JP5546506B2 - ELECTRIC DEVICE IDENTIFICATION DEVICE, ELECTRIC DEVICE IDENTIFICATION METHOD, AND ELECTRIC DEVICE IDENTIFICATION PROGRAM - Google Patents

Description

  The present invention relates to an electric device identification device, an electric device identification method, and an electric device identification program, and in particular, identifies an electric device in operation based on an observation result of a total value of electric consumption of a plurality of electric devices. The present invention relates to a possible electric device identification device, an electric device identification method, and an electric device identification program.

  In recent years, energy management system HEMS (Home Energy Management System) technology has been studied and developed to install electric power sensors in the homes of electric power consumers such as general households and offices, and to achieve efficient use of electric power and remote control of electrical equipment. Progressing.

  In general, in the energy management system HEMS, necessary information is collected by collecting and transferring information from power sensors attached to each electrical device such as home appliances to a center device using wired or wireless communication means. The method of collecting is considered.

  On the other hand, M. et al. As described in Katsukura et al. “Life Pattern Sensor with Non-intrusive Appliance Monitoring” (ICCE '09 pp.1-2, Jan.2009), etc. Using only one sensor (power sensor or current sensor) installed at the position of the line or distribution board, the total power consumption and current waveform of each electrical device are monitored and used based on their characteristics. There has also been proposed a method for identifying and grasping the type, operating state, power consumption, and the like of each electric device.

  In the case of the method described in Non-Patent Document 1, since only one sensor is required, the cost is high, and a new sensor is added to an existing electric product, or a new electric Since it is not necessary to embed a sensor with a uniform standard in the product, it is considered to be a practical and effective means of implementation because there are few barriers to introduction.

  However, the information obtained from one sensor is limited to data related to the accumulated current and voltage values, and it is not possible to directly grasp the power consumption status of each electrical device. It is necessary to infer power data of individual electrical equipment from the collected data.

  Therefore, as a method for identifying electrical devices and estimating the respective operating states, in Non-Patent Document 1, data such as power consumption and current waveforms of individual electrical devices are previously stored in various operating states. In advance, the data is stored as a database, and the data of each electrical device stored in advance as a database is compared with the actual measurement value to identify the type of each electrical device and estimate the operating state. The method is considered.

  However, the method of comparing and comparing the data of each electrical device stored in advance as a database with the actual measurement value causes a combined explosion when the number of electrical devices that are assumed to be connected to one sensor increases. There is a problem that the amount is enormous and the calculation does not end in a realistic time.

Katsukura et al. “Life Pattern Sensor with Non-intrusive Appliance Monitoring,” ICCE '09 pp.1-2, Jan.2009.

  As described above, in the conventional method for identifying an electric device as described in Non-Patent Document 1, when performing comparison and collation between data of each electric device stored in advance as a database and an actual measurement value, 1 As the number of electrical devices that are assumed to be connected to one sensor increases, a combination explosion occurs, the amount of calculation becomes enormous, and the calculation does not end within a realistic time, so it cannot be actually used. There's a problem.

  The present invention has been made in view of such circumstances, and it is possible to quickly and accurately identify an operating electrical device regardless of the number of electrical devices connected to one sensor. An object of the present invention is to provide an electrical device identification device, an electrical device identification method, and an electrical device identification program.

  The present invention comprises the following technical means in order to solve the above-mentioned problems.

  The first technical means is based on a result of observing a total value of power consumption of a plurality of electric appliances installed in a home or office of a power consumer with a single sensor. Of these, an electric device identification device for identifying an electric device in an operating state, and measuring a total power consumption vector amount as a total power consumption vector amount measured by a single sensor. A measuring unit, a data storage unit that preliminarily stores the amount of power consumed when each of the electrical devices operates independently as a single power consumption vector amount, and the operating electrical device among the plurality of electrical devices is unknown The total power consumption vector amount measured by the power measurement unit in the state, and the single power consumption vector amount of each of the electric devices stored in the data storage unit in advance Then, by performing a non-negative matrix factorization (NMF) on a non-negative matrix factorization (NMF) with respect to the non-negative matrix synthesized by the measurement power synthesis unit synthesized as one non-negative matrix , An NMF operation unit that decomposes into a vector product of a non-negative basis vector matrix and a non-negative value weight matrix, and based on a result of performing the non-negative value matrix decomposition in the NMF operation unit, An electrical device extraction unit that extracts an electrical device that is in an operating state when the total power consumption vector amount is measured by the power measurement unit in an unknown state of the operating electrical device. It is characterized by being.

  According to a second technical means, in the electrical equipment identification device according to the first technical means, a power spectrum of a current supplied to the corresponding electrical equipment as the total power consumption vector quantity and the single power consumption vector quantity. Alternatively, only the half period of the positive part of the current value itself of the current supplied to the electric device is used.

  The third technical means is the electrical device identification device according to the first or second technical means, wherein when the NMF computing unit decomposes the non-negative matrix, the base vector matrix is used as the base vector matrix. Each of the electric devices is configured so that each arbitrary column vector composed of the single power consumption vector amount among the column vectors of the non-negative matrix can be substantially expressed only by any specific basis vector of the basis vector matrix. A basis vector composed of the single power consumption vector amount is obtained, and a weight element corresponding to the electric device in an operating state in each arbitrary column vector of the non-negative matrix to be decomposed as each weight element of the weight matrix Only, a significant value in the vicinity of '1' is set, and each arbitrary column vector of the non-negative matrix to be decomposed is set. For the weight factor corresponding to the electrical equipment that is not in the state during operation Te, wherein the value nearly "0" is set.

  According to a fourth technical means, based on a result of observing a total value of power consumption of a plurality of electric devices installed in a home or office of a power consumer with one sensor, the plurality of electric devices An electrical device identification method for identifying an electrical device in an operating state, wherein the power consumption amount when each of the electrical devices operates independently is stored in advance in the data storage unit as an individual power consumption vector amount, A total power consumption vector amount, which is a total value of power consumption amounts of the plurality of electrical devices observed by one sensor in an unknown state of an operating electrical device among the plurality of electrical devices, is stored in the data storage. Combined with the single power consumption vector amount of each of the electrical devices stored in advance in the unit, synthesized as a single non-negative matrix, then non-negative with respect to the synthesized non-negative matrix By performing non-negative matrix factorization (NMF), the vector product of a non-negative basis vector matrix and a non-negative weight matrix is decomposed, and the result of the non-negative matrix decomposition decomposed into the vector product is obtained. On the basis of this, it is characterized in that an electrical device that is in an operating state when the total power consumption vector amount is observed when an operating electrical device is unknown among a plurality of the electrical devices is extracted.

  According to a fifth technical means, in the electric device identification method according to the fourth technical means, a power spectrum of a current supplied to the corresponding electric device as the single power consumption vector amount and the total power consumption vector amount. Alternatively, only the half period of the positive part of the current value itself of the current supplied to the electric device is used.

  Sixth technical means, in the electrical device identification method according to the fourth or fifth technical means, when decomposing the non-negative matrix, as a basis vector matrix, a column of the non-negative matrix to be decomposed The single power consumption vector quantity of each of the electric devices so that each arbitrary column vector consisting of the single power consumption vector quantity of the vectors can be substantially expressed only by any specific base vector of the base vector matrix. Only for the weight element corresponding to the electric device in the operating state in each column vector of the non-negative matrix to be decomposed as each weight element of the weight matrix, '1' Significant values in the vicinity are set, and each column vector of the non-negative matrix to be decomposed is in an active state. For the weight factor corresponding to the non the electric device, characterized in that a value substantially close to '0' is set.

  A seventh technical means is the electric equipment identification method according to the sixth technical means, wherein a column vector of the weight matrix corresponding to a column vector composed of the total power consumption vector amount in the non-negative matrix to be decomposed is calculated. Among the weighting elements to be configured, the electrical equipment corresponding to the weighting element having a significant value in the vicinity of “1” is the total of the electrical equipment that is operating among the plurality of electrical equipments in an unknown state. When the power consumption vector amount is observed, it is identified as an electric device in an operating state.

  An eighth technical means is an electrical equipment identification program that implements the electrical equipment identification method according to any of the fourth to seventh technical means as a program executable by a computer. .

  According to the electrical device identification device, the electrical device identification method, and the electrical device identification program of the present invention, when there are a plurality of electrical devices in the home or office of a general household of power consumers, the power supply line that supplies power The total power consumption, current, and voltage of multiple electrical devices are observed by a single sensor installed at a position such as a lead-in wire or a power line in a distribution board, and the operation is based on the observed results. In contrast to the conventional method of matching each feature amount of each electrical device by a process such as a neural network as a method for identifying the electrical device in the present invention, in the present invention, in the present invention, a multivariate analysis called non-negative matrix decomposition NMF is performed. Since this method is adopted, the following effects can be obtained.

  First, the computational cost required to identify each electrical device can be significantly reduced, and the currently operating electrical device can be grasped quickly and reliably regardless of the number of installed electrical devices. Therefore, it can be suitably used as an electric device identification device (monitor device) for grasping the type, operation status, power consumption, and the like of each electric device used by an electric power consumer.

  Secondly, by grasping the type and operation status of each electric appliance used by the electric power consumer and the power consumption by the electric appliance identification device (monitor device) installed in the home or office of the electric power consumer, It is also possible to transfer the grasped information to the center device etc. via the communication network. Even in a remote place where the center device etc. is located, Power control and the like based on the operation status can be reliably performed.

  Third, by introducing an electrical equipment identification device (monitoring device) like this, it is possible to provide information services for effectively reducing power consumption in the homes and offices of power consumers, It is also possible to provide grasping / notification services and monitoring services for elderly people who need to live alone or those who need care for electric power consumers.

It is a block block diagram which shows an example of the block configuration of the electric equipment identification device by this invention. It is explanatory drawing for demonstrating the concept of nonnegative matrix decomposition | disassembly NMF. It is explanatory drawing for demonstrating in more detail the concept of nonnegative matrix decomposition | disassembly NMF. It is explanatory drawing explaining the specific example of the mode of the nonnegative matrix decomposition | disassembly NMF in the NMF calculating part of the electric equipment identification device shown in FIG. It is explanatory drawing which generalizes and demonstrates the mode of the nonnegative matrix decomposition | disassembly NMF in the NMF calculating part of the electric equipment identification device shown in FIG. It is explanatory drawing for demonstrating an example of the specific implementation result of the identification process of the electric equipment by the electric equipment identification device of FIG.

  Hereinafter, an example of a preferred embodiment of an electrical equipment identification device, electrical equipment identification method, and electrical equipment identification program according to the present invention will be described in detail with reference to the drawings. In the following description, the electric device identification device and the electric device identification method according to the present invention will be described. However, the electric device identification method may be implemented as an electric device identification program that can be executed by a computer. Furthermore, it goes without saying that such an electric device identification program may be recorded on a computer-readable recording medium.

(Features of the present invention)
Prior to the description of the embodiments of the present invention, an outline of the features of the present invention will be described first. In general, when observing the power consumption of each of a plurality of electrical devices, the simplest method is to install a power sensor or the like for each electrical device and observe the power consumption for each electrical device. good. However, in such a method, since it is necessary to mount power sensors etc. for each electric device, it is necessary to prepare sensors for the number of electric devices, and the cost for each sensor is required. become.

  Therefore, in order to reduce costs, only one sensor is installed on the distribution board of the building of the power consumer, and the total power consumption of all the electrical equipment that is supplied with power from the distribution board etc. It is effective to measure this. However, in this case, although the cost of the sensor is only required for one unit, it becomes impossible to directly grasp the type and power consumption state for each electric device.

  Therefore, as described in Non-Patent Document 1, data such as power consumption and current waveform for each electrical device is measured in advance under various conditions, and a technique such as a neural network is used for each electrical device. The data indicating the feature value stored in the database is obtained by storing the data indicating the feature value of the sensor and storing it as a database, and the measured values measured by one sensor in an unknown state of the operating electrical equipment. A technique has been proposed in which the type of the electric device in operation is identified and the power consumption state is estimated by performing comparison and comparison one by one.

However, in the case of the electrical device identification method according to Non-Patent Document 1 as described above, as described above, as the number of electrical devices connected to one sensor increases, the database rapidly grows and the calculation amount increases. There is a problem that will explode rapidly. For example, when the number of electrical devices is N (N: a positive integer), all the data when each electrical device takes two states of power ON / OFF is prepared as a database in a permutation combination beforehand. A database with a data amount of ^2N is required, and as the number N increases, the calculation amount required for comparison with the actual measurement value increases explosively.

  In order to solve such a situation, the present invention provides a non-negative matrix decomposition NMF (Non-negative) in order to infer the operating state of each electric device from the total value of power consumption measured by one sensor. By using a multivariate analysis technique called Matrix Factorization, the main feature is to simultaneously reduce the size of the database and reduce the amount of calculation.

Specifically, the present invention provides:
(1) State of power consumption at the time of individual operation of each electric device (that is, corresponding to the above-mentioned database, and prepared in advance as known data on the amount of power consumption vector for each electric device at the time of single operation ) And the state of total power consumption observed by one sensor (corresponding to the total power consumption vector amount in the state where one or more electric devices are operating at the same time, and the operating electric device is unknown) Are combined together as one non-negative matrix X (that is, a matrix in which each element has a value of 0 or more).

  (2) The synthesized non-negative matrix X is decomposed into a vector product of a non-negative basis vector matrix H and a non-negative value weight matrix U by non-negative matrix decomposition NMF.

(3) By using the decomposed non-negative weight matrix U, a feature indicating the operating state of each electrical device when the simultaneously operating electrical device is observed in an unknown state is extracted, thereby operating the electrical device in operation. Identify.
It is characterized by consisting of

  In other words, the present invention refers to a non-negative matrix decomposition NMF that can be realized only by repeating four arithmetic operations for the process of identifying the characteristics of an electric device that is operating simultaneously, which has been conventionally realized by a sequential matching process such as a neural network. By using the multivariate analysis method, it is possible to greatly reduce the calculation cost required for identifying the electric device, and even if many electric devices are connected to one sensor, The effect is obtained that the operating state of the device can be grasped quickly and reliably.

(Configuration example of embodiment)
Next, an example of a block configuration of the electrical equipment identification device according to the present invention will be described in detail with reference to FIG. FIG. 1 is a block configuration diagram showing an example of a block configuration of an electrical equipment identification device according to the present invention for supplying power to a plurality of electrical equipment installed in the home of a power consumer such as a home or office. The example of a structure of the electric equipment identification device arrange | positioned in positions, such as a distribution board, is shown.

  1 includes at least an electric device identification unit 10, an input unit 20, and an output unit 30, and as described above, the electric power from the power source 200 is installed in the house. It arrange | positions in positions, such as a distribution board for supplying to each of the some electric equipment 300. FIG. Here, the plurality of electric devices 300 are configured from general home appliances such as an air conditioner, a lighting fixture, a TV receiver, a refrigerator, a washing machine, a personal computer, and a printer, and office equipment. As long as the electrical control is performed, a device using an energy source other than electric power, such as a fan heater device, is also included.

  The electric device identification unit 10 of the electric device identification apparatus 100 is a part constituting the core of the electric device identification apparatus 100, and includes a power measurement unit 11, a measured power synthesis unit 12, an NMF calculation unit 13, and an electric device extraction unit 14. And the current operation based on the measurement result of the total power consumption of the whole of the plurality of electrical devices 300 measured by only one sensor provided with at least the data storage unit 15 and installed at a position such as a distribution board This is a part for identifying the electric device 300 therein.

  The power measurement unit 11 is a part that measures the result of observing the total power consumption amount supplied from the power source 200 to each of the plurality of electrical devices 300 by one sensor as the total power consumption vector amount, The unit 12 includes a total power consumption vector amount that is a total value of power consumption amounts measured by the power measurement unit 11 in a state where the operating electric device 300 among the plurality of electric devices 300 is unknown, and the data storage unit 15. Is a portion that combines the single power consumption vector amount that is the power consumption amount during the single operation of each electrical device 300 accumulated in advance into one non-negative matrix X.

  Also, the NMF calculation unit 13 performs non-negative matrix decomposition NMF on the non-negative matrix X synthesized by the measured power synthesis unit 12 to obtain a vector of a non-negative basis vector matrix H and a non-negative value weight matrix U. The electrical device extraction unit 14 is an unknown device that is operating among the plurality of electrical devices 300 based on the result of the non-negative matrix decomposition NMF performed by the NMF calculation unit 13. This is a part for extracting the electric device 300 that is in the operating state when the total power consumption vector amount, which is the total value of the power consumption amount, is observed by the power measurement unit 11 in the state.

  The data storage unit 15 measures in advance the power consumption amount when each of the plurality of electric devices 300 operates independently as a single power consumption vector amount in the form of a power vector amount, and the electric device at the time of single operation It is stored as known data indicating the amount of power vector, which is the amount of power consumed for each of 300. On the other hand, as the non-negative matrix X, the total power consumption vector amount of the total power consumption amount synthesized by the measured power combining unit 12 with the known data indicating the single power consumption vector amount relating to each of the electric devices 300 during the single operation. (Total power consumption measured by the power measurement unit 11 when the electric devices 300 operating at the same time are unknown) means that any one or a plurality of electric devices 300 among all the electric devices 300 operate simultaneously. Is the amount of power consumed in

  The input unit 20 inputs data to be set and registered in the data storage unit 15 in advance, or various service information transmitted from a remote center device via a communication network (for power saving for power consumers) Or information regarding remote control or failure diagnosis of the electric device 300). In addition, the output unit 30 outputs information on the identification result of each individual electric device output from the electric device identification unit 10 and the power consumption amount to the output device in order to notify the power consumer, This is a part that transmits to a center device at a remote location via a communication network.

(Operation example of embodiment)
Next, an example of operation | movement of the electric equipment identification device 100 shown in FIG. 1 is demonstrated.

  First, a method of multivariate analysis called non-negative matrix factorization NMF (Non-negative Matrix Factorization) used in the NMF calculation unit 13 of the electrical device identification apparatus 100 shown in FIG. 1 will be described. In the present invention, as described above, the non-negative matrix decomposition NMF, which is one of the multivariate analysis methods, is adopted as the electric device identification method, so that the total power consumption observed by one sensor is obtained. By estimating the operation state of each individual electric device from the value, the scale of the database stored in the data storage unit 15 in advance and the calculation in the NMF calculation unit 13 and the electric device extraction unit 14 for identifying the electric device in operation are calculated. The amount can be reduced.

  Non-negative matrix decomposition NMF is a multivariate analysis technique that makes it possible to generate meaningful patterns and themes by appropriately combining multiple attributes when there are multiple attributes with low predictability. In particular, it has been widely used as a method for extracting characteristic patterns from non-negative data such as image signal and sound signal spectra, and has been widely applied to image feature extraction and sound source separation. Has been.

  Here, the non-negative matrix is a matrix in which all elements for configuring the matrix are composed of non-negative (0 or more) values, and the non-negative matrix decomposition NMF is all non-negative data that is observation data. Is decomposed into a non-negative basis vector matrix and a non-negative weight matrix constituting an appropriate basis set so that can be approximated by a non-negative combination of non-negative basis vectors.

  That is, the non-negative matrix decomposition NMF includes all observation data (for example, N L-dimensional non-negative vectors).

を、それぞれの基底系を構成するＭ個のＬ次元非負値基底ベクトル

, M L-dimensional non-negative basis vectors constituting each basis set

の非負結合ｙ_ｎ（１≦ｎ≦Ｎ）

Non-negative bond y _n (1 ≦ n ≦ N)

によって良く近似できるような適当な基底系ｈ_ｉ（１≦ｉ≦Ｍ）とその重みｕ_ｉ，ｎとに分解するものである。

Is decomposed into an appropriate basis set h _i (1 ≦ i ≦ M) and its weights u _{i, n} that can be approximated by

  That is, a non-negative matrix X in which N L-dimensional non-negative vectors to be decomposed are arranged is

とし、Ｍ個のＬ次元非負値基底ベクトルを並べた基底ベクトル行列Ｈを

And a basis vector matrix H in which M L-dimensional non-negative basis vectors are arranged

とし、非負結合の係数つまり非負結合の重みを要素とした重み行列Ｕを

And a weight matrix U whose elements are non-negative coupling coefficients, that is, non-negative coupling weights.

とし、分解時における誤差を示す誤差行列をＤとした場合、非負値行列分解ＮＭＦとは、

And an error matrix indicating an error at the time of decomposition is D, the non-negative matrix decomposition NMF is:

の関係が成立するような、非負値の基底ベクトル群からなる基底ベクトル行列Ｈと、非負値行列Ｘの各列ベクトルを近似するために非負結合時における基底ベクトル行列Ｈの各基底ベクトルに関する重み付けを行う重み行列Ｕと、を求めるものである。

In order to approximate each column vector of the non-negative matrix X and a base vector matrix H composed of a group of non-negative base vectors so that the relationship of A weight matrix U to be obtained is obtained.

  Here, the error matrix D is made as small as possible,

の関係が成立するためには、例えば、中鹿亘他による「確率スペクトルを用いた基底生成モデルとＮＦＭによる混合楽音解析」（日本音響学会講演論文集、２０１１年３月）等にも記載されているように、一般に、教師なし非負値行列分解ＮＭＦを実施する場合においては、二乗誤差基準によって各行列要素の更新を行う。すなわち、式（９）の条件下において、次の式（１０）の二乗誤差を最小化するような基底ベクトル行列Ｈ、重み行列Ｕを求める。

For example, Watanabe Nakaga et al., “A Basis Generation Model Using Probability Spectrum and Mixed Music Analysis Using NFM” (The Acoustical Society of Japan Proceedings, March 2011), etc. In general, when unsupervised non-negative matrix decomposition NMF is performed, each matrix element is updated according to a square error criterion. That is, a basis vector matrix H and a weight matrix U that minimize the square error of the following equation (10) are obtained under the condition of equation (9).

このための各行列要素の更新式は、Ｈ^Ｔを基底ベクトル行列Ｈの転置行列、Ｕ^Ｔを重み行列Ｕの転置行列とすると、次の式（１１）、式（１２）として与えられる。

Update equation of each matrix element for this is the transposed matrix of basis vectors matrix H and H ^T, when the transposed matrix of weight matrix U and U ^T, the following equation (11) is given as an expression (12).

式（１１）、式（１２）の四則演算を繰り返すことによって、式（８）の関係が近似的に成立する基底ベクトル行列Ｈ（および重み行列Ｕ）を求めることができる。

By repeating the four arithmetic operations of Equation (11) and Equation (12), a basis vector matrix H (and weight matrix U) in which the relationship of Equation (8) is approximately established can be obtained.

  As described above, the non-negative matrix decomposition NMF includes weighting elements that perform weighting on each base vector of the base vector matrix H characterizing each column vector of the non-negative value matrix X to be decomposed and the base vector matrix H at the time of non-negative coupling. This is an algorithm for approximation as a vector product of two non-negative matrixes with a weight matrix U.

  Next, a mechanism for identifying an operating electrical device by applying a non-negative matrix decomposition NMF calculation to the electrical device identification apparatus according to the present invention will be described.

  As described above, the non-negative matrix decomposition NMF is a technique that has been conventionally applied to perform transcription in an acoustic signal or the like. In the present invention, the non-negative matrix decomposition NMF is a multivariate of the non-negative matrix decomposition NMF as described above. Applying the analysis method to the electric current signal (current power spectrum) flowing through the electric device as a non-negative electric signal indicating the electric energy consumption of the electric device, and so on to identify the electric device Yes. In the following description, a case will be described in which a power spectrum of a current from which a non-negative value is obtained is used as a power consumption vector amount indicating a power consumption amount of an electrical device.

When applying the non-negative matrix decomposition NMF technique to the identification of electrical equipment, as shown in FIG. 2, the non-negative matrix X to be decomposed is subjected to various operating conditions as observation data by a single sensor in the power measurement unit 11. A matrix of a column vector group consisting of each column vector of a current power spectrum indicating the power consumption data measured in the above, a matrix consisting of a base vector group having the basis vector matrix H as a column vector with features relating to the power consumption of each of a plurality of electrical devices, When the weight matrix U is viewed as a matrix composed of weight elements u _{i, j} that perform weighting on each base vector of the base vector matrix H at the time of non-negative coupling so as to approximate each column vector of the non-negative matrix X to be decomposed, It can be interpreted as follows.

  That is, for example, a non-negative matrix X composed of observation data obtained by measuring a power spectrum of a current flowing through the electric device 300 is used as a characteristic characterizing the operating state of each electric device 300. Can be interpreted as being approximated as a linear combination of non-negative values of the basis vector matrix H composed of the measured basis vectors.

That is, for example, when focusing on any certain column vector x _i of non-negative matrix X of decomposed, as shown in FIG. 2, it is a current power spectrum during independent operation which characterizes the operating state of each of the electrical equipment 300 A non-negative value to be decomposed for weighting each base vector of the base vector matrix H so as to approximate the entire matrix H of the base vector group consisting of and the observation data (current power spectrum) indicated by the column vector x _i It can be expressed as a vector product with the column vector u _i of the weight matrix U corresponding to the column vector x _i of the matrix X. FIG. 2 is an explanatory diagram for explaining the concept of non-negative matrix decomposition NMF. Correspondence between an arbitrary column vector x _{i of} a non-negative matrix X, a matrix H of a basis vector group, and a weight matrix U is shown. Explains.

Here, an arbitrary certain column vector x _i of the non-negative matrix X to be decomposed indicating each observation data of the current power spectrum flowing through the electric device 300 is used as data indicating the characteristics of each electric device 300. assuming that can be substantially represented by only one basis vector h _i of the basis vector group of the matrix H indicating a current power spectrum at each independent operation, as shown in FIG. 3, of the weight matrix U, specific It can be expected that only the weight elements u _{i, j} have a significant value in the vicinity of “1”, and the remaining weight elements are set to almost “0”.

FIG. 3 is an explanatory diagram for explaining the concept of the non-negative matrix decomposition NMF in more detail. An arbitrary column vector x _i of the non-negative matrix X is converted into a specific base vector h of the matrix H of the base vector group. _If only _i can be represented approximately, then the weight matrix U is such that only a particular weight element u _{i, j} has a significant value in the vicinity of '1', and the remaining weight elements are almost ' It shows that it becomes 0 '.

  In the NMF calculation unit 13 of the electric appliance identification apparatus 100 shown in FIG. 1, the concept of non-negative matrix decomposition NMF is used.

  That is, first, in the data storage unit 15 of the electrical device identification apparatus 100 shown in FIG. 1, the power consumption amount when each of the multiple electrical devices 300 is independently operated is measured in advance, for example, in the form of a current power spectrum. Thus, it is stored in advance as a known database indicating the characteristics related to each electric device 300. That is, data indicating the power spectrum of the current at the time of single operation for each electrical device 300 is stored in the data storage unit 15 in advance.

  Next, when the total power consumption of one or more electrical devices 300 is measured as a total value of the current power spectrum in the power measuring unit 11 by one sensor, which one of the one or more electrical devices 300 is selected. In order to identify whether or not the device 300 is in operation, the measured power combining unit 12 has a power spectrum of current that is the power consumption amount of each known electrical device 300 stored in the data storage unit 15 during single operation. And the total value of the current power spectrum measured by the power measuring unit 11 (the total value of the power spectrum of the current, which is the amount of power consumption in an unknown state as to which of the plurality of electrical devices 300 is operating). Are grouped as column vectors to create a non-negative matrix X to be decomposed.

Thereafter, the NMF calculation unit 13 performs non-negative matrix decomposition NMF on the non-negative matrix X to be decomposed, thereby obtaining a non-negative value as a base vector matrix H composed of base vectors indicating the characteristics of each electrical device 300. Operation of each electric device 300 such that each of column vectors of the matrix X can be substantially represented by any specific base vector of any one of the base vector matrices H, each of which is a current power spectrum during single operation. A weight for obtaining a column vector composed of a power spectrum of current that is power consumption at the time of single operation characterizing the state, and weighting each non-negative matrix X to be decomposed as a weighting matrix U for each base vector The element u _{i, j} is obtained, and the non-negative matrix X to be decomposed is converted to the basis vector matrix H and the weight. Decompose into a vector product (H × U) with the matrix U.

  Here, the number of columns of the basis vector matrix H indicating the characteristics of each electrical device 300 as the current power spectrum of each electrical device 300 during independent operation corresponds to the number of electrical devices 300 and is an unknown power consumption. The number of columns is smaller than the number of columns of the non-negative matrix X to be decomposed including the column vector of the current power spectrum, for example, by one. The number of rows of the basis vector matrix H (that is, the number of elements of the column vector) corresponds to the vector order of the current power spectrum of the electric device 300 and is the same as the number of rows of the non-negative matrix X to be decomposed. .

  Further, the number of columns of the weight matrix U is the same as the number of columns of the non-negative matrix X to be decomposed, which is a target to be generated by weighting the entire basis vector matrix H, and each column vector is a non-negative matrix. Corresponds to the column vector of X. The number of rows of the weight matrix U corresponds to the number of electrical devices 300 and is the same as the number of columns of the basis vector matrix H.

For each weight element u _{i, j} of the weight matrix U, as described in FIG. 3, with respect to the column vector u _i of the weight matrix corresponding to the column vector x _i of the non-negative matrix X, the non-negative matrix X Weighting is performed on the characteristics of the current power spectrum of each electric device 300 so that the column vector x _i can be substantially expressed, and the electric current in the operation state in the column vector x _i of the non-negative matrix X For the device 300 only, the weight element corresponding to the column vector h _j of the basis vector matrix H indicating the characteristics of the electric device 300 indicates a significant value near “1”, and the other weight elements are substantially “0”. The value is close to '.

  When the NMF calculation unit 13 performs the non-negative matrix decomposition NMF regarding the measurement data of the power consumption, the electrical device extraction unit 14 determines the column vector of the weight matrix U based on the result of the non-negative matrix decomposition NMF. Among the weight elements of the column vector corresponding to the unknown column vector of the non-negative matrix X to be decomposed, the weight element having a significant value close to '1' instead of a value close to '0' When the power consumption is measured by the power measuring unit 11, the electrical device 300 that is assumed to be actually operating is identified, and the operating state of each electrical device 300 is estimated. can do.

  As described above, FIG. 4 illustrates a state in which the NMF calculation unit 13 performs the non-negative matrix decomposition NMF in order to identify the electric device 300 that is operating. FIG. 4 is an explanatory diagram illustrating a specific example of the state of non-negative value matrix decomposition NMF in the NMF calculation unit 13 of the electrical device identification apparatus 100 illustrated in FIG. 1.

In FIG. 4, the electrical equipment _E A as electrical apparatus 300 to be _identified, E _B, E C, though there are four electrical equipment _{E D,} the respective electric devices _{E A, E B, E C} , E D The power spectrum of the current, which is the power consumption during the single operation, is registered in advance in the data storage unit 15 as A, B, C, and D, respectively. Further, FIG. 4 shows a case where the power spectrum of the current actually measured by the power measuring unit 11 in an unknown state of the operating electric device 300 is (B + D), and the non-negative matrix decomposition NMF is calculated. as a result, of the four electrical equipment 300, power equipment during operation, shows an example in which identifying electrical equipment E _B, that there were two electrical apparatus E _D by the electric equipment extractor 14 Yes.

That is, in FIG. 4, in the non-negative matrix X to be decomposed, as described above, each of the column numbers X1 to X4 has four E _A , E _B , The power spectra A, B, C, and D of the currents during the single operation of E _C and E _D are arranged. Furthermore, in the column number X5, the electric power measuring unit 11 is in an unknown state in the power measuring unit 11. The power spectrum (B + D) of the actually measured current is arranged and collected by the measured power combining unit 12 as a non-negative matrix X consisting of a total of five column vectors.

By performing non-negative value matrix decomposition NMF on the non-negative value matrix X to be decomposed by the NMF calculation unit 13, each of the column numbers X1 to X4 of the non-negative value matrix X is substantially expressed as a basis vector matrix H as shown in FIG. four electrical equipment _E a that may _be, E _B, E _C, 4 single basis vector corresponding to the power spectrum of the _{E D} respective current ^{a * a -1, B * b} -1, C * c -1, D * d ^-1
Are generated in column numbers H1 to H4, and the current power of each of the four electric devices E _A , E _B , E _C , E _D for weighting each column vector of the basis vector matrix H at the time of non-negative coupling By extracting a, b, c, and d as weighting elements related to the spectrum, each of the column numbers U1 to U5 of the weight matrix U (column numbers corresponding to the column numbers X1 to X5 of the non-negative matrix X to be decomposed) As the weight elements, a, b, c, and d are set only for the weight elements corresponding to the column numbers H1 to H4 of the basis vector matrix H, respectively, and values of almost “0” are set for the other weight elements. become.

Accordingly, in the non-negative matrix X to be decomposed, each weight element of the column number U5 of the weight matrix U corresponding to the column number X5 in which the power spectrum of the current measured in an unknown state of the operating electric device is arranged. , A significant value is obtained by searching the electrical device extraction unit 14 for weight elements (b and d in FIG. 4) that are not close to “0” but have significant values set. basis (in FIG. 4, the basis vector matrix column number H2 and H4 of H corresponding to the electrical equipment E _B and electrical equipment E _D) vector matrix H column number determination to the that but corresponding to the weight elements set , it is possible to identify that four electrical equipment _{E a, E B, E C} , of the _{E D,} 2 one electrical equipment of the electrical device _{E B} and the electrical equipment _{E D} is in operation.

Further, by setting in advance registered power consumption amount at the time of the electrical device 300 each alone operation, the current consumption amount of power each of the two electrical appliances are consuming electrical equipment E _B and the electrical equipment E _D Can also be estimated.

As electrical equipment 300 to be identified, four electrical equipment _{E A, E B, E C} , the NMF operation of FIG. 4 described for the case of _{E D,} further generalized, for the case of M base of the electrical device, FIG. 5 shows. FIG. 5 is an explanatory diagram that generalizes and describes the state of the non-negative matrix decomposition NMF in the NMF calculation unit 13 of the electrical device identification apparatus 100 illustrated in FIG. 1. In the example shown in FIG. 5, the power consumption amount during the single operation of the electric device is measured by measuring the power consumption amount during single operation for any one or more of the M electric devices. That is, the current power spectrum is measured a total of N times (N ≧ M) and set and registered in the data storage unit 15 in advance. The vector order of the measured power spectrum of the current is shown as being L-dimensional.

  In other words, in FIG. 5, the non-negative matrix X to be decomposed is a plurality of times for any one to a plurality of M electric devices 300 (including all M devices). As a measurement result of a total of N times (N ≧ M) in which the L-dimensional current power spectrum is measured during operation, N pieces of L-dimensional power spectrum set and registered in advance in the data storage unit 15 are used as electric devices in operation. Are combined with the L-dimensional power spectrum measured by the power measuring unit 11 in an unknown state, so that the measured power combining unit 12 generates a total of (N + 1) column vectors.

Thereafter, the non-negative matrix X to be decomposed is subjected to non-negative matrix decomposition NMF by the NMF calculation unit 13, so that a base vector matrix H composed of base vectors indicating the characteristics of each electrical device 300 as shown in FIG. As follows: M units of electric power such that each column vector consisting of N L-dimensional power spectra in the single operation of the non-negative matrix X can be expressed almost only by any specific basis vector of the basis vector matrix H. It is possible to obtain M basis vectors composed of L-dimensional power spectra at the time of single operation that characterize the operation states of the devices 300 as column vectors, and more accurately approximate the non-negative matrix X to be decomposed as a weight matrix U. A weight element u _i for weighting each basis vector at the time of non-negative coupling _{, Find j} .

  As a result, (N + 1) columns of the weight matrix U (columns corresponding to the columns of the non-negative matrix X to be decomposed) correspond to the column vectors of the basis vector matrix H as the M weight elements. Only the weighting element at the position to be set is set to a weighting element having a significant value close to “1”, and the values of “0” are set to the other weighting elements.

  Therefore, in the electrical device extraction unit 14 that has inherited the result of the non-negative value matrix decomposition NMF performed in the NMF calculation unit 13, the electrical device 300 in operation in the non-negative value matrix X to be decomposed is measured in an unknown state. By searching M weight elements constituting the column vector of the weight matrix U corresponding to the column vector consisting of the current power spectrum, and extracting the weight elements having a significant value close to '1', the weight elements It is possible to identify the electrical device 300 corresponding to the above, and accurately identify which electrical device 300 is operating.

  Since the non-negative matrix decomposition NMF is an algorithm based on the assumption of non-negative additivity, the power consumption data to be calculated by the non-negative matrix decomposition NMF needs to be non-negative. There is a restriction that there is. Therefore, in the present embodiment, as described above, the power consumption vector amount indicating the power consumption amount of each electrical device 300 (total power consumption vector amount, single power consumption vector amount) is supplied to each corresponding electrical device 300. The power spectrum of the current is used. However, the present invention is not limited to such a case. For example, only the half period of the positive part of the current value of the current supplied to each corresponding electric device 300 may be extracted and used. Absent.

  Further, in the non-negative matrix decomposition NMF of the present embodiment, the power spectrum of the current flowing through each electrical device 300 is all treated the same regardless of the amount of power consumption, and the non-negative matrix X to be decomposed is determined. Although the case where it was created and approximated by the vector product of the base vector matrix H after decomposition and the weight matrix U has been described, the present invention is not limited to such a case.

  When all of the current power spectrum is treated the same regardless of the amount of power consumption, the error between the non-negative matrix X to be decomposed and the vector product of the basis vector matrix H and the weight matrix U is all Since the absolute evaluation format is such that the current error spectrum has almost the same error value, the ratio of error in the power consumption is lower for the spectrum with less power consumption than for the spectrum with more power consumption. growing.

  For example, in the calculation of the non-negative matrix decomposition NMF, when approximation is made so as to be within an absolute error range of 1 W, the error in the power spectrum of 100 W power consumption is 1 W and the power consumption is 10 W less than that. An error of 1 W in the power spectrum is equivalently handled.

  Therefore, although not described in the above-described embodiment, the current power spectrum flowing through each electrical device 300 is multiplied by a predetermined coefficient in consideration of the power consumption, according to the power consumption. Normalization of the current power spectrum may be performed, and the non-negative matrix X to be decomposed may be created for the normalized current power spectrum. As described above, the normalization process is performed in advance prior to the calculation of the non-negative matrix decomposition NMF, so that the non-negative matrix X to be decomposed, the basis vector matrix H and the weight after the non-negative matrix decomposition NMF are performed. The error between the vector product of the matrix U can be approximated in the form of a relative evaluation in the form of the ratio of error to the power consumption (error rate), regardless of the power consumption. The rate can be set to a substantially uniform value, and even in the case of the electric device 300 with a small amount of power consumption, the electric device 300 in operation can be accurately determined as in the case of the electric device 300 with a large amount of power consumption. Can be extracted.

Next, as a specific example of the electric device identification apparatus 100 of FIG. 1, when the electric device 300 to be identified is three electric devices E _A , E _B , E _C , three electric devices About the identification result actually implemented about the case where the current power spectrum at the time of single operation under the same conditions of each of E _A , E _B , and E _{C is} measured three times and set and registered in advance in the data storage unit 15 This will be further described with reference to FIG. FIG. 6 is an explanatory diagram for explaining an example of a specific implementation result of the identification processing of the electric device 300 by the electric device identification apparatus 100 of FIG. 1 according to the embodiment of the present invention.

In FIG. 6, the power spectrum of the current, which is the measurement result of the power consumption when each of the three electric devices E _A , E _B , E _C is operated independently, is the three electric devices E _A , E _B of the _{E C,} for electrical equipment _{E a,} as the third measurement result from the first round is shown in column number X1, X2, X3 of the non-negative matrix X of decomposed, respectively, A1, A2 is A3, for electrical equipment _{E B,} as the third measurement result from the first round is shown in column number X4, X5, X6 nonnegative value matrix X of decomposed, respectively, B1, B2, B3, and for the electrical equipment E _C , the first to third measurement results are C1, C2, C3, respectively, as indicated by column numbers X7, X8, X9 of the non-negative matrix X to be decomposed. As known data in the data storage unit 15 Settings are registered in advance. Furthermore, as indicated by the column number X10 of the non-negative matrix X to be decomposed, the power spectrum of the current measured by the power measuring unit 11 in a state where the operating electric device is unknown is (B + C). .

In such a case, in the measured power combining unit 12, as shown in column numbers X1 to X10 of the non-negative matrix X in FIG. 6, three types of three electric devices E _A , E _B , and E _C at the time of single operation are used. A non-negative matrix X is generated by synthesizing the current power spectra A1 to A3, B1 to B3, and C1 to C3 under the above measurement conditions and the current power spectrum (B + C) measured in an unknown operating state.

Next, when the non-negative matrix decomposition NMF is performed in the NMF calculation unit 13, as shown in FIG. 6, the electrical devices 300 to be identified are the three electrical devices E _A , E _B , and E _C. Three base vectors (column vectors) in the matrix H of the vector group are also extracted. Furthermore, each weight element of the weight matrix U has a significant value only in the weight element at the position corresponding to each column of the basis vector matrix H, as indicated by column numbers U1 to U10 in FIG. A value is set, and values close to “0” are set for the other weight elements.

In other words, among the column number U1~U10 the weight matrix U, the column number U1~U3 the column vector during independent operation of the electric device _{E A} corresponds to the column number X1~X3 of configured non-negative value matrix X , the first row corresponding to the basis vectors of the electrical equipment _{E a,} respectively, '1.395''1.376''1.386' is set, the second, the third row remaining In each case, a value in the vicinity of “0” is set.

Similarly, the column number U4~U6 corresponding to each electric equipment _{E B} and the electrical equipment _{E C} each single operation when the column vector set nonnegative matrix X column number X4~X6 and column number of X7~X9 and for column number U7~U9 also in the second row and the third row corresponds to the electrical equipment _{E B} and the electrical equipment _{E C} each basis vectors, respectively, '1.282''1.298' , '1.277' and '1,009', '1.015', '1.012' are set, and the remaining first, third and first and second rows are respectively Also, a value near “0” is set.

In the column number U10 operation state corresponds to the column number X10 nonnegative value matrix X column vector is set in the current power spectrum measured in an unknown state (B + C), corresponding to the basis vectors of the electric device E _A to the first row, but a value nearly "0" and "0.017" is set, the second row corresponds to the electrical device E _B and the electrical equipment E _C each basis vectors and the In the third row, large significant values close to “1.305”, “0.829”, and “1” are set, respectively.

Therefore, the electric device extraction unit 14 searches for each weight element in the tenth column of the weight matrix U in FIG. 6, thereby obtaining a power spectrum of the current measured by the power measurement unit 11 in an unknown operating state ( B + C) is measured when two of the three electric devices E _A , E _B , and E _C , that is, the electric device E _B and the electric device E _C are in operation. A quick and reliable determination can be made.

In the present embodiment, for each of the three electric devices E _A , E _B , E _C , the result of measuring the current power spectrum at the time of single operation as many as three times is set in the data storage unit 15. has been described as being registered, three electrical equipment E _a, E _B, have a different number for each E _C, it is possible to obtain the same determination result. Further, considering the case where the number of current power spectra is different for each of the three electric devices E _A , E _B , and E _C , as described above in (Example of operation of the embodiment), the three electric devices E _The current power spectrum flowing through each of _{A 1} , E _B , and E _C is multiplied by a predetermined coefficient in consideration of the power consumption, thereby normalizing the current power spectrum in accordance with the power consumption. If the non-negative matrix X to be decomposed is created for the converted current power spectrum, even if the number of current power spectra is different for each of the three electrical devices E _A , E _B , E _C An accurate discrimination result can be obtained.

  Further, in the calculation of the non-negative matrix decomposition NMF, generally, random values are set as the values set as initial values in the elements of the basis vector matrix H and the weight matrix U. As described above, the column vector consisting of the current power spectrum at the time of single operation that characterizes the operation state of each electrical device to be identified is used as the basis vector of the basis vector matrix H. By setting the current power spectrum at the time of such single operation as the base vector of the corresponding electric device to the column vector of the base vector matrix H, it becomes possible to converge the calculation of the non-negative matrix decomposition NMF at high speed. As a result, the calculation cost as a whole is further reduced.

(Description of effects in the present embodiment)
As described above in detail as embodiments and examples, the following effects can be obtained in this embodiment.

  That is, the calculation of the non-negative matrix decomposition NMF can be performed only by repeating four arithmetic operations, and the calculation cost for identifying the electric device 300 in operation is greatly increased regardless of the number of electric devices 300 to be identified. Therefore, in this embodiment, even in an environment where more electrical devices 300 are used, the operating state of a specific electrical device 300 can be quickly and reliably grasped.

  More specifically, in the case where there are a plurality of electrical devices 300 in the home or office of a general household of power consumers, the positions of power supply lines for supplying power, power lines in a distribution board, etc. As a technique for observing the total amount of power consumption, current, and voltage of the plurality of electric devices 300 with a single sensor installed in the network, and identifying the operating electric device 300 based on the observed results, a neural network is used. In contrast to the conventional method of performing matching with the feature value of each electric device 300 one by one by processing such as a network, the present embodiment and example include an NMF calculation unit 13 and a multivariate called non-negative matrix decomposition NMF. Since the analysis method is employed, the following effects can be obtained.

  First, it is possible to greatly reduce the calculation cost required to identify each electric device 300 in operation, and the currently operating electric device can be determined regardless of the number of electric devices 300 installed. The effect of being able to grasp quickly and surely is obtained, and it is suitable as an electric device identification device (monitor device) for grasping the type, operation state, power consumption, etc. of each electric device 300 used by electric power consumers. Can be used.

  Secondly, the type, operation status, power consumption, etc. of each electric appliance used by the electric power consumer are grasped by the electric appliance identification device 100 (monitor device) installed in the home or office of the electric power consumer. It is also possible to transfer the grasped information from the output unit 30 to the center device or the like via the communication network, and even in a remote place where the center device or the like is located, the electrical equipment in each power consumer Thus, it is possible to surely grasp the operation status, power control based on the operation status, and the like.

  Thirdly, by introducing the electrical device identification device 100 (monitor device) like this, from a center device or the like that performs various processes based on information output by the electrical device identification device 100 (monitor device), etc. It is also possible to receive various types of information as processing results via the input unit 20, information providing services for effectively reducing power consumption in the homes and offices of power consumers, and grasping of failures in electrical equipment -It will also be possible to provide power consumers with notification services and monitoring services for elderly people living alone and those requiring care.

DESCRIPTION OF SYMBOLS 10 ... Electric equipment identification part, 11 ... Electric power measurement part, 12 ... Measuring power synthetic | combination part, 13 ... NMF calculating part, 14 ... Electric equipment extraction part, 15 ... Data storage part, 20 ... Input part, 30 ... Output part, 100 ... Electrical device identification device, 200 ... Power source, 300 ... Electric device.

Claims (8)

  Based on the result of observing the total power consumption of a plurality of electric devices installed in the home or office of a power consumer with one sensor, the electric device is in an operating state. An electric device identification device for identifying an electric device, wherein a power measurement unit that measures a total value of power consumption amounts of a plurality of the electric devices observed by one sensor as a total power consumption vector amount, and the electric device A data storage unit that stores in advance the amount of power consumed when each unit operates independently as each unit power consumption vector amount, and the power measuring unit in an unknown state of the operating electrical device among the plurality of electrical devices. The measured total power consumption vector amount is combined with the single power consumption vector amount of each of the electric devices stored in advance in the data storage unit, thereby obtaining one non-negative value. A non-negative basis vector matrix by performing a non-negative matrix factorization (NMF) on the measured power combiner to be combined as a matrix and the non-negative matrix combined by the measured power combiner An NMF operation unit that decomposes into a vector product of a non-negative value weight matrix and an electric device that is operating among the plurality of electric devices based on a result of performing the non-negative value matrix decomposition in the NMF operation unit. An electrical device extraction unit that extracts an electrical device that is in an operating state when the total power consumption vector amount is measured by the power measurement unit in an unknown state, and is configured to include at least Electric device identification device.   2. The electric device identification device according to claim 1, wherein the total power consumption vector amount and the single power consumption vector amount are supplied to a power spectrum of a current supplied to the corresponding electric device or to the corresponding electric device. Only the half period of the positive part of the current value of the current is used.   3. The electric device identification device according to claim 1, wherein when the NMF computing unit decomposes the non-negative matrix, the single vector among the column vectors of the non-negative matrix to be decomposed is used as the basis vector matrix. A base vector composed of the individual power consumption vector amount of each of the electrical devices is represented so that each arbitrary column vector composed of the power consumption vector amount can be substantially expressed only by any one specific base vector of the base vector matrix. The significant value in the vicinity of '1' is obtained only with respect to the weight element corresponding to the electrical device that is in operation in each column vector of the non-negative matrix to be decomposed as each weight element of the weight matrix. Is set, and any column vector of the non-negative matrix to be decomposed is not in operation. Serial regard to the appropriate weight factor to the electrical device, almost electric device identification and wherein the value is set close to '0'.   Based on the result of observing the total power consumption of a plurality of electric devices installed in the home or office of a power consumer with one sensor, the electric device is in an operating state. An electrical device identification method for identifying an electrical device, wherein power consumption when each of the electrical devices operates independently is stored in advance in a data storage unit as an individual power consumption vector amount, and a plurality of the electrical devices are The electric power stored in advance in the data storage unit is a total power consumption vector amount that is a total value of the power consumption amounts of the plurality of electric devices observed by one sensor in an unknown state of the operating electric device. The non-negative matrix decomposition (NMF) is performed on the combined non-negative matrix after combining the single power consumption vector quantities of the devices together as one non-negative matrix. Non-negative Matrix Factorization) is performed to decompose a vector product of a non-negative basis vector matrix and a non-negative weight matrix, and based on the result of the non-negative matrix decomposition decomposed into the vector product, A method for identifying an electric device, comprising: extracting an electric device that is in an operating state when the total electric power consumption vector amount is observed while the operating electric device is in an unknown state.   5. The electric device identification method according to claim 4, wherein the single power consumption vector amount and the total power consumption vector amount are supplied to a power spectrum of a current supplied to the corresponding electric device or to the corresponding electric device. An electrical equipment identification method characterized by using only the half period of the positive portion of the current value of the current itself.   6. The electric device identification method according to claim 4, wherein, when decomposing the non-negative matrix, the basis vector matrix includes the single power consumption vector amount among column vectors of the non-negative matrix to be decomposed. A base vector composed of the single power consumption vector amount of each of the electric devices is obtained so that each arbitrary column vector can be substantially expressed by only one specific base vector of the base vector matrix. As each weighting element, a significant value in the vicinity of '1' is set only for the weighting element corresponding to the electric device in the operating state in each of the arbitrary column vectors of the non-negative matrix to be decomposed. Corresponding to the electrical device that is not in operation in each of the arbitrary column vectors of the non-negative matrix of With respect to the weight element, the electrical appliance identification method, wherein a value close to approximately "0" is set.   The electric device identification method according to claim 6, wherein, among the weight elements constituting the column vector of the weight matrix corresponding to the column vector composed of the total power consumption vector amount in the non-negative value matrix to be decomposed, '1 'Electrical equipment corresponding to a weight element having a significant value in the vicinity is operated when the total power consumption vector amount is observed in a state where the operating electrical equipment is unknown among the plurality of electrical equipments A method for identifying an electrical device, wherein the electrical device is identified as an electrical device in a medium state.   8. An electrical equipment identification program, wherein the electrical equipment identification method according to claim 4 is implemented as a program executable by a computer.

Images