Non-intrusive load decomposition system and method based on deep learning

Abstract

The invention relates to the technical field of non-intrusive load decomposition, in particular to a non-intrusive load decomposition system and method based on deep learning, and the system comprises a data collection module, an event detection module, a feature enhancement module, a load recognition module, a load decomposition module, a communication module, a data storage and analysis module, and a user interaction module. The data acquisition module is used for providing sensors and data preprocessing; the event detection module is used for load event detection and feature extraction; the feature enhancement module is used for recurrence plot feature optimization and high-dimensional implicit information representation; the load identification module is used for pattern matching and deep learning identification; the load decomposition module is used for further extracting load features and optimizing an attention mechanism; the system has the advantages of high precision, high generalization ability, strong real-time performance and the like, and can provide more accurate and personalized power consumption management and energy-saving suggestion services for users.

Description

Technical Field

[0001] The present invention relates to the technical field of non-invasive load decomposition, and specifically to a non-invasive load decomposition system and method based on deep learning. Background Art

[0002] Non-Intrusive Load Monitoring (NILM) is a technology that identifies and decomposes the operating status and energy consumption of each electrical device by analyzing the voltage, current and other electrical signals of the user's main line without interfering with the user's normal electricity consumption. Non-Intrusive Load Monitoring can identify the energy consumption of each electrical device, so that users can control and manage electricity consumption more accurately. The decomposed electricity consumption data can be used to analyze the user's electricity consumption behavior pattern and provide users with personalized electricity consumption suggestions and energy-saving solutions.

[0003] However, in general, traditional non-intrusive load decomposition methods often rely on simple feature extraction and pattern matching, which makes it difficult to accurately identify the energy consumption of equipment in complex power consumption environments. They are only designed for specific types of power equipment and are difficult to adapt to power equipment of different types and specifications. The data processing and analysis process is often more complicated, making it difficult to achieve real-time online load decomposition.

[0004] In summary, it is necessary to propose a non-intrusive load decomposition system and method based on deep learning to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide a non-intrusive load decomposition system and method based on deep learning to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A non-intrusive load decomposition system based on deep learning, including a data acquisition module, an event detection module, a feature enhancement module, a load identification module, a load decomposition module, a communication module, a data storage and analysis module, and a user interaction module;

[0008] The data acquisition module is used to provide sensors and data preprocessing;

[0009] The event detection module is used for load event detection and feature extraction;

[0010] The feature enhancement module is used for recursive graph feature optimization and high-dimensional implicit information representation;

[0011] The load identification module is used for pattern matching and deep learning identification;

[0012] The load decomposition module is used to further extract load features and optimize the attention mechanism;

[0013] The communication module is used for data transmission and security encryption;

[0014] The data storage and analysis module is used to build a distributed database and intelligent analysis;

[0015] The user interaction module is used to provide Web / APP interaction and personalized recommendations.

[0016] Preferably, the data acquisition module further includes a sensor unit and a preprocessing unit;

[0017] The sensor unit uses a high-precision power sensor and an analog-to-digital converter to collect voltage and current electrical signals of the user's trunk line in real time to ensure high accuracy and anti-interference of the data;

[0018] The preprocessing unit is based on a sliding window data preprocessing algorithm, which is used to perform denoising, filtering, and standardization preprocessing operations on the collected electrical signals, thereby improving data quality and laying a foundation for subsequent analysis.

[0019] Preferably, the event detection module further includes a load event detection unit and a feature extraction unit;

[0020] The load event detection unit is based on a change point detection algorithm (including CUSUM, EWMA) to detect load start and stop events in the electrical signal in real time and mark the start and stop time points of each electrical device;

[0021] The feature extraction unit combines the time-frequency analysis method of Fourier transform and wavelet transform to extract load features, including steady-state and transient features, from the electrical signal after event detection, providing key information for subsequent load decomposition.

[0022] Preferably, the feature enhancement module further includes a recursive graph feature optimization unit and a high-dimensional implicit information representation unit;

[0023] The recursive graph feature optimization unit is used to convert the current time series signal into TFRP, and realize power-type scaling through voltage-current correlation, improve feature discrimination, and reduce hyperparameter dependence;

[0024] The high-dimensional implicit information representation unit is used to map the optimized recursive graph features into images, and utilize the Swin-Transformer network to efficiently represent the high-dimensional implicit information and enhance the feature expression capability.

[0025] Preferably, the load identification module further includes a pattern matching unit and a deep learning identification unit;

[0026] The pattern matching unit is used to classify and identify the extracted load features through a predefined load feature template library;

[0027] The deep learning recognition unit is used to automatically learn load characteristics and perform classification, thereby improving the accuracy and generalization ability of load recognition.

[0028] Preferably, the load decomposition module further includes a decomposition unit and an attention mechanism optimization unit;

[0029] The decomposition unit uses Bi-LSTM to model the time series data for further extracting load characteristics and performing load decomposition;

[0030] The attention mechanism optimization unit is used to introduce the attention mechanism through the Transformer layer, enhance the model's attention to key load features, and improve the accuracy of load decomposition.

[0031] Preferably, the communication module further comprises a data transmission unit and a security encryption unit;

[0032] The data transmission unit is used to transmit the decomposed load data to a remote database in real time for storage;

[0033] The security encryption unit is based on the AES encryption algorithm and is used to encrypt the transmitted data to ensure the security of the data during transmission.

[0034] Preferably, the data storage and analysis module further includes a distributed database and an intelligent analysis unit;

[0035] The distributed database uses the Cassandra distributed database system to support high-concurrency access and achieve efficient storage and query of large amounts of data;

[0036] The intelligent analysis unit is based on a data analysis algorithm for machine learning, and is used to automatically analyze load data, identify power consumption behavior patterns, and provide energy-saving suggestion services to users.

[0037] Preferably, the user interaction module further includes a Web / APP interaction unit and a personalized recommendation unit;

[0038] The Web / APP interaction unit is based on the React front-end development framework and is used to provide users with a Web or APP interface to realize real-time query of power consumption and historical data analysis functions;

[0039] The personalized recommendation unit is used to provide personalized electricity usage suggestions and energy-saving solutions according to the user's electricity usage behavior and load decomposition results through collaborative filtering and content-based recommendation algorithms.

[0040] Based on the above system, the present invention also proposes a non-invasive load decomposition method, comprising the following steps:

[0041] S1. Data collection: Real-time collection of voltage and current electrical signals of user trunk lines to ensure high accuracy and anti-interference of data, and perform denoising, filtering, and standardization preprocessing operations on the collected electrical signals to improve data quality;

[0042] S2. Event detection and feature extraction: Real-time detection of load start and stop events in electrical signals, marking the start and stop time points of each electrical device, and extracting load features from electrical signals after event detection, including steady-state and transient features, to provide key information for subsequent load decomposition;

[0043] S3. Feature enhancement: Convert the current time series signal into a specific form of recursive graph, optimize the features through voltage-current correlation, improve feature discrimination, map the optimized features into images, and use deep learning networks to efficiently represent high-dimensional implicit information and enhance feature expression capabilities;

[0044] S4. Load identification: The extracted load features are classified and identified through the predefined load feature template library, and the load features are automatically learned and classified using the deep learning algorithm to improve the accuracy and generalization ability of identification;

[0045] S5. Load decomposition: Use a bidirectional long short-term memory network to model time series data, further extract load features and perform load decomposition, introduce an attention mechanism, enhance the model's attention to key load features, and improve the accuracy of decomposition;

[0046] S6. Data transmission and security encryption: The decomposed load data is transmitted to the remote database in real time for storage, and the transmitted data is encrypted to ensure the security of the data during transmission;

[0047] S7. Data storage and intelligent analysis: Use distributed database systems to support high concurrent access, achieve efficient storage and query of large amounts of data, and automatically analyze load data based on machine learning data analysis algorithms to identify power consumption behavior patterns;

[0048] S8. User interaction and personalized recommendations: Provide users with a Web or APP interface to realize real-time query of electricity usage and historical data analysis functions, and provide personalized electricity usage suggestions and energy-saving solutions based on the user's electricity usage behavior and load decomposition results.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention uses a sliding window data preprocessing algorithm to denoise, filter, and standardize the collected electrical signals to improve data quality, and adopts a change point detection algorithm and a time-frequency analysis method to detect load start and stop events in electrical signals in real time, and extract steady-state and transient characteristics to provide key information for load decomposition. Through recursive graph feature optimization and high-dimensional implicit information characterization technology, the current time series signal is converted into a specific form of recursive graph, and a deep learning network is used to efficiently characterize high-dimensional implicit information, enhance feature expression capabilities, and improve the accuracy of load identification. The deep learning algorithm is used to automatically learn load characteristics and perform classification and identification, and a bidirectional long short-term memory network is used to model time series data and decompose the load. The attention mechanism is introduced to enhance the model's attention to key load features, improve the accuracy and robustness of decomposition, and use wireless or wired transmission methods to transmit the decomposed load data to a remote database for storage in real time. At the same time, the AES encryption algorithm is used to encrypt the transmitted data to ensure data security. The distributed database system is used to support high-concurrency access, achieve efficient storage and query of large amounts of data, provide users with a Web or APP interface, realize real-time query of power consumption and historical data analysis functions, and provide personalized power consumption recommendations and energy-saving plans based on the user's power consumption behavior and load decomposition results. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A topological diagram of a non-intrusive load decomposition system based on deep learning according to the present invention;

[0051] Figure 2 This is a flow chart of the non-invasive load decomposition method of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Example 1

[0054] See also Figure 1 ,The present invention proposes a non-intrusive load decomposition system based on deep learning.,The system includes a data acquisition module, an event detection module, a feature enhancement module, a load identification module, a load decomposition module, a communication module, a data storage and analysis module, and a user interaction module;

[0055] Among them, it should be noted that the data acquisition module of this system is used to provide sensors and data preprocessing, the event detection module of this system is used for load event detection and feature extraction, the feature enhancement module of this system is used for recursive graph feature optimization and high-dimensional implicit information representation, the load identification module of this system is used for pattern matching and deep learning recognition, the load decomposition module of this system is used to further extract load features and optimize the attention mechanism, the communication module of this system is used for data storage and security encryption, the data storage and analysis module of this system is used to build a distributed database and intelligent analysis, and the user interaction module of this system is used to provide Web / APP interaction and personalized recommendations.

[0056] In this embodiment, it should also be noted that the data acquisition module of the system also includes a sensor unit and a preprocessing unit. The sensor unit uses a high-precision power sensor and an analog-to-digital converter to collect the voltage and current electrical signals of the user's trunk line in real time to ensure the high accuracy and anti-interference of the data. The preprocessing unit is based on a sliding window data preprocessing algorithm to perform denoising, filtering, and standardization preprocessing operations on the collected electrical signals to improve data quality and lay the foundation for subsequent analysis.

[0057] In this embodiment, it should also be noted that the event detection module of the system also includes a load event detection unit and a feature extraction unit. The load event detection unit is based on a change point detection algorithm (including CUSUM, EWMA) to detect load start and stop events in electrical signals in real time and mark the start and stop time points of each electrical device. The feature extraction unit combines the time-frequency analysis method of Fourier transform and wavelet transform to extract load characteristics from the electrical signal after event detection, including steady-state and transient characteristics, to provide key information for subsequent load decomposition.

[0058] In this embodiment, it should also be noted that the feature enhancement module of the system also includes a recursive graph feature optimization unit and a high-dimensional implicit information representation unit. The recursive graph feature optimization unit is used to convert the current time series signal into TFRP, and realize power scaling through voltage-current correlation, thereby improving feature discrimination and reducing hyperparameter dependence. The high-dimensional implicit information representation unit is used to map the optimized recursive graph features into images, and use the Swin-Transformer network to efficiently represent high-dimensional implicit information and enhance feature expression capabilities.

[0059] In this embodiment, it should also be noted that the load identification module of the system also includes a pattern matching unit and a deep learning identification unit. The pattern matching unit is used to classify and identify the extracted load features through a predefined load feature template library, and the deep learning identification unit is used to automatically learn and classify load features to improve the accuracy and generalization ability of load identification.

[0060] In this embodiment, it should also be noted that the load decomposition module of this system also includes a decomposition unit and an attention mechanism optimization unit. The decomposition unit uses Bi-LSTM to model the time series data for further extracting load features and performing load decomposition. The attention mechanism optimization unit is used to introduce the attention mechanism through the Transformer layer to enhance the model's attention to key load features and improve the accuracy of load decomposition.

[0061] In this embodiment, it should also be noted that the communication module of the system also includes a data transmission unit and a security encryption unit. The data transmission unit is used to transmit the decomposed load data to a remote database in real time for storage. The security encryption unit is based on the AES encryption algorithm and is used to encrypt the transmitted data to ensure the security of the data during the transmission process.

[0062] In this embodiment, it should also be noted that the data storage and analysis module of the system also includes a distributed database and an intelligent analysis unit. The distributed database uses the Cassandra distributed database system to support high-concurrency access and achieve efficient storage and query of large amounts of data. The intelligent analysis unit is based on a machine learning data analysis algorithm to automatically analyze load data, identify electricity consumption behavior patterns, and provide users with energy-saving recommendation services.

[0063] In this embodiment, it should also be noted that the user interaction module of the system also includes a Web / APP interaction unit and a personalized recommendation unit. The Web / APP interaction unit is based on the React front-end development framework and is used to provide users with a Web or APP interface to realize real-time query and historical data analysis of power consumption. The personalized recommendation unit uses collaborative filtering and content-based recommendation algorithms to provide personalized power consumption suggestions and energy-saving plans based on the user's power consumption behavior and load decomposition results.

[0064] Example 2

[0065] See also Figure 2 Based on the above system, in practical application, the present invention also proposes a non-invasive load decomposition method, comprising the following steps:

[0066] S1. Data collection

[0067] S1.1. Real-time signal acquisition:

[0068] Use high-precision power sensors and analog-to-digital converters (ADCs) to collect voltage and current electrical signals of user trunk lines in real time;

[0069] Ensure high accuracy and anti-interference of data, and provide a reliable basis for subsequent analysis;

[0070] Through this step, high-precision collection ensures the accuracy of the data, which is the prerequisite for subsequent analysis;

[0071] S1.2. Data preprocessing:

[0072] Apply the sliding window data preprocessing algorithm to denoise the collected electrical signals;

[0073] Perform filtering operations to remove high-frequency noise and low-frequency interference in the signal;

[0074] Standardize the data to make it have uniform dimension and distribution;

[0075] Through this step, preprocessing improves data quality and facilitates subsequent feature extraction and model training;

[0076] S2. Event detection and feature extraction

[0077] S2.1. Load event detection:

[0078] Use change point detection algorithms (such as CUSUM, EWMA) to detect load start and stop events in electrical signals in real time;

[0079] Mark the start and shut down time points of each electrical equipment to provide a time basis for load decomposition;

[0080] Through this step, accurate detection of load events helps to identify the operating status of electrical equipment;

[0081] S2.2. Feature extraction:

[0082] The time-frequency analysis method combining Fourier transform and wavelet transform is used to extract load characteristics from the electrical signal after event detection;

[0083] The extracted features include steady-state features (such as power and current effective value) and transient features (such as starting current and voltage fluctuation);

[0084] Through this step, the extracted features provide key information for load decomposition and improve the accuracy of decomposition;

[0085] S3. Feature Enhancement

[0086] S3.1. Recursive graph feature optimization:

[0087] Convert the current time series signal into a threshold-free recursive graph (TFRP) or an adaptive compressed recursive graph (ASRP);

[0088] Power-type scaling is achieved through voltage-current correlation to improve feature differentiation;

[0089] Through this step, the optimized features are more discriminative, which helps to improve the accuracy of load identification;

[0090] S3.2. High-dimensional implicit information representation:

[0091] Map the optimized recursive graph features into image form;

[0092] Use the Swin-Transformer network to efficiently represent high-dimensional implicit information and enhance feature expression capabilities;

[0093] Through this step, the deep learning network can efficiently extract and represent high-dimensional features and improve model performance;

[0094] S4. Load identification

[0095] S4.1. Template matching recognition:

[0096] Establish a predefined load feature template library, including feature templates for various electrical equipment;

[0097] Perform template matching on the extracted load features to preliminarily classify and identify electrical equipment;

[0098] Through this step, template matching can quickly identify common electrical equipment and improve recognition efficiency;

[0099] S4.2. Deep learning recognition:

[0100] Using deep learning algorithms such as convolutional neural networks (CNN) or gated recurrent neural networks (GRU);

[0101] Automatically learn load characteristics and perform classification and recognition to improve recognition accuracy and generalization ability;

[0102] Through this step, deep learning can automatically extract complex features and adapt to different power usage environments and equipment changes;

[0103] S5. Load decomposition

[0104] S5.1. Time series data modeling:

[0105] Use Bi-LSTM to model time series data;

[0106] Further extract load features,considering temporal dependencies and contextual information;

[0107] Through this step, Bi-LSTM can process time series data and improve the accuracy of load decomposition;

[0108] S5.2. Attention mechanism optimization:

[0109] Introducing Transformer layers or similar attention mechanisms;

[0110] Enhance the model's focus on key load characteristics and improve the accuracy and robustness of decomposition;

[0111] Through this step, the attention mechanism can highlight important features and improve the model's adaptability to complex power consumption environments;

[0112] S6. Data transmission and security encryption

[0113] S6.1. Data transmission:

[0114] Use wireless transmission protocols (such as Wi-Fi, LoRa) or wired transmission methods;

[0115] The decomposed load data is transmitted to the remote database in real time for storage;

[0116] Through this step, real-time transmission ensures the timeliness and availability of data;

[0117] S6.2. Security encryption:

[0118] Apply AES encryption algorithm or other secure encryption technology;

[0119] Encrypt the transmitted data to ensure the security and privacy of the data during transmission;

[0120] Through this step, encryption processing prevents data leakage and illegal access, ensuring the data security of users;

[0121] S7.Data storage and intelligent analysis

[0122] S7.1. Distributed database storage:

[0123] Use distributed database systems such as Cassandra and HBase;

[0124] Support high concurrent access to achieve efficient storage and query of large amounts of data;

[0125] Through this step, distributed storage improves data scalability and access efficiency;

[0126] S7.2. Intelligent data analysis:

[0127] Data analysis algorithms based on machine learning (e.g. cluster analysis, anomaly detection);

[0128] Automatically analyze load data, identify electricity usage patterns, and provide users with energy-saving suggestions and other services;

[0129] Through this step, intelligent analysis can mine the value of data and provide users with personalized electricity usage suggestions and energy-saving solutions;

[0130] S8. User interaction and personalized recommendation

[0131] S8.1.Web / APP interactive interface:

[0132] Based on front-end development frameworks such as React and Vue.js;

[0133] Provide users with a Web or APP interface to realize real-time query of electricity consumption, historical data analysis and other functions;

[0134] Through this step, the interactive interface allows users to check power usage at any time, thus improving user experience;

[0135] S8.2. Personalized recommendation service:

[0136] Apply technologies such as collaborative filtering and content-based recommendation algorithms;

[0137] Provide personalized electricity usage suggestions and energy-saving solutions based on the user's electricity usage behavior and load decomposition results;

[0138] Through this step, personalized recommendations can meet the personalized needs of users and improve their satisfaction and energy-saving awareness.

[0139] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A non-intrusive load decomposition system based on deep learning, characterized by: It includes data acquisition module, event detection module, feature enhancement module, load identification module, load decomposition module, communication module, data storage and analysis module, and user interaction module; The data acquisition module is used to provide sensors and data preprocessing; The event detection module is used for load event detection and feature extraction; The feature enhancement module is used for recursive graph feature optimization and high-dimensional implicit information representation; The load identification module is used for pattern matching and deep learning identification; The load decomposition module is used to further extract load features and optimize the attention mechanism; The communication module is used for data transmission and security encryption; The data storage and analysis module is used to build a distributed database and intelligent analysis; The user interaction module is used to provide Web / APP interaction and personalized recommendations.

2. The non-intrusive load decomposition system based on deep learning according to claim 1, characterized in that: The data acquisition module also includes a sensor unit and a preprocessing unit; The sensor unit uses a high-precision power sensor and an analog-to-digital converter to collect voltage and current electrical signals of the user's trunk line in real time to ensure high accuracy and anti-interference of the data; The preprocessing unit is based on a sliding window data preprocessing algorithm, which is used to perform denoising, filtering, and standardization preprocessing operations on the collected electrical signals, thereby improving data quality and laying a foundation for subsequent analysis.

3. The non-intrusive load decomposition system based on deep learning according to claim 2, characterized in that: The event detection module also includes a load event detection unit and a feature extraction unit; The load event detection unit is based on a change point detection algorithm (including CUSUM, EWMA) to detect load start and stop events in the electrical signal in real time and mark the start and stop time points of each electrical device; The feature extraction unit combines the time-frequency analysis method of Fourier transform and wavelet transform to extract load features, including steady-state and transient features, from the electrical signal after event detection, providing key information for subsequent load decomposition.

4. The non-intrusive load decomposition system based on deep learning according to claim 3, characterized in that: The feature enhancement module also includes a recursive graph feature optimization unit and a high-dimensional implicit information representation unit; The recursive graph feature optimization unit is used to convert the current time series signal into TFRP, and realize power-type scaling through voltage-current correlation, improve feature discrimination, and reduce hyperparameter dependence; The high-dimensional implicit information representation unit is used to map the optimized recursive graph features into images, and utilize the Swin-Transformer network to efficiently represent the high-dimensional implicit information and enhance the feature expression capability.

5. The non-intrusive load decomposition system based on deep learning according to claim 4, characterized in that: The load identification module also includes a pattern matching unit and a deep learning identification unit; The pattern matching unit is used to classify and identify the extracted load features through a predefined load feature template library; The deep learning recognition unit is used to automatically learn load characteristics and perform classification, thereby improving the accuracy and generalization ability of load recognition.

6. The non-intrusive load decomposition system based on deep learning according to claim 5, characterized in that: The load decomposition module also includes a decomposition unit and an attention mechanism optimization unit; The decomposition unit uses Bi-LSTM to model the time series data for further extracting load characteristics and performing load decomposition; The attention mechanism optimization unit is used to introduce the attention mechanism through the Transformer layer, enhance the model's attention to key load features, and improve the accuracy of load decomposition.

7. The non-intrusive load decomposition system based on deep learning according to claim 6, characterized in that: The communication module also includes a data transmission unit and a security encryption unit; The data transmission unit is used to transmit the decomposed load data to a remote database in real time for storage; The security encryption unit is based on the AES encryption algorithm and is used to encrypt the transmitted data to ensure the security of the data during transmission.

8. The non-intrusive load decomposition system based on deep learning according to claim 7, characterized in that: The data storage and analysis module also includes a distributed database and an intelligent analysis unit; The distributed database uses the Cassandra distributed database system to support high-concurrency access and achieve efficient storage and query of large amounts of data; The intelligent analysis unit is based on a data analysis algorithm for machine learning, and is used to automatically analyze load data, identify power consumption behavior patterns, and provide energy-saving suggestion services to users.

9. The non-intrusive load decomposition system based on deep learning according to claim 8, characterized in that: The user interaction module also includes a Web / APP interaction unit and a personalized recommendation unit; The Web / APP interaction unit is based on the React front-end development framework and is used to provide users with a Web or APP interface to realize real-time query of power consumption and historical data analysis functions; The personalized recommendation unit is used to provide personalized electricity usage suggestions and energy-saving solutions according to the user's electricity usage behavior and load decomposition results through collaborative filtering and content-based recommendation algorithms.

10. A non-intrusive load decomposition method, according to any one of claims 1 to 9, wherein the non-intrusive load decomposition system based on deep learning is characterized in that: The following steps are involved: S1. Data collection: Real-time collection of voltage and current electrical signals of user trunk lines to ensure high accuracy and anti-interference of data, and perform denoising, filtering, and standardization preprocessing operations on the collected electrical signals to improve data quality; S2. Event detection and feature extraction: Real-time detection of load start and stop events in electrical signals, marking the start and stop time points of each electrical device, and extracting load features from electrical signals after event detection, including steady-state and transient features, to provide key information for subsequent load decomposition; S3. Feature enhancement: Convert the current time series signal into a specific form of recursive graph, optimize the features through voltage-current correlation, improve feature discrimination, map the optimized features into images, and use deep learning networks to efficiently represent high-dimensional implicit information and enhance feature expression capabilities; S4. Load identification: The extracted load features are classified and identified through the predefined load feature template library, and the load features are automatically learned and classified using the deep learning algorithm to improve the accuracy and generalization ability of identification; S5. Load decomposition: Use a bidirectional long short-term memory network to model time series data, further extract load features and perform load decomposition, introduce an attention mechanism, enhance the model's attention to key load features, and improve the accuracy of decomposition; S6. Data transmission and security encryption: The decomposed load data is transmitted to the remote database in real time for storage, and the transmitted data is encrypted to ensure the security of the data during transmission; S7. Data storage and intelligent analysis: Use distributed database systems to support high concurrent access, achieve efficient storage and query of large amounts of data, and automatically analyze load data based on machine learning data analysis algorithms to identify power consumption behavior patterns; S8. User interaction and personalized recommendations: Provide users with a Web or APP interface to realize real-time query of electricity usage and historical data analysis functions, and provide personalized electricity usage suggestions and energy-saving solutions based on the user's electricity usage behavior and load decomposition results.

Images