Noise model construction method based on machine learning method

Abstract

The invention discloses a noise model construction method based on a machine learning method, and the method comprises the steps: S1, setting a noise sampling point of an old transformer substation, monitoring and collecting the noise value of the sampling point and the operation data of the old transformer substation, and carrying out the preprocessing of the obtained noise value and operation data, the operation data comprises current and voltage values and noise duration T; and S2, carrying out transformation processing on the noise value and the operation data which are preprocessed in the S1. The invention belongs to the technical field of noise model construction, and aims to solve the problem that accurate and effective analysis and prediction cannot be provided for noise of an old transformer substation due to the fact that it is difficult to accurately capture features of noise changing along with time in the prior art. The method has the technical effects that useful information is learned and extracted from a large amount of operation data and noise values of the old transformer substation by utilizing a machine learning algorithm, so that accurate and effective analysis and prediction are provided for the noise.

Description

Technical Field

[0001] The present application relates to the technical field of noise model construction, and in particular to a noise model construction method based on a machine learning method. Background Art

[0002] In recent years, my country's economy has continued to develop rapidly, the industrialization process has continued to advance, the living standards of urban residents have significantly improved, and the penetration rate of various electrical appliances has increased significantly, resulting in a sharp increase in electricity demand. Especially in summer and winter, the use of high-power electrical appliances such as air conditioners and heaters has further aggravated the contradiction between electricity supply and demand. The demand for electricity in cities has increased significantly, resulting in some substations having to be built in densely populated areas. At the same time, with the accelerated advancement of urbanization, a large amount of farmland and vacant land have been developed into residential areas or industrial parks. Substations originally located in the suburbs and far away from residential areas are gradually surrounded by new residential areas.

[0003] As the service time of substations increases, transformers and other key components in some old substations have shown varying degrees of wear and aging. This not only reduces the efficiency of the equipment, but also increases the mechanical vibration and aerodynamic noise generated during operation. Studies have shown that people who are in a noise pollution environment for a long time are more likely to suffer from insomnia or other related diseases, especially children and the elderly, who are more sensitive to sound and more easily disturbed. Therefore, it is necessary to carry out sound insulation and noise reduction treatment on old substations, so that effective noise control technology can significantly reduce the level of noise pollution in the surrounding areas of substations and improve the quality of life and health of local residents; before carrying out sound insulation and noise reduction treatment on old substations, it is necessary to effectively model and predict the noise of old substations, so as to facilitate the subsequent formulation of detailed sound insulation and noise reduction treatment plans.

[0004] At present, traditional noise modeling methods for old substations mostly rely on physical models or statistical models for construction. These methods often have limitations when dealing with complex and changeable noise environments. It is difficult to accurately capture the characteristics of noise changing over time, resulting in the inability to provide accurate and effective analysis and prediction of noise in old substations. It cannot meet the high-precision requirements of modern substations for noise management and is not conducive to laying the foundation for subsequent noise control. Summary of the invention

[0005] To this end, the present application provides a noise model construction method based on machine learning methods to solve the existing problem that it is difficult to accurately capture the characteristics of noise changes over time, resulting in the inability to provide accurate and effective analysis and prediction of the noise of old substations.

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

[0007] A method for constructing a noise model based on a machine learning method, a method for constructing a noise model based on a machine learning method, the method comprising:

[0008] S1: Set noise sampling points of old substations, monitor and collect noise values ​​at sampling points and operation data of old substations, and pre-process the acquired noise values ​​and operation data, wherein the operation data includes current and voltage values ​​and noise duration T;

[0009] S2: transform the preprocessed noise value and operation data in S1, convert the signal from the time domain to the spectrum distribution in the frequency domain, extract the statistical characteristics of the noise, and obtain the noise data group 1;

[0010] S3: Calculate the noise energy E corresponding to each sampling point based on a machine learning algorithm, and obtain a noise data group 2 consisting of the noise energy E of each sampling point in different time periods;

[0011] S4: Build a machine learning model based on the machine learning algorithm, make a sequence curve graph of data set 2 and time series, and then insert data set 1 of the same sampling point into the machine learning model;

[0012] S5: Optimizing the data set 1 and the sequence curve graph respectively to obtain a training set;

[0013] S6: Integrate the framework group into the machine learning model, determine the final machine learning model, and apply the machine learning model to noise prediction and identification of old substations.

[0014] Furthermore, the preprocessing refers to using the IQR rule to identify and remove those noise values ​​and operating data that are obviously deviated from the normal range, and integrating the noise data and substation operating data from different sampling points to facilitate subsequent analysis; after integration, the noise values ​​and operating data are mean-normalized to eliminate the dimensional differences between different features; and through data segmentation processing, the noise values ​​and operating data are divided into different time series.

[0015] Furthermore, the transformation processing in S2 refers to decomposing the collected noise signal into a superposition of sine waves and cosine waves of different frequencies, converting the noise signal from the time domain to the frequency domain, so as to better analyze and process the signal; it is not limited to wavelet transform, but can also be Fourier transform.

[0016] Furthermore, the calculation formula of the noise energy E is:

[0017]

[0018] Where P is the noise signal power, T is the noise duration, t1 is the start time of the noise, and t2 is the end time of the noise.

[0019] Furthermore, the calculation formula of the noise signal power P is:

[0020] P=V(T)×I(T)

[0021] Wherein, V(T) is the total voltage value during the noise duration, and I(T) is the total current value during the noise duration.

[0022] Furthermore, the sequence curve diagram enables each noise energy E in the data group 2 to have a corresponding time stamp, and accurate synchronization of data can be achieved by taking the time series as a common coefficient.

[0023] Furthermore, the specific steps of the optimization process are as follows:

[0024] Step 1: Smooth the sequence curve chart to reduce the impact of random fluctuations and make the trend more obvious;

[0025] Step 2: Integrate data set 1 with the data of the sequence curve graph to form a unified data set for subsequent analysis and processing;

[0026] Step 3: Identify the noise characteristics in data set 1 and the sequence curve graph respectively through machine learning algorithms;

[0027] Step 4: Perform a joint analysis on the data set 1 and the sequence curve chart based on the spectrum analysis method, extract the characteristics of each noise, and obtain a framework group composed of various noise characteristics and noise properties.

[0028] Furthermore, the time range of the same sampling point data group 1 in S4 is consistent with the time series of the sequence curve graph.

[0029] Compared with the prior art, this application has at least the following beneficial effects:

[0030] The model building method that combines machine learning algorithms with machine learning models aims to use machine learning algorithms to learn and extract useful information from a large number of operating data and noise values ​​of old substations, deeply analyze and process the characteristics of complex noise signals generated by old substations that change over time, and extract characteristic frequencies to capture the characteristics of noise, so as to provide accurate and effective analysis and prediction of noise, facilitate more accurate identification and classification of different types of noise in old substations, and lay the foundation for subsequent noise control. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] 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.

[0032] Figure 1 A flowchart of a noise model construction method based on a machine learning method is provided for one embodiment of the present application. DETAILED DESCRIPTION

[0033] The present application is further described below in detail through specific embodiments in conjunction with the accompanying drawings.

[0034] like Figure 1 As shown, a noise model construction method based on a machine learning method in the first aspect of the present application is specifically implemented as follows:

[0035] S1: Set noise sampling points of old substations, monitor and collect noise values ​​at sampling points and operation data of old substations, and pre-process the acquired noise values ​​and operation data; the operation data of old substations include current and voltage values ​​and noise duration T;

[0036] The preprocessing uses the IQR rule to identify and remove noise values ​​and operating data that are obviously deviated from the normal range. At the same time, the noise data and substation operating data from different sampling points are integrated to facilitate subsequent analysis. After integration, the noise values ​​and operating data are mean-normalized to eliminate the dimensional differences between different features. And through data segmentation, the noise values ​​and operating data are divided into different time series to better capture the patterns of changes in noise values ​​and operating data over time.

[0037] S2: transform the preprocessed noise value and operation data in S1, convert the signal from the time domain to the spectrum distribution in the frequency domain, extract the statistical characteristics of the noise, and obtain the noise data group 1;

[0038] Among them, the transformation processing in S2 refers to decomposing the collected noise signal into a superposition of sine waves and cosine waves of different frequencies, converting the noise signal from the time domain to the frequency domain, so as to better analyze and process the signal; it is not limited to wavelet transform, but can also be Fourier transform; spectral analysis of the noise signal is performed to identify and display the various frequency components contained in the signal. This analysis allows us to understand the intensity distribution of the noise signal at different frequencies, and can effectively capture the changes in different frequency components, which is very important for identifying and distinguishing different types of noise sources. In addition to spectral information, transformation processing can also provide time-frequency localization information, that is, the changes in frequency components within a specific time period, which helps to understand the dynamic change characteristics of noise over time and can reflect the statistical characteristics of noise;

[0039] S3: Calculate the noise energy E corresponding to each sampling point based on the machine learning algorithm, and obtain a noise data set 2 consisting of the noise energy E of each sampling point in different time periods; the data set 2 is used as the basis for training the machine learning model; wherein the machine learning algorithm in S3 can be a convolutional neural network or a recurrent neural network;

[0040] The calculation formula of noise energy E is:

[0041]

[0042] Where P is the noise signal power, T is the noise duration, t1 is the start time of the noise, and t2 is the end time of the noise.

[0043] The calculation formula of noise signal power P is:

[0044] P=V(T)×I(T)

[0045] Wherein, V(T) is the total voltage value during the noise duration, and I(T) is the total current value during the noise duration.

[0046] S4: Build a machine learning model based on the machine learning algorithm, make a sequence curve graph of data group 2 and the time series, and then insert data group 1 of the same sampling point into the machine learning model to ensure that the time range of data group 1 is consistent with the time series of the sequence curve graph;

[0047] The sequence curve chart ensures that each noise energy E in data group 2 has a corresponding timestamp. With the time series as the common coefficient, accurate synchronization of data can be achieved, which is convenient for subsequent processing and analysis. Data group 2 can describe the characteristics and distribution patterns of noise at sampling points at different times, which is convenient for machine learning models to understand the inherent patterns of noise.

[0048] S5: Optimize the data set 1 and the sequence curve graph respectively to obtain a training set, so that the machine learning model can learn the characteristics of noise. Data set 1 is used as additional feature auxiliary information to help the model better learn the noise characteristics in the data. The optimization of the dual data of data set 1 and the sequence curve graph enables the machine learning model to learn different noise patterns, which is convenient for identifying complex noise and avoiding overfitting.

[0049] The specific steps of optimization are as follows:

[0050] Step 1: Smooth the sequence curve chart to reduce the impact of random fluctuations and make the trend more obvious;

[0051] Step 2: Integrate data set 1 with the data of the sequence curve graph to form a unified data set for subsequent analysis and processing;

[0052] Step 3: Identify the noise characteristics in data set 1 and the sequence curve graph respectively through machine learning algorithms;

[0053] Step 4: Perform a joint analysis on the data set 1 and the sequence curve chart based on the spectrum analysis method, extract the characteristics of each noise, and obtain a framework group composed of various noise characteristics and noise properties.

[0054] S6: Integrate the framework group into the machine learning model, determine the final machine learning model, and apply the machine learning model to the noise prediction and identification of old substations. The final machine learning model can monitor and collect the current noise values ​​and operating parameters of old substations, accurately predict noise characteristics, and identify different types of noise sources, providing a scientific basis and support for the subsequent development of targeted noise reduction measures.

[0055] In the noise model construction method based on machine learning methods, a model construction method combining machine learning algorithms and machine learning models is adopted, aiming to use machine learning algorithms to learn and extract useful information from a large number of operating data and noise values ​​of old substations, and deeply analyze and process the characteristics of complex noise signals generated by old substations that change over time. By performing detailed characteristic frequency extraction on these noise signals, the characteristics of the noise are captured, and based on these characteristics, a highly accurate noise model is constructed, which can accurately identify and classify different types of noise in old substations, laying the foundation for subsequent noise control.

[0056] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A noise model construction method based on machine learning method, characterized in that: The method comprises: S1: Set noise sampling points of old substations, monitor and collect noise values ​​at sampling points and operation data of old substations, and pre-process the acquired noise values ​​and operation data, wherein the operation data includes current and voltage values ​​and noise duration T; S2: transform the preprocessed noise value and operation data in S1, convert the signal from the time domain to the spectrum distribution in the frequency domain, extract the statistical characteristics of the noise, and obtain the noise data group 1; S3: Calculate the noise energy E corresponding to each sampling point based on a machine learning algorithm, and obtain a noise data group 2 consisting of the noise energy E of each sampling point in different time periods; S4: Build a machine learning model based on the machine learning algorithm, make a sequence curve graph of data set 2 and time series, and then insert data set 1 of the same sampling point into the machine learning model; S5: Optimizing the data set 1 and the sequence curve graph respectively to obtain a training set; S6: Integrate the framework group into the machine learning model, determine the final machine learning model, and apply the machine learning model to noise prediction and identification of old substations.

2. The noise model construction method based on machine learning method according to claim 1, characterized in that: The preprocessing refers to using the IQR rule to identify and remove those noise values ​​and operating data that are obviously deviated from the normal range, and at the same time integrate the noise data and substation operating data from different sampling points to facilitate subsequent analysis; after integration, the noise value and operating data are normalized by the mean to eliminate the dimensional differences between different features; And through data segmentation processing, the noise value and operating data are divided into different time series.

3. The noise model construction method based on machine learning method according to claim 1, characterized in that: The transformation processing in S2 refers to decomposing the collected noise signal into a superposition of sine waves and cosine waves of different frequencies, converting the noise signal from the time domain to the frequency domain, so as to better analyze and process the signal; it is not limited to wavelet transformation, but can also be Fourier transformation.

4. The noise model construction method based on machine learning method according to claim 1, characterized in that: The calculation formula of the noise energy E is: Where P is the noise signal power, T is the noise duration, t1 is the start time of the noise, and t2 is the end time of the noise.

5. The noise model construction method based on machine learning method according to claim 4 is characterized in that: The calculation formula of the noise signal power P is: P=V(T)×I(T) Wherein, V(T) is the total voltage value during the noise duration, and I(T) is the total current value during the noise duration.

6. The noise model construction method based on machine learning method according to claim 1, characterized in that: The sequence curve diagram enables each noise energy E in the data group 2 to have a corresponding time stamp, and accurate synchronization of data can be achieved by taking the time series as a common coefficient.

7. The noise model construction method based on machine learning method according to claim 1, characterized in that: The specific steps of the optimization process are as follows: Step 1: Smooth the sequence curve chart to reduce the impact of random fluctuations and make the trend more obvious; Step 2: Integrate data set 1 with the data of the sequence curve graph to form a unified data set for subsequent analysis and processing; Step 3: Identify the noise characteristics in data set 1 and the sequence curve graph respectively through machine learning algorithms; Step 4: Perform a joint analysis on the data set 1 and the sequence curve chart based on the spectrum analysis method, extract the characteristics of each noise, and obtain a framework group composed of various noise characteristics and noise properties.

8. The noise model construction method based on machine learning method according to claim 1, characterized in that: The time range of the same sampling point data group 1 in S4 is consistent with the time series of the sequence curve graph.

Images