Intelligent noise detection method based on multiple parameters

Abstract

The invention provides an intelligent noise detection method based on multiple parameters. The method comprises the steps that an objective function and constraint conditions of the objective function are constructed, the influence of noise parameters on an objective function value is determined through a gradient descent method, and the noise parameters are adjusted to minimize the objective function value. And establishing a voiceprint library, and collecting, classifying and storing voiceprint features of different noise types. Noise signals are collected in real time, and multi-parameter features, including frequency, amplitude, time domain and frequency domain features, of the noise signals are extracted. And matching the extracted features with a voiceprint library, identifying noise types, and selecting a corresponding monitoring mode. And reducing the dimension of the initial noise parameter through a principal component analysis algorithm, optimizing the feature vector by adopting a genetic algorithm, and iteratively updating the noise parameter until the optimal detection parameter is achieved. According to the invention, different types of noise signals can be accurately identified and monitored, and the precision and efficiency of noise detection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of noise detection, and more specifically, to a noise intelligent detection method based on multiple parameters. Background Art

[0002] Existing noise detection methods mainly rely on single signal features or simple algorithms for noise identification and monitoring. For example, traditional spectrum analysis methods focus on the frequency components of noise signals, while time domain analysis focuses on the amplitude changes of signals. These methods often have limitations when dealing with complex noise environments and have difficulty accurately identifying multiple noise types. In addition, traditional noise detection methods also have deficiencies in real-time and adaptability, and cannot effectively cope with the dynamic changes of noise environments.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: it is impossible to integrate multi-parameter features for noise identification, resulting in low detection accuracy; there is a lack of intelligent optimization algorithms to dynamically adjust detection parameters, resulting in poor real-time and adaptability of noise detection. Summary of the invention

[0004] The present invention provides a noise intelligent detection method based on multiple parameters, comprising: Construct the objective function and constraints of noise detection, The influence of noise parameters on the objective function value is determined by the gradient descent method, including: Calculate the partial derivative of the objective function with respect to each noise parameter to obtain the gradient value of each parameter; Adjust the noise parameter according to the gradient value so that it moves in the direction of gradient descent until the objective function value reaches the minimum; Adjust the noise parameters according to the gradient descent method to obtain multiple groups of initial noise parameters; Establish a voiceprint database, collect voiceprint features of different noise types, and classify and store the voiceprint features; Real-time acquisition and preprocessing of noise signals to extract multi-parameter features of noise signals, including frequency features, amplitude features, time domain features and frequency domain features; Match the extracted multi-parameter features with the voiceprint features in the voiceprint library to identify the type of noise signal; According to the identified noise type, select the corresponding monitoring method to monitor and analyze the noise signal; The following update operations are iteratively performed on the multiple groups of initial noise parameters: the multiple groups of initial noise parameters are reduced to multiple feature vectors using a principal component analysis algorithm; According to the preset optimization operator, a genetic algorithm is used to optimize each feature vector in turn, and then multiple groups of global noise parameters are obtained by merging them; If the total objective function value of the multiple sets of global noise parameters is less than the total objective function value of the multiple sets of initial noise parameters, then the multiple sets of initial noise parameters are updated using the multiple sets of global noise parameters; The next round of updating operation is performed until the stop condition is reached, and the updated multiple sets of initial noise parameters constitute the optimal noise detection parameters.

[0005] Furthermore, the extracted multi-parameter features are matched with the voiceprint features in the voiceprint library to identify the type of noise signal, including: The similarity between the extracted multi-parameter features and each voiceprint feature in the voiceprint library is calculated using Euclidean distance, cosine similarity or dynamic time warping algorithm; According to the preset similarity threshold, the noise type corresponding to the voiceprint feature with the highest similarity is selected as the preliminary recognition result; Conduct confidence assessment on the preliminary recognition results and calculate their confidence values; If the confidence value is lower than the preset confidence threshold, the machine learning model is used for classification and identification to obtain the final noise type identification result.

[0006] Furthermore, the selecting a corresponding monitoring method according to the identified noise type includes: For low-frequency noise, spectrum analysis is used for monitoring. The specific steps are as follows: Perform fast Fourier transform on the noise signal to obtain its spectrum, and analyze the frequency components and energy distribution in the spectrum; For high-frequency noise, the time-frequency analysis method is used for monitoring. The specific steps are as follows: Perform short-time Fourier transform or wavelet transform on the noise signal to obtain its time-frequency diagram, and analyze the time domain characteristics and frequency domain characteristics of the noise signal at the same time; For sudden noise, the event detection method is used for monitoring. The specific steps are as follows: Set the threshold of the noise signal and monitor the amplitude or energy of the noise signal in real time. When it exceeds the threshold, it is judged as a noise event. For continuous noise, statistical analysis method is used for monitoring. The specific steps are as follows: Calculate the statistical characteristics of the noise signal, including mean, variance, standard deviation, and probability distribution, and analyze its overall characteristics and changing patterns.

[0007] Furthermore, the principal component analysis algorithm is used to reduce the dimensions of multiple groups of initial noise parameters into multiple feature vectors, including: Calculate the covariance matrix of each set of initial noise parameters; Perform eigenvalue decomposition on the covariance matrix to obtain eigenvectors and eigenvalues; According to the size of the eigenvalue, select the first few main eigenvectors to form the principal component eigenvector matrix; Each set of initial noise parameters is multiplied by the principal component eigenvector matrix to obtain the eigenvector after dimensionality reduction, which is used as the input for subsequent genetic algorithm optimization.

[0008] Furthermore, the method of optimizing each feature vector in sequence by using a genetic algorithm according to a preset optimization operator includes: The following iterative optimization operation is performed on each feature vector: Initialize the global optimal solution, the optimal solution of the previous iteration, and the weight of the optimization operator, and set the maximum number of executions; In each execution, an optimization operator is selected using the roulette wheel selection algorithm according to the weights of each optimization operator; According to the selected optimization operator, the first solution of each eigenvector is calculated; Apply crossover and mutation operations to the first solution to obtain the second solution; According to the global optimal solution, the optimal solution of the current iteration, the optimal solution of the previous iteration and the second solution, the performance record value of the selected optimization operator, the global optimal solution and the optimal solution of the current iteration are updated. After updating each eigenvector with the global optimal solution, the next execution is performed until the maximum number of executions is reached. For each eigenvector, the optimal solution of the previous iteration is updated using the optimal solution of the current iteration. According to the preset weight update rate and the performance record value of each optimization operator, the weight of each optimization operator is updated, and the next round of iterative optimization is performed until the maximum number of iterations of the eigenvector is reached, and the global optimal solution of each eigenvector is obtained, that is, each eigenvector after optimization.

[0009] Further, the updating of the performance record value of the selected optimization operator, the global optimal solution and the optimal solution of the current round of iteration according to the global optimal solution, the optimal solution of the current round of iteration, the optimal solution of the previous round of iteration and the second solution includes: If the optimal solution of the current iteration is empty, the second solution is used to update the optimal solution of the current iteration; If the objective function value of the second solution is less than the objective function value of the global optimal solution, then according to the value of the first weight update factor, the performance record value of the selected optimization operator is increased, and the global optimal solution is updated using the second solution; If the objective function value of the second solution is less than the objective function value of the optimal solution of the current iteration, then the performance record value of the selected optimization operator is increased according to the value of the second weight update factor, and the second solution is used to update the optimal solution of the current iteration; If the objective function value of the second solution is less than the objective function value of the optimal solution of the previous iteration, the performance record value of the selected optimization operator is increased according to the value of the third weight update factor.

[0010] Furthermore, the weight of each optimization operator is updated by the following formula according to the preset weight update rate and the performance record value of each optimization operator: in, represents the weight of the p-th optimization operator at the t-th iteration, represents the weight of the p-th optimization operator at the t+1-th iteration, represents the performance record value of the p-th optimization operator at the t-th iteration, represents the weight update rate, , Represents the total number of optimization operators.

[0011] Furthermore, the optimization operator includes: The crossover operator is used to exchange gene segments between two feature vectors. The mutation operator is used to randomly select a gene position in the feature vector for mutation. The selection operator is used to select the optimal feature vector for genetic operation according to the fitness function.

[0012] Furthermore, the real-time acquisition and preprocessing of the noise signal includes: Use a microphone array to collect noise signals in real time to obtain the original data of the noise signals; Filter, reduce noise and normalize the collected noise signal to eliminate the interference and noise components in the noise signal and retain the effective information of the noise signal; Feature extraction is performed on the processed noise signal to extract its multi-parameter features, providing basic data for subsequent noise type identification and monitoring analysis.

[0013] Furthermore, the objective function takes the signal-to-noise ratio of the noise signal as the optimization target, and its mathematical form is: in, represents the power of the signal, Represents the power of the noise; constraints include the frequency range of the noise signal , Amplitude range , Noise Type and the duration of the noise signal .

[0014] The above-mentioned embodiments of the present invention have at least the following beneficial effects: by constructing the objective function and its constraints and optimizing the noise parameters by the gradient descent method, the method can effectively determine the influence of the noise parameters on the objective function value, thereby realizing accurate optimization of noise detection. In addition, by establishing a voiceprint library and collecting voiceprint features of different noise types, the method can collect noise signals in real time and extract multi-parameter features, further improving the accuracy of noise type identification. Combining principal component analysis and genetic algorithm to reduce the dimension and optimize the noise parameters can effectively reduce the computational complexity and improve the efficiency and adaptability of noise detection.

[0015] In addition, the method can also select corresponding monitoring methods according to the identified noise types, such as spectrum analysis, time-frequency analysis, event detection and statistical analysis, to adapt to the characteristics and monitoring needs of different noise signals. By iteratively updating the noise parameters until the optimal detection parameters are reached, the method can continuously optimize the noise detection performance, improve the accuracy and reliability of noise monitoring, and is suitable for real-time monitoring and analysis of various complex noise environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, in which: Figure 1 A schematic diagram of a flow chart of a multi-parameter-based intelligent noise detection method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0018] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, device, apparatus, method or computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0019] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0020] Reference below Figure 1 , Figure 1 The following is a flow chart of a noise intelligent detection method based on multiple parameters provided by an embodiment of the present invention. Figure 1 As shown, a noise intelligent detection method 100 based on multiple parameters includes: Step 101, construct the objective function and constraints of noise detection, The influence of noise parameters on the objective function value is determined by the gradient descent method, including: Step 102, calculating the partial derivative of the objective function with respect to each noise parameter to obtain the gradient value of each parameter; Step 103, adjusting the noise parameter according to the gradient value so that it moves along the direction of gradient descent until the objective function value reaches the minimum; Step 104, adjusting the noise parameters according to the gradient descent method to obtain multiple groups of initial noise parameters; Step 105, establishing a voiceprint database, collecting voiceprint features of different noise types, and classifying and storing the voiceprint features; Step 106, real-time acquisition and preprocessing of the noise signal, extracting multi-parameter features of the noise signal, including frequency features, amplitude features, time domain features, and frequency domain features; Step 107, matching the extracted multi-parameter features with the voiceprint features in the voiceprint library to identify the type of the noise signal; Step 108, selecting a corresponding monitoring method according to the identified noise type, and monitoring and analyzing the noise signal; Step 109, iteratively performing the following update operations on the multiple groups of initial noise parameters: using a principal component analysis algorithm to reduce the dimensions of the multiple groups of initial noise parameters into multiple feature vectors; Step 110, using a genetic algorithm according to a preset optimization operator, optimizing each feature vector in turn, and merging to obtain multiple sets of global noise parameters; Step 111, if the total objective function value of the multiple sets of global noise parameters is less than the total objective function value of the multiple sets of initial noise parameters, then use the multiple sets of global noise parameters to update the multiple sets of initial noise parameters; Step 112, executing the next round of updating operation until the stop condition is reached, and the updated multiple groups of initial noise parameters constitute the optimal noise detection parameters.

[0021] It should be noted that the core of this method is to optimize noise detection by constructing an objective function and its constraints. The objective function is a mathematical expression used to quantify the effect of noise detection. The signal-to-noise ratio is usually used as the optimization target, that is, the ratio of signal power to noise power. Constraints are used to limit parameters such as the frequency range, amplitude range, noise type, and duration of the noise signal to ensure the accuracy and effectiveness of the detection process. The gradient descent method is a commonly used optimization algorithm that adjusts the noise parameters by calculating the partial derivative (i.e., gradient value) of the objective function with respect to each noise parameter, so that it moves along the direction of gradient descent until the objective function value reaches the minimum value.

[0022] Specifically, when constructing the objective function, the signal-to-noise ratio of the noise signal can be expressed as in, and Represent the power of the signal and the power of the noise respectively. The constraints can be set according to the actual application scenario. For example, the frequency range can be set to , the amplitude range is , the noise type can include low-frequency noise, high-frequency noise, burst noise and continuous noise, etc., and the duration can be set to In the gradient descent method, the partial derivative of the objective function with respect to each noise parameter is first calculated to obtain the gradient value of each parameter, and then the noise parameter is adjusted according to the gradient value so that it moves in the direction of gradient descent until the objective function value reaches the minimum value.

[0023] Preferably, when adjusting the noise parameters, a step size control strategy can be used to improve the efficiency and accuracy of the optimization. For example, an initial step size can be set, and then the step size can be dynamically adjusted according to the change of the objective function value. When the objective function value decreases rapidly, the step size can be appropriately increased to speed up the convergence speed; when the objective function value decreases slowly or approaches the minimum value, the step size can be appropriately reduced to improve the optimization accuracy.

[0024] Furthermore, a momentum term can be introduced to accelerate the gradient descent process. The momentum term can be expressed as the weighted sum of the gradient values ​​in the previous iterations. By combining the momentum term and the current gradient value to update the noise parameter, it is possible to effectively avoid falling into the local minimum and improve the effect of global optimization.

[0025] In some embodiments, matching the extracted multi-parameter features with the voiceprint features in the voiceprint library to identify the type of the noise signal includes: The similarity between the extracted multi-parameter features and each voiceprint feature in the voiceprint library is calculated using Euclidean distance, cosine similarity or dynamic time warping algorithm; According to the preset similarity threshold, the noise type corresponding to the voiceprint feature with the highest similarity is selected as the preliminary recognition result; Conduct confidence assessment on the preliminary recognition results and calculate their confidence values; If the confidence value is lower than the preset confidence threshold, the machine learning model is used for classification and identification to obtain the final noise type identification result.

[0026] It should be noted that this method identifies the type of noise signal by matching the extracted multi-parameter features with the voiceprint features in the voiceprint library. The voiceprint library is a database that stores voiceprint features of different noise types. Voiceprint features refer to feature information extracted from noise signals and are used to characterize the characteristics of noise. Euclidean distance, cosine similarity, and dynamic time warping algorithms are commonly used similarity calculation methods to measure the similarity between the extracted features and the features in the voiceprint library. By calculating the similarity and selecting the noise type corresponding to the voiceprint feature with the highest similarity as the preliminary identification result, the preliminary classification of the noise signal can be achieved.

[0027] Specifically, Euclidean distance is a method to measure the distance between two feature vectors, and the calculation formula is in, and Represents the first elements. Cosine similarity is used to measure the directional similarity of two feature vectors. The calculation formula is Dynamic time warping algorithm is a method for comparing the similarity of two time series, which is particularly suitable for time domain feature matching of noise signals. In practical applications, we can select a suitable similarity calculation method according to the characteristics of the noise signal and set the corresponding similarity threshold to improve the accuracy of recognition.

[0028] Preferably, when performing noise type identification, a machine learning model can be further introduced for classification and identification. For example, a classification model such as a support vector machine (SVM), a random forest or a neural network can be used to perform confidence assessment and classification identification on the preliminary identification results. First, the confidence value of the preliminary identification result is calculated based on the similarity value of the preliminary identification result. If the confidence value is lower than the preset confidence threshold, the feature is input into the machine learning model for further classification and identification to obtain the final noise type identification result. In addition, the voiceprint library can be regularly updated and expanded to include more types and characteristics of noise signals, thereby improving the accuracy and robustness of identification.

[0029] In some embodiments, selecting a corresponding monitoring method according to the identified noise type includes: For low-frequency noise, spectrum analysis is used for monitoring. The specific steps are as follows: Perform fast Fourier transform on the noise signal to obtain its spectrum, and analyze the frequency components and energy distribution in the spectrum; For high-frequency noise, the time-frequency analysis method is used for monitoring. The specific steps are as follows: Perform short-time Fourier transform or wavelet transform on the noise signal to obtain its time-frequency diagram, and analyze the time domain characteristics and frequency domain characteristics of the noise signal at the same time; For sudden noise, the event detection method is used for monitoring. The specific steps are as follows: Set the threshold of the noise signal and monitor the amplitude or energy of the noise signal in real time. When it exceeds the threshold, it is judged as a noise event. For continuous noise, statistical analysis method is used for monitoring. The specific steps are as follows: Calculate the statistical characteristics of the noise signal, including mean, variance, standard deviation, and probability distribution, and analyze its overall characteristics and changing patterns.

[0030] It should be noted that this method selects the corresponding monitoring method according to the identified noise type to adapt to the characteristics and monitoring needs of different noise signals. Spectral analysis, time-frequency analysis, event detection and statistical analysis are commonly used noise monitoring methods, which are respectively applicable to different types of noise signals. Spectral analysis is mainly used to analyze the frequency components and energy distribution of noise signals, time-frequency analysis is suitable for simultaneously analyzing the time domain and frequency domain characteristics of noise signals, event detection is used to monitor the occurrence of noise events in real time, and statistical analysis is used to analyze the overall characteristics and change laws of noise signals by calculating their statistical characteristics.

[0031] Specifically, for low-frequency noise, the spectrum analysis method can obtain its spectrum diagram by performing a fast Fourier transform (FFT) on the noise signal, and analyze the frequency components and energy distribution in the spectrum diagram. The frequency range of low-frequency noise is usually low, for example, below 100 Hz, and detailed spectrum information can be obtained by setting the sampling frequency and window length of the FFT. For high-frequency noise, the time-frequency analysis method can obtain its time-frequency diagram through short-time Fourier transform (STFT) or wavelet transform, and analyze the time domain characteristics and frequency domain characteristics of the noise signal. The frequency range of high-frequency noise is usually high, for example, above 1000 Hz, and the accuracy of time-frequency analysis can be improved by selecting appropriate window functions and transformation parameters.

[0032] More specifically, for sudden noise, the event detection method can monitor the amplitude or energy of the noise signal in real time by setting the threshold of the noise signal. When it exceeds the threshold, it is judged as a noise event. The threshold can be set according to the background noise level of the noise signal and the application scenario. For continuous noise, the statistical analysis method can calculate the statistical characteristics of the noise signal such as the mean, variance, standard deviation and probability distribution, analyze its overall characteristics and change patterns, and thus realize long-term monitoring and analysis of the noise signal.

[0033] Preferably, when selecting a monitoring method, it can be optimized according to the specific characteristics of the noise signal and the application scenario. For example, for a complex noise environment, a variety of monitoring methods can be combined for comprehensive analysis to improve the accuracy and reliability of monitoring. For application scenarios with high real-time requirements, event detection methods and spectrum analysis methods can be given priority to quickly respond to noise events and analyze the frequency characteristics of noise signals. In addition, adaptive algorithms can be introduced to dynamically adjust monitoring parameters to adapt to changes in noise signals and changes in monitoring requirements, further improving the flexibility and adaptability of monitoring.

[0034] In some embodiments, the principal component analysis algorithm is used to reduce the dimensions of multiple groups of initial noise parameters into multiple feature vectors, including: Calculate the covariance matrix of each set of initial noise parameters; Perform eigenvalue decomposition on the covariance matrix to obtain eigenvectors and eigenvalues; According to the size of the eigenvalue, select the first few main eigenvectors to form the principal component eigenvector matrix; Each set of initial noise parameters is multiplied by the principal component eigenvector matrix to obtain the eigenvector after dimensionality reduction, which is used as the input for subsequent genetic algorithm optimization.

[0035] It should be noted that the principal component analysis (PCA) algorithm is used to reduce the dimensions of multiple sets of initial noise parameters into multiple eigenvectors for subsequent optimization processing. PCA is a commonly used dimensionality reduction technique that extracts the main eigenvectors by calculating the covariance matrix of the data and performing eigenvalue decomposition, thereby reducing the dimension of the data while retaining most of the information. The eigenvector is a vector extracted from PCA that represents the main direction of the data, and the eigenvalue represents the importance or variance of each eigenvector. By selecting the main eigenvectors, the complexity of the data can be effectively reduced and the efficiency of subsequent processing can be improved.

[0036] Specifically, we first calculate the covariance matrix of each set of initial noise parameters. The covariance matrix is ​​a square matrix used to represent the covariance relationship between the variables in the data. , whose covariance matrix It can be expressed as in, is the mean of the noise parameters, is the number of samples. Then the covariance matrix is ​​decomposed to obtain the eigenvectors and eigenvalues.

[0037] More specifically, the result of eigenvalue decomposition can be expressed as ,in is the eigenvector matrix, is the eigenvalue diagonal matrix. According to the size of the eigenvalue, select the first few main eigenvectors to form the principal component eigenvector matrix. Usually, the first few eigenvectors with the largest eigenvalues ​​can be selected to retain most of the information. Finally, multiply each set of initial noise parameters by the principal component eigenvector matrix to obtain the eigenvector after dimensionality reduction as the input for subsequent genetic algorithm optimization.

[0038] Preferably, when selecting eigenvectors, the number of eigenvectors to be selected can be determined based on the cumulative contribution rate of the eigenvalues. The cumulative contribution rate refers to the ratio of the sum of the first few eigenvalues ​​to the total of all eigenvalues. Usually, the number of eigenvectors whose cumulative contribution rate reaches a certain threshold (such as 85% or 90%) can be selected.

[0039] Furthermore, other dimensionality reduction techniques can be considered as alternatives, such as linear discriminant analysis (LDA) or t-distributed stochastic neighbor embedding (t-SNE) to further improve the effect and adaptability of dimensionality reduction.

[0040] In some embodiments, the optimization of each feature vector in sequence using a genetic algorithm according to a preset optimization operator includes: The following iterative optimization operation is performed on each feature vector: Initialize the global optimal solution, the optimal solution of the previous iteration, and the weight of the optimization operator, and set the maximum number of executions; In each execution, an optimization operator is selected using the roulette wheel selection algorithm according to the weights of each optimization operator; According to the selected optimization operator, the first solution of each eigenvector is calculated; Apply crossover and mutation operations to the first solution to obtain the second solution; According to the global optimal solution, the optimal solution of the current iteration, the optimal solution of the previous iteration and the second solution, the performance record value of the selected optimization operator, the global optimal solution and the optimal solution of the current iteration are updated. After updating each eigenvector with the global optimal solution, the next execution is performed until the maximum number of executions is reached. For each eigenvector, the optimal solution of the previous iteration is updated using the optimal solution of the current iteration. According to the preset weight update rate and the performance record value of each optimization operator, the weight of each optimization operator is updated, and the next round of iterative optimization is performed until the maximum number of iterations of the eigenvector is reached, and the global optimal solution of each eigenvector is obtained, that is, each eigenvector after optimization.

[0041] It should be noted that the genetic algorithm is an optimization algorithm based on natural selection and genetic mechanisms, which is used to optimize each eigenvector. In this method, the genetic algorithm searches for the global optimal solution of the eigenvector through iterative optimization operations. The global optimal solution refers to the solution that can minimize the objective function value under a given optimization goal. The optimization operator is an operation used to generate new solutions in the genetic algorithm, including the crossover operator, mutation operator, and selection operator. The crossover operator is used to exchange gene fragments between two eigenvectors, the mutation operator is used to randomly select a gene position in the eigenvector for mutation, and the selection operator is used to select the optimal eigenvector for genetic operations according to the fitness function. Through the combined use of these optimization operators, the solution space can be effectively explored and the optimization effect can be improved.

[0042] Specifically, the global optimal solution, the optimal solution of the previous iteration, and the weight of the optimization operator are initialized first, and the maximum number of executions is set. The global optimal solution and the optimal solution of the previous iteration are used to record the optimal solutions in the current and previous iterations, and the weight of the optimization operator is used to control the selection probability of different optimization operators. In each execution, an optimization operator is selected by the roulette selection algorithm according to the weights of each optimization operator. The roulette selection algorithm is a probability-based selection method that can randomly select an optimization operator according to the weight of the optimization operator. Then, according to the selected optimization operator, the first solution of each eigenvector is calculated. The first solution is the initial solution generated by applying the optimization operator. Then, the first solution is subjected to crossover and mutation operations to obtain the second solution. The second solution is the solution after further optimization. Finally, according to the global optimal solution, the optimal solution of the current iteration, the optimal solution of the previous iteration, and the second solution, the performance record value of the selected optimization operator, the global optimal solution, and the optimal solution of the current iteration are updated, and each eigenvector is updated with the global optimal solution, and the next execution is performed until the maximum number of executions is reached.

[0043] Preferably, when updating the performance record value of the optimization operator, a fitness function can be introduced to evaluate the quality of the solution. The fitness function is used to measure the quality of the solution and is usually related to the objective function. For example, the reciprocal of the objective function value can be used as the fitness value. The smaller the objective function value, the larger the fitness value, indicating that the quality of the solution is higher.

[0044] Furthermore, the weight of the optimization operator can be dynamically adjusted according to the performance record value of the optimization operator and the preset weight update rate to adapt to different optimization requirements. The weight update rate is a parameter between 0 and 1, which is used to control the speed of weight update. By dynamically adjusting the weight, the ability of global search and local search can be better balanced, and the efficiency and accuracy of optimization can be improved.

[0045] In some embodiments, updating the performance record value of the selected optimization operator, the global optimal solution and the optimal solution of the current round of iteration according to the global optimal solution, the optimal solution of the current round of iteration, the optimal solution of the previous round of iteration and the second solution includes: If the optimal solution of the current iteration is empty, the second solution is used to update the optimal solution of the current iteration; If the objective function value of the second solution is less than the objective function value of the global optimal solution, then according to the value of the first weight update factor, the performance record value of the selected optimization operator is increased, and the global optimal solution is updated using the second solution; If the objective function value of the second solution is less than the objective function value of the optimal solution of the current iteration, then the performance record value of the selected optimization operator is increased according to the value of the second weight update factor, and the second solution is used to update the optimal solution of the current iteration; If the objective function value of the second solution is less than the objective function value of the optimal solution of the previous iteration, the performance record value of the selected optimization operator is increased according to the value of the third weight update factor.

[0046] It should be noted that this method improves the optimization process of the genetic algorithm by updating the performance record value of the optimization operator, the global optimal solution and the optimal solution of the current iteration. The performance record value is used to record the performance of the optimization operator in each iteration. The global optimal solution is the best solution currently found, and the optimal solution of the current iteration is the best solution found in the current iteration. By comparing the objective function value of the second solution with the objective function values ​​of the global optimal solution and the optimal solution of the current iteration, it can be determined whether the second solution is better than the current optimal solution, and the performance record value and the optimal solution are updated accordingly. The weight update factor is a parameter used to adjust the performance record value of the optimization operator, and the amount of weight increase is determined based on the comparison result between the second solution and the optimal solution.

[0047] Specifically, when the optimal solution of the round iteration is empty, the second solution is directly used to update the optimal solution of the current round iteration, because there is no better solution for comparison at this time. If the objective function value of the second solution is less than the objective function value of the global optimal solution, it means that the second solution is a better solution. At this time, the performance record value of the selected optimization operator is increased according to the value of the first weight update factor, and the global optimal solution is updated using the second solution. The first weight update factor is usually set to a value greater than 1 to ensure that the performance record value can be significantly increased, thereby increasing the probability of the optimization operator being selected in future iterations. Similarly, if the objective function value of the second solution is less than the objective function value of the optimal solution of the current round iteration, the performance record value is increased according to the value of the second weight update factor, and the second solution is used to update the optimal solution of the current round iteration. If the objective function value of the second solution is less than the objective function value of the optimal solution of the previous round iteration, the performance record value is increased according to the value of the third weight update factor. The second and third weight update factors are usually set to values ​​less than the first weight update factor to reflect the degree of influence of different comparison results on the performance record value.

[0048] Preferably, when updating the performance record value, different weight update factors can be set according to the specific performance of the optimization operator and the optimization target. For example, for some optimization operators, if they can find a solution better than the global optimal solution in multiple iterations, the value of the first weight update factor can be appropriately increased to encourage the use of the optimization operator. In addition, other update mechanisms such as fitness sharing or covariance matrix adaptation can be introduced to further improve the performance of the optimization operator and the overall effect of the algorithm.

[0049] In some embodiments, the weight of each optimization operator is updated by the following formula according to the preset weight update rate and the performance record value of each optimization operator: in, represents the weight of the p-th optimization operator at the t-th iteration, represents the weight of the p-th optimization operator at the t+1-th iteration, represents the performance record value of the p-th optimization operator at the t-th iteration, represents the weight update rate, , Represents the total number of optimization operators.

[0050] It should be noted that this formula is used to update the weight of the optimization operator to improve the optimization effect of the genetic algorithm. The weight of the optimization operator determines the probability of it being selected in the algorithm. The higher the weight, the greater the chance of the optimization operator being selected. In the formula, Indicates The optimization operator is in the The weight at the iteration, Indicates The optimization operator is in the The weight at the iteration, is the weight update rate, which is used to control the speed of weight update. It is The optimization operator is in the The performance record value at the iteration, are all optimization operators in the The sum of the performance records at the iterations, is the total number of optimization operators. Through this formula, the weight of the optimization operator can be dynamically adjusted according to its performance record value, thereby optimizing the algorithm's search process.

[0051] Specifically, the weight update rate It is usually set to a value between 0 and 1, such as 0.1 or 0.2, indicating the proportion of weight updates in each iteration. It is calculated based on the performance of the optimization operator in the current iteration. It can be the inverse of the objective function value of the solution generated by the optimization operator or other indicators related to the optimization effect.

[0052] More specifically, in each iteration, the new weight is calculated according to the performance record value of the optimization operator and the weight update rate according to the formula , and update the weights of the optimization operators. In this way, the optimization operators with better performance can be selected more frequently in subsequent iterations, thereby improving the optimization effect of the algorithm.

[0053] Preferably, when updating the weights of the optimization operators, other factors may be considered to further optimize the weight distribution. For example, the diversity of the optimization operators may be considered, and the weights of the optimization operators with higher diversity may be increased to avoid the algorithm from converging to the local optimal solution too early.

[0054] Furthermore, an adaptive mechanism can be introduced to dynamically adjust the weight update rate according to the convergence of the algorithm and the changes in the optimization objectives. , in order to better balance the capabilities of global search and local search and improve the adaptability and robustness of the algorithm.

[0055] In some embodiments, the optimization operator includes: The crossover operator is used to exchange gene segments between two feature vectors. The mutation operator is used to randomly select a gene position in the feature vector for mutation. The selection operator is used to select the optimal feature vector for genetic operation according to the fitness function.

[0056] It should be noted that the optimization operators involved in this method include crossover operators, mutation operators, and selection operators, which are used to generate and select new solutions in genetic algorithms. The crossover operator generates a new feature vector by exchanging gene fragments between two feature vectors, the mutation operator increases the diversity of solutions by randomly selecting a gene position in the feature vector for mutation, and the selection operator selects the optimal feature vector for genetic operation according to the fitness function. The combined use of these operators can effectively explore the solution space and improve the optimization effect and efficiency.

[0057] Specifically, the crossover operator can use single-point crossover, multi-point crossover or uniform crossover. For example, in single-point crossover, a crossover point is randomly selected, and then the gene segments of the two feature vectors after that point are exchanged. The mutation operator can use bit flipping, insertion, deletion or replacement. For example, in bit flip mutation, a gene bit is randomly selected and its value is reversed. The selection operator usually uses roulette selection, tournament selection or elite selection. For example, in roulette selection, the probability of the feature vector being selected is calculated according to its fitness value, and the feature vector of the next generation is determined by random selection. In practical applications, the appropriate operator type and parameter settings can be selected according to the characteristics and requirements of the optimization problem to improve the optimization effect.

[0058] Preferably, when using the crossover operator, an adaptive crossover rate can be introduced to dynamically adjust the probability of crossover, and the size of the crossover rate can be adjusted according to the diversity of the current population and the change of the optimization target. For example, when the diversity of the population is high, the crossover rate can be appropriately reduced to reduce unnecessary computational overhead; when the diversity of the population is low, the crossover rate can be appropriately increased to increase the diversity of solutions and avoid the algorithm from falling into the local optimal solution. In addition, it is also possible to consider combining multiple mutation operators and selection operators to further improve the optimization effect and adaptability.

[0059] In some embodiments, the real-time acquisition and preprocessing of the noise signal includes: Use a microphone array to collect noise signals in real time to obtain the original data of the noise signals; Filter, reduce noise and normalize the collected noise signal to eliminate the interference and noise components in the noise signal and retain the effective information of the noise signal; Feature extraction is performed on the processed noise signal to extract its multi-parameter features, providing basic data for subsequent noise type identification and monitoring analysis.

[0060] It should be noted that the method involves real-time acquisition and preprocessing of noise signals to obtain the original data of the noise signals and perform necessary processing to facilitate subsequent feature extraction and noise detection. Real-time acquisition refers to data acquisition immediately when the noise signal occurs to ensure the integrity and accuracy of the data. Preprocessing includes steps such as filtering, noise reduction and normalization, which are used to eliminate interference components and noise components in the noise signal and retain effective signal information. Feature extraction is to extract multi-parameter features from the processed noise signal to provide basic data for noise type identification and monitoring analysis.

[0061] Specifically, real-time acquisition can be achieved using a microphone array, which consists of multiple microphones and can collect noise signals from different directions and angles to improve the signal acquisition accuracy and coverage. Filtering can use low-pass filters, high-pass filters, or band-pass filters, etc., and select appropriate filter parameters according to the frequency range of the noise signal and the application scenario to remove unnecessary frequency components.

[0062] More specifically, noise reduction processing can use noise gating, spectral subtraction or wavelet denoising to reduce the noise level and retain useful signal components by analyzing the characteristics of the noise signal. Normalization processing adjusts the amplitude or energy of the signal to a uniform range to facilitate comparison and analysis between different signals. Feature extraction can include frequency features, amplitude features, time domain features and frequency domain features. For example, frequency features can extract the main frequency components of the signal, and amplitude features can extract the maximum, minimum and average values ​​of the signal.

[0063] Preferably, during the real-time acquisition process, the sampling frequency and sampling time of the microphone array can be dynamically adjusted according to the strength and change of the noise signal to improve the efficiency and accuracy of data acquisition. For example, when the noise signal is strong or changes rapidly, the sampling frequency and sampling time can be increased to capture more signal details; when the noise signal is weak or changes slowly, the sampling frequency and sampling time can be reduced to reduce the amount of data and computing overhead. In addition, a variety of preprocessing methods and feature extraction techniques can be combined to further improve the processing effect of the noise signal and the accuracy of feature extraction.

[0064] In some embodiments, the objective function takes the signal-to-noise ratio of the noise signal as the optimization target, and its mathematical form is: in, represents the power of the signal, Represents the power of the noise; constraints include the frequency range of the noise signal , Amplitude range , Noise Type and the duration of the noise signal .

[0065] It should be noted that the objective function of this method takes the signal-to-noise ratio of the noise signal as the optimization target. The signal-to-noise ratio is the ratio of signal power to noise power, which is used to measure the clarity and quality of the signal. The objective function is a mathematical expression used to quantify the effect of noise detection. By optimizing the objective function, the accuracy and efficiency of noise detection can be improved. Constraints are used to limit parameters such as the frequency range, amplitude range, noise type and duration of the noise signal to ensure the accuracy and reliability of the detection process. The frequency range and amplitude range represent the upper and lower limits of the frequency and amplitude of the noise signal, respectively. The noise types include low-frequency noise, high-frequency noise, burst noise and continuous noise, etc. The duration represents the duration range of the noise signal.

[0066] Specifically, the objective function can be expressed as in, is the power of the signal, is the power of the noise. The signal power can be obtained by calculating the sum of the squares of the signal, and the noise power can be obtained by calculating the sum of the squares of the noise. The constraints can be set according to the actual application scenario. For example, the frequency range can be set to , the amplitude range can be set to , the noise type can include multiple types of noise signals, and the duration can be set to .

[0067] In practical applications, the parameter settings of the objective function and constraints can be adjusted according to the specific noise environment and detection requirements to achieve the best detection effect.

[0068] Preferably, when setting the objective function and constraint conditions, other optimization objectives and constraint conditions may be considered to further improve the performance of noise detection. For example, the stability and consistency of the noise signal may be used as the optimization objective, and the variation range and fluctuation degree of the noise signal may be limited by adding corresponding constraint conditions.

[0069] Furthermore, a variety of optimization algorithms and detection techniques, such as particle swarm optimization algorithm, neural network detection technology, etc., can be combined with objective functions and constraints to achieve more efficient and accurate noise detection.

[0070] The above-mentioned embodiments of the present invention have the following beneficial effects: The beneficial effects of the present invention are mainly reflected in the improvement of the accuracy and efficiency of noise detection. By constructing the objective function and its constraints, and optimizing the noise parameters by using the gradient descent method, the influence of the noise parameters on the objective function value can be accurately determined, thereby realizing the optimization of noise detection. In addition, by establishing a voiceprint library and collecting voiceprint features of different noise types, the noise signal can be collected in real time and multi-parameter features can be extracted, further improving the accuracy of noise type identification. Combining principal component analysis and genetic algorithm to reduce the dimension and optimize the noise parameters can effectively reduce the computational complexity and improve the efficiency and adaptability of noise detection. At the same time, according to the identified noise type, the corresponding monitoring method is selected, such as spectrum analysis, time-frequency analysis, event detection and statistical analysis, etc., to adapt to the characteristics and monitoring requirements of different noise signals, further enhancing the flexibility and applicability of noise detection.

[0071] By optimizing the weight update mechanism in the genetic algorithm, the present invention can more effectively select and adjust the optimization operator, thereby improving the optimization effect of the feature vector and further improving the accuracy and stability of noise detection. In addition, real-time acquisition and preprocessing of noise signals, including filtering, noise reduction and normalization, can effectively eliminate interference components and noise components in the noise signal, retain effective information, and provide more accurate basic data for subsequent noise type identification and monitoring analysis. The objective function takes the signal-to-noise ratio of the noise signal as the optimization target, which can better reflect the actual effect of noise detection and ensure the reliability and effectiveness of the detection results.

[0072] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, or a terminal device such as a computer or a server.

[0073] The above descriptions are only some preferred embodiments of the present invention and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A noise intelligent detection method based on multiple parameters, characterized in that: The following steps are involved: Construct the objective function and constraints of noise detection, The influence of noise parameters on the objective function value is determined by the gradient descent method, including: Calculate the partial derivative of the objective function with respect to each noise parameter to obtain the gradient value of each parameter; Adjust the noise parameter according to the gradient value so that it moves in the direction of gradient descent until the objective function value reaches the minimum; Adjust the noise parameters according to the gradient descent method to obtain multiple groups of initial noise parameters; Establish a voiceprint database, collect voiceprint features of different noise types, and classify and store the voiceprint features; Real-time acquisition and preprocessing of noise signals to extract multi-parameter features of noise signals, including frequency features, amplitude features, time domain features and frequency domain features; Match the extracted multi-parameter features with the voiceprint features in the voiceprint library to identify the type of noise signal; According to the identified noise type, select the corresponding monitoring method to monitor and analyze the noise signal; The following update operations are iteratively performed on the multiple groups of initial noise parameters: the multiple groups of initial noise parameters are reduced to multiple feature vectors using a principal component analysis algorithm; According to the preset optimization operator, a genetic algorithm is used to optimize each feature vector in turn, and then multiple groups of global noise parameters are obtained by merging them; If the total objective function value of the multiple sets of global noise parameters is less than the total objective function value of the multiple sets of initial noise parameters, then the multiple sets of initial noise parameters are updated using the multiple sets of global noise parameters; The next round of updating operation is performed until the stop condition is reached, and the updated multiple sets of initial noise parameters constitute the optimal noise detection parameters.

2. The multi-parameter based noise intelligent detection method according to claim 1, characterized in that: The extracted multi-parameter features are matched with the voiceprint features in the voiceprint library to identify the type of noise signal, including: The similarity between the extracted multi-parameter features and each voiceprint feature in the voiceprint library is calculated using Euclidean distance, cosine similarity or dynamic time warping algorithm; According to the preset similarity threshold, the noise type corresponding to the voiceprint feature with the highest similarity is selected as the preliminary recognition result; Conduct confidence assessment on the preliminary recognition results and calculate their confidence values; If the confidence value is lower than the preset confidence threshold, the machine learning model is used for classification and identification to obtain the final noise type identification result.

3. The multi-parameter based noise intelligent detection method according to claim 2, characterized in that: The method of selecting a corresponding monitoring method according to the identified noise type includes: For low-frequency noise, spectrum analysis is used for monitoring. The specific steps are as follows: Perform fast Fourier transform on the noise signal to obtain its spectrum, and analyze the frequency components and energy distribution in the spectrum; For high-frequency noise, the time-frequency analysis method is used for monitoring. The specific steps are as follows: Perform short-time Fourier transform or wavelet transform on the noise signal to obtain its time-frequency diagram, and analyze the time domain characteristics and frequency domain characteristics of the noise signal at the same time; For sudden noise, the event detection method is used for monitoring. The specific steps are as follows: Set the threshold of the noise signal and monitor the amplitude or energy of the noise signal in real time. When it exceeds the threshold, it is judged as a noise event. For continuous noise, statistical analysis method is used for monitoring. The specific steps are as follows: Calculate the statistical characteristics of the noise signal, including mean, variance, standard deviation, and probability distribution, and analyze its overall characteristics and changing patterns.

4. The multi-parameter based noise intelligent detection method according to claim 3 is characterized in that: The principal component analysis algorithm is used to reduce the dimensions of multiple groups of initial noise parameters into multiple feature vectors, including: Calculate the covariance matrix of each set of initial noise parameters; Perform eigenvalue decomposition on the covariance matrix to obtain eigenvectors and eigenvalues; According to the size of the eigenvalue, select the first few main eigenvectors to form the principal component eigenvector matrix; Each set of initial noise parameters is multiplied by the principal component eigenvector matrix to obtain the eigenvector after dimensionality reduction, which is used as the input for subsequent genetic algorithm optimization.

5. The multi-parameter based noise intelligent detection method according to claim 4, characterized in that: The method of optimizing each feature vector in sequence by using a genetic algorithm according to a preset optimization operator includes: The following iterative optimization operation is performed on each feature vector: Initialize the global optimal solution, the optimal solution of the previous iteration, and the weight of the optimization operator, and set the maximum number of executions; In each execution, an optimization operator is selected using the roulette wheel selection algorithm according to the weights of each optimization operator; According to the selected optimization operator, the first solution of each eigenvector is calculated; Apply crossover and mutation operations to the first solution to obtain the second solution; According to the global optimal solution, the optimal solution of the current iteration, the optimal solution of the previous iteration and the second solution, the performance record value of the selected optimization operator, the global optimal solution and the optimal solution of the current iteration are updated. After each eigenvector is updated with the global optimal solution, the next execution is performed until the maximum number of executions is reached. For each eigenvector, the optimal solution of the previous iteration is updated with the optimal solution of the current iteration. According to the preset weight update rate and the performance record value of each optimization operator, the weight of each optimization operator is updated, and the next round of iterative optimization is performed until the maximum number of iterations of the eigenvector is reached, and the global optimal solution of each eigenvector is obtained, that is, each eigenvector after optimization.

6. The multi-parameter based noise intelligent detection method according to claim 5, characterized in that: The updating of the performance record value of the selected optimization operator, the global optimal solution and the optimal solution of the current round of iteration according to the global optimal solution, the optimal solution of the current round of iteration, the optimal solution of the previous round of iteration and the second solution comprises: If the optimal solution of the current iteration is empty, the second solution is used to update the optimal solution of the current iteration; If the objective function value of the second solution is less than the objective function value of the global optimal solution, then according to the value of the first weight update factor, the performance record value of the selected optimization operator is increased, and the global optimal solution is updated using the second solution; If the objective function value of the second solution is less than the objective function value of the optimal solution of the current iteration, then the performance record value of the selected optimization operator is increased according to the value of the second weight update factor, and the second solution is used to update the optimal solution of the current iteration; If the objective function value of the second solution is less than the objective function value of the optimal solution of the previous iteration, the performance record value of the selected optimization operator is increased according to the value of the third weight update factor.

7. The multi-parameter based noise intelligent detection method according to claim 6, characterized in that: According to the preset weight update rate and the performance record value of each optimization operator, the weight of each optimization operator is updated by the following formula: in, represents the weight of the p-th optimization operator at the t-th iteration, represents the weight of the p-th optimization operator at the t+1-th iteration, represents the performance record value of the p-th optimization operator at the t-th iteration, represents the weight update rate, , Represents the total number of optimization operators.

8. The multi-parameter based noise intelligent detection method according to claim 7, characterized in that: The optimization operator includes: The crossover operator is used to exchange gene segments between two feature vectors. The mutation operator is used to randomly select a gene position in the feature vector for mutation. The selection operator is used to select the optimal feature vector for genetic operation according to the fitness function.

9. The multi-parameter based noise intelligent detection method according to claim 8, characterized in that: The real-time acquisition and preprocessing of the noise signal includes: Use a microphone array to collect noise signals in real time to obtain the original data of the noise signals; Filter, reduce noise and normalize the collected noise signal to eliminate the interference and noise components in the noise signal and retain the effective information of the noise signal; Feature extraction is performed on the processed noise signal to extract its multi-parameter features, providing basic data for subsequent noise type identification and monitoring analysis.

10. The method according to claim 6, characterized in that The objective function takes the signal-to-noise ratio of the noise signal as the optimization target, and its mathematical form is: in, represents the power of the signal, Represents the power of the noise; constraints include the frequency range of the noise signal , Amplitude range , Noise Type and the duration of the noise signal .

Images