Cell evaluation method and system based on noise map

Abstract

The invention discloses a cell evaluation method and system based on a noise map, and relates to the field of intelligent perception and Internet of Things, and the method comprises the steps: constructing a digital twin model of a cell, dividing the cell into a plurality of sub-regions through employing a trained multi-agent reinforcement learning model, and carrying out the processing of all sub-regions, outputting a dynamic noise distribution diagram of the cell, arranging the dynamic noise distribution diagram into a causal analysis data set, generating an initial causal relationship diagram through causal analysis software, collecting and identifying an implicit noise source in the initial causal relationship diagram, and adding the implicit noise source into the causal analysis data set to obtain a perfect causal relationship diagram; according to the method, the implicit noise source data is collected by analyzing the difference between the initial causal relationship graph and the dynamic noise distribution graph, and the perfect causal relationship graph is generated after the causal analysis data set is added, so that the implicit noise source is identified and integrated.

Description

Technical Field

[0001] The present invention relates to the field of intelligent perception and Internet of Things, and particularly to a method and system for evaluating a community based on a noise map. Background Art

[0002] As an increasingly prominent environmental problem in the process of urbanization, noise pollution has received wide attention. In recent years, with the rapid development of sensor technology, geographic information system, and artificial intelligence algorithms, significant progress has been made in noise monitoring and assessment technology.

[0003] However, there are still many problems in the existing community noise assessment. First of all, the existing digital twin models often rely on default values in parameter calibration and do not fully combine actual environmental data for dynamic adjustment, resulting in a deviation between the simulation results and the real noise distribution. In addition, the analysis of noise sources mostly focuses on obvious factors, and the identification of hidden noise sources is insufficient, which limits the comprehensiveness of causal relationships, making it difficult to accurately reflect the complex driving mechanism and regional differences of noise. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for evaluating a community based on a noise map to solve the problems of deviation between the simulation results and the real noise distribution and insufficient identification of hidden noise sources.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a method for evaluating a community based on a noise map, which includes: Constructing a digital twin model of the community; Using a trained multi-agent reinforcement learning model to divide the community into multiple sub-regions, processing all sub-regions, and outputting a dynamic noise distribution map of the community; Sorting the dynamic noise distribution map into a causal analysis data set and generating an initial causal relationship diagram through causal analysis software; Collecting and identifying hidden noise sources in the initial causal relationship diagram and adding them to the causal analysis data set to obtain a complete causal relationship diagram; Identifying key noise sources according to the causal strength in the complete causal relationship diagram, setting intervention measures and simulating the effects to obtain an optimal noise reduction plan; Combining the dynamic noise distribution map with the optimal noise reduction plan to obtain a comprehensive noise map, and then evaluating the noise situation based on the comprehensive noise map and generating a community score.

[0007] As a preferred solution of the method for evaluating a community based on a noise map according to the present invention, wherein: constructing a digital twin model of the community specifically includes the following steps: Collect the noise data, traffic flow data and meteorological data of the community to form a multi-dimensional data set; Establish and adjust the three-dimensional virtual model of the community through the multi-dimensional data set and the historical multi-dimensional data set.

[0008] As a preferred solution of the community evaluation method based on the noise map according to the present invention, wherein: use the trained multi-agent reinforcement learning model to divide the community into multiple sub-regions, and process all sub-regions to output the dynamic noise distribution map of the community, specifically including the following steps: Use the historical multi-dimensional data set to train the multi-agent reinforcement learning model, and divide the community into multiple sub-regions; Each sub-region is responsible for an agent, and the agent outputs the noise prediction value of the responsible sub-region; Combine the noise prediction values of the agents in each sub-region to form a dynamic noise distribution map.

[0009] As a preferred solution of the community evaluation method based on the noise map according to the present invention, wherein: the generation of the initial causal relationship diagram by the causal analysis software refers to using pcalg to generate the initial causal relationship diagram based on the causal analysis data set.

[0010] As a preferred solution of the community evaluation method based on the noise map according to the present invention, wherein: identify and collect the hidden noise sources in the initial causal relationship diagram and add them to the causal analysis data set to obtain a complete causal relationship diagram, specifically including the following steps: When it is identified that there are hidden noise sources in the initial causal relationship diagram, collect the hidden noise sources; Add the collected hidden noise sources to the causal analysis data set, and use pcalg to obtain a complete causal relationship diagram according to the new causal analysis data set.

[0011] As a preferred solution of the community evaluation method based on the noise map according to the present invention, wherein: identify the key noise sources according to the causal strength in the complete causal relationship diagram, set intervention measures and simulate the effects to obtain the optimal noise reduction plan, specifically including the following steps: Identify the key noise sources by analyzing the causal strength in the complete causal relationship diagram, and set intervention measures; Simulate the intervention measures, generate a comprehensive evaluation score table, and obtain the optimal noise reduction plan.

[0012] As a preferred solution of the community evaluation method based on the noise map according to the present invention, wherein: the re-evaluation of the noise condition based on the comprehensive noise map and the generation of the community score refer to analyzing the noise decibel value based on the comprehensive noise map using a spatial analysis algorithm to obtain the proportion of the exceeded area; Use the weighted scoring method to analyze the impact of the exceeded standard areas on residents according to the regional importance, and obtain the proportion of the weighted exceeded standard area. According to the proportion of the exceeded standard area and the proportion of the weighted exceeded standard area, generate the community score by linear deduction.

[0013] In a second aspect, the present invention provides a community evaluation system based on a noise map, including: A construction module for constructing a digital twin model of the community; A prediction module that uses the trained multi-agent reinforcement learning model to divide the community into multiple sub-regions, processes all sub-regions, and outputs the dynamic noise distribution map of the community; An analysis module that organizes the dynamic noise distribution map into a causal analysis data set and generates an initial causal relationship diagram through causal analysis software; A refinement module that collects and identifies the hidden noise sources in the initial causal relationship diagram and adds them to the causal analysis data set to obtain a refined causal relationship diagram; An identification module that identifies the key noise sources according to the causal intensity in the refined causal relationship diagram, sets intervention measures and simulates the effects, and obtains the optimal noise reduction plan; A scoring module that combines the dynamic noise distribution map with the optimal noise reduction plan to obtain a comprehensive noise map, and then evaluates the noise condition based on the comprehensive noise map and generates the community score.

[0014] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the community evaluation method based on the noise map as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the community evaluation method based on the noise map as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: The accurate construction of the three-dimensional virtual model of the community is realized by calibrating the digital twin model through the multi-dimensional dataset and the historical multi-dimensional dataset. The deviation between the simulation and the actual situation is eliminated through calibration, providing a reliable virtual mirror for noise prediction and distribution analysis. The accuracy of the three-dimensional virtual model of the community is greatly improved, which can accurately simulate the physical characteristics of the acoustic environment in the community, laying a solid foundation for subsequent analysis. In addition, by analyzing the differences between the initial causal relationship diagram and the dynamic noise distribution map, collecting the data of hidden noise sources, and generating a complete causal relationship diagram after adding them to the causal analysis dataset, the identification and integration of hidden noise sources are realized, filling the noise increment that cannot be explained by explicit factors, and providing a complete perspective for noise source analysis. The interpretability of the causal relationship diagram is also improved, making the description of the noise cause from partial to comprehensive, and avoiding analysis deviation caused by missing key factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of the community evaluation method based on the noise map in Embodiment 1.

[0019] Figure 2 It is a schematic diagram of the system module composition in Embodiment 1.

[0020] Figure 3 It is a schematic diagram of the digital twin model construction in Embodiment 1.

[0021] Figure 4 It is a schematic diagram of multi-agent collaborative prediction in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to make the above-mentioned objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings in the specification.

[0023] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0024] Second, the "one embodiment" or "embodiment" referred to herein means a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments.

[0025] Embodiment 1, referring to Figures 1 to 4 , this embodiment provides a cell evaluation method based on a noise map, including the following steps: S1. Construct a digital twin model of the cell.

[0026] Specifically, it includes the following steps: S1.1. Deploy a noise sensor network throughout the target cell. Select high-precision microphone array devices and arrange them at key positions, such as the boundaries near main roads, near the windows of residential buildings, internal green spaces, and community activity squares. Set up a monitoring point every 50 meters to cover the main areas of the cell in total. The noise sensors automatically collect noise data every 5 minutes, record the decibel values, and upload them to the cloud server in real time through the 5G network. At the same time, monitor the traffic flow data in real time through cameras, such as the number of vehicles passing through per hour, and obtain environmental data such as wind speed, wind direction, and humidity from the local weather station. The collected noise data, traffic flow data, and meteorological data will be aggregated to the cloud server to form a multi-dimensional data set containing timestamps and spatial positions.

[0027] Preferably, by deploying a noise sensor network throughout the target cell and combining traffic flow data and meteorological data, a real-time updated multi-dimensional data set is formed, providing rich basic information for constructing the digital twin model. And the noise data is collected every 5 minutes and quickly uploaded through the 5G network, ensuring the real-time and continuity of the data. Compared with traditional static noise measurement, this high-frequency and dynamic acquisition method can capture the instantaneous changes of noise, enabling the digital twin model to truly reflect the dynamic characteristics of the cell noise environment.

[0028] S1.2. Take the generated and real-time updatable multi-dimensional dataset as the input and start constructing the digital twin model of the cell under test. First, input the collected environmental variables such as noise data, traffic flow, wind speed, wind direction, humidity, etc., together with the geographical information file of the cell (including building layout, road distribution, and green space location) into a digital twin platform, such as using Unity or similar software. On the digital twin platform, create a three-dimensional virtual model of the cell, map the real-time noise data collected by the sensors to the corresponding spatial positions. For example, the monitoring points near the main road show higher decibel values, while those in the green space are lower, forming an initial noise distribution baseline. To ensure the accuracy of the three-dimensional virtual model of the cell, calibrate it using the noise data, traffic flow data, and meteorological data collected in the past 24 hours, and adjust the parameters in the three-dimensional virtual model of the cell. For example, there is a parameter in the three-dimensional virtual model of the cell called the sound attenuation rate, which represents the degree to which sound weakens with distance when propagating in the air. The default value may be set to 6 decibels attenuation per 100 meters. However, in the data of the past 24 hours, the actual noise at 100 meters beside the main road dropped from 60 decibels to 50 decibels (attenuation of 10 decibels), while the simulation in the three-dimensional virtual model of the cell only dropped to 54 decibels (attenuation of 6 decibels), with a difference of 4 decibels, exceeding the error range of 2 decibels. According to the actual data of the past 24 hours, adjust the sound attenuation rate to 10 decibels attenuation per 100 meters and run the simulation again. The result shows that the noise at 100 meters drops to 50 decibels, which is consistent with the measured value. This adjustment reflects that the air humidity or vegetation density in the cell may be higher than the default value, resulting in faster sound attenuation. Also, the building reflection coefficient in the three-dimensional virtual model of the cell represents the intensity of sound reflected back after hitting the building. The default value may be 0.8 (i.e., 80% of the sound is reflected). However, the data of the past 24 hours show that the noise in the area near a certain high-rise residential building is 65 decibels, but the simulation in the three-dimensional virtual model of the cell is 60 decibels, with a difference of 5 decibels. Analysis finds that in the actual environment, the building surface (such as glass curtain wall) reflects more sound, while the three-dimensional virtual model of the cell underestimates the reflection effect. So increase the building reflection coefficient from 0.8 to 0.95 (95% reflection). After re-simulation, the noise in this area rises to 64 decibels, and the difference from the actual value of 65 decibels is reduced to 1 decibel, meeting the error requirements. Such a modification and adjustment method takes into account the real reflection characteristics of the building materials in the cell. After such calibration is completed, a digital twin model is obtained, which is named the three-dimensional virtual model of the cell for appropriate use.

[0029] Furthermore, calibrating using the data collected in the past 24 hours and adjusting the parameters of the three-dimensional virtual model of the cell further reduces the error between the simulation value and the actual measurement value, significantly improving the reliability of the three-dimensional virtual model of the cell and providing a solid foundation for subsequent noise prediction and evaluation.

[0030] S2. Use the trained multi-agent reinforcement learning model to divide the community into multiple sub-regions, process all sub-regions, and output the dynamic noise distribution map of the community.

[0031] Specifically, it includes the following steps: S2.1 First, train the multi-agent reinforcement learning model.

[0032] First, extract the noise data, traffic flow data, and meteorological data of the past 7 days from the cloud server, which are accompanied by timestamps and spatial locations, forming a continuous time series that records the changes in community noise and environmental variables in the past 7 days. For example, the data may show that the noise at 8 am on weekdays is higher than at the same time on weekends, or that the noise in certain areas increases significantly when the wind speed increases.

[0033] Use the multi-agent reinforcement learning model to divide the community into multiple sub-regions. For example, each building or each road is used as an independent sub-region, and each sub-region is responsible for by an agent. Then clarify the tasks of each agent. The agent predicts the noise change trend of the sub-region in the next period (e.g., 2 hours) based on the input multi-dimensional data set (noise level, number of vehicles, wind speed, etc.). To enable the agent to learn how to complete the task of predicting the noise change trend of the sub-region in the next period (e.g., 2 hours), the training process simulates the real environment and allows the agent to trial and error and summarize the rules from the historical data. For example, an agent responsible for the main road sub-region needs to learn to identify the pattern that an increase in the number of vehicles will lead to an increase in noise. The training is carried out in a virtual environment, which is constructed based on the historical data of the past 7 days. Set a starting time, for example, starting from 00:00 on the first day, and then play the historical data step by step in chronological order. The goal of the agent is to predict the noise level in the next 2 hours as accurately as possible. For example, given the noise of 60 dB, the number of vehicles of 200, and the wind speed of 3 m / s at 08:00 on the first day, the agent needs to predict the noise value at 10:00 and compare it with the actual data (e.g., 62 dB). The goal of the training is to minimize the gap between the predicted value and the actual value, for example, controlling it within 1.5 dB. To motivate the agent to learn, set a reward mechanism: if the prediction is close to the actual value, the agent gets a high score; if the deviation is large, the score is low.

[0034] At the beginning of training, each agent does not know the pattern of noise change and can only randomly guess the future noise value based on the input data (noise, number of vehicles, wind speed). For example, the main road agent may first guess that the noise at 10:00 is 65 dB, while the actual value is 62 dB, with a deviation of 3 dB. Feedback (such as a low score) is given according to the size of the deviation, and the results of the attempts are recorded. As time progresses, the historical data is played from 00:00 on the first day to 23:55 on the seventh day, and the agent will repeatedly try to predict and adjust its strategy according to the actual data after each attempt. For example, it may find that for every 100 additional vehicles, the noise increases by an average of 2 dB, or for every 2 m / s increase in wind speed, the noise propagation range expands by 10 m. Through the attempts in the past seven days, the agent gradually summarizes the pattern of noise change. For example, when the number of vehicles surges at 8 am, the noise will increase by 5 dB, or the noise is stable when the wind speed is low at night.

[0035] Preferably, through repeated trial and error and the feedback mechanism, the agent gradually optimizes the predicted value to be close to the actual value. The dynamic and high-precision output significantly improves the reliability and foresight of the noise distribution map, providing more accurate data support for the management of the community environment.

[0036] Moreover, it should be considered that the noise distribution in the community is holistic. For example, the noise on the main road may spread to the residential buildings, so the agents also need to cooperate during training. Therefore, a sharing mechanism is set up so that each agent not only learns about its own sub-region but also refers to the prediction results of neighboring regions. For example, after the main road agent predicts a 5 dB increase in noise, it will notify the residential building agent, prompting the residential building agent to consider the propagation impact. By integrating the predictions of all agents, the strategy is adjusted to ensure the noise distribution throughout the community Figure 1 is consistent. For example, if the main road prediction is too high and the residential building prediction is too low, the results of the two are coordinated to optimize the overall accuracy. Cooperative training enables the multi-agent reinforcement learning model to capture the spatial propagation pattern of noise, such as how noise affects one sub-region from another when the wind direction changes.

[0037] Preferably, the cooperation mechanism between agents is used. By setting up a sharing mechanism, each agent not only focuses on its own sub-region but also refers to the prediction results of neighboring regions. Cooperative training ensures the consistency of the dynamic noise distribution map.

[0038] After the training of the multi-agent reinforcement learning model is completed, it starts to be put into use. Every 15 minutes, the agents in each sub-region will adjust their prediction strategies according to the latest multi-dimensional data set, such as real-time noise levels and traffic flows, and predict the noise change trend in this region within the next 2 hours. For example, the noise near the main road may increase due to the increase in vehicles, while the inner courtyard may remain at a relatively low level. During the prediction process, factors such as wind direction and building occlusion are considered to further refine the results. For example, when the wind speed increases, the noise may spread to a farther area, or the noise in some areas may be weakened due to the obstruction of high-rise buildings. The prediction results are completed through the collaboration of all agents, forming a complete dynamic noise distribution map for the next 2 hours. The dynamic noise distribution map not only shows the noise level of each sub-region, but also reflects the trend of noise propagation from high-noise sources (such as the main road) to low-noise areas (such as the inner courtyard). Finally, the dynamic noise distribution map is presented in the 3D virtual model of the community in the form of a heat map, where the red area indicates that the noise exceeds the standard, and the green area indicates quietness.

[0039] Furthermore, since the trained multi-agent reinforcement learning model can generate the latest dynamic noise distribution map according to the latest multi-dimensional data set and also reflect the noise propagation trend, it greatly improves the timeliness and practicality of the response.

[0040] S3. Organize the dynamic noise distribution map into a causal analysis data set and generate an initial causal relationship diagram through causal analysis software.

[0041] Specifically, it includes the following steps: Based on the dynamic noise distribution map, extract the predicted noise data (such as 65 decibels at 10:00 on the main road), and integrate the predicted noise data with relevant environmental variables (such as traffic flow, wind speed, and residents' activity time, such as square dance or construction period). The environmental variables are sourced from the sensor network, cameras (providing real-time traffic flow and wind speed), and community management records (providing activity time). Organize the predicted noise data and relevant environmental variables into a causal analysis data set, and ensure that each piece of data has a timestamp and a spatial location, such as the noise prediction value at a certain point on the main road at a certain moment corresponding to the number of vehicles in the same period.

[0042] S3.1 First, prepare a data set in the form of a table, for example, including columns such as predicted noise level, traffic flow, wind speed, and residents' activity time. Each row is a data point of a specific time and location. For example, at 10:00, on the east side of the main road, the predicted noise is 65 decibels, the number of vehicles is 300, the wind speed is 5 m / s, and there is construction activity. Subsequently, load the causal analysis data set into pcalg in R language (pcalg is an R package for causal structure learning, which can automatically infer causal relationship graphs from observational data. It is based on the PC algorithm and identifies the causal direction between variables through conditional independence tests). The software will automatically recognize each column in the table as a variable (also called a node). For example, traffic flow and noise level each become an independent variable.

[0043] After loading the causal analysis data set, pcalg will assume that there may be connections between all variables. For example, traffic flow may affect the noise level, wind speed may affect the noise level, or residents' activity time may also have an impact. Subsequently, it will analyze the statistical dependence between variables through statistical tests. Generally speaking, it is to see whether traffic flow and noise level always change together: when the number of vehicles increases from 200 to 300, does the noise increase from 60 decibels to 65 decibels? If the statistical dependence between variables repeatedly appears in the data, pcalg will consider that the two may be related. After finding the association between variables, pcalg further determines the direction of the association between variables. For example, the data may show that the increase in traffic flow always precedes the increase in noise level, and the traffic flow does not necessarily decrease after the noise level increases. The PC algorithm will then infer that traffic flow is the cause and noise level is the result, and thus determine that the arrow points from traffic flow to noise level. The inference that the arrow points from traffic flow to noise level is based on the timeliness and statistical laws of the data. For example, if the noise level changes following the change in wind speed, but not vice versa, the arrow points from wind speed to noise level. The PC algorithm will try to avoid cycles (such as A affecting B and B affecting A) to ensure that a clear causal network is generated. Moreover, during the generation process, pcalg will verify the reliability of each causal relationship through statistical tests. For example, check whether the change in traffic flow significantly affects the noise level. After passing the verification, an initial causal relationship graph will be obtained. However, the initial causal relationship graph currently only reflects the main association relationships between variables in the causal analysis data set and establishes the basic connection between noise and environmental variables, which is not yet perfect.

[0044] Further explanation: By integrating the predicted noise data with multi-source environmental variables into a causal analysis dataset and using the pcalg software to generate an initial causal relationship diagram, various factors affecting noise can be comprehensively captured. pcalg analyzes the dependencies between variables through statistical tests and infers the causal direction based on temporality and statistical laws, ensuring that the generated causal relationship diagram is not limited to a single variable but reflects the complex associations among multiple variables. In addition, the conditional independence test of pcalg verifies the reliability of each causal relationship, reducing the possibility of misjudgment.

[0045] S4. Collect and identify the hidden noise sources in the initial causal relationship diagram and add them to the causal analysis dataset to obtain a refined causal relationship diagram.

[0046] Specifically, it includes the following steps: As mentioned before, the generated initial causal relationship diagram only establishes the basic connection between noise and environmental variables. Now, start to identify the hidden noise sources and refine the causal analysis dataset.

[0047] First, analyze whether the existing causal relationships in the initial causal relationship diagram can fully explain the prediction results in the dynamic noise distribution map. For example, the initial causal relationship diagram may show that traffic flow and wind speed point to the noise level, but the predicted noise increases by 5 decibels, while the traffic flow and wind speed pointing to the noise level can only explain an increase of 3 decibels, and the remaining 2 decibels cannot be explained. This indicates the existence of hidden factors that are not directly measured, such as infrasound (low-frequency sound waves below 20 Hz) caused by building structure resonance. To verify the hypothesis, use a low-frequency microphone as a tool to detect in the potentially affected areas (such as near high-rise buildings). The detection method is to arrange low-frequency microphones at specific positions and continuously record for a period of time (such as several hours), measure whether there is infrasound, record the intensity (decibel value) and distribution (which areas are more obvious) of the infrasound. For example, it is found that the infrasound intensity near high-rise buildings is 30 decibels and is concentrated during the morning traffic peak. The newly collected infrasound data will be organized in tabular form, such as 10:00, west side of the high-rise building, infrasound 30 decibels, and then added to the causal analysis dataset, supplemented as a new column of infrasound level, so that the new causal analysis dataset contains both explicit variables (traffic flow, wind speed, etc.) and the newly added hidden variable (such as infrasound).

[0048] Then, the causal analysis dataset containing latent variables is input into pcalg in R language again, and causal inference is re-run using the PC algorithm. The specific process is as follows: First, analyze the statistical dependencies among all variables (including the newly added infrasound level), for example, check whether the noise level also increases when the infrasound intensity increases, and then infer the causal direction, such as infrasound resonance → noise level, and verify the reliability of the inferred causal direction through conditional independence tests (for example, confirm whether the influence of infrasound is independent of traffic flow). In multiple iterations, pcalg will update the causal relationship graph. For example, a new node "infrasound resonance" is added, and an arrow is drawn pointing to the noise level, indicating that infrasound contributes to the noise increment. Then, use statistical tests (such as significance tests) to continuously verify the accuracy of each relationship in the updated causal relationship graph. For example, combined with the prediction of the dynamic noise distribution map (the main road noise will increase by 5 decibels in the next 2 hours), analyze the proportion of the sources of the increment: perhaps 80% comes from traffic flow, 10% comes from wind speed, and 10% comes from infrasound resonance. After adjustment, the finally obtained causal relationship graph clearly shows the main sources of noise. The final causal relationship graph (i.e., the refined causal relationship graph) includes both explicit factors (traffic flow, wind speed) and latent influences (infrasound) are also identified.

[0049] Preferably, by analyzing the comparison between the initial causal relationship graph and the prediction results of the dynamic noise distribution map, identify the noise increment that cannot be fully explained by explicit factors (such as traffic flow and wind speed) (for example, 2 decibels out of 5 decibels are unexplained), and then supplement the data of latent noise sources (such as infrasound) to improve the causal analysis dataset. Input the updated causal analysis dataset into pcalg and iteratively generate the refined causal relationship graph, which significantly improves the explanatory power of the graph. Compared with the initial causal relationship graph that only reflects basic connections (such as traffic flow → noise level), the refined causal relationship graph not only includes explicit factors but also identifies latent influences (such as the contribution of infrasound to noise), making the description of the noise sources more comprehensive. This approach of supplementation ensures that the causal relationship graph can accurately reflect the true driving mechanism of community noise and avoids analysis biases caused by omitting key factors.

[0050] S5. Identify the key noise sources according to the causal strength in the refined causal relationship graph, set intervention measures and simulate the effects to obtain the optimal noise reduction plan.

[0051] Specifically, it includes the following steps: Use a visualization tool to analyze the refined causal relationship diagram and identify the key noise sources that have the greatest impact on the noise in the community. For example, it is found that traffic flow has the thickest arrow pointing to the noise level, and the data shows that for every 100 additional vehicles, the noise level increases by an average of 2 decibels. The arrows for wind speed and residents' activities are thinner, indicating less impact (for example, for every 2 m / s increase in wind speed, the noise only increases by 0.5 decibels). Confirm that traffic flow is the main cause, while wind speed and residents' activities are secondary factors.

[0052] It should be noted that since the refined causal relationship diagram consists of nodes and arrows, for example, traffic flow pointing to the noise level indicates that an increase in vehicles leads to a rise in noise. The thickness of the arrow or the marked value (such as the correlation coefficient) reflects the connection strength. And an impact threshold is set based on the noise change (for example, a contribution to the noise change > 1 decibel). The traffic flow (2 decibels) exceeds the impact threshold, while the wind speed (0.5 decibels) is below the impact threshold but still has an impact. The same is true for residents' activities and wind speed, which do not exceed the impact threshold. Factors that exceed the impact threshold are also identified as the main causes (i.e., key nodes).

[0053] After confirming that traffic flow is the main cause and wind speed and residents' activities are secondary factors, multiple intervention measures need to be set. Specifically, they should be formulated according to the key noise sources and combined with domain knowledge. Taking traffic flow as the main cause and wind speed and residents' activities as secondary factors as examples, first, install sound insulation screens. Set up sound insulation screens 2 meters high beside the main roads, which are expected to block 50% of the traffic noise. Then, limit the vehicle speed. Reduce the vehicle speed on the main roads from 50 km / h to 40 km / h during peak hours to reduce vehicle noise. Finally, increase the green belt in the upwind direction (such as planting trees 3 meters high) to weaken the amplification effect of the wind on the noise. The mentioned intervention measures should ensure coverage of both main and secondary factors.

[0054] Preferably, use a visualization tool to analyze the refined causal relationship diagram and identify the key noise sources that have the greatest impact on the noise in the community. The subsequent intervention measures (such as installing sound insulation screens, limiting vehicle speed, and adjusting the green belt) are directly set for the key nodes, combined with the knowledge in the field of noise control, ensuring the pertinence of the measures. Moreover, with the clear guidance of the refined causal relationship diagram, resources are concentrated to address the main driving factors (such as traffic flow), significantly improving the scientific nature and efficiency of the noise reduction measures.

[0055] After obtaining the intervention measures, each intervention measure is input into R language one by one. Based on the improved causal relationship diagram and combined with Monte Carlo simulation, the causal effect simulation algorithm is used to simulate the effect of the intervention measure. The complete process is to first input the intervention hypothesis. For example, the vehicle speed is reduced by 20% (that is, the vehicle speed is reduced from 50 km / h to 40 km / h as mentioned before), and the value of the traffic flow node is adjusted to reduce the vehicle noise contribution by 20%. Then, the causal effect simulation algorithm calculates the noise change according to the arrow relationship in the improved causal relationship diagram (for example, traffic flow → noise level, with an intensity of 2 dB / 100 vehicles). For example, assuming that the number of vehicles during the peak period is 500, after a 20% reduction, it becomes 400, a decrease of 100 vehicles, and the noise drops by 2 dB. Finally, the secondary effects are analyzed, and the changes in other nodes are checked. For example, the sound insulation screen may change the influence path of the wind speed (the wind flow is blocked, and the noise propagation range is reduced), and the secondary effect (such as an additional 0.5 dB drop in the courtyard noise) is simulated through the interaction relationship in the improved causal relationship diagram. The causal effect simulation algorithm will run multiple times (such as 1000 times of Monte Carlo simulation), considering the uncertainty of variables (such as vehicle number fluctuations), to obtain the average effect of each measure. For example, for the sound insulation screen, the overall noise reduction is 5 dB, and the secondary effect is a 0.3 dB increase in the courtyard; for speed limit, the overall noise reduction is 3 dB, and there is no secondary effect; for green belt adjustment, the overall noise reduction is 3 dB, and the courtyard benefits first. And the effects of these noise reduction measures are sorted into a table and saved (that is, the preliminary evaluation result table).

[0056] Through the preliminary evaluation result table, compare the noise reduction effect and implementation cost of each measure. For example, for the sound insulation screen, the noise reduction is 5 dB, the construction cost is about 500,000 yuan (2 meters high, covering 200 meters on the east side of the main road), and the maintenance cost is 50,000 yuan per year; for speed limit, the noise reduction is 3 dB, and the implementation cost is almost zero (only need to adjust traffic rules and set signs), and there is no maintenance cost; for green belt adjustment, the noise reduction is 3 dB, the planting cost is about 100,000 yuan (3-meter-high trees, covering 100 meters upstream of the wind direction), and the maintenance cost is 20,000 yuan per year.

[0057] After comparing the noise reduction effects and implementation costs of each measure, the multi-criteria decision analysis method is used to comprehensively consider three dimensions: noise reduction effect, implementation cost, and feasibility (specially noted that the dimensions can be increased or decreased according to actual needs and are not fixed). Among them, the noise reduction effect is extracted from the simulation results. For example, the sound insulation screen is 5 decibels, and the speed limit is 3 decibels; the implementation cost is estimated in combination with the budget, including initial investment and long-term maintenance costs; the feasibility is to evaluate potential problems through research. For example, speed limits may cause traffic congestion (a 10% increase in vehicle queues during peak hours), the sound insulation screen needs to comply with the community plan (approval is required for occupying green spaces), and the adjustment of the green belt may be restricted by seasons (planting needs to be carried out in spring). After determining the dimensions, values and weights should be assigned to each dimension (for example, the effect accounts for 50%, the cost accounts for 30%, and the feasibility accounts for 20%), and the scores of each measure are calculated. For example, for the sound insulation screen, the effect score is 5 (full score 5), the cost score is 2 (expensive, on a 5-point scale), and the feasibility score is 4 (approval required), and the total score = 5×0.5 + 2×0.3 + 4×0.2 = 3.9; for the speed limit, the effect score is 3, the cost score is 5 (low cost), and the feasibility score is 3 (congestion risk), and the total score = 3×0.5 + 5×0.3 + 3×0.2 = 3.6. For the green belt planting, the effect score is 3, the cost score is 4 (medium), and the feasibility score is 4 (seasonal restriction), and the total score = 3×0.5 + 4×0.3 + 4×0.2 = 3.5. And it is organized into a table (that is, the comprehensive evaluation score table).

[0058] The solution with the highest score in the comprehensive evaluation score table is used as the optimal noise reduction solution. At the same time, a causal inference algorithm (such as do-calculus) is used to generate counterfactual predictions. The content is that without intervention, the noise increases by 2 decibels (from 65 decibels to 67 decibels), and after intervention, it drops to 61 decibels, that is, a reduction of 4 decibels.

[0059] To further illustrate, by using the solution with the highest comprehensive evaluation score (such as the sound insulation screen) as the optimal noise reduction solution and using the causal inference algorithm (do-calculus) to generate counterfactual predictions (without intervention, the noise increases by 2 decibels to 67 decibels, and after intervention, it drops to 61 decibels, that is, a reduction of 4 decibels), it provides a strong basis for comparison in decision-making. The counterfactual prediction shows the consequences of not taking measures (noise increase), which is in sharp contrast to the effect after intervention (a 4-decibel noise reduction), proving the necessity of the optimal noise reduction solution. The prediction based on the improved causal relationship diagram not only enhances the persuasiveness of the optimal noise reduction solution but also provides managers with an intuitive analysis of the consequences of whether to intervene.

[0060] S6. Combine the dynamic noise distribution map with the optimal noise reduction solution to obtain a comprehensive noise map, and then evaluate the noise situation based on the comprehensive noise map and generate a community score.

[0061] Specifically, it includes the following steps: Use GIS tools (geographic information system tools) to overlay the dynamic noise distribution map with the optimal noise reduction solution. In layman's terms, the optimal noise reduction solution is applied to the corresponding area on the dynamic noise distribution map, for example, the main road is reduced from 65 decibels to 61 decibels, generating a new noise value. At the same time, adjust the color of the heat map according to the decibel range, for example, 65 decibels is red, 61 decibels is yellow, and 45 decibels is green, mark the improvement area (such as the change from red to yellow near the main road), and obtain a comprehensive noise map.

[0062] After the application is completed, according to the national standard GB3096-2008 (acoustic environment quality standard), the spatial analysis algorithm (raster analysis) is used to calculate the proportion of the area where the noise exceeds the standard in the community. The specific steps are as follows: define the noise exceeding standard threshold, and the daytime standard in the residential area is 55 decibels (determined according to the acoustic environment quality standard). Convert the comprehensive noise map into a spatial data format and divide it into regular grids, for example, each grid unit is 5 meters × 5 meters (area 25 square meters). Assuming that the total area of ​​the community is 1000 square meters, it is divided into 40 grids (1000÷25=40). Each grid is assigned a value according to the comprehensive noise map, for example, the grid near the main road is 61 decibels, and the courtyard grid is 45 decibels. Rasterization converts the continuous noise distribution into discrete spatial units for easy statistics. Traverse all grids and find the units with noise values ​​exceeding 55 decibels. For example, the 4 grid values ​​near the main road are 61 decibels, which exceeds the standard; the 36 grid values ​​in the courtyard are 45 decibels, which meet the standard. The area of ​​each grid is 25 square meters, and the total number of grids that exceed the standard is multiplied by the area of ​​a single grid. For example, 4 grids with excessive noise × 25 square meters = 100 square meters. Divide the excessive area by the total area of ​​the community to get the percentage. For example, 100 square meters ÷ 1000 square meters = 10%. This means that 10% of the area in the community has excessive noise, mainly concentrated near the main roads.

[0063] However, the simple proportion of the exceeded area cannot fully reflect the impact of noise on residents. Therefore, the weighted scoring method needs to be used to consider the importance of different regions. For example, the exceedance in areas with frequent activities such as kindergartens and residential buildings requires more attention. First, the community is divided into key regions (such as kindergartens and residential buildings) and ordinary regions (such as the edge of the road). Suppose the kindergarten occupies 2 grids (50 square meters), and other exceeded areas beside the main road occupy 2 grids (50 square meters). Then, weights are set according to the importance of residents' activities. The weight for the exceeded kindergarten is set to 2 (highly sensitive), and the weight for other regions is set to 1 (generally sensitive). Next, weighted calculations are carried out. For example, the exceeded area of the kindergarten is 2 grids × 25 square meters × weight 2 = 100 weighted square meters; the exceeded area of other parts beside the main road is 2 grids × 25 square meters × weight 1 = 50 weighted square meters; the total weighted exceeded area is 100 + 50 = 150 weighted square meters. The weighted proportion is 150 ÷ (1000 × 1) = 15% (if the total area is standardized by weight 1). The 15% after weighting is higher than the simple area proportion of 10%, reflecting the severity of the exceedance in the kindergarten. Based on this, the impact degree of the exceedance is obtained as 15%, emphasizing the priority of treatment in key regions.

[0064] Preferably, by using the weighted scoring method and considering the importance of different regions (such as the weight of the kindergarten is 2 and the weight of other regions is 1), the comprehensiveness of the evaluation and the sensitivity to the actual impact on residents are further improved. For example, the 50 square meters of exceeded area in the kindergarten is 100 weighted square meters after weighting, the total weighted exceeded area is 150 weighted square meters, and the proportion is 15%, which is higher than the simple area proportion of 10%. The weighted scoring method highlights the priority of the exceedance in key regions (such as kindergartens and residential buildings), reflecting the true impact degree of noise on residents' lives.

[0065] Now, a quantitative score needs to be given to the community. First, set the scoring rules. The full score is set to 10 points, representing the ideal state, that is, the noise in all regions of the community is lower than the GB3096-2008 standard (55 decibels) and there is no exceedance. 1 to 3 points indicate serious exceedance, 4 to 6 points indicate relatively serious exceedance, 7 to 9 points indicate mild exceedance, and greater than 9 points and less than 10 points indicate approaching compliance. The deduction rule is to deduct points linearly according to the proportion of the exceeded area. For every 10% increase in the exceeded area, 1 point is deducted. This is because the exceeded area directly reflects the scope of the noise problem, and the higher the proportion, the worse the environmental quality. An additional deduction also needs to be added, which means considering the importance of key regions. For example, the exceedance near the kindergarten mentioned earlier has a greater impact on residents. Therefore, an additional deduction item is set to reflect the priority of key regions. Example: 10% exceedance deducts 1 point, kindergarten exceedance deducts 0.5 points, score = 10 - 1 - 0.5 = 8.5 points. This 8.5 is the quantitative score of the community and also corresponds to mild exceedance. Thus, the evaluation of the community is completed.

[0066] This embodiment also provides a cell evaluation system based on a noise map, including: Build modules to construct a digital twin model of the community; The prediction module uses the trained multi-agent reinforcement learning model to divide the cell into multiple sub-areas, processes all sub-areas, and outputs the dynamic noise distribution map of the cell; The analysis module organizes the dynamic noise distribution map into a causal analysis data set and generates an initial causal relationship diagram through the causal analysis software; The improvement module collects and identifies the hidden noise sources in the initial causal relationship graph and adds them to the causal analysis data set to obtain a perfect causal relationship graph; The identification module identifies the key noise sources according to the causal strength in the improved causal relationship diagram, designs intervention measures and simulates the effects to obtain the optimal noise reduction solution; The scoring module combines the dynamic noise distribution map with the optimal noise reduction solution to obtain a comprehensive noise map, and then evaluates the noise conditions based on the comprehensive noise map and generates a cell score.

[0067] This embodiment also provides a computer device, which is applicable to the case of a cell evaluation method based on a noise map, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to implement the cell evaluation method based on the noise map proposed in the above embodiment.

[0068] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0069] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for evaluating a community based on a noise map as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, a magnetic disk or an optical disc.

[0070] In summary, the present invention realizes the accurate construction of a three-dimensional virtual model of a community through a multidimensional dataset and a historical multidimensional dataset to calibrate a digital twin model. By calibration, the deviation between simulation and reality is eliminated, providing a reliable virtual mirror for noise prediction and distribution analysis. It greatly improves the accuracy of the three-dimensional virtual model of the community, enabling accurate simulation of the physical characteristics of the community acoustic environment and laying a solid foundation for subsequent analysis; in addition, by analyzing the differences between the initial causal relationship diagram and the dynamic noise distribution diagram, collecting hidden noise source data, and generating a complete causal relationship diagram after adding it to the causal analysis dataset, the identification and integration of hidden noise sources are realized, filling the noise increment that cannot be explained by explicit factors and providing a complete perspective for noise source analysis. It also improves the interpretability of the causal relationship diagram, making the description of the noise cause from partial to comprehensive and avoiding analysis deviation caused by missing key factors.

[0071] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A cell evaluation method based on noise map, characterized in that: include: Build a digital twin model of the community; Use the trained multi-agent reinforcement learning model to divide the community into multiple sub-areas, process all sub-areas, and output the dynamic noise distribution map of the community; The dynamic noise distribution map is organized into a causal analysis data set, and an initial causal relationship diagram is generated by the causal analysis software; Collect and identify the hidden noise sources in the initial causal relationship graph and add them to the causal analysis data set to obtain a complete causal relationship graph; According to the causal strength in the improved causal relationship diagram, the key noise sources are identified, intervention measures are set and the effects are simulated to obtain the optimal noise reduction solution; The dynamic noise distribution map is combined with the optimal noise reduction solution to obtain a comprehensive noise map. The noise condition is then evaluated based on the comprehensive noise map and a cell score is generated.

2. The cell evaluation method based on noise map according to claim 1, characterized in that: Building a digital twin model of a community includes the following steps: Collect noise data, traffic flow data and meteorological data of the community to form a multi-dimensional data set; Establish and adjust the three-dimensional virtual model of the community through multi-dimensional data sets and historical multi-dimensional data sets.

3. The cell evaluation method based on noise map as claimed in claim 2, characterized in that: The trained multi-agent reinforcement learning model is used to divide the cell into multiple sub-areas, and all sub-areas are processed to output the dynamic noise distribution map of the cell. The specific steps include the following: Use historical multidimensional datasets to train a multi-agent reinforcement learning model and divide the community into multiple sub-regions; Each sub-region is managed by an agent, which outputs the noise prediction value of the sub-region it is responsible for; The noise prediction values ​​of the agents in each sub-area are combined to form a dynamic noise distribution map.

4. The cell evaluation method based on noise map as claimed in claim 3, characterized in that: The generating of the initial causal relationship diagram by the causal analysis software refers to generating the initial causal relationship diagram by using pcalg based on the causal analysis data set.

5. The cell evaluation method based on noise map as claimed in claim 4, characterized in that: Identify and collect the hidden noise sources in the initial causal relationship diagram and add them to the causal analysis data set to obtain a complete causal relationship diagram, which specifically includes the following steps: In the case where an implicit noise source is identified in the initial causal relationship diagram, the implicit noise source is collected; The collected implicit noise sources are added to the causal analysis data set, and causal inference is performed using pcalg based on the new causal analysis data set to obtain a complete causal relationship diagram.

6. The cell evaluation method based on noise map as claimed in claim 5, characterized in that: According to the causal strength in the improved causal relationship diagram, the key noise sources are identified, intervention measures are set and the effects are simulated to obtain the optimal noise reduction solution, which specifically includes the following steps: Identify key noise sources by analyzing the causal strength in the refined causal relationship diagram and set intervention measures; The intervention measures are simulated, a comprehensive evaluation score sheet is generated, and the optimal noise reduction solution is obtained.

7. The cell evaluation method based on noise map according to claim 6, characterized in that: The said evaluating the noise condition based on the comprehensive noise map and generating a cell score means analyzing the noise decibel value based on the comprehensive noise map using a spatial analysis algorithm to obtain the percentage of the area exceeding the standard; The weighted scoring method was used to analyze the impact of the over-standard areas on residents according to the regional importance, and the weighted over-standard area ratio was obtained; Based on the proportion of area exceeding the standard and the proportion of weighted area exceeding the standard, a linear deduction method is used to generate a community score.

8. A cell evaluation system based on noise maps, based on the cell evaluation method based on noise maps according to any one of claims 1 to 7, characterized in that: include: Build modules to construct a digital twin model of the community; The prediction module uses the trained multi-agent reinforcement learning model to divide the cell into multiple sub-areas, processes all sub-areas, and outputs the dynamic noise distribution map of the cell; The analysis module organizes the dynamic noise distribution map into a causal analysis data set and generates an initial causal relationship diagram through the causal analysis software; The improvement module collects and identifies the hidden noise sources in the initial causal relationship graph and adds them to the causal analysis data set to obtain a perfect causal relationship graph; The identification module identifies the key noise sources according to the causal strength in the perfect causal relationship diagram, sets intervention measures and simulates the effects to obtain the optimal noise reduction solution; The scoring module combines the dynamic noise distribution map with the optimal noise reduction solution to obtain a comprehensive noise map, and then evaluates the noise conditions based on the comprehensive noise map and generates a cell score.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the cell evaluation method based on the noise map according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the cell evaluation method based on the noise map described in any one of claims 1 to 7 are implemented.

Images