Power data service system and power data service method

Abstract

The challenge is to provide time-series electricity data that is tailored to the user's intended use. [Solution] The problem was solved by a power data service system comprising a power data separation means installed in a building that separates and estimates measured power for each electrical device, and a cloud server that processes the power separation data, wherein the cloud server's feature extraction unit extracts primary features from time-series data that represent the portion where changes in the usage status of electrical devices occur, and further extracts secondary features that show the strength of the relationship between the primary features and at least one of the following: time information, calendar information, address information, temperature information, and humidity information, and the trained model generated by the trained model generation unit receives time-series power data and input information regarding the purpose of use, and further receives condition information that defines the output target, and outputs time-series power data suitable for the purpose of use.

Description

Technical Field

[0001] The present invention relates to a power data service system and a power data service method for providing users with power data suitable for various usage purposes, each of which is considered to be related to the usage status of the power consumption of each electrical device used in each building of a home, a building, and a factory.

Background Art

[0002] Patent Document 1 discloses a power demand prediction device including: weather information receiving means for receiving weather information including a predicted temperature value, a predicted humidity value, and a predicted solar radiation amount in a predetermined area from a weather information management server for managing weather information; storage means for storing an outside air enthalpy coefficient representing the degree of influence of the outside air enthalpy corresponding to the predicted temperature value and the predicted humidity value on the power demand, and a solar radiation coefficient representing the degree of influence of the predicted solar radiation amount on the power demand; and power demand prediction value calculating means for calculating a power demand prediction value, which is a predicted value of the power demand in the predetermined area, based on the weather information received by the weather information receiving means, the outside air enthalpy coefficient and the solar radiation coefficient stored in the storage means. The storage means further has a building information database in which a delay time until solar radiation to a building is reflected in an increase in the indoor temperature of the building is stored. The power demand prediction value calculating means calculates a correlation coefficient representing the correlation between a graph in which the temporal change of the solar radiation amount is delayed by the delay time candidate and a graph showing the temporal change of the power demand for each of a plurality of delay time candidates that are candidates for the delay time, sets the delay time candidate having the maximum correlation coefficient among the plurality of delay time candidates as the delay time, and refers to the building information database when calculating the power demand prediction value for a future predetermined time, and calculates the power demand prediction value using the predicted solar radiation amount at a time retrogressed by the delay time from the predetermined time.

[0003] Patent Document 2 discloses an action estimation device comprising: an information input unit that inputs input information including equipment-separated power and individual circuit power obtained from power measured by a power sensor installed in the residence of a person to be monitored; an action estimation unit that estimates the behavior of the person based on the input information input by the information input unit using machine learning results; and an information output unit that outputs information regarding the estimation results by the action estimation unit. The equipment-separated power is obtained by performing a predetermined equipment separation process based on information on the total amount of power in the residence obtained by the power sensor, and includes information representing the amount of power used for one or more devices used in the residence, and the room corresponding to the individual circuit power is a room in which no devices that can obtain power using the equipment-separated power do not exist.

[0004] Patent Document 3 describes a detection step for detecting vibrations emitted from a rotating machine, a waveform shaping step for broadening the base of the output of the detection step, a recording step for recording the output of the waveform shaping step, and a signal for extracting a predetermined signal indicating damage to the rotating machine from the signal recorded in the recording step. A method for diagnosing the equipment of a rotating machine, comprising a processing step, is disclosed. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Japanese Patent Publication No. 6130055 [Patent Document 2] Patent No. 7596695 [Patent Document 3] Japanese Patent Publication No. 2013-160749 [Overview of the project] [Problems that the invention aims to solve]

[0006] The invention described in Patent Document 1 is a power demand forecasting technology that calculates predicted power demand values ​​in a predetermined area. However, there was a problem in that even if the amount of electricity measured in a predetermined area was used only for the power demand of that area.

[0007] The invention described in Patent Document 2 is a technology aimed at people with dementia and those receiving care, and it uses isolated power of equipment and individual circuit power, as well as sensors such as toilet room sensors and bedroom sensors, to estimate the behavior of people with dementia and those receiving care. However, there was a problem in that even when isolated power of equipment was measured, it was only used to estimate the behavior of people with dementia and those receiving care.

[0008] The invention described in Patent Document 3 is a technology for diagnosing equipment based on power waveforms obtained by detecting vibrations in rotating machinery. However, there was a problem in that even though power waveforms were measured, they were only used for equipment diagnosis.

[0009] This invention was conceived in view of these problems, and aims to provide a power data service system and a power data service method that provide power data tailored to the individual user's purpose, for users who find it difficult to invest in equipment to grasp power consumption but want to utilize power consumption information. [Means for solving the problem]

[0010] In the present invention, a node refers to a primary feature extracted from time-series power data for each electrical device, environmental conditions related to power consumption (e.g., time information, calendar information, address information (region, building), temperature information, and humidity information), and information representing the purpose of use, such as power supply and demand forecasting, power adjustment, power distribution network fault detection, monitoring alerts, frailty signals, push advertising timing, industrial equipment failure prediction, or energy saving targets. An edge refers to a connection relationship that shows the relationship between the nodes, for example, the relationship between primary features, the relationship between a primary feature and at least one of the time information, calendar information, address information, temperature information, and humidity information, and the relationship between a primary feature and the purpose of use information. The strength of the relationship between the nodes is assigned to the edge. The strength may be expressed as numerical data, or as a qualitative index that shows a stepped relationship such as strong, medium, or weak.

[0011] The power data service system described in claim 1 is a power data service system that provides users with power data suitable for their intended use, the power data service system comprising: power data separation means for separating and estimating measured power per power meter unit installed in a building for each electrical device; and a cloud server for receiving power separation data from the power data separation means, the storage unit of the cloud server records time-series data of power for each electrical device installed in multiple buildings, which is the power separation data transmitted in real time at predetermined time intervals from the power data separation means, linked to at least one of time information, calendar information, address information, temperature information and humidity information; the feature extraction unit extracts primary feature quantities from the time-series data that represent the portion where changes in the usage status of each electrical device occur, and further extracts the primary feature quantities for each electrical device and the time information, calendar information, address information, temperature information and humidity information from the time information The system extracts secondary features that indicate the strength of the relationship with at least one of the following: the learning data generation unit generates learning data using the time-series data of power, the primary features, and the secondary features; the training data generation unit generates training data by labeling the learning data and assigning usage purpose labels to the primary and secondary features that are determined to be related to a predetermined usage purpose; the trained model generation unit performs machine learning processing using the training data to generate a trained model; the estimation unit inputs the time-series data of power for each electrical device and usage purpose information into the trained model, and further inputs at least one of the following as condition information defining the output target: time information, calendar information, temperature information, humidity information, and address information; estimates the time-series data of power that includes the primary and secondary features best suited to the usage purpose based on the condition information; and outputs the estimated time-series data of power.

[0012] The power data service system according to claim 2 is characterized in that, in claim 1, the secondary feature further indicates the strength of the relationship between the primary feature quantities.

[0013] The power data service system according to claim 3 is a machine learning model in which the trained model generates graph structure data in which the primary features are represented as nodes representing environmental conditions such as time information, calendar information, address information, temperature information, and humidity information, and purpose of use information, and the relationships between the primary features are represented as edges representing the relationship between the primary features and at least one of the time information, calendar information, address information, temperature information, and humidity information, and the relationship between the primary features and the purpose of use information, and estimation is performed based on the graph structure data, wherein the edges indicate the strength of the relationship.

[0014] The power data service system according to claim 4 is characterized in that, in claim 1 or 2, the visualization unit generates and outputs visualization information for displaying the estimated power time series data corresponding to the power time series data as a graph.

[0015] The power data service system according to claim 5 is, in claim 1 or 2, the Power separation data The time interval is characterized by having a lower limit of one of the periods between 0.1 seconds and 1 second, and an upper limit of several years.

[0016] The power data service method described in claim 6 is a power data service method that provides a user with power data suitable for its intended use, comprising: a power data separation step of separating and estimating the measured power of each power meter installed in a building for each electrical device; a receiving step of a cloud server receiving the power separation data separated and estimated by the power data separation step; a storage step of recording the time-series data of power for each electrical device installed in multiple buildings, which is the power separation data received in the receiving step and transmitted in real time at predetermined time intervals, in association with at least one of time information, calendar information, address information, temperature information and humidity information; and extracting primary feature quantities from the time-series data that represent the portion where changes in the usage status of each electrical device occur, and further extracting secondary feature quantities that show the strength of the relationship between the primary feature quantities for each electrical device and at least one of the time information, calendar information, address information, temperature information and humidity information. The system is characterized by comprising: a feature extraction step for extracting features; a training data generation step for generating training data using the time-series data of power, the primary features, and the secondary features; a training data generation step for generating training data by labeling the training data and assigning usage purpose labels to the primary features and secondary features determined to be related to a predetermined usage purpose; a trained model generation step for generating a trained model by performing machine learning processing using the training data; and an estimation step for inputting the time-series data of power for each electrical device and usage purpose information into the trained model, and further inputting at least one of time information, calendar information, temperature information, humidity information, and address information as condition information defining the output target, estimating the time-series data of power containing the primary features and secondary features best suited to the usage purpose based on the condition information, and outputting the estimated time-series data of power.

[0017] The power data service system according to claim 7 is characterized in that, in claim 6, the secondary feature further indicates the strength of the relationship between the primary feature quantities.

[0018] The power data service system according to claim 8 is a machine learning model according to claim 6 or 7 in which the trained model generates graph structure data in which the primary features are represented as nodes representing environmental conditions such as time information, calendar information, address information, temperature information, and humidity information, and purpose of use information, and the relationships between the primary features are represented as edges representing the relationship between the primary features and at least one of the time information, calendar information, address information, temperature information, and humidity information, and the relationship between the primary features and the purpose of use information, and estimates are made based on the graph structure data, wherein the edges indicate the strength of the relationship. [Effects of the Invention]

[0019] The power data service system and power data service method of the present invention can grasp the power usage status for each region, building, or electrical equipment, and transmit the power usage status for each electrical equipment via a network. As a result, the power data service provided by the present invention can be used anytime, anywhere, and has the effect of allowing businesses that do not own a system to acquire the power usage status to easily grasp the power usage status.

[0020] The power data service system and power data service method of the present invention have the effect of providing time-series power data that corresponds to the user's purpose of use, by extracting features from time-series power data, labeling the features estimated to correspond to the aforementioned purpose of use, and generating training data, for use purposes such as power supply and demand forecasting, power adjustment, power distribution network fault detection, monitoring alerts, frailty signals, push advertising timing, industrial equipment failure prediction, or energy saving targets.

[0021] In addition, the power data service system and the power data service method of the present invention structure a plurality of factors related to the magnitude and change of power consumption as nodes, and generate a graph structure model represented as edges indicating the strength of the relationship between the factors, thereby making it possible to analyze the factors highly influential on the magnitude and change of power consumption and the dependency relationship between the factors. Further, by using white-box machine learning, it is possible to visualize the relationship structure between the magnitude and change of power consumption and each factor, which was difficult to grasp in black-box machine learning.

[0022] The power data service system and the power data service method of the present invention can acquire the power data time interval at any period from 0.1 second period to 1 second period, store a huge amount of power data, and utilize it. Therefore, compared with the case where the power data time interval exceeds 1 second period, instantaneous power changes can also be grasped, so that the prediction accuracy of distribution network fault detection, industrial equipment fault prediction, etc. can be improved. Even for enterprises that have difficulty storing and keeping a huge amount of power data with the power data time interval being any period from 0.1 second period to 1 second period, they can easily utilize the huge amount of power data.

[0023] The power data service system and the power data service method of the present invention provide time-series power data within the range corresponding to the usage purpose to enterprises that can process power data according to the usage purpose for users. At the same time, for enterprises that have difficulty processing power data according to the usage purpose in their own companies, it is also possible to provide visualized power data in which time-series power data for each electrical device or a combination of electrical devices includes feature quantities suitable for the usage purpose. Therefore, it has the effect that it can be used by both enterprises that can process power data and enterprises that cannot process it in their own companies.

Brief Description of the Drawings

[0024] [Figure 1] It is a configuration explanatory diagram of the power data service system of the present invention. [Figure 2] This is an explanatory diagram illustrating an example of power data separation means and electrical equipment layout in a single building. [Figure 3] This is an explanatory diagram of a line graph that visualizes the time-series data of power consumption for each electrical appliance in a specified building over a day. [Figure 4] This is a flowchart illustrating the power data service method of the present invention. [Figure 5] This is an explanatory diagram illustrating the relationship between primary features, secondary features, environmental conditions, and intended use. [Figure 6] This diagram shows the weekly electricity consumption of a Washlet (registered trademark) in a certain building (a detached house). [Modes for carrying out the invention]

[0025] The power data service system 1 and power data service method 2 of the present invention can provide time-series power data suitable for use by companies that want to understand and analyze the power usage status of each electrical device 10 in each region and each building 6, for purposes such as power supply and demand forecasting, power adjustment, power distribution network fault detection, monitoring alerts, frailty signals, push advertising timing, industrial equipment failure prediction, or energy saving, but do not have a system in place to understand and analyze the power usage status of each electrical device 10 in each region or each building 6.

[0026] As shown in Figure 1, the power data service system 1 includes a power data separation means 5 that receives instruction information from a cloud server 4 equipped with a non-intrusive load monitoring technology application and separates and estimates the measured power of power meters such as distribution boards installed in many buildings 6 for each electrical device 10, and a cloud server 3 that receives the combined, organized, and compressed power separation data from many of the power data separation means 5 via the cloud server 4. The cloud server 3 includes a transmitting / receiving unit 11, a storage unit 12, a feature extraction unit 13, a training data generation unit 14, a teacher data generation unit 15, a trained model generation unit 16, an estimation unit 17, and a visualization unit 18. The power data separation means 5, the cloud server 3, and the cloud server 4 can communicate with each other via a network 8.

[0027] Furthermore, as shown in Figure 1, the power data service system 1 is able to communicate via network 8 with the service processing unit (energy saving) 21 of cloud server 7a, the service processing unit (monitoring) 22 of cloud server 7b, the service processing unit (power demand forecasting) 23 of cloud server 7c, the service processing unit (power adjustment) 24 of cloud server 7d, the service processing unit (fault detection) 25 of cloud server 7e, the service processing unit (frailty) 26 of cloud server 7f, the service processing unit (push-type advertising timing) 27 of cloud server 7g, and the service processing unit (software development) 28 of cloud server 7h, which constitute the user-side cloud server group 7, and are destinations for time-series power data containing feature quantities suitable for each purpose of use from the cloud server 3.

[0028] Each user can access any of the cloud servers 7a to 7h that make up the cloud server group 7 and have the same purpose of use, to receive time-series power data from the power data service system 1 for each electrical device 10 or for combinations of electrical devices 10. Users can use the time-series power data as is, or they can use the time-series power data to perform further processing using their own proprietary programs to create new added value.

[0029] Users can utilize the time-series power data from the power data service system 1 for each electrical device 10, or for combinations of electrical devices 10, in their business operations. For example, a company that provides monitoring services can contract with households that want to receive monitoring services and develop their monitoring business, and small and medium-sized power companies that want to reflect power demand forecasts in their business plans can make efficient capital investments.

[0030] The present invention provides a power data service system 1 that provides users with time-series power data suitable for their intended use, comprising: a power data separation means 5 that separates and estimates the measured power of each power meter unit installed in a building 6 for each electrical device 10; and a cloud server 3 that receives the power separation data from the power data separation means 5, wherein the storage unit 12 of the cloud server 3 records the time-series power data for each electrical device 10 installed in each of the multiple buildings 6, which is the power separation data transmitted in real time at predetermined time intervals from the power data separation means 5, in association with at least one of the following: time information, calendar information, address information, temperature information, and humidity information; and a feature extraction unit 13 extracts primary feature quantities from the time-series data that represent the portion where changes in the usage status of each electrical device 10 occur, and further extracts the primary feature quantities for each electrical device 10, along with the time information, calendar information, address information, and The learning data generation unit 14 extracts secondary features indicating the strength of the relationship with at least one of the temperature information and humidity information, generates learning data using the power time series data, the primary features and the secondary features, generates training data using the power time series data, the primary features and the secondary features, generates training data using the training data, labels the primary features and secondary features determined to be related to a predetermined purpose of use, and generates training data using the training data. The trained model generation unit 16 performs machine learning processing using the training data and generates a trained model, and the estimation unit 17 inputs the power time series data and purpose of use information into the trained model, and further inputs at least one of the time information, calendar information, temperature information, humidity information and address information as condition information that defines the output target, estimates the power time series data containing the primary features and secondary features that are optimal for the purpose of use based on the condition information, and outputs the estimated power time series data.

[0031] The structure representing the aforementioned relationships does not necessarily have to be explicitly constructed as a graph structure; it may be a black-box type that learns the relationships between features as internal parameters but makes it difficult to directly interpret those relationships, or it may be a white-box type that is explicitly constructed as a graph structure and learns as a model capable of explicitly representing the relationships between features. Examples of such machine learning models include Transformer, Attention, and DNN (Deep Neural Network) for the black-box type, and graph AI for the white-box type.

[0032] If the structure representing the aforementioned relationships is of the white-box type, the trained model is a machine learning model that generates graph structure data in which primary features extracted from time-series power data for each electrical device, environmental conditions related to power consumption (e.g., time information, calendar information, address information (region, building), temperature information, and humidity information), and information representing the purpose of use, such as power supply and demand forecasting, power adjustment, power distribution network fault detection, monitoring alerts, frailty signals, push advertising timing, industrial equipment failure prediction, or energy saving targets are represented as nodes, the relationships between the primary features are represented as edges, the relationships between the primary features and at least one of the time information, calendar information, address information, temperature information, and humidity information are represented as edges, and the relationships between the primary features and the purpose of use are represented as edges, and estimates are made based on the graph structure data, with the edges indicating the strength of the relationships. In other words, the graph structure data is a data structure that includes nodes and edges representing the strength of the relationships between nodes. Examples of machine learning models that process the graph structure data include graph AI such as graph neural networks and graph transformers.

[0033] The power data separation means 5 is installed in each building 6 and can be easily attached to power measuring instrument units such as distribution boards connected to each electrical equipment 10, and has a transmission and reception function. The power data separation means 5 receives separation instructions from the power data separation application on a cloud server 4 equipped with a non-intrusive load monitoring technology application that separates and estimates the measured power for each electrical equipment 10, separates and estimates the measured power for each electrical equipment 10, and transmits the time-series data of the separated and estimated power to the cloud server 4. The time-series data of many separated and estimated powers from many locations are compiled, organized, and compressed, and the time-series data of the power is continuously transmitted in real time from the cloud server 4 to the cloud server 3. As an example of the intrusive load monitoring technology application on the cloud server 4, there is a non-intrusive load monitoring application from SENSE, but any non-intrusive load monitoring application is acceptable as long as the time interval of the time-series data of the separated and estimated power is 0.1 seconds to 1 second or close to that period.

[0034] The power data separation means 5 can be any device capable of collecting time-series data of power separated and estimated at time intervals of 0.1 seconds to 1 second, for example, 0.1 seconds, 0.3 seconds, 0.5 seconds, 0.8 seconds, 1 second, etc. The power data separation means 5 separates and estimates the power consumption of devices such as power measuring instruments, such as distribution boards, for each electrical device 10 at one of the aforementioned time intervals. An example of the power data separation means 5 is the CK sensor (manufactured by China Instruments Industry Co., Ltd., product name), which is scheduled to be released. The power data separation means 5 can separate and estimate the measured power of power measuring instruments for each electrical device 10 at time intervals of 0.1 seconds to 1 second.

[0035] The aforementioned electrical equipment 10 can be any electrical equipment 10 that uses electricity, for example, as shown in Figure 2, there are microwave ovens 10a, electric rice cookers 10b, air conditioners 10c, refrigerators 10d, washing machines 10e, vacuum cleaners 10f, etc. Furthermore, although not shown, factories may have equipment such as compressors, painting equipment, machining centers, etc.

[0036] In this invention, the address information includes both a region and a building 6, as the region can be set from the address information. The building 6 can be any building 6 that uses electricity, such as a commercial facility, factory, office building, apartment building, or general house including a detached house. The region can be a geographical area such as all of Japan, a region, prefecture, city, town, or any region, and can be freely set by setting the address.

[0037] The transmitting / receiving unit 11 receives time-series power data for each electrical device 10 from the power data separation means 5 via the network 8 and the cloud server 4, and transmits time-series power data for each electrical device 10, or for combinations of electrical devices 10, which contains feature quantities suitable for their respective purposes, to each of the cloud servers 7a to 7h in the cloud server group 7.

[0038] The storage unit 12 records the time-series power data for each electrical device 10 installed in each of the multiple buildings 6, which is the power separation data transmitted in real time at predetermined time intervals from the power data separation means 5, and associates it with at least one of the following: time information, calendar information, address information, temperature information, and humidity information.

[0039] The storage unit 12 stores in real time and sequentially the time-series power data of each electrical device 10 that has been separated and estimated from the power data separation means 5 at a predetermined time period, for example, a period of 0.1 seconds to 1 second, so it stores a vast amount of power data.

[0040] The aforementioned time information is synchronized with the OS time, which is automatically corrected for advance or lag by time synchronization; the aforementioned calendar information is acquired, for example, by information automatically calculated from the OS time by a predetermined program; the aforementioned address information is input, for example, when the power data separation means 5 is installed; and the aforementioned temperature information and humidity information are input, for example, by selecting a predetermined region and obtaining the aforementioned temperature information and humidity information from the Japan Meteorological Agency. Any method is acceptable as long as the aforementioned time information, calendar information, address information, temperature information, or humidity information can be obtained, and is not limited to the methods described above.

[0041] Next, the feature extraction unit 13 extracts primary feature quantities from the time-series power data for each electrical device 10, representing the portion where changes in power usage occur for each electrical device 10. Furthermore, it extracts secondary feature quantities that indicate the strength of the relationship between the primary feature quantities for each electrical device 10 and at least one of the following: time information, calendar information, address information (regional information, building 6 information), temperature information, and humidity information.

[0042] The primary feature is a feature of the change in time-series data of power for each electrical device 10. As the primary feature, at least one feature is extracted from among, for example, the extraction of a trend / momentum feature that is characterized by a change in the amount of power that is rapidly or slowly increasing or rapidly or slowly decreasing over time, the extraction of a frequency domain feature that is characterized by the period of change in the amount of power, the extraction of a pattern complexity feature that is characterized by the regularity or irregularity observed in the waveform of the amount of power, and the extraction of an energy distribution feature that is characterized by where the use of power is concentrated.

[0043] From the time-series data of the power for each of the electrical devices 10, the trend / momentum features are extracted based on, for example, a moving average or time difference; the frequency domain features are extracted based on, for example, a spectrum estimated by FFT (Fast Fourier Transform); the pattern complexity features are extracted based on, for example, graph AI or entropy, which is an indicator of the degree of irregularity in the time-series pattern of power consumption; and the energy distribution features are extracted based on, for example, graph AI or a bandwidth power distribution based on PSD (Power Spectral Density). The methods for extracting the trend / momentum features, frequency domain features, pattern complexity features, or energy distribution features are not limited to these methods, and any method capable of extracting each of the features may be used.

[0044] Examples of primary features of the time-series power data for each of the electrical devices 10 include on / off characteristics associated with the start or stop of power use, characteristics of a rapid increase or decrease from normal power use, characteristics of a gradual increase or decrease from normal power use, characteristics of instantaneous increase or decrease in power use, characteristics of maintaining approximately the same power use over a certain period, characteristics of repeatedly increasing or decreasing power use periodically, characteristics of irregular power use, characteristics of turning off even though approximately constant power use should continue automatically, characteristics of observing power use patterns different from the norm, and characteristics of high or low power use.

[0045] The primary features of the aforementioned time-series power data show a relationship with environmental conditions, such as temperature information, humidity information, time of day information, calendar information, and address information (region, building). The amount of power used by air conditioners fluctuates depending on the temperature, the amount of power used by dehumidifiers fluctuates depending on the humidity, the amount of power used by televisions fluctuates depending on whether it is a sleeping or awake time, the amount of power used in homes and factories fluctuates depending on whether it is a holiday or a weekday, or on the calendar information of spring, summer, autumn, or winter, and there are differences in power usage depending on whether it is an industrial area or a residential area. Therefore, there is a relationship between the primary features of the aforementioned time-series power data for each of the aforementioned electrical devices 10 and the items of environmental conditions, and the strength of this relationship differs for each of the aforementioned electrical devices 10 and the items of environmental conditions.

[0046] Furthermore, the primary features of the aforementioned time-series power data show relationships with the intended uses, such as power supply and demand forecasting, power adjustment, power distribution network fault detection, monitoring alerts, frailty signals, push advertising timing, industrial equipment failure prediction, or energy saving targets. For example, power supply and demand forecasting is estimated to be related to the characteristics of a gradual increase or decrease from normal power usage; power adjustment is estimated to be related to the characteristics of a rapid increase or decrease from normal power usage, and the characteristics of a gradual increase or decrease from normal power usage; power distribution network fault detection is estimated to be related to the characteristics of a gradual increase or decrease from normal power usage, the characteristics of a momentary increase or decrease in power usage, the characteristics of maintaining almost the same power usage over a certain period, the characteristics of repeatedly increasing and decreasing power usage periodically, and the characteristics of irregular power usage; monitoring alerts are estimated to be related to the characteristics of irregular power usage, the characteristics of turning off even though power usage should automatically remain almost constant, and the characteristics of observing power usage patterns different from the norm; and frailty signals are estimated to be related to the characteristics of periodic power usage Relationships are estimated between the characteristics of repeated increases and decreases in dosage, irregular power consumption, automatic shut-off despite expected to maintain a nearly constant power consumption, and unusual power consumption patterns. The timing of push advertising is estimated to be related to the characteristics of high or low power consumption. Industrial equipment failure prediction is estimated to be related to the characteristics of a sudden increase or decrease from normal power consumption, a gradual increase or decrease from normal power consumption, a momentary increase or decrease in power consumption, maintaining nearly the same power consumption over a certain period, periodic increases and decreases in power consumption, and irregular power consumption. Energy saving targets are estimated to be related to the characteristics of a gradual increase or decrease from normal power consumption and high or low power consumption.

[0047] To explain the aforementioned relationship in more detail, examples of the primary features include: when the purpose of use is power supply and demand forecasting, for example, the time-series data portion for each electrical device 10 where there is an increase or decrease in power usage; when the purpose of use is power adjustment, for example, the time-series data portion for each electrical device 10 where there is a period of high power usage; when the purpose of use is power distribution network fault detection, for example, the time-series data portion for each electrical device 10 where the power usage of a specific wiring network fluctuates instantaneously; when the purpose of use is monitoring alerts, for example, the time-series data portion for each electrical device 10 where the usage status of a specific electrical device in a specific house differs from normal; and when the purpose of use is frailty signaling In some cases, for example, the portion of the time-series data for each electrical device 10 where the usage status of a specific electrical device in a specific house differs from normal conditions would be relevant; in cases where the purpose of use is push advertising timing, for example, the portion of the time-series data for each electrical device 10 where the power usage level of a specific electrical device in a selected area is higher or lower than a generally expected level would be relevant; in cases where the purpose of use is industrial equipment failure prediction, for example, the portion of the time-series data for each electrical device 10 where the power usage of a specific piece of equipment fluctuates instantaneously would be relevant; and in cases where the purpose of use is energy saving, for example, the portion of the time-series data for each electrical device 10 where the power usage level of a specific electrical device in a selected area is higher than a generally expected level would be relevant. The above examples of primary features are merely illustrative and are not limited thereto.

[0048] The secondary features indicate the strength of the relationship between the primary features of each electrical device 10 and at least one of the environmental conditions considered to be related to the amount of electricity used, namely the time information, calendar information, address information, temperature information, and humidity information, and further indicate the strength of the relationships between the primary features. The strength may be expressed as numerical data, or as a qualitative index showing a stepped relationship such as strong, medium, or weak.

[0049] The amount of electricity used, whether high or low, or increasing or decreasing, is related to time information indicating whether it is daytime or nighttime, calendar information indicating the season (spring, summer, autumn, winter) and working days / holidays, regional information indicating whether buildings are densely packed or scattered, and whether there are many residences or factories, and temperature information indicating the degree of coldness or heat, and humidity information indicating the degree of rain or sunshine. Therefore, secondary features are extracted that show the strength of the relationship between the primary features and at least one of the following: the time information, the calendar information, the address information (regional information, building information), the temperature information, and the humidity information.

[0050] Furthermore, the secondary features also indicate the strength of the relationships between each of the primary features among the trend / momentum features, frequency domain features, pattern complexity features, and energy distribution features. For example, when the temperature changes, some features change relatively large, while others do not change much. This allows us to understand whether the relationship between each feature is strong or weak. This makes it possible to cover all the primary and secondary features that have relationships depending on the purpose of use.

[0051] For example, when electricity usage is on an increasing trend, trend / momentum features and energy distribution features become prominent in the time series data, and these features can be expected to have a meaningful effect on purposes such as electricity supply and demand forecasting and electricity adjustment. For example, when electricity usage is irregular, frequency domain features and pattern complexity features become prominent in the time series data, and these features can be expected to have a meaningful effect on purposes such as monitoring alerts and frailty signals. For example, when electricity usage shows an abnormal value with a sudden high, trend / momentum features, pattern complexity features, or frequency domain features become apparent in the time series data, and these features can be expected to have a meaningful effect on purposes such as power distribution network fault detection or industrial equipment failure prediction. Furthermore, comparing the daily electricity consumption of each of the aforementioned electrical devices 10 in each region using, for example, trend-momentum features or energy distribution features, can be expected to have a meaningful effect on the purpose of using, for example, the timing of push-type advertisements. Similarly, comparing the electricity consumption of each piece of equipment within a single factory using, for example, trend-momentum features or energy distribution features, can be expected to have a meaningful effect on the purpose of using, for example, energy conservation.

[0052] As shown in Figure 5, the relationship between the aforementioned secondary features is illustrated as a relationship diagram between primary features extracted from time-series data of electrical equipment 10 (including facilities) such as air conditioners 10c, washing machines 10e, vacuum cleaners 10f, lighting equipment, and microwave ovens 10a, and secondary features that show the strength of the relationship with environmental conditions related to power consumption, such as time information, calendar information, address information (region, building), temperature information, and humidity information. Figure 5 also illustrates the relationship between secondary features that show the relationship between the primary features. Furthermore, Figure 5 is an explanatory diagram for explaining the relationship between the primary and secondary features and their intended uses, such as power supply and demand forecasting, power adjustment, power distribution network failure detection, monitoring alerts, frailty signals, push advertising timing, industrial equipment failure prediction, or energy saving. For estimating such relationships, machine learning methods such as DNN (Deep Neural Network), LSTM (Long Shot Term Memory), or graph AI can be used. Furthermore, the electrical equipment 10 (including facilities) can be increased or decreased as needed, the environmental conditions can be increased or decreased or changed as needed, and the purpose of use can be increased or decreased as needed.

[0053] As shown in Figure 5, the primary feature, environmental conditions, and purpose of use can be represented as nodes, and the lines connecting the primary feature, environmental conditions, and purpose of use can be represented as edges indicating the relationships between nodes. The relationships between multiple nodes can be represented as a graph structure using the nodes and edges. The strength of the relationships between each node can be indicated by the edges. This makes it possible to analyze the nodes that are factors influencing power consumption, and the dependencies between each node, as a graph structure, and the strength of the relationships can be displayed, for example, by the difference in line thickness. For analyzing such a graph structure, graph AI can be used, for example. In addition, all of the nodes can be arbitrarily changed or added or removed.

[0054] Next, the learning data generation unit 15 generates learning data using the time-series data of the power for each electrical device 10, the primary features, and the secondary features.

[0055] Next, the training data generation unit 16 generates training data by labeling the training data, assigning purpose labels to the primary and secondary features that are determined to be related to a predetermined purpose of use. This labeling is performed using, for example, a neural network or a random forest.

[0056] Furthermore, the labeling process involves streaming, which sequentially labels the time-series power data for each electrical device 10 at predetermined time intervals as it is received. As a result, the most recent time-series power data for each electrical device 10 is sequentially labeled, allowing the user to understand the time-series power data for each electrical device 10.

[0057] For example, the same purpose label is assigned to the primary and secondary features that are determined to be related to each user's purpose of use, such as power supply and demand forecasting, power adjustment, power distribution network fault detection, monitoring alerts, frailty signals, push advertising timing, industrial equipment failure prediction, or energy saving.

[0058] Next, the trained model generation unit 17 performs machine learning processing using the training data to generate a trained model.

[0059] The trained model is a machine learning model that performs estimation using input data as input, which includes the primary features, the relationships between the primary features, and information showing the relationship between the primary features and at least one of the following: time information, calendar information, address information (region, building), temperature information, and humidity information. The input data includes the primary features and the secondary features.

[0060] Furthermore, the trained model can also be a machine learning model that performs estimation using input data configured as a graph structure, where the primary features are represented as nodes, the relationships between the primary features are represented as edges, and the relationships between the primary features and at least one of the following are represented as edges: time information, calendar information, address information (region, building), temperature information, and humidity information. The edges indicate the strength of the relationships.

[0061] Next, the estimation unit 17 inputs time-series data of power for each electrical device 10 and information on the purpose of use into the trained model, and further inputs at least one of the following as condition information defining the output target: time information, calendar information, temperature information, humidity information, and address information. Based on the condition information, it estimates time-series data of power that includes the primary and secondary features best suited to the purpose of use, and outputs the estimated time-series data of power. The items of the condition information are the same as the items of the environmental conditions.

[0062] The time information is the time period of the power time series data that the trained model is to output, and the calendar information is the period of the power time series data that the trained model is to output.

[0063] The aforementioned information on purposes of use includes, for example, power supply and demand forecasting, power adjustment, power distribution network fault detection, monitoring alerts, frailty signals, push advertising timing, industrial equipment failure prediction, or energy saving. The aforementioned purposes of use are not limited to these and can be added or deleted at will.

[0064] The aforementioned region is, for example, the range of any area such as all of Japan, a region, a prefecture / city / town / village, and a local area. The aforementioned building 6 is, for example, a commercial facility, a factory, an office building, an apartment building, or a general house including a detached house, and it is also possible to specify one or more buildings 6. Furthermore, the target of the aforementioned electrical equipment 10 can be any electrical equipment 10 that uses electricity, such as an electrical equipment 10 such as a washing machine 10e installed in a home or an electrical equipment 10 such as a machining center installed in a factory. The aforementioned power data time interval can be set to the desired time interval for summarizing power time series data, ranging from a 0.1-second period to a 1-second period. For example, it can be set to a 0.1-second period, a 0.3-second period, a 0.5-second period, a 0.8-second period, a 1-second period, a daily period, a weekly period, a monthly period, a yearly period, or a period of several years.

[0065] By inputting information about the intended use and conditional information defining the output target into the aforementioned trained model, it is possible to output time-series power data suitable for the intended use.

[0066] For example, when using the power data service system 1 for power supply and demand forecasting, if address information (regional information) and temperature information for a certain region are input as conditional information to the trained model, it is possible to estimate and output time-series data of power that includes features effective for power demand forecasting.

[0067] For example, the trained model may accept time-series data of power for each electrical device 10 and information on the purpose of use for power supply and demand forecasting, and perform estimation based on data including the primary features, the relationships between the primary features, and information showing the relationship between the primary features and at least one of the following: time information, calendar information, address information (region, building 6), temperature information, and humidity information. In this case, for example, by inputting information on the purpose of use, such as power demand forecasting, and further inputting address information (regional information) and temperature information for a certain region as conditional information, power demand forecast data can be output based on time-series data of power including trend and momentum features, which are primary features effective for power demand forecasting.

[0068] Furthermore, for example, the trained model may perform estimation based on graph structure data. In this case, the graph structure data is represented by nodes for the primary features, at least one of the following as nodes: time information, calendar information, address information (regional information, building information), temperature information, and humidity information, and purpose of use information, with the relationships between these nodes represented as edges. For example, by inputting the purpose of use information, as well as conditional information for a certain region's address information (regional information, building information) and temperature information, into the trained model, estimation is performed based on the relationship between the primary features and the conditional information, and the relationship between the primary features and the purpose of use information, and power demand forecast data is output. In the case of the trained model, since it is based on graph structure data, the strength of the relationships can be displayed, for example, by the thickness of the lines.

[0069] Next, the trained model receives time-series data of power consumption for each electrical device 10 and information on the purpose of use for power supply and demand forecasting. If the model performs estimation based on data including primary features, relationships between primary features, and information showing the relationship between primary features and at least one of the following: time information, calendar information, address information (region, building 6), temperature information, and humidity information, for example, by inputting information on the purpose of use, such as a monitoring alert, and further inputting address information for a detached house and calendar information for a certain week as condition information, it can output time-series data of power consumption estimated to be related to the condition information and the purpose of use information, based on time-series data of power consumption including frequency domain features and pattern complexity features, which are primary features effective for monitoring alerts. For example, as shown in Figure 6, time-series data of power consumption that has the characteristic of showing a different power consumption pattern from the normal day, such as the power consumption of the bidet being high in the early morning during the specified week, is visualized as a bar graph for each time period and output.

[0070] Furthermore, if the trained model performs estimation based on graph structure data, the graph structure data is represented as follows: the primary features are nodes, at least one of the following is nodes: time information, calendar information, address information (regional information, building information), temperature information, and humidity information; and purpose of use information is also a node, with the relationships between these nodes represented as edges. For example, if the trained model is input with purpose of use information such as monitoring alerts, and further input with address information of a detached house and calendar information for a week as condition information, estimation is performed based on the graph structure data including the nodes and edges, based on the relationship between the primary features and the condition information, and the relationship between the primary features and the purpose of use information, and time-series data of power estimated to be related to monitoring alerts is output. In the case of the trained model, since it is based on graph structure data, the strength of the relationships can be displayed, for example, by the thickness of the lines.

[0071] The output is transmitted by the transmitting / receiving unit 11 to one of the cloud servers 7a to 7h in the cloud server group 7 that serve the same purpose.

[0072] Next, the visualization unit 18 generates visualization information for displaying the estimated time-series data of the power as a graph, and transmits it from the transmitting / receiving unit 11.

[0073] Figure 3 shows an example of time-series power data visualized in a graph by the visualization unit 18. The example is a trained model that receives time-series power data for each electrical appliance 10 and information on the purpose of use of the monitoring alert, and performs estimation based on data including the primary features, the relationships between the primary features, and information showing the relationship between the primary features and at least one of the following: time information, calendar information, address information (regional information, building information), temperature information, and humidity information. If the range of the building 6 is selected as a single-family home, and the range of electrical appliances 10 is selected as the individual display of microwave oven 10a, electric rice cooker 10b, air conditioner 10c, refrigerator 10d, washing machine 10e, vacuum cleaner 10f, and solar power generation, and one day is selected as the calendar information, then, as shown in Figure 3, a line graph 9 of the power usage at one-minute intervals for the selected day can be created for each electrical appliance 10 of the selected single-family home. For example, time-series electricity data that includes complexity features, such as the pattern of changes in time-series complexity, indicating that a microwave oven is used twice a day, reveals that the person is living their life as usual.

[0074] Next, the power data service method 2 will be described. As shown in Figure 4, the power data service method 2 is a power data service method 2 that provides users with power data suitable for their intended use, and includes a power data separation step 31 that separates and estimates the measured power of each power meter unit installed in the building 6 for each electrical equipment 10, a receiving step 32 in which the power separation data separated and estimated by the power data separation step 31 is received by the cloud server 3, a storage step 33 that records the time-series data of power for each electrical equipment 10 installed in each of the multiple buildings 6, which is the power separation data received in the receiving step 32 and transmitted in real time at a predetermined time period, in association with at least one of the following: time information, calendar information, address information, temperature information and humidity information, and a primary feature quantity that represents the portion where the usage status of each electrical equipment 10 changes from the time-series data, and further the relationship between the primary feature quantity for each electrical equipment 10 and at least one of the following: time information, calendar information, address information, temperature information and humidity information. The system includes: a feature extraction step 34 for extracting secondary features that indicate the intensity of; a training data generation step 35 for generating training data using the power time series data, the primary features, and the secondary features; a training data generation step 36 for generating training data by labeling the training data and assigning usage purpose labels to the primary features and secondary features that are determined to be related to a predetermined usage purpose; a trained model generation step 37 for generating a trained model by performing machine learning processing using the training data; and an estimation step 38 for inputting the power time series data and usage purpose information for each of the electrical devices 10 into the trained model, and further inputting at least one of time information, calendar information, temperature information, humidity information, and address information as condition information that defines the output target, estimating the power time series data containing the primary features and secondary features that are optimal for the usage purpose based on the condition information, and outputting the estimated power time series data.

[0075] In the power data service method 2 described above, the secondary feature further indicates the strength of the relationship between the primary feature. The strength may be expressed as numerical data, or as a qualitative index indicating a stepped relationship such as strong, medium, or weak.

[0076] In the power data service method 2 described above, the trained model is a machine learning model that generates graph structure data in which the primary features are represented as nodes, the relationships between the primary features, and the relationships between the primary features and at least one of the time information, calendar information, address information, temperature information, and humidity information as edges, and performs estimation based on the graph structure data, wherein the edges have the strength of the relationship. The strength may be represented as numerical data, or as a qualitative index indicating a stepped relationship such as strong, medium, or weak.

[0077] The system also includes a visualization step 39 for generating visualization information to display the estimated time-series data of power as a graph. The visualization step 39 visualizes the time-series data of power, which includes the primary and secondary features according to the purpose of use, generated in the training data generation step 37, and / or visualizes the time-series data of power, which includes the primary and secondary features according to the purpose of use, generated in the trained model generation step 37.

[0078] The system also includes a power time-series data output step 40 that transmits power time-series data to the user's cloud server group 7. The power time-series data output step 40 transmits power time-series data for each purpose of use, labeled in the training data generation step 37 which assigns purpose of use labels, to the cloud server group 7, transmits power demand forecast data, which predicts power demand using the trained model generated in the trained model generation step 38, to the cloud server group 7, and transmits power time-series data visualized in graphs 9 such as line graphs and bar graphs in the visualization step 39 to the cloud server group 7.

[0079] In the power time-series data output process 40, for example, the time-series power data containing the primary and secondary features for each usage purpose is transmitted to the service processing unit (energy saving) 21 of cloud server 7a, the service processing unit (monitoring) 22 of cloud server 7b, the service processing unit (power demand forecasting) 23 of cloud server 7c, the service processing unit (power adjustment) 24 of cloud server 7d, the service processing unit (fault detection) 25 of cloud server 7e, the service processing unit (frailty) 26 of cloud server 7f, the service processing unit (push advertising timing) 27 of cloud server 7g, and the service processing unit (software development) 28 of cloud server 7h. [Explanation of Symbols]

[0080] 1. Power Data Service System 2. Power Data Service Method 3. Cloud Server 4. Cloud Server 5. Power Data Separation Means 6 Buildings 7 Cloud Server Cluster 8 Networks 9 Graphs 10 Electrical equipment 11 Transmitter / Receiver 12 Storage section 13 Feature Extraction Unit 14. Training Data Generation Unit 15. Training Data Generation Unit 16. Pre-trained model generation unit 17 Estimation part 18 Visualization part 21 Service Processing Unit (Energy Saving) 22 Service Processing Unit (Monitoring) 23. Service Processing Unit (Power Demand Forecasting) 24. Service Processing Unit (Power Adjustment) 25. Service Processing Unit (Fault Prediction) 26. Service Processing Unit (Frailty) 27. Service Processing Unit (Push Ad Timing) 28. Service Processing Unit (Software Development) 31 Power Data Separation Process 32 Receiving process 33 Memory process 34 Feature Extraction Process 35. Training Data Generation Process 36. Training Data Generation Process 37. Process for generating trained models 38 Estimation process 39 Visualization process 40 Power Time-Series Data Output Process

Claims

1. A power data service system that provides users with power data suitable for their intended use, The aforementioned power data service system comprises a power data separation means for separating and estimating the measured power of each power meter installed in a building, and a cloud server for receiving the power separation data from the power data separation means. The aforementioned cloud server is equipped with, The storage unit records the time-series power data for each electrical device installed in multiple buildings, which is the power separation data transmitted in real time at predetermined time intervals from the power data separation means, linked to at least one of the following: time information, calendar information, address information, temperature information, and humidity information. The feature extraction unit extracts primary features from the time-series data that represent the portion where changes in the usage status of each electrical device occur, and further extracts secondary features that show the strength of the relationship between the primary features of each electrical device and at least one of the following: time information, calendar information, address information, temperature information, and humidity information. The training data generation unit generates training data using the time-series data of power, the primary features, and the secondary features. The training data generation unit generates training data by labeling the training data by assigning purpose labels to the primary and secondary features that are determined to be related to a predetermined purpose of use. The trained model generation unit performs machine learning processing using the training data to generate a trained model. The power data service system is characterized in that the estimation unit inputs time-series data of power for each electrical device and usage purpose information to the trained model, and further inputs at least one of time information, calendar information, temperature information, humidity information, and address information as condition information that defines the output target, estimates time-series data of power that includes the primary and secondary features best suited to the usage purpose based on the condition information, and outputs the estimated time-series data of power.

2. The power data service system according to claim 1, characterized in that the secondary features further indicate the strength of the relationship between the primary features.

3. The power data service system according to claim 1 or 2, characterized in that the trained model generates graph structure data in which the primary features are represented as nodes representing environmental conditions such as time information, calendar information, address information, temperature information, and humidity information, and purpose of use information, and the relationships between the primary features are represented as edges representing the relationship between the primary features and at least one of the time information, calendar information, address information, temperature information, and humidity information, and the relationship between the primary features and the purpose of use information, and performs estimation based on the graph structure data, the edges indicating the strength of the relationship.

4. The power data service system according to claim 1 or 2, characterized in that the visualization unit generates and outputs visualization information for displaying the estimated power time series data corresponding to the power time series data as a graph.

5. The power data service system according to claim 1 or 2, characterized in that the time interval of the power separation data is limited to a period of 0.1 seconds to 1 second, with a lower limit of a period of several years and an upper limit of several years.

6. A power data service method that provides users with power data suitable for their intended use, A power data separation process that separates and estimates the measured power of each electrical device from the power meters installed within the building, A receiving step in which a cloud server receives the power separation data separated and estimated by the power data separation step, A storage step in which the power separation data received in the receiving step, transmitted in real time at predetermined time intervals, is time-series data of power for each electrical device installed in multiple buildings, is recorded in association with at least one of the following: time information, calendar information, address information, temperature information, and humidity information. A feature extraction step is performed to extract primary feature quantities from the aforementioned time-series data that represent the portion where changes in the usage status of each electrical device occur, and further extract secondary feature quantities that show the strength of the relationship between the primary feature quantities for each electrical device and at least one of the following: time information, calendar information, address information, temperature information, and humidity information. A training data generation step that generates training data using the aforementioned time-series data of power, the aforementioned primary features, and the aforementioned secondary features, A training data generation step involves generating training data by labeling the training data by assigning purpose labels to the primary and secondary features that are determined to be related to a predetermined purpose of use. A trained model generation step involves performing machine learning processing using the aforementioned training data to generate a trained model, A power data service method characterized by comprising: inputting time-series data of power for each electrical device and usage purpose information into the trained model; further inputting at least one of time information, calendar information, temperature information, humidity information, and address information as condition information defining the output target; estimating time-series data of power containing the primary and secondary features best suited to the usage purpose based on the condition information; and outputting the estimated time-series data of power.

7. The power data service method according to claim 6, characterized in that the secondary feature further indicates the strength of the relationship between the primary feature.

8. The power data service method according to 6 or 7, characterized in that the trained model is a machine learning model that generates graph structure data in which the primary features are represented as nodes, with the environmental conditions of the time information, calendar information, address information, temperature information, and humidity information as nodes, and the relationships between the primary features are represented as edges, with the relationships between the primary features and at least one of the time information, calendar information, address information, temperature information, and humidity information, and the relationships between the primary features and the purpose of use information as edges, and performs estimation based on the graph structure data, wherein the edges indicate the strength of the relationships.

Images