Non-intrusive load monitoring method and system based on dual attention mechanism

Abstract

The application discloses a non-intrusive load monitoring method and system based on a double attention mechanism, belongs to the intelligent power grid and artificial intelligence technical field, and collects electrical quantities of total loads and electrical quantities of each device through a measuring device to construct a non-intrusive load monitoring data set; for a target electrical signal, a self-attention mechanism is adopted to extract target electrical quantity features; for a measuring noise signal, a noise attention mechanism is adopted to extract measuring noise features; based on the above double attention mechanism, a non-intrusive load intelligent monitoring model is trained; and the trained non-intrusive load intelligent monitoring model is used to realize real-time monitoring of the working state of the electrical equipment. The application improves the robustness of the non-intrusive load monitoring model to noise, improves the calculation efficiency of the non-intrusive load monitoring, and can support the intelligent perception of the electrical equipment of a new power system, thereby promoting the digitalization and intelligentization development of the new power system.

Description

Technical Field

[0001] The present invention belongs to the field of smart grid and artificial intelligence technology, and particularly relates to a non-invasive load monitoring method and system based on a dual attention mechanism. Background Art

[0002] With the increasing rate of electrification, optimizing energy consumption has become a crucial measure for improving energy efficiency, promoting sustainable development, and enhancing economic returns. A deep understanding of energy consumption patterns is crucial for developing effective energy management strategies. Furthermore, energy consumption analysis can benefit both supply and demand, reducing costs while also improving grid resilience and reliability.

[0003] Non-intrusive load monitoring technology, a method for decomposing electricity consumption without installing measurement equipment on each device, has become a crucial tool for device-level energy consumption awareness and monitoring. This technology plays a crucial role in supporting load forecasting, demand response, and grid status assessment.

[0004] In recent years, deep learning methods have demonstrated strong potential in the field of non-intrusive load monitoring. They can automatically extract multi-level features from complex raw data and adapt to high-dimensional and nonlinear power signals. However, despite the various deep learning models proposed in existing research, a systematic suppression mechanism for measurement noise has yet to be established. This noise has become a significant factor limiting the accuracy of non-intrusive load monitoring. Summary of the Invention

[0005] To address the shortcomings of existing technologies, achieve the goal of ensuring real-time load monitoring while significantly reducing the interference of measurement noise on the results, and improving the accuracy and stability of non-intrusive load monitoring, the present invention adopts the following technical solutions:

[0006] The non-intrusive load monitoring method based on the dual attention mechanism includes the following steps:

[0007] Step S1: Collecting electrical quantity data of the total load and electrical quantity data of each device to construct a non-intrusive load monitoring data set;

[0008] Step S2: for the total load electrical quantity data including the target electrical signal and the measurement noise signal, a self-attention mechanism is used to extract the target electrical quantity features;

[0009] Step S3: To effectively suppress the interference of measurement noise on load decomposition, a noise self-attention mechanism is used to extract measurement noise features from the total signal. This captures the randomly fluctuating measurement noise signal and generates a measurement noise feature vector from the total signal, thereby assisting the model in adaptively suppressing measurement noise interference.

[0010] Step S4: fuse the target electrical quantity feature, the measurement noise feature and the feature of the original sequence to construct a feature fusion sequence for training the non-intrusive load monitoring model;

[0011] Step S5: use the trained non-intrusive load monitoring model to perform real-time non-intrusive load monitoring of the working state of the electrical equipment.

[0012] Further, in the step S2, based on the aggregated power of the electrical equipment, a multi-head self-attention mechanism is used to extract features of the target electrical signal to capture the dependency relationship between the target electrical signals in the total signal at different time points, generate a target electrical signal feature vector, effectively distinguish the running modes of different devices, and improve the accuracy of load monitoring.

[0013] Further, in the step S2, the aggregated power of all electrical equipment at a certain moment is obtained by multiplying the on-off state of the electrical equipment at a certain moment and the power, then adding the measurement noise based on the number of electrical equipment.

[0014] Further, in the step S2, the aggregated power sequence of all electrical equipment in a period of time is multiplied by the first embedding matrix as the query vector of the attention mechanism, and the aggregated power of all electrical equipment at a certain moment is multiplied by the second first embedding matrix as the key vector of the attention mechanism. Then, the two products are multiplied, and the result is divided by the square root of the dimension of the aggregated power vector of all electrical equipment at a certain moment. The result is activated by an activation function to obtain the target electrical quantity feature extracted based on the self-attention mechanism.

[0015] Further, in the step S3, the result is taken as a negative value, and then activated by an activation function to obtain the measurement noise signal feature.

[0016] Further, in the step S4, an improved information extraction network Informer deep learning method is used for data-driven non-intrusive load monitoring. Through a three-channel parallel processing mechanism, the target electrical quantity feature and the measurement noise feature are analyzed synchronously, and a sparse attention mechanism is used to focus on key data segments with high correlation and significant noise interference, realizing high-precision and real-time intelligent load monitoring.

[0017] Further, in the step S4, the target electrical quantity feature, the measurement noise feature and the original sequence are fused by a fusion feature function to obtain a feature fusion sequence. After the feature fusion sequence passes through a one-dimensional convolution network of three channels, it is activated by an activation function to obtain the parallel processed fusion feature.

[0018] Further, in the step S4, the sparse attention mechanism of the information extraction network reduces the time and spatial complexity from the standard Transformer Significantly reduced to , where O represents the time complexity and L represents the sequence length of the fused features after parallel processing, thereby focusing on key data segments with high correlation and significant noise interference.

[0019] Furthermore, in step S1, the electrical quantity data of the total load is sampled at low or high frequency, and the electrical quantity data of each device is recorded by an independent measuring device. The data is preprocessed by methods such as data cleaning and filling to construct a non-intrusive load monitoring data set for typical power consumption scenarios, providing data support for subsequent model training.

[0020] A non-intrusive load monitoring system based on a dual attention mechanism includes a data set construction module, a target electrical quantity feature extraction module, a measurement noise feature extraction module, a feature fusion module, and a non-intrusive load monitoring module. The non-intrusive load monitoring method based on the dual attention mechanism is used to construct a non-intrusive load monitoring data set, extract target electrical quantity features and measurement noise features respectively, and then perform feature fusion in combination with the features of the original sequence. The non-intrusive load monitoring model in the non-intrusive load monitoring module is trained so that the trained non-intrusive load monitoring model can perform real-time non-intrusive load monitoring of the working status of electrical equipment.

[0021] The advantages and beneficial effects of the present invention are:

[0022] Under real-time monitoring conditions, the present invention uses a dual attention mechanism to separate the target electrical characteristics and the measurement noise characteristics, thereby improving the robustness of the non-invasive load monitoring model to noise; through the sparse attention mechanism, it focuses on key time periods and important features, further improving the computational efficiency of non-invasive load monitoring; achieving high-precision, real-time non-invasive load monitoring, and supporting the intelligent perception and control of new power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 4 is a flow chart of a non-intrusive load monitoring method in an embodiment of the present invention.

[0024] Figure 2 Schematic diagram of the structure of a non-intrusive load monitoring system in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following describes the specific embodiments of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention and are not intended to limit the present invention.

[0026] In view of the real-time requirements of current non-intrusive load monitoring methods and the problem of insufficient consideration of measurement noise of measurement equipment, this paper proposes a non-intrusive load monitoring method based on a dual attention mechanism to achieve higher accuracy under real-time monitoring conditions, such as Figure 1 As shown, the following steps are included:

[0027] Step S1: The electrical quantity data of the total load and the electrical quantity data of each device are collected through measuring equipment to construct a non-intrusive load monitoring data set. The electrical quantity data of the total load can be sampled at low or high frequency, and the electrical quantity data of each device is recorded by an independent measuring device. Data preprocessing is performed through methods such as data cleaning and filling to construct a non-intrusive load monitoring data set for typical power consumption scenarios, providing data support for subsequent model training.

[0028] Step S2: For the total load electrical quantity data containing the target electrical signal and the measurement noise signal, a self-attention mechanism is used to extract the target electrical quantity features; specifically, a multi-head self-attention mechanism is used to extract the features of the target electrical signal in the signal. This mechanism can capture the dependency between the target electrical signals in the total signal at different time points and generate a target electrical signal feature vector, thereby effectively distinguishing the operating modes of different devices and improving the accuracy of load monitoring.

[0029] In one embodiment, the non-intrusive load monitoring model can be expressed as follows:

[0030]

[0031] in, represents the aggregate power of all electrical devices at time t, Indicates the switch status of the i-th electrical device at time t, represents the power of the i-th electrical device at time t, represents the measurement noise of the measurement device, and N represents the total number of electrical devices.

[0032] In one embodiment, the multi-head self-attention mechanism can be expressed as follows:

[0033]

[0034] in, represents the target electrical signal features extracted based on the self-attention mechanism, represents the aggregate power sequence from time t-L+1 to t, and represents the embedding matrix, q and k represent the query vector in the attention mechanism With key vector The index of express The dimension of the vector, Represents the activation function.

[0035] Step S3: To effectively suppress the interference of measurement noise on load decomposition, a noise self-attention mechanism is used to extract the measurement noise signal features from the total signal. This mechanism can capture the randomly fluctuating measurement noise signal and generate a measurement noise feature vector in the total signal, thereby assisting the model in adaptively suppressing measurement noise interference.

[0036] In one embodiment, the noisy self-attention mechanism can be expressed as follows:

[0037]

[0038] in, represents the measurement noise signal features extracted based on the noise self-attention mechanism, represents the aggregate power sequence from time t-L+1 to t, and represents the embedding matrix, q and k represent the query vector in the attention mechanism With key vector The index of express The dimension of the vector, Represents the activation function.

[0039] Step S4: Based on the target electrical signal features and measurement noise signal features extracted by the above-mentioned dual attention mechanism, the two are intelligently fused with the original sequence to construct a feature fusion sequence for training the non-intrusive load monitoring model; further, deep learning methods such as the improved Informer (Information Extractor Network) are used to realize data-driven non-intrusive load monitoring. Specifically, this method synchronously analyzes the target electrical signal features and the measurement noise signal features through a three-channel parallel processing mechanism, and adopts a sparse attention mechanism to focus on key data segments with high correlation and significant noise interference, thereby realizing high-precision and real-time intelligent load monitoring.

[0040] In one embodiment, the intelligent fusion of the features extracted by the two branches and the original sequence can be expressed as follows:

[0041]

[0042] in, represents the fused feature sequence, Represents the function used to fuse features.

[0043] In one embodiment, the three-channel parallel processing mechanism can be expressed as follows:

[0044]

[0045] Among them, Conv1D represents a three-channel one-dimensional convolutional network. Represents the feature vector after parallel processing, which serves as the input of Informer.

[0046] In one implementation, the Informer's sparse attention mechanism reduces the time and space complexity from the standard Transformer to Significantly reduced to , where L represents The sequence length is shortened to focus on the key data segments with high correlation and significant noise interference.

[0047] Step S5: Using the trained non-intrusive load monitoring model, perform real-time non-intrusive load monitoring of the working status of the electrical equipment.

[0048] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A non-intrusive load monitoring method based on a dual attention mechanism, characterized by The steps include: Step S1: Collecting electrical quantity data of the total load and electrical quantity data of each device to construct a non-intrusive load monitoring data set; Step S2: for the total load electrical quantity data including the target electrical signal and the measurement noise signal, a self-attention mechanism is used to extract the target electrical quantity features; Step S3: For the measurement noise signal, a noise self-attention mechanism is used to extract the measurement noise features in the total signal; Step S4: Fusing the target electrical quantity features, the measurement noise features, and the features of the original sequence to construct a feature fusion sequence for training the non-intrusive load monitoring model; Step S5: Using the trained non-intrusive load monitoring model, perform real-time non-intrusive load monitoring of the working status of the electrical equipment.

2. The non-intrusive load monitoring method based on the dual attention mechanism according to claim 1, characterized in that: In step S2, based on the aggregated power of the electrical equipment, a multi-head self-attention mechanism is used to extract features of the target electrical signal to capture the dependencies between the target electrical signals in the total signal at different time points, and generate a target electrical signal feature vector to distinguish the operating modes of different devices.

3. The non-intrusive load monitoring method based on the dual attention mechanism according to claim 2, characterized in that: In step S2, the switch state of the electrical device at a certain moment is multiplied by the power, and then accumulated based on the number of electrical devices. After adding the measured noise, the aggregate power of all electrical devices at a certain moment is obtained.

4. The non-intrusive load monitoring method based on the dual attention mechanism according to claim 2, characterized in that: In step S2, the aggregate power sequence of all electrical devices over a period of time is used as the query vector of the attention mechanism and multiplied by the first embedding matrix, and the aggregate power of all electrical devices at a certain moment is used as the key vector of the attention mechanism and multiplied by the second first embedding matrix. The two products are then multiplied and divided by the square root of the dimension of the aggregate power vector of all electrical devices at a certain moment. The result is activated by the activation function to obtain the target electrical quantity feature extracted based on the self-attention mechanism.

5. The non-intrusive load monitoring method based on the dual attention mechanism according to claim 4, characterized in that: In step S3, the obtained result is negated and then activated by an activation function to obtain the measurement noise signal feature.

6. The non-intrusive load monitoring method based on the dual attention mechanism according to claim 1, characterized in that: In step S4, an information extraction network is used to perform data-driven non-intrusive load monitoring. A three-channel parallel processing mechanism is used to synchronously analyze the target electrical quantity characteristics and the measurement noise characteristics. A sparse attention mechanism is used to focus on key data segments with high correlation and significant noise interference.

7. The non-intrusive load monitoring method based on the dual attention mechanism according to claim 6, characterized in that: In step S4, the target electrical quantity feature, the measurement noise feature, and the original sequence are fused by a fusion feature function to obtain a feature fusion sequence. The feature fusion sequence is passed through a three-channel one-dimensional convolutional network and then activated by an activation function to obtain a fusion feature after parallel processing.

8. The non-intrusive load monitoring method based on the dual attention mechanism according to claim 6, characterized in that: In step S4, the sparse attention mechanism of the information extraction network reduces the time and space complexity to , where O represents the time complexity and L represents the sequence length of the fusion features after parallel processing.

9. The non-intrusive load monitoring method based on the dual attention mechanism according to claim 1, characterized in that: In step S1, the electrical quantity data of the total load is sampled at a low frequency or a high frequency, and the electrical quantity data of each device is recorded by an independent measuring device.

10. A non-intrusive load monitoring system based on a dual attention mechanism, comprising a dataset construction module, a target electrical quantity feature extraction module, a measurement noise feature extraction module, a feature fusion module, and a non-intrusive load monitoring module, characterized by: A non-invasive load monitoring method based on a dual attention mechanism as described in any one of claims 1 to 9 is adopted to construct a non-invasive load monitoring data set, extract the target electrical quantity features and the measurement noise features respectively, and then perform feature fusion in combination with the features of the original sequence, and train a non-invasive load monitoring model in a non-invasive load monitoring module so that the trained non-invasive load monitoring model can perform real-time non-invasive load monitoring of the working status of electrical equipment.

Images