Method and system for using a graphical categorization of electrical loads to identify one of a large number of different electrical load types.

Abstract

A system (100) for identifying electrical load types for a variety of different electrical loads (102, 103, 104), wherein the system comprises: a hierarchical load attribute database (108) comprising a plurality of layers (110), wherein a (112) of the layers contains a plurality of different load categories, wherein a first layer (112, L1) of the plurality of layers contains the different load categories; wherein a second layer (L2) of the plurality of layers contains a plurality of different load subcategories for each of the different load categories; and wherein a third layer (L3) of the plurality of layers contains the different electrical load types for the different load subcategories; and a processor (114) which is constructed (116) to: Acquiring a voltage waveform and a current waveform for a corresponding one of the different electrical loads; Mapping a voltage-current trajectory onto a grid containing a plurality of cells, each of which has a binary value of zero or One is assigned based on whether the voltage-current trajectory passes through each of the cells; Extracting a variety of different properties from the depicted grid of cells as a graphical signature of the corresponding one of the different electrical loads; Deriving a category of the corresponding one of the different electrical loads from the hierarchical load attribute database; and Identifying one of a variety of different electrical load types for the corresponding one of the different electrical loads, where the voltage waveform and the current waveform each represent a total number of K data points of the form (v k , i k ) contained, where: k = 1,...,K; where v k and i k correspondingly, a voltage and a current value of a sampled data point k; where maximum and minimum values ​​of the voltage waveform and the current waveform are calculated from: v max = max v k , V min = min v k , i max = max i k , i min = min i k , v 0 = 1 2 ( v max + v min ) , and i 0 = 1 2 ( i max + i min ) ; where v0 and i0 are both averages of the corresponding maximum and minimum values ​​that form the central points of the grid of cells, where Δ defines a size of the grid; where dv = v max − v 0 Δ di = i max − i 0 Δ ; the processor is further configured to generate two sequences from the voltage waveform and the current waveform as: { v 0 − dv ⋅ Δ , v 0 − dv ⋅ ( Δ − 1 ) , … , v 0 , … , v 0 + dv ⋅ ( Δ − 1 ) , v 0 + dv ⋅ Δ} and { i 0 − di ⋅ Δ , i 0 − di ⋅ ( Δ − 1 ) , … , i 0 , … , i 0 + di ⋅ ( Δ − 1 ) , i 0 + di ⋅ Δ} where each of the two sequences has N = 2Δ+1 elements; where the grid of cells includes a first axis having N cells and a second axis having N cells; and where each of the cells has a position value (v0 + dv · x,i0 + di · y) and a binary model value B x,y is assigned, which is initialized to 0.

Description

CROSS-REFERENCE TO RELATED REGISTRATION

[0001] This application claims the priority and rights of U.S. patent application serial no. 13 / 912,819, filed on June 7, 2013, which is incorporated herein by reference.

[0002] The invention was made with government support under patent DE-EE0003911, granted by the Department of Energy National Energy Technology Laboratory. The government holds certain rights to this invention. BACKGROUND area

[0003] The disclosed concept relates generally to electrical loads or consumers and, in particular, to methods for identifying different types of electrical loads or consumers. The disclosed concept also relates to systems for identifying different types of electrical loads. Background information

[0004] Electrical loads, or consumers, in commercial and residential buildings consumed approximately 75% of total electricity in the United States in 2012. However, a significant portion of this electrical use was wasted, and its management was often overlooked. Many electrical devices with external power supplies, remote controls, continuous displays, or battery chargers draw power continuously even when switched off or in standby mode. Electrical loads with external power supplies are also referred to as plugged-in loads (PELs) (or, in some contexts, as miscellaneous or other electrical loads). PELs are one of the major load categories and account for a larger share of electricity consumption than any other end-user service, such as heating or ventilation.

[0005] Standby power consumption in the US exceeds 100 billion kWh and costs over $10 billion annually. Up to 75% of these costs can be saved through appropriate energy management. To achieve the net-zero energy building targets defined by the Department of Energy (DOE) for residential buildings by 2020 and for commercial buildings by 2025, effective monitoring and management of power-enabled energy sources (PELs) must be considered. Understanding the type of PEL is essential for implementing an effective solution.

[0006] Since the introduction of non-invasive load monitoring (NILM) in the 1980s, numerous proposals have attempted to find various NILM solutions. A wide selection of well-known solutions is revealed by Du et al., “A review of identification and monitoring methods for electric loads in commercial and residential buildings”, Proc. 2010 IEEE Energy Conversion Conf. and Expo, 2010, pages 4527-33.

[0007] A load or consumer identification system consists of several modules, including data acquisition, data processing, event detection, property extraction, and identity display. The identity display module compares the extracted properties with a database of properties of known loads or consumers and identifies unknown loads based on predefined rules, such as maximum similarity or training results from artificial neural networks (ANNs).

[0008] The performance of almost all existing load identification methods depends heavily on the electrical signatures of loads, which are defined as "the electrical expression or representation that a load device or equipment possesses in a unique or individual way." The goal is to extract useful properties that can uniquely distinguish the individual PEL types or classes within a predetermined load configuration.

[0009] Many property extraction methods have been proposed. For example, real power (active power) and reactive power are used to identify load types for property exploration of a continuous state. Peak current, average current, and RMS current values ​​can also be used for load identification. Current oscillations are used as the core identification property, primarily to address loads with a non-linear power supply. Furthermore, a voltage-current or VI trajectory modeling method uses purely graphical shape properties of each load's VI trajectory for load identification. Some transition state features, such as immediate admittance curves and transition power curves, can also be employed.

[0010] The development of property extraction and the assignment of each load type to a corresponding load group was purely data-driven. While many previous proposals demonstrate that satisfactory performance can be achieved by selecting a suitable set of features for a targeted load combination, there are no known guidelines on how to perform optimized property selection, and there may be redundancy in information within each set of features. Furthermore, identification performance typically depends on the specific load combination being studied. It is assumed that the generalizability of the performance of a developed classifier to other load combinations has not yet been investigated, and that no such set of electrical signatures exists, such that each load has a "unique" expression.has a "unique" expression or representation.

[0011] Due to the complexity and subtle differences of installations and devices, it is often challenging, if not impossible, to distinguish between loads or consumers that use the same interface circuitry to a power line. For example, those PELs that use a standardized DC power supply with current oscillation reduction, such as DVD players, cable or satellite set-top boxes, and PC monitors, exhibit very similar electrical signatures and are indistinguishable by using steady-state characteristics alone.

[0012] Therefore, a truly meaningful load categorization method is often still desirable.

[0013] US 2013 / 0138651A1 describes a method for identifying electrical load types among a variety of different electrical loads.The method involves providing a self-organizing mapping load feature database with a multitude of different electrical load types and a multitude of neurons, each load type corresponding to a number of neurons; using a weight vector for each neuron; acquiring a voltage signal and a current signal for each load; determining a load feature vector containing at least four different load features from the acquired voltage signal and the acquired current signal for a given load; and identifying one of the load types by a processor by relating the load feature vector to the neurons in the database by identifying the weight vector of one of the neurons corresponding to the load type that has a minimum distance to the load feature vector.

[0014] US 2013 / 0138669A1 describes a method for identifying electrical load types among a variety of different electrical loads.The method comprises providing a hierarchical load characteristic database with a plurality of layers; inserting a corresponding load characteristic set into each of the plurality of layers, wherein the corresponding load characteristic set of at least one of the layers differs from the corresponding load characteristic set of at least one other of the layers; inserting a plurality of different electrical load types into one of the layers; acquiring a voltage signal and a current signal for each of the different electrical loads; determining at least four different load characteristics from the acquired voltage signal and the acquired current signal for a corresponding one of the different electrical loads; and identifying one of the different electrical load types by a processor by relating the different load characteristics to the hierarchical load characteristic database.

[0015] US 2013 / 0138661A1 describes a method for identifying electrical load types from a multitude of different electrical loads. The method includes providing a load characteristic database containing a multitude of different electrical load types, each of which comprises a first load characteristic vector with at least four distinct load characteristics; acquiring a voltage signal and a current signal for each of the different electrical loads; determining a second load characteristic vector, comprising at least four distinct load characteristics, from the acquired voltage signal and the acquired current signal for one of the different electrical loads; and identifying one of the different electrical load types by a processor by determining a minimum distance between the second load characteristic vector and the first load characteristic vector of the different electrical load types in the load characteristic database.

[0016] Furthermore, reference should be made to an article by Costa, Jose Alfredo F., and De Andrade Netto, Marcia L., entitled "Automatic data classification by a hierarchy of self-organizing maps." Published in: IEEE SMC'99 Conference Proceedings: 1999 IEEE International Conference on Systems, Man, and Cybernetics, October 12-15, 1999, Tokyo, Japan. Volume V. Piscataway, NJ, USA: IEEE-E, 1999, pp. 419-424. The article describes clustering, the process by which discrete objects are assigned to groups with similar properties. Self-organizing maps (SOMs) are frequently used as a data visualization tool. Among their advantages are information compression and density estimation while preserving the topological and metric relationships of the primary data elements. To use SOMs as a clustering tool, additional procedures are required to interpret the assignments obtained through unsupervised learning.The purpose of this article is to improve the clustering process in order to further detail the underlying structure obtained in an initial attempt. Groups of neurons associated with clusters are further subdivided into new subnetworks, resulting in a tree-like structure of SOMs. Unlike other hierarchical SOM approaches, the number of subnetworks for a given SOM at a specific height in the tree is not predetermined. The process can be viewed as a dynamic strategy for cluster detection.

[0017] There is potential for improvement in methods for identifying different types of electrical loads.

[0018] There is also the possibility of improving systems for identifying different load or consumer types. SUMMARY

[0019] These and other requirements are met by the embodiments of the disclosed concept which map a voltage-current trajectory onto a grid containing a plurality of cells, each of which has a binary value; extract a plurality of distinct properties from the mapped grid of cells as a graphical signature of a corresponding one of a plurality of distinct electrical loads; derive a category of the corresponding one of the distinct electrical loads from a hierarchical load property database; and identify one of a plurality of distinct electrical load types for the corresponding one of the distinct electrical loads.

[0020] According to one aspect of the disclosed concept, a system for a multitude of different electrical loads comprises: a multitude of sensors configured to sense a voltage signal and a current signal for each of the different electrical loads; a hierarchical load attribute database comprising a multitude of levels or layers, one of which comprises a multitude of different load categories; and a processor configured to: acquire orAcquiring a voltage waveform and a current waveform from the sensors for a corresponding one of the different electrical loads; mapping a voltage-current trajectory onto a grid containing a plurality of cells, each cell being assigned a binary value of zero or one; extracting a plurality of different properties from the mapped grid of cells as a graphical signature of the corresponding one of the different electrical loads; deriving a category of the corresponding one of the different electrical loads from the hierarchical load property database; and identifying one of a plurality of different electrical load types for the corresponding one of the different electrical loads.

[0021] Another aspect of the disclosed concept is a method for identifying load types for a multitude of different electrical loads, wherein the method comprises: sensing a voltage signal and a current signal for each of the different electrical loads; providing a hierarchical load property database having a multitude of layers, one of the layers containing a multitude of different load categories; acquiring a voltage waveform and a current waveform from a corresponding one of the different electrical loads; mapping a voltage-current trajectory onto a grid containing a multitude of cells, each of the cells being assigned a binary value of zero or one; extracting a multitude of different properties from the mapped grid of cells as a graphical signature of the corresponding one of the different electrical loads;Deriving a category of the corresponding one of the different electrical loads from the hierarchical load attribute database; and identifying one of a multitude of different electrical load types for the corresponding one of the different electrical loads.

[0022] The embodiments of the present invention are described in main claim 1 and in dependent claim 8. Further embodiments of the invention are described in the dependent claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] A full understanding of the disclosed concept can be obtained from the following description of preferred embodiments when read in conjunction with the accompanying drawings, which show the following: Fig.Figures 1A-1G are graphical representations of current versus voltage and normalized current versus normalized voltage for VI trajectories of representative loads in seven load categories according to embodiments of the disclosed concept. Fig. 2A-2D are graphical representations of normalized current versus normalized voltage for the VI trajectories of four specific example loads. Fig. Figure 3 is a mapping of a graphical representation of a VI trajectory onto a binary cell grid according to exemplary embodiments of the disclosed concept. Fig. Figure 4A is a graphical representation of a sampled voltage versus a discrete sample for a given load, which includes an average of the maximum and minimum voltage values, according to an embodiment of the disclosed concept. Fig.Figure 4B is a graphical representation of a sampled current versus a discrete sample for the specified load of the Fig. 4A, which includes an average of the maximum and minimum current values. Fig. 4C is a graphical representation of current versus voltage for the VI trajectory of the given load. Fig. 4A, which shows the average values ​​of the maximum and minimum voltage and current values. Fig. Figure 5A is a graphical representation of sampled voltage versus discrete sampling for a given load, which includes a specific voltage sampling according to an embodiment of the disclosed concept. Fig. 5B is a graphical representation of sampled voltage versus discrete sampling for the specified load of the Fig. 5A, which includes a specific current sampling rate. Fig.5C is a graphical representation of current versus voltage for the VI trajectory of the specific load of the Fig. 5A, which indicates the specific voltage and current sampling. Fig. 6A and Fig. Figure 6B are exemplary graphical representations of binary cell grids according to embodiments of the disclosed concept. Fig. Figure 7 is a graphical representation of a self-intersecting interface that includes a VI trajectory, according to an embodiment of the disclosed concept. Fig. Figure 8 is a block diagram of a system that uses graphical categorization of electrical loads to identify one of a multitude of different electrical load types according to embodiments of the disclosed concept. DESCRIPTION OF PREFERRED EXECUTION EXAMPLES

[0024] As used herein, the term "number" is intended to mean one or an integer greater than one (i.e., a multitude).

[0025] As used herein, the term "processor" shall mean: a programmable analog and / or digital device capable of storing, retrieving, and processing data; a computer; a digital signal processor (DSP); a controller; a workstation; a personal computer or PC; a microprocessor; a microcontroller, a microcomputer; a central processing unit or CPU; a mainframe computer, a minicomputer; a server; a network processor; or any suitable processing device or apparatus.

[0026] According to the disclosed concept, categorizing an electrical load using graphical methods investigates the relationship between physical electrical circuits and their corresponding functions. With a thorough understanding of electrical devices, property extraction is facilitated by understanding the relationship between different steady-state current waveforms and their corresponding circuit topologies. The resulting features are defined in a way that is defined within a load model rather than solely through targeted data searches (also known as a purely data-driven approach). Electrical signatures of loads are extracted from VI trajectories. The VI trajectories are first mapped onto a grid of cells, each assigned a binary value. A set of simple but efficient properties is then extracted from the mapped cell grid containing binary values.The established relationship is very helpful in organizing the feature space and defining simple properties. The disclosed mapping to cell grids with binary values ​​aims to avoid performing a discrete Fourier transform (DFT) and reduce the required computing resources. It also provides a description of the boundaries of properties or features of the stationary state that were used in previous proposals.

[0027] US patent application Publication number 2013 / 0138669 entitled: “System And Method Employing A Hierarchical Load Feature Database To Identify Electric Load Types Of Different Electric Loads”, incorporated herein by reference, discloses a system and method that employs a hierarchical load characteristic database and classification structure as a model-driven orientation for optimized characteristic selections.

[0028] The disclosed concept fits into the Level 1 categorization in the hierarchical load identification framework disclosed in Publication No. 2013 / 0138669 and focuses on property extraction from the steady state. Due to the potential limitation of using only steady-state properties, a higher level of detail for load identification can be achieved by introducing Level 2 and Level 3 load identification / categorization in the hierarchical load identification framework of Publication 2013 / 0138669.

[0029] A hierarchical load attribute database has three levels or layers, although more than three layers can be used. The first layer or level (Level 1) is the load category, the second layer or level (Level 2) is the load subcategory, and the third layer or level (Level 3) is the load type, which includes a variety of different load types.

[0030] Non-restrictive examples of first-level load categories include resistive loads, reactive loads, non-linear loads with power factor correction, non-linear loads without power factor correction, non-linear loads with transformer, non-linear loads with phase angle control, and complex structures.

[0031] Non-restrictive examples of second-level load subcategories include resistive loads, such as lighting equipment, kitchen appliances, and personal care devices; reactive loads, such as linear reactive loads and non-linear loads with saturation of devices; non-linear loads with power factor correction, such as large monitors, television equipment, and other large consumer electronics equipment; non-linear loads without power factor correction, such as imaging or image capture equipment, small monitors and televisions, personal computers or PCs, electrical loads with battery chargers, lighting loads, and other small electronic equipment; non-linear loads with transformers, such as small electronics without a battery charger and others with a battery charger; and complex structures, such as a microwave oven.

[0032] A few non-restrictive examples of third-level load types are incandescent light bulbs (<100W) for lighting appliances, and a bread toaster, a space heater, and other appliances for the kitchen and personal care. Load categorization through front-end electronic circuit topologies

[0033] The electrical signals, i.e., voltage and current waveforms, of PELs during steady-state operation are directly related to the circuit topology of their front-end power supply units. The first level, i.e., Level 1 in Table 1 below, includes seven load categories: resistive loads (R); reactive, predominant loads (X); electronic loads (E-loads) with a power factor correction circuit (P); electronic loads without a power factor correction circuit (NP); linear power supplies that utilize a transformer for voltage amplification (T); phase angle controllable (PAC) loads; and complex structures (M).

[0034] The majority of resistive loads (R) are used for heating, cooking, and lighting. Non-restrictive examples of such loads include space heaters, coffee makers, and incandescent light bulbs. For reactive loads (X), the equipment often consists of compressors, motors, or refrigeration units. The motors commonly used in these devices are often small DC motors. Non-restrictive examples of such loads in this subcategory include fans, washing machines / dishwashers, refrigerators, and shredders / paper shredders. The next two major groups of equipment are all electronic loads, designated in Table 1 as categories P and NP. Since IEC standard 61000-3-2 limits the harmonic current level for all loads with a power rating above 75 watts, it can be assumed that a power factor correction (PFC) module is required to meet this requirement.Therefore, category P refers to electronic loads with PFC (Power Factor Correction). Personal computers (PCs) over 75 W, projectors, LCD TVs, LED TVs (operating in "high quality mode"), plasma TVs, home theaters, and game consoles all fall into category P. In contrast, category NP refers to electronic loads that do not use power factor correction techniques. Small devices, such as mobile phone chargers, portable DVD players, adapters for portable printers, scanners, fax machines, and multifunction devices (MFDs) that use inkjet technology, PC monitors, LED TVs (when operating in power-saving mode), and PCs (when operating in low-power mode) are the main loads or consumers in this subcategory.Category T loads refer to low-power devices that use linear DC power supplies with a relatively small front-end transformer. Battery chargers, hole punchers, and staplers are non-restrictive examples of characteristic loads in this category. Devices such as light dimmers that use thyristor phase-angle voltage control are listed in the PAC category. Category M includes devices that often have relatively high power consumption and multiple electrical systems, such as microwave ovens and laser printers. Category M loads also include power electrical appliances (PEAs) that operate at different power levels and repeatedly switch between these power levels during use. These PEAs are programmed to operate in this repeatedly switching mode because their functional performance may require repeated processes in a specific sequence.For example, most high-volume printers have two print units in a single unit and are capable of printing both sides of the paper in a single pass. A double-sided print job is a repeated process of feeding a sheet of paper, printing, and rolling the paper forward, holding the paper to allow the ink to dry, moving or reversing the paper to print the other side, and feeding the next sheet of paper for faster printing. The two units are programmed to operate in different combinations during this repeated process, and these combinations could fall into one or more of the categories listed above. Table 1 Seven load categories according to front-end electronic circuit topologies Examples of pluggable loads in each category R: Resistive loads R1: Incandescent lamps (< 100W) R2: Space heaters R3: Bread Toaster R4: Coffee machines - Other kitchen appliances X: Reactive, predominant loads X1: Fans X2: Cooling unit (any with a cooling unit) X3: Vending machine X4: Paper Shredder P: E-loads with PFC P1: PC (Desktop / Laptop) (>75W) P2: Projectors P3: Large TVs (LCD / LED) (>75 W) P4: Home cinema / game consoles (70-80W) NP: E-loads without PFC nP1: PC (Laptop) (< 75 W) nP2: Charger (any with battery) nP3: Other small electronic devices nP3: FL / CFL nP5: Portable Multifunction Device (MFD) / Printer / Scanner nP6: PC monitors T: Linear loads T1: Small electronics (e.g., staplers) T2: AA battery charger PAC: Phase Angle Controlled Loads PAC1: Dimmer PAC2: Others with thyristor-controlled rectifier M: Complex structure M1: Microwave oven M2: Printer / Copier / Fax machine / MFD Typical VI trajectories of pluggable load categories

[0035] Fig.Figures 1A-1G show graphical representations 2, 6, 10, 14, 18, 22, and 26 of current versus voltage and graphical representations 4, 8, 12, 16, 20, 24, and 28 of normalized current versus normalized voltage for the VI trajectories of representative loads in the load categories R, X, P, NP, T, PAC, and M, respectively. Since only Level 1 subcategories are considered in the first step, all loads exhibit very different characteristics from one category to another. By applying a relatively simple property space, it is possible to perform Level 1 load category identification, which represents a potential solution for a cost-effective implementation of pluggable load identification with embedded systems.

[0036] From the Fig.In sections 1A-1G, it can be seen that the normalized VI trajectories appeared different for the various load categories. An earlier proposal, based on 126 sets of operational data from different PEL types and modes, summarized that there are eight shape signatures that can be considered to describe the VI trajectory: asymmetry, loop direction, area, mean line curvature, self-intersections, middle segment slope, area of ​​left and right segments, and the middle segment peak. However, calculating these graphical signatures still requires significant computational resources, as the entire VI trajectory must be traversed or transformed in a specific order or direction. Furthermore, these signatures are designed for a taxonomy or classification of loads similar to the load groups defined by IEC standard 61000-3-2.Therefore, they are not suitable for the proposed seven load categories. Furthermore, as will be discussed, these signatures extracted from the VI trajectories cannot effectively handle the diversity within each type of PEL and the similarity between different types of PELs. Instead, the disclosed concept employs a different set of signatures, which can be extracted from the VI trajectories, for the purpose of load identification. Furthermore, such signatures of different categories are unique and distinct. Limitations of existing graphical load signatures

[0037] The existing graphical load signatures discussed above are based solely on shape properties. However, different PEL models within the same category may have similar (but not identical) front-end power supply topologies. Therefore, such PELs will exhibit similar (but not identical) current waveforms, as well as VI trajectories. In this case, there may be significant differences in some existing graphical load signatures that should be identical, since these PELs belong to the same type or category. Furthermore, some existing graphical signatures may no longer be accurate or useful.

[0038] Several exemplary graphical representations 30, 32, 34, 36 of normalized current versus normalized voltage for the VI trajectories of certain loads are shown in the Fig. 2A-2D shown. Fig. 2A and Fig.Figures 2B and 32 show the graphical representations of the VI trajectories of two portable fans (e.g., 32 inches and 9 inches, respectively). It can be observed that these two VI trajectories have a similar shape, but quite different area values ​​for both the overall VI trajectory and the left and right segments, as well as peak values ​​for the middle segments. As a further example, the Fig. 2C and Fig. 2D graphical profiles 34, 36 for the VI trajectories of two flat-screen televisions (e.g., an LED and an LCD). It can be observed that these two VI trajectories have a similar shape but quite different zero-crossing times (and therefore different values ​​for the left, middle, and right segments). It is also relatively complicated to determine the asymmetry, the loop direction, and a region of the graphical representation 34 of the Fig.Determining 2C requires a relatively long calculation time due to the oscillation in the VI trajectory. Binary mapping of VI trajectories

[0039] Fig. Figure 3 shows the mapping of a graphical representation 38 of a VI trajectory onto a binary cell grid 40. To handle the difference between the VI trajectories of PELs within the same load category and to reduce errors, the disclosed concept first maps the VI trajectory onto a grid of cells. Each cell is assigned a binary number (i.e., zero or one). When the VI trajectory passes through a cell, that cell is considered to be occupied by that VI trajectory, is assigned a digital value of 1, and is, for example, filled in black. Fig. 3 shown.

[0040] The binary cell lattice 40 is a generalization of VI trajectories. VI trajectories with similar but not identical shapes can have identical mapped binary cell lattices. This is because two VI trajectories can pass through a cell on different paths, but that cell is still considered occupied and assigned a binary value of 1. The following defines a binary cell lattice mapping algorithm according to the disclosed concept.

[0041] First, the voltage and current data are loaded, assuming there is a total of K data points of the form (v k , i k ) gives where: k=1,…,K; and v k and i k The voltage and current values ​​of sample data point k are.

[0042] Secondly, the maximum and minimum values ​​of the voltage and current waveforms are determined, i.e., v max , v min , imax and i min calculated from: v max = max v k , v min = min v k , i max = max i k , i min = min i k , v0=12(vmax+vmin), and i0=12(imax+imin); where: v0 and i0 are both averages of the corresponding maximum and minimum values ​​of voltage and current, respectively, and form the central points of the cell grid.

[0043] Physically, the v0 and i0 values ​​are the DC bias values ​​of the respective voltage and current waveforms, which are usually determined by the DC offset of the voltage and current sensors (which are not shown, but see Sensors 106 of the Fig. 8) and / or the asymmetry between positive and negative half-cycles of the waveforms is introduced. In an ideal scenario, these DC offsets are relatively small or even close to zero.

[0044] The Fig. 4A-4C show an example with a DC bias voltage (i0) of 0.032 A on the current waveform ( Fig. 4B) and a DC bias voltage (v0) of 0.7 V on the voltage waveform ( Fig. 4A). Fig. Figure 4A shows a graphical representation 42 of the sampled voltage compared to a discrete sample for a specific load. Fig. Figure 4B shows a graphical representation 44 of sampled current versus discrete sampling for the specific load of the Fig. 4A. Fig. Figure 4C shows a graphical representation 46 of the current versus voltage for the VI trajectory of the specific load of the Fig. 4A and includes a point 48 for the v0 and i0 values.

[0045] Thirdly, if a predefined parameter Δ is given, the width (or size) of the grid is defined and calculated by: dv=vmax−v0Δ di=imax−i0Δ and the two sequences are generated: {v0−dv⋅Δ,v0−dv⋅(Δ−1),…,v0,…,v0+dv⋅(Δ−1),v0+dv⋅Δ} and {i0−di⋅Δ,i0−di⋅(Δ−1),…,i0,…,i0+di⋅(Δ−1),i0+di⋅Δ}

[0046] Here, both of these sequences have 2Δ+1 elements.

[0047] Fourthly, a square NxN cell grid is defined, where: N=2Δ+1.

[0048] The (x-th,y-th) cell has a position value (v0 + dv · x,i0 + di · y) and a binary model value B. x,y assigned, which is initialized to be 0.

[0049] Fifthly, a half-line cycle of data points is loaded, as in Fig. 5A-5C is shown. The half-cycle of the voltage waveform 50 ( Fig. 5A) and the half-cycle of the current waveform 52 ( Fig. 5B) begins at the voltage zero crossing point 51V ( Fig.5A) with a positive gradient (i.e., the voltage value passes through zero from a negative value to a positive value) and ends at another voltage zero crossing point of 53V ( Fig. 5A) with a negative gradient (i.e., the voltage value passes through zero from a positive value to a negative value). Similar start and end points 51A, 53A ( Fig. 5B) for the current waveform 52 are in Fig. 5B is shown. The starting points are 51V, 51A, the endpoints are 53V, 53A, and an example current / voltage sampling value (v) is shown. k , i k ) 55 are in the graphical representation 54 of voltage versus current in Fig. 5C shown.

[0050] Sixthly, starting with the first data point 51V, 51A of the data points that were loaded in the previous step, the one with (v1h,i1h) The following loop is indicated:

[0051] In the execution loop above, for each cell (Δ+1, y) for y = Δ+1, Δ+2, ..., 2Δ+1 in the grid, it is determined whether it passes through a specific data point. (v1h,i1h) is occupied. If it is determined that a cell is occupied by this data point, then that cell is designated as the winner for that data point. Once the winning cell is determined, the loop is STOPPED for that data point (which is also known as the end of the loop). If the data point is the first in the data sequence (i.e., the half-cycle of data points from step five), then this step marks the occupied cell as the starting cell.

[0052] As a seventh step, the sixth step is repeated by searching and determining a cell assignment for the remaining half-cycle data points from the fifth step. To speed up the execution process, for example, only the eight adjacent cells of a previous winner are considered for each search loop.

[0053] Eighthly, repetitions from step six should be carried out for a predetermined number (e.g., the number of data points in the half-cycle; ten or one hundred; any suitable number; without being restricted to that).

[0054] The coefficient Δ defines the width of each cell and therefore the number of cells within the cell grid. The grid size should be chosen based on the specific application. If there are too many cells, mapping VI trajectories to the binary cell grid may not effectively handle the variance of similar VI trajectories. Conversely, if the number of cells is insufficient, the mapped binary cell grid may not accurately represent the VI trajectories. Property extraction based on a binary VI cell grid

[0055] In addition to reducing the error introduced by the difference between VI trajectories of PELs within the same load category, mapping VI trajectories to binary cell grids can also reduce the effect of distortion while preserving graphical characteristics. For each category of PELs, the disclosed concept employs a set of novel signatures that can be directly identified from the binary cell grid.

[0056] Fig. 6A and Fig. Figure 6B graphically represents three important points or cells (P1, P2, P3) 56, 58, 60 and four important lines (L1, L2, L3, L4) 62, 64, 66, 68 as properties of a binary cell grid. The following provides an example of a set of eight properties or features that can be used to represent each load category identically: (1) Property 1: the binary value of the left horizontal cell (1,Δ+1), labeled as cell P1 56 in Fig.6A, where the applicable values ​​include: 0 (cell unoccupied) and 1 (cell occupied); (2) Property 2: the binary value of the central cell (Δ+1,Δ+1), labeled as cell P2 58 in Fig. 6A, where the applicable values ​​include: 0 (cell unoccupied) and 1 (cell occupied); (3) Property 3: the multiplication of antidiagonal grid cell values, i.e. the multiplication of the binary values ​​of all cells along the diagonal line (labeled as line L2 64 in Fig. 6A) in the grid from the lower left corner to the upper right corner. This number is also a binary value and indicates whether the VI trajectory is linear, or in other words, whether the VI trajectory is aligned with the diagonal line. The applicable values ​​of this property3 include: 1 (Linear), if the VI trajectory has the form of a straight line from the lower left corner to the upper right corner, as in the examples shown in Fig. 1A are shown; and 0 (non-linear) if at least one of the antidiagonal cells is unoccupied and the VI trajectory is not a straight line, as in the examples shown in the Fig. 1B-1G are shown; (4) Property 4: the number of continua of grid cells with value 1 within all cells (Δ+1,[1:2Δ+1]) that indicates the number of intersection points of the VI trajectory and the basis stress v0 line (labeled as line L1 62 in Fig. 6A); where the designation 1:2Δ+1 denotes all integers from 1 to 2Δ+1; the applicable values ​​include: 1 (one-cell) and 2 (two-cell); (5) Property 5: whether there are any intersections of the VI trajectory with itself; the applicable values ​​include: 0 (None), 1 (One intersection), 2 (Two intersections), and so on; (6) Property 6: the number of intersection points of the VI trajectory with the 1.3v0 line (marked as the line L3 66 in Fig.6A); the applicable values ​​include 1 (one intersection point) and 2 (two intersection points). (7) Property 7: the existence of a central, horizontal line segment (marked as line L4 68 in Fig. 6B): This line L4 occupies 30% of the total horizontal line, where y=0 in the grid; the existence of such a line is determined if 50% of the line overlaps with the part of the VI trajectory; the applicable values ​​include: 0 (no horizontal line) and 1 (with horizontal line); and (8) Property 8: the binary value of the upper middle cell (Δ+1,1), labeled as cell P3 60 in Fig. 6B; the applicable values ​​include: 0 (cell-unoccupied) and 1 (cell-occupied). Number of intersections with itself

[0057] The VI trajectories of some electrical loads intersect themselves, as for example in Fig.1G is shown. An earlier proposal might suggest that the number of self-intersections contained in a VI trajectory could be related to the order of the harmonics. For example, a simulated load with a significant third (or fifth) harmonic component in the current has two (or four) self-intersections. This could also be caused by loads in category M (i.e., loads with multiple independent front-end power supply units). Therefore, the disclosed concept employs a general yet cost-effective algorithm to determine the number of self-intersections contained in a VI trajectory, as shown in Fig. 7 is shown.

[0058] First, a half-cycle (e.g. 1 / 120 second) of the sampled data points [0 - , 0 +] read, starting with the zero-crossing data point from negative voltage values ​​to positive voltage values ​​(labeled 0-) and ending with the zero-crossing data point from the positive voltage values ​​to the negative voltage values ​​(labeled 0 + ).

[0059] Secondly, for each data point j within the range [0-, peak value], + ], where peak value + a data point within [0 - , 0 + ] denoted by the maximum positive voltage value, the data point k has been found whose voltage value is closest to point j.

[0060] Thirdly, a data point j is assigned a voltage value v. j and a current value i jdenoted by a vector j, and check whether the values ​​of the stream of the data point sequence {j-1, j, j+1} and {k-1, k, k+1} are monotonically increasing; if so, proceed to the fourth step below, and if not, repeat this third step starting with j+1.

[0061] Fourthly, check whether the data points k-1= (v k-1 , i k-1 ) and k+1= (v k+1 , i k+1 ) are on opposite sides of the line, which is determined by j-1 = (v j-1 , i j-1 ) and j+1 = (v j+1 , i j+1 ) using the following criterion: {(j+1¯−j−1¯)×(j+1¯−k−1¯)}⋅{(j+1¯−j−1¯)×(j+1¯−k+1¯)}<0 where: x denotes the vector product; and ▪ denotes the scalar product.

[0062] In other words, for any j and k, a case where the criterion of the fourth step is met is considered to be an intersection with itself. Numerical test results

[0063] The disclosed concept may be used in combination with the teachings of any or all of the following documents: (1) U.S. Patent Application Publication No. 2013 / 0138651, entitled “System and Method Employing a Self-Organizing Map Load Feature Database to Identify Electric Load Types of Different Electric Loads”; (2) U.S. Patent Application Publication No. 2013 / 0138661, entitled “System and Method Employing a Minimum Distance and a Load Feature Database to Identify Electric Load Types of Different Electric Loads”; and (3) U.S. Patent Application Serial No. 13 / 597,324, filed on August 29, 2012, entitled “System and Method for Electric Load Identification and Classification Employing Support Vector Machine”, all of which are incorporated herein by reference.

[0064] According to the teachings of the disclosed concept, the resulting binary VI properties extracted from the depicted cell grid of binary values ​​can be used as inputs to any or all classification and identification systems disclosed in the three patent applications above to derive the category of load under consideration. With reference to the hierarchical load identification architecture disclosed in Publication No. 2013 / 0138669, the disclosed concept can be applied to provide the properties required by Level 1 load category identification. Load categorization can be performed by applying a supervised self-organizing map (SSOM) or a self-organizing map (SOM).SOFM (SOFM = self-organizing feature map)), which is a type of unsupervised artificial neural network trained using competitive learning to produce a discretized representation of relatively small dimension (typically two-dimensional) the input space of training examples, referred to as a map, as disclosed in Publication No. 2013 / 0138651. Tests regarding five main load categories

[0065] Five of the load categories (i.e., R, X, NP, P, and M) cover the majority of existing PELs. The following discusses the success rate of identifying loads from these five load categories using the first five properties disclosed herein. It is expected that the proposed graphical signatures from the binary mapping of VI trajectories for these five categories of PELs will have values ​​(where "X" represents either 0 or 1), as shown in Table 2. Table 2 category Property 1 Property 2 Property 3 Property 4 Property 5 R 0 1 1 1 0 X 0 0 0 2 0 NP 1 1 0 1 0 P 0 1 0 1 0 M 0 X 0 2 1 or more

[0066] For each category, a number of PELs are tested, and each PEL is tested independently 100 times. The results are shown in Table 3. Table 3 category Total number of loads Total number of tests Total number of correct results Success rate R 6 600 597 99,5% X 10 1000 991 99,1% NP 15 1500 1493 99,5% P 11 1100 1091 99,2% M 4 400 395 98,8%

[0067] In summary, the proposed graphical signatures derived from a binary mapping of VI trajectories achieve an average accuracy rate of over 99%. The identification of loads from category M (i.e., multiple independent front-end power supply units) has the lowest accuracy in Table 3. This is primarily due to the wide diversity of loads in this category. Tests regarding all seven load categories

[0068] This test considers all seven load categories. The proposed graphical signatures, derived from a binary mapping of VI trajectories, are expected to have values ​​for the seven PEL categories as shown in Table 4. Table 4 category Property 1 Property 2 Property 3 Property 4 Property 5 Property 6 Property 7 Property 8 R 0 1 1 1 0 1 0 0 X 0 0 0 2 0 2 0 0 NP 1 1 0 1 0 1 1 0 P 0 1 X 1 0 1 1 0 M 0 X 0 2 1 or more 2 0 0 T 0 0 0 2 0 2 0 1 PAC X 1 1 1 0 2 1 0

[0069] In this test, a total of 20 load types (with one to seven load models for each load type) were tested. For each dataset, approximately 900 to 3000 VI trajectories were selected and mapped onto a 64x64 cell grid. The results are shown in Table 5. Table 5 Target load category load type Complete models Total number of tests Success rate (%) NP Battery charger 1 3000 83,4 DVD player 4 3000 100 Desktop computer 2 3000 99,8 LCD monitor 7 3000 99,5 Printer 1 3000 99,9 Electronic circuit board 1 3000 98,7 P LCD television 8 3000 98,5 LED TV 3 3000 99,2 Plasma TV 2 3000 99 Multifunctional device 3 3000 93 projector 4 3000 99,9 Complex M microwave oven 4 1800 99 R space heater 4 1800 93 coffee machine 2 1800 98 lightbulb 4 1800 99,2 Electric frying pan 2 1800 98,6 T stapler 1 1800 98,9 adapter 5 1800 100 X fan 5 3600 98,5 Refrigerator 4 3600 100 Water dispenser 1 3600 100 Paper shredder 2 3600 65 PAC Incandescent light bulb with dimmer 1 1800 50

[0070] The test results validate that the proposed graphical signatures derived from the binary mapping of the VI trajectories can achieve an average accuracy rate of over 90% with a relatively large load set and seven target load categories. The main errors originate from some PAC loads where the phase angle is less than 90°, making the load characteristics similar to those expected for resistive loads, thus incorrectly categorizing them as if they were in the R category. Increasing the sampling rate of the sensed voltage and current signals could help improve performance, although a trade-off should be considered regarding memory availability and computational load.At the same time, from an application point of view, if an incandescent lamp with a dimmer with a relatively small phase angle is identified as a resistive load, then the resulting categorization will still be acceptable. Summary

[0071] Fig.Figure 8 shows a system 100 for different electrical consumers or loads 102, 103, 104. The system includes sensors 106 configured to detect voltage and current signals 107 for each of the different electrical loads 102, 103, 104; a hierarchical load attribute database 108 with a plurality of levels or layers (L1, L2, L3) 110, wherein a first layer 112 (L1) of the layers 110 contains a plurality of different load categories; and a processor 114. The processor 114 includes a routine 116 which, according to the teachings of the disclosed concept, acquires voltage and current waveforms from the sensors 106 for a corresponding one of the different electrical loads 102, 103, 104. maps a voltage-current trajectory onto a grid containing a multitude of cells, each assigned a binary value of zero or one (see, for example, Fig.3); extracts a variety of different properties from the depicted grid of cells as a graphic signature of the corresponding one of the different electrical loads 102, 103, 104; derives a category of the corresponding one of the different electrical loads 102, 103, 104 from the database 108; and identifies one of a variety of different electrical load types for the corresponding one of the different electrical loads 102, 103, 104.

[0072] The main advantages of the proposed binary VI property reduction include reducing harmonic and noise effects on load current and voltage waveforms, providing a relatively simple abstraction of graphical forms of trajectories, and simplifying graphical property extraction.

[0073] The binary VI properties are relatively easy to compute and require less memory, since the property values ​​are all integers. The initial computation and memory requirements were evaluated, and the results show that the computational effort for calculating the graphical features and the memory requirement are on the order of x% of what is required by a fast Fourier transform (FFT).

[0074] The disclosed concept employs a relatively low-cost, yet accurate, method and system for extracting electrical load identification signatures. Instead of using digital signal processing and frequency domain analysis, the disclosed concept leverages the similarity of VI trajectories between loads and maps these trajectories onto a cell grid with binary cell values. Graphical properties are then extracted for numerous applications.

[0075] The revealed concept significantly reduces computational effort compared to existing frequency domain property extraction and analysis techniques. Test results show that an average success rate of over 99% can be achieved using the proposed signatures.

[0076] While specific embodiments of the disclosed concept have been described in detail, it will be clear to the person skilled in the art that various modifications and alternatives to those details could be developed with regard to the entire teachings of the disclosure. Accordingly, the specific arrangements disclosed are intended to be illustrative only and are not intended to limit the scope of the disclosed concept, which is to be granted the full breadth of the appended claims and any and all equivalent embodiments thereof. 36325

Claims

[1] A system (100) for identifying electrical load types for a variety of different electrical loads (102, 103, 104), wherein the system comprises: a hierarchical load attribute database (108) comprising a plurality of layers (110), wherein a (112) of the layers contains a plurality of different load categories, wherein a first layer (112, L1) of the plurality of layers contains the different load categories; wherein a second layer (L2) of the plurality of layers contains a plurality of different load subcategories for each of the different load categories; and wherein a third layer (L3) of the plurality of layers contains the different electrical load types for the different load subcategories; and a processor (114) which is constructed (116) to: Acquiring a voltage waveform and a current waveform for a corresponding one of the different electrical loads; Mapping a voltage-current trajectory onto a grid containing a plurality of cells, each of which has a binary value of zero or One is assigned based on whether the voltage-current trajectory passes through each of the cells; Extracting a variety of different properties from the depicted grid of cells as a graphical signature of the corresponding one of the different electrical loads; Deriving a category of the corresponding one of the different electrical loads from the hierarchical load attribute database; and Identifying one of a variety of different electrical load types for the corresponding one of the different electrical loads, where the voltage waveform and the current waveform each represent a total number of K data points of the form (v k , i k ) contained, where: k = 1,...,K; where v k and i k correspondingly, a voltage and a current value of a sampled data point k; where maximum and minimum values ​​of the voltage waveform and the current waveform are calculated from: v max = max v k , V min = min v k , i max = max i k , i min = min i k , v0=12(vmax+vmin),and i0=12(imax+imin); where v0 and i0 are both averages of the corresponding maximum and minimum values ​​that form the central points of the grid of cells, where Δ defines a size of the grid; where dv=vmax−v0Δ di=imax−i0Δ; the processor is further configured to generate two sequences from the voltage waveform and the current waveform as: {v0−dv⋅Δ,v0−dv⋅(Δ−1),…,v0,…,v0+dv⋅(Δ−1),v0+dv⋅Δ} and {i0−di⋅Δ,i0−di⋅(Δ−1),…,i0,…,i0+di⋅(Δ−1),i0+di⋅Δ} where each of the two sequences has N = 2Δ+1 elements; where the grid of cells includes a first axis having N cells and a second axis having N cells; and where each of the cells has a position value (v0 + dv · x,i0 + di · y) and a binary model value B x,y is assigned, which is initialized to 0. [2] System (100) according to claim 1 or method according to claim 7, wherein the different load categories include resistive loads, reactive predominant loads, electronic loads with a power factor correction circuit, electronic loads without a power factor correction circuit, electronic loads which include a linear power supply using a transformer to amplify the voltage, phase angle controlled loads and complex structures. [3] System (100) according to claim 1, wherein the processor is further configured to map a half-cycle of the voltage waveform and the current waveform onto the grid of the cells and to assign each of the K data points to a corresponding one of the cells with the binary model value B x,y from 1. [4] System (100) according to claim 1, wherein the processor is further configured to determine a number of intersection points with itself that are contained in the voltage-current trajectory shown. [5] System (100) according to claim 1 or method according to claim 11, wherein the category of the corresponding one of the different electrical load types is derived from a supervised self-organizing map, SSOM. [6] System (100) according to claim 1 or method according to claim 7, wherein the category of the corresponding one of the different electrical load types is derived from a self-organizing map, SOM, (SOM = self-organizing map) or a self-organizing feature map, SOFM, (SOFM = self-organizing feature map) which are trained using competitive learning. [7] A method for identifying load types for a variety of different electrical loads (102, 103, 104) wherein the method comprises: Sensing (106) a voltage signal (107) and a current signal (107) for each of the different electrical loads; providing a processor; Provide, with the processor, a hierarchical load attribute database (108) having a plurality of layers (110), wherein one (112) of the layers has a plurality of different load categories, wherein a first layer (112, L1) of the plurality of layers contains the different load categories; wherein a second layer (L2) of the plurality of layers contains a plurality of different load subcategories for each of the different load categories; and wherein a third layer (L3) of the plurality of layers contains the different electrical load types for the different load subcategories; Acquire, with the processor, a voltage waveform and a current waveform for a corresponding one of the different electrical loads; Mapping, using the processor, a voltage-current trajectory onto a grid containing a plurality of cells, where each of the cells is assigned a binary value of zero or one based on whether the voltage-current trajectory passes through each of the cells; Extracting, with the processor, a variety of different properties from the depicted grid of cells as a graphical signature of the corresponding one of the different electrical loads; Derive, using the processor, a category of the corresponding one of the different electrical loads from the hierarchical load attribute database; and Identifying one of a variety of different electrical load types for the corresponding one of the different electrical loads; Including a total of K data points of the form (v k , i k ) with both the voltage waveform and the current waveform, where: k = 1,...,K; where v k and i k correspondingly, a voltage value and a current value of a sampled data point k; where maximum and minimum values ​​of the voltage waveform and the current waveform are calculated from: v max = max v k , V min = min v k , i max = max i k , i min = min i k , v0=12(vmax+vmin), and i0=12(imax+imin); where v0 and i0 are both averages of the corresponding maximum and minimum values ​​that form the central points of the grid of cells, where Δ defines a size of the grid; where dv=vmax−v0Δ di=imax−i0Δ; and Generating two sequences from the voltage waveform and the current waveform as: {v0−dv⋅Δ,v0−dv⋅(Δ−1),…,v0,…,v0+dv⋅(Δ−1),v0+dv⋅Δ} and {i0−di⋅Δ,i0−di⋅(Δ−1),…,i0,…,i0+di⋅(Δ−1),i0+di⋅Δ} where each of the two sequences has N = 2Δ+1 elements; where the grid of cells includes a first axis having N of the cells and a second axis having N of the cells; and where each of the cells has a position value (v0 + dv · x,i0 + di · y) and a binary model value B x,y is assigned, which is initialized to 0. [8] The method of claim 7, further comprising: Mapping one half-cycle of the voltage waveform and the current waveform onto the grid of the cells; and Assign each of the K data points to a corresponding cell with the binary model value B. x.y from 1. [9] The method of claim 7, further comprising: determining a number of intersection points with itself that are contained in the voltage-current trajectory shown.

Images