System for testing noise disturbance degree in high-speed train tunnel based on electroencephalogram signals

Abstract

The invention discloses a test system for noise disturbance degree in a high-speed train tunnel based on electroencephalogram signals. The test system comprises a noise acquisition and separation module, a dynamic noise simulation module and a multi-modal evaluation module, wherein the noise acquisition and separation module is used for separating aerodynamic noise, wheeltrack noise and cavity resonance noise in a tunnel environment in real time; the dynamic noise simulation module is used for generating an adjustable sound field based on the parameterized noise database and supporting noise switching and dynamic reverberation control; the multi-modal evaluation module is used for calculating a dynamic annoyance index DFI according to the physiological signals, the behavior data and subjective feedback and outputting a quantitative evaluation result; according to the method, the subconsciousness stress intensity caused by noise is quantified through EEG (electroencephalogram), eye movement tracking and other multi-dimensional physiological signals, the memory deviation defect of a subjective questionnaire is made up, and total factor evaluation of tunnel noise dynamic characteristics is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-speed trains, and in particular to a high-speed train tunnel noise annoyance degree test system based on electroencephalogram signals. BACKGROUND

[0002] With the extension of high-speed railway network to complex terrain, the proportion of tunnels has increased significantly (such as the tunnel mileage of more than 40% in the "eight vertical and eight horizontal" high-speed rail network), and the unique acoustic environment in the tunnel makes the traditional noise evaluation method face serious challenges. As a semi-closed space, the sound field characteristics of the tunnel are essentially different from those of the open environment: the air pressure pulse wave (micro-pressure wave) generated when the high-speed train enters can cause a transient explosion sound of more than 120 dB; the rigid wall surface causes the standing wave effect of low-frequency noise (63-250 Hz), and the sound pressure level is increased by 3-5 times compared with the open section; the multiple reflections of wheel-rail and aerodynamic noise form a prolonged reverberation time (>2s), which aggravates the auditory fatigue of the driver and crew. The existing noise evaluation system relies on equivalent continuous sound level (Leq) and A-weighted measurement, which cannot capture the dynamic time-domain characteristics and spatial perception effects (such as sound field directionality suppression) of tunnel noise, resulting in a large deviation in annoyance prediction.

[0003] The existing subjective evaluation method (such as the ICBEN standard questionnaire) has significant limitations in the tunnel scene: the passenger's annoyance emotion to the transient high-pressure noise decays by 37% within 10 minutes after the event (Cambridge University psychoacoustic experiment data), and the traditional laboratory simulation can only reproduce steady-state noise, lacking dynamic sound exposure simulation of the whole process of tunnel passing. At the same time, physiological studies have confirmed that tunnel noise can trigger subconscious stress reactions (such as anxiety caused by amygdala activation), but the existing technology does not integrate neural feedback mechanisms. SUMMARY

[0004] To solve the problems in the prior art, the purpose of the present application is to provide a high-speed train tunnel noise annoyance degree test system based on electroencephalogram signals. The present application quantifies the subconscious stress intensity caused by noise through multi-dimensional physiological signals such as EEG (electroencephalogram), eye tracking, etc., makes up for the recall bias defects of subjective questionnaire, and realizes the full-factor evaluation of the dynamic characteristics of tunnel noise.

[0005] To achieve the above-mentioned purpose, the technical solution adopted by the present application is: a high-speed train tunnel noise annoyance degree test system based on electroencephalogram signals, comprising: a noise collection and separation module, a dynamic noise simulation module, and a multi-modal evaluation module; wherein: The noise collection and separation module: collects the original signal, separates the aerodynamic noise 200Hz-5kHz wideband spectrum, the wheel-rail noise 63-500Hz line spectrum, and the cavity noise narrowband peak value through the improved Fast ICA blind source separation algorithm combined with feature recognition technology; The dynamic noise simulation module uses parametric sound field modeling technology and a real-time sound field synthesis system to achieve accurate sound field reconstruction using an improved ray acoustics / FDTD hybrid algorithm, and supports noise switching and dynamic reverberation control. The multimodal assessment module integrates physical acoustic indicators, physiological and psychological indicators, and behavioral indicators, and uses the dynamic entropy weight method to adaptively adjust the weights, calculates the dynamic annoyance index (DFI) on a scale of 0-100 in real time, and outputs quantitative assessment results.

[0006] As a further improvement of the present invention, the noise acquisition and separation module adopts a 16-channel ring microphone array, and combines the improved Fast ICA blind source separation algorithm to introduce a time-frequency domain joint analysis method. Wavelet packet transform preprocessing and time-frequency masking technology are used to improve the signal-to-noise ratio. Physical characteristic constraints of aerodynamic noise are added, including frequency band limitation and time-domain sparsity penalty term, to reduce separation error and separate aerodynamic noise.

[0007] As a further improvement of the present invention, the noise database of the dynamic noise simulation module includes 12 types of noise prototypes and supports a dynamic range of 65-115dB sound pressure level.

[0008] As a further improvement of the present invention, the DFI calculation model of the multimodal evaluation module includes: physical layer indicators, physiological layer indicators, and behavioral layer indicators.

[0009] As a further improvement of the present invention, the weight allocation of the DFI calculation model is dynamically adjusted using the entropy weight method, and the information entropy is calculated based on the dispersion of each indicator. Through formula The weights are dynamically allocated inversely proportional to the entropy value, and are updated once every preset time. When the variation of a certain indicator data increases, its entropy value decreases and its weight automatically increases, ensuring the model's sensitive response to sudden disturbances. The weights of the physical layer, physiological layer and behavioral layer are 30%, 40% and 30%, respectively.

[0010] As a further improvement of the present invention, the testing system utilizes real-time feedback control to dynamically adjust the noise exposure parameters based on the DFI results. Based on the deviation between the real-time DFI value and a preset threshold, the sound pressure level adjustment is calculated using a PID fuzzy algorithm. Furthermore, by combining the weights of each indicator layer, the spectrum is dynamically reshaped, forming a closed-loop testing process.

[0011] As a further improvement of the present invention, the dynamic noise simulation module uses an FIR filter to correct the frequency response in real time and adopts a multi-level optimization design: by analyzing the 1 / 24 octave band spectrum of the noise signal in real time, the deviation from the target frequency response curve is dynamically calculated; and the LMS adaptive algorithm is used to update the filter coefficients.

[0012] As a further improvement of the application, the physiological signal processing of the multi-modal evaluation module adopts ICA denoising, wavelet packet decomposition and moving average filtering, first separates the 64-channel EEG signal through independent component analysis (ICA), identifies and removes the electrooculogram / electromyogram artifacts using the correlation coefficient threshold method; second, decomposes the brain waves by 5 layers using db4 wavelet packet, and retains the effective components through threshold shrinkage; finally, the GSR signal is subjected to sliding average filtering.

[0013] As a further improvement of the application, the test system integrates VR headsets and 3D audio equipment to restore the three-dimensional sound field in the tunnel.

[0014] The beneficial effects of the present application are: The present application combines VR virtual tunnel environment + 3D spatial audio to realize sound-visual multi-channel synchronous simulation, breaking through the ecological validity limit of laboratory static test; through the mobile acoustic array, the three-dimensional sound field characteristics (including transient air pressure wave, low-frequency standing wave, high-frequency howling) in the tunnel are restored, solving the problem of insufficient capture of dynamic events in traditional acoustic measurement; through multi-dimensional physiological signals such as EEG (electroencephalogram) and eye movement tracking, the subconscious stress intensity caused by noise is quantified, the recall bias defect of subjective questionnaire is made up, and full-factor evaluation of tunnel noise dynamic characteristics is realized. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A schematic diagram of typical sound field data collected in the embodiment of the present application; Figure 2 A 64-channel electroencephalogram signal acquisition point distribution diagram in the embodiment of the present application; Figure 3 A 64-channel electroencephalogram signal processing schematic diagram in the embodiment of the present application; Figure 4 A noise annoyance score change diagram in the embodiment of the present application; Figure 5 A test method technical route diagram of the present application. DETAILED DESCRIPTION

[0016] The embodiments of the present application will be described in detail below with reference to the accompanying drawings. EMBODIMENT

[0017] As Figures 1-5 shown, a tunnel noise dynamic immersive annoyance evaluation system, comprising: a noise collection and separation module for separating the aerodynamic noise, wheel-rail noise and cavity resonance noise in the tunnel environment in real time; a dynamic noise simulation module for generating adjustable sound field based on parameterized noise database, supporting noise switching and dynamic reverberation control; a multi-modal evaluation module for calculating dynamic annoyance index (DFI) through physiological signals, behavior data and subjective feedback, and outputting quantitative evaluation results.

[0018] The noise acquisition and separation module uses a 16-channel ring microphone array, combined with an improved blind source separation algorithm, to achieve aerodynamic noise separation accuracy of ±1.5dB (200Hz-5kHz frequency band).

[0019] The noise database of the dynamic noise simulation module contains 12 types of noise prototypes and supports a dynamic range of 65-115dB sound pressure level.

[0020] The DFI calculation model of the multimodal assessment module includes: physical layer indicators, physiological layer indicators, and behavioral layer indicators.

[0021] The dynamic noise simulation module uses an FIR filter to correct the frequency response in real time, with a noise switching transition time of <10ms and a reverberation time control accuracy of ±0.1s (adjustable from 0.5 to 2.5s).

[0022] The system integrates a VR headset and 3D audio equipment to recreate the three-dimensional sound field of the tunnel.

[0023] The physiological signal processing of the multimodal assessment module employs ICA denoising, wavelet packet decomposition (EEG), and moving average filtering (GSR), with a data acquisition delay of <2ms.

[0024] The weight allocation of the DFI model is dynamically adjusted using the entropy weight method, with the weights of the physical layer, physiological layer, and behavioral layer being 30%, 40%, and 30%, respectively.

[0025] The system supports real-time feedback control, dynamically adjusting noise exposure parameters based on DFI results to form a closed-loop testing process.

[0026] This embodiment relies on the collaborative work of three main modules: the noise acquisition and separation module uses a spherical microphone array and deep learning algorithms to achieve high-fidelity separation and parameterized characterization of noise components; the dynamic noise simulation module constructs an experimental environment with high ecological validity through 3D audio rendering, virtual reality technology, and a moving acoustic array; and the multimodal assessment module integrates multimodal physiological and psychological data to construct a dynamic annoyance index model. These three modules form a closed-loop system, in which noise parameters can be dynamically adjusted based on real-time feedback from subjects, thereby achieving accurate assessment of tunnel noise annoyance.

[0027] The core principle of this embodiment is as follows: The goal is to establish a precise mapping relationship between "physical acoustic characteristics and physiological and psychological responses." Raw noise signals from a tunnel environment are acquired using a high-precision acoustic acquisition system. Deep neural networks are then employed for noise component separation and feature extraction, decomposing the complex tunnel noise into components with clear physical meaning (aerodynamic noise, wheel-rail noise, cavity resonance noise, etc.). Based on this, a dynamically adjustable virtual noise environment is constructed through parametric modeling, accurately reproducing the time-frequency characteristics of tunnel noise in an immersive experimental setting. The system monitors the multidimensional responses of subjects under noise exposure in real time, including physiological indicators such as EEG and skin conductance, behavioral characteristics such as head movements, and subjective evaluation data. Finally, machine learning algorithms are used to establish a quantitative predictive model from noise physical parameters to human annoyance levels.

[0028] The advantages of using the above methods are: high-precision noise acquisition and reconstruction, immersive exposure simulation, and multi-dimensional physiological and psychological monitoring.

[0029] 1. High-precision noise acquisition and reconstruction (noise acquisition and modeling subsystem) A 32-channel spherical microphone array was used to acquire three-dimensional sound field data in the tunnel at a sampling rate of 192 kHz, with a dynamic range of 30-140 dB.

[0030] Develop a deep learning-based noise separation algorithm that can accurately separate aerodynamic noise components, wheel-rail noise components, cavity resonance noise, and establish a parameterized noise database.

[0031] 2. Immersive Exposure Simulation (Immersive Simulation Subsystem) The high-fidelity 3D audio system and 4K VR headset (120Hz refresh rate) can render the acoustic characteristics of the tunnel in real time (reverberation time adjustable from 0.5 to 2.5 seconds) to simulate the entire process of a train passing through (acceleration, constant speed, deceleration). 3. Multi-dimensional physiological and psychological monitoring (assessment and feedback subsystem) Accurate quantification of tunnel noise annoyance is achieved through multimodal data fusion. A three-dimensional assessment system, encompassing physical, physiological, and behavioral aspects, is constructed using a 64-channel EEG combined with head motion tracking and interactive behavior recording in a VR environment. An innovative Dynamic Annoyance Index (DFI) model is employed, with a weighted algorithm (30% physical indicators, 40% physiological indicators, and 30% behavioral indicators) to output quantified results in real time. This subsystem not only supports real-time data visualization but also dynamically adjusts test parameters based on monitoring results, forming a closed-loop assessment process of "stimulus-response-optimization," ultimately generating a comprehensive assessment report including noise component analysis and annoyance level.

[0032] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A high-speed train tunnel noise annoyance degree test system based on electroencephalogram signals, characterized in that, Comprise: Noise acquisition and separation module, dynamic noise simulation module and multi-modal evaluation module; wherein: Noise acquisition and separation module: collect original signals, separate aerodynamic noise 200Hz-5kHz broadband spectrum, wheel-rail noise 63-500Hz line spectrum and cavity noise narrowband peak through improved Fast ICA blind source separation algorithm combined with feature recognition technology; Dynamic noise simulation module, through parameterized sound field modeling technology and real-time sound field synthesis system, uses improved ray acoustics / FDTD hybrid algorithm to realize accurate reconstruction of sound field, supports noise switching and dynamic reverberation control; Multi-modal evaluation module, through fusion of physical acoustics index, physiological and psychological index and behavior index, uses dynamic entropy weight method to adaptively adjust weight, calculates dynamic annoyance index DFI of 0-100 point system in real time, and outputs quantitative evaluation results. 2.The EEG-based test system for the annoyance degree of the noise in the tunnel of the high-speed train according to claim 1, characterized in that, The noise acquisition and separation module uses a 16-channel ring microphone array, introduces time-frequency domain joint analysis method through improved Fast ICA blind source separation algorithm, uses wavelet packet transform preprocessing and time-frequency masking technology to improve signal-to-noise ratio; adds physical feature constraint conditions of aerodynamic noise, including frequency band limitation and time domain sparsity penalty term, reduces separation error, and separates aerodynamic noise. 3.The EEG-based test system for the annoyance degree of the noise in the high-speed train tunnel according to claim 1, wherein, The noise database of the dynamic noise simulation module includes 12 types of noise prototypes, and supports 65-115dB sound pressure level dynamic range. 4.The EEG-based test system for the annoyance degree of the noise in the tunnel of the high-speed train according to claim 1, wherein, The DFI calculation model of the multi-modal evaluation module includes: physical layer index, physiological layer index and behavior layer index.

5. The EEG-based test system for the annoyance degree of the noise in the tunnel of the high-speed train according to claim 4, characterized in that, The weight distribution of the DFI calculation model is dynamically adjusted by using an entropy weight method, and information entropy is calculated based on the dispersion degree of each index The weight is inversely proportional to the entropy value and is dynamically distributed by a formula , and each grid is updated once every preset time When the variation of certain index data increases, the entropy value decreases, and the weight automatically increases, ensuring the sensitive response of the model to sudden interference, and the weights of the physical layer, physiological layer and behavior layer are 30%, 40% and 30% respectively.

6. The EEG-based test system for the annoyance degree of the noise in the tunnel of the high-speed train according to claim 5, characterized in that, The test system utilizes real-time feedback control to dynamically adjust noise exposure parameters according to DFI results, and calculates a sound pressure level adjustment amount through a PID fuzzy algorithm based on the deviation of a real-time DFI value from a preset threshold value: And a closed-loop test process is formed by combining the index layer weights to realize dynamic spectrum reshaping. 7.The EEG-based test system for the annoyance degree of the noise in the tunnel of the high-speed train according to claim 1, wherein, The dynamic noise simulation module uses FIR filter to correct frequency response in real time, and uses multi-stage optimization design: through real-time analysis of 1 / 24 octave spectrum of noise signal, the deviation from target frequency response curve is calculated dynamically; LMS adaptive algorithm is used to update filter coefficients. 8.The EEG-based test system for the annoyance degree of the noise in the tunnel of the high-speed train according to claim 1, wherein, The physiological signal processing of the multi-modal evaluation module uses ICA denoising, wavelet packet decomposition and moving average filtering: first, separate 64-channel EEG signals through independent component analysis ICA, and identify and remove electrooculogram / electromyogram artifacts using correlation coefficient threshold method; second, decompose brain waves by 5 layers using db4 wavelet packet, and retain effective components through threshold shrinkage; finally, sliding average filtering is performed on GSR signal. 9.The EEG-based test system for the annoyance degree of the noise in the tunnel of the high-speed train according to claim 1, wherein, The test system integrates VR head-mounted display and 3D audio equipment, and restores three-dimensional sound field of tunnel.

Images