Complex sound source noise characteristic identification method for transformer substation

Abstract

The invention discloses a complex sound source noise characteristic identification method for a transformer substation, and the method comprises the following steps: S1, arranging an acoustic sensor in the transformer substation, carrying out the noise measurement of a key sound source through the acoustic sensor, and collecting the noise data in the transformer substation, including noise signals in different time periods and under different working conditions; according to the method, the time domain, frequency domain and spatial features of the noise signals are comprehensively extracted, so that multi-feature fusion and multi-dimensional noise feature recognition are realized, and the recognition accuracy of noise features of each sound source in the transformer substation is improved; a deep learning model is used for training a data set synthesized by noise signals and corresponding feature tags, so that the model can accurately identify noise characteristics of different sound sources, the accuracy of noise identification is improved, and the noise of a transformer substation can be monitored and identified in real time. And timely and effective information support is provided for noise treatment of a complex sound source of the transformer substation.

Description

Technical Field

[0001] The present application relates to the technical field of sound source characteristic identification, and in particular to a method for identifying complex sound source noise characteristics for a substation. Background Art

[0002] With the rapid economic development and accelerated urbanization, the construction of substations is gradually approaching densely populated areas. The noise problem of old substations has become increasingly prominent and has become an important factor affecting the urban environment. Substation noise mainly comes from vibration and electromagnetic radiation of equipment operation, cooling system and transformer and other components. It is complex and diverse. Traditional noise control methods often cannot take into account multiple needs such as noise reduction efficiency, heat dissipation, fire protection and maintenance.

[0003] The prior art discloses a method for real-time prediction and identification of long-distance radiated noise of a substation based on multi-point virtual sound sources, which belongs to the field of measuring sound waves. The present invention can find the most accurate noise source and the arrangement of the sound source monitoring points, and can remove the loudspeaker and only keep the sound source monitoring points, which is used for 24-hour real-time monitoring of the noise spectrum signal under the actual working conditions of the substation.

[0004] When the above-mentioned prior art identifies the noise of the substation, it is difficult to accurately identify the noise characteristics of each sound source in the substation due to the complexity of the sound source noise of the substation, resulting in low accuracy of actual noise identification and difficulty in real-time monitoring of the substation noise. Therefore, it is particularly important to develop a method that can accurately identify the noise characteristics of complex sound sources in the substation. Summary of the invention

[0005] To this end, the present application provides a method for identifying noise characteristics of complex sound sources in substations to solve the problems in the prior art that it is difficult to accurately identify the noise characteristics of each sound source in the substation, the accuracy of actual noise identification is low, and it is difficult to monitor the substation noise in real time.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] A method for identifying noise characteristics of complex sound sources in a substation, the method comprising the following steps:

[0008] S1 arranges acoustic sensors in the substation, measures the noise of key sound sources through acoustic sensors, and collects noise data in the substation, including noise signals in different time periods and under different working conditions;

[0009] S2 filters the collected noise signals to remove background noise and interference signals, and processes the continuous noise signals in segments, marking the sound source and working condition information corresponding to each segment of the signal;

[0010] S3 extracts the time domain features, frequency domain features and spatial features of each noise signal;

[0011] S4 selects a deep learning model according to the specific needs of noise characteristic recognition, integrates the preprocessed noise signal and its corresponding features into a data set, inputs the data set into the deep learning model for training, and adjusts the model parameters so that the model can accurately identify the noise characteristics of different sound sources;

[0012] S5 deploys the trained deep learning model to the noise monitoring system of the substation to monitor and identify the noise of the substation in real time.

[0013] Optionally, in step S1, a plurality of acoustic sensors are arranged to form a sensor array, and the sensor array comprehensively covers a target area in the substation, and noise measurements are performed on key sound sources in the target area, such as transformers, switchgear, cooling systems, etc.

[0014] Optionally, in step S2, the continuous noise signal is segmented based on a variety of segmentation methods, including segmentation based on time length, segmentation based on sound level change, segmentation based on events, and segmentation based on spectrum features.

[0015] Optionally, in step S3,

[0016] The extracted time domain features include the peak, mean, variance, skewness and kurtosis characteristics of the signal;

[0017] The extracted frequency domain features include the signal’s spectrum and power spectrum density features;

[0018] The extracted time-space features include the signal's sound source localization and sound pressure level distribution characteristics.

[0019] Optionally, in step S4, the deep learning model has an input layer, a hidden layer and an output layer;

[0020] Input layer: used to receive pre-processed sound signals, which include time domain waveforms, frequency domain spectrograms or feature vectors;

[0021] Hidden layer: including convolution layer, pooling layer, and fully connected layer, which are used for feature extraction and classification;

[0022] Output layer: includes classifiers for identifying different types of noise and regressors for identifying noise characteristics.

[0023] Optionally, in step S4, the model training steps are as follows:

[0024] (1) Divide the data set into training set, validation set and test set. The training set is used to train the model, the validation set is used to evaluate the model performance during the training process, and the test set is used to finally evaluate the generalization ability of the model. At the same time, the data needs to be preprocessed, such as data enhancement and normalization.

[0025] (2) The input data is forward-calculated through the model. During the forward-calculation process, the model calculates the output of each neuron layer by layer and passes the result to the next layer.

[0026] (3) Use the loss function to calculate the error between the model output and the target value. This error value will be used for subsequent gradient calculation and parameter update.

[0027] (4) Based on the gradient of the loss function, the gradient of each neuron is calculated back-propagated layer by layer starting from the output layer, and the gradient is passed to the previous layer. During the back-propagation process, the chain rule is used to calculate the gradient of each parameter;

[0028] (5) Based on the calculated gradient, use the optimizer to update the model parameters. The optimizer adjusts the parameter values ​​according to the size and direction of the gradient to reduce the value of the loss function.

[0029] (6) Regularly evaluate the performance of the model on the validation set. If the performance on the validation set no longer improves or overfitting occurs, adjust the training parameters or optimizer settings to improve the performance of the model.

[0030] Optionally, in step S5, a corresponding noise threshold is set according to the noise standard and safety requirements of the substation, and when the monitored noise exceeds the preset threshold, the system triggers an early warning or alarm mechanism.

[0031] Optionally, the noise threshold includes a normal noise range, an abnormal noise warning threshold and a severe noise alarm threshold;

[0032] When the monitored noise exceeds the normal noise range and reaches the abnormal noise warning threshold, the warning mechanism is triggered, and the substation noise monitoring system uses sound and light alarms, and sends text messages or emails to remind relevant personnel to take measures;

[0033] When the monitored noise exceeds the abnormal noise warning threshold and reaches the severe noise alarm threshold, the alarm mechanism is triggered and the substation noise monitoring system starts the emergency response process, such as automatically cutting off the power supply of related equipment.

[0034] Compared with the prior art, this application has at least the following beneficial effects:

[0035] (1) By comprehensively extracting the time domain, frequency domain and spatial characteristics of noise signals, multi-feature fusion and multi-dimensional noise characteristic identification are realized, which is conducive to improving the identification accuracy of noise characteristics of various sound sources in the substation;

[0036] (ii) training the entire synthetic data set of noise signals and their corresponding feature labels through a deep learning model, so that the model can accurately identify the noise characteristics of different sound sources, thereby improving the accuracy of noise identification;

[0037] (3) Deploy the trained deep learning model to the noise monitoring system of the substation to monitor and identify the noise of the substation in real time, and provide timely and effective information support for the noise control of complex sound sources in the substation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more intuitively illustrate the prior art and the present application, exemplary drawings are given below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing the present application; for example, those skilled in the art are capable of easily making conventional adjustments or further optimizations to the addition / reduction / attribution division, specific shapes, positional relationships, connection methods, dimensional ratios, etc. of certain units (components) based on the technical concepts and exemplary drawings disclosed in the present application.

[0039] Figure 1 A schematic flow chart of a method for identifying noise characteristics of complex sound sources in a substation provided in the present application. DETAILED DESCRIPTION

[0040] The present application is further described below in detail through specific embodiments in conjunction with the accompanying drawings.

[0041] In one embodiment, Figure 1 As shown, a method for identifying noise characteristics of complex sound sources for a substation is provided, the method comprising the following steps:

[0042] S1 arranges acoustic sensors in the substation, measures the noise of key sound sources through acoustic sensors, and collects noise data in the substation, including noise signals in different time periods and under different working conditions;

[0043] S2 filters the collected noise signals to remove background noise and interference signals, and processes the continuous noise signals in segments, marking the sound source and working condition information corresponding to each segment of the signal;

[0044] S3 extracts the time domain features, frequency domain features and spatial features of each noise signal;

[0045] S4 selects a deep learning model according to the specific needs of noise characteristic recognition, integrates the preprocessed noise signal and its corresponding features into a data set, inputs the data set into the deep learning model for training, and adjusts the model parameters so that the model can accurately identify the noise characteristics of different sound sources;

[0046] S5 deploys the trained deep learning model to the noise monitoring system of the substation to monitor and identify the noise of the substation in real time.

[0047] Among them, in step S1, multiple acoustic sensors (such as microphones) are arranged to form a sensor array, and the sensor array comprehensively covers the target area in the substation, and noise measurements are performed on key sound sources in the target area such as transformers, switchgear, cooling systems, etc.

[0048] Wherein, in step S2, the continuous noise signal is segmented based on a plurality of segmentation methods, including segmentation based on time length, segmentation based on sound level change, segmentation based on events, and segmentation based on spectrum features;

[0049] After the continuous noise signal is segmented, it is easier to conduct detailed analysis on each segment of the signal, which helps to identify key information such as the source, type and intensity of the noise;

[0050] Segment processing can make signal processing more modular, and each segment of the signal can be processed and analyzed independently, thereby improving the overall processing efficiency;

[0051] The noise signal may change with the change of working conditions. Through segmented processing, these changes can be better captured, and corresponding processing measures can be taken for the noise signals under different working conditions;

[0052] The following is a detailed description of the above four segmentation methods:

[0053] Segmentation based on time length: Cut the continuous noise signal into several segments according to a fixed time length. For example, a one-hour noise signal can be cut into multiple 10-minute or 5-minute segments.

[0054] Segmentation based on sound level changes: Segmentation is performed based on the sound level changes of the noise signal. When the sound level exceeds a preset threshold, it can be considered as the beginning of a new segment.

[0055] Event-based segmentation: In some cases, the noise signal may be caused by a specific event. For example, during the operation of a machine, different operations or failures may cause changes in the noise signal. In this case, the segments can be divided according to these events.

[0056] Segmentation based on spectrum characteristics: The spectrum characteristics of the noise signal are used as the basis for segmentation. By analyzing the changes in the spectrum characteristics, the source or type of the noise signal is determined and segmented accordingly;

[0057] During the segment processing, the sound source and working condition information corresponding to each signal segment also need to be marked. This information will help in the subsequent analysis and identification of the noise signal, for example:

[0058] There is a machine in the target area of ​​the substation that generates noise during operation. In order to identify the noise characteristics of this machine, we can perform the following segmentation and marking processing:

[0059] a) Using a time-length-based segmentation method, the continuous noise signal is cut into segments of 1 second in length to form multiple signal segments of equal length;

[0060] b) Assume that we have obtained 10 signal segments, labeled as S1, S2, ..., S10.

[0061] c) Through on-site investigation and equipment inventory, we determined that these signal segments were all generated by the same machine, so the sound source information can be uniformly marked as "machine A";

[0062] d) Obtain the operating condition information of the machine during operation through sensor monitoring and equipment monitoring system, for example, at the first second (signal segment S1), the machine speed is 1500 rpm and the load is 80%, at the second second (signal segment S2), the machine speed is 1600 rpm and the load is 85%, and so on;

[0063] In summary, through segmentation and labeling, we can decompose continuous noise signals into units that are easier to analyze and process, and associate these units with specific sound sources and operating conditions. This information provides an important basis for subsequent analysis and identification of noise signals.

[0064] In step S3, the noise signal is subjected to Fourier transform, and the extracted time domain features include the signal

[0065] Peak value: The maximum value of a signal, used to indicate the strongest part of the signal;

[0066] Mean: The average value of all sampled values ​​of a signal, reflecting the overall trend of the signal;

[0067] Variance measures the dispersion of signal values ​​and reflects the volatility of the signal;

[0068] Skewness: A statistic that describes the distribution shape of a signal and reflects the symmetry of the signal distribution;

[0069] Kurtosis: A statistic that describes the sharpness of the signal distribution pattern and reflects the concentration of the signal distribution;

[0070] In addition, the time domain features of the signal, such as short-time energy and short-time average zero-crossing rate, can also be extracted. These features can reflect the changing laws and characteristics of the signal in the time domain.

[0071] The extracted frequency domain features include the signal

[0072] Spectrum: reflects the energy distribution and phase information of the signal at different frequencies;

[0073] Power spectral density: reflects the power distribution characteristics of the signal in the frequency domain;

[0074] The extracted spatiotemporal features include the signal

[0075] Sound source localization: Determine the location of the sound source by using the spatial distribution of the microphone array and the time difference or phase difference of the signal reaching different microphones;

[0076] Sound pressure level distribution: describes the sound pressure level distribution of the signal in space, which can reflect the radiation and propagation characteristics of the sound source in space.

[0077] Wherein, in step S4, the deep learning model has an input layer, a hidden layer and an output layer;

[0078] Input layer: used to receive pre-processed sound signals, which include time domain waveforms, frequency domain spectrograms or feature vectors;

[0079] Hidden layer: including convolution layer, pooling layer, and fully connected layer, which are used for feature extraction and classification;

[0080] The convolution layer extracts local features from the input data through convolution operations. The convolution kernel is designed to capture specific frequency components or time structures in the sound signal. Through multiple layers of convolution, higher-level features are gradually extracted.

[0081] The pooling layer is used to reduce the spatial dimension of the data, which helps to reduce redundant information in the data and improve the robustness of the model;

[0082] The fully connected layer is located at the end of the hidden layer. It combines and classifies the features extracted by the previous layer. In the fully connected layer, each neuron is connected to all neurons in the previous layer. New feature representations are output through weighted summation and activation function operations. These feature representations are ultimately used for classification or regression tasks.

[0083] Output layer: includes classifiers for identifying different types of noise and regressors for identifying noise characteristics;

[0084] If the task is to identify different types of noise (such as mechanical noise, traffic noise, environmental noise, etc.), the output layer can be designed as a classifier, and softmax classifiers and support vector machine (SVM) classifiers can be selected. Preferably, softmax classifier can output the probability distribution of each category, so as to achieve accurate identification of noise types;

[0085] If the task is to predict a certain characteristic of noise (such as noise intensity, frequency distribution, etc.), the output layer is designed as a regressor. A linear regressor and a neural network regressor can be selected, preferably a linear regressor, which can output continuous numerical prediction results, thereby achieving accurate quantification of noise characteristics.

[0086] Among them, in step S4, the model training steps are as follows:

[0087] (1) Divide the data set into training set, validation set and test set. The training set is used to train the model, the validation set is used to evaluate the model performance during the training process, and the test set is used to finally evaluate the generalization ability of the model. At the same time, the data needs to be preprocessed, such as data enhancement and normalization.

[0088] (2) The input data is forward-calculated through the model. During the forward-calculation process, the model calculates the output of each neuron layer by layer and passes the result to the next layer.

[0089] (3) Use the loss function to calculate the error between the model output and the target value. This error value will be used for subsequent gradient calculation and parameter update.

[0090] (4) Based on the gradient of the loss function, the gradient of each neuron is calculated back-propagated layer by layer starting from the output layer, and the gradient is passed to the previous layer. During the back-propagation process, the chain rule is used to calculate the gradient of each parameter;

[0091] (5) Based on the calculated gradient, use the optimizer to update the model parameters. The optimizer will adjust the parameter values ​​according to the size and direction of the gradient to reduce the value of the loss function.

[0092] (6) Regularly evaluate the performance of the model on the validation set. If the performance on the validation set no longer improves or overfitting occurs, you need to adjust the training parameters or optimizer settings to improve the performance of the model.

[0093] During the training process, the best hyperparameter combination can be found through grid search, random search, or Bayesian optimization, which facilitates the tuning of model parameters. By continuously optimizing and improving the training process, a high-performance deep learning model can be built.

[0094] In step S5, a corresponding noise threshold is set according to the noise standard and safety requirements of the substation. When the noise exceeds the preset threshold, the system triggers an early warning or alarm mechanism;

[0095] Further, the noise threshold includes a normal noise range, an abnormal noise warning threshold, and a severe noise alarm threshold;

[0096] Normal noise range: set to a relatively low value to represent the noise level generated by the equipment in the substation during normal operation;

[0097] Abnormal noise warning threshold: When the noise exceeds the normal noise range and reaches this threshold, it indicates that the equipment may be operating abnormally or malfunctioning, which requires attention;

[0098] Severe noise alarm threshold: When the noise exceeds the abnormal noise warning threshold and reaches this threshold, it indicates that the equipment may have a serious fault or accident and immediate measures need to be taken;

[0099] When the monitored noise exceeds the normal noise range and reaches the abnormal noise warning threshold, the warning mechanism is triggered, and the substation noise monitoring system uses sound and light alarms, and sends text messages or emails to remind relevant personnel to take measures;

[0100] When the monitored noise exceeds the abnormal noise warning threshold and reaches the severe noise alarm threshold, the alarm mechanism is triggered and the substation noise monitoring system starts the emergency response process, such as automatically cutting off the power supply of related equipment.

[0101] The technical features of the above embodiments may be arbitrarily combined (as long as there is no contradiction in the combination of these technical features). To make the description concise, not all possible combinations of the technical features in the above embodiments are described; these embodiments that are not explicitly written should also be considered to be within the scope of this specification.

Claims

1. A method for identifying noise characteristics of complex sound sources in a substation, characterized in that: The method comprises: S1 arranges acoustic sensors in the substation, measures the noise of key sound sources through the acoustic sensors, and collects noise data in the substation; S2 filters the collected noise signals to remove background noise and interference signals, and processes the continuous noise signals in segments, marking the sound source and working condition information corresponding to each segment of the signal; S3 extracts the time domain features, frequency domain features and spatial features of each noise signal; S4 selects a deep learning model based on the specific needs of noise characteristic recognition, integrates the preprocessed noise signal and its corresponding features into a data set, and inputs the data set into the deep learning model for training, so that the model can accurately identify the noise characteristics of different sound sources; S5 deploys the trained deep learning model to the noise monitoring system of the substation to monitor and identify the noise of the substation in real time.

2. A method for identifying complex sound source noise characteristics for a substation according to claim 1, characterized in that: In step S1, a plurality of acoustic sensors are arranged to form a sensor array, and the sensor array is used to fully cover a target area in the substation, and noise measurement is performed on key sound sources in the target area.

3. A method for identifying complex sound source noise characteristics for a substation according to claim 1, characterized in that: In step S2, the continuous noise signal is segmented based on a variety of segmentation methods, including segmentation based on time length, segmentation based on sound level change, segmentation based on events, and segmentation based on spectrum features.

4. A method for identifying complex sound source noise characteristics for a substation according to claim 1, characterized in that: In step S3, The extracted time domain features include the peak, mean, variance, skewness and kurtosis characteristics of the signal; The extracted frequency domain features include the signal’s spectrum and power spectrum density features; The extracted time-space features include the signal's sound source localization and sound pressure level distribution characteristics.

5. The method for identifying complex sound source noise characteristics for a substation according to claim 1, characterized in that: In step S4, the deep learning model has an input layer, a hidden layer and an output layer; Input layer: used to receive pre-processed sound signals, which include time domain waveforms, frequency domain spectrograms or feature vectors; Hidden layer: including convolution layer, pooling layer, and fully connected layer, which are used for feature extraction and classification; Output layer: includes classifiers for identifying different types of noise and regressors for identifying noise characteristics.

6. A method for identifying complex sound source noise characteristics for a substation according to claim 1, characterized in that: In step S4, the model training steps are as follows: (1) Divide the dataset into training set, validation set and test set. The training set is used to train the model, the validation set is used to evaluate the model performance during the training process, and the test set is used to finally evaluate the generalization ability of the model. (2) The input data is forward-calculated through the model. During the forward-calculation process, the model calculates the output of each neuron layer by layer and passes the result to the next layer. (3) Use the loss function to calculate the error between the model output and the target value; (4) Based on the gradient of the loss function, reversely calculate the gradient of each neuron layer by layer starting from the output layer, and pass the gradient to the previous layer; (5) Use the optimizer to update the model parameters based on the calculated gradient; (6) Regularly evaluate the performance of the model on the validation set. If the performance on the validation set no longer improves or overfitting occurs, adjust the training parameters or optimizer settings to improve the performance of the model.

7. A method for identifying complex sound source noise characteristics for a substation according to claim 1, characterized in that: In step S5, a corresponding noise threshold is set according to the noise standard and safety requirements of the substation. When the monitored noise exceeds the preset threshold, the system triggers an early warning or alarm mechanism.

8. A method for identifying complex sound source noise characteristics for a substation according to claim 7, characterized in that: The noise threshold includes a normal noise range, an abnormal noise warning threshold and a severe noise alarm threshold; When the monitored noise exceeds the normal noise range and reaches the abnormal noise warning threshold, the warning mechanism is triggered, and the substation noise monitoring system uses sound and light alarms, and sends text messages or emails to remind relevant personnel to take measures; When the monitored noise exceeds the abnormal noise warning threshold and reaches the severe noise alarm threshold, the alarm mechanism is triggered and the substation noise monitoring system starts the emergency response process.

Images