A sound source localization and management system and method based on multi-source information fusion

Abstract

The present invention relates to a sound source positioning and management system and method based on multi-source information fusion, which adopts multi-source information fusion and Bayesian estimation model to locate noise sources by integrating acoustic, vibration and meteorological information. Acoustic sensors capture noise, vibration sensors assist in judgment, and meteorological sensors optimize sound speed calculation, which has higher accuracy in complex environments and multi-source scenarios. Based on real-time positioning and noise characteristic monitoring, the dynamic control execution unit adjusts equipment parameters such as active noise reduction devices according to the decision algorithm, and the user interface facilitates operation and monitoring, thereby improving operability and flexibility. It has been verified in multiple environments such as industrial plants, traffic sections, and construction sites, and can effectively control various types of noise. It has strong environmental adaptability and wide practicability, and can provide a reliable solution for noise control in multiple fields.

Description

Technical Field

[0001] The present invention relates to the fields of acoustic engineering and environmental management, and in particular to a sound source localization and management system based on multi-source information fusion. Background Art

[0002] At a time when urbanization is accelerating, noise pollution has become an extremely serious environmental problem. In industrial production activities, high-intensity noise generated by the continuous operation of various mechanical equipment permeates factory workshops and surrounding areas; in the field of transportation, the roar of vehicles, trains, airplanes and other means of transportation fills the city's main roads, stations, airports and other places; in the construction process, the sharp noise generated by equipment such as pile drivers, cranes, and concrete mixers seriously interferes with the normal lives of surrounding residents. These noises not only cause great trouble to people's daily lives, such as affecting sleep quality, interfering with normal conversations and learning, but also have many negative effects on people's physical and mental health. For example, long-term exposure to noise environments may cause hearing damage, cardiovascular diseases, increased psychological pressure, and reduced work efficiency. CN110531023A "Noise source positioning method based on microphone array" proposes a noise source positioning method based on microphone array, which determines the direction of the noise source by calculating the time delay difference of the noise signal received by the microphone array. However, in actual complex environments, such as when there are multipath reflections, obstacles blocking, and background noise interference, its positioning accuracy will be greatly reduced. After actual testing, in a more complex industrial plant environment, the average positioning error of this method can reach about 8 meters, and under certain special working conditions, the error may even exceed 10 meters, which is difficult to meet the needs of high-precision positioning. CN106932357A "A noise source positioning system for a distributed acoustic sensor network" This patent uses a distributed acoustic sensor network to collect noise signals and adopts a positioning algorithm based on signal strength. However, the algorithm has extremely high requirements for the arrangement density and position accuracy of the sensors, and it is difficult to achieve the ideal state in actual deployment. At the same time, environmental factors (such as the impact of temperature and humidity changes on sound propagation) can also lead to inaccurate signal strength measurements, which in turn affects positioning accuracy. In urban traffic scene tests, the positioning error is usually between 5-10 meters, and the positioning results are unstable and prone to large deviations. CN107830934A "An adaptive sound insulation material and its preparation method" This patent provides an adaptive sound insulation material that can automatically adjust the sound insulation performance according to the noise frequency. However, its adjustment range is limited, and once the material is prepared, its adaptive ability is relatively fixed, and it cannot be flexibly adjusted according to the dynamic changes of the actual noise source. In practical applications, the noise reduction effect is limited in complex and changeable noise environments. Generally, it can only increase the noise reduction by 5-10dB within a specific frequency range, which makes it difficult to cope with broadband noise. CN109540478A "An Intelligent Ventilation Noise Reduction System" This system reduces the noise generated by the ventilation system by adjusting the size and angle of the vents. However, the system is only optimized for ventilation noise, has a narrow scope of application, and lacks the ability to perceive and process other types of noise sources. In actual environments, when there are multiple noise sources acting together, the noise reduction effect is minimal and cannot effectively improve the overall acoustic environment.

[0003] In summary, the existing noise control technology has many shortcomings. In terms of noise source positioning, whether it is based on microphone arrays or distributed sensor networks, it is difficult to achieve high-precision and stable positioning in complex environments. In terms of noise reduction technology, whether it is adaptive sound insulation materials or specific ventilation noise reduction systems, there are problems such as poor adaptability, limited effects and single functions. These limitations have led to the inability of existing noise control technologies to effectively cope with complex and changeable noise pollution conditions. There is an urgent need for an innovative sound source positioning and management system based on multi-source information fusion to achieve accurate positioning of noise sources, comprehensive monitoring of noise characteristics, and the ability to dynamically adjust noise reduction strategies according to real-time conditions, thereby effectively improving noise reduction effects and protecting people's quality of life and health. Summary of the invention

[0005] The present invention provides a sound source localization and management system based on multi-source information fusion, comprising a sensor network unit, a data processing center, a dynamic control execution unit and a user interaction interface, wherein the sensor network unit is used to collect noise-related data and transmit the data to the data processing center, the data processing center uses a noise source localization model based on multi-source information fusion and Bayesian estimation to locate and analyze the noise source, and formulates a control strategy according to the positioning result and noise characteristics, the dynamic control execution unit executes the control strategy to reduce noise, and the user interaction interface is used for system operation and information display;

[0006] According to the Bayesian formula, the posterior probability distribution P(L|x a (t), x υ (t), w(t)) is:

[0007]

[0008] The location estimate of the noise source is determined by solving the maximum value of the posterior probability distribution Right now:

[0009]

[0010] Furthermore, in the noise source location model based on multi-source information fusion and Bayesian estimation, the noise signal collected by the acoustic sensor is preprocessed to extract the acoustic feature vector, the vibration signal collected by the vibration sensor is preprocessed to extract the vibration feature vector, the meteorological parameters measured by the meteorological sensor are used to calculate the speed of sound, and the Bayesian estimation method is used to fuse the acoustic, vibration and meteorological information to determine the location of the noise source, and the likelihood function is calculated based on the acoustic and vibration signal propagation model and the influence of meteorological conditions on noise propagation. Further, the noise source location model is based on the improved particle swarm optimization-support vector machine (IPSO-SVM) model:

[0011]

[0012] Among them, k is the number of iterations, w is the inertia weight, c1 and c2 are acceleration coefficients, and r1 and r2 are random numbers uniformly distributed in the interval [0, 1].

[0013] In the IPSO algorithm, the following improvements are made to the traditional PSO algorithm:

[0014] οDynamic inertia weight adjustment: The inertia weight w is no longer a fixed value, but changes dynamically with the increase of the number of iterations. Initially, w is large, which is conducive to the particles searching the solution space in a larger range. As the number of iterations increases, w gradually decreases, allowing particles to perform fine searches in local areas. The specific calculation formula is: where w max and w min are the maximum and minimum values ​​of the inertia weight, respectively, k max is the maximum number of iterations.

[0015] ο Adaptive mutation strategy: In order to avoid particles falling into the local optimal solution, an adaptive mutation strategy is introduced. In each iteration, the particle's fitness value and the average fitness value of the group are used to determine whether the particle needs to mutate. If the particle's fitness value is lower than a certain proportion (such as 80%) of the group's average fitness value, the position of the particle is mutated. The mutation method is to randomly generate a new value within the particle's parameter value range to replace the original parameter value, which can increase the diversity of particles and improve the algorithm's ability to jump out of the local optimal solution.

[0016] After optimizing the parameters of SVM through the IPSO algorithm, the optimal γ and C values ​​are obtained, thus building a noise source location model with better performance. Using this model, the location coordinates of the noise source can be predicted based on the noise signal feature vector collected by the acoustic sensor.

[0017] Furthermore, the acoustic feature vector includes energy, power spectrum density and zero-crossing rate of the noise signal, and the vibration feature vector includes vibration frequency, amplitude and acceleration.

[0018] Furthermore, the acoustic propagation model calculates the sound wave propagation path and attenuation based on the ray tracing principle, and the vibration propagation model is constructed according to the vibration propagation theory and the structural characteristics of the equipment.

[0019] Furthermore, the data processing center adopts an expectation maximization algorithm to solve the optimization problem in the noise source localization model based on multi-source information fusion and Bayesian estimation, and iteratively updates the model parameters and noise source position estimation by alternately executing expectation steps and maximization steps until the convergence conditions are met.

[0020] Furthermore, the dynamic control execution unit includes a movable active noise reduction device, an intelligent sound insulation barrier and a ventilation system noise controller, which can adjust working parameters in real time according to instructions from the data processing center to achieve noise reduction.

[0021] Furthermore, the system conducts model training and optimization through long-term data collection in different noise environment scenarios. The collected data includes acoustic, vibration and meteorological data, and performs real-time preprocessing on the collected data, including outlier removal, data interpolation and data synchronization.

[0022] Furthermore, when evaluating the noise reduction effect, noise monitoring points are arranged in multiple test areas to measure the difference in equivalent continuous sound levels and spectrum changes before and after noise reduction. The positioning accuracy is tested by simulating noise sources. The system adaptability is evaluated by changing the state of noise sources. At the same time, relevant parameters such as equipment cost, energy consumption, installation and maintenance difficulty are recorded for comprehensive system evaluation.

[0023] On the other hand, the present invention also provides a noise source positioning and noise reduction control method based on multi-source information fusion, comprising the following steps: Step 1: using a sensor network to collect data, the sensor network includes an acoustic sensor, a vibration sensor and a meteorological sensor, collecting noise signals in the environment, vibration signals of noise source equipment and meteorological parameter information, performing preprocessing operations on various types of collected data, including removing interference components in noise signals, extracting effective features in vibration signals and calibrating meteorological parameter data, and then transmitting the processed data to a data processing center;

[0024] Step 2: In the data processing center, noise source location is performed based on multi-source information fusion technology;

[0025] Step 3: According to the noise source positioning results and the noise characteristics of real-time monitoring, combined with the preset noise reduction strategy library, a dynamic noise reduction control strategy is formulated, wherein the noise reduction strategy library contains a variety of control schemes for different types of noise sources, different locations and different noise characteristics;

[0026] Step 4: Send the formulated control strategy to the dynamic control execution unit. The execution unit operates the corresponding noise reduction equipment according to the strategy instructions. During the control process, it continuously collects feedback information on the noise reduction effect, including changes in noise intensity and spectrum distribution. The control strategy is adjusted in real time according to the feedback information to ensure that the noise reduction effect reaches the optimal state.

[0027] Furthermore, in the step 2, firstly, according to the noise signal characteristics collected by the acoustic sensor, the vibration signal characteristics detected by the vibration sensor and the meteorological parameter information obtained by the meteorological sensor, a noise signal propagation model, a vibration signal propagation model and a noise source position probability model considering the meteorological influence are established respectively, and then the position estimation of the noise source is determined by fusing the information of the three models using the Bayesian estimation method. In this process, the prior probability distribution of the noise source position is set according to prior knowledge and historical data, and the probability of collecting multi-source data under a given noise source position is evaluated by calculating the likelihood function, and finally the noise source position estimation result based on the posterior probability distribution is obtained.

[0028] Beneficial technical effects of the invention:

[0029] Adopting multi-source information fusion and Bayesian estimation model, the noise source is located by integrating acoustic, vibration and meteorological information. Acoustic sensors capture noise, vibration sensors assist in judgment, and meteorological sensors optimize sound speed calculation. Compared with traditional single information positioning, it has higher accuracy in complex environments and multi-source scenarios. Based on real-time positioning and noise characteristic monitoring, the dynamic control execution unit adjusts the parameters of equipment such as active noise reduction devices according to the decision algorithm. Real-time adaptation to changes in different noise sources, such as industrial equipment, traffic, construction noise, etc., avoids waste of resources and improves noise reduction effect. The new mathematical model uses Bayesian formula to fuse information and EM algorithm to solve, which can handle complex problems more accurately. Model parameters can be optimized to adapt to different environments and noise sources, laying the foundation for accurate noise reduction. Integrated multi-unit collaborative work, sensor network collects data, data processing center makes decisions, control unit executes, user interface facilitates operation and monitoring, and improves operability and flexibility. It has been verified in multiple environments such as industrial plants, traffic sections, and construction sites, effectively controlling various types of noise, with strong environmental adaptability and wide practicality, and can provide a reliable solution for noise control in multiple fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 : Flow chart of the working principle of the present invention. DETAILED DESCRIPTION

[0031] Example 1

[0032] Sensor network unit: It consists of multiple acoustic sensors, vibration sensors and meteorological sensors distributed in the monitoring area. Acoustic sensors are used to collect noise signals. They have high sensitivity and wide-band response characteristics, and can accurately capture noise of different frequencies and intensities; vibration sensors can detect the vibration of noise source equipment and assist in determining the type and location of noise sources; meteorological sensors measure meteorological parameters such as temperature, humidity, air pressure and wind in the environment. These parameters are crucial for accurately calculating the speed of sound and the noise propagation path. The sensor nodes transmit the collected data to the data processing center in real time through wireless communication modules (such as Wi-Fi, ZigBee or Bluetooth, etc.).

[0033] Data Processing Center: Equipped with high-performance computing equipment and large-capacity storage systems, it is responsible for receiving and processing data from sensor network units. The center runs noise source location and analysis algorithms, noise propagation models, and dynamic control decision algorithms to analyze and process data in real time, locate noise sources, evaluate noise characteristics, and formulate control strategies.

[0034] Dynamic control execution unit: executes corresponding noise reduction control measures according to the instructions of the data processing center. This unit includes various types of noise reduction equipment, such as movable active noise reduction devices, intelligent sound insulation barriers, ventilation system noise controllers, etc. These devices can adjust the working parameters in real time according to the instructions received, such as the noise reduction frequency and amplitude of the active noise reduction device, the position and angle of the sound insulation barrier, the air volume and wind speed of the ventilation system, etc., to achieve effective control of noise.

[0035] User interaction interface: provides operators with an intuitive and convenient operation platform for system parameter setting, status monitoring and control command issuance. Operators can view information such as noise source location, noise intensity distribution, equipment operation status, etc. through this interface. They can also manually adjust the noise reduction strategy according to actual needs, or view the system's historical data and operation reports, realize human-computer interaction functions, and improve the system's operability and management efficiency.

[0036] Noise source localization model based on multi-source information fusion and Bayesian estimation;

[0037] Assume that the noise signal collected by acoustic sensor i at time t is s i (t), the vibration signal detected by vibration sensor j is υ j (t), the environmental parameters measured by the meteorological sensor are w(t) = {T(t), H(t), P(t), F(t)} (representing temperature, humidity, air pressure and wind respectively).

[0038] First, the acoustic signal is preprocessed, including removing the DC component, filtering (using adaptive filtering algorithm to remove background noise interference) and feature extraction. The extracted acoustic features include signal energy, power spectrum density, zero-crossing rate, etc. Suppose the acoustic feature vector is x a (t) = {E(s i (t)), PSD(s i (t)), ZCR(s i (t))}.

[0039] For the vibration signal, preprocessing and feature extraction are also performed. The vibration features can include the frequency, amplitude, acceleration, etc. The vibration feature vector is x υ (t) = {f(υ j (t)), A(υ j (t)), a(υ j (t))}.

[0040] The sound speed c(t) is calculated based on the meteorological parameters. The sound speed calculation formula is:

[0041]

[0042] Where c0 is the speed of sound under standard conditions and T0 is the standard temperature.

[0043] The Bayesian estimation method is used to integrate acoustic, vibration and meteorological information to locate the noise source. Assume that the location of the noise source is L = (x, y, z), and the prior probability distribution is P (L) (it can be estimated based on the characteristics of the monitoring area and historical data. For example, in industrial areas, the probability of noise sources appearing at certain equipment locations is higher).

[0044] Likelihood function P(x a (t), x υ (t), w(t)|L) represents the probability of observing acoustic, vibration and meteorological data given the noise source location L. Assuming that the acoustic, vibration and meteorological information are independent of each other at a given noise source location, the likelihood function can be expressed as:

[0045] P(x a (t), x υ (t), w(t)|L)=P(x a (t)|L)P(x υ (t)|L)P(w(t)|L)

[0046] For P(x a (t)|L), based on the acoustic propagation theory and the characteristics of the sensor, an acoustic signal propagation model is established, taking into account the attenuation, reflection, diffraction and other factors of the sound waves emitted by the noise source during the propagation process, and calculating the acoustic characteristic vector x generated at position L aFor example, using an acoustic propagation model based on ray tracing, the acoustic wave propagation path is calculated according to the noise source position, sensor position and sound speed, and then P(x a (t)|L).

[0047] Similarly, for P(x υ (t)|L), based on the vibration propagation theory and the structural characteristics of the equipment, a vibration signal propagation model is established to calculate the vibration characteristic vector x generated at position L υ The probability of (t).

[0048] P(w(t)|L) mainly considers the impact of meteorological conditions on noise propagation, such as the impact of wind on the direction of sound wave propagation, and can be calculated based on meteorological principles and empirical formulas.

[0049] According to the Bayesian formula, the posterior probability distribution P(L|x a (t), x υ (t), w(t)) is:

[0050]

[0051] The location estimate of the noise source is determined by solving the maximum value of the posterior probability distribution Right now:

[0052]

[0053] In actual calculations, an iterative algorithm (such as the expectation maximization algorithm - EM algorithm) is used to solve the above optimization problem. The EM algorithm gradually approaches the optimal solution by alternating between the expectation step (E-step) and the maximization step (M-step). In the E-step, the expected likelihood function is calculated; in the M-step, the expected likelihood function is maximized to update the noise source position estimation and model parameters (such as the attenuation coefficient in the acoustic propagation model, the damping coefficient in the vibration propagation model, etc.).

[0054] During the modeling and solving process, sensor network units are deployed in different noise environment scenarios (such as factories, transportation hubs, construction sites, residential areas, etc.) for long-term data collection. During the collection process, ensure the accurate installation and calibration of the sensors, and record the location information and collection time of the sensors. Perform real-time preprocessing on the collected acoustic, vibration and meteorological data, including removing outliers (such as abnormal data points caused by sensor failure or sudden interference), data interpolation (for a small number of missing data points, use appropriate interpolation methods to supplement, such as linear interpolation or spline interpolation) and data synchronization (ensure that the data collected by different sensors are aligned in time).

[0055] Model initialization and parameter estimation phase

[0056] ο Based on the prior knowledge of the monitoring area (such as statistical information on equipment distribution, building structure, and meteorological conditions in the area), the parameters of the noise source location model are initialized, including the prior probability distribution P(L), the initial attenuation coefficient in the acoustic propagation model, the initial damping coefficient in the vibration propagation model, etc. At the same time, the initial parameters of the EM algorithm are set, such as the number of iterations, the convergence threshold, etc.

[0057] Iterative solution phase (EM algorithm)

[0058] οE-step (expected step): Estimate the current noise source position and model parameters, calculate the expected likelihood function where θ represents the model parameter vector, θ old Represents the model parameters of the previous iteration.

[0059] οM-step (maximization step): maximize the expected likelihood function Q(θ,θ old ), solve to get the updated model parameters θ new Then, based on the updated model parameters and observed data, the posterior probability distribution P(L|x a (t), x υ (t), w(t)), and obtain the new noise source position estimate

[0060] Repeat the E-step and M-step until the convergence condition is met (such as the change in the posterior probability distribution is less than the set threshold or the maximum number of iterations is reached).

[0061] During the operation of the system, the sensor network unit continuously collects data, which is input into the trained noise source positioning model after preprocessing to calculate the position estimate of the noise source in real time. The data processing center generates control instructions using the dynamic control decision algorithm based on the positioning results and the noise characteristics monitored in real time (such as noise intensity, frequency distribution, etc.), combined with the preset noise reduction strategy library, and sends them to the dynamic control execution unit.

[0062] The dynamic control execution unit adjusts the working parameters of the noise reduction equipment in real time according to the instructions to achieve dynamic control of noise. At the same time, the system continuously collects noise feedback data after control to further optimize the model and control strategy.

[0063] For the comparative test, three typical noise test areas were selected: Area 1: Large industrial plants, with many different types of machinery and equipment inside, complex and diverse noise sources, and equipment operating status and location may change. Area 2: Busy urban traffic sections, with large traffic flow and various types of vehicles, the main noise source is traffic noise, and traffic flow and vehicle speed change over time. Area 3: Construction sites, with a variety of construction machinery used during construction, high noise source intensity, and constantly changing locations.

[0064] In each test area, the following three noise reduction schemes were deployed for comparative tests: Scheme 1: Traditional fixed noise reduction measures, such as installing fixed sound barriers and silencers, do not have real-time monitoring and dynamic control functions. Scheme 2: The existing noise source positioning and noise reduction system based on a single sensor (such as only an acoustic sensor), using conventional positioning algorithms (such as positioning algorithms based on arrival time difference-TDOA) and simple control strategies (such as active noise reduction equipment with fixed parameters). Scheme 3: The acoustic governance noise reduction management system of the present invention with real-time noise source positioning and dynamic control functions. In each test area, multiple noise monitoring points are evenly arranged, and a high-precision sound level meter is used to measure the noise levels at different locations before and after noise reduction. At the same time, the relevant parameters such as equipment cost, energy consumption, installation and maintenance difficulty of each scheme are recorded. The present invention adopts a noise source positioning model based on multi-source information fusion and Bayesian estimation, integrating acoustic, vibration and meteorological information. Compared with the traditional positioning method that only relies on acoustic signals or a single sensor, this model can locate the noise source more comprehensively and accurately. By comprehensively considering the impact of multiple factors on noise propagation, such as vibration signals to assist in determining the type and location of noise sources, and meteorological parameters to accurately calculate the speed of sound and propagation path, the positioning accuracy is improved, especially in complex environments and scenarios with multiple noise sources. Based on real-time positioning results and noise characteristic monitoring, the system uses a dynamic control decision algorithm to generate a highly adaptive control scheme. According to the characteristics of different noise sources (such as industrial equipment noise, traffic noise, construction noise, etc.) and changes (such as intensity fluctuations, frequency changes, position movement, etc.), the working parameters of equipment such as active noise reduction devices, intelligent sound insulation barriers, and ventilation system noise controllers can be adjusted in real time. This precise dynamic control can effectively improve the noise reduction effect, so that the noise reduction measures always match the actual noise conditions, avoid resource waste, and improve the overall performance of the system. The proposed mathematical model based on multi-source information fusion and Bayesian estimation fuses multi-source information through the Bayesian formula, and uses the EM algorithm to iteratively solve the posterior probability distribution to determine the location of the noise source. The model and algorithm can effectively handle complex noise propagation and multi-source information fusion problems. Compared with traditional simple positioning and control models, the calculation results are more accurate and reliable. The model parameters can be continuously optimized according to actual data, which improves the system's adaptability to different environments and noise sources, and provides a solid theoretical basis for achieving precise noise reduction. The system integrates a sensor network unit, a data processing center, a dynamic control execution unit, and a user interaction interface, and each part works together to achieve efficient noise reduction management. The sensor network unit collects multi-source data in real time, the data processing center performs complex calculations and decisions, the control execution unit accurately executes control instructions, and the user interaction interface facilitates operators to monitor and manage the system. This intelligent integrated design makes the system run smoothly and efficiently, and the human-computer interaction function enhances the operability and flexibility of the system, making it easy to adjust the noise reduction strategy according to actual needs.After testing and verification in different typical noise environments such as large industrial plants, busy urban traffic sections, and construction sites, the system of the present invention has shown good environmental adaptability. Whether facing complex and changeable industrial noise sources, dynamic traffic noise, or high-intensity and position-changing construction noise, the system can effectively locate the noise source and implement dynamic regulation and noise reduction. This shows that the system can be widely used in a variety of noise pollution scenarios, has strong practicality, can provide reliable solutions for noise control in different fields, and effectively improve the quality of the acoustic environment.

[0065] Example 2

[0066] Acoustic sensor network: It is composed of a large number of distributed acoustic sensor nodes, covering the areas that need to be monitored and managed. The sensor nodes use high-precision, wide-bandwidth microphones to accurately collect acoustic signals in the environment. Each node has independent data collection, preprocessing and wireless transmission functions, and can transmit the collected acoustic signals to the central processing unit in real time. The distribution of sensor nodes is optimized according to the characteristics of the monitoring area. For example, in industrial plants, they are mainly arranged in equipment concentrated areas and areas with frequent personnel activities; around major traffic arteries, they are evenly distributed along the road direction to ensure that noise information can be obtained comprehensively and accurately.

[0067] Central Processing Unit (CPU): As the core control and computing part of the system, it is responsible for receiving massive amounts of data from the acoustic sensor network. The unit is equipped with a high-performance processor, large-capacity memory and high-speed storage devices, and has powerful data processing capabilities. It runs noise source location algorithms, noise analysis programs and dynamic control decision algorithms, performs real-time analysis and processing of acoustic signals, and realizes noise source location, noise characteristic evaluation and the formulation of control strategies.

[0068] Dynamic control execution unit: According to the instructions of the central processing unit, it executes the corresponding noise reduction control operation. This unit contains various types of noise reduction equipment, such as intelligent sound insulation barriers with adjustable angles and heights, active noise reduction speaker arrays that can emit anti-phase sound waves, and ventilation system noise regulators that can control air volume and wind speed. These devices realize dynamic adjustment of working parameters through motor drive, signal generator control, etc., such as angle changes of sound insulation barriers, frequency and amplitude adjustments of active noise reduction speakers, and air volume adjustments of ventilation systems, so as to change the noise propagation path and reduce noise intensity.

[0069] Communication network: responsible for establishing a stable, high-speed data transmission link between the acoustic sensor network, the central processing unit, and the dynamic control execution unit. A combination of wired and wireless communication methods is used. For example, in an industrial environment, data can be transmitted between the sensor node and the central processing unit via industrial Ethernet or Wi-Fi; in some areas with wiring restrictions, such as along urban roads, wireless sensor networks (such as ZigBee, LoRa, etc.) can be used for data communication to ensure real-time and accurate data transmission and the normal operation of the system.

[0070] User interaction interface: Provides an intuitive and convenient operation platform for operators to set system parameters, monitor status and issue control instructions. Through the graphical interface, operators can view information such as the location of noise sources, noise intensity distribution, and the operating status of noise reduction equipment in real time. They can also manually adjust the noise reduction strategy according to actual needs, or view the system's historical data and operation reports, realizing human-computer interaction functions and improving the operability and management efficiency of the system.

[0071] Noise source localization model based on improved particle swarm optimization-support vector machine (IPSO-SVM)

[0072] Let x i =(x i1 , x i2 , …, x in ) is the noise signal feature vector collected by the i-th acoustic sensor (n is the feature dimension, which may include signal strength, frequency, phase and other information), y i is the corresponding noise source position label (which can be expressed as a two-dimensional or three-dimensional coordinate). The basic idea of ​​the support vector machine (SVM) is to separate data points of different categories as much as possible by constructing a hyperplane, and its decision function is:

[0073]

[0074] Among them, l is the number of training samples, α i is the Lagrange multiplier, K(x i , x) is the kernel function (common kernel functions include linear kernel, polynomial kernel, radial basis function (RBF) kernel, etc. Here we choose RBF kernel function K(x i , x) = exp(-γ||x i -x|| 2 ), γ is the kernel parameter), and b is the bias term.

[0075] When training the SVM model, you need to solve the optimization problem:

[0076]

[0077] Among them, C is the penalty parameter, which is used to control the degree of penalty for misclassified samples.

[0078] In order to improve the performance of the SVM model, the improved particle swarm optimization (IPSO) algorithm is used to optimize the parameters γ and C of the SVM. In the particle swarm optimization (PSO) algorithm, each particle represents a set of possible parameter solutions, and the position vector of the particle is represented by X i =(γ i , C i ), the velocity vector is V i =(υ γi , Ci ). The particle is based on its own experience (the individual optimal position P i ) and group experience (global optimal position G) to update its own position and speed. The update formula is as follows:

[0079]

[0080] Among them, k is the number of iterations, w is the inertia weight, c1 and c2 are acceleration coefficients, and r1 and r2 are random numbers uniformly distributed in the interval [0, 1].

[0081] In the IPSO algorithm, the following improvements are made to the traditional PSO algorithm:

[0082] οDynamic inertia weight adjustment: The inertia weight w is no longer a fixed value, but changes dynamically with the increase of the number of iterations. Initially, w is large, which is conducive to the particles searching the solution space in a larger range. As the number of iterations increases, w gradually decreases, allowing particles to perform fine searches in local areas. The specific calculation formula is: where w max and w min are the maximum and minimum values ​​of the inertia weight, respectively, k max is the maximum number of iterations.

[0083] Adaptive mutation strategy: In order to prevent particles from falling into the local optimal solution, an adaptive mutation strategy is introduced. In each iteration, the particle's fitness value and the average fitness value of the group are used to determine whether the particle needs to mutate. If the particle's fitness value is lower than a certain proportion of the group's average fitness value (such as 80%), the particle's position is mutated. The mutation method is to randomly generate a new value within the particle's parameter value range to replace the original parameter value, which can increase the diversity of particles and improve the algorithm's ability to jump out of the local optimal solution.

[0084] After optimizing the parameters of SVM through the IPSO algorithm, the optimal γ and C values ​​are obtained, thus building a noise source location model with better performance. Using this model, the location coordinates of the noise source can be predicted based on the noise signal feature vector collected by the acoustic sensor.

[0085] In the modeling and solving process, an acoustic sensor network is deployed in different types of noise environments (such as industrial noise environment, traffic noise environment, construction noise environment, etc.) to collect large-scale data. During the collection process, the location information of the sensor, the collection time, and environmental related parameters (such as temperature, humidity, air pressure, etc., which will affect the speed of sound and thus affect the accuracy of noise source positioning) are recorded synchronously. The collected original acoustic signal is preprocessed, including removing the DC component and normalizing the signal to eliminate the baseline drift and dimensional difference in the signal; the signal is filtered using digital filtering technology (such as Butterworth filter, Chebyshev filter, etc.) to remove high-frequency interference and low-frequency fluctuations in the signal; signal enhancement processing is performed, such as using wavelet transform technology to decompose and reconstruct the signal to improve the signal-to-noise ratio, ensure the quality and reliability of the data, and provide an accurate data basis for subsequent model training and positioning calculations. Extract feature vectors from the preprocessed acoustic signal. Feature selection includes signal time domain features (such as mean, variance, root mean square value, kurtosis, etc.), frequency domain features (such as spectrum center of gravity, spectrum bandwidth, harmonic factor, etc.) and time-frequency domain features (such as energy distribution after short-time Fourier transform, energy coefficient after wavelet packet decomposition, etc.). The extracted feature vectors and the corresponding noise source position labels (obtained through actual measurement or other precise positioning methods) form a training sample set. The training sample set is divided into training set, validation set and test set. The noise source location model based on IPSO-SVM is trained using the training set data. During the training process, the position and velocity of the particle swarm are first initialized, and the position and velocity of the particles are updated according to the rules of the improved particle swarm optimization algorithm (IPSO) to find the optimal SVM parameters. Then, the SVM model is trained using the optimized parameters, and the accuracy, mean square error and other evaluation indicators of the model on the validation set are calculated. According to the evaluation results, adjust the parameters of the IPSO algorithm (such as the value range of the inertia weight, the acceleration coefficient, etc.), and continue iterative training until the performance of the model on the validation set reaches stability or meets the preset training termination conditions (such as reaching the maximum number of iterations or the accuracy no longer improves). Finally, use the test set data to test the trained model to evaluate the generalization ability and positioning accuracy of the model.

[0086] In actual operation, the acoustic sensor network collects noise signals in real time, extracts feature vectors after preprocessing, and inputs them into the trained noise source positioning model based on IPSO-SVM to obtain the prediction results of the noise source location. The central processing unit generates a control plan based on the positioning results and the noise characteristics monitored in real time (by performing fast Fourier transform (FFT) on the acoustic signal to analyze the frequency components, intensity distribution and other information of the noise), combined with the preset noise reduction strategy library, using the dynamic control decision algorithm. For example, if the main frequency of the noise source located in a certain area is concentrated in the low frequency band and the intensity is large, the decision algorithm may instruct the dynamic control execution unit to adjust the operating frequency of the active noise reduction speaker array to the low frequency band, increase the emission amplitude, and adjust the angle of the intelligent sound insulation barrier to better block the propagation of low-frequency noise. During the control process, the system continuously monitors the noise changes, compares the actual noise reduction effect with the expected target, and adjusts the control strategy in real time according to the feedback information to achieve dynamic optimization of noise reduction.

[0087] Test scenario selection, industrial manufacturing factory: The production workshop of a large machinery manufacturing factory was selected as the test area. There are many large stamping machines, CNC machine tools, fans and other equipment running at the same time in the workshop, generating high-intensity, wide-band and dynamically changing noise. The workshop space is large, the equipment layout is complex, there are a large number of reflective surfaces, and the acoustic environment is extremely complex. Urban traffic arterial road: Select a city arterial road with large traffic volume and busy traffic for testing. Various types of motor vehicles (cars, buses, trucks, etc.) frequently travel on this road section. The noise sources mainly come from vehicle engines, tire friction with the ground, and horn sounds. Traffic flow changes significantly at different times (peak hours and non-peak hours), and the noise characteristics change accordingly. Construction site: A construction site where a high-rise building is under construction is used as the test point. There are many noise sources involved in the construction process, such as pile drivers, tower cranes, concrete mixers, electric saws, etc. The noise intensity is high and has sudden and intermittent characteristics. At the same time, the construction site environment is open, surrounded by buildings and obstacles, and the sound propagation is complex. School playground: The school playground was selected as a relatively simple noise test scene. The main noise sources were the noise generated by students' activities during recess and the noise of vehicles passing by occasionally. This scene was used to compare the performance of the system in a relatively simple environment.

[0088] Test Scheme Scheme 1: The acoustic governance noise reduction management system of the present invention reasonably arranges the acoustic sensor network in the test area according to the system design requirements to ensure full coverage of the noise source area. The sensor nodes are connected to the central processing unit by wired or wireless means to realize real-time data transmission. The central processing unit runs the noise source positioning model and dynamic control decision algorithm based on the improved particle swarm optimization-support vector machine (IPSO-SVM). The dynamic control execution unit controls the intelligent sound insulation barrier, active noise reduction speaker array, ventilation system noise regulator and other equipment to perform dynamic noise reduction operations according to the instructions. Scheme 2: Traditional fixed noise reduction measures combination (sound insulation barrier + silencer) Install a sound insulation barrier of fixed height and angle around the test area, and select sound insulation materials of conventional specifications and materials. For noise-generating equipment (such as some equipment in the factory and some machinery on the construction site), install a silencer, and the silencer parameters are pre-set according to the equipment type and cannot be adjusted. During the entire test process, the noise reduction measures remain fixed and do not have real-time monitoring and dynamic control functions. Solution 3: A regular noise reduction system based on manual experience. The acoustic sensor network is also deployed, but it is only used to collect noise data regularly (such as every hour or every half hour), without real-time analysis and positioning. Based on past experience, the staff manually adjusts the angle of the sound barrier (if adjustable), the working status of the muffler and other noise reduction measures in specific time periods (such as according to factory production shifts, peak traffic hours, etc.). The control process is relatively rough and lacks real-time and precision.

[0089] Monitoring point arrangement: In each test scenario, multiple noise monitoring points are evenly arranged according to the size of the area and the distribution of noise sources. For example, 20-30 monitoring points are arranged in the workshop of an industrial manufacturing factory, covering locations such as around equipment, workers' operating areas, and workshop corners; a monitoring point is set up every 50-100 meters on both sides of urban traffic arteries, with a total of 10-15 monitoring points; 15-20 monitoring points are arranged on construction sites according to the construction area and surrounding environment; 5-10 monitoring points are arranged on school playgrounds. All monitoring points use high-precision sound level meters (such as sound level meters that comply with the IEC 61672-1 standard) for noise measurement. The sound level meter is installed on a tripod at a height of 1.5-2 meters from the ground to avoid the influence of surrounding objects on the measurement results.

[0090] According to data collection and recording, a data acquisition device (connected to a sound level meter) is used to record in real time the noise data of each monitoring point before and after the implementation of noise reduction measures, including parameters such as equivalent continuous sound level (Leq), sound pressure level (SPL), and spectrum characteristics. The data collection frequency is 1-5 times per second, and the continuous collection time depends on different scenarios, such as 8-10 hours for industrial scenarios (covering a complete production shift), 24 hours for traffic scenarios (a complete day), 6-8 hours for construction sites based on construction time, and 2-3 hours for school playgrounds (inter-class activity time). Record the equipment operating parameters (such as the adjustment parameters of the intelligent equipment in the system of the present invention, the initial setting parameters of the equipment in the traditional scheme), environmental parameters (such as temperature, humidity, wind speed, etc., recorded every 15-30 minutes) and the control operation records (including time, adjustment content, etc.) in the manual control scheme during the implementation of each test scheme.

[0091] The test schedule is arranged so that the tests of each test scenario are conducted in different time periods to cover different working conditions and environmental conditions. For example, the tests of industrial manufacturing plants are conducted during the normal production shift during the day (8 am - 5 pm) and the overtime production period at night (7 pm - 2 am); the tests of urban traffic arteries are conducted for 24 hours, covering different traffic flow periods such as the morning peak (7 am - 9 am), the flat peak (9 am - 4 pm) and the evening peak (5 pm - 7 pm); the tests of construction sites are conducted at different construction stages according to the construction schedule, and each test lasts for one working day; the tests of school playgrounds are conducted during the morning break, lunch break and afternoon break on normal working days.

[0092] Test indicators, noise reduction effect evaluation:

[0093] 1. Equivalent continuous sound level (Leq) noise reduction: Calculate the Leq difference before and after the implementation of noise reduction measures at each monitoring point, that is, the noise reduction amount. Compare the noise reduction amounts of different solutions at each monitoring point and in different test scenarios to evaluate their overall noise reduction effect.

[0094] 2. Spectral noise reduction effect: Perform spectrum analysis on the noise signal before and after noise reduction to observe the noise attenuation in different frequency bands. Calculate the sound pressure level reduction value in each frequency band to evaluate the system's ability to control noise of different frequencies. For example, for industrial noise, focus on the noise reduction effect in the mid- and low-frequency bands (50-500Hz); for traffic noise, analyze the changes in the mid- and high-frequency bands (500-5000Hz); for construction noise, examine the improvement of the spectrum characteristics of the entire frequency band.

[0095] 2. Adaptability Assessment

[0096] 1. Response time: In industrial manufacturing plants and construction sites, sudden changes in the intensity and characteristics of noise sources are created by artificially changing the operating status of equipment (such as starting or stopping a large piece of equipment) or construction procedures (such as switching from the piling stage to the concrete pouring stage); on urban traffic arteries, observe the impact of natural changes in traffic flow during peak and non-peak hours on the system. Record the time interval from when the noise source changes to when the system takes effective control measures (such as when the equipment in the system of the present invention starts to adjust the working parameters, and in other solutions, when there are adjustable equipment, human intervention and adjustment begins) as the system's response time.

[0097] 2. Stability and recovery time: After the system makes control measures, continue to monitor noise changes and record the time required from the start of control to the noise level and recovery to a stable state (that is, the noise fluctuation range is less than the set threshold within a certain period of time, such as Leq fluctuation is less than 1-2dB). Evaluate the system's stability recovery ability after responding to changes in noise sources.

[0098] 3. Resource Utilization Efficiency Assessment

[0099] 1. Equipment cost-effectiveness ratio: Count the purchase cost of the noise reduction equipment used in each test scheme (including sensors, noise barriers, mufflers, active noise reduction equipment, etc.), and calculate the equipment cost-effectiveness ratio (, where is the total equipment cost). Compare the cost-effectiveness ratios of different schemes and evaluate the rationality of equipment cost investment to achieve the same noise reduction effect.

[0100] 2. Energy efficiency ratio: Use a power monitor to measure the energy consumption of each solution during operation (including equipment operation power and control system energy consumption), and calculate the energy efficiency ratio (, where is the total energy consumption). Analyze the energy consumption of different solutions during the noise reduction process and evaluate the energy efficiency of the system.

[0101] 3. Comparative test data and results

[0102] 1. Industrial manufacturing factory scene

[0103] 1. Noise reduction effect

[0104] 1. The system of the present invention: The Leq noise reduction at different monitoring points in the workshop reaches an average of 15-25dB, and the noise reduction at some monitoring points close to the noise source and optimized and regulated can reach more than 30dB. Spectral analysis shows that the noise reduction effect is particularly significant in the low and medium frequency bands (50-500Hz), with the sound pressure level reduced by 10-20dB, and the noise reduction in the high frequency band (500-5000Hz) can also reach 5-10dB, which effectively improves the overall acoustic environment in the workshop and reduces the noise level to a more comfortable range (Leq is around 70-80dB).

[0105] 2. Combination of traditional fixed noise reduction measures: The average Leq noise reduction is 5-10dB, and the noise reduction effects at different monitoring points vary greatly. The noise reduction effect at monitoring points close to the sound barrier and the installation location of the silencer is better, but the noise reduction effect is limited in areas inside the workshop far away from the noise reduction equipment. In terms of the spectrum, the high-frequency noise is mainly attenuated to a certain extent (the sound pressure level in the 500-5000Hz frequency band is reduced by 3-5dB), and the control ability of the medium and low-frequency noise is weak, and the overall noise level in the workshop is still high (Leq is around 85-95dB).

[0106] 3. Regularly controlled noise reduction system based on manual experience: The average Leq noise reduction is 8-12dB. Since manual control cannot adapt to the dynamic changes of noise sources in a timely and accurate manner, the noise reduction effect is unstable. In some periods of time, due to improper control timing or inaccurate adjustment parameters, the noise reduction effect may be even worse than traditional fixed measures. The spectrum characteristics are not significantly improved, the noise attenuation of each frequency band fluctuates greatly, and the noise level in the workshop fluctuates between 80-90dB.

[0107] 2. Adaptability

[0108] 1. The system of the present invention has an average response time of 1-3 seconds, which can quickly detect changes in noise sources (such as equipment startup or shutdown) and initiate control measures. The stability recovery time is short, and the noise level can generally be restored to stability within 10-20 seconds, effectively coping with the challenges brought about by frequent startup and shutdown of equipment and changes in working conditions in the factory.

[0109] 2. Traditional fixed noise reduction measures: It does not have adaptive capabilities and cannot respond to changes in noise sources in real time. The response time and stability recovery time are both meaningless (can be considered infinite), and it cannot provide additional noise reduction improvements when the noise source changes.

[0110] 3. Regularly controlled noise reduction system based on manual experience: The response time depends on the timeliness of manual intervention, which is generally about 5-15 minutes (the time interval between manual discovery of noise changes and operation). The stability recovery time is long, usually 30-60 minutes to stabilize the noise level, and the noise fluctuates greatly during the recovery process, which cannot effectively adapt to the complex and changeable acoustic environment of the factory.

[0111] 3. Resource Utilization Efficiency

[0112] The system of the present invention has a relatively high equipment cost, mainly due to the use of intelligent sensor networks, active noise reduction equipment and complex control systems, but the equipment cost-effectiveness ratio is relatively high, reaching 0.8-1.2dB / 10,000 yuan (that is, 0.8-1.2dB of noise reduction can be obtained for every 10,000 yuan of equipment cost invested). The energy efficiency ratio is 0.5-0.8dB / kWh (0.5-0.8dB of noise reduction can be achieved for every 1kWh of electricity consumed). Although the energy consumption is relatively high, considering its significant noise reduction effect, the resource utilization efficiency is still advantageous.

[0113] The acoustic control and noise reduction management system of the present invention has broad application prospects, especially in the fields of urban transportation, industrial production, construction, etc. Through real-time monitoring and dynamic regulation, the system can effectively reduce noise pollution and improve people's living and working environment. In the future, the system can be further expanded to emerging fields such as smart homes and smart cities, providing technical support for more intelligent and environmentally friendly noise control.

Claims

1. A sound source localization and management system based on multi-source information fusion, characterized in that: It includes a sensor network unit, a data processing center, a dynamic control execution unit and a user interaction interface. The sensor network unit is used to collect noise-related data and transmit it to the data processing center. The data processing center uses a noise source positioning model based on multi-source information fusion and Bayesian estimation to locate and analyze the noise source, and formulates a control strategy based on the positioning result and noise characteristics. The dynamic control execution unit executes the control strategy to reduce noise. The user interaction interface is used for system operation and information display; According to the Bayesian formula, the posterior probability distribution P(L|x a (t), x v (t), w(t)) is: The location estimate of the noise source is determined by solving the maximum value of the posterior probability distribution Right now:

2. The system according to claim 1, characterized in that In the noise source localization model based on multi-source information fusion and Bayesian estimation, the noise signal collected by the acoustic sensor is preprocessed to extract the acoustic feature vector, the vibration signal collected by the vibration sensor is preprocessed to extract the vibration feature vector, the meteorological parameters measured by the meteorological sensor are used to calculate the sound speed, the Bayesian estimation method is used to fuse the acoustic, vibration and meteorological information to determine the location of the noise source, and the likelihood function is calculated based on the acoustic propagation model and the influence of meteorological conditions on noise propagation.

3. The system according to claim 1, characterized in that The acoustic feature vector includes energy, power spectrum density and zero-crossing rate of the noise signal, and the vibration feature vector includes vibration frequency, amplitude and acceleration.

4. The system according to claim 3, characterized in that The acoustic propagation model calculates the sound wave propagation path and attenuation based on the ray tracing principle, and the vibration propagation model is constructed according to the vibration propagation theory and the structural characteristics of the equipment. The acoustic propagation model calculates the sound wave propagation path and attenuation based on the ray tracing principle. Assuming the speed of sound is c(t), the calculation formula is: Where c0 is the speed of sound under standard conditions, T0 is the standard temperature, and the sound wave propagation path is calculated based on the speed of sound, the noise source position L, and the sensor position, thereby determining P(x a (t)|L).

5. The system according to claim 1, characterized in that The data processing center adopts the expectation maximization algorithm to solve the optimization problem in the noise source localization model based on multi-source information fusion and Bayesian estimation, and iteratively updates the model parameters and the noise source position estimation by alternately executing the expectation step and the maximization step until the convergence condition is met.

6. The system according to claim 1, characterized in that The dynamic control execution unit includes a movable active noise reduction device, an intelligent sound insulation barrier and a ventilation system noise controller, which can adjust working parameters in real time according to instructions from a data processing center to achieve noise reduction.

7. The system according to claim 1, characterized in that The system conducts model training and optimization through long-term data collection in different noise environment scenarios. The collected data includes acoustic, vibration and meteorological data, and performs real-time preprocessing on the collected data, including outlier removal, data interpolation and data synchronization.

8. The system according to claim 1, characterized in that When evaluating the noise reduction effect, noise monitoring points are arranged in multiple test areas to measure the difference in equivalent continuous sound levels before and after noise reduction and the change in spectrum. The positioning accuracy is tested by simulating noise sources. The system adaptability is evaluated by changing the state of noise sources. At the same time, relevant parameters such as equipment cost, energy consumption, installation and maintenance difficulty are recorded for comprehensive system evaluation. Where T is the measurement time, L(t) is the sound pressure level that changes with time; the positioning accuracy is tested by simulating the noise source, and the positioning error is calculated using the Euclidean distance formula, that is, (where (x est ,y est , z est ) is the estimated noise source position, (x true ,y true , z true ) is the actual noise source location); the system adaptability is evaluated by changing the noise source state, and related parameters such as equipment cost, energy consumption, installation and maintenance difficulty are recorded for comprehensive system evaluation.

9. A noise source location and noise reduction control method based on multi-source information fusion, characterized in that: The following steps are involved: Step 1: Collect data using a sensor network, which includes acoustic sensors, vibration sensors, and meteorological sensors. Collect noise signals in the environment, vibration signals of noise source equipment and meteorological parameter information, and pre-process each type of data collected Processing operations include removing interference components from noise signals, extracting effective features from vibration signals, and calibrating meteorological parameter data. Transmit the processed data to the data processing center; Step 2: In the data processing center, noise source location is performed based on multi-source information fusion technology; Step 3: According to the noise source location results and the noise characteristics of real-time monitoring, combined with the preset noise reduction strategy library, formulate dynamic noise reduction control Strategy, the noise reduction strategy library contains a variety of control schemes for different types of noise sources, different locations and different noise characteristics; Step 4: Send the formulated control strategy to the dynamic control execution unit, which will control the corresponding noise reduction equipment according to the strategy instructions. During the control process, we continuously collect feedback information about the noise reduction effect, including changes in noise intensity and spectrum distribution, and then Information is used to adjust control strategies in real time to ensure optimal noise reduction effects.

10. The system according to claim 9, characterized in that In the step 2, firstly, according to the noise signal characteristics collected by the acoustic sensor, the vibration signal characteristics detected by the vibration sensor and the meteorological parameter information obtained by the meteorological sensor, a noise signal propagation model, a vibration signal propagation model and a noise source position probability model considering the meteorological influence are established respectively, and then the position estimation of the noise source is determined by fusing the information of the three models using the Bayesian estimation method. In this process, the prior probability distribution of the noise source position is set according to prior knowledge and historical data, and the probability of collecting multi-source data under a given noise source position is evaluated by calculating the likelihood function, and finally the noise source position estimation result based on the posterior probability distribution is obtained.

Images