(X,a,b) and o”(Xi a,b) are determined experimentally and stored in the memory of the device according to the invention. Unlike in the potential event detection module 13, in which one value of o is used, i.e. o ’ and o ” are equal, and these values are selected for specific (user-defined) types of devices, rather than specific devices. In the change vector determination module 15 and in the pattern vector determination module 16, the o ’ and o” values may be different than those substituted in the potential event detection module (13). The purpose of detection in the potential event detection module 13 is only to detect changes in the signal, the purpose of calculations in modules 15 and 16 is to determine the vector of changes, assuming that in a given period the state k changed EA from state a to state b.

[0090] • O is the set of all o’(Xi a,b) and o”(Xi a,b) - for all x e <1 and for all pairs (a, b) e <1,5X> i a b - in practice, the number of possible pairs (a, b), i.e. the number of different changes n EA states with ordinal number x is limited by the technical implementation of this EA. In an embodiment of the device according to the present invention, a finite set O is assumed.

[0091] The operation of the pattern vector determination module 16 consists in determining the EA feature vector with ordinal number x when changing the state of EA from state a to state b, marked as: is determined on the basis of changes that occurred in the current signal as a result of changes in state from a to state b by EA with ordinal number x. This vector is determined upon receipt of a signal to turn on this EA. This signal comes from the user of the device who has put the device in the state of adding new patterns.

[0092] During its operation, the pattern vector determination module 16 defines the characteristic vector Mref ’ for EA with ordinal number x by subtracting the values of matrix I samples from the period b before the change in the state of EA no. x forced by the user of the device from the values of the corresponding matrix I samples from the period k” registered when EA no. x will already be in the new steady state b. The formula for determining the reference feature vector of EA with ordinal number x takes the form dla k The pattern memory (database) module 17 at its input adopts from the output of the pattern vector determination module 16 and stores in its non-volatile memory a designated Afyey ’ EA reference feature vector with an ordinal number x along with the parameters dT^b), o\^a,b), and o”(X,a,b).

[0093] The device according to the invention, after determining the reference feature vector, A / ^’^ sends a signal to the user of this device about the need to turn off or change the state of EA with ordinal number x.

[0094] Repeated use of the device according to the invention in its teaching mode makes it possible to further determine the reference feature vectors A / ^ ’ for the number A of all EAs in the electrical circuit covered by the monitoring of the device according to the invention. Determining the reference feature vectors &Iref ’ for all A EAs in a given electrical circuit allows the creation of a database of reference patterns for all EAs in the form of four: {A / ^’b\ dT^^, o’(X,a,b), o”(X,a,b)} - The determination of each reference feature vector A / ^’ must take place with EA other than EA with ordinal number x excluded.

[0095] The operation of the pattern database module 17 consists in permanently saving, storing and providing access to other modules of the device according to the invention the patterns for each EA that the device is able to identify. The pattern of each EA with the sequence number x are feature vectors Mref’ consisting of a series of numbers with the number resulting from the ratio of the sampling frequency in the A / D converter of the converter module 7 and the voltage frequency of the LV network, i.e. for the network in Europe: Fs / 50. The parameters of each reference feature vector Afyey ’ are information about which EA and which of its states a and b apply to this vector. Additionally, the pattern memory module 17 stores information for each feature vector about how many periods of o’<Xi a,b) and o”<Xi a,b) of the basic voltage component of the LV power supply network have elapsed between the subtracted periods when determining this vector. In the pattern data module 17, a parameter defining the threshold value dT^b) is also stored, on the basis of which the decision module 19 decides (in the manner described below) whether the potential event concerns the change of state from a to b of EA with the order number x, or whether it is an event not related to any change in the state of EA in a given LV network power supply circuit. Thus, each pattern of each state change of each EA that is stored in the non-volatile memory of the pattern memory module 17 of the device of the present invention is four: {A / ^’b\ dT^b), o’(X,a,b), o”(x,a,b)} assigned to a change in a specific state of operation a to another specific state of operation b EA with ordinal number x. In the identification mode circuit of the device according to the invention, the operating mode selection module 14 is connected via the change vector determination module 15 and via the similarity determination module 181 to the decision module 19.

[0096] Additionally, the pattern memory module 17 is connected directly to the decision module 19. The change vector determination module 15 at its input receives from the output of the operating mode selection module 14 data in the form of matrix I and period numbers k in which the state change event of one of the EA may have occurred.

[0097] The change vector determination module 15 takes the values of samples from matrix I from the period k" before the change in the state of EA and the values of samples from matrix I from the period k' recorded after the analysed state change, when the EA which changed the state and will already be in the new steady state.

[0098] The following designations shall be adopted: X, xim ko' <x,a,b), o ”(x,a,b) identical to the pattern vector determination module 16 described above. The change vector determination module 15 performs calculations k k", im,k, Ikalso in the same way as in the case of the standard vector determination module 16, so the description of the method of these calculations will not be repeated here.

[0099] The method of operation of the change vector determination module 15 consists in determining the change vector AIk’ , which refers to changes that have occurred in the signal as a result of a change in state from a to state b by EA with the number x. This vector is determined at the moment of receiving a signal about a potential event (that is, during the period k of the voltage of the LV power supply network) of a change in the state of EA. This signal comes from the potential event detection module 13 propagated by the operating mode selection module 14 and is supplied to the input of the change vector determination module 15 in the form of the value of the period number k in which the state change event of one of the EA may have occurred.

[0100] The method of determining the vector of changes in the change vector determination module 15 consists in subtracting the values of the samples of matrix I from the period before the change in the state of EA from the values of the corresponding samples of matrix I from the period recorded after the analysed state change, when the EA that changed status will already be in a new steady state.

[0101] The difference between the / ??-th values of the samples from the period k''=k+o" and the period k,=k-o' ^x,a,b)wasdetermined as Aim feand the change vector determination module 15 calculates according to the formula (W.5). Based on the definition Aim feand the / fechange vector determination module 15 calculates the change vector according to the formula (W.6).

[0102] (x CL b )

[0103] The change Mk’ ’ vector is at the same time a feature vector from which the change in the state of the EA with the ordinal number is identified x by comparison with the reference feature vectors of Mref ’ all EA in a given electrical circuit. The device of the invention performs this comparison in the similarity determination module 181.

[0104] The similarity determination module 181 at its inputs receives from the output of the change vector determination module 15 the calculated values of the change vector and from the output of the pattern memory module 17 the matrix A / re^ containing the pattern vectors of all X EA, as well as the matrix O containing the parameters o’(X,a,b) and o”(X,a,b) appropriate for these pattern vectors. The operation of the similarity determination module 18 consists in comparing the variation vector obtained at its input with each pattern for the respective values of the parameter o’(X,a,b) and o”(X,a,b).

[0105] The similarity determination module 18 may use various known similarity measures, for example, it is a mid-square measure. The result of the operation of the similarity determination module 18 is the formation of a vector of similarity coefficients d. which contains in its elements the calculated similarity coefficients <7(x,a’b)for each EA pattern.

[0106] The decision module 19 at its inputs adopts from the input of the similarity determination module 18 the vector of similarity coefficients d and from the output of the pattern database module 17 the threshold value dT(X^ characteristic for each EA with the number x and for each pair of state changes from a to b of this EA.

[0107] The operation of the decision module 19 consists in determining the maximum value of the coefficient < / x a b)for the entire vector d and comparing this maximum value with the threshold value dT^b). The decision module 19 then makes a decision based on that comparison. If the maximum value of the coefficient < / x a b)for the vector d exceeds the threshold value dT^a_b), then the system decides that the tested event concerns a change in the state of EA with the ordinal number x and a change in its state from a to another state b.

[0108] If the maximum value of the coefficient < / x a b)for the vector d does not exceed the threshold value dT(Xt a,b), then the decision module 19 decides that the previously detected potential event is not an event indicating a change in the state of any EA in the tested LV network power supply circuit.

Claims

Claims1. Device for detecting changes in the operating mode or identifying electrical appliances EA in the power network, which contains: a voltage track that includes a voltage measurement module (22), the input of which is connected in parallel to the power network and the output through the phase detection module (10) to the sample selection module (9), and a current track that includes a current measurement module (32), the input of which is connected in series to the power network, and the output of which is connected to the sample selection module (9), the output of the sample selection module (9) is attached to the matrix forming module (11), wherein the sample matrix of the matrix forming module (11) comprises a predetermined number of the last K periods of voltage signal and m samples of instantaneous current value for each of the last K periods, at the same time, the output of the matrix forming module (11) via the filtration module (12) is connected to the input of the change vector determination module (15), wherein the change vector determination module (15) is adapted to determine change vectors, wherein the change vectors are transmitted to the input of the decision-making module (18) by the output of the change vector determination module (15), in addition, the output of the EA pattern memory module (17) is attached to the input of the decision-making module (18), wherein the output of the decision-making module (18) provides information for detecting changes in the operating mode or identifying electrical appliances EA, characterized in that the filtration module (12) is connected to the change vector determination module (15) via the potential event detection module (13), wherein the potential event detection module (13) is arranged to transmit to the input of the change vector determination module (15) a sample matrix from the matrix forming module (11) and additional information about the period number in which the potential EA mode change occurred.

2. The device according to claim 1, characterized in that the decision-making module (18) comprises a similarity determination module (181) and a decision module (19), wherein data from the change vector determination module (15) and data from the pattern memory module (17) are fed to the input of the similarity determination module (181), the output of the similarity determination module (181) is connected to the first input of the decision module (19), and the second input of the decision module (19) is connected to the output of the pattern memory module (17).

3. The device according to claim 2, characterized in that the similarity determination module (181) in the decision-making module (18) is adapted to compare the values of the change vectors with the pattern definition vectors that are stored in the EA pattern memory module (17).

4. The device according to claim 2 or 3, characterized in that the decision module (19) in the decisionmaking module (18) is adapted to identify the EA and / or identify the change in state of the EA based on data from the similarity determination module (181) and the pattern memory module (17).

5. The device according to claim 1-4, characterized in that a mode selection module (14) is placed between the potential event detection module (13) and the change vector determination module (15), wherein one of the outputs of the mode selection module (14) is connected to the decision module (19) via the pattern vector determination module (16) and the pattern memory module (17), and the other output of the mode selection module (14) is connected to the decision module (19) via the change vector determination module (15) and via the similarity determination module (181), wherein when the operation mode of the new EA pattern recording is active, the mode selection module (14) is adapted to transmit information from the input to the pattern determination module (16), when the operation mode of the new EA pattern recording is not active, the operation mode selection module (14) is adapted to transmit data only to the change vector determination module (15).

6. The device according to claim 1 - 5, characterized in that the phase detection module (10) is adapted to determine the time at which the voltage signal value of the electrical network reaches a predefined phase value, preferably when the voltage signal value of the electrical network changes from a negative to a positive value.

7. The device according to claim 1 - 6, characterized in that the sample selection module (9) is arranged to transmit to the matrix forming module ( 11) the values of the current samples, starting with the sample number determined by the phase detection module (10) for each subsequent period of the voltage signal, wherein the number of transmitted samples results from the duration of one period of the voltage signal and the sampling frequency in the current measurement module (32).

8. The device according to claim 1 - 7 characterized in that the matrix forming module (11) is adapted to store the values of the current samples for a certain number K of the last periods of the voltage signal that form the sample matrix that is fed to the input of the filtration module (12).

9. The device according to claim 8, characterized in that the filtration module (12) is adapted to perform median filtration or averaging filtration or mode filtration.

10. The device according to claim 1-9, characterized in that the filtration module (12) is adapted to perform filtration for a series of samples taken at moments equally distant from the phase determined by the phase detection module (4).

11. The device according to claim 1-10, characterized in that the voltage measurement module (22) includes a voltage transformer (2), to which a low-pass filter (4) is connected, and the filter output (4) through the amplifier module (6) is connected to the analogue-to-digital converter module (8), the output of which is the output of the voltage measurement module (32), or the voltage measurement module (22) comprises a voltage transformer (2), to the secondary winding of which an amplifier module (6) is connected, which is coupled to the analogue-to-digital converter module (8) andfurther to the low-pass digital filter module (42), and the output of the low -pass digital filter (42) is the output of the voltage measurement module (22).

12. The device according to claim 1-12, characterized in that the current measurement module (32) comprises a current transformer (1), to the secondary winding of which an amplifier module (5) is connected, which is coupled to the analogue-to-digital converter module (7), preferably an anti-aliasing filter (3) is placed between the secondary winding of the current transformer (1) and the amplifier module (5), wherein the throughput bandwidth of the anti-aliasing filter (3) is related to the frequency of sampling of the converter in the analogue-to-digital converter module (7), or the current measurement module (32) comprises a current transformer (1) to the secondary winding of which an amplifier module (5) is connected, which is coupled to the analogue-to-digital converter module (7), preferably the analogue-to-digital converter module (7) is connected to the anti-aliasing digital filter module (52), the output of the digital anti-aliasing filter (52) or the analogue-to-digital converter (7) is the output of the current measurement module (32).

13. Method for detecting changes in the operating mode and identifying electrical appliances EA using measurements of instantaneous values of current signals and voltage phase detection, including: a current measurement step in the power supply network and a voltage measurement step, wherein, after the current measurement step, the current signal is converted into a digital form thereof, which is transmitted to the sample selection module (9), and after the voltage measurement step, the voltage signal is converted into its digital form on the basis of which, using the phase detection module (10), the detection of a specific value of the voltage signal phase is carried out, and then at the stage of matrix formation using the matrix formation module (11), appropriate values of samples from the sample selection module (9) are selected, which are further formed into a matrix in which subsequent m rows contain samples of instantaneous current values in k columns of subsequent voltage periods starting from the moment of reaching a specific phase value by the voltage signal, and then at the filtration stage, using the filtration module (12), the matrix is filtered by its rows to obtain a sample matrix after filtration, based on which change vectors are determined in the change vector determination module (15), covering the differences in the values of the samples of the sample matrix after filtration from the period before the change in the state of the EA from the values of the samples of the sample matrix after filtration from the period recorded after the analysed change in state, when the EA has changed the state and already is in the new steady state, and then the identification step follows, in which the designated change vectors by means of a decision-making module (18) are compared with the reference feature vectors of the EAs in the given electrical circuit placed in the memory module (17), identifying the EA based on the most suitable patterncharacterized in that after filtration of the sample matrix in the filtration module (12), there is a stage of detection of potential events, wherein the detection consists in determining in the potential event detection module (13) the vectors of changes for at least some of the K voltage periods, which vectors allow to determine the proportionality coefficient proportional to the mean of the absolute values of the elements of these vectors (W.2), wherein the period k, in which the designated coefficient locally reaches the maximum value, is considered to be the period in which the potential change in the mode or state of EA occurred, then the information about the number of the period in which the potential change in the mode or state of EA occurred is transferred to the change vector determination module (15), wherein the change vectors determined in the change vector determination module (15) include the differences between the values of the sample matrix samples after filtration for the period before the period in which the potential change in the mode or state of the EA occurred, and the values of the sample matrix samples after filtration for the period after the period in which the potential change in the mode or state of the EA occurred, whereby only those periods for which the proportionality coefficient has reached the maximum value are taken into account, only the selected change vectors determined in the change vector determination module (15) are further compared in the decision-making module (18) and are used to identify EA based on the most suitable pattern.

14. The method according to claim 13, characterized in that the determination of the proportionality coefficient in the potential event detection module (13) includes the determination of change vectors including differences in the values of the sample matrix samples after filtration from the period k- o’ before the change in the state of the EA from the values of the sample matrix samples after filtration from the period k+o” recorded after the analysed period k, whereby only selected periods before and after are analysed for a predefined at least one pair of offset values o’ and o”, preferably the offset o’ and o” corresponds to the category of EA and the offset values o’ and o” may not be equal, the proportionality coefficient is determined on the basis of the determined difference in the values of the matrix samples for each predefined at least one pair of offsets o’ and o”, preferably the proportionality coefficient is the average of the absolute value of the determined difference in the values of the samples.

15. The method according to claim 13 or 14, characterized in that, by means of the phase detection module (10), a time moment is determined in which the voltage value of the electrical network changes from a negative to a positive value.

16. The method according to claim 13-15, characterized in that, by means of the matrix forming module(11), the values of the current samples are transmitted from the sample selection module (9) in the form of a sample matrix, whereby for each period k, starting from the sample determined by the voltage detection module (10), the number of m samples corresponding to each of the voltage periods is transmitted, preferably the matrix forming module (11) contains the matrix buffer on K of the last voltage periods and transfers the sample matrix to the output every each voltage period, which the sample matrix is given to the input of the filtration module(12).

17. The method according to claim 13-16, characterized in that filtration is carried out for a series of samples taken at equally distant moments in time from the phase determined by the phase detection module (10).

18. The method according to claim 13-17, characterized in that all steps are performed on signal samples in the time domain.

19. The method according to claim 13-18, characterized in that the moments of sampling the current signal are correlated in time with the moment of reaching a specific phase by the voltage signal.

20. The method according to claim 13-19, characterized in that the filtration is a median filtration or averaging filtration or mode filtration.

21. The method according to claim 13-20, characterized in that it comprises:- determining in the similarity determination module (181) similarities between all reference change vectors taken from the pattern memory module (17), wherein the pattern memory module (17) contains reference change vectors characteristic of EA, and a set of vectors of changes determined in the change vector determination module (15), wherein the determined similarities are transmitted to the decision module (19);- making decisions by the decision module (19) on the basis of the similarities determined in the similarity determination module (181) and on the basis of data from the pattern memory module (17) with a threshold value characteristic for each type of EA and a possible change in the state of this EA, the pattern of which is stored in the pattern memory module (17).

22. The method according to claim 13-21, characterized in that the decision making by the decision module (19) comprises: determining the maximum value of the similarity coefficient for the entire similarity vector, and comparing this maximum value with the threshold value taken from the standard memory module (17), then decision making by the decision module (19) based on this comparison, if the maximum value of the similarity coefficient for the similarity vector exceeds the threshold value, then the decision module (19) decides that the potential event in the period k concerns a change in the state of the specified EA, while if the maximum value of the similarity coefficient for the similarity vector does not exceed the threshold value, then the decision module (19) decides that the previously detected potential event is not an event indicating a change in the state of any EA in the tested power supply circuit of the power network.

23. The method according to claim 13-22, characterized in that in the decision-making stage in the decision module (19), in a situation where the given decision selection criteria are insufficient, statistical algorithms are used to select the most likely EA from among the recorded patterns, preferably it is a decisiontree algorithm or a random forest, or K-nearest neighbours (KNN), or convolutional neural networks (CNN), or support vector machines (SVM), or deep neural networks (DNN), or Naive Bayes.

24. The method according to claim 13-23, characterized in that before the decision-making stage in the decision module (19) a stage of adding patterns of changes to the pattern memory module (17) is performed, wherein the stage of adding patterns depends on the predefined and provided by the user data on the category of the device being added and the moment of its connection to the network, the stages of adding patterns include: setting the operating mode selection module (14) to the pattern addition mode (output duplicate mode), which supplies the sample matrix to the pattern determination module (16), determining the reference change vectors, wherein the determination of the reference change vectors takes place similarly to the determination of the change vectors, wherein in the determination of the reference change vectors, predefined values of offsets o’ and o” and a predefined threshold value characteristic for the given category of EA are adopted, then memorizing a reference change vector, along with corresponding o’ and o” offset values, and a predefined threshold characteristic of a given EA category in a pattern memory module (17), wherein after adding the reference change vector to the pattern memory module (16), the recognition of the EA as defined in claim 13 begins or another device is added.

Images