System and method for identifying electric equipment by using electromagnetic wave form of null line

Abstract

The invention discloses a system and a method for identifying electric equipment by using a null line electromagnetic waveform, and belongs to the technical field of non-intrusive load monitoring. The system comprises a high-frequency current sensor which is installed on a main zero line of a distribution box and is used for collecting an instantaneous current signal in a non-contact manner; the signal conditioning module comprises a 3kHz-150kHz band-pass filter and an amplifying circuit; the sampling rate of the analog-to-digital converter is not lower than 500kS / s, and the quantization bit is greater than or equal to 12 bits; the embedded processing unit is internally provided with a one-dimensional convolutional neural network model combined with a channel attention mechanism, and the model comprises at least three one-dimensional convolutional layers, two pooling layers, an attention layer and a full-connection output layer; and the storage module is used for storing equipment identity, waveform characteristics and type information. According to the method, live wire interference is effectively avoided by analyzing the high-frequency electromagnetic waveform of the null line, high-precision and high-robustness equipment identification can still be realized in complex scenes such as multi-equipment concurrency and power grid disturbance, and the method is suitable for smart home and energy management systems.

Description

Technical Field

[0001] This invention belongs to the field of non-intrusive load monitoring technology, and in particular relates to a system and method for identifying electrical equipment using the electromagnetic waveform of the neutral wire. Background Technology

[0002] Existing non-intrusive electrical equipment identification technologies are mostly based on voltage or current signals at the main incoming line, identifying equipment by analyzing active power changes, harmonic components, or transient events. However, live wire current is susceptible to grid fluctuations and interference from multiple devices, leading to feature confusion. Furthermore, conventional one-dimensional convolutional neural network models have limited ability to distinguish key frequency band features in time series data, resulting in insufficient identification stability in scenarios with multiple concurrent devices. Currently, there is a lack of an embedded identification system and method specifically designed for high-frequency electromagnetic waveforms on the neutral wire, incorporating an attention mechanism to enhance feature selection capabilities. Summary of the Invention

[0003] The purpose of this invention is to provide a system and method for identifying electrical equipment using the electromagnetic waveform of the neutral wire. By collecting high-frequency electromagnetic waveform signals on the main neutral wire of the distribution box, and combining a one-dimensional convolutional neural network model with bandpass filtering, high sampling rate analog-to-digital conversion, and fusion attention mechanism for feature extraction and classification, this invention solves the problems of existing non-intrusive load monitoring technologies, which rely on live wire current and are therefore susceptible to power grid fluctuations and interference from multiple devices, resulting in severe feature confusion. It also addresses the issues of traditional models having insufficient sensitivity to key frequency bands and low identification accuracy and robustness in scenarios with multiple devices operating concurrently.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0005] This invention relates to a system and method for identifying electrical equipment using neutral wire electromagnetic waveforms, comprising:

[0006] A high-frequency current sensor is installed on the main neutral line inside the distribution box to collect the instantaneous current signal flowing through the main neutral line;

[0007] A signal conditioning module is electrically connected to the high-frequency current sensor, and the signal conditioning module includes a bandpass filter and an amplifier circuit.

[0008] An analog-to-digital converter, whose input is connected to the output of the signal conditioning module, is used to convert analog signals into digital sampling sequences;

[0009] An embedded processing unit, communicatively connected to the analog-to-digital converter, has a built-in one-dimensional convolutional neural network model incorporating an attention mechanism. This model includes at least three one-dimensional convolutional layers, two pooling layers, one attention layer, and one fully connected output layer. The embedded processing unit is configured to receive the digital sampling sequence, divide the digital sampling sequence into multiple sub-sequences with fixed time windows, and input each sub-sequence into the one-dimensional convolutional neural network model incorporating an attention mechanism for classification operations, and output the identification of at least one electrical device currently in operation during the current time period.

[0010] The storage module, connected to the embedded processing unit, is used to store the identity identifier, the corresponding waveform feature parameters, and the device type information.

[0011] The present invention is further configured such that the high-frequency current sensor is a Rogowski coil, which is arranged around the outer periphery of the main neutral line and has no electrical contact with the main neutral line.

[0012] The present invention is further configured such that the passband frequency range of the bandpass filter is 3kHz to 150kHz.

[0013] The present invention is further configured such that the sampling frequency of the analog-to-digital converter is not less than 500 kS / s and the quantization bit depth is not less than 12 bits.

[0014] The present invention is further configured such that the fixed time window has a duration of 200 milliseconds and adjacent time windows have a 50% time overlap.

[0015] The present invention is further configured to include a wireless communication module, which is connected to the embedded processing unit and is used to upload the identity identifier to a remote server.

[0016] The present invention is further configured such that the one-dimensional convolutional neural network model incorporating the attention mechanism is trained based on a dataset of neutral electromagnetic waveforms generated by multiple known electrical devices in individual operating states collected in history.

[0017] A method for using a system to identify electrical equipment using electromagnetic waveforms of a neutral wire, characterized by the following steps: acquiring instantaneous current signals flowing through the main neutral wire using a high-frequency current sensor installed on the main neutral wire of a distribution box; sequentially performing bandpass filtering and amplitude amplification on the instantaneous current signals to obtain conditioned analog signals; converting the conditioned analog signals into digital sampling sequences at a sampling rate of not less than 500 kS / s; dividing the digital sampling sequences into multiple subsequences according to a 200-millisecond time window, with 50% overlap between adjacent subsequences; inputting each subsequence into a pre-trained one-dimensional convolutional neural network model incorporating an attention mechanism for classification operations, wherein the one-dimensional convolutional neural network model incorporating an attention mechanism includes at least three one-dimensional convolutional layers, two pooling layers, one attention layer, and one fully connected output layer; and outputting the identification identifier of at least one electrical device currently in operation based on the classification results, wherein the identification identifier includes a unique device code, rated power range, and electrical appliance category.

[0018] The present invention has the following beneficial effects:

[0019] 1. This invention effectively avoids feature confusion caused by voltage fluctuations, harmonic interference, and multi-device coupling on the live wire side by collecting high-frequency electromagnetic waveforms on the main neutral wire, and significantly improves the accuracy of equipment identification in complex power environments.

[0020] 2. The present invention employs a one-dimensional convolutional neural network with a fusion channel attention mechanism, which can adaptively focus on the key frequency bands and time domain features of different electrical devices in the neutral waveform, and accurately identify each device in a scenario where multiple devices are operating simultaneously.

[0021] 3. The system provided by this invention is based on a training model of general equipment waveform data collected in the laboratory. When deployed under different home users, different brand models and line impedance conditions, the average recognition accuracy still reaches 91.3%, which has good cross-user generalization performance.

[0022] 4. The system provided by this invention is non-invasive, easy to deploy, and low in cost: it only requires the installation of a Rogowski coil sensor on the main neutral line of the distribution box, without the need to modify the original circuit or install a monitoring device at each device end, making it suitable for the rapid transformation and large-scale application of existing home and building systems.

[0023] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0025] Figure 1This is a schematic diagram of the principle of the present invention.

[0026] Figure 2 This is a schematic diagram of the method flow of the present invention.

[0027] Figure 3 This is a schematic diagram of the model training of the present invention.

[0028] The attached diagram lists the components represented by each number as follows:

[0029] 1. High-frequency current sensor; 2. Signal conditioning module; 21. Bandpass filter; 22. Amplifier circuit; 3. Analog-to-digital converter; 4. Embedded processing unit; 5. Storage module; 6. Wireless communication module. Detailed Implementation

[0030] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0031] Please see Figure 1 - The electrical equipment identification system provided by this invention is integrated and installed inside the user's single-phase 220V distribution box. Its hardware components include: a high-frequency current sensor 1, a signal conditioning module 2, an analog-to-digital converter 3, an embedded processing unit 4, a storage module 5, and an optional wireless communication module 6.

[0032] in:

[0033] The high-frequency current sensor 1 uses an openable Rogowski coil, which is surrounded around the main neutral line of the distribution box and has no electrical contact with the neutral line, thus achieving non-invasive sensing. The frequency response range of this sensor is 1kHz to 200kHz, which can effectively capture the high-frequency transient current signal generated on the neutral line during the start-up, shutdown or operation of electrical equipment.

[0034] The signal conditioning module 2 is electrically connected to the output of the high-frequency current sensor 1 and includes a bandpass filter 21 and an amplifier circuit 22. The passband of the bandpass filter 21 is set to 3kHz–150kHz to filter out 50Hz power frequency and its low-frequency harmonic interference. The amplifier circuit 22 adjusts the amplitude of the filtered weak signal to the range of 0–3.3V to match the input requirements of the subsequent analog-to-digital converter.

[0035] The analog-to-digital converter 3 uses the high-precision chip ADS8556, with a sampling frequency of 500kS / s and a quantization bit depth of 12 bits. It converts the conditioned analog signal into a digital sampling sequence in real time and transmits it to the embedded processing unit 4 via the SPI bus.

[0036] Embedded processing unit 4 uses an STM32H743 microcontroller with an ARM Cortex-M7 core and runs the lightweight inference engine TensorFlow LiteMicro. It is pre-loaded with a one-dimensional convolutional neural network model (CA-1D-CNN) incorporating a channel attention mechanism. The model structure is as follows:

[0037] The input layer receives a digital sampled subsequence with a length of 10,000 points (corresponding to 200ms, sampling rate 500kS / s);

[0038] First convolutional layer: 64 one-dimensional convolutional kernels with a kernel length of 64 and a stride of 8, activated by ReLU;

[0039] First max pooling layer: window length 16, stride 16;

[0040] The second convolutional layer consists of 128 convolutional kernels with a kernel length of 32 and a stride of 4, activated by ReLU.

[0041] Second max pooling layer: window length 8, stride 8;

[0042] The third convolutional layer consists of 256 convolutional kernels with a kernel length of 16 and a stride of 2, activated by ReLU.

[0043] Attention layer: Employs an SE (Squeeze-and-Excitation) structure, which uses global average pooling, two fully connected layers (compression ratio 1 / 16), and Sigmoid to generate channel weights to achieve feature recalibration;

[0044] After global average pooling, a fully connected output layer is formed, which outputs K-dimensional classification probabilities (K is the preset number of device categories), which are then normalized by Softmax.

[0045] When the system is working, the embedded processing unit 4 divides the continuously acquired digital sampling sequence into 200ms time windows, with adjacent windows overlapping by 50% (i.e., a new subsequence is generated every 100ms), and inputs each subsequence into the CA-1D-CNN model for real-time inference, outputting the identification of one or more electrical devices that are currently running in the current time period, including the device's unique code, the type of electrical appliance (such as heating, motor, or electronic), and the rated power range.

[0046] Storage module 5 is a W25Q64 non-volatile Flash chip, divided into two partitions: a device feature library partition for storing waveform templates and category labels of known devices; and an operation log partition for recording identification results, timestamps, and abnormal events.

[0047] The wireless communication module 6 uses the ESP8266 Wi-Fi module and communicates with the embedded processing unit 4 through the UART interface. It uploads the recognition results to the remote cloud server in real time, supporting the energy management platform to perform visual analysis and electricity behavior modeling.

[0048] Model training process: In a laboratory environment, neutral current signals of typical household appliances such as electric kettles, microwave ovens, air conditioner compressors, and LED lighting were collected under stable operating conditions. No less than 50 data segments were collected for each type of device, with each segment lasting 2 seconds. Training samples were generated from the raw data by dividing it into 200ms windows (50% overlap) and labeling them with the corresponding device labels. The CA-1D-CNN model was trained end-to-end using the cross-entropy loss function and the Adam optimizer until the accuracy of the validation set converged. After training, the model was quantized and deployed to the embedded processing unit.

[0049] This invention has been validated in multiple home scenarios: the highest recognition accuracy reached 99.3% in single-device testing; the F1-score was 94.8% in multi-device concurrent scenarios; the average accuracy across user deployments was 91.3%; and it maintained stable recognition performance even under injected harmonics or voltage sag interference. In summary, this system requires no modification to the user's existing wiring and achieves high-precision, robust, non-intrusive device-level power consumption monitoring solely through the neutral wire waveform. It is suitable for applications such as smart homes, building energy efficiency management, and power demand-side response.

[0050] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

Claims

1. A system for identifying electrical equipment using neutral wire electromagnetic waveforms, characterized in that, The utility model relates to a kind of power utilization identification method based on high-frequency current sensor, including: High-frequency current sensor (1) is installed on the main zero line in distribution box, for collecting the instantaneous current signal flowing through the main zero line; Signal conditioning module (2) is electrically connected with the high-frequency current sensor (1), and the signal conditioning module (2) includes band-pass filter (21) and amplification circuit (22); Analog-digital converter (3) is connected with the output end of the signal conditioning module (2), for converting analog signal into digital sampling sequence; Embedded processing unit (4) is communicatively connected with the analog-digital converter (3), and one-dimensional convolutional neural network model combined with attention mechanism is built-in, which includes at least three one-dimensional convolutional layers, two pooling layers, an attention layer and a fully connected output layer, the embedded processing unit (4) is configured to receive the digital sampling sequence, divide the digital sampling sequence into a plurality of fixed time window subsequences, and input each subsequence into the one-dimensional convolutional neural network model combined with attention mechanism for classification operation, and output the identity of at least one electrical equipment running in the current period; Storage module (5) is connected with the embedded processing unit (4), for storing the identity, corresponding waveform characteristic parameter and device type information.

2. The system for identifying an electrical device using a zero-line electromagnetic waveform according to claim 1, characterized in that, The high-frequency current sensor (1) is a Rogowski coil, which is arranged around the outer periphery of the main zero line and has no electrical contact with the main zero line.

3. The system for identifying an electrical device using a zero-line electromagnetic waveform according to claim 1, characterized in that, The passband frequency range of the band-pass filter (21) is 3kHz to 150kHz.

4. The system for identifying an electrical device using a zero-line electromagnetic waveform according to claim 1, characterized by, The sampling frequency of the analog-digital converter (3) is not less than 500kS / s, and the quantization bit number is not less than 12 bits.

5. The system for identifying electrical devices using a zero-line electromagnetic waveform of claim 1, wherein, The time length of the fixed time window is 200 milliseconds, and there is a 50% time overlap between adjacent time windows.

6. The system for identifying electrical devices using a zero-line electromagnetic waveform of claim 1, wherein, It also includes a wireless communication module (6) connected with the embedded processing unit (4) for uploading the identity to a remote server.

7. The system for identifying electrical devices using a zero-line electromagnetic waveform according to claim 1, wherein The one-dimensional convolutional neural network model combined with attention mechanism is trained based on the zero line electromagnetic waveform data set generated by a plurality of known electrical equipment in separate running state.

8. The method of using a system for identifying electrical devices using a zero-line electromagnetic waveform according to any of claims 1-7, wherein, The utility model relates to a kind of power utilization identification method based on high-frequency current sensor, including: Collecting the instantaneous current signal flowing through the main zero line by the high-frequency current sensor (1) installed on the main zero line in distribution box;The instantaneous current signal is sequentially subjected to band-pass filtering and amplitude amplification to obtain the analog signal after conditioning;The analog signal after conditioning is converted into a digital sampling sequence at a sampling rate of not less than 500kS / s;The digital sampling sequence is divided into a plurality of subsequences according to a 200 millisecond time window, and there is a 50% overlap between adjacent subsequences;Each subsequence is input into a pre-trained one-dimensional convolutional neural network model combined with attention mechanism for classification operation, and the one-dimensional convolutional neural network model combined with attention mechanism includes at least three one-dimensional convolutional layers, two pooling layers, an attention layer and a fully connected output layer;According to the classification result, the identity of at least one electrical equipment running in the current period is output, and the identity includes device unique code, rated power interval and belonging appliance category.

Images