CN110889465B - Method and system for identification of power demand side equipment based on adaptive resonant network - Google Patents

Abstract

The invention discloses a method and system for identifying power demand side equipment based on an adaptive resonance network. By collecting multi-dimensional characteristic data of unknown equipment on the power demand side, and extracting unknown equipment load events according to an event detection algorithm, the unknown equipment load events are used to obtain unknown equipment start-up events. Stop time, transient variation of each characteristic and steady-state operation characteristic data; classify the characteristic data to obtain input data and test data, and normalize and encode the input data and test data; The input data is trained and learned to obtain an adaptive resonance network training model; the test data is identified according to the adaptive resonance network training model to obtain an identification result; the identification results are matched according to a pre-stored database to obtain an unknown device matching result. The self-adaptive resonance network of the present invention can use standard data to train and perform supervised learning, and can also classify unlabeled data to identify unknown equipment.

Description

Method and system for identification of power demand side equipment based on adaptive resonant network

technical field

The invention belongs to the field of non-intrusive load identification, in particular to a method and system for identifying power demand side equipment based on an adaptive resonance network.

Background technique

Non-intrusive load identification technology is currently a research hotspot in the field of smart metering of power systems, and has developed rapidly in recent years. By installing a non-intrusive power identification device on the power demand side, intelligent analysis technology can be used to obtain the type, operating status and energy consumption of unknown equipment within the power consumption range, so as to realize real-time monitoring of the start-stop and power consumption of different equipment. This technology can help users and power grid companies analyze the power consumption behavior of the power demand side, and provide strong data support for energy conservation and emission reduction, energy consumption management, smart home management and power demand side maintenance. Under the current advocacy of the ubiquitous power Internet of Things, it is of great significance for power grid companies to understand the power consumption composition of residents and industries, make overall planning of power dispatching plans, and reduce peaks and valleys.

Non-intrusive load identification technology is an important direction of the current smart grid development, and many different types of identification methods have been studied in this field. However, due to the wide variety of unknown devices on the power demand side, the newly added devices by users are more complicated, and the brands and models are endless. The existing identification methods cannot update the local database in real time, and it is difficult to adapt to the unknown devices newly accessed by users, resulting in the current non-intrusive The recognition accuracy of the unknown device by the conventional identification device is not high, and the existing technology cannot perform self-learning and reclassification for the unknown device that has not been identified, resulting in a reduction in the recognition efficiency.

SUMMARY OF THE INVENTION

In view of the above problems, the present invention is proposed in order to provide a method and system for identifying power demand side equipment based on an adaptive resonant network that overcomes the above problems or at least partially solves the above problems.

A method for identifying power demand side equipment based on an adaptive resonant network, characterized in that it includes:

S100. Collect multi-dimensional feature data of unknown equipment on the power demand side, extract unknown equipment load events according to an event detection algorithm, and obtain unknown equipment start-stop time, transient changes of various characteristics, and steady-state operation characteristic data through the unknown equipment load events;

S200. Classify the unknown equipment start-stop time, each characteristic transient change and steady-state operation characteristic data to obtain input data and test data, and normalize and encode the input data and test data;

S300. Perform training and learning on the input data through an adaptive resonant network to obtain an adaptive resonant network training model;

S400. Identify the test data according to the adaptive resonance network training model, and obtain the identification result;

S500. Match the identification results according to a pre-stored database to obtain a matching result of an unknown device;

S600. Correct the identification result through the human-computer interaction interface, add the correction result to the pre-stored database, and update the pre-stored database.

Further, classify unknown equipment start and stop time, characteristic transient variation and steady-state operation characteristic data, and obtain the input data and test data as follows: obtain the number l of unknown equipment on the power demand side, when each equipment is running. Turns on and off randomly. The m features are recorded at the acquisition frequency f, and then n feature points are randomly selected between two transient events, so a matrix A _n×m composed of n m-dimensional data can be obtained; the data used for testing is denoted as a _{= { a 1} , _{a 2} , .

Further, the input data and the test data are normalized using the minimum-maximum feature method, and the normalization formula is as formula (1):

Among them, a' _i is the ith sampling point after normalization.

Further, the input data and test data are encoded, and the input data and test data encoding methods are calculated according to formula (2) and formula (3) respectively:

x ^a =[a, ^ac ]=[a,1-a] (2)

x ^b =[b,b ^c ]=[b,1-b] (3)

Further, the adaptive resonance network is horizontally divided into a test data identification area ARTa, a standard data learning area ARTb and a mapping area x ^ab connecting the two areas, and is vertically divided into an input layer, an encoding layer and an identification layer.

Further, the mapping area x ^ab is expressed as

where the operator ∧ is defined as w _i ∧y _i =min(ω _i ,y _i ).

Further, S400 includes:

S401 determines the test data identification area ARTa selection function T _i ^a , and the selection function T _i ^a is as in formula (4):

Among them, α is the selection parameter, x ^α is the input data, and w _i ^α is the weight vector of the ARTa layer;

S402 determines the winning category J, and the winning category J is as follows in formula (5):

S403 judges the category of the data sequence x according to the winning category J, the data sequence x, the weight vector W _J related to the winning node, and the resonance alert parameter ρ, if the following formula (6) is satisfied:

Then resonance can be generated, and the sequence x belongs to the category J; if the formula (6) is not satisfied, then T _i ^α is reset to 0, and the weight w is updated to w _J ^new according to the learning rate β, as shown in the following formula (7)

Further, S500 includes:

S501. Perform basic identification on category J by using a large category database in a pre-stored database, where the category database includes temporary steady state power data, harmonic data, electrical appliance operating cycle, common time period and corresponding specific electrical appliance types, and identify the specific electrical appliances of unknown equipment type;

S502. Use the fine database in the pre-stored database to accurately identify the category J, where the fine database includes a single electrical characteristic of the electrical equipment model, and is used to identify the specific model of the unknown equipment.

On the other hand, the present invention also discloses a power demand side equipment identification system based on an adaptive resonance network, comprising: a power data acquisition and detection module, a feature input module, a learning and training module, an identification module and a database matching module;

The power data acquisition and detection module is used to collect the multi-dimensional characteristic data of unknown equipment on the power demand side, and extract the unknown equipment load events according to the event detection algorithm. state operating characteristic data;

The feature input module classifies the unknown equipment start-stop time, the transient variation of each feature and the steady-state operation feature data, obtains input data and test data, and normalizes and encodes the input data and test data;

The learning and training module is used to train and learn the input data through the adaptive resonant network to obtain the adaptive resonant network training model;

The identification module identifies the test data according to the new training model of the adaptive resonant network, and obtains the identification result;

The database matching module matches the identification results through a pre-stored database to obtain matching results of unknown devices.

Further, it also includes a human-computer interaction module, which corrects the identification result through the human-computer interaction module, adds the correction result to a pre-stored database, and updates the pre-stored database.

The beneficial effects of the present invention are: compared with the prior art, a method and system for identifying power demand side equipment based on an adaptive resonant network provided by the present invention take adaptive resonant network mapping as the main identification method, and according to its competitive characteristics and scalability, the unknown device can be separated from the existing device features for individual identification. This method processes real-time data through pre-trained results, and when detecting access to unknown devices, a large-class database can be enabled for class identification, and then the user can correct the feature data according to the actual value and add it to the fine database for the next step. The secondary algorithm can perform autonomous identification. This method can perform high-precision power load identification when multiple devices are running in aliased operation, and can distinguish autonomously when new devices are connected, and provide category recommendations for users' reference. Therefore, the method can effectively improve the accuracy and ease of use of the current non-invasive identification device, expand the compatibility of traditional equipment with new equipment and complex equipment, and has broad application prospects.

Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description, claims, and drawings.

The technical solutions of the present invention will be further described in detail below through the accompanying drawings and embodiments.

Description of drawings

The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification, and are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. In the attached image:

1 is a flowchart of a method for identifying power demand side equipment based on an adaptive resonant network in Embodiment 1 of the present invention;

Fig. 2 is the original acquisition waveform of active power data in the first embodiment of the present invention;

FIG. 3 is the result of the active power data reconstruction waveform and event detection algorithm in the first embodiment of the present invention;

4 is a structural diagram of an adaptive resonant network model in Embodiment 1 of the present invention;

Fig. 5 is the identification result of the unknown device in the first embodiment of the present invention;

FIG. 6 is a structural diagram of a power demand side equipment identification system based on an adaptive resonant network in Embodiment 2 of the present invention.

Detailed ways

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure will be more thoroughly understood, and will fully convey the scope of the present disclosure to those skilled in the art.

In order to solve the problem in the prior art, due to the variety of power demand side power equipment, the new equipment purchased by the user is more complex, and the brands and models are endless. To solve the problem of low accuracy, embodiments of the present invention provide a method and system for identifying power demand side equipment based on an adaptive resonant network.

Example 1

As shown in FIG. 1 , this embodiment discloses a method for identifying power demand side equipment based on an adaptive resonant network, including:

S100. Collect multi-dimensional characteristic data of unknown equipment on the power demand side, extract unknown equipment load events according to an event detection algorithm, and obtain unknown equipment start-stop time, characteristic transient changes and steady-state operation characteristic data through the unknown equipment load events.

Specifically, a non-intrusive load monitoring device is used to collect the total power data on the user side, including at least multi-dimensional characteristic data such as voltage, current, active power, reactive power, and harmonics within a certain period of time. Taking active power as an example, the original acquisition waveform of power data is shown in Figure 2. Then, the event detection algorithm is used to extract the events of the load state change, record the equipment start and stop time, the transient change of each feature and the steady-state operation characteristics and other information, and filter the original data to remove noise and fluctuations. The power data reconstruction waveform and event detection algorithm results are shown in Figure 3.

S200. Classify the unknown equipment start-stop time, each characteristic transient change and steady-state operation characteristic data, obtain input data and test data, and normalize and encode the input data and test data;

Specifically, S200 includes:

S201: Classify the recorded information such as the start-stop time of the unknown equipment, the transient variation of each feature, and the steady-state operation feature by category, and distinguish between standard data and test data for learning. The specific method is as follows: there are l power load devices, and each device is randomly turned on and off when running in the room. The m features are recorded at the acquisition frequency f, and then n feature points are randomly selected between two transient events, so a matrix A _n×m composed of n m-dimensional data can be obtained. The test vector is denoted as a={a ₁ ,a ₂ ,...,am }, and the standard input data for learning is denoted as b={b ₁ ,b ₂ ,...,b _i }, where _i is the value of the data sequence The maximum amount.

S202: Form the obtained multi-dimensional feature data into an input vector, and use the minimum-maximization feature method to normalize to the [0,1] interval, and the normalization formula is formula (1):

where is the i-th sampling point after normalization, it is understandable, assuming α={1,2,3,4,5}, after the normalization formula is α={0,0.25,0.5,0.75, 1}.

Step 203: Perform complement processing on the obtained input vector. The input data and test data are encoded, and the input data and test data encoding methods are calculated according to formula (2) and formula (3) respectively:

x ^a =[a, ^ac ]=[a,1-a] (2)

x ^b =[b,b ^c ]=[b,1-b] (3)

S300. Perform training and learning on the input data through an adaptive resonant network to obtain an adaptive resonant network training model;

Figure 4 is a structural diagram of the adaptive resonance network mapping module. The essence of the adaptive resonance network mapping module is a neural network, and its horizontal structure includes a test data identification area ARTa, a standard data learning area ARTb and a mapping area x ^ab connecting the two areas ; its vertical region contains three layers, namely input layer, coding layer and recognition layer. Each layer relies on the update of weights to advance. The learning process of the adaptive resonant network is:

The standard data is learned and trained. The standard data is the pre-collected and correct power characteristic data and the type and model of the power equipment. Only through the pre-learning of the ARTb area can the input ARTa data be classified and identified. The standard data for learning is input from the input layer 206 in the adaptive resonance network mapping identification module, and is formed after being encoded, and its standard characteristics are recorded through competitive learning. At the beginning of learning, set each weight, learning rate, and vigilance parameters. The activation of the mapped regions needs to rely on both the output of the encoding layer 205 in ARTb. x ^ab can be expressed as:

The operator symbol ∧ represents w _i ∧y _i =min(ω _i ,y _i ). The standard data includes the input vector b and its corresponding label set l, that is, there is a standard training set, and the data in ARTa can be identified and classified after the training is completed.

S400. Identify the test data according to the adaptive resonance network training model, and obtain the identification result.

Specifically, S400 includes:

S401 determines the test data identification area ARTa selection function T _i ^a , and the selection function T _i ^a is as follows in formula (4):

Among them, α is the selection parameter, x ^α is the input data, and w _i ^α is the weight vector of the ARTa layer;

S402 determines the winning category J, and the winning category J is as follows in formula (5):

S403 According to the winning category J, the data sequence x, the weight vector W _J related to the winning node, and the resonance alert

The parameter ρ determines the category of the data sequence x, if it satisfies the following formula (6):

Then resonance can be generated, and the sequence x belongs to the category J; if the formula (6) is not satisfied, then T _i ^a is reset to 0, and the weight w is updated to w _J ^new according to the learning rate β, as shown in the following formula (7):

S500. Match the identification results according to a pre-stored database to obtain a matching result of an unknown device;

Specifically, according to the data generated in step 400, real-time data storage is performed, and based on the data principle, the data is compared with the local database. Local databases include large-scale databases and fine-grained databases. The data stored in the database includes a wide range of electrical appliance types, temporary stable power data range, harmonic data range, electrical appliance operating cycle characteristics and common time period characteristics, etc., which are used for basic identification, such as identifying air conditioners, TVs, water heaters, etc. Appliance category. The refined database includes the characteristics of a single electrical appliance accurate to the equipment model, which is used to identify the specific model of the power demand side electrical appliance, which can identify the number of a certain type of electrical appliance and distinguish its model. If the detected data features match a sample feature in the user's local accurate database, the device can be accurately identified. If the detected data features are not in the accurate data samples, it will be marked as an unknown device and a large-class database will be enabled for class identification, and the range will be specific to the corresponding class of electrical appliances for user reference. FIG. 5 is the matching identification result provided by the present invention, and it can be seen that the test data points falling within the area are assigned the identification result of the corresponding category.

S600. Correct the identification result through the human-computer interaction interface, add the correction result to the pre-stored database, and update the pre-stored database.

Specifically, according to the database matching result in step 4, the identification result is corrected through the human-computer interaction interface, the characteristic data of the unknown device is refined, and the device is added to the refined database for algorithm training by the learning training module. The human-computer interaction interface can include mobile phone applications or web pages for interaction. The correct identification results will be stored as standard data for use in the identification process within a certain period of time in the future, so that more accurate results can be output when the data features are detected again.

A method for identifying power demand side equipment based on an adaptive resonant network provided by an embodiment of the present invention uses adaptive resonant network mapping as the main identification method. According to its competitive characteristics and scalability, unknown equipment and existing equipment characteristics can be distinguished. Separate and identify individually. This method processes real-time data through pre-trained results, and when detecting access to unknown devices, a large-class database can be enabled for class identification, and then the user can correct the feature data according to the actual value and add it to the fine database for the next step. The secondary algorithm can perform autonomous identification. This method can perform high-precision power load identification when multiple devices are running in aliased operation, and can distinguish autonomously when new devices are connected, and provide category recommendations for users' reference. Therefore, the method can effectively improve the accuracy and ease of use of the current non-invasive identification device, expand the compatibility of traditional equipment with new equipment and complex equipment, and has broad application prospects.

Embodiment 2

The present invention also discloses a power demand-side equipment identification system based on an adaptive resonant network applied in the first embodiment, including: a power data acquisition and detection module 1, a feature input module 2, a learning and training module 3, and an identification module 4 and database matching module 5;

Power data acquisition and detection module 1 is used to collect multi-dimensional feature data of unknown equipment on the power demand side, and extract unknown equipment load events according to an event detection algorithm. Steady-state operating characteristic data;

The feature input module 2 classifies the unknown equipment start-stop time, the transient variation of each feature and the steady-state operation feature data to obtain input data and test data, and normalizes and encodes the input data and test data;

Specifically, the feature input module 2 classifies the startup and shutdown time of the unknown equipment, the transient variation of each feature and the steady-state operation feature data, obtains input data and test data, and normalizes the input data and test data to sum The encoding process includes:

S201: Classify the recorded information such as the start-stop time of the unknown equipment, the transient variation of each feature, and the steady-state operation feature by category, and distinguish between standard data and test data for learning. The specific method is as follows: there are l power load devices, and each device is randomly turned on and off when running in the room. The m features are recorded at the acquisition frequency f, and then n feature points are randomly selected between two transient events, so a matrix A _n×m composed of n m-dimensional data can be obtained. The test vector is denoted as a={a ₁ ,a ₂ ,...,am }, the standard input data used for learning is denoted as b={b ₁ ,b ₂ ,...,b _i }, and _i is the maximum value of the data sequence. a lot.

S202: Form the obtained multi-dimensional feature data into an input vector, and use the minimum-maximization feature method to normalize to the [0,1] interval, and the normalization formula is formula (1):

where is the i-th sampling point after normalization, it is understandable, assuming α={1,2,3,4,5}, after the normalization formula is α={0,0.25,0.5,0.75, 1}.

Step 203: Encoding (complement) processing of the obtained input vector. For the input test vector, its encoding formula is calculated according to formula (2):

x ^a =[a, ^ac ]=[a,1-a] (2)

The encoding formula of the standard input data used for learning is calculated according to formula (3):

x ^b =[b,b ^c ]=[b,1-b] (3)

Learning and training module 3, training and learning the input data through an adaptive resonant network to obtain an adaptive resonant network training model; performing learning and training on standard data, the standard data being pre-collected, correct power characteristic data and power equipment categories , model, through the pre-learning of the ARTb area, the input ARTa data can be classified and identified. The standard data for learning is input from the input layer 206 in the adaptive resonance network mapping identification module 4, and is formed after encoding, and its standard characteristics are recorded through a competitive learning method. At the beginning of learning, set each weight, learning rate, and vigilance parameters. The activation of the mapped regions needs to rely on both the output of the encoding layer 205 in ARTb. x ^ab can be expressed as:

where the operator symbol represents. The standard data includes the input vector b and its corresponding label set l, that is, there is a standard training set, and the data in ARTa can be identified and classified after the training is completed.

The identification module 4 identifies the test data according to the new training model of the adaptive resonant network, and obtains the identification result;

Specifically, the identification module 4 identifies the training data including:

S401 determines the test data identification area ARTa selection function T _i ^a , and the selection function T _i ^a is as follows in formula (4):

Among them, α is the selection parameter, x ^α is the input data, and w _i ^α is the weight vector of the ARTa layer;

S402 determines the winning category J, and the winning category J is as follows in formula (5):

S403 judges the category of the data sequence x according to the winning category J, the data sequence x, the weight vector W _J related to the winning node, and the resonance alert parameter ρ, if the following formula (6) is satisfied:

Then resonance can be generated, and the sequence x belongs to the category J; if the formula (6) is not satisfied, then T _i ^α is reset to 0, and the weight w is updated to w _J ^new according to the learning rate β, as shown in the following formula (7):

The database matching module 5 matches the identification results through a pre-stored database to obtain matching results of unknown devices.

Specifically, the database matching module 5 performs real-time data storage according to the data generated in step 400, and compares it with the local database based on the data principle. Local databases include large-scale databases and fine-grained databases. The data stored in the large-scale database includes a wide range of electrical appliance types, the range of temporary and steady-state power data, the range of harmonic data, the characteristics of electrical appliance operating cycles, and the characteristics of common time periods. and other electrical categories. The refined database includes the characteristics of a single electrical appliance accurate to the equipment model, which is used to identify the specific model of the power demand side electrical appliance, which can identify the number of a certain type of electrical appliance and distinguish its model. If the detected data features match a sample feature in the user's local accurate database, the device can be accurately identified. If the detected data features are not in the accurate data samples, it will be marked as an unknown device and a large-class database will be enabled for class identification, and the range will be specific to the corresponding class of electrical appliances for user reference. FIG. 5 is the matching identification result provided by the present invention, and it can be seen that the test data points falling within the area are assigned the identification result of the corresponding category.

In some preferred embodiments, the system further includes a human-computer interaction module, through which the identification result is corrected, the characteristic data of the unknown device is refined, and the device is added to the refined database for the learning and training module 3 Perform algorithm training. The human-computer interaction interface can include mobile phone applications or web pages for interaction. The correct identification results will be stored as standard data for use in the identification process within a certain period of time in the future, so that more accurate results can be output when the data features are detected again.

An adaptive resonance network-based power demand-side equipment identification system provided by the embodiment of the present invention uses adaptive resonance network mapping as the main identification method, and can separate unknown equipment from existing equipment characteristics according to its competitive characteristics and scalability. on for individual identification. This method processes real-time data through pre-trained results, and when detecting access to unknown devices, a large-class database can be enabled for class identification, and then the user can correct the feature data according to the actual value and add it to the fine database for the next step. The secondary algorithm can perform autonomous identification. This method can perform high-precision power load identification when multiple devices are running in aliased operation, and can distinguish autonomously when new devices are connected, and provide category recommendations for users' reference. Therefore, the system can effectively improve the accuracy and ease of use of the current non-invasive identification device, expand the compatibility of traditional equipment to new equipment and complex equipment, and has broad application prospects.

It is understood that the specific order or hierarchy of steps in the disclosed processes is an example of a sample approach. Based upon design preferences, it is understood that the specific order or hierarchy of steps in the processes may be rearranged without departing from the scope of the present disclosure. The accompanying method claims present elements of the various steps in a sample order, and are not meant to be limited to the specific order or hierarchy presented.

In the foregoing Detailed Description, various features are grouped together in a single embodiment for the purpose of simplifying the disclosure. This method of disclosure should not be construed as reflecting an intention that embodiments of the claimed subject matter require more features than are expressly recited in each claim. Rather, as the following claims reflect, present invention lies in less than all features of a single disclosed embodiment. Thus, the following claims are hereby expressly incorporated into the Detailed Description, with each claim standing on its own as a separate preferred embodiment of this invention.

Those skilled in the art will also appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments herein may be implemented as electronic hardware, computer software, or combinations thereof. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether this functionality is implemented as hardware or software depends on the specific application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, however, such implementation decisions should not be interpreted as a departure from the scope of the present disclosure.

The steps of a method or algorithm described in connection with the embodiments herein may be directly embodied in hardware, a software module executed by a processor, or a combination thereof. A software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and storage medium may reside in an ASIC. The ASIC may be located in the user terminal. Of course, the processor and the storage medium may also exist in the user terminal as discrete components.

For a software implementation, the techniques described in this application may be implemented in modules (eg, procedures, functions, etc.) that perform the functions described in this application. These software codes may be stored in a memory unit and executed by a processor. The memory unit may be implemented within the processor or external to the processor, in which case it is communicatively coupled to the processor via various means, as is known in the art.

The above description includes examples of one or more embodiments. Of course, it is not possible to describe all possible combinations of components or methods in order to describe the above embodiments, but one of ordinary skill in the art will recognize that further combinations and permutations of the various embodiments are possible. Accordingly, the embodiments described herein are intended to cover all such changes, modifications and variations that fall within the scope of the appended claims. Furthermore, with respect to the term "comprising," as used in the specification or claims, the word is encompassed in a manner similar to the term "comprising," as if "comprising," were construed as a conjunction in the claims. Furthermore, any use of the term "or" in the specification of the claims is intended to mean a "non-exclusive or."

Claims (6)

1. A method for identifying equipment on a power demand side based on an adaptive resonant network is characterized by comprising the following steps:

s100, collecting multidimensional characteristic data of unknown equipment on a power demand side, extracting load events of the unknown equipment according to an event detection algorithm, and obtaining starting and stopping time, transient variation of each characteristic and steady-state operation characteristic data of the unknown equipment through the load events of the unknown equipment;

s200, classifying the starting and stopping time of unknown equipment, the transient variation of each characteristic and the steady-state operation characteristic data to obtain input data and test data, and carrying out normalization and coding processing on the input data and the test data; the method for classifying the starting and stopping time, the transient variation of each characteristic and the steady-state operation characteristic data of unknown equipment and acquiring the input data and the test data comprises the following steps: acquiring the number l of unknown equipment on the power demand side, and randomly turning on and off each equipment when running; recording m characteristics of the data with an acquisition frequency f, and randomly taking n characteristic points between two transient events, thereby obtaining a matrix A consisting of n m-dimensional data_n×m(ii) a Wherein the data used for the test is designated as a ═ a₁,a₂,…,a_mData for learning is denoted as b ═ b₁,b₂,…,b_iWhere i is the maximum amount of the data sequence;

s300, training and learning input data through a self-adaptive resonant network to obtain a self-adaptive resonant network training model; the self-adaptive resonant network is transversely divided into a test data identification area ARTa, a standard data learning area ARTb and a mapping area x connecting the two areas^abThe device is longitudinally divided into an input layer, an encoding layer and an identification layer; mapping region x^abIs shown as

Wherein the operator ^ is defined as w_i∧y_i＝min(ω_i,y_i)；

S400, identifying the test data according to the adaptive resonant network training model to obtain an identification result; s400 includes:

s401, determining ARTa selection function T of test data identification area_i ^aSelecting a function T_i ^aAs shown in formula (1):

wherein alpha is a selection parameter, x^αTo input data, w_i ^αIs an ARTa layer weight vector;

s402 determines a winning class J, which is given by the following formula (2):

s403, according to the winning category J, the data sequence x and the weight vector W related to the winning node_JAnd judging the category of the data sequence x by the resonance warning parameter rho, if the following formula (3) is satisfied:

resonance can occur and the sequence x belongs to class J; if equation (3) is not satisfied, T_i ^αIs reset to 0, the weight w is updated to w according to the learning rate beta_J ^newThe following formula (4)

S500, matching the identification result according to a pre-stored database to obtain a matching result of the unknown equipment;

s600, the identification result is corrected through a human-computer interaction interface, the correction result is added into a pre-stored database, and the pre-stored database is updated.

2. The adaptive resonant network-based power demand side equipment identification method according to claim 1, wherein the input data and the test data are normalized by using a min-max characteristic method, and the normalization formula is as follows (5):

wherein, a_i' is the normalized ith sample point.

3. The adaptive resonant network-based power demand side equipment identification method according to claim 1, wherein input data and test data are encoded, and the input data and test data encoding methods are calculated according to equations (6) and (7), respectively:

x^a＝[a,a^c]＝[a,1-a](6)

x^b＝[b,b^c]＝[b,1-b](7)。

4. the method for identifying a power demand side device based on an adaptive resonant network as claimed in claim 1, wherein S500 comprises:

s501, performing basic identification on the category J by utilizing a large database in a pre-stored database, wherein the large database comprises transient power data, harmonic data, an electric appliance operation cycle, a common time period and a corresponding specific electric appliance type, and identifying the specific electric appliance type of unknown equipment;

s502, accurately identifying the category J by using a fine database in a pre-stored database, wherein the fine database comprises single electrical appliance characteristics of electrical appliance equipment models and is used for identifying specific models of unknown equipment.

5. An adaptive resonant network-based power demand side equipment identification system, comprising: the device comprises a power data acquisition and detection module, a feature input module, a learning and training module, an identification module and a database matching module;

the power data acquisition and detection module is used for acquiring multi-dimensional characteristic data of unknown equipment on a power demand side, extracting load events of the unknown equipment according to an event detection algorithm, and obtaining start-stop time, transient variation of each characteristic and steady-state operation characteristic data of the unknown equipment through the load events of the unknown equipment;

the characteristic input module is used for classifying the starting and stopping time of unknown equipment, the transient variation of each characteristic and the steady-state operation characteristic data, acquiring input data and test data, and normalizing and encoding the input data and the test data;

the learning training module is used for training and learning the input data through the self-adaptive resonant network to obtain a self-adaptive resonant network training model;

the identification module is used for identifying the test data according to the new training model of the adaptive resonant network to obtain an identification result;

and the database matching module is used for matching the identification result through a pre-stored database to obtain a matching result of the unknown equipment.

6. The system according to claim 5, further comprising a human-computer interaction module, wherein the human-computer interaction module is used for modifying the identification result, adding the modified result into the pre-stored database, and updating the pre-stored database.

Images