KR20190078952A - Appliance monitoring system for predicting service condition of each device and monitoring method thereof - Google Patents

Abstract

The present invention relates to a home appliance monitoring system for identifying a usage state of a plurality of devices receiving power from a distribution board, and a monitoring method thereof. The home appliance monitoring system comprises: a first meter measuring used power or voltage of the devices at a first position adjacent to the distribution board on an internal wiring to which the devices are connected; a second meter measuring the used power or voltage of the devices at a second position different from the first position on the internal wiring; and a monitoring device identifying the usage state of the devices by using first power-voltage data measured by the first meter and second power-voltage data measured by the second meter. The second meter includes a composite sensor for measuring physical environment information at a time point when the first and second power-voltage data are measured. The monitoring device identifies the usage state of the devices in consideration of the first and second power-voltage data and the physical environment information. Accordingly, the present invention can correctly identify the usage state of devices.

Description

TECHNICAL FIELD [0001] The present invention relates to a home appliance monitoring system for predicting a usage state of each appliance, and a monitoring method thereof. [0002]

The present invention relates to a home appliance monitoring method, and more particularly, to a household appliance monitoring system and a monitoring method thereof for predicting a usage state of each appliance using a total electric energy used in a home.

In order to solve the problem of energy supply and demand in the future, it is necessary to secure technology for energy saving and efficient use. Accordingly, there is an increasing interest in building a smart grid environment for efficient use of energy. In addition, understanding and participation of the consumers as well as the producers is important for the successful introduction of Smart Grid in homes and buildings. In order to save energy on the customer side, the user can be aware of the current usage of power so that the user can participate in the usage reduction by himself.

A system that measures and monitors the amount of electricity used by each appliance is a straightforward and effective method, but has a disadvantage of high construction cost. Therefore, using non-intrusive load monitoring (NILM) technology that measures power consumption using a single measurement device can reduce facility complexity and cost. The NILM system uses a single measuring device to monitor the power consumption of each device and can monitor the signals connected to the measuring device to detect the electrical change caused by the operation of the individual device and notify the user. However, due to the noise and interference generated between the measuring device and the device, the NILM system using one measuring device has difficulty in accurately grasping the usage pattern of the individual device.

Korean Patent No. 10-1219416

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned technical problems, and it is an object of the present invention to provide a method for measuring power or voltage used in a home using a main meter close to a distribution board and an auxiliary meter installed at a position different from the main meter, The present invention provides a home appliance monitoring system and a monitoring method thereof that can acquire state information other than power and voltage by using a hybrid sensor.

The household appliance monitoring system for identifying the use state of a plurality of appliances which receive power from the distribution board according to the present invention is characterized in that it measures the use power or voltage of the appliances at a first position adjacent to the distribution panel on the internal wiring connected to the appliances A second meter for measuring a power or voltage used by the devices at a second location different from the first location on the internal wiring, and a second meter for measuring the first power-voltage data measured by the first meter, And a monitoring device for identifying the use state of the devices using second power-voltage data measured by the first and second meters, wherein the second meter is configured to monitor the physical state of the first and second power- And a hybrid sensor for measuring environmental information, wherein the monitoring device is configured to monitor the first and second power- In consideration of the physical environment with the information it may identify a use state of the device.

In an embodiment, the first and second power-voltage data may include a total usable power pattern or a total usable voltage pattern of the devices currently in use.

In an embodiment, the physical environment information may include data on illuminance, sound, temperature or humidity.

In an embodiment, the first meter receives the second power-voltage data and the physical environment information from the second meter via wired or wireless communication, and the monitoring device receives the first power- Voltage data, the second power-voltage data, and the physical environment information.

In an embodiment, the monitoring device is configured to encode the first power-voltage data, the second power-voltage data, and the physical environment information according to an exponential smoothing technique, a frequency division method and a peak comparison method to generate comparison data A determination unit that includes a home appliance determination system that applies a weight to the comparison data through a learning algorithm and compares the weighted data with reference data for each device received from the database to identify a device currently in use; And an output unit displaying the identification result of the determination unit on a display or storing the result in the database.

In an embodiment, the monitoring device may include a learning unit that learns the home appliance determination system through the learning algorithm and corrects the weight.

As an embodiment, the learning unit may input the plurality of measurement data measured from the first and second meters in the same situation to the learning algorithm to determine the weight.

In an embodiment, the encoding unit may generate frequency domain data by Fourier transform on the time domain data measured by the first and second meters, and may remove noise from the frequency domain data through the exponential smoothing scheme.

In an embodiment, the encoding unit divides the frequency-domain data from which noise has been removed by the frequency-division-partitioning analysis into a predetermined number of regions within a predetermined frequency band, divides the frequency-domain data into a predetermined number of regions by using the sum of squares of frequency components, The comparison data can be generated.

In an embodiment, the encoding unit may detect the peak frequency in the noise-removed frequency-domain data using the peak analysis method, and generate the comparison data using the array of the peak frequencies.

A household appliance monitoring method for identifying a state of use of a plurality of appliances receiving power from a distribution board according to the present invention includes the steps of generating a first power- Measuring data; measuring second power-voltage data comprising power or voltage information used by the devices at a second location other than the first location; determining, at the second location, Generating first and second weighted data by applying weights to the first and second power-voltage data, generating first and second weighted data based on the device-specific power-voltage reference stored in the database, Comparing the first and second weighted data with the device-specific power-voltage reference data, Comparing the physical state reference data stored in the database with the physical state information, and if the physical state information does not match the physical state reference data, comparing the first weighted data, the second weighted data, And can be stored in the database with new reference data.

In an embodiment, when the first and second weighted data match the device-specific power-voltage reference data, the device currently in use can be identified based on the device-specific power-voltage reference data.

In an embodiment, when the physical state information matches the physical state reference data, the device currently in use can be identified based on the device-specific power-voltage reference data.

In an embodiment, the first power-voltage data, the second power-voltage data, and the physical state information may be measured at the same time.

According to the embodiment of the present invention, it is possible to accurately identify the usage state of each device based on the power or voltage information measured by the main meter close to the distribution board and the auxiliary meter installed at a position different from the main meter.

In addition, the composite sensor included in the auxiliary meter can acquire additional information such as illumination, sound, temperature, and humidity to more accurately identify the usage state of each device.

In addition, the appliance monitoring system of the present invention can accurately identify the usage state of each appliance by applying a learning algorithm to the home appliance determination system.

1 is a block diagram illustrating a home appliance monitoring system according to an embodiment of the present invention.
Figure 2 is a block diagram illustrating an exemplary monitoring device of Figure 1;
FIG. 3 is a block diagram illustrating an exemplary first and second meters of FIG. 1;
4 is a flowchart showing a learning method of a home appliance monitoring system according to an embodiment of the present invention.
5 is a flowchart showing a learning method of a home appliance monitoring system according to another embodiment of the present invention.
6 is a flowchart illustrating a method of determining a home appliance monitoring system according to an embodiment of the present invention.
7A to 7G are views for explaining an encoding method used in the encoding unit of FIG.
8A and 8B are diagrams for explaining a learning algorithm used in the determination unit and the learning unit in FIG.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and should provide a further description of the claimed invention. Reference numerals are shown in detail in the preferred embodiments of the present invention, examples of which are shown in the drawings. Wherever possible, the same reference numbers are used in the description and drawings to refer to the same or like parts.

The terms first, second, or the like may be used to describe various elements, but the elements should not be limited by the terms. The terms may be named for the purpose of distinguishing one element from another, for example without departing from the scope of the right according to the concept of the present invention, the first element being referred to as the second element, Similarly, the second component may also be referred to as the first component.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between. Expressions that describe the relationship between components, for example, "between" and "immediately" or "directly adjacent to" should be interpreted as well.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises ", or" having ", and the like, are used to specify one or more of the features, numbers, steps, operations, elements, But do not preclude the presence or addition of steps, operations, elements, parts, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the meaning of the context in the relevant art and, unless explicitly defined herein, are to be interpreted as ideal or overly formal Do not.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. However, the scope of the patent application is not limited or limited by these embodiments. Like reference symbols in the drawings denote like elements.

1 is a block diagram illustrating a home appliance monitoring system according to an embodiment of the present invention. Referring to FIG. 1, the home appliance monitoring system 100 may include a monitoring device 110, a first meter 120, and a second meter 130. In addition, the home appliance monitoring system 100 may itself include a database 140, or may use an external database.

The devices 21 to 2n used in the home can be connected to the internal wiring 11 to receive power. For example, the appliances 21 to 2n may include household appliances that are continuously used for a long time, such as a refrigerator and an air purifier, and appliances used intermittently such as a TV, a light, a washing machine, and an air conditioner. The distribution board 10 can transmit electric power supplied from an external power supply source to the internal wiring 11. [

The first meter 120 may be connected to the first point P1 of the internal wiring 11 close to the distribution board 10. At the first point P1, the first meter 120 can measure the first power-voltage data for the power or voltage used in the first device 21 to the n-th device 2n. However, there is interference between each of the devices 21 to 2n and the first meter 120 due to impedances of other devices and electric wires. Therefore, it is difficult to distinguish individual devices by only the first power-voltage data.

The second meter 130 may be connected to any position of the internal wiring 11. For example, the second meter 130 may be connected to the second point P2, which is different from the first meter 120, between the devices 21 to 2n. At the second point P2, the second meter 130 may measure the second power-voltage data for the power or voltage used in the devices 21-2n. The impedance interference at the first point P1 may be different from the impedance interference at the second point P2. Thus, the first power-voltage data can be measured differently from the second power-voltage data. In addition, the second meter 130 may include a composite sensor for measuring illuminance, sound, temperature, humidity, and the like. The second meter 130 can measure the complex sensor data including environmental information such as illumination, sound, temperature, and humidity through the composite sensor. In one embodiment, a home control device or a smart secretary acting as a hub in a smart home may be used as the second meter 130 by installing the corresponding function.

The second meter 130 may communicate with the first meter 120 via wire or wireless. The second meter 130 may transmit the measured second power-voltage data and the composite sensor data to the first meter 120. The first meter 120 may transmit the measured first power-voltage data, the received second power-voltage data, and the received composite sensor data to the monitoring device 110.

The monitoring device 110 may perform learning and judgment to identify the devices 21 to 2n using the first power-voltage data, the second power-voltage data, and the composite sensor data. For example, the monitoring device 110 may include a home appliance determination system.

First, in the learning step, the monitoring device 110 can learn the home appliance determination system using the first power-voltage data, the second power-voltage data, and the composite sensor data to collect data for determination . For example, the monitoring device 110 may learn the home appliance determination system using measured data in various states. The monitoring device 110 may store in the database 140 reference data for the power and voltage of each device collected through learning.

On the other hand, in the determining step, the monitoring device 110 calculates the first power-voltage data based on the first power-voltage data, the second power-voltage data, and the composite sensor data measured through the first meter 120 and the second meter 130 It is possible to judge the current state of the devices 21 to 2n. For example, the monitoring device 110 compares the measured first power-voltage data and the measured second power-voltage data with the reference data of the database 140 to determine whether the devices 21-2n ) Can be identified. That is, the monitoring device 110 uses the first power-voltage data of the first meter 120 as well as the second power-voltage data of the second meter 130 to determine the identification accuracy of the devices 21 to 2n Can be improved. In addition, the monitoring device 110 may compare the measured composite sensor data with the reference data of the database 140 to identify the usage state of each of the devices 21 to 2n. The monitoring device 110 may further utilize the composite sensor data measured by the second meter 130 to further improve the identification accuracy of the devices 21 to 2n.

Figure 2 is a block diagram illustrating an exemplary monitoring device of Figure 1; Referring to FIG. 2, the monitoring apparatus 110 may include a receiving unit 111, an encoding unit 112, a determination unit 113, a learning unit 114, and an output unit 115.

The receiving unit 111 may receive the first power-voltage data, the second power-voltage data, and the composite sensor data from the first meter 120. For example, the first meter 120 may receive the second power-voltage data and the composite sensor data from the second meter 130. The first meter 120 may transmit the first power-voltage data, the second power-voltage data, and the composite sensor data to the receiving unit 111.

The encoding unit 112 may encode the received first power-voltage data, second power-voltage data, and composite sensor data for use in the determination unit 113 and the learning unit 114. [ For example, the encoding unit 112 may perform encoding using an exponential smoothing technique, a frequency band division method, and a peak comparison method. The encoding unit 112 may encode the first power-voltage data, the second power-voltage data, and the composite sensor data to generate first power-voltage comparison data, second power-voltage comparison data, have. The exponential smoothing technique, frequency band division method and peak comparison method are described in detail in FIGS. 7A to 7G.

The determination unit 113 may include a home appliance determination system that is learned through a learning algorithm. For example, the learning algorithm may include a Multinomial Logistic Regression, a Recurrent Neural Network (RNN), and a Convolutional Neural Network (CNN). The determination unit 113 identifies the device currently in use among the devices 21 to 2n based on the first power-voltage comparison data, the second power-voltage comparison data, and the composite sensor comparison data through the home appliance determination system . For example, the determination unit 113 may compare the first power-voltage comparison data, the second power-voltage comparison data, and the hybrid sensor comparison data with the reference data stored in the database 140 via the home appliance determination system. Based on the comparison result, the determination unit 113 can identify the device currently in use among the devices 21 to 2n. The multiple logistic regression method is described in detail in FIGS. 8A and 8B.

The learning unit 113 can learn the home appliance determination system of the determination unit 112 using the learning algorithm. For example, the learning unit 113 may control the first and second meters 120 and 130 to acquire various first power-voltage comparison data, second power-voltage comparison data, and composite sensor comparison data . The learning unit 113 may synchronize the measurement time points of the first and second meters 120 and 130. [ On the other hand, the learning unit 113 may store a list of devices capable of on / off control among the devices 21 to 2n. The learning unit 113 can perform learning by setting various operation states of the devices 21 to 2n. The learning unit 113 can perform a learning algorithm for the home appliance determination system based on the collected various measurement data.

The output unit 115 may display the power usage status of each device identified by the determination unit 113 on the display or may store the status in the database 140. [ The output unit 115 may output the usage history for each device corresponding to each device using the data stored in the database 140 at the request of the user.

FIG. 3 is a block diagram illustrating an exemplary first and second meters of FIG. 1; Referring to FIGS. 1 and 3, the first meter 120 may include a first communication unit 121 and a first power-voltage meter 122. The second meter 130 may include a second communication unit 131, a second power-voltage meter 132, and a composite sensor 133.

The first power-voltage meter 122 may measure the first power-voltage data at a first point P1 proximate to the distribution board 10. The second power-voltage meter 132 may measure the second power-voltage data at a second point P2 proximate to the instruments 21-2n. The first power-voltage data and the second power-voltage data may include a change in the invalid / active power level before and after turning on / off the devices 21-2n. The first power-voltage data and the second power-voltage data remove low-frequency components through the high-pass filter in the measured voltage signal and can include SMPS (Switched Mode Power Supply) noise of the devices 21-2n have. The devices 21 to 2n can be identified using the unique peak frequency (Reak Frequency) of each device in the SMPS noise. On the other hand, since the first power-voltage data and the second power-voltage data are measured at different points, device-specific voltage waveforms, frequency patterns, peak frequencies, and SMPS noise are different. Since the second power-voltage meter 132 is located close to the devices 21 to 2n, it is easy to measure the voltage signal compared to the first power-voltage meter 122.

The composite sensor 133 can measure the complex sensor data including the illuminance, sound, temperature, and humidity of the devices 21 to 2n in operation. The composite sensor data may be used to identify devices having similar first power-voltage data and second power-voltage data (e.g., two devices with similar operating voltages and power usage).

The second communication unit 131 can perform wired or wireless communication with the first communication unit 121. The second communication unit 131 may transmit the measured second power-voltage data and the composite sensor data to the first communication unit 121. [ The first communication unit 121 may transmit a command (for example, a synchronization signal) received from the monitoring device 110 to the second communication unit 131.

4 is a flowchart showing a learning method of a home appliance monitoring system according to an embodiment of the present invention. 1 and 4, the CE monitoring system 100 measures the power-voltage data and the complex sensor data through the first and second meters 120 and 130, and the power-voltage data and the complex sensor data The learning of the monitoring device 110 can be performed.

In step S110, the monitoring device 110 may receive the measured power-voltage data from the first and second meters 120, For example, the first meter 120 may measure the first power-voltage data at a first point P1. The second meter 130 may measure the second power-voltage data at the second point P2. The first and second meters 120, 130 may measure the first and second power-voltage data in synchronization with each other. The second meter 130 may transmit the second power-voltage data to the first meter 120. The first meter 120 may transmit the first and second power-voltage data to the monitoring device 110. The monitoring device 110 may encode the first and second power-voltage data using an exponential smoothing technique, a frequency divisional analysis method, and a peak comparison method.

In step S120, the monitoring device 110 may receive the device-specific power-voltage reference data from the database 140. [ For example, the power-voltage reference data may be data input by the user or previously measured and stored data. The power-voltage reference data may be pre-encoded data.

In step S130, the monitoring device 110 performs learning about the CE device based on the encoded first and second power-voltage data and the power-voltage reference data, Weights can be generated. For example, the monitoring device 110 may compare the result of applying the default weights to the first and second power-voltage data based on a learning algorithm (e.g., multiple logistic regression) with the power-voltage reference data . According to the comparison result, the monitoring device 110 may generate the first weight by modifying the basic weight in a direction in which the resultant value matches the power-voltage reference data.

In step 140, the monitoring device 110 may receive the composite sensor measurement data from the second meter 130. For example, the second meter 130 may simultaneously measure the composite sensor measurement data at the time of measuring the second power-voltage data. The second meter 130 may transmit the composite sensor measurement data to the first meter 120. The first meter 120 may transmit the composite sensor measurement data to the monitoring device 110. The monitoring device 110 may encode complex sensor measurement data using an exponential smoothing technique, a frequency divisional analysis method, and a peak comparison method.

In step S150, the monitoring device 110 may receive the composite sensor reference data from the database 140. [ For example, the composite sensor reference data may be data entered by the user or previously measured and stored data. The composite sensor reference data may be pre-encoded data.

In step S160, the monitoring apparatus 110 learns the home appliance determination system based on the encoded hybrid sensor measurement data and the hybrid sensor reference data, and modifies the first weight used in the home appliance determination system to the second weight . For example, the result of applying the first weight to the composite sensor measurement data based on the learning algorithm can be compared to the composite sensor reference data. According to the comparison result, the monitoring device 110 may generate the second weight by modifying the first weight in a direction in which the resultant value and the complex sensor reference data match. The second weight generated according to the learning method of FIG. 4 may be stored in the database 140 and used in the determination method of FIG.

5 is a flowchart showing a learning method of a home appliance monitoring system according to another embodiment of the present invention. 1 and 5, the CE monitoring system 100 measures power-voltage data and complex sensor data through the first and second meters 120 and 130, and generates power-voltage data and complex sensor data The learning of the monitoring device 110 can be performed. Also, the home appliance monitoring system 100 can set on / off of the devices 21 to 2n.

In step S210, the monitoring device 110 may select a device to be operated among the devices 21 to 2n for map learning. For example, the monitoring device 110 may store or receive from the database 140 a list of devices that can be controlled on / off among the devices 21 to 2n. The monitoring device 110 can set various operating states of the devices 21 to 2n based on the list of devices that can be controlled on / off. That is, the monitoring device 110 can operate only the devices desired to be measured.

In step S220, the monitoring device 110 may receive the measured first and second power-voltage data from the first and second meters 120 and 130 in a set state. For example, the monitoring device 110 may encode the first and second power-voltage data using exponential smoothing techniques, frequency-division techniques, and peak comparison techniques. In step S230, the monitoring device 110 may receive the power-voltage reference data of the selected device from the database 140. [ For example, the power-voltage reference data may be pre-encoded data. In step S240, the monitoring device 110 performs learning based on the first and second power-voltage data and the power-voltage reference data for the selected device, and generates a first weight used in the home appliance determination system .

In step S250, the monitoring device 110 may receive the composite sensor measurement data measured by the second meter 130 in the set state. The monitoring device 110 may encode complex sensor measurement data using an exponential smoothing technique, a frequency divisional analysis method, and a peak comparison method. In step S260, the monitoring device 110 may receive the composite sensor reference data from the database 140 when the selected device operates. For example, the composite sensor reference data may be pre-encoded data. In step S270, the monitoring device 110 learns the selected device based on the composite sensor measurement data and the composite sensor reference data, and modifies the first weight used in the home appliance determination system to the second weight. The second weight generated according to the learning method of FIG. 5 may be stored in the database 140 and used in the determination method of FIG.

6 is a flowchart illustrating a method of determining a home appliance monitoring system according to an embodiment of the present invention. Referring to FIGS. 1 and 6, the home appliance monitoring system 100 uses the power-voltage data and the composite sensor data measured from the first and second meters 120 and 130 to determine the current usage state of each device It can be judged.

In step S310, the monitoring device 110 may receive the measured power-voltage data from the first and second meters 120, For example, the first meter 120 may measure the first power-voltage data at a first point P1. The second meter 130 may measure the second power-voltage data at the second point P2. The first and second meters 120, 130 may measure the first and second power-voltage data in synchronization with each other. The second meter 130 may transmit the second power-voltage data to the first meter 120. The first meter 120 may transmit the first and second power-voltage data to the monitoring device 110. The monitoring device 110 may encode the first and second power-voltage data using an exponential smoothing technique, a frequency divisional analysis method, and a peak comparison method.

In step S320, the monitoring device 110 may input the encoded first and second power-voltage data to the consumer electronics determination system and apply weights to generate the first and second power-voltage weighted data. For example, the weight may be a second weight generated by the learning method of FIG. 4 or FIG. The monitoring device 110 may apply the learning algorithm used in FIG. 4 or 5 to generate first and second power-voltage weighted data.

In step S330, the monitoring device 110 may receive device-specific power-voltage reference data from the database 140. For example, the power-voltage reference data may be data input by the user or previously measured and stored data. The power-voltage reference data may be pre-encoded data.

In step S340, the monitoring device 110 may compare the power-voltage reference data with the first and second power-voltage weighted data. If the first and second power-voltage weighted data match the power-voltage reference data, the process proceeds to step S390, where the monitoring device 110 may identify the currently operating device. If the first and second power-voltage weighted data do not match the power-voltage reference data, the flow proceeds to step S350.

In step S350, the monitoring device 110 may receive the composite sensor measurement data from the second meter 130. [ For example, the second meter 130 may simultaneously measure the composite sensor measurement data at the time of measuring the second power-voltage data. The second meter 130 may transmit the composite sensor measurement data to the first meter 120. The first meter 120 may transmit the composite sensor measurement data to the monitoring device 110. The monitoring device 110 may encode complex sensor measurement data using an exponential smoothing technique, a frequency divisional analysis method, and a peak comparison method.

In step S360, the monitoring device 110 may receive the composite sensor reference data from the database 140. [ For example, the composite sensor reference data may be data entered by the user or previously measured and stored data. The composite sensor reference data may be pre-encoded data.

In step S370, the monitoring device 110 may compare the composite sensor measurement data with the composite sensor reference data. If the composite sensor measurement data coincides with the composite sensor reference data, the monitoring device 110 may identify the device currently operating, even if there is a slight difference in the power-voltage data, in step S390. If the composite sensor measurement data does not match the composite sensor reference data, the flow proceeds to step S380.

In step S380, the monitoring device 110 may register the measured first and second power-voltage data and the combined sensor measurement data as new device data and store the new device data in the database 140. [ The new device data can then be used as reference data.

7A to 7G are views for explaining an encoding method used in the encoding unit of FIG. 7A to 7C are diagrams for explaining an exponential smoothing technique according to an embodiment of the present invention. FIG. 7A is a graph showing Fourier transform performed on the basic voltage signal measured by the first meter or the second meter of FIG. 1. FIG. FIG. 7B is a graph in which the Fourier transform data of FIG. 7A is subjected to exponential smoothing. FIG. 7C is a graph of the exponential smoothing technique performed twice on the Fourier transform data of FIG. 7A. FIG.

Referring to FIGS. 7A to 7C, the noise generated when the voltage value is obtained appears as noise even after FFT (Fast Fourier Transform) conversion. Some of the noise is white noise, but most of the noise is caused by the switching operation of the internal SMPS (Switched Mode Power Supply) circuit, which makes frequency analysis difficult. This noise can be significantly reduced by simply smoothing the graph after FFT conversion. Since the exponential smoothing technique reflects the center value most sensitively, it does not impair the peak value to be distinguished by the characteristic. The exponential smoothing technique can be expressed by Equation (1).

In Equation (1),? Can be set to a value between 0 and 1, and the closer to 1, the larger the weight for the center value. As an example, FIGS. 7B and 7C are graphs showing that the exponential smoothing technique is performed after 0.5 is input to?. Referring to FIG. 7C, in which the exponential smoothing technique is performed twice on the Fourier transform data of FIG. 7A, distortion of the graph due to noise is reduced, and the characteristic can be changed into a state that is easy to extract the characteristic. Thus, the encoding unit 112 of FIG. 2 may encode the measured first and second power-voltage data using exponential smoothing techniques.

FIGS. 7D and 7E are views for explaining a frequency-division analysis method according to an embodiment of the present invention. FIG. 7D is a graph in which the frequency fuzz size of the exponential smoothed Fourier transform graph of FIG. 7C is converted in decibels. FIG. 7E is a graph in which the frequency band segmentation analysis is applied to the graph of FIG. 7D. FIG.

Referring to FIGS. 7D and 7E, the frequency generated when the home appliance operates does not appear to be focused on several characteristic frequencies due to the noise and the characteristics of the device, but spreads little by little about the characteristic frequencies. The degree of spreading differs from device to device and measurement environment, making it difficult to set standards and it is often difficult to extract device-specific features.

Using frequency domain partitioning, this difficulty is largely solved, and characteristic frequencies can be easily extracted. The frequency band segmentation method is a method based on the Philips Audio Fingerprinting technique used for song recognition and is a method of calculating the energy of each region by dividing the frequency band to be observed in the FFT graph into several regions. If the energy of each region thus calculated is normalized by the energy of the entire observation region, a specific pattern appears. If the range of the equal number and the observation frequency band is appropriately adjusted, the specific pattern is uniquely displayed for each device. . The application of the frequency domain segmentation method can calculate the energy by grouping the frequency components appearing in the vicinity of a specific frequency into the same domain, and the aforementioned problems can be solved.

As an example, the graph of FIG. 7E is a graph in which frequency division analysis is performed by setting an equal number to 50 and a frequency of fwindow (frequency band to be observed) to 1 kHz in the graph of FIG. The frequency band partitioning analysis can be expressed by Equation (2).

In Equation (2), exponential smoothing is omitted for the sake of convenience. In Equation (2), X represents an FFT-transformed signal, Xwindowed represents X covered by a window, N represents a number of measured voltage data, m represents a sampling rate, and fwindow represents a frequency band to be observed. Also, k denotes a domain value corresponding to each frequency after the FFT transform, and the corresponding relationship is expressed by Equation (3).

Then, the frequency band is divided into p regions, and the frequency components of each region are squared and added, and normalized to the total energy value of the frequency band. When Xwindowed is divided by p, the number of elements q of each region can be expressed by Equation (4).

The band power (BP) per area can be expressed by Equation (5).

This BP array is an array with an equal number of elements, which can be used for feature extraction by device.

The encoding unit 112 of FIG. 2 may use two comparison methods based on a frequency band division method. First, the encoding unit 112 of FIG. 2 can compare the value of BP (i) itself. BPref, which is an array serving as a comparison reference, can be obtained from the voltage data of each device registered in the database 140. The encoding unit 112 of FIG. 2 can compare the BP array of the newly measured voltage data and the reference array BPref as shown in Equation (6).

Equation (6) adds the error rate of the BP value, and the smaller the Yval value is, the more the FFT spectrum of the two voltage data can be judged to be similar. The comparison of the BP values themselves shows the overall similarity of the observation frequency band.

Second, the encoding unit 112 in FIG. 2 can compare the amount of change in BP. The encoding unit 112 of FIG. 2 can compare the amount of change of the neighboring values in the BP array made of the measured voltage data and the amount of change between the intervals of the reference array BPref corresponding to Equation (7).

Equation (7) adds the error rate of the variation amount of the BP, and the smaller the Ydiff value is, the more similar the FFT spectrum of the two voltage data can be judged to be. This comparison of BP variations can be used to compare and capture sudden changes in the frequency spectrum, such as peaks, which are easily missed in the BP value self comparison method.

7F and 7G are views for explaining a peak comparison method according to an embodiment of the present invention. FIG. 7F is a graph of performing the Fourier transform on the measured fundamental voltage signal and performing the exponential smoothing technique twice on the Fourier transform data. FIG. 7G is a graph showing a peak detection technique applied to the graph of FIG. 7F. FIG.

The peak comparison method is a function that detects and compares the peaks of the frequency spectrum with an exponential smoothed FFT as input. The encoding unit 112 in FIG. 2 can detect a peak based on the following two conditions.

(Condition 1) In FIG. 7F, the magnitude of X (i) is larger than the median value * of the entire spectrum.

(Condition 2) In Fig. 7F, the rate of increase from X (i) to X (i + 1) or X (i + 2) and X (i + 3) is greater than?.

The encoding unit 112 of FIG. 2 increases i, finds X (i) satisfying the above two conditions, and then finds X (k), X (k + 1), X (J + N * m) to X (j + gamma * N / m), where j is the minimum value of k that is smaller than the median value * . Where N is the number of data samples, m is the sampling rate, and α, β, and γ are sensitivity constants and are set to values that accurately select peaks according to the measurement environment. As α and β become larger, the criterion for the peak becomes strict. For a frequency judged to be a potential peak, gamma is checked to see if there is another peak at a distance of? Hz back and forth so that the peak is not detected as redundant due to noise. As an example, FIG. 7G shows a result of detecting a peak by setting? = 2,? = 0.8 and? = 10.

The encoding unit 112 of FIG. 2 compares the peak frequencies extracted by the peak detection technique of FIG. 7G with P (i) and compares it with P (i) from the device-specific data stored in the database 140, The comparison value Ypeak can be obtained. The encoding unit 112 in FIG. 2 compares the two frequencies with the same frequency value (when there is a difference in the calculated peak frequency even though the same peak is present, the sensitivity constant of +/- Hz is set) After the addition, you can divide two arrays by the number of arguments in the array with more arguments. On the other hand, in order to have a tendency similar to the comparison method of Equations (6) and (7), this value is subtracted from 1 and set to Ypeak. As the peak comparison value (Ypeak) is smaller, the FFT spectrum of the two voltage data can be judged to be similar. Since a peak is generated in a unique pattern for each device, it can be compared to determine whether or not the peak is coincident.

The encoding unit 112 in FIG. 2 can generate the comparison values (Yval, Ydiff) based on the frequency band division method and can generate the comparison value (Ypeak) based on the peak comparison method. The generated comparison values (Yval, Ydiff, Ypeak) may be transmitted to the determination unit 113 of FIG.

8A and 8B are diagrams for explaining a learning algorithm used in the determination unit and the learning unit in FIG. 8A is a diagram showing a basic system of a logistic regression. FIG. 8B is a diagram showing a Sigmoid function used in the logistic regression method. FIG.

Logistic regression is a method of predicting the likelihood of an event through linear combination of independent variables. The fact that output (dependent variable) appears as a linear combination of inputs (independent variable) is similar to linear regression, but differs in that the dependent variable is divided into two categories and can be used for binomial classification.

8A, the correction factor B may be interpreted as a weight W0 for one input. The logistic regression is simply a weighting function to classify the input data into a desired value (Y) through equations dividing the points of the n-dimensional space consisting of n independent variables into two regions (two-dimensional lines and three- (W1, W2, ..., Wn). Since the linear combination (Z) of the input variable and the weight has a value in a range that is not constant, it is transformed to have a value between 0 and 1 for convenience of calculation. This process is called activation. The activation of the logistic regression method generally uses the sigmoid function of FIG. 8b. The sigmoid function can be expressed as Equation (8).

The sigmoid function has a value of 0.5 at x = 0, and is a function that is a circle-symmetric based on this. Using the sigmoid function, A has a value between 0 and 1 after the activation process (A = f (Z)) regardless of the Z value. If you convert the resulting value of A to a value closer to 0 and 1, the result will be the output of the system, Y '. If you know the output Y for a particular input X, you can adjust W so that Y 'equals Y. This process is called Back-Propagation, which allows the system to perform Supervised Learning . If you then enter a new input X, you can guess which of the two groups 0 and 1 the result belongs to. At this time, if you learn the system using a sufficient number of I / O data, the W value becomes quite elaborate and the guess value becomes closer to the correct answer. In this way, logistic regression produces a binomial classification system that responds to one of two groups for the input factor.

If the output of the basic logistic regression method is expressed as one of two groups, Multinomial Logistic Regression (or Softmax Regression) is an algorithm that classifies the output into three or more groups by applying it. W Combination One output classifies the output into two types, so if the W combination is equal to the number of groups (g) to be classified, the output can be classified into 2 ^g regions. For example, when the output should be divided into A, B, and C groups, if there are three W combinations, each combination is divided into A or Not A, B or Not B, and C or Not C, of the ²³ zones will be divided into one place. Therefore, when there are m input variables and there are g groups, the weight arrangement can be expressed as Equation (9).

In this case, the 0th factor of W is a simple expression of the correction value B as a weight for one input. In this case, a response Z to n inputs of data composed of m input variables can be expressed as Equation (10).

The activation process of polynomial logistic regression can use softmax function instead of sigmoid function. If the result of the sigmoid function is a value between 0 and 1, the result of the soft max function is g (number of groups), each of which has a value between 0 and 1, If the soft max function is represented by a matrix, it can be expressed as Equation (11).

By definition, each element of A matrix is a value between 0 and 1, and the sum of each column is 1. Where A _{i, j} means the probability that the j-th input is an i-th group. The object of the present invention is to match one of g devices with respect to each input, so the largest value should be selected for each column of the matrix A. Therefore, in the polynomial logistic regression method, the 'One-Hot Encoding' technique should be applied instead of the binomial classification of FIG. 8A. Applying this technique to each column yields a matrix with the largest value of 1 and the remainder of 0, which is the output of the logistic regression system, Y '. In summary, we assume that Y _{i, j} = 1 in the output Y 'of the g * n matrix when inputting n data with m input variables into the system and guessing that the j-th input corresponds to the i group Is the Forward Propagation of the polynomial logistic regression.

Now, the system constructed by the polynomial logistic regression method should obtain the weighting matrix W that accurately maps each input to one of g groups. Since W is not known at the beginning, it is set to an arbitrary value. Compute Y with the input X here and compare it with the Y created by the actual corresponding value, and modify W so that the two matrices are equal. This process is called Back-Propagation. In the polynomial logistic regression method, a cross entropy technique is used as a back propagation algorithm.

A cost function is a numerical representation of a difference between Y and Y '. In a cross entropy technique, a cost function H can be expressed by Equations (12) and (13).

In Equation (11), -logA _{i, j} is 0 when A _{i, j} is 1, and diverges to positive infinity when it is 0. If we add Y _{i, j} to this, we can see from the equation that the cost function (H) is 0 when Y _{i, j} and A _{i, j} are equal, and the cost function (H) .

In the back propagation process, the matrix X is a constant that is not changed as an input, so it can be thought of as a constant multiplied by the weighting factors, and the cost function (H) is a function composed of independent variables Wa0 to Wgm. Therefore, it is possible to take a gradient for W in the cost function H (Equation 14).

Since ∇H is a vector that increases the cost function H, to reduce the cost function H, let W0 be a point on the g * (m + 1) dimension produced by the W matrix. _{W = W0} direction. Therefore, when learning rate is α, W after the back propagation process can be changed as shown in equation (15).

This implies that each factor W _{i, j} of the W matrix changes as shown in equation (16).

The learning rate α is reduced to 0.1, 0.01, and 0.001, and the cost function can be adjusted to the fastest decrease. If you put the input X back into the system where W is adjusted, the output A and hence the Y 'will be a little closer to the Y value.

When the polynomial logistic regression method and the cross entropy method are sufficiently performed, ∇H reaches almost 0, and the W and the cost function (H) hardly change. In the present invention, this point is defined as a point at which the sum of the absolute values of all the element values of | Wnew-Wo | becomes smaller than?. ε is assumed to be the same as the learning rate α as an example of a system parameter. At this point, we can determine that the cost function (H) reaches a local minimum, where A approaches the local closest to Y at local. In addition, Y 'coincides with Y in an ideal case. Y 'may not be equal to Y if the cost function H is strictly concave or there are many bad data. In this case, W can be set as a new arbitrary value, another minimal value can be found, bad data can be removed, and the learning can be resumed.

If W is determined so that Y '= Y without doing so, it can be judged that the learning of the system has been successfully completed. If the number of learned data is large enough, then the post-learning W will be able to successfully classify new inputs with high probability.

The learning unit 114 of FIG. 2 may perform learning using a polynomial logistic regression method and a cross entropy technique. The determination unit 113 of FIG. 2 can identify the device currently in use based on the new input by using the polynomial logistic regression method and the cross entropy method.

The embodiments have been disclosed in the drawings and specification as described above. Although specific terms have been employed herein, they are used for purposes of illustration only and are not intended to limit the scope of the invention as defined in the claims or the claims. Therefore, those skilled in the art will appreciate that various modifications and equivalent embodiments are possible without departing from the scope of the present invention. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

10: Distribution board
11: Internal wiring
21 ~ 2n: Devices
100: Home appliance monitoring device
110: Monitoring device
111: Receiver
112: encoding section
113:
114:
115:
120: First meter
121: first communication section
122: First power-voltage meter
130: Second meter
131:
132: Second power-voltage meter
133: Composite sensor
140: Database

Claims (14)

A household appliance monitoring system for identifying a use state of a plurality of appliances receiving power from a distribution board,
A first meter for measuring the power or voltage used by the devices at a first location adjacent to the distribution board on an internal wiring to which the devices are connected;
A second meter for measuring a usage power or voltage of the devices at a second location different from the first location on the internal wiring; And
And a monitoring device for identifying the use state of the devices using the first power-voltage data measured by the first meter and the second power-voltage data measured by the second meter,
Wherein the second meter includes a composite sensor for measuring physical environment information at a time when the first and second power-voltage data are measured,
Wherein the monitoring device identifies the use state of the devices by considering the first and second power-voltage data and the physical environment information together. The method according to claim 1,
Wherein the first and second power-voltage data comprise a total used power pattern or a total used voltage pattern of the devices currently in use. The method according to claim 1,
Wherein the physical environment information includes data on illuminance, sound, temperature, or humidity. The method according to claim 1,
Wherein the first meter receives the second power-voltage data and the physical environment information from the second meter via wired or wireless communication,
Wherein the monitoring device comprises a receiver for receiving the first power-voltage data, the second power-voltage data and the physical environment information from the first meter. 5. The method of claim 4,
The monitoring device includes:
An encoding unit encoding the first power-voltage data, the second power-voltage data, and the physical environment information according to an exponential smoothing scheme, a frequency band division scheme, and a peak comparison scheme to generate comparison data;
A judgment unit including a household appliance judgment system for applying a weight to the comparison data through a learning algorithm and comparing the weighted data with reference data for each appliance received from the database to identify the appliance currently in use; And
And an output unit for displaying the identification result of the determination unit on the display or storing the result in the database. 6. The method of claim 5,
Wherein the monitoring apparatus includes a learning unit that learns the home appliance determination system through the learning algorithm and corrects the weight. The method according to claim 6,
Wherein the learning unit inputs the plurality of measurement data measured from the first and second meters in the same situation to the learning algorithm to determine the weight. 6. The method of claim 5,
Wherein the encoding unit generates frequency domain data by Fourier transforming the time domain data measured by the first and second meters and removes noise from the frequency domain data through the exponential smoothing technique. 9. The method of claim 8,
The encoding unit divides the noise-removed frequency-domain data into a predetermined number of regions in the frequency band by the frequency-division-based analysis method, and generates the comparison data using the sum or the absolute sum of the frequency component values in each region Home appliance monitoring system. 9. The method of claim 8,
Wherein the encoding unit detects the peak frequency in the noise-removed frequency-domain data through the peak analysis and generates the comparison data using the array of the peak frequencies. A household appliance monitoring method for identifying a use state of a plurality of appliances receiving power from a distribution panel,
Measuring first power-voltage data comprising power or voltage information used by the devices at a first location proximate to the distribution board;
Measuring second power-voltage data comprising power or voltage information used by the devices at a second location other than the first location;
Measuring physical state information of the devices at the second location;
Applying a weight to the first and second power-voltage data to produce first and second weighted data;
Comparing the first and second weighted data with device-specific power-voltage reference data stored in a database;
Comparing the physical state information with physical state reference data stored in the database if the first and second weighted data do not match the device specific power-voltage reference data; And
Wherein the first weighted data, the second weighted data, and the physical state information are stored in the database as new reference data when the physical state information does not match the physical state reference data. 12. The method of claim 11,
And if the first and second weighted data coincide with the device-specific power-voltage reference data, identifies the device currently in use based on the device-specific power-voltage reference data. 12. The method of claim 11,
And if the physical state information matches the physical state reference data, identifies the device currently in use based on the device-specific power-voltage reference data. 12. The method of claim 11,
Wherein the first power-voltage data, the second power-voltage data, and the physical state information are measured at the same time.

Images