Urban comprehensive traffic noise dynamic monitoring and environmental impact analysis model optimization method and system

Abstract

The application relates to the field of urban noise monitoring and artificial intelligence technology, and specifically discloses a method and system for optimizing a dynamic monitoring and environmental impact analysis model of urban comprehensive traffic noise, which comprises the following steps: acquiring high-precision map data of a target area and constructing a three-dimensional digital environment; acquiring real-time multi-source traffic flow data of the target area; constructing an intelligent agent model based on a physical information neural network; determining noise monitoring points according to the distribution of sensitive targets and standard layout rules, and acquiring measured noise data; comparing simulated noise data output by the intelligent agent model with the measured noise data in multiple dimensions, calculating model errors and generating model verification parameters; iteratively training the intelligent agent model through a self-supervised learning method based on the model verification parameters to optimize the operation parameters of the intelligent agent model; and judging whether the noise at the sensitive target exceeds a preset warning threshold based on the simulation results of the optimized intelligent agent model, and if so, generating and pushing traffic control suggestions.

Description

Technical Field

[0001] This disclosure generally relates to the fields of urban noise monitoring and artificial intelligence technology, and specifically to a method and system for optimizing a dynamic monitoring and environmental impact analysis model for urban integrated traffic noise. Background Technology

[0002] With the acceleration of urbanization and the continuous growth of motor vehicle ownership, road traffic noise has become one of the major sources of environmental pollution affecting the quality of life of urban residents. Schools, hospitals, residential areas, and other noise-sensitive locations have high requirements for noise control. Currently, the monitoring, simulation, and management systems for urban road traffic noise still have significant technical shortcomings. In particular, most existing simulation systems are "open-loop" architectures, which cannot use measured data to back-optimize the core computational logic of the model. As the environment changes, the accuracy of the model will continue to decline, losing its long-term predictive ability. Summary of the Invention

[0003] In view of the above-mentioned defects or deficiencies in the existing technology, it is desirable to provide a method and system for optimizing the dynamic monitoring and environmental impact analysis model of urban integrated traffic noise. Through a closed-loop mechanism of "simulation-measurement-verification-optimization", the model can continuously adapt to the dynamically changing environment. The optimized model can be linked with the traffic control system to achieve forward-looking early warning and automatic control of noise pollution.

[0004] In a first aspect, embodiments of this application provide a method for optimizing a dynamic monitoring and environmental impact analysis model for urban integrated traffic noise, including:

[0005] Acquire high-precision map data of the target area, construct a three-dimensional digital environment, and automatically identify and label sensitive targets in the acoustic environment based on a pre-set sensitive target feature library;

[0006] Access and analyze the acoustic environment quality standards, extract the noise limit standards and technical constraints, and transform them into standardized data operation rules;

[0007] Real-time acquisition of multi-source traffic flow data for the target area;

[0008] A smart agent model based on a physical information neural network is constructed. The smart agent model is driven by the multi-source traffic flow data and constrained by the three-dimensional spatial digital environment. It dynamically simulates and deduces the three-dimensional spatial distribution intensity of the noise field in the target area and its impact on sensitive targets.

[0009] Based on the distribution of the sensitive targets and the standard layout rules, noise monitoring points are determined, and measured noise data are obtained.

[0010] The simulated noise data output by the intelligent agent model is compared with the measured noise data in multiple dimensions to calculate the model error and generate model verification parameters.

[0011] Based on the model validation parameters, the agent model is iteratively trained using a self-supervised learning method to optimize its computational parameters.

[0012] Based on the simulation results of the optimized agent model, it is determined whether the noise at the sensitive target exceeds the preset warning threshold. If so, traffic control suggestions are generated and pushed.

[0013] In some embodiments, constructing an intelligent agent model based on a physical information neural network includes:

[0014] A multimodal architecture combining convolutional neural networks and a vision-language model is adopted to extract spatial environmental detail features in the three-dimensional digital environment and construct a multimodal perception and recognition subunit.

[0015] A physical information neural network architecture based on graph neural networks is adopted, and acoustic physical equations are used as pre-constraints for network training to construct a noise field simulation and deduction sub-unit.

[0016] A hybrid training strategy combining semi-supervised and self-supervised training is adopted. Supervised fine-tuning is performed using labeled measured noise verification data, and self-supervised training is performed using unlabeled data, thus constructing a self-supervised iterative training sub-unit.

[0017] In some embodiments, the step of iteratively training the agent model using self-supervised learning based on the model validation parameters to optimize its computational parameters includes:

[0018] Based on the model verification parameters, determine whether the preset iteration triggering conditions are met; if so, start the iterative optimization process.

[0019] In response to the initiation of the iterative optimization process, training samples are extracted from the multi-source data fusion training library to construct a self-supervised training sample set;

[0020] Input the self-supervised training sample set into the agent model and calculate the composite loss function;

[0021] Based on the composite loss function, the operational parameters of the agent model are updated using the backpropagation algorithm, and it is determined whether the updated model satisfies the convergence condition.

[0022] The agent model determined by convergence will be solidified and switched to the official running version.

[0023] In some embodiments, the step of extracting training samples from a multi-source data fusion training library and constructing a self-supervised training sample set in response to the initiation of the iterative optimization process includes:

[0024] Labeled training data is extracted from the model validation parameters to form a labeled sample set. The labeled training data includes the input feature vectors of each monitoring point and their corresponding simulation errors.

[0025] Unlabeled training data within the same spatiotemporal range as the labeled training data are extracted from the multi-source data fusion training library to form an unlabeled sample set. The unlabeled training data includes input feature vectors in areas where no monitoring points are set.

[0026] Spatiotemporal alignment and feature standardization are performed on all samples in the labeled sample set and the unlabeled sample set.

[0027] The processed labeled and unlabeled sample sets are randomly divided into several small batches, which are used as training batches.

[0028] In some embodiments, updating the operational parameters of the agent model based on the composite loss function using the backpropagation algorithm and determining whether the updated model satisfies the convergence condition includes:

[0029] The gradient of the composite loss function with respect to the operational parameters of the agent model is calculated using automatic differentiation techniques.

[0030] An adaptive moment estimation optimizer is used to update the operational parameters of the agent model based on the gradient:

[0031] The predicted noise values ​​of the same batch of verification points are recalculated using the updated operational parameters, and the updated global mean error and the maximum error of a single point are calculated.

[0032] When the global mean error and the maximum error at a single point meet the preset conditions, the convergence condition is met, and the updated operation parameters are accepted; otherwise, the parameter update is rejected, and the operation parameters are restored to their previous state.

[0033] In some embodiments, the process of solidifying the model determined by convergence and switching to the production version includes:

[0034] The updated agent model was tested for accuracy and stability on an independent test dataset, and the mean error and standard deviation of the error on the test set were calculated.

[0035] When the mean error and standard deviation of the error on the test set meet the test conditions, the updated core operation parameters of the model and its corresponding network architecture configuration file and normalization parameters are packaged, written into the model repository and assigned a new version number.

[0036] Record the version metadata information that has been solidified this time. The version metadata information includes version number, solidification time, trigger type, number of training samples, training data time range, error comparison before and after update, composite loss value and number of iteration rounds.

[0037] The pointer to the intelligent agent model called by the dynamic early warning system has been switched from the old version to the new version.

[0038] Secondly, embodiments of this application provide a system for dynamic monitoring and environmental impact analysis model optimization of urban integrated traffic noise, including:

[0039] The first acquisition module is used to acquire high-precision map data of the target area, construct a three-dimensional spatial digital environment, and automatically identify and label sensitive targets in the acoustic environment based on a preset sensitive target feature library.

[0040] The access module is used to access and parse the acoustic environment quality standards, extract the noise limit standards and technical constraints, and convert them into standardized data operation rules.

[0041] The second acquisition module is used to acquire multi-source traffic flow data of the target area in real time.

[0042] The construction module is used to construct an intelligent agent model based on a physical information neural network. The intelligent agent model is driven by the multi-source traffic flow data and constrained by the three-dimensional spatial digital environment. It dynamically simulates and deduces the three-dimensional spatial distribution intensity of the noise field in the target area and its impact on sensitive targets.

[0043] The third acquisition module is used to determine the noise monitoring points and acquire measured noise data based on the distribution of the sensitive targets and the standard layout rules.

[0044] The generation module is used to compare the simulated noise data output by the intelligent agent model with the measured noise data in multiple dimensions, calculate the model error, and generate model verification parameters.

[0045] The self-supervised optimization module is used to iteratively train the agent model based on the model validation parameters and optimize its computational parameters through self-supervised learning.

[0046] The dynamic early warning module is used to determine whether the noise at the sensitive target exceeds the preset early warning threshold based on the simulation results of the optimized intelligent agent model. If so, it generates and pushes traffic control suggestions.

[0047] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in embodiments of this application.

[0048] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in embodiments of this application.

[0049] Fifthly, embodiments of this application provide a computer program product, including a computer program, characterized in that, when the computer program is executed by a processor, it implements the method described in embodiments of this application.

[0050] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0051] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0052] Figure 1 A flowchart illustrating the optimization method for dynamic monitoring and environmental impact analysis model of urban integrated traffic noise provided in an embodiment of this application is shown.

[0053] Figure 2 This invention provides a schematic diagram of the structure of an urban integrated traffic noise dynamic monitoring and environmental impact analysis model optimization system according to an embodiment of this application.

[0054] Figure 3 A schematic diagram of the structure of a computer system suitable for implementing an electronic device or server according to embodiments of this application is shown. Detailed Implementation

[0055] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0056] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0057] To further illustrate the technical solutions provided in the embodiments of this application, a detailed description is provided below in conjunction with the accompanying drawings and specific implementation methods. Although the embodiments of this application provide method operation instruction steps as shown in the following embodiments or drawings, the method may include more or fewer operation instruction steps based on conventional or non-creative effort. In steps where there is no logically necessary causal relationship, the execution order of these steps is not limited to the execution order provided in the embodiments of this application. In actual processing or when the device executes the method, it may be executed sequentially or in parallel according to the method shown in the embodiments or drawings.

[0058] It should be noted that the acquisition or use of data in the embodiments of this application requires the user's consent. The relevant data can only be obtained after the user's authorization and permission, and the acquisition or use of the data complies with the laws and regulations of the relevant regions.

[0059] Please refer to Figure 1 , Figure 1 This illustration shows a flowchart of an embodiment of the method for optimizing a dynamic monitoring and environmental impact analysis model of urban integrated traffic noise provided in this application. Figure 1 As shown, the method includes:

[0060] Step 101: Obtain high-precision map data of the target area, construct a three-dimensional spatial digital environment, and automatically identify and label sensitive targets in the acoustic environment based on a preset sensitive target feature library.

[0061] It should be noted that the three-dimensional spatial digital environment is a virtual environment for simulating traffic noise based on an intelligent agent model. The three-dimensional spatial digital environment can access, parse, and process high-precision map data that conforms to relevant national surveying and mapping standards through a data interface. The high-precision map data includes at least road geometric alignment data, three-dimensional outline data of buildings (structures) around the road, topographic data, road surface and ancillary facilities data, satellite image layer data, and land use and land use type data.

[0062] Among them, the three-dimensional spatial digital environment can also combine satellite image layers and a preset sensitive target feature library to automatically identify and label sensitive targets in the urban sound environment.

[0063] Specifically, the 3D spatial digital environment accesses high-precision map data of the target area, which includes at least road geometry, building 3D outlines, terrain features, road surface materials, and satellite imagery layers. Then, all data is uniformly converted into vector data in a standard spatial coordinate system to construct a real-world 3D digital model. Based on a pre-defined sensitive target feature library, image recognition and spatial overlay analysis technologies are used to automatically match and label all acoustically sensitive targets within the scene, such as schools, hospitals, and densely populated residential areas, and output their location and extent information to a multi-source data fusion training library.

[0064] Step 102: Access and parse the acoustic environment quality standards, extract the noise limit standards and technical constraints, and convert them into standardized data operation rules.

[0065] In other words, the three-dimensional spatial digital environment further accesses and analyzes relevant national and industry standards for acoustic environment quality, motor vehicle noise emission standards, and laws and regulations, such as GB 1495-2002, GB 3096-2008, and JT / T 1198-2018, to extract the rules for dividing acoustic environment functional zones, noise limit standards, noise measurement specifications, sensitive target protection requirements, and technical constraints. The extracted standard constraints are then transformed into standardized data operation rules that can be recognized and executed by the intelligent agent model, as well as standard limit reference data that can be presented in a visual interface.

[0066] Step 103: Obtain multi-source traffic flow data for the target area in real time.

[0067] It should be noted that multi-source traffic flow data refers to traffic flow data obtained through multiple types of data sources. Specifically, traffic flow monitoring data can be acquired in real time from various types of traffic flow monitoring terminals (such as roadside radar, geomagnetic coils, checkpoint cameras, and floating car GPS, etc.) to form multi-source real-time urban traffic flow monitoring data.

[0068] It should be understood that after acquiring the raw data of multi-source real-time urban traffic flow monitoring, all accessed multi-source data are uniformly standardized and processed, including time synchronization (aligning all data to the same timestamp), spatial matching (mapping vehicle positions to the road network coordinate system), and missing value handling (using linear interpolation to fill in the missing values). After these data preprocessing methods, multi-source real-time urban traffic flow monitoring data are obtained for the intelligent agent model to perform noise field simulation and deduction.

[0069] Step 104: Construct an intelligent agent model based on a physical information neural network. The intelligent agent model is driven by multi-source traffic flow data and constrained by a three-dimensional spatial digital environment. It dynamically simulates and deduces the three-dimensional spatial distribution intensity of noise in the target area and its impact on sensitive targets.

[0070] Optionally, an intelligent agent model based on a physical information neural network can be formed by constructing multiple model sub-units.

[0071] In one feasible embodiment, a physical information neural network-based intelligent agent model is constructed, including: using a multimodal architecture combining a convolutional neural network and a vision-language model to extract spatial environment detail features in the three-dimensional digital environment and constructing a multimodal perception and recognition subunit; using a physical information neural network architecture based on a graph neural network, with acoustic physics equations as pre-constraints for network training, to construct a noise field simulation and deduction subunit; and using a hybrid training strategy combining semi-supervised and self-supervised training, using labeled measured noise verification data for supervised fine-tuning and unlabeled data for self-supervised training, to construct a self-supervised iterative training subunit.

[0072] Specifically, the multimodal perception and recognition subunit is used to extract deep features from satellite imagery and 3D contour data in high-precision maps, accurately identifying spatial details such as building layout, vegetation density, and terrain slope, which serve as boundary constraints for subsequent simulations. The noise field simulation and deduction subunit is used to embed acoustic physics equations into a neural network as prior constraints. The self-supervised iterative training subunit is used to converge the agent model to a preset initial threshold through training.

[0073] It should be understood that the intelligent agent model based on the physical information neural network is the core of realizing the dynamic monitoring of urban integrated traffic noise and environmental impact analysis in the embodiments of this application. In the early stage of construction, the intelligent agent model is initially iteratively trained by the self-supervised iterative training sub-unit to meet the initial conditions for deployment and application.

[0074] Step 105: Determine the noise monitoring points based on the distribution of sensitive targets and standard deployment rules, and obtain measured noise data.

[0075] Specifically, based on the monitoring point layout rules stipulated in national standards (such as GB 3096-2008), and combined with the identified distribution of sensitive targets and the characteristics of the three-dimensional spatial environment, a standardized monitoring point layout scheme is intelligently generated to guide the installation of standard-compliant noise monitoring terminals at corresponding locations. This allows for the real-time collection of actual feedback data, such as equivalent sound level (Leq), statistical sound level (LN), and meteorological data, which can be used to optimize the intelligent agent model. The acquired measured data will serve as the basis for model verification and iterative optimization.

[0076] Step 106: Compare the simulated noise data output by the intelligent agent model with the measured noise data in multiple dimensions, calculate the model error, and generate model verification parameters.

[0077] Multi-dimensional comparisons include, but are not limited to, time-dimensional comparisons, spatial-dimensional comparisons, and statistical-dimensional comparisons. Time-dimensional comparisons calculate the absolute error between simulated and measured values ​​at the same monitoring point and time. Spatial-dimensional comparisons analyze the distribution characteristics of simulated and measured errors at multiple different points within the same region. Statistical-dimensional comparisons calculate the mean error within a continuous time window. and standard deviation .

[0078] Furthermore, based on the analysis results, a model validation parameter file containing detailed error information is generated.

[0079] Step 107: Based on the model validation parameters, iteratively train the agent model using self-supervised learning to optimize its computational parameters.

[0080] In other words, after obtaining the measured noise data of the noise monitoring points, the model verification data based on the measured noise data is used to perform self-supervised learning through the self-supervised iterative training subunit in the agent model, so as to achieve iterative training of the agent model and optimize the operational parameters of the agent model.

[0081] It should be understood that this parameter file reflects the current simulation accuracy of the agent model within the target area. That is, when the error is within an acceptable range, the agent model can continue to be used for subsequent dynamic simulation and early warning analysis; when the error exceeds the preset threshold or shows a continuous deterioration trend, it indicates that the model's ability to fit the current urban traffic environment and noise propagation conditions has declined, and it is necessary to update the model's operational parameters through iterative optimization to make it adapt to environmental changes again.

[0082] Therefore, this application establishes an iterative optimization mechanism for intelligent agent models based on self-supervised learning, enabling the intelligent agent model to automatically correct its own parameters using measured noise data, continuously adapting to the dynamically changing urban traffic environment without human intervention, and ensuring that the simulation accuracy of the intelligent agent model does not decrease with running time.

[0083] In a feasible embodiment, based on the model validation parameters, it is determined whether the preset iteration triggering conditions are met. If so, the iterative optimization process is initiated. In response to the initiation of the iterative optimization process, training samples are extracted from the multi-source data fusion training library to construct a self-supervised training sample set. The self-supervised training sample set is input into the agent model, and the composite loss function is calculated. Based on the composite loss function, the operational parameters of the agent model are updated through the backpropagation algorithm, and it is determined whether the updated model meets the convergence condition. The agent model that passes the convergence judgment is solidified and switched to the official running version.

[0084] Specifically, the latest validation parameters are read from the model validation feedback information. These parameters include at least the single-point absolute error sequence for each monitoring point. } (where i is the monitoring point number and t is time), the global error mean μΔ and its trend over time, and the global error standard deviation σΔ. The calculation method for the single-point absolute error ΔLi(t) is as follows:

[0085]

[0086] in,( ) represents the three-dimensional spatial coordinates (unit: meters) of the i-th monitoring point, z represents the altitude, used to distinguish the noise impact of different floors; t represents the time. The noise value (in dB(A)) predicted by the agent model at the i-th monitoring point and time t. This represents the measured noise value (unit: dB(A)) collected by the noise monitoring terminal at the i-th monitoring point and time t. The global error mean μ Δ The calculation method is as follows:

[0087]

[0088] Where M represents the total number of monitoring points.

[0089] In a feasible embodiment, the single-point absolute error ΔL for any monitoring point in K consecutive verifications is... i (t) exceeds the preset allowable threshold ΔL th The percentage of times exceeded R th Determine if the preset iteration start conditions are met.

[0090] Preferably, K=3 (number of consecutive verifications), ΔLth=2.5 dB(A) (single-point absolute error allowable threshold), and Rth=80% (percentage threshold for number of times exceeding the standard).

[0091] It should be understood that this condition is used to capture significant simulation biases in local areas. That is, when the error at a specific monitoring point is consistently high, it indicates that there is a systematic bias in the extraction of environmental features or the calculation of noise propagation in that area by the agent model, which needs to be corrected accordingly.

[0092] In another feasible embodiment, the global error mean μ is calculated over the most recent N verification periods. Δ Perform linear regression, with a regression slope b > b th Determine if the preset iteration start conditions are met.

[0093] Preferably, N=7 (number of verification periods, corresponding to 7 days), b th =0.15 dB(A) / day (maximum allowable ramp rate).

[0094] It should be understood that this condition is used to capture the overall degradation trend of model accuracy. That is, when the model error shows a statistically significant and continuous increase, it indicates that the urban environment (such as changes in traffic flow, road construction, new buildings, etc.) has undergone substantial changes that are sufficient to affect the model accuracy.

[0095] In another feasible embodiment, the time interval since the last iteration of optimization reaches a preset period Tcycle = 30 days. It should be understood that this condition serves as a routine maintenance mechanism to ensure that the model undergoes a comprehensive parameter calibration periodically, even if the error does not exceed the limit.

[0096] Preferably, when at least one of the above triggering conditions is met, it is determined that the iteration triggering condition is met and the iteration optimization process is started; otherwise, monitoring continues and the process waits for the next verification cycle.

[0097] After determining that the iteration triggering condition is met, the self-supervised iterative training submodule enters the training sample set construction phase. This phase involves extracting training data for this iteration from the multi-source data fusion training library, forming a sample set that combines labeled and unlabeled data, providing a data foundation for subsequent loss calculation and parameter updates.

[0098] In one feasible embodiment, labeled training data is extracted from model validation parameters to form a labeled sample set. The labeled training data includes the input feature vectors of each monitoring point and their corresponding simulation errors. Unlabeled training data within the same spatiotemporal range as the labeled training data is extracted from the multi-source data fusion training library to form an unlabeled sample set. The unlabeled training data includes the input feature vectors of the regions where no monitoring points are set. All samples in the labeled and unlabeled sample sets are subjected to spatiotemporal alignment and feature standardization. The processed labeled and unlabeled sample sets are divided into several small batches in a random order as training batches.

[0099] Specifically, labeled training data is extracted from the model validation parameter file. Each labeled sample contains the input feature vector for that monitoring point. and its corresponding simulation error ΔL j Input feature vector Including the point at time t j Multi-source data: spatial coordinates (x j , y j , z j Traffic flow parameters (vehicle volume Q) j Vehicle speed v j Vehicle model ratio p i,j Meteorological parameters (wind speed w) j Temperature T j Humidity Hj and spatial environment characteristics Simulation error All labeled samples constitute the labeled sample set D. labeled ={( ΔL j )}, j=1,…,M, where M is the total number of labeled samples available in this iteration (valid data pairs from all monitoring points within the verification period). Labeled data is the core signal driving the model parameters to converge to the measured environment.

[0100] Regional data within the same spatiotemporal range as the labeled data, but without monitoring points, are extracted from the multi-source data fusion training library and used as unlabeled training data. Each unlabeled sample contains only the input feature vector. This does not include measured noise values. Unlabeled sample set D unlabeled ={ Let M be the number of unlabeled samples (spatial points randomly sampled within the simulation region, typically much larger than M). Unlabeled data is used to calculate the physical consistency loss, reinforcing the model's adherence to physical laws and preventing overfitting to finite labeled data.

[0101] All extracted samples undergo uniform spatiotemporal alignment and feature standardization. Temporal alignment involves converting timestamps to the system's standard time (UTC+8) and normalizing them to the same sampling interval (e.g., 1 minute). Spatial alignment involves converting all coordinates to the same geographic coordinate system (CGCS2000). Feature standardization involves Z-score standardization of continuous values ​​(e.g., traffic flow, temperature) in the input feature vector X, ensuring a mean of 0 and a standard deviation of 1 to accelerate gradient descent convergence.

[0102]

[0103] Where μ X and σ X represents the mean and standard deviation of this feature on the training set.

[0104] Then, the processed labeled sample set D labeled and unlabeled sample set D unlabeled The gradient is divided into several mini-batches in random order, with each batch size set to B (e.g., B=64). Mini-batch training can improve the stability of gradient estimation and accelerate convergence.

[0105] After constructing the self-supervised training sample set, the next step is to calculate the composite loss function. By inputting the training samples into the agent model, the predicted values ​​are obtained through forward propagation. Based on these predicted values, a composite loss function containing multiple optimization objectives is calculated, which serves as the optimization objective for subsequent parameter updates.

[0106] In one feasible embodiment, the composite loss function includes, but is not limited to, calibration loss, physical consistency loss, and smoothness constraint loss.

[0107] Specifically, the input features of the labeled samples in the current batch Input features of unlabeled samples The noise field simulation and deduction submodule (based on a graph neural network-based physical information neural network architecture) is input separately, and forward propagation is performed to obtain the corresponding predicted noise values. :

[0108]

[0109] in Let θ represent the forward propagation function of the agent model with parameter θ.

[0110] Calibration loss is calculated only for labeled samples. :

[0111]

[0112] Among them, B labeled This represents the number of tagged samples in the current batch.

[0113] It should be understood that the calibration loss is used to directly drive the model predictions to converge to the measured values. That is, the larger the error, the larger the loss term, and the greater the adjustment of the parameters in that region during backpropagation.

[0114] Physical consistency loss is applied to both labeled and unlabeled samples. Calculation:

[0115]

[0116] Among them, B labeled This represents the number of tagged samples in the current batch. This represents the number of unlabeled samples in the current batch. Let be the acoustic physics equation residual for the q-th sample. This residual is calculated as follows:

[0117]

[0118] in, The predicted noise value (in dB(A)) for the q-th sample is output by forward propagation; The sound power level (in dB(A)) of the sound source near the qth sample, for vehicle type i at vehicle speed v_q, is obtained by looking up a table in the vehicle noise database. Let q be the straight-line distance (in meters) from the q-th sample point to the nearest main sound source. This distance is calculated based on the sound source location sq and the sample point location r. q Calculate: r q =r q -s q ; Air absorption attenuation (unit: dB), based on propagation distance r q Temperature T q Humidity H q Calculated according to ISO 9613-1 standard; This is the ground effect attenuation (unit: dB), calculated based on the ground type along the propagation path (grass, concrete, asphalt, etc., considering spatial environmental characteristics). Extracted from ( ), calculated according to ISO 9613-2 standard; Add attenuation (in dB) to obstacles (buildings, sound barriers) based on the three-dimensional obstacle distribution between the sample point and the sound source (calculated from building profile data) according to diffraction theory.

[0119] It should be understood that the physical consistency loss ensures that the model's output does not deviate from the basic laws of acoustics, even in regions where measured data is lacking, thereby effectively suppressing the physically unreasonable predictions that a purely data-driven model may produce.

[0120] Calculate the smoothness constraint loss :

[0121]

[0122] in, This is the parameter vector of the agent model to be updated in this iteration (including the weights and biases of each layer of the neural network). This is the parameter vector of the agent model at the end of the previous iteration.

[0123] It should be understood that the smoothness constraint loss term is used to suppress parameter abrupt changes and ensure the stability of model evolution. That is, the greater the parameter change, the larger the loss term will be, and it will penalize large parameter jumps during backpropagation.

[0124] Furthermore, by weighting and summing the calibration loss, physical consistency loss, and smoothness constraint loss, we obtain the composite loss function:

[0125]

[0126] Where α is the calibration loss weight coefficient, β is the physical consistency loss weight coefficient, and γ is the smoothness constraint loss weight coefficient.

[0127] Optionally, the three weight coefficients can be dynamically adjusted at different stages of iterative optimization. For example, in the first three rounds of iterative optimization, α=0.8, β=0.15, and γ=0.05 are set to prioritize rapid fitting of the measured data (α>β and α>γ). After the third round of iterative optimization, α=0.5, β=0.4, and γ=0.1 are set, increasing β relative to α to strengthen the constraints imposed by physical laws. Through this dynamic weight adjustment strategy, the model prioritizes learning the distribution characteristics of the measured data in early iterations, while in later iterations it focuses more on maintaining physical consistency, achieving a smooth transition from "data fitting" to "physical constraints."

[0128] After calculating the composite loss function, the next step is parameter update and convergence determination. The backpropagation algorithm is used to calculate the gradient of the loss function with respect to each network parameter. Based on this, the model parameters are updated, the update effect is evaluated, and it is determined whether the model has reached convergence.

[0129] In one feasible embodiment, the gradient of the composite loss function with respect to the operational parameters of the agent model is calculated using automatic differentiation technology; an adaptive moment estimation optimizer is used to update the operational parameters of the agent model based on the gradient; the predicted noise values ​​of the same batch of verification points are recalculated with the updated operational parameters, and the updated global mean error and the maximum error at a single point are calculated; when the global mean error and the maximum error at a single point meet the preset conditions, it is determined that the convergence condition is met, and the updated operational parameters are accepted; otherwise, the parameter update is rejected, and the operational parameters are restored to those before the update.

[0130] Specifically, automatic differentiation is used to calculate the gradient of the composite loss function with respect to each network parameter. The gradient direction indicates how the parameters should be adjusted to reduce the loss.

[0131] The adaptive moment estimator optimizer (Adam) is used to perform parameter updates. The Adam optimizer, through its adaptive learning rate, can apply different update step sizes to different parameters, thus accelerating convergence. Its update rule is as follows:

[0132]

[0133] in, For the updated calculation parameters, These are the computational parameters before the update, where η is the learning rate, initially set to 1×10. -4 It decays with each iteration; The bias correction estimate for the first moment (mean) of the gradient is accumulated by the historical gradient direction; The bias correction estimate for the second moment (variance) of the gradient is used, and the learning rate of each parameter is adaptively adjusted; ε is the numerical stability constant, which is taken as 10. -8 To prevent division by zero.

[0134] With updated operation parameters Recalculate the predicted noise values ​​for the same batch of verification points, and calculate the updated global error mean μ. Δ (new) and single-point maximum error max(ΔL(t))(new).

[0135] Furthermore, if μΔ(new) < μΔ(old) and max(ΔL(t))(new) ≤ ΔLth, then the iteration is successful and the new parameters are accepted. The loss value and error improvement of this iteration are recorded. If μΔ(new) ≥ μΔ(old) (i.e., the error has not improved), the parameter update is rejected, the value is restored to θ(old), and the learning rate η is reduced (e.g., reduced to 0.8 times the original value) and then retried. The maximum number of retryings is 3. If the error does not improve significantly for 10 consecutive iterations (improvement < 0.05 dB(A)), the model is determined to be trapped in a local optimum or the data is abnormal. The iteration is terminated and an anomaly report is generated.

[0136] After successful iteration and convergence, the model solidification and version management phase begins. This phase finalizes the updated model, records complete version metadata, and switches the running version to the optimized new model, providing a more precise computing engine for subsequent dynamic simulations and early warning linkage.

[0137] In a feasible embodiment, the updated agent model is subjected to accuracy and stability testing on an independent test dataset, and the mean error and standard deviation of the error on the test set are calculated. When the mean error and standard deviation of the error on the test set both meet the test conditions, the updated model's core operational parameters and their corresponding network architecture configuration files and normalization parameters are packaged, written into the model repository, and assigned a new version number. The version metadata of this fixation is recorded, including the version number, fixation time, trigger type, number of training samples, training data time range, error comparison before and after the update, composite loss value, and number of iterations. The agent model pointer called by the dynamic warning is switched from the old version to the new version.

[0138] Specifically, the updated model is tested for accuracy and stability on an independent test dataset (real-world data that has not been used for training in the last 24 hours). The mean error μΔ(test) and standard deviation σΔ(test) on the test set are calculated. Only when μΔ(test) ≤ 1.5 dB(A) and σΔ(test) ≤ 1.0 dB(A) is the model allowed to enter the solidification stage; otherwise, more training data is needed before iterating again.

[0139] Then, the model parameters θ that have passed the stability test, along with their corresponding network architecture configuration files and normalized parameters (μ), will be used. X σ X The model is packaged, written to the model repository, and assigned a new version number. The complete version metadata for this fixation is recorded in the data storage interaction module, including: version number, fixation time, trigger type (condition A / condition B / condition C), number of training samples (M labeled samples, P unlabeled samples), training data time range, and error comparison before and after the update (μ). Δ (old)→μ Δ (new)), composite loss value, and number of iterations.

[0140] Finally, the pointers to the intelligent agent model used by dynamic early warning and other downstream applications were switched from the old version to the new version. From then on, subsequent noise simulations and early warning judgments were all performed based on the optimized model. After the version switch was completed, the model accuracy was monitored, awaiting the fulfillment of the trigger conditions for the next iteration.

[0141] Thus, through the above process, the accuracy and adaptability of the intelligent agent model improve with each "test-verification-optimization" cycle. As the running time accumulates, the adaptability of the intelligent agent model to changes in urban traffic flow and the surrounding environment will continue to improve, and the simulation accuracy will not decrease due to environmental changes, achieving a self-evolving capability of "getting more accurate with each run".

[0142] Step 108: Based on the simulation results of the optimized agent model, determine whether the noise at the sensitive target exceeds the preset warning threshold. If so, generate and push traffic control suggestions.

[0143] Specifically, real-time multi-source traffic flow data of the target area is input into the optimized agent model, which continuously and dynamically predicts the noise field changes over the next hour and outputs the predicted noise values ​​at sensitive targets. .

[0144] Dynamic early warning function makes real-time judgments .like or If the noise level is too high, an early warning will be triggered immediately. The system will also extract suggestions from the pre-set traffic noise reduction and control strategy knowledge base that match the current excessive scenario (such as the location of the excessive noise, the amount of the excessive noise, and the type of sensitive target). These suggestions (such as "adjust the timing of the traffic lights at intersection A" or "restrict heavy trucks from passing through section B") will be pushed to the traffic control platform along with detailed warning information, thus completing the closed loop from monitoring and simulation to coordinated governance.

[0145] In summary, the urban integrated traffic noise dynamic monitoring and environmental impact analysis model optimization method provided in this application acquires high-precision map data of the target area, constructs a three-dimensional spatial digital environment, and automatically identifies and labels sensitive targets in the sound environment based on a preset sensitive target feature library; it accesses and parses sound environment quality standards, extracts noise limit standards and technical constraints, and transforms them into standardized data operation rules; it acquires multi-source traffic flow data of the target area in real time; and it constructs an intelligent agent model based on a physical information neural network. This intelligent agent model is driven by multi-source traffic flow data and constrained by the three-dimensional spatial digital environment, dynamically simulating and deducing the three-dimensional spatial distribution intensity of the noise field within the target area and its impact on sensitive targets. Based on the distribution of sensitive targets and standard deployment rules, noise monitoring points are determined, and measured noise data is obtained. The simulated noise data output by the intelligent agent model is compared with the measured noise data in multiple dimensions to calculate the model error and generate model verification parameters. Based on the model verification parameters, the intelligent agent model is iteratively trained through self-supervised learning to optimize its computational parameters. Based on the simulation results of the optimized intelligent agent model, it is determined whether the noise at the sensitive target exceeds the preset warning threshold. If so, traffic control suggestions are generated and pushed. Through the closed-loop mechanism of "simulation-measurement-verification-optimization", the model can continuously adapt to the dynamically changing environment. The optimized model is linked with the traffic control system to achieve proactive early warning and automatic control of noise pollution.

[0146] It should be noted that although the operation of the method of the present invention is described in a specific order in the accompanying drawings, this does not require or imply that the operations must be performed in that specific order, or that all the operations shown must be performed in order to achieve the desired result.

[0147] Figure 2 A schematic diagram of the structure of an urban integrated traffic noise dynamic monitoring and environmental impact analysis model optimization system provided in an embodiment of this application is shown.

[0148] like Figure 2 As shown, the urban integrated traffic noise dynamic monitoring and environmental impact analysis model optimization system 10 includes:

[0149] The first acquisition module 11 is used to acquire high-precision map data of the target area, construct a three-dimensional spatial digital environment, and automatically identify and label sensitive targets in the acoustic environment based on a preset sensitive target feature library.

[0150] The access module 12 is used to access and parse the acoustic environment quality standards, extract the noise limit standards and technical constraints therein, and convert them into standardized data operation rules;

[0151] The second acquisition module 13 is used to acquire multi-source traffic flow data of the target area in real time.

[0152] Module 14 is used to construct an intelligent agent model based on a physical information neural network. The intelligent agent model is driven by the multi-source traffic flow data and constrained by the three-dimensional spatial digital environment. It dynamically simulates and deduces the three-dimensional spatial distribution intensity of the noise field in the target area and its impact on sensitive targets.

[0153] The third acquisition module 15 is used to determine the noise monitoring points and acquire measured noise data based on the distribution of the sensitive targets and the standard layout rules.

[0154] The generation module 16 is used to compare the simulated noise data output by the intelligent agent model with the measured noise data in multiple dimensions, calculate the model error, and generate model verification parameters.

[0155] The self-supervised optimization module 17 is used to iteratively train the agent model based on the model verification parameters and optimize its computational parameters through self-supervised learning.

[0156] The dynamic early warning module 18 is used to determine whether the noise at the sensitive target exceeds the preset early warning threshold based on the simulation results of the optimized intelligent agent model. If so, it generates and pushes traffic control suggestions.

[0157] In some embodiments, the construction module 14 is specifically used for:

[0158] A multimodal architecture combining convolutional neural networks and a vision-language model is adopted to extract spatial environmental detail features in the three-dimensional digital environment and construct a multimodal perception and recognition subunit.

[0159] A physical information neural network architecture based on graph neural networks is adopted, and acoustic physical equations are used as pre-constraints for network training to construct a noise field simulation and deduction sub-unit.

[0160] A hybrid training strategy combining semi-supervised and self-supervised training is adopted. Supervised fine-tuning is performed using labeled measured noise verification data, and self-supervised training is performed using unlabeled data, thus constructing a self-supervised iterative training sub-unit.

[0161] In some embodiments, the self-supervised optimization module 17 is specifically used for:

[0162] Based on the model verification parameters, determine whether the preset iteration triggering conditions are met; if so, start the iterative optimization process.

[0163] In response to the initiation of the iterative optimization process, training samples are extracted from the multi-source data fusion training library to construct a self-supervised training sample set;

[0164] Input the self-supervised training sample set into the agent model and calculate the composite loss function;

[0165] Based on the composite loss function, the operational parameters of the agent model are updated using the backpropagation algorithm, and it is determined whether the updated model satisfies the convergence condition.

[0166] The agent model determined by convergence will be solidified and switched to the official running version.

[0167] In some embodiments, the self-supervised optimization module 17 is specifically used for:

[0168] Labeled training data is extracted from the model validation parameters to form a labeled sample set. The labeled training data includes the input feature vectors of each monitoring point and their corresponding simulation errors.

[0169] Unlabeled training data within the same spatiotemporal range as the labeled training data are extracted from the multi-source data fusion training library to form an unlabeled sample set. The unlabeled training data includes input feature vectors in areas where no monitoring points are set.

[0170] Spatiotemporal alignment and feature standardization are performed on all samples in the labeled sample set and the unlabeled sample set.

[0171] The processed labeled and unlabeled sample sets are randomly divided into several small batches, which are used as training batches.

[0172] In some embodiments, the self-supervised optimization module 17 is specifically used for:

[0173] The gradient of the composite loss function with respect to the operational parameters of the agent model is calculated using automatic differentiation techniques.

[0174] An adaptive moment estimation optimizer is used to update the operational parameters of the agent model based on the gradient:

[0175] The predicted noise values ​​of the same batch of verification points are recalculated using the updated operational parameters, and the updated global mean error and the maximum error of a single point are calculated.

[0176] When the global mean error and the maximum error at a single point meet the preset conditions, the convergence condition is met, and the updated operation parameters are accepted; otherwise, the parameter update is rejected, and the operation parameters are restored to their previous state.

[0177] In some embodiments, the self-supervised optimization module 17 is specifically used for:

[0178] The updated agent model was tested for accuracy and stability on an independent test dataset, and the mean error and standard deviation of the error on the test set were calculated.

[0179] When the mean error and standard deviation of the error on the test set meet the test conditions, the updated core operation parameters of the model and its corresponding network architecture configuration file and normalization parameters are packaged, written into the model repository and assigned a new version number.

[0180] Record the version metadata information that has been solidified this time. The version metadata information includes version number, solidification time, trigger type, number of training samples, training data time range, error comparison before and after update, composite loss value and number of iteration rounds.

[0181] The pointer to the intelligent agent model called by the dynamic early warning system has been switched from the old version to the new version.

[0182] It should be understood that the modules or modules recorded in the Urban Integrated Traffic Noise Dynamic Monitoring and Environmental Impact Analysis Model Optimization System 10 and the reference modules Figure 1 The steps in the described method correspond accordingly. Therefore, the operations and features described above for the method are also applicable to the Urban Integrated Traffic Noise Dynamic Monitoring and Environmental Impact Analysis Model Optimization System 10 and its included modules, and will not be repeated here. The Urban Integrated Traffic Noise Dynamic Monitoring and Environmental Impact Analysis Model Optimization System 10 can be pre-implemented in the browser or other secure applications of an electronic device, or it can be loaded into the browser or its secure applications of an electronic device through download or other means. The corresponding modules in the Urban Integrated Traffic Noise Dynamic Monitoring and Environmental Impact Analysis Model Optimization System 10 can cooperate with the modules in the electronic device to implement the solution of the embodiments of this application.

[0183] The division of modules or units mentioned in the detailed description above is not mandatory. In fact, according to the embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0184] The following is for reference. Figure 3 , Figure 3 A schematic diagram of the structure of a computer system suitable for implementing the embodiments of this application is shown.

[0185] like Figure 3As shown, the computer system 300 includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 302 or programs loaded from storage section 308 into random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the system's operating instructions. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0186] The following components are connected to I / O interface 305: an input section 306 including a keyboard, mouse, etc.; an output section 307 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN card, modem, etc. The communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0187] Specifically, according to embodiments of this application, the flowchart above refers to... Figure 2 The described process can be implemented as a computer software program. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. In such an embodiment, the computer program contains program code for performing the methods shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the functions defined in the system of this application.

[0188] It should be noted that the computer-readable medium shown in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0189] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operational instructions of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two connected blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified functions or operational instructions, or using a combination of dedicated hardware and computer instructions.

[0190] The units or modules described in the embodiments of this application can be implemented in software or hardware. The described units or modules can also be housed in a processor. For example, a processor can be described as including a first acquisition module, an access module, a second acquisition module, a construction module, a third acquisition module, a generation module, a self-supervised optimization module, and a dynamic early warning module. The names of these units or modules do not necessarily limit the specific unit or module itself. For example, the first acquisition module can also be described as "acquiring high-precision map data of the target area, constructing a three-dimensional spatial digital environment, and automatically identifying and labeling acoustic environment sensitive targets based on a preset sensitive target feature library."

[0191] In another aspect, this application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments, or may exist independently and not assembled into the electronic device. The aforementioned computer-readable storage medium stores one or more programs that, when used by one or more processors, execute the urban integrated traffic noise dynamic monitoring and environmental impact analysis model optimization method described in this application.

[0192] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for optimizing a dynamic monitoring and environmental impact analysis model for urban integrated traffic noise, characterized in that, include: Acquire high-precision map data of the target area, construct a three-dimensional digital environment, and automatically identify and label sensitive targets in the acoustic environment based on a preset sensitive target feature library; Access and analyze the acoustic environment quality standards, extract the noise limit standards and technical constraints, and transform them into standardized data operation rules; Real-time acquisition of multi-source traffic flow data for the target area; A smart agent model based on a physical information neural network is constructed. The smart agent model is driven by the multi-source traffic flow data and constrained by the three-dimensional spatial digital environment. It dynamically simulates and deduces the three-dimensional spatial distribution intensity of the noise field in the target area and its impact on sensitive targets. Based on the distribution of the sensitive targets and the standard layout rules, noise monitoring points are determined, and measured noise data are obtained. The simulated noise data output by the intelligent agent model is compared with the measured noise data in multiple dimensions to calculate the model error and generate model verification parameters. Based on the model validation parameters, the agent model is iteratively trained using a self-supervised learning method to optimize its computational parameters. Based on the simulation results of the optimized agent model, it is determined whether the noise at the sensitive target exceeds the preset warning threshold. If so, traffic control suggestions are generated and pushed.

2. The method for optimizing the dynamic monitoring and environmental impact analysis model of urban integrated traffic noise according to claim 1, characterized in that, The construction of the intelligent agent model based on physical information neural networks includes: A multimodal architecture combining convolutional neural networks and a vision-language model is adopted to extract spatial environmental detail features in the three-dimensional digital environment and construct a multimodal perception and recognition subunit. A physical information neural network architecture based on graph neural networks is adopted, and acoustic physical equations are used as pre-constraints for network training to construct a noise field simulation and deduction sub-unit. A hybrid training strategy combining semi-supervised and self-supervised training is adopted. Supervised fine-tuning is performed using labeled measured noise verification data, and self-supervised training is performed using unlabeled data, thus constructing a self-supervised iterative training sub-unit.

3. The method for optimizing the dynamic monitoring and environmental impact analysis model of urban integrated traffic noise according to claim 1, characterized in that, The step of iteratively training the agent model using self-supervised learning based on the model validation parameters to optimize its computational parameters includes: Based on the model verification parameters, determine whether the preset iteration triggering conditions are met; if so, start the iterative optimization process. In response to the initiation of the iterative optimization process, training samples are extracted from the multi-source data fusion training library to construct a self-supervised training sample set; Input the self-supervised training sample set into the agent model and calculate the composite loss function; Based on the composite loss function, the operational parameters of the agent model are updated using the backpropagation algorithm, and it is determined whether the updated model satisfies the convergence condition. The agent model determined by convergence will be solidified and switched to the official running version.

4. The method for optimizing the dynamic monitoring and environmental impact analysis model of urban integrated traffic noise according to claim 3, characterized in that, In response to the initiation of the iterative optimization process, training samples are extracted from the multi-source data fusion training library to construct a self-supervised training sample set, including: Labeled training data is extracted from the model validation parameters to form a labeled sample set. The labeled training data includes the input feature vectors of each monitoring point and their corresponding simulation errors. Unlabeled training data within the same spatiotemporal range as the labeled training data are extracted from the multi-source data fusion training library to form an unlabeled sample set. The unlabeled training data includes input feature vectors in areas where no monitoring points are set. Spatiotemporal alignment and feature standardization are performed on all samples in the labeled sample set and the unlabeled sample set. The processed labeled and unlabeled sample sets are randomly divided into several small batches, which are used as training batches.

5. The method for optimizing the dynamic monitoring and environmental impact analysis model of urban integrated traffic noise according to claim 3, characterized in that, The step of updating the operational parameters of the agent model based on the composite loss function using the backpropagation algorithm, and determining whether the updated model satisfies the convergence condition, includes: The gradient of the composite loss function with respect to the operational parameters of the agent model is calculated using automatic differentiation techniques. An adaptive moment estimation optimizer is used to update the operational parameters of the agent model based on the gradient: The predicted noise values ​​of the same batch of verification points are recalculated using the updated operational parameters, and the updated global mean error and the maximum error of a single point are calculated. When the global mean error and the maximum error at a single point meet the preset conditions, the convergence condition is met, and the updated operation parameters are accepted; otherwise, the parameter update is rejected, and the operation parameters are restored to their previous state.

6. The method for optimizing the dynamic monitoring and environmental impact analysis model of urban integrated traffic noise according to claim 3, characterized in that, The process of solidifying the model based on convergence criteria and switching it to the official running version includes: The updated agent model was tested for accuracy and stability on an independent test dataset, and the mean error and standard deviation of the error on the test set were calculated. When the mean error and standard deviation of the error on the test set meet the test conditions, the updated core operation parameters of the model and its corresponding network architecture configuration file and normalization parameters are packaged, written into the model repository and assigned a new version number. Record the version metadata information that has been solidified this time. The version metadata information includes version number, solidification time, trigger type, number of training samples, training data time range, error comparison before and after update, composite loss value and number of iteration rounds. The pointer to the intelligent agent model called by the dynamic early warning system has been switched from the old version to the new version.

7. A system for dynamic monitoring and environmental impact analysis model optimization of urban integrated traffic noise, characterized in that, include: The first acquisition module is used to acquire high-precision map data of the target area, construct a three-dimensional spatial digital environment, and automatically identify and label sensitive targets in the acoustic environment based on a preset sensitive target feature library. The access module is used to access and parse the acoustic environment quality standards, extract the noise limit standards and technical constraints, and convert them into standardized data operation rules. The second acquisition module is used to acquire multi-source traffic flow data of the target area in real time. The construction module is used to construct an intelligent agent model based on a physical information neural network. The intelligent agent model is driven by the multi-source traffic flow data and constrained by the three-dimensional spatial digital environment. It dynamically simulates and deduces the three-dimensional spatial distribution intensity of the noise field in the target area and its impact on sensitive targets. The third acquisition module is used to determine the noise monitoring points and acquire measured noise data based on the distribution of the sensitive targets and the standard layout rules. The generation module is used to compare the simulated noise data output by the intelligent agent model with the measured noise data in multiple dimensions, calculate the model error, and generate model verification parameters. The self-supervised optimization module is used to iteratively train the agent model based on the model validation parameters and optimize its computational parameters through self-supervised learning. The dynamic early warning module is used to determine whether the noise at the sensitive target exceeds the preset early warning threshold based on the simulation results of the optimized intelligent agent model. If so, it generates and pushes traffic control suggestions.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for optimizing the dynamic monitoring and environmental impact analysis model of urban integrated traffic noise as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the method for optimizing the dynamic monitoring and environmental impact analysis model of urban integrated traffic noise as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for optimizing the dynamic monitoring and environmental impact analysis model of urban integrated traffic noise as described in any one of claims 1-6.

Images