Dynamic optimization method and system for urban air pollution monitoring

Abstract

The invention discloses a dynamic optimization method and system for urban air pollution monitoring, relates to the technical field of environment monitoring and intelligent control, and discloses a dynamic optimization method and system for urban air pollution monitoring. A deep reinforcement learning model is utilized to generate a scheduling strategy in real time and optimize position distribution of monitoring units, and meanwhile, multi-source data is fused to generate a pollution map, so that the problems of limited coverage range, response lag and untimely model updating of a traditional monitoring method are effectively solved; monitoring resource distribution can be dynamically adjusted, a pollution diffusion model is updated in real time, intelligent fusion of monitoring data is achieved, and therefore the coverage range, the response speed and the precision of pollution monitoring are improved.

Description

Technical Field

[0001] This application relates to the field of environmental monitoring and intelligent control technology, and in particular to a dynamic optimization method and system for urban air pollution monitoring. Background Technology

[0002] Urban air pollution is becoming increasingly serious, significantly impacting people's health and quality of life. Traditional air pollution monitoring methods primarily rely on fixed monitoring stations. While these stations can provide pollution data within a certain range, they suffer from limitations such as limited monitoring coverage and untimely data updates. Particularly in the face of sudden air pollution events, fixed monitoring stations often fail to quickly capture the dynamic changes in pollution diffusion, leading to delayed emergency responses. Furthermore, with the acceleration of urbanization, the spatial distribution and temporal variations of urban air pollution are becoming increasingly complex, making traditional monitoring methods inadequate for the needs of refined management. Existing technologies lack solutions capable of dynamically adjusting the distribution of monitoring resources, updating pollution diffusion models in real time, and achieving intelligent data fusion, resulting in significant deficiencies in pollution source tracing, emergency response speed, and monitoring accuracy. Simultaneously, existing systems generally suffer from low data processing efficiency and delayed model updates, failing to adapt to the rapidly changing characteristics of urban air pollution.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main objective of this application is to provide a dynamic optimization method and system for urban air pollution monitoring, aiming to improve the coverage, response speed and accuracy of pollution monitoring.

[0005] To achieve the above objectives, this application proposes a dynamic optimization method for urban air pollution monitoring, the method comprising: Raw environmental data is collected through a sensor network consisting of fixed sensors and mobile monitoring units; The raw environmental data is preprocessed to generate feature parameter data; The feature parameter data is input into a deep reinforcement learning model to generate pollution source diffusion model data and scheduling strategy data. The scheduling strategy data is parsed to generate position control commands for the motion monitoring unit; The location distribution of the mobile monitoring units is adjusted according to the location control command, and optimized monitoring data is collected through the adjusted mobile monitoring units. By integrating the optimized monitoring data, fixed sensor data, and pollution source diffusion model data, the final pollution map data is generated.

[0006] In one embodiment, the step of inputting the feature parameter data into a deep reinforcement learning model to generate pollution source diffusion model data and scheduling strategy data includes: Construct a dynamic environment state matrix based on feature parameter data; The dynamic environment state matrix is ​​processed by a policy network to output initial scheduling action data. The initial scheduling action data is optimized based on the reward function to generate scheduling strategy data and pollution source diffusion model data.

[0007] In one embodiment, the step of processing the dynamic environment state matrix through a policy network and outputting initial scheduling action data includes: The dynamic environmental state matrix is ​​split into a meteorological submatrix and a pollution submatrix; The meteorological submatrix is ​​subjected to a first convolution process, and the pollution submatrix is ​​subjected to a second convolution process to generate a fused feature vector; The fused feature vector is mapped through a fully connected layer to output initial scheduling action data.

[0008] In one embodiment, the step of parsing the scheduling strategy data to generate location control instructions for the motion monitoring unit includes: Parse the target location coordinate set and movement priority identifier from the scheduling strategy data; The current location data of the mobile monitoring unit is obtained through a positioning device; A position control command is generated based on the target location coordinate set, the movement priority identifier, and the current location data.

[0009] In one embodiment, the step of generating a position control command based on the target location coordinate set, the movement priority identifier, and the current location data includes: When the movement priority is marked as an emergency, the shortest path algorithm is used to process the target location coordinate set and the current location data to generate the first path coordinate data. When the movement priority is set to normal, an energy consumption optimization algorithm is used to process the target location coordinate set and the current location data to generate the second path coordinate data. Position control commands are generated based on the first path coordinate data or the second path coordinate data.

[0010] In one embodiment, the method further includes: Monitor the rate of change of pollution concentration parameters and generate an event trigger signal when the rate exceeds a preset threshold. The reward function weights of the deep reinforcement learning model are reset in response to the event trigger signal.

[0011] In one embodiment, the step of resetting the reward function weights of the deep reinforcement learning model in response to the event trigger signal includes: Identify areas exceeding pollution gradient limits from raw environmental data and generate coordinates of hotspot areas; Calculate the corresponding response time reduction factor based on the coordinates of the hotspot area; The reward function weights are adjusted based on the response time reduction factor.

[0012] In one embodiment, the step of integrating the optimized monitoring data, fixed sensor data, and pollution source diffusion model data to generate the final pollution map data includes: Noise filtering is applied to the optimized monitoring data to generate calibration monitoring data; The calibration monitoring data is spatiotemporally aligned with the fixed sensor data to generate spatiotemporally aligned monitoring data. The final pollution map data is generated by fusing the spatiotemporally aligned monitoring data with the pollution source diffusion model data using a Gaussian interpolation algorithm.

[0013] In one embodiment, the method further includes: The final pollution map data is compared with the pollution source diffusion model data to generate model error data. The steps are: adjusting the parameters of the deep reinforcement learning model based on the model error data, and then returning to the process of inputting the feature parameter data into the deep reinforcement learning model to generate pollution source diffusion model data and scheduling strategy data.

[0014] Furthermore, to achieve the above objectives, this application also proposes a dynamic optimization system for urban air pollution monitoring, the dynamic optimization system for urban air pollution monitoring comprising: a memory, a processor, and a dynamic optimization program for urban air pollution monitoring stored in the memory and executable on the processor, the dynamic optimization program for urban air pollution monitoring being configured to implement the steps of the dynamic optimization method for urban air pollution monitoring.

[0015] The dynamic optimization method and system for urban air pollution monitoring proposed in this application constructs a dynamic network by combining fixed sensors and mobile monitoring units. It utilizes a deep reinforcement learning model to generate scheduling strategies in real time and optimize the location distribution of monitoring units. At the same time, it integrates multi-source data to generate a pollution map. This effectively solves the problems of limited coverage, slow response, and untimely model updates in traditional monitoring methods. It can dynamically adjust the distribution of monitoring resources, update the pollution diffusion model in real time, and achieve intelligent fusion of monitoring data, thereby improving the coverage, response speed, and accuracy of pollution monitoring. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an embodiment of the dynamic optimization method for urban air pollution monitoring in this application; Figure 2 This is a schematic diagram of a structure provided for an embodiment of the dynamic optimization system for urban air pollution monitoring in this application.

[0019] Explanation of icon numbers: 10. Memory; 20. Processor.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of this application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0022] It should be understood that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0023] Urban air pollution is becoming increasingly serious, significantly impacting people's health and quality of life. Traditional air pollution monitoring methods primarily rely on fixed monitoring stations. While these stations can provide pollution data within a certain range, they suffer from limitations such as limited monitoring coverage and untimely data updates. Particularly in the face of sudden air pollution events, fixed monitoring stations often fail to quickly capture the dynamic changes in pollution diffusion, leading to delayed emergency responses. Furthermore, with the acceleration of urbanization, the spatial distribution and temporal variations of urban air pollution are becoming increasingly complex, making traditional monitoring methods inadequate for the needs of refined management. Existing technologies lack solutions capable of dynamically adjusting the distribution of monitoring resources, updating pollution diffusion models in real time, and achieving intelligent data fusion, resulting in significant deficiencies in pollution source tracing, emergency response speed, and monitoring accuracy. Simultaneously, existing systems generally suffer from low data processing efficiency and delayed model updates, failing to adapt to the rapidly changing characteristics of urban air pollution.

[0024] Based on this, embodiments of this application provide a dynamic optimization method for urban air pollution monitoring, referring to... Figure 1 The dynamic optimization method for urban air pollution monitoring includes steps S100 to S600, wherein: Step S100: Collect raw environmental data through a sensor network consisting of fixed sensors and mobile monitoring units; Step S200: Preprocess the raw environmental data to generate feature parameter data; Step S300: Input the feature parameter data into the deep reinforcement learning model to generate pollution source diffusion model data and scheduling strategy data; Step S400: Parse the scheduling strategy data to generate position control instructions for the motion monitoring unit; Step S500: Adjust the position distribution of the mobile monitoring units according to the position control command, and collect optimized monitoring data through the adjusted mobile monitoring units; Step S600: The optimized monitoring data, fixed sensor data, and pollution source diffusion model data are integrated to generate the final pollution map data.

[0025] In this embodiment, fixed sensors can be understood as devices installed at specific locations, primarily used for continuously collecting pollution-related data in the environment, such as sulfur dioxide concentration and particulate matter concentration. Furthermore, fixed sensors can transmit the collected data to a data processing center via wired or wireless means. Mobile monitoring units refer to flexibly deployable monitoring devices, which can be implemented through vehicles, drones, or other mobile platforms. Specifically, the deployment of mobile monitoring units can be dynamically adjusted according to needs, such as rapidly covering high-risk areas during sudden pollution events. Preprocessing of raw environmental data can be achieved in various ways. For example, filtering the raw data to remove noise interference, or normalizing data from different sources to have a unified dimension. Its main purpose is to provide structured input data for subsequent analysis. Feature parameter data refers to key information extracted from raw environmental data, which can be generated through statistical analysis, frequency domain transformation, etc. For example, extracting the changing trend of pollution concentration based on time series analysis, or generating identification information of pollution hotspot areas through spatial distribution analysis.

[0026] In this embodiment, the deep reinforcement learning model is an algorithmic framework combining deep learning and reinforcement learning, which can be implemented using various architectures. For example, it can be based on convolutional neural networks to process spatially distributed data, or on recurrent neural networks to capture the dynamic characteristics of time-series data. Specifically, the model learns the mapping relationship between environmental states and actions through a training process, thereby generating pollution source diffusion model data and scheduling strategy data. The pollution source diffusion model data is used to describe the spatial and temporal propagation patterns of pollutants, while the scheduling strategy data is used to guide the dynamic deployment of mobile monitoring units. The parsing of scheduling strategy data can be achieved in various ways. For example, rule-based methods can be used to convert scheduling strategy data into specific execution instructions, or optimization algorithms can be used to generate path planning schemes that meet constraints. Location control instructions refer to specific operational commands used to guide the mobile monitoring unit to adjust its position, which can be directly sent to the control system of the mobile monitoring unit through a communication protocol. Furthermore, optimizing the acquisition of monitoring data can also be achieved in various ways. For example, the mobile monitoring unit can adjust the sampling frequency according to the pollution concentration gradient at the current location, or generate a more comprehensive dataset through multi-point sampling. Its main purpose is to improve the spatial resolution and representativeness of the monitoring data. The final pollution map data can be generated through various fusion algorithms, such as integrating multi-source data based on weighted averaging, or generating a continuous pollution distribution view through interpolation algorithms. In this embodiment, this application integrates the monitoring capabilities of fixed sensors and mobile monitoring units, and combines them with a deep reinforcement learning model to achieve dynamic optimization of the monitoring process. Compared with traditional fixed-site monitoring methods, this application can effectively expand the monitoring range and improve the timeliness of data updates. Especially in the face of sudden pollution events, the dynamic deployment of mobile monitoring units can quickly respond to changes in pollution diffusion, thereby providing more comprehensive data support for pollution monitoring and management.

[0027] This application embodiment collects raw environmental data through a sensor network composed of fixed sensors and mobile monitoring units. Fixed sensors continuously monitor pollution levels in specific areas, while mobile monitoring units leverage their flexibility to extend monitoring coverage, enabling rapid deployment to critical areas, especially during sudden pollution events, effectively addressing the limited monitoring range of traditional fixed stations. Furthermore, the collected raw environmental data is preprocessed to generate feature parameter data. This process, through denoising and key information extraction, ensures the quality and reliability of subsequent model input data, avoiding interference from redundant data in the analysis results. The generated feature parameter data is input into a deep reinforcement learning model. Based on the structured characteristics of the feature parameter data, the model can more accurately capture the dynamic changes in pollution, thereby generating pollution source diffusion model data and scheduling strategy data, providing highly adaptive diffusion prediction and scheduling decision support for pollution monitoring. Specifically, the scheduling strategy data is parsed to generate location control commands for the mobile monitoring units, transforming abstract scheduling strategies into executable instructions, ensuring that the actions of the mobile monitoring units closely match actual monitoring needs, and enhancing the practical feasibility of the strategy. The location distribution of mobile monitoring units is adjusted according to location control commands, and optimized monitoring data is collected through these adjusted units. This provides real-time guidance for the mobile units to focus on pollution hotspots, significantly improving the targeting and efficiency of data collection. Furthermore, by integrating optimized monitoring data, fixed sensor data, and pollution source diffusion model data, a final pollution map is generated. Through the collaborative integration of multi-source data, combining the advantages of real-time monitoring and model prediction, the shortcomings of single data sources are overcome, resulting in a pollution distribution view with higher spatial resolution and more timely updates. Ultimately, this improves the comprehensiveness and accuracy of monitoring results, solving the technical problems of untimely data updates and inability to quickly respond to sudden pollution events in traditional methods.

[0028] In one feasible implementation, the step of inputting the feature parameter data into a deep reinforcement learning model to generate pollution source diffusion model data and scheduling strategy data includes: constructing a dynamic environmental state matrix based on the feature parameter data; processing the dynamic environmental state matrix through a policy network to output initial scheduling action data; and optimizing the initial scheduling action data based on a reward function to generate scheduling strategy data and pollution source diffusion model data.

[0029] In this embodiment, the dynamic environmental state matrix refers to a multi-dimensional data structure that can reflect environmental changes in real time. It can be implemented as a matrix containing multi-dimensional parameters such as meteorological information and pollution concentration distribution. The policy network can be understood as an intelligent decision-making module based on neural networks, whose purpose is to generate a preliminary scheduling scheme adapted to the current environmental conditions based on the input dynamic environmental state matrix. The reward function is a mechanism for evaluating and optimizing the performance of scheduling actions, and its optimization objective can be achieved by defining key indicators such as monitoring efficiency and response time.

[0030] In this embodiment, the above-mentioned technical solution effectively responds to dynamic environmental changes through the organic combination of multiple steps. First, a dynamic environmental state matrix is ​​constructed based on feature parameter data. This process ensures the real-time nature and comprehensiveness of environmental parameters, providing a reliable data foundation for subsequent processing. Second, the dynamic environmental state matrix is ​​processed by a policy network, which outputs initial scheduling action data. The policy network can analyze the complex correlations within the matrix, thereby generating a preliminary scheduling scheme adapted to real-time changes. Finally, the initial scheduling action data is optimized based on a reward function. This process introduces a feedback mechanism, continuously optimizing the scheduling actions to ensure that the generated scheduling strategy data and pollution source diffusion model data are more efficient and reliable. The entire process, through gradual refinement and optimization, solves the problem of suboptimal strategy generation under dynamic environmental changes, significantly improving the timeliness and accuracy of monitoring response.

[0031] In this embodiment, the above method is closely integrated with the steps of sensor network acquiring raw environmental data and subsequent parsing of scheduling strategy data to generate position control commands. Through the construction of a dynamic environmental state matrix and processing by the strategy network, the accuracy and adaptability of the scheduling strategy data are ensured, thus providing a scientific basis for the position adjustment of the mobile monitoring unit. This overall design effectively solves the response lag problem caused by environmental fluctuations in traditional monitoring methods, demonstrating the innovation and practicality of the technical solution.

[0032] In one feasible implementation, the step of processing the dynamic environment state matrix through a policy network and outputting initial scheduling action data includes: splitting the dynamic environment state matrix into a meteorological sub-matrix and a pollution sub-matrix; performing a first convolution process on the meteorological sub-matrix and a second convolution process on the pollution sub-matrix to generate a fused feature vector; and mapping the fused feature vector through a fully connected layer to output the initial scheduling action data.

[0033] In this embodiment, the dynamic environmental state matrix refers to a comprehensive information matrix containing meteorological and pollution data, which can be implemented using multidimensional arrays or tensors. In practical applications, the meteorological submatrix refers to the meteorological-related data extracted from the dynamic environmental state matrix. It can be implemented using specific data segmentation algorithms or indexing operations, aiming to process meteorological data independently to avoid information interference. The pollution submatrix refers to the pollution-related data extracted from the dynamic environmental state matrix, which can also be implemented using data segmentation algorithms or indexing operations, aiming to specifically extract pollution data features. Specifically, the first convolutional processing refers to the spatial feature extraction operation performed on the meteorological submatrix, which can be implemented using standard convolutional neural networks or depthwise separable convolution, aiming to capture spatial features in the meteorological data such as wind direction and temperature changes. The second convolutional processing refers to the distribution feature extraction operation performed on the pollution submatrix, which can be implemented using standard convolutional neural networks or dilated convolution, aiming to process the distribution features of pollution data such as concentration gradients. The fused feature vector refers to the feature representation integrating meteorological and pollution processing results, which can be generated through concatenation operations or weighted summation, aiming to provide a comprehensive feature representation.

[0034] In this embodiment, the problem of insufficient feature extraction caused by data mixing in the dynamic environmental state matrix can be effectively solved by decomposing the dynamic environmental state matrix. First, the dynamic environmental state matrix is ​​split into a meteorological sub-matrix and a pollution sub-matrix, allowing independent processing of meteorological and pollution data and ensuring that their respective features are extracted in a targeted manner. Next, convolution processing is performed on the meteorological sub-matrix and the pollution sub-matrix respectively. The former captures the spatial features of meteorological data, and the latter extracts the distribution features of pollution data, thereby achieving efficient processing of the two types of data. Subsequently, by generating a fused feature vector, the processing results of the two types of data are integrated, providing comprehensive feature support for the subsequent mapping of the action space. Finally, the fused feature vector is mapped through a fully connected layer, and the integrated information is transformed into optimized initial scheduling action data, improving the accuracy and efficiency of scheduling actions. Based on this, the above method not only solves the problem of insufficient feature extraction, but also significantly improves the quality of initial scheduling action data through independent processing and integration, providing a reliable foundation for the generation of subsequent pollution monitoring and scheduling strategies.

[0035] In one feasible implementation, the step of parsing the scheduling strategy data to generate a position control command for the motion monitoring unit includes: parsing a target location coordinate set and a motion priority identifier from the scheduling strategy data; obtaining the current location data of the motion monitoring unit through a positioning device; and generating a position control command based on the target location coordinate set, the motion priority identifier, and the current location data.

[0036] In this embodiment, the scheduling strategy data refers to the dataset generated by the deep reinforcement learning model, which contains key information to guide the dynamic adjustment of the mobile monitoring unit. In practical applications, the target location coordinate set can be understood as a set of preset geographic coordinate points, which typically correspond to pollution hotspots or optimized monitoring target areas. The purpose is to ensure that the mobile monitoring unit can efficiently cover key monitoring areas. The movement priority identifier is a classification label used to distinguish between task requirements in emergency and normal states. It can be implemented numerically or in encoded form, for example, using binary encoding "1" to represent an emergency state and "0" to represent a normal state, thus providing a basis for subsequent path planning decisions. The positioning device is a device capable of acquiring the geographic location information of the mobile monitoring unit in real time. It can be implemented using a Global Positioning System (GPS) module, a BeiDou navigation module, or other high-precision positioning technologies. The introduction of this device aims to ensure the real-time nature and accuracy of location data, thereby avoiding response delays or path planning errors caused by location deviations. In this embodiment, the generation process of the location control command involves the comprehensive processing of the target location coordinate set, the movement priority identifier, and the current location data. Specifically, when the movement priority indicator indicates an emergency, the shortest path algorithm (such as Dijkstra's algorithm or A* algorithm) can be used to calculate the optimal path to quickly respond to pollution events; while under normal conditions, energy-saving paths can be generated through energy consumption optimization algorithms (such as genetic algorithms or dynamic programming algorithms) to balance monitoring efficiency and energy consumption.

[0037] In this embodiment, the above-described scheme achieves efficient parsing of scheduling strategy data through a clearly defined step design, and generates optimized position control commands by combining target location, priority, and current location. First, the target location coordinate set and movement priority identifier are extracted from the scheduling strategy data; this step provides a fundamental guarantee for the extraction of core information. Second, real-time location data is acquired using a positioning device, ensuring the accuracy and adaptability of command generation. Finally, position control commands are generated based on the target location, priority, and current location. This process not only considers the actual needs of pollution monitoring but also optimizes energy consumption and response speed. This step-by-step design effectively solves the problems of inefficiently parsing strategy data and the inability to combine priority and current location, thereby improving the dynamic adaptability and overall efficiency of the mobile monitoring unit. Furthermore, the above scheme, combined with steps such as sensor network acquisition of raw environmental data and generation of feature parameter data, forms a complete dynamic optimization method. By combining scheduling strategy data with real-time location information, not only is the real-time performance of pollution monitoring improved, but the system's energy consumption is also significantly reduced, demonstrating the overall advantages of the technical solution.

[0038] In one feasible implementation, the step of generating a position control command based on the target location coordinate set, the movement priority identifier, and the current location data includes: when the movement priority identifier is in an emergency state, processing the target location coordinate set and the current location data using a shortest path algorithm to generate first path coordinate data; when the movement priority identifier is in a normal state, processing the target location coordinate set and the current location data using an energy consumption optimization algorithm to generate second path coordinate data; and generating a position control command based on the first path coordinate data or the second path coordinate data.

[0039] In this embodiment, the mobile priority identifier refers to the marking information used to distinguish the urgency of monitoring tasks. It can be a status label automatically generated according to preset rules or a manually set task level. In practical applications, this identifier can be dynamically adjusted according to the severity of the pollution event, thereby ensuring the rational allocation of monitoring resources. Its purpose is to select the optimal path planning strategy according to different monitoring needs and improve overall monitoring efficiency. The shortest path algorithm refers to a path planning method with the goal of minimizing time or distance, which can be implemented using Dijkstra's algorithm, A* search algorithm, etc. Its purpose is to ensure that in emergency situations, the mobile monitoring unit can quickly reach the target location and reduce response delay. The energy consumption optimization algorithm refers to a path planning method with the goal of reducing energy consumption, which can be implemented through dynamic programming algorithms or heuristic energy-saving algorithms. Its purpose is to extend the equipment's endurance and maintain long-term monitoring stability.

[0040] In this embodiment, the above technical solution introduces a conditional branching mechanism based on movement priority indicators to optimize the path planning algorithm for different monitoring states, effectively solving the problems of response lag and energy waste in the process of generating location control commands. When the movement priority indicator indicates an emergency state, the system automatically calls the shortest path algorithm to ensure that the mobile monitoring unit can reach the target location in the shortest time and capture the dynamic changes in pollution diffusion in a timely manner. Under normal monitoring conditions, the system switches to an energy consumption optimization algorithm to reduce energy consumption by optimizing path selection and avoid unnecessary equipment damage. On this basis, location control commands are generated based on the first path coordinate data or the second path coordinate data, directly converting the algorithm output into executable commands to ensure the accuracy and real-time performance of location adjustment. In addition, the above technical solution, combined with the aforementioned collection of raw environmental data by a sensor network composed of fixed sensors and mobile monitoring units, and the parsing of scheduling strategy data to generate location control commands for mobile monitoring units, forms a complete dynamic optimization monitoring system. This not only improves the response speed and resource utilization efficiency of the mobile monitoring unit, but also further enhances the overall performance of urban air pollution monitoring and meets the needs of complex dynamic monitoring.

[0041] In one feasible implementation, the method further includes: monitoring the rate of change of pollution concentration parameters, generating an event trigger signal when the rate exceeds a preset threshold, and resetting the reward function weights of the deep reinforcement learning model in response to the event trigger signal.

[0042] In this embodiment, the pollution concentration parameter change rate refers to the magnitude of change in pollution concentration per unit time, which can be calculated by differential calculation of continuously collected pollution concentration data. In practical applications, a sliding window algorithm or an exponentially weighted moving average algorithm can be used to smooth noise interference, thereby improving the accuracy of change rate detection. The purpose is to capture abnormal fluctuations in pollution dynamics. The preset threshold can be understood as the critical point that distinguishes normal fluctuations from abnormal events. Its setting method can be determined based on historical data analysis or expert experience, aiming to define the standard for emergency states. The event trigger signal is a distinctive control signal used to indicate that the system enters emergency response mode. Its generation logic can be implemented by combining multi-level threshold judgments or time window constraints to avoid false triggering.

[0043] In this embodiment, a rapid response to sudden pollution events is achieved by introducing an event-triggered mechanism. First, the rate of change of pollution concentration parameters is monitored as an input signal, reflecting the dynamic characteristics of pollution diffusion in real time. When the rate of change exceeds a preset threshold, the system generates an event-triggered signal, ensuring timely identification of abnormal events. Subsequently, in response to the event-triggered signal, the system resets the weights of the reward function, enabling the deep reinforcement learning model to quickly adjust its learning strategy. This dynamic adjustment mechanism effectively improves the model's adaptability to scenarios with rapidly changing pollution concentrations, solving the response lag problem caused by fixed reward function weights in traditional methods. Furthermore, this scheme, combined with other steps in the aforementioned dynamic optimization method for urban air pollution monitoring, such as generating location control commands through scheduling strategy data, further improves the overall response efficiency of the monitoring system, thereby better meeting the needs of refined management. Thus, not only is the real-time performance and accuracy of pollution monitoring improved, but the system's ability to respond to sudden pollution events is also significantly enhanced, providing reliable technical support for the dynamic optimization of urban air pollution.

[0044] In one feasible implementation, the step of resetting the reward function weights of the deep reinforcement learning model in response to the event trigger signal includes: identifying pollution gradient excess areas from raw environmental data and generating hotspot area coordinates; calculating a corresponding response time reduction factor based on the hotspot area coordinates; and adjusting the reward function weights according to the response time reduction factor.

[0045] In this embodiment, raw environmental data refers to unprocessed air pollution-related parameters collected through a sensor network, which may include information such as particulate matter concentration and gas composition. The pollution gradient exceeding limit area can be understood as an area within a certain spatial range where the rate of change in pollution concentration exceeds a preset threshold; its purpose is to quickly locate key areas of pollution spread. The response time reduction factor refers to a time adjustment coefficient quantified based on the location information of hotspot areas, used to optimize scheduling priorities; its purpose is to ensure that the model can respond quickly to emergency areas.

[0046] In this embodiment, the system extracts pollution gradient information from raw environmental data, identifies areas with significant changes in pollution concentration, and generates corresponding hotspot coordinates, ensuring accurate positioning of key pollution diffusion areas. Based on the generated hotspot coordinates, the system calculates a response time reduction factor using a specific algorithm. This factor reflects the urgency of response in different areas. According to the calculated response time reduction factor, the reward function weights of the deep reinforcement learning model are dynamically adjusted, enabling the model to prioritize severely polluted areas and improve overall response efficiency. This method, combined with the aforementioned pollution monitoring and scheduling strategy generation process, solves the response lag problem caused by a lack of targeted optimization in traditional methods. It not only improves the model's response speed to pollution hotspots but also optimizes resource allocation efficiency, thereby enhancing the overall performance of the urban air pollution monitoring system.

[0047] In one feasible implementation, the step of fusing the optimized monitoring data, fixed sensor data, and pollution source diffusion model data to generate the final pollution map data includes: filtering noise from the optimized monitoring data to generate calibrated monitoring data; spatiotemporally aligning the calibrated monitoring data with the fixed sensor data to generate spatiotemporally aligned monitoring data; and fusing the spatiotemporally aligned monitoring data with the pollution source diffusion model data using a Gaussian interpolation algorithm to generate the final pollution map data.

[0048] In this embodiment, noise filtering refers to removing random interference components from the collected data using signal processing techniques. This can be achieved using methods such as low-pass filters, Kalman filters, or wavelet transforms, aiming to eliminate noise interference that may be introduced when the mobile monitoring unit collects data in complex environments. Spatiotemporal alignment refers to uniformly matching and correcting data from different sources in both time and space dimensions. This can be achieved through timestamp synchronization technology and geographic coordinate transformation algorithms, aiming to solve the data inconsistency problem caused by differences in sampling frequency and location between fixed sensors and mobile monitoring units. Gaussian interpolation is a spatial interpolation method based on probability distribution. It uses a Gaussian kernel function to smooth discrete data points, thereby achieving a continuous representation of the data, aiming to improve the spatial resolution and accuracy of pollution map data.

[0049] In this embodiment, the above-mentioned technical solution achieves efficient fusion of multi-source data through a hierarchical data processing flow. Noise filtering of the optimized monitoring data ensures the quality of the input data, laying a reliable foundation for subsequent processing. Furthermore, spatiotemporal alignment integrates the calibrated monitoring data with fixed sensor data into a unified time and space reference system, resolving the spatiotemporal inconsistency problem caused by different data sources in traditional methods. Finally, Gaussian interpolation is used to fuse the spatiotemporally aligned data with pollution source diffusion model data, fully leveraging the advantages of Gaussian interpolation in handling spatial uncertainties to generate final pollution map data with high accuracy and reliability. This process not only effectively overcomes the problems of data noise interference, spatiotemporal inconsistency, and insufficient fusion accuracy, but also significantly enhances the practical value of pollution map data, providing support for the refined management of urban air pollution.

[0050] In one feasible implementation, the method further includes: comparing the final pollution map data with the pollution source diffusion model data to generate model error data; adjusting the deep reinforcement learning model parameters based on the model error data; and returning to the step of inputting the feature parameter data into the deep reinforcement learning model to generate pollution source diffusion model data and scheduling strategy data.

[0051] In this embodiment, the final pollution map data refers to the comprehensive pollution distribution result generated by fusing and optimizing monitoring data, fixed sensor data, and pollution source diffusion model data. This can be achieved using Gaussian interpolation algorithms or similar data fusion techniques, aiming to provide a more accurate reference benchmark for the actual pollution distribution. Model error data refers to the quantitative index generated by comparing the differences between the final pollution map data and the pollution source diffusion model data. This can be achieved by calculating the mean squared error or other error assessment methods, aiming to provide a basis for subsequent model parameter adjustments. Adjusting the deep reinforcement learning model parameters refers to the process of dynamically correcting the model's internal weights or structural parameters based on the generated model error data. This can be achieved using gradient descent or other optimization algorithms, aiming to reduce systematic biases in model predictions and improve adaptability.

[0052] In this embodiment, the above-mentioned technical solution addresses the problem of accumulated prediction errors in deep reinforcement learning models during continuous execution due to initial parameter deviations or dynamic environmental changes by introducing model error feedback and iterative optimization mechanisms. Specifically, the final pollution map data is compared with pollution source diffusion model data to generate model error data. This step quantifies the deviation between the actual fused monitoring results and the model prediction output by directly comparing them, providing a precise error basis for subsequent adjustments. Adjusting the deep reinforcement learning model parameters based on the model error data dynamically corrects the model's internal parameters, reducing systematic deviations in subsequent predictions. The process then returns to the step of inputting feature parameter data into the deep reinforcement learning model, forming a closed-loop feedback loop. This ensures the model can continuously optimize itself based on the latest error data, enhancing its adaptability to the complex dynamics of urban air pollution. Furthermore, this process, combined with the aforementioned steps of collecting data through sensor networks and generating scheduling strategy data, further improves the overall system's monitoring accuracy and response efficiency, thereby better meeting the needs of dynamic urban air pollution monitoring. It not only effectively solves the problem of model prediction error accumulation but also significantly enhances the system's adaptability and long-term operational stability, thus providing more reliable support for urban air pollution monitoring.

[0053] In the embodiments of this application, the dynamic optimization method for urban air pollution monitoring constructs a dynamic network by combining fixed sensors and mobile monitoring units, uses a deep reinforcement learning model to generate scheduling strategies in real time and optimize the location distribution of monitoring units, and integrates multi-source data to generate a pollution map. This effectively solves the problems of limited coverage, slow response, and untimely model updates in traditional monitoring methods. It can dynamically adjust the distribution of monitoring resources, update the pollution diffusion model in real time, and achieve intelligent fusion of monitoring data, thereby improving the coverage, response speed, and accuracy of pollution monitoring.

[0054] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the dynamic optimization method for urban air pollution monitoring in this application. Any simple modifications based on this technical concept are within the scope of protection of this application.

[0055] This application also provides a dynamic optimization system for urban air pollution monitoring, with reference to... Figure 2 The dynamic optimization system for urban air pollution monitoring includes: a memory 10, a processor 20, and a dynamic optimization program for urban air pollution monitoring stored on the memory 10 and executable on the processor 20. The dynamic optimization program for urban air pollution monitoring is configured to implement the steps of the dynamic optimization method for urban air pollution monitoring.

[0056] The dynamic optimization system for urban air pollution monitoring provided in this application, employing the dynamic optimization method for urban air pollution monitoring described in the above embodiments, can improve the coverage, response speed, and accuracy of pollution monitoring. Compared with the prior art, the beneficial effects of the dynamic optimization system for urban air pollution monitoring provided in this application are the same as those of the dynamic optimization method for urban air pollution monitoring provided in the above embodiments, and other technical features of the dynamic optimization system for urban air pollution monitoring are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0057] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0058] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. All equivalent structural transformations made under the technical concept of this application using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included within the scope of patent protection of this application.

Claims

1. A dynamic optimization method for urban air pollution monitoring, characterized in that, The method comprises: Collecting environment raw data through a sensor network composed of fixed sensors and mobile monitoring units; Preprocessing the environment raw data to generate feature parameter data; Inputting the feature parameter data into a deep reinforcement learning model to generate pollution source diffusion model data and scheduling strategy data; Analyzing the scheduling strategy data to generate position control instructions for the mobile monitoring units; Adjusting the position distribution of the mobile monitoring units according to the position control instructions, and collecting optimized monitoring data through the adjusted mobile monitoring units; Fusing the optimized monitoring data, fixed sensor data, and pollution source diffusion model data to generate final pollution map data.

2. The dynamic optimization method for urban air pollution monitoring of claim 1, wherein, The step of inputting the feature parameter data into a deep reinforcement learning model to generate pollution source diffusion model data and scheduling strategy data comprises: Constructing a dynamic environment state matrix based on the feature parameter data; Processing the dynamic environment state matrix through a policy network to output initial scheduling action data; Optimizing the initial scheduling action data based on a reward function to generate scheduling strategy data and pollution source diffusion model data.

3. The dynamic optimization method for urban air pollution monitoring of claim 1, wherein, The step of processing the dynamic environment state matrix through a policy network to output initial scheduling action data comprises: Splitting the dynamic environment state matrix into a weather sub-matrix and a pollution sub-matrix; Performing first convolution processing on the weather sub-matrix and second convolution processing on the pollution sub-matrix to generate a fusion feature vector; Mapping the fusion feature vector through a fully connected layer to output initial scheduling action data.

4. The dynamic optimization method for urban air pollution monitoring of claim 1, wherein, The step of analyzing the scheduling strategy data to generate position control instructions for the mobile monitoring units comprises: Analyzing target position coordinate sets and mobile priority identifiers from the scheduling strategy data; Obtaining current position data of the mobile monitoring units through a positioning device; Generating position control instructions based on the target position coordinate sets, the mobile priority identifiers, and the current position data.

5. The dynamic optimization method for urban air pollution monitoring of claim 4, wherein, The step of generating position control instructions based on the target position coordinate sets, the mobile priority identifiers, and the current position data comprises: When the mobile priority identifier is in an emergency state, processing the target position coordinate sets and the current position data using a shortest path algorithm to generate first path coordinate data; When the mobile priority identifier is in a regular state, processing the target position coordinate sets and the current position data using an energy consumption optimization algorithm to generate second path coordinate data; Generating position control instructions based on the first path coordinate data or the second path coordinate data.

6. The dynamic optimization method for urban air pollution monitoring of claim 1, wherein, The method further comprises: Monitoring the change rate of pollution concentration parameters, and generating an event trigger signal when the change rate exceeds a preset threshold; Resetting the reward function weight of the deep reinforcement learning model in response to the event trigger signal.

7. The dynamic optimization method for urban air pollution monitoring of claim 6, wherein, The step of resetting the reward function weight of the deep reinforcement learning model in response to the event trigger signal comprises: Identifying a pollution gradient overrun area from the environment raw data to generate hotspot area coordinates; Calculating a corresponding response time reduction factor based on the hotspot area coordinates; Adjusting the reward function weight according to the response time reduction factor.

8. The dynamic optimization method for urban air pollution monitoring of claim 1, wherein, The step of fusing the optimized monitoring data, fixed sensor data, and pollution source diffusion model data to generate final pollution map data comprises: The optimized monitoring data is filtered to remove noise to generate calibrated monitoring data; The calibrated monitoring data is spatio-temporally aligned with the fixed sensor data to generate spatio-temporally aligned monitoring data; The spatio-temporally aligned monitoring data and the pollution source diffusion model data are fused by a Gaussian interpolation algorithm to generate final pollution map data.

9. The dynamic optimization method for urban air pollution monitoring of claim 1, wherein, The method further comprises: The final pollution map data is compared with the pollution source diffusion model data to generate model error data; Based on the model error data, the parameters of the deep reinforcement learning model are adjusted, and the step of inputting the feature parameter data into the deep reinforcement learning model to generate the pollution source diffusion model data and the scheduling strategy data is returned.

10. A dynamic optimization system for urban air pollution monitoring, characterized in that, The dynamic optimization system for urban air pollution monitoring comprises a memory, a processor, and a dynamic optimization program for urban air pollution monitoring stored on the memory and executable on the processor, and the dynamic optimization program for urban air pollution monitoring is configured to implement the steps of the dynamic optimization method for urban air pollution monitoring according to any one of claims 1 to 9.

Images