Zero-crossing point detection system for distorted sine signals

Abstract

A zero-crossing detection system for distorted sine signals, comprising: a signal acquisition unit for capturing real-time signals, a data preprocessing module for filtering and segmenting signals, a module for zero-crossing detection based on a deep neural network and a user interface for displaying detected zero crossing points.

Description

AREA OF INVENTION

[0001] The present invention relates to the field of signal processing, in particular the detection of zero crossings in distorted sine signals for applications such as grid synchronization, power conversion, and switchgear protection. It utilizes advanced machine learning techniques to improve detection accuracy in the presence of noise and harmonic distortion. BACKGROUND OF THE INVENTION

[0002] Precise detection of the zero crossing point in a distorted sine wave signal for grid synchronization, power conversion and protection of switchgear. • The approach described in US patent no. 6,285,140B1 involves the use of a microcontroller equipped with a zero-crossing point detector to generate an AC-synchronized time-delay pulse that controls a triac to an on or off state in order to send varying electrical power to a light-emitting diode load. • In Chinese patent no. CN101871965B, unary linear regression theory is used to determine the time of the zero crossing point in the sinusoidal power signal. • Currently, electronic comparator circuits are used to identify the zero crossing point in the sine signal.

[0003] Electronic comparator circuits can effectively detect the zero crossing point in a pure sine wave signal, but in a heavily distorted sine wave signal with noise and harmonic components, the detection of the zero crossing point may fail.

[0004] None of the above-mentioned points of the prior art disclose, either alone or in combination, what the present invention discloses. This invention relates to a process for constructing a deep neural network model that detects the zero-crossing point in a distorted sine wave signal. Summary of the invention

[0005] The invention provides a system for the accurate detection of the zero crossing point in a distorted sine signal using machine learning techniques, in particular deep neural networks (DNNs).

[0006] The proposed solution captures and processes sinusoidal signals with noise and harmonic distortion in real time. A dataset of signals with varying levels of noise and harmonic distortion (THD) is used to train the deep neural network (DNN). The system comprises several hardware and software components to ensure robust and accurate zero-crossing detection in real-world applications.

[0007] An approach based on Deep Neural Network8 (DNN), which is capable of developing a highly complex and nonlinear relationship between input and output features, is used to detect the zero crossing point in a distorted sine signal.

[0008] Heavily distorted sine waves are generated based on the noise level and total harmonic distortion (THD), as shown below. The data extracted from these signals is used to train the DNN model, enabling it to effectively detect the zero-crossing point of distorted sine waves in real time. DETAILED DESCRIPTION OF THE INVENTION

[0009] An approach based on Deep Neural Network8 (DNN), which is capable of developing a highly complex and nonlinear relationship between input and output features, is used to detect the zero crossing point in a distorted sine signal.

[0010] Heavily distorted sine waves are generated based on the noise level and total harmonic distortion (THD), as shown below. The data extracted from these signals is used to train the DNN model, enabling it to effectively detect the zero-crossing point of distorted sine waves in real time. data set Noise Total harmonic distortion Samples 1 10%-50% - 4936 2 - 10%-50% 4436 3 10% - 40% 50% 3949 4 5% - 20% - 3949

[0011] Hardware components: 1. Signal acquisition unit: ◯ Sensors: Capture sinusoidal signals in real time, including voltage and current waveforms. ◯ Analog-to-digital converter (ADC): Converts analog signals into a digital form for processing. 2. Data processing unit: ◯ Microcontroller or digital signal processor (DSP): Controls the signal acquisition process and performs preprocessing of the signal, such as filtering and noise reduction. 3. Zero-crossing detection module: ◯ Deep Neural Network Processor: This processor executes the trained DNN model, which is responsible for detecting zero crossing points from distorted signals. ◯ Storage unit: Stores the trained DNN model and parameters such as weights, biases, and hyperparameters. 4. Power supply: Provides power to the entire system, including sensors, processors and other components. 5. User interface: ◯ A display or communication interface for showing real-time information about zero crossing points and any signal distortion statistics. Software components: 1. Data preprocessing module: Normalizes and segments the captured signals to extract meaningful features. It prepares the data for DNN input by removing high-frequency noise and low-frequency drift and decomposing the signal into relevant subcomponents such as amplitude and frequency. 2. Feature Selection Module: Identifies key signal features useful for zero-crossing detection. These include features such as intercept, slope, correlation, and RMSE (Root Mean Square Error), which are calculated for specific window sizes. 3. Model training module: Uses machine learning algorithms, in particular deep learning models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Generative Adversarial Networks (GANs). The DNN is trained on a dataset that includes signals with varying levels of noise and harmonic distortion. 4. Zero-crossing detection module: Implements the trained DNN to detect zero-crossing points in real time. This module is optimized through hyperparameter tuning, which includes adjusting the number of hidden layers, neurons, and activation functions to improve accuracy. Data set:

[0012] A dataset is created using sinusoidal signals with varying noise levels (10%–50%) and harmonic distortion (THD from 10% to 50%). The dataset contains features extracted from the signals, such as intercept, slope, correlation, and RMSE, and is used to train the DNN. data set Noise Total harmonic distortion Samples 1 10% - 50% - 4936 2 - 10% - 50% 4436 3 10% - 40% 50% 3949 4 5% - 20% - 3949 How the system works: 1. Signal acquisition: The signal acquisition unit captures the distorted sine signal in real time and forwards it to the data processing unit. 2. Preprocessing: The signal is filtered and segmented by the data preprocessing module. Features such as slope, intercept, and correlation are extracted. 3. Zero-crossing detection: The processed signal is fed into the DNN-based zero-crossing detection module, where the trained model identifies the zero-crossing points. 4. Output: The detected zero crossings are forwarded to the user interface and to all connected devices (e.g., network synchronization or protection systems). The system can also log the data for later analysis. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] US 6,285,140B1

[0002] CN 101871965B

[0002]

Claims

[1] A zero-crossing point detection system for distorted sine signals, comprising: a signal acquisition unit for capturing real-time signals, a data preprocessing module for filtering and segmenting signals, a module for zero-crossing detection based on a deep neural network and a user interface for displaying detected zero crossing points. [2] System according to claim 1, wherein the recognition module based on a deep neural network is trained using signals with different noise and harmonic distortion levels. [3] System according to claim 1, wherein the data preprocessing module extracts features such as intercept, slope, correlation and RMSE from the signal. [4] System according to claim 1, wherein the recognition module based on a deep neural network uses hyperparameter tuning to improve recognition accuracy.