Atmospheric Pollutant Tracing and Investigation Method, Terminal and System Based on Intelligent Street Lights

Abstract

The present invention discloses a method, a terminal and a system for tracing and investigating atmospheric pollutants based on intelligent street lamps, including a smart lamp post sensing module, a 5G transmission module, a model calculation module, and a GIS visualization platform. The smart lamp post sensing module includes various sensors for collecting atmospheric pollution sensing information coupled with longitude and latitude coordinates; the 5G transmission module includes a communication base station, a synchronous clock source and an intelligent gateway deployed on the smart lamp post for transmitting atmospheric pollution sensing information and longitude and latitude coordinate information with a unified timestamp; the model calculation module is used to construct a basic wind field model; fit a pollutant range model through monitoring data; calculate the pollutant diffusion center; the GIS visualization platform is used to visually output the three-dimensional position of the pollution source. This automatic detection algorithm can be used for environmental detection in different places such as chemical industrial parks, parks, and communities. It can quickly locate pollutants, save manpower and material resources, and ensure environmental safety.

Description

Technical Field

[0001] The present invention belongs to the field of environmental monitoring devices, and particularly relates to a method, a terminal, and a system for tracing and investigating atmospheric pollutants based on intelligent street lamps. Background Art

[0002] Since there are a large number of ordinary people near places such as chemical industrial parks, parks, and communities, rapid evacuation is required in case of an accident, and there are extremely high requirements for immediate environmental information during accident handling. It has become an important problem to be solved urgently to use more flexible and efficient devices and technologies for detection instead of sending people to the front line for investigation.

[0003] Most of the existing pollutant detection systems rely on manpower. Most of the existing intelligent street lamps only detect atmospheric pollutants such as PM2.5. Although the existing rescue robots have relatively complete functions, they still need to basically perceive the rescue environment in advance and their functions are relatively single.

[0004] In summary, the existing technology has certain defects and inconveniences in the above usage scenarios, so it is necessary to improve. Summary of the Invention

[0005] The technical problem to be solved by the present invention is: to provide a method for tracing and investigating atmospheric pollutants based on intelligent street lamps, which solves the problems of single function, slow speed, and low efficiency of regional air pollution detection technology in the existing technology.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] A method for tracing and investigating atmospheric pollutants based on intelligent street lamps includes the following steps:

[0008] Step 1: Use the sensing module installed on the smart light pole to obtain various sensor data around each smart light pole, assign time stamps and position stamps, and construct a distributed sensing network based on the smart light pole;

[0009] Step 2: Each smart light pole constructs an original convolutional neural network model, and performs deep learning on the convolutional neural network model of its own light pole in combination with its own sensor data;

[0010] Step 3: Use a Gaussian correction model based on deep learning to predict the sensor data of neighboring smart light poles, and obtain and output the surrounding atmospheric concentration and pollution source prediction information;

[0011] Step 4: Update the training weight values of its own convolutional neural network model according to the neighbor sensor data values, use the current own sensor data to train the convolutional neural network model of each, and update the parameters in the Gaussian correction model based on deep learning;

[0012] Step 5: Repeat Steps 3 to 4.

[0013] The sensor data includes, but is not limited to, location, time, wind speed, wind direction, rainfall, air humidity, and concentrations of various pollutants.

[0014] The specific representation of the Gaussian correction model based on deep learning is as follows:

[0015] R i =(b, V i , C ij ), where R i is the relative distance between the smart light pole D i and the pollutant diffusion center, b = (V i , H i , f) are the parameters inside the Gaussian correction model based on deep learning, V i is the wind speed detected by the smart light pole D i , H i is the humidity detected by the smart light pole D i , f is whether there is rainfall, C ij is the concentration of the j-th pollutant detected by the smart light pole D i , and both i and j are positive integers.

[0016] Calculate the pollutant emission Q per unit time according to the following formula:

[0017] Q = 2πC ij * V * σ y * σ z / {e^(-y 2 / 2σ y 2 ) * {e^[-(z - H) 2 / 2σ y 2 + e^[-(z + H) 2 / 2σ z 2}}

[0018] where V is the average wind speed at the emission port, σ y is the lateral diffusion coefficient, σ z is the vertical diffusion coefficient, y is the horizontal distance from the pollutant diffusion center to the nearest light pole in the downwind direction, z is the height of the light pole required for y from the ground, and H is the height of the pollutant diffusion center.

[0019] The initial value of the parameter b in the Gaussian correction model based on deep learning is obtained from the collection of historical big data. During the training process, it is a dynamically updated value, which is calculated based on the data of neighboring sensors. And after each round of training, the current sensor data of at least 3 smart light poles with the closest distance are obtained for verification.

[0020] In order to further solve the problems of single function and inconvenient detection of regional air pollution detection equipment, the present invention also provides an air pollutant traceability and investigation terminal and system based on intelligent street lights. The specific technical solutions are as follows:

[0021] An air pollutant traceability and investigation terminal based on intelligent street lights includes a data processing module and a communication module. The data processing module executes the air pollutant traceability and investigation method described above; the communication module is used to realize data interaction.

[0022] An air pollutant traceability and investigation system based on intelligent street lights includes a visualization platform and several air pollutant traceability and investigation terminals. Among them, the visualization platform serves as the central server, and the air pollutant traceability and investigation terminal serves as a sub-server; data interaction is carried out between the central server and all sub-servers, and each sub-server has the function of independently analyzing air pollution data and self-organizing a network with neighboring sub-servers.

[0023] The central server controls the switches of each sub-server and provides an initial global model to each sub-server; during operation, it receives the data sent by each sub-server, processes it, and sends training weight values to each sub-server;

[0024] The sub-server collects air data with timestamps and spatial stamps of its own, receives the detection data of neighboring sub-servers, calculates and adjusts the training weights according to the training weight values provided by the central server and the detection data of neighboring sub-servers, retrains and updates the parameters in the model, and returns the processed data model to the central server.

[0025] The central server receives the pollution source prediction information of multiple sub-servers, constructs a federated learning neural network model, globally searches for pollution sources within the coverage range of smart light poles, determines the final pollution source location, and feeds back the results to the convolutional neural network models of each smart light pole.

[0026] When the sub-server sends data to the central server, it first compresses the data into the form of data packets and then sends them to the central server multiple times.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. This method assigns timestamps and location stamps to various types of sensing information, and realizes data sharing and interaction in a distributed sensing network based on the TCP / IP handshake protocol. Based on the Gaussian wind point source diffusion model, a deep learning model is constructed by combining the multi-modal sensing big data of each lamp post to estimate the surrounding concentration and predict and track pollution sources. The original deep learning models of their respective lamp posts are trained through the multi-modal sensing big data of neighbor node lamp posts to improve the accuracy of surrounding concentration estimation and pollution source prediction and tracking.

[0029] 2. By installing relevant sensors on intelligent street lamps, the chain of information acquisition, processing, and decision-making is shortened, the real-time performance of execution control is improved, and at the same time, the convenience of intelligent street lamps is utilized to facilitate real-time detection and provide environmental data.

[0030] 3. The sub-server and the central server perform data interaction in real time. The central server provides the initial model for the sub-server and feedbacks the training weights, making the detection and prediction of air pollution by the sub-server more accurate.

[0031] 4. This automatic detection algorithm and terminal system can be used for environmental detection in different places such as chemical industrial parks, parks, and communities. It can quickly locate pollutants, save manpower and material resources, and ensure environmental safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 FIG. is a schematic diagram of the module connection of the air pollutant tracing and investigation system based on intelligent street lamps of the present invention.

[0033] Figure 2 FIG. is a schematic diagram of the geographical location distribution of the air pollutant tracing and investigation system based on intelligent street lamps of the present invention.

[0034] Among them, the markings in the figure are: 01-28 - ordinary lamp post; 001-004 - intelligent lamp post. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] In order to enable those skilled in the art of this technology to better understand the solution of this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0036] The present invention relates to an atmospheric pollutant distributed monitoring system and a traceability and investigation method based on a smart light pole, including a smart light pole sensing module, a 5G transmission module, a model calculation module, and a GIS visualization platform. The smart light pole sensing module includes a high-precision positioning sensor, a wind speed and direction sensor, and a pollutant monitor, which are used to collect atmospheric pollution sensing information coupled with longitude and latitude coordinates. The 5G transmission module includes a communication base station, a synchronous clock source, and an intelligent gateway deployed on the smart light pole, which are used to transmit atmospheric pollution sensing information and longitude and latitude coordinate information with a unified timestamp. The model calculation module is used to construct a basic wind field model, fit a pollutant range model through monitoring data, and calculate the pollutant diffusion center. The GIS visualization platform is used to intuitively output the three-dimensional position of the pollution source.

[0037] An atmospheric pollutant traceability and investigation method based on a smart street lamp includes the following steps:

[0038] Step 1: Use the sensing module installed on the smart light pole to obtain various sensor data around each smart light pole, assign a timestamp and a position stamp, and construct a distributed sensing network based on the smart light pole.

[0039] Step 2: Each smart light pole constructs an original convolutional neural network model and performs deep learning on the convolutional neural network model of its own light pole in combination with its own sensor data.

[0040] Step 3: Use a Gaussian correction model based on deep learning to predict the sensor data of neighboring smart light poles, and obtain and output the surrounding atmospheric concentration and pollution source prediction information.

[0041] Step 4: Update the training weight value of its own convolutional neural network model according to the neighbor sensor data value, use the current own sensor data to train the convolutional neural network model of each, and update the parameters in the Gaussian correction model based on deep learning.

[0042] Step 5: Repeat steps 3 to 4.

[0043] An atmospheric pollutant traceability and investigation terminal based on a smart street lamp includes a data processing module and a communication module. The data processing module executes the above-mentioned atmospheric pollutant traceability and investigation method; the communication module is used to realize data interaction.

[0044] An atmospheric pollutant traceability and investigation system based on a smart street lamp includes a visualization platform and several atmospheric pollutant traceability and investigation terminals. Among them, the visualization platform serves as the central server, and the atmospheric pollutant traceability and investigation terminal serves as the sub-server; data interaction is carried out between the central server and all sub-servers, and each sub-server has the function of independently analyzing atmospheric pollution data and self-organizing a network with neighboring sub-servers.

[0045] Specific embodiments are asFigure 1 , Figure 2 as shown in

[0046] An air pollutant traceability and investigation system based on intelligent street lights, including a visualization platform and several air pollutant traceability and investigation terminals. Each air pollutant traceability and investigation terminal is installed on a smart lamp post, and the smart lamp posts are sequentially set as D1, D2, …, D i , …, D n , where both i and n are positive integers. The visualization platform serves as the central server, and the air pollutant traceability and investigation terminals serve as sub-servers; data interaction occurs between the central server and all sub-servers, and each sub-server has the function of independently analyzing air pollution data and self-organizing a network with neighboring sub-servers.

[0047] The air pollutant traceability and investigation terminal includes a smart lamp post sensing module, a 5G transmission module, and a model calculation module. Among them, the smart lamp post sensing module includes, but is not limited to, high-precision positioning sensors, timers, wind speed and direction sensors, multi-component pollutant monitors, rainfall, air humidity, various pollutant concentration devices, etc., for collecting big data on air pollution sensing that couples spatio-temporal parameters and the near-surface atmospheric dynamics state; the model calculation module includes a Gaussian correction model and a convolutional neural network model based on deep learning, with edge AI computing capabilities, processes the data collected by the smart lamp post sensing module according to the process of the air pollutant traceability and investigation method, and outputs the processing results to the visualization platform through the 5G transmission module.

[0048] The 5G transmission module includes a communication base station, a synchronous clock source, and an intelligent gateway deployed on the smart lamp post, for transmitting information such as near-surface air pollutant data, high-dimensional semantic extraction features, and pollution source inference results that couple spatio-temporal dynamics information under a unified timestamp, realizing the interconnection and interoperability of different regional smart lamp posts as sensing nodes, and constructing a distributed intelligent sensing network driven by data sharing.

[0049] The visualization platform is a GIS visualization platform, which accesses the distributed sensing network of smart lamp posts and intuitively outputs the location of pollution sources and the long-term dynamic diffusion process of different air pollutants by using the method of digital twin. Through the method of combining distributed spatio-temporal data federated learning with the near-surface air pollution diffusion model, pollutant diffusion deduction and traceability are realized.

[0050] The traceability and investigation method of the air pollutant traceability and investigation system specifically includes the following steps:

[0051] First step, apply the installation on smart lamp posts D1, D2, …, D i , …, D nThe perception module on it obtains various sensor data around each smart light pole, assigns timestamps and location stamps, and constructs a distributed perception network between the smart light pole and the GIS visualization platform; and realizes data sharing and interaction of the distributed perception network based on the TCP / IP handshake protocol.

[0052] Second, the GIS visualization platform sends the initial convolutional neural network model and the Gaussian correction model based on deep learning to the terminals of each smart light pole. The terminals on each smart light pole construct the original convolutional neural network model and perform deep learning on the convolutional neural network models of their respective light poles in combination with their own sensor data.

[0053] Third, apply the Gaussian correction model based on deep learning to predict the sensor data of neighboring smart light poles, and obtain and output the predicted information of the surrounding atmospheric concentration and pollution sources.

[0054] During prediction, the wind speed, humidity, and whether it is raining in the current environment will all affect the Gaussian correction model based on deep learning, and they are all in a state of non-linear change. The specific representation of its Gaussian correction model based on deep learning is as follows:

[0055] R i =(b, V i , C ij ), where R i is the relative distance between the smart light pole D i and the pollutant diffusion center, b=(V i , H i , f) are the parameters inside the Gaussian correction model based on deep learning, V i is the wind speed detected by the smart light pole D i , H i is the humidity detected by the smart light pole D i , f is whether it is raining, C ij is the concentration of the j-th pollutant detected by the smart light pole D i , and both i and j are positive integers. The initial value of the parameter b inside the Gaussian correction model based on deep learning comes from historical big data collection, is a dynamically updated value during the training process, is calculated based on the neighbor sensor data, and after each round of training is completed, the current sensor data of at least 3 smart light poles with the closest distance are obtained for verification;

[0056] At the same time, calculate the pollutant emission Q per unit time according to the following formula:

[0057] Q = 2πC ij *V*σ y *σ z / {e^(-y 2 / 2σ y 2)*{e^[-(z - H) 2 / 2σ y 2 +e^[-(z + H) 2 / 2σ z 2}}

[0058] where V is the average wind speed at the discharge outlet, σ y is the lateral diffusion coefficient, σ z is the vertical diffusion coefficient, y is the horizontal distance from the pollutant diffusion center to the nearest lamp post in the downwind direction, z is the height of the lamp post from the ground required by y, and H is the height of the pollutant diffusion center.

[0059] Step 4: Update the training weights of its own convolutional neural network model according to the neighbor sensor data values, train each convolutional neural network model with the current own sensor data, and update the parameters in the Gaussian correction model based on deep learning; the terminal of each smart lamp post collects the atmospheric data with timestamps and spatial stamps of its own, receives the detection data of the adjacent smart lamp post terminals, calculates and adjusts the training weights according to the training weight values provided by the central server and the detection data of the adjacent smart lamp post terminals, retrains and updates the parameters in the model, and returns the processed data model to the central server. When sending data to the central server, due to the large amount of data, to avoid data loss during sending and receiving, it should be considered to first compress the data into the form of data packets and then send it to the central server multiple times.

[0060] Furthermore, the process of predicting data judgment can also be included in this step, that is, receiving the detection data transmitted from other lamp posts, comparing it with the trained data, and when the error value between the two is within the preset threshold range, sending the predicted value to the central server, otherwise, not sending it, and continuing to adjust the weights until the parameters in the trained model satisfy that its predicted value conforms to the detection data of the surrounding lamp posts.

[0061] Step 5: The central server receives the pollution source prediction information of multiple sub - servers, constructs a federated learning neural network model, globally searches for pollution sources within the coverage area of smart lamp posts, determines the final pollution source location, and feeds back the results to the convolutional neural network models of each smart lamp post.

[0062] Such as Figure 2As shown in the figure, within a certain area, there are multiple ordinary lamp posts. The terminals installed on lamp posts numbered 001 - 004 are used as smart lamp posts. When the terminal of No. 001 conducts data processing, it can obtain its own sensor data and the sensor data of terminals No. 002 - 004 in real time. First, it receives the data from its own sensors and inputs them into a convolutional neural network model, and then applies a Gaussian correction model based on deep learning to predict terminals No. 002 - 004, obtaining the predicted information of the atmospheric concentration and pollution sources of terminals No. 002 - 004, and judging the error between the predicted value and the actual sensor values of terminals No. 002 - 004. If the error range is within the threshold, the predicted value is sent to the central server. If the error is large, it calculates the training weights of its own convolutional neural network model according to the sensor values of terminals No. 002 - 004, and trains according to the new weights to update the parameters in the Gaussian correction model based on deep learning. The above process is repeated until the error range is reached. The working processes of other terminals are the same.

[0063] The data listed in Table 1 below are the predicted values and the errors between the predicted values and the true values of four smart lamp posts with respect to multiple pollution sources at different distances according to the above - mentioned method. All distance units are in meters. Among them,

[0064] The first column is the lamp post numbers corresponding to 001 - 004 respectively; the second column is the distances between multiple preset pollution sources and the four smart lamp posts; the third column is the position of the pollution source relative to the lamp post inferred by the Gaussian correction model based on deep learning updated through the convolutional neural network; the fourth column is the relative error between the predicted value and the preset value.

[0065] Only a few groups of data are listed in Table 1 for the purpose of illustration and verification of the results obtained by this method. It can be seen from the error values in Table 1 that the largest value also appears in the second digit after the decimal point. Therefore, it can be seen that the prediction accuracy obtained by applying this method is very high. With the continuous increase of the data volume and running time, the prediction accuracy of this system will be higher and higher.

[0066] Table 1

[0067] Lamp post number True relative distance Predicted value Error 1 20 19.48095 0.026644 2 50 48.72426 0.026183 3 40 39.38646 0.015577 4 70 69.31296 0.009912 1 40 39.27716 0.018404 2 20 19.67016 0.016768 3 10 9.534927 0.048776 4 50 49.97158 0.000569 1 60 59.87853 0.002029 2 20 19.12972 0.045494 3 10 9.325307 0.072351 4 30 29.94676 0.001778 1 60 59.40178 0.010071 2 30 28.68822 0.045725 3 10 9.662353 0.034945 4 20 19.15549 0.044087 1 50 49.30998 0.013994 2 40 39.7244 0.006938 3 80 78.83078 0.014832 4 20 19.59321 0.020762 1 40 39.4894 0.01293 2 30 28.70473 0.045124 3 40 39.07558 0.023657 4 70 68.66055 0.019508

[0068] In order to improve the accuracy of the surrounding concentration estimation and pollution source prediction and tracking, the number of smart lamp posts can be increased, and the terminals of this solution can be installed on the corresponding ordinary lamp posts to achieve this.

[0069] The central server controls the switches of each sub - server, evaluates the performance of each sub - server, provides training weights for each sub - server, and processes the data packets provided by each sub - server. Since there is a threshold for the detection distance, the central server needs to read the data to determine whether the sub - server exceeds the threshold and select the sub - server for subsequent processing.

[0070] After the training of each intelligent street lamp terminal is completed, the trained model and training data are uploaded to the central server. Due to the problems of building occlusion and diffusion threshold, the central server finally reconfirms the final pollutant diffusion center according to the prediction accuracy of each intelligent street lamp and the final prediction result, and uploads it to the visualization interface.

[0071] In order to further improve the prediction accuracy, after determining the information such as the pollution source and type concentration, the traceability prediction method of the pollution source can further use drones or unmanned vehicles equipped with pollution detection equipment to conduct more refined detection and investigation of the pollution source to determine the accuracy of its prediction.

[0072] It should be understood that the data used in this way can be interchanged under appropriate circumstances for the embodiments of the present application described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0073] In the present application, the orientation or positional relationship indicated by the terms "upper", "lower", "left", "right", "front", "rear", "top", "bottom", "inner", "outer", "middle", "vertical", "horizontal", "lateral", "longitudinal", etc. is based on the orientation or positional relationship shown in the drawings. These terms are mainly used to better describe the present application and its embodiments, and are not used to limit that the indicated devices, elements or components must have a specific orientation or be constructed and operated in a specific orientation.

[0074] Moreover, in addition to being able to represent the orientation or positional relationship, some of the above terms may also be used to represent other meanings. For example, the term "upper" may also be used to represent a certain attachment relationship or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in the present application can be understood according to specific circumstances.

[0075] The preferred embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and the devices and structures not described in detail should be understood to be implemented in a common manner in the art; any person skilled in the art can make many possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above, or modify it into an equivalent embodiment with equivalent changes, which does not affect the essence of the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for tracing and investigating atmospheric pollutants based on intelligent street lamps, characterized in that: It includes the following steps: Step 1: Use the sensing module installed on the smart light pole to obtain various sensor data around each smart light pole, assign time stamps and location stamps, and construct a distributed sensing network based on the smart light pole; Step 2: Each smart light pole constructs an original convolutional neural network model and performs deep learning on the convolutional neural network model of its own light pole in combination with its own sensor data; Step 3: Use the Gaussian correction model based on deep learning to predict the sensor data of neighboring smart light poles, and obtain and output the predicted information of the surrounding atmospheric concentration and pollution sources; the specific representation of the Gaussian correction model based on deep learning is as follows: R i = (b, V i , C ij ), where R i is the relative distance between the intelligent street lamp D i and the pollutant diffusion center, b = (V i , H i , f) are the parameters in the Gaussian correction model based on deep learning, V i is the wind speed detected by the intelligent street lamp D i , H i is the humidity detected by the intelligent street lamp D i , f is whether it is raining, C ij is the concentration of the j-th pollutant detected by the intelligent street lamp D i , and both i and j are positive integers; Calculate the pollutant emission Q per unit time according to the following formula: Q = 2πC ij *V*σ y *σ z / {e ^ (-y 2 / 2σ y 2 )*{e^[-(z - H) 2 / 2σ y 2 +e^[-(z + H) 2 / 2σ z 2}} where V is the average wind speed at the discharge outlet, σ y is the lateral diffusion coefficient, σ z is the vertical diffusion coefficient, y is the horizontal distance from the pollutant diffusion center to the nearest lamp post in the downwind direction, z is the height of the lamp post required for y from the ground, and H is the height of the pollutant diffusion center; Step 4: Update the training weights of its own convolutional neural network model according to the neighbor sensor data values, use the current own sensor data to train the convolutional neural network model of each, and update the internal parameters of the Gaussian correction model based on deep learning; Step 5: Repeat Steps 3 to 4; according to multiple pollution source prediction information, construct a federated learning neural network model, globally search for pollution sources within the coverage of the smart light poles, determine the final pollution source location, and feedback the results to the convolutional neural network models of each smart light pole.

2. The method for tracing and investigating atmospheric pollutants based on intelligent street lamps according to claim 1, wherein: The sensor data includes but is not limited to location, time, wind speed, wind direction, rainfall, air humidity, and concentrations of various pollutants.

3. The method for tracing and investigating atmospheric pollutants based on intelligent street lamps according to claim 1, wherein: The initial value of the internal parameter b of the Gaussian correction model based on deep learning is collected from historical big data, and it is a dynamically updated value during the training process, calculated according to the neighbor sensor data, and after each round of training is completed, the current sensor data of at least 3 smart light poles closest in distance are obtained for verification.

4. An atmospheric pollutant tracing and investigation terminal based on intelligent street lamps, characterized in that: It includes a data processing module and a communication module. The data processing module executes the atmospheric pollutant tracing and investigation method described in any one of claims 1 to 3; the communication module is used to realize data interaction.

5. An atmospheric pollutant traceability and investigation system based on intelligent street lights, characterized in that: It includes a visualization platform and several atmospheric pollutant tracing and investigation terminals described in claim 4, where the visualization platform is used as the central server and the atmospheric pollutant tracing and investigation terminal is used as a sub-server; data interaction is carried out between the central server and all sub-servers, and each sub-server has the function of independently analyzing atmospheric pollution data and self-organizing a network with neighboring sub-servers.

6. The atmospheric pollutant tracing and investigation system based on intelligent street lamps according to claim 5, characterized in that: The central server controls the switches of each sub-server and provides an initial global model; during operation, it receives the data sent by each sub-server, processes it, and sends the training weight values to each sub-server; The sub-server collects its own atmospheric data with time stamps and space stamps, receives the detection data of neighboring sub-servers, calculates and adjusts the training weights according to the training weight values provided by the central server and the detection data of neighboring sub-servers, retrains and updates the internal parameters of the model, and returns the processed data model to the central server.

7. The atmospheric pollutant tracing and investigation system based on intelligent street lamps according to claim 5, characterized in that: The central server receives the pollution source prediction information of multiple sub-servers, constructs a federated learning neural network model, globally searches for pollution sources within the coverage of the smart light poles, determines the final pollution source location, and feedbacks the results to the convolutional neural network models of each smart light pole.

8. The atmospheric pollutant traceability and investigation system based on intelligent street lamps according to claim 5, characterized in that: When the sub-server sends data to the central server, it first compresses the data into the form of data packets and then sends them to the central server multiple times.

Images