CN103018611B - Non-invasive load monitoring method and system based on current decomposition - Google Patents

Abstract

The present invention provides a non-intrusive load monitoring method and system based on current decomposition. The method includes: step 1, train each electric device, obtain the statistical manifold of each electric device, and The above statistical manifold is stored in the database; step 2, install the acquisition unit at the entrance of the power grid to obtain the power consumption data of all electrical equipment in real time, and convert the power consumption data into a vector representation of the calculation statistical manifold to obtain the total current The set of vectors; step 3, the calculation unit loads the statistical manifold of each electric device, and decomposes the total statistical manifold according to the statistical manifold, so as to obtain the real-time current vector of each electric device. The present invention directly decomposes the collected total current to each device, no longer estimates the complex operating state, but directly uses the current information to calculate the detailed power consumption information on the relevant equipment, so as to ensure accurate and detailed power consumption information .

Description

A non-intrusive load monitoring method and system based on current decomposition

technical field

The invention relates to the technical field of computer applications, in particular to a non-invasive load monitoring method and system based on current decomposition.

Background technique

In 2008, IBM proposed the concept of "Smart Earth", describing it as "more thorough perception, more comprehensive interconnection, and deeper intelligence". The US government also put smart grid as an important part of its "new energy rescue plan" on the agenda shortly after Smart Earth was proposed. Compared with the traditional grid that we are familiar with, the smart grid means to obtain as much information as possible, pay more attention to the interaction between users, and provide better services by analyzing information. User-end devices such as smart meters directly contact users and guide users' electricity consumption behaviors, which is an important manifestation of the difference between smart grids and traditional grids. By giving detailed power consumption information on the relevant equipment, it can effectively reduce the user's understanding of the power consumption behavior, optimize the user's usage habits, and obtain better power saving effects.

Therefore, effectively obtaining detailed power consumption information on related devices in the power consumption environment (family, production environment, etc.) is a key technology for user-side information collection in the smart grid field. The monitoring technology that obtains detailed power consumption information on each device from the outside without affecting the power consumption environment is called non-intrusive load monitoring (NILM) technology. So far, non-intrusive load monitoring technologies mainly fall into two categories:

1. Load monitoring technology based on steady-state analysis: This type of technology first defines multiple operating states (steady-state) for each electrical equipment, and establishes corresponding characteristics for each operating state during the training phase; after starting monitoring, By comparing the collected global information with the known feature set, the operating status of all electrical equipment in the current state is obtained; finally, according to the predefined operating status, detailed electrical information on the equipment is given.

2. Load monitoring technology based on transient events: This type of technology first defines multiple transient events for each electrical equipment, and establishes corresponding characteristics for each transient event in the training phase; Compare the collected global information with the known feature set to judge whether a transient event occurs in the current state; when a transient event occurs, modify the operating state of the corresponding device according to the event definition, and finally give the information on the relevant device. detailed electricity usage information.

Although both of the above two technologies support non-intrusive load monitoring and meet the needs of user-end information collection in the smart grid field, both technologies make similar assumptions about electrical equipment: the equipment has a relatively stable operating state, and the After the running state, the detailed power consumption information of the equipment can be obtained according to the known information. With the progress of the times, the behavior of electrical equipment is becoming more and more flexible, which makes this assumption no longer applicable, and accurate electricity consumption information cannot be obtained. For example: in the same period of time, a computer running game programs consumes more power than simply browsing the web, and the difference between idle and busy power consumption may exceed 30%.

Contents of the invention

Aiming at the flexible power consumption behavior of electrical equipment, this method models the phase space of the current passing through the equipment, and directly decomposes the collected total current into each equipment. Instead of estimating the complex operating state, the current information is directly used to calculate the detailed power consumption information on the equipment.

In order to achieve the above object, the present invention provides a non-intrusive load monitoring method based on current decomposition, the method comprising:

Step 1, train each electrical device, obtain the statistical manifold of each electrical device, and store the statistical manifold in a database;

Step 2, installing a collection unit at the entrance of the power grid to obtain the power consumption data of all power consumption equipment in real time, and converting the power consumption data into a vector representation of the calculation statistical manifold to obtain a vector set of the total current;

Step 3, the calculation unit loads the statistical manifold of each electric device, and decomposes the total statistical manifold according to the statistical manifold, so as to obtain the real-time current vector of each electric device.

Further, the step 1 includes:

Step 11, collecting the current and voltage signals of each electrical device when it is working normally;

Step 12, vectorizing the current and voltage signals to obtain a vector set for each electrical device;

Step 13, according to the set of vectors, construct the statistical manifold of the electrical equipment until the training of all electrical equipment is completed.

Further, said step 2 includes:

Step 21, the acquisition unit acquires the current and voltage signals of all electrical equipment in real time;

Step 22, the calculation unit vectorizes the current and voltage signals to obtain a vector set of the total current.

Further, said step 3 includes:

Step 31, the calculation unit loads the statistical manifold of each electric device, and selects the optimization algorithm used for decomposition according to the properties of the statistical manifold;

Step 32, using the optimization algorithm, combined with the statistical manifold of each electric device, decomposes the vector set of the total current to obtain the current on each device.

The non-intrusive load monitoring method also includes:

Step 4, after completing the decomposition, perform error analysis on the decomposition results;

Step 5, if the error is within the acceptable range, the calculation unit calculates the power of each electric device according to the real-time current vector of each electric device, and transmits the power to the display unit; otherwise, abnormal processing is performed.

To achieve the above purpose, the present invention also provides a non-invasive load monitoring system based on current decomposition, which includes:

A training module, which trains each electrical device, obtains the statistical manifold of each electrical device, and stores the statistical manifold in a database;

The preprocessing module installs the collection unit at the grid entrance to obtain the electricity consumption data of all electrical equipment in real time, and converts the electricity consumption data into a total current vector;

In the decomposition module, the calculation unit loads the statistical manifold of each electric device, and decomposes the total current vector according to the statistical manifold, so as to obtain the real-time current vector of each electric device.

Further, the training module includes:

The first acquisition module collects the current and voltage signals of each electrical device when it is working normally;

The first vectorization module vectorizes the current and voltage signals to obtain a vector set for each electrical device;

The first manifold construction module constructs the statistical manifold of the electric device according to the set of vectors until the training of all electric devices is completed.

Further, the preprocessing module includes:

The second acquisition module, the acquisition unit acquires the current and voltage signals of all electrical equipment in real time;

In the second vectorization module, the calculation unit vectorizes the current and voltage signals to obtain a vector set of all electrical equipment.

Further, the decomposition module includes:

Algorithm selection module, the calculation unit loads the statistical manifold of each electrical equipment, and selects the optimization algorithm used when decomposing according to the properties of the statistical manifold;

The algorithm execution module decomposes the vector set of the total current by using the optimization algorithm combined with the statistical manifold of each electric device to obtain the current on each device.

The non-intrusive load monitoring system also includes:

The analysis module, after completing the decomposition, performs error analysis on the decomposition results;

Processing module, if the error is within the acceptable range, the calculation unit calculates the power of each electric device according to the real-time current vector of each electric device, and transmits the power to the display unit; otherwise, abnormal processing is performed.

The beneficial effect of the present invention is that,

1. The present invention does not rely on the judgment of the equipment state, and directly decomposes the current: most of the existing non-intrusive load monitoring technologies first estimate the equipment state, and then complete the analysis of the power consumption information. This type of method has a direct flaw, when there are elastic devices with dynamically changing loads in the home, the analysis accuracy will drop significantly. However, the present invention directly models the current phase space of the equipment, does not depend on the judgment of the operating state of the equipment, and can effectively decompose the current generated by various elastic equipment with dynamically changing loads.

2. In the calculation method of the decomposed current, the present invention uses a variety of optimization methods to greatly improve the accuracy of the current decomposition: most of the existing non-intrusive load monitoring technologies first estimate the state of the equipment, and the estimation accuracy of this step is about About 95%, the accuracy of electricity consumption information analysis based on this is about 90%, and when there are multiple elastic devices in the family, the accuracy will drop by about 10%. However, the present invention simplifies the calculation steps, and at the same time uses the optimization method to directly decompose the equipment current, effectively track the current changes of each equipment, and can increase the average accuracy of the current decomposition to 95%.

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments, but not as a limitation of the present invention.

Description of drawings

Fig. 1 is the flow chart of the non-intrusive load monitoring method based on current decomposition of the present invention;

2 is a schematic diagram of a non-invasive load monitoring system based on current decomposition of the present invention;

Fig. 3 is the training stage flowchart of the present invention;

Fig. 4 is the decomposition stage flowchart of the present invention;

Fig. 5 is a schematic structural diagram of the electrical equipment of the present invention.

Detailed ways

Fig. 5 is a schematic structural diagram of the electrical equipment of the present invention. The realization of this structure includes three main components: acquisition unit, calculation unit and display unit, among which:

The acquisition unit is the main input device, and the current decomposition system of electrical equipment obtains the current and voltage information in the entire home by installing the acquisition unit near the entrance of the household grid. The current sensor in the acquisition unit can use a magnetic field sensing chip based on the Hall effect, and use the principle of electromagnetic induction to realize non-invasive current data acquisition. The voltage sensor can be directly connected to the mains circuit to be connected in parallel with all electrical equipment to measure the voltage on it. In the decomposition algorithm, when calculating the vector representation of the signal, the acquisition frequency of the data is required to be higher than a certain threshold. In one embodiment of the present invention: when the algorithm uses the short-time Fourier transform, it is required to include 256 Sampling point, that is, a sampling frequency of 12.8kHz is required for 50Hz alternating current.

The calculation unit is the core of the entire decomposition system, where the current and voltage information of the entire family is decomposed to each electrical device, and the real-time power consumption information of each device is obtained. The calculation unit is located in the center of the whole decomposition system, and the real-time collected data is obtained from the acquisition unit, and the decomposition result is transmitted to the display unit after calculation. In the embodiment of the present invention, any device with sufficient computing capability can be used as a computing unit in the system. Therefore, the calculation unit can be implemented independently, or share a processor with the acquisition unit to form a new acquisition device similar to a smart meter; it can also share a processor with the display unit to form a new display device similar to an intelligent terminal.

The display unit is the main unit for the interaction between the system and the user. In addition to the basic display and interaction functions, it also needs to be able to count and analyze real-time power consumption information, and support data storage and management functions. The display unit obtains the real-time power consumption information of each electric device in the family from the computing unit, updates the historical and real-time power consumption information of each electric device and the whole family, and saves these contents to the database or other In the media; on the other hand, the display unit needs to implement a set of effective user interaction interface, including a graphical display interface, to convert the power consumption information into a form that the user can understand, and to feed back to the user; at the same time, the user can point out the distortion of the system analysis and feed back to the decomposition system, so as to improve the accuracy of subsequent decomposition.

The key technology of the present invention is concentrated in the current decomposition algorithm used by the calculation unit. The algorithm flow can be divided into two phases: training phase and decomposition phase.

The training phase is a necessary phase to build a computational model for each electrical device. In this stage, the current and voltage signals of each device during normal operation are collected respectively; the collected data is then converted into a preset vector representation; the statistical manifold of the current phase space of the current device is constructed from the collected actual data . At this stage, it is not necessary to train all electrical appliances at the same time, so it can be completed by the user at home, or by an equipment provider or a third-party organization. However, from the perspective of the effect of training and the amount of labor required to achieve this stage, it is more accurate and effective to complete the training stage by the equipment provider or a third-party organization and send the results to each user. Each statistical manifold obtained in this stage will be stored in the database and used in the next stage.

The decomposition stage is the main stage for the current decomposition of electrical equipment. In this stage, it is necessary to install the acquisition unit at the entrance of the power grid to collect the current and voltage signals of the whole family; then transfer the collected data to the calculation unit and convert it into a vector representation; The statistical manifold of the equipment, and select the appropriate optimization method according to the mathematical characteristics of each statistical manifold; in the case of knowing the total power consumption information, by calculating the most likely current distribution on each device in the whole family; by The obtained current information of each device deduces the specific power consumption information, and finally transmits the information to the display unit to complete the display and subsequent statistical processing.

Fig. 1 is a flowchart of the non-intrusive load monitoring method based on current decomposition of the present invention. As shown in Figure 1, the method includes:

Step 1, train each electrical device, obtain the statistical manifold of each electrical device, and store the statistical manifold in a database;

Step 2, installing a collection unit at the entrance of the power grid to obtain the power consumption data of all power consumption equipment in real time, and converting the power consumption data into a vector representation of the calculation statistical manifold to obtain a vector set of the total current;

Step 3, the calculation unit loads the statistical manifold of each electric device, and decomposes the total statistical manifold according to the statistical manifold, so as to obtain the real-time current vector of each electric device.

Vector representation of the completion signal. Using methods such as short-time Fourier transform or wavelet transform, the obtained signals such as current and voltage are transformed from the time domain to the frequency domain, and a basic periodic signal waveform is converted into an isolated point in a high-dimensional vector space. The high-dimensional vector or matrix of the domain signal represents the circuit signal waveform for a period of time; technical effect: after transformation, the signal can express its physical meaning more intuitively, for example, it can support the analysis of physical quantities such as harmonic power and reactive power Calculate directly.

Establish a statistical model of the equipment current phase space. Using the concept of manifold in differential geometry and statistical methods, according to the pre-collected power consumption information of a single device, the current signal phase space of each device (in the high-dimensional vector space where the signal may appear) is expressed as a statistical Manifold, that is, a series of smooth probability distribution functions that support the differential geometry method, in which the geometric structure of the manifold describes the device from the perspective of circuit theory, and the probability distribution on it expresses the current of the device during operation Value distribution; technical effect: For each electric device, all the values of the current on it are included in a statistical manifold, and this manifold can be used to calculate the probability of the corresponding device working at a specific current.

Computational methods for constructing current decompositions. According to a variety of optimization calculation methods in operations research, combined with the known statistical manifolds corresponding to each device, a variety of current decomposition methods are established; when decomposition is required, the most effective method can be selected according to the situation to complete the calculation of the total current Decomposition; technical effect: For the complex household electricity environment, through this compound decomposition method, the current passing through each device is decomposed from the total current collected, and the average current accuracy obtained by decomposing during the normal operation of the system is guaranteed to be within 95%. %above.

Further, the step 1 includes:

Step 11, collecting the current and voltage signals of each electrical device when it is working normally;

Step 12, vectorizing the current and voltage signals to obtain a vector set for each electrical device;

Step 13, according to the set of vectors, construct the statistical manifold of the electrical equipment until the training of all electrical equipment is completed.

Further, said step 2 includes:

Step 21, the acquisition unit acquires the current and voltage signals of all electrical equipment in real time;

Step 22, the calculation unit vectorizes the current and voltage signals to obtain a vector set of the total current.

Further, said step 3 includes:

In step 31, the calculation unit loads the statistical manifold of each electric device, and selects an optimization algorithm for decomposition according to the properties of the statistical manifold.

Step 32, using the optimization algorithm, combined with the statistical manifold of each electric device, decomposes the vector set of the total current to obtain the current on each device.

The optimization algorithm mentioned in the step 31 mainly includes: the least square method, the maximum likelihood estimation, the minimum risk method and the method of minimizing the maximum entropy.

The non-intrusive load monitoring method also includes:

Step 4, after completing the decomposition, perform error analysis on the decomposition results;

Step 5, if the error is within the acceptable range, the calculation unit calculates the power of each electric device according to the real-time current vector of each electric device, and transmits the power to the display unit; otherwise, abnormal processing is performed.

Fig. 3 is a flow chart of the training phase of the present invention. As shown in FIG. 3 , after the training phase starts, it is necessary to judge whether the training of all electric devices has been completed. If not, select a device to be tested and deploy the training environment. In the training phase, it is necessary to collect the current and voltage information of each electrical equipment during normal operation, and then the calculation unit performs short-time Fourier transform on the collected current information, so as to obtain the vector representation corresponding to the current at each moment in the frequency domain space.

A 256-dimensional frequency domain vector can be used to represent the current information in one cycle of the device, and more than 256 sampling points per cycle must be guaranteed for data collection, so 12.8kHz is selected as the sampling frequency. For a specific electrical device, after obtaining the current vector for a sufficiently long time (the so-called "sufficiently long time" refers to including all possible working states of the device during normal use), the calculation of its work is started. The statistical manifold to which the current belongs.

This method uses the linear manifold with uniform distribution of probability density as the mathematical abstraction to describe the phase space of the equipment current. The calculation method is principal component analysis: all the collected current vectors (row vectors) are formed into a matrix, and the principal component analysis is performed on this matrix. Select the eigenvector corresponding to the principal component whose variance contribution rate is greater than 95% to construct the linear manifold of this device, that is, the current vector must satisfy the linear manifold equation. The resulting eigenvectors are stored in a database for use in the decomposition stage. The training phase is over until all electrical equipments are trained.

Fig. 4 is a flowchart of the decomposition stage of the present invention. As shown in Figure 4, the decomposition phase decomposes the real-time collected total household current according to the current phase space of all devices obtained in the training phase. In the decomposition phase, the first step is to load the statistical manifold of all electrical appliances (electrical consumers) in the household from the database. Turn on the acquisition unit to obtain real-time total current and voltage data.

Then judge whether to receive the decomposition stop command at this time, if not received, then use the acquisition unit to obtain the total current data in the household at the grid entrance, the data acquisition frequency is the same as the training stage, and the calculation unit will pass the data through the short-time Fourier The leaf transformation obtains the frequency-domain vector representation corresponding to the total current at the current moment. Then select the decomposition algorithm according to the field situation, and carry out the current decomposition. The decomposition method in the present embodiment selects the least squares method,

According to Kirchhoff's law, at any time, the total current is equal to the sum of the currents on each device, and the current vector of each device belongs to its linear manifold. Therefore, in the decomposition stage, the total current and the linear manifold of each electrical appliance are known, and an equation system consisting of two types of equations can be obtained: Kirchhoff's equation and the equation corresponding to the linear manifold. This system of equations is overdetermined and cannot be solved using the usual methods. In this decomposition stage, the least squares method is used to solve the system of equations for the currents on each device that minimizes the error. Anomalies that may occur during the solution process (use of new equipment, equipment damage, etc.) need to be taken into account in this decomposition phase.

After the decomposition calculation is completed, the error analysis of the decomposition results is carried out: compare the numerical relationship between the sum of the current values obtained by the decomposition and the total current, and judge whether there is any abnormality through the analysis of the deviation value, and analyze the possible abnormalities. deal with. For the normal decomposition results, the final data processing is passed from the calculation unit to the display unit, and the power and energy consumption of each device are calculated according to the current vector on each device, combined with the voltage information in the home, and the results are stored in the database. for user review.

In this decomposition stage, the current vector obtained after decomposition is expressed in the current frequency domain, and the power information can be directly calculated with the voltage vector at the corresponding time without converting back to the time domain. In this embodiment, the decomposition stage will continue to be executed until the user issues a command to the decomposition system to stop the decomposition.

Fig. 2 is a flowchart of the non-intrusive load monitoring system based on current decomposition of the present invention. As shown in Figure 2, the system includes:

The training module 100 trains each electric device, obtains the statistical manifold of each electric device, and stores the statistical manifold in a database;

The pre-processing module 200 installs an acquisition unit at the entrance of the power grid, acquires the power consumption data of all power consumption equipment in real time, and converts the power consumption data into a vector representation of the calculation statistical manifold to obtain a vector set of the total current;

In the decomposition module 300, the calculation unit loads the statistical manifold of each electric device, and decomposes the total statistical manifold according to the statistical manifold, so as to obtain the real-time current vector of each electric device.

Vector representation of the completion signal. Using methods such as short-time Fourier transform or wavelet transform, the obtained signals such as current and voltage are transformed from the time domain to the frequency domain, and a basic periodic signal waveform is converted into an isolated point in a high-dimensional vector space. The high-dimensional vector or matrix of the domain signal represents the circuit signal waveform for a period of time; technical effect: after transformation, the signal can express its physical meaning more intuitively, for example, it can support the analysis of physical quantities such as harmonic power and reactive power Calculate directly.

Establish a statistical model of the equipment current phase space. Using the concept of manifold in differential geometry and statistical methods, according to the pre-collected power consumption information of a single device, the current signal phase space of each device (in the high-dimensional vector space where the signal may appear) is expressed as a statistical Manifold, that is, a series of smooth probability distribution functions that support the differential geometry method, in which the geometric structure of the manifold describes the device from the perspective of circuit theory, and the probability distribution on it expresses the current of the device during operation Value distribution; technical effect: For each electric device, all the values of the current on it are included in a statistical manifold, and this manifold can be used to calculate the probability of the corresponding device working at a specific current.

Computational methods for constructing current decompositions. According to a variety of optimization calculation methods in operations research, combined with the known statistical manifolds corresponding to each device, a variety of current decomposition methods are established; when decomposition is required, the most effective method can be selected according to the situation to complete the calculation of the total current Decomposition; technical effect: For the complex household electricity environment, through this compound decomposition method, the current passing through each device is decomposed from the total current collected, and the average current accuracy obtained by decomposing during the normal operation of the system is guaranteed to be within 95%. %above.

Further, the training module 100 includes:

The first collection module 110 collects the current and voltage signals of each electric device when it is working normally;

The first vectorization module 120 vectorizes the current and voltage signals to obtain a vector set for each electric device;

The first manifold construction module 130 constructs the statistical manifold of the electric device according to the set of vectors until the training of all electric devices is completed.

Further, the preprocessing module 200 includes:

The second acquisition module 210, the acquisition unit acquires the current and voltage signals of all electrical equipment in real time;

The second vectorization module 220, the calculation unit vectorizes the current and voltage signals to obtain a vector set of the total current.

Further, the decomposition module 300 includes:

Algorithm selection module 310, the calculation unit loads the statistical manifold of each electric device, and selects the optimization algorithm used in decomposition according to the properties of the statistical manifold.

The algorithm execution module 320 uses the optimization algorithm to decompose the vector set of the total current in combination with the statistical manifold of each electrical device, to obtain the current on each device.

The optimization algorithms mentioned in the algorithm selection module 310 mainly include: the least square method, the maximum likelihood estimation, the minimum risk method and the method of minimizing the maximum entropy.

The non-intrusive load monitoring system also includes:

The analysis module 400, after completing the decomposition, performs error analysis on the decomposition result;

Processing module 500, if the error is within the acceptable range, the calculation unit calculates the power of each electric device according to the real-time current vector of each electric device, and transmits the power to the display unit; otherwise, abnormal processing is performed .

Fig. 3 is a flow chart of the training phase of the present invention. As shown in FIG. 3 , after the training phase starts, it is necessary to judge whether the training of all electric devices has been completed. If not, select a device to be tested and deploy the training environment. In the training phase, it is necessary to collect the current and voltage information of each electrical equipment during normal operation, and then the calculation unit performs short-time Fourier transform on the collected current information, so as to obtain the vector representation corresponding to the current at each moment in the frequency domain space.

A 256-dimensional frequency domain vector can be used to represent the current information in one cycle of the device, and more than 256 sampling points per cycle must be guaranteed for data collection, so 12.8kHz is selected as the sampling frequency. For a specific electrical device, after obtaining the current vector for a sufficiently long time (the so-called "sufficiently long time" refers to including all possible working states of the device during normal use), the calculation of its work is started. The statistical manifold to which the current belongs.

This method uses the linear manifold with uniform distribution of probability density as the mathematical abstraction to describe the phase space of the equipment current. The calculation method is principal component analysis: all the collected current vectors (row vectors) are formed into a matrix, and the principal component analysis is performed on this matrix. Select the eigenvectors corresponding to the principal components whose variance contribution rate is greater than 95% to construct the linear manifold of this device. The resulting eigenvectors are stored in a database for use in the decomposition stage. The training phase is over until all electrical equipments are trained.

Fig. 4 is a flowchart of the decomposition stage of the present invention. As shown in Figure 4, the decomposition phase decomposes the real-time collected total household current according to the current phase space of all devices obtained in the training phase. In the decomposition phase, the first step is to load the statistical manifold of all electrical appliances (electrical consumers) in the household from the database. Turn on the acquisition unit to obtain real-time total current and voltage data.

Then judge whether to receive the decomposition stop command at this time, if not received, then use the acquisition unit to obtain the total current data in the household at the grid entrance, the data acquisition frequency is the same as the training stage, and the calculation unit will pass the data through the short-time Fourier The leaf transformation obtains the frequency-domain vector representation corresponding to the total current at the current moment. Then select the decomposition algorithm according to the field situation, and carry out the current decomposition. According to Kirchhoff's law, at any time, the total current is equal to the sum of the currents on each device, and the current vector of each device belongs to its linear manifold.

Therefore, in the decomposition stage, the total current and the linear manifold of each electrical appliance are known, and an equation system consisting of two types of equations can be obtained: Kirchhoff's equation and the equation corresponding to the linear manifold. In this decomposition stage, the least squares method is used to solve the system of equations for the currents on each device that minimizes the error. Anomalies that may occur during the solution process (use of new equipment, equipment damage, etc.) need to be taken into account in this decomposition phase.

After the decomposition calculation is completed, the error analysis is performed on the decomposition results, and the numerical relationship between each current value and the total current obtained by the decomposition is compared to determine whether there is an abnormality, and to deal with the possible abnormality. For the normal decomposition results, the final data processing is passed from the calculation unit to the display unit, and the power and energy consumption of each device are calculated according to the current vector on each device, combined with the voltage information in the home, and the results are stored in the database. for user review.

In this decomposition stage, the current vector obtained after decomposition is expressed in the current frequency domain, and the power information can be directly calculated with the voltage vector at the corresponding time without converting back to the time domain. In this embodiment, the decomposition stage will continue to be executed until the user issues a command to the decomposition system to stop the decomposition.

Certainly, the present invention also can have other multiple embodiments, without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and deformations according to the present invention, but these corresponding Changes and deformations should belong to the scope of protection of the appended claims of the present invention.

Claims (8)

1. A non-intrusive load monitoring method based on current decomposition, characterized in that, comprising: Step 1, train each electrical device, obtain the statistical manifold of each electrical device, and store the statistical manifold in a database; Step 2, install a collection unit at the entrance of the power grid to obtain the power consumption data of all power consumption equipment in real time, and transfer the collected power consumption data to a calculation unit, and the calculation unit converts the power consumption data into calculation statistics The vector representation of the manifold, the vector set of the total current is obtained; Step 3, the calculation unit loads the statistical manifold of each electric device, and decomposes the vector set of the total current according to the statistical manifold, so as to obtain the real-time current vector of each electric device; Said step 1 includes: Step 11, collecting the current and voltage signals of each electrical device when it is working normally; Step 12, vectorizing the current and voltage signals to obtain a vector set for each electrical device; Step 13, according to the set of vectors, construct the statistical manifold of the electrical equipment until the training of all electrical equipment is completed. 2. The non-intrusive load monitoring method according to claim 1, wherein said step 2 comprises: Step 21, the acquisition unit acquires the current and voltage signals of all electrical equipment in real time; Step 22, the calculation unit vectorizes the current and voltage signals to obtain a vector set of the total current. 3. The non-intrusive load monitoring method according to claim 1, wherein said step 3 comprises: Step 31, the calculation unit loads the statistical manifold of each electric device, and selects the optimization algorithm used for decomposition according to the properties of the statistical manifold; Step 32, using the optimization algorithm, combined with the statistical manifold of each electric device, decomposes the vector set of the total current to obtain the current on each device. 4. The non-intrusive load monitoring method according to claim 1, wherein the non-intrusive load monitoring method further comprises: Step 4, after completing the decomposition, perform error analysis on the decomposition results; Step 5, if the error is within the acceptable range, the calculation unit calculates the power of each electric device according to the real-time current vector of each electric device, and transmits the power to the display unit; otherwise, abnormal processing is performed. 5. A non-intrusive load monitoring system based on current decomposition, characterized in that it comprises: A training module, which trains each electrical device, obtains the statistical manifold of each electrical device, and stores the statistical manifold in a database; The preprocessing module installs the collection unit at the entrance of the power grid to obtain the power consumption data of all power consumption equipment in real time, and transfers the collected power consumption data to the calculation unit, and the calculation unit converts the power consumption data into calculation The vector representation of the statistical manifold obtains the vector set of the total current; Decomposition module, the calculation unit loads the statistical manifold of each electric device, and decomposes the vector set of the total current according to the statistical manifold, so as to obtain the real-time current vector of each electric device; The training modules include: The first acquisition module collects the current and voltage signals of each electrical device when it is working normally; The first vectorization module vectorizes the current and voltage signals to obtain a vector set for each electrical device; The first manifold construction module constructs the statistical manifold of the electric device according to the set of vectors until the training of all electric devices is completed. 6. The non-intrusive load monitoring system according to claim 5, wherein the preprocessing module comprises: The second acquisition module, the acquisition unit acquires the current and voltage signals of all electrical equipment in real time; In the second vectorization module, the calculation unit vectorizes the current and voltage signals to obtain a vector set of the total current. 7. The non-intrusive load monitoring system according to claim 5, wherein the decomposition module comprises: Algorithm selection module, the calculation unit loads the statistical manifold of each electrical equipment, and selects the optimization algorithm used when decomposing according to the properties of the statistical manifold; The algorithm execution module decomposes the vector set of the total current by using the optimization algorithm combined with the statistical manifold of each electric device to obtain the current on each device. 8. The non-intrusive load monitoring system of claim 5, wherein the non-intrusive load monitoring system further comprises: The analysis module, after completing the decomposition, performs error analysis on the decomposition results; Processing module, if the error is within the acceptable range, the calculation unit calculates the power of each electric device according to the real-time current vector of each electric device, and transmits the power to the display unit; otherwise, abnormal processing is performed.