Construction dust real-time monitoring and control system based on intelligent sensor

Abstract

The application relates to the field of environment monitoring and intelligent sensor technology, and particularly discloses a construction dust real-time monitoring and regulation system based on an intelligent sensor, which comprises an edge intelligent node arranged on a site, a regional collaborative controller and a cloud management platform. The edge node integrates sensing, calculation and execution, and realizes millisecond-level autonomous regulation based on a local prediction model; the regional controller gathers multi-node data, and through diffusion prediction and collaborative optimization algorithms, the regional controller commands joint defense and joint control among nodes; and the cloud platform is responsible for data storage, model training and global supervision. Through the above-mentioned three-layer distributed collaborative architecture, the application realizes real-time monitoring, rapid response and precise collaborative regulation of construction dust.

Description

Technical Field

[0001] This invention belongs to the field of environmental monitoring and intelligent sensor technology, specifically relating to a real-time monitoring and control system for construction dust based on intelligent sensors. Background Technology

[0002] In the fields of environmental protection and refined urban management, real-time monitoring and intelligent control of various pollution sources are key technologies for improving environmental quality and protecting public health. Construction dust, as a significant source of urban air pollution, directly impacts the effectiveness of regional environmental governance through efficient monitoring and control.

[0003] Real-time monitoring and control of construction dust is an important branch of environmental monitoring technology. This technology aims to effectively suppress dust pollution by deploying sensor networks to collect environmental parameters such as particulate matter concentration at construction sites in real time, and automatically or semi-automatically activating dust suppression equipment based on the monitoring results.

[0004] The system employs a model that uploads raw data collected by sensors to a cloud server for centralized processing and decision-making. This architecture results in significant communication and computational delays from data acquisition to the issuance of control commands, making it difficult to meet the stringent requirements of rapidly changing construction site environments and the need for immediate response. The latency issue is particularly pronounced when multiple sensor nodes are transmitting data concurrently or when network conditions are poor, potentially causing delays in control measures and hindering timely intervention in the early stages of dust concentration exceeding standards, thereby weakening the overall pollution control effectiveness. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time monitoring and control system for construction dust based on intelligent sensors, so as to solve the technical contradiction in the prior art where the communication and computing delays caused by centralized cloud processing of data are significant and it is difficult to meet the requirements of real-time response at the construction site.

[0006] To achieve the above objectives, the present invention provides the following technical solution. The core of this real-time monitoring and control system for construction dust based on intelligent sensors lies in constructing a three-layer distributed collaborative processing architecture consisting of edge intelligent nodes, regional collaborative controllers, and a cloud management platform. By decentralizing the decision-making logic to the edge side adjacent to the physical space and introducing a prediction and collaborative mechanism based on spatiotemporal correlation, a millisecond-level closed-loop response from perception to control is achieved.

[0007] The system includes multiple edge intelligent nodes deployed at the construction site, at least one regional collaborative controller, and a cloud management platform.

[0008] The edge intelligent node is an embedded integrated device that integrates sensing, computing, and execution functions. Each edge intelligent node includes a micro-meteorological sensor array, a main control computing unit, a local control actuator, and a first wireless communication module. The micro-meteorological sensor array is used to collect real-time data on aerosol particulate matter concentration, temperature, humidity, wind speed, and wind direction at its deployment location. The main control computing unit has a built-in first-level dust risk prediction model and first-level control decision logic. The first-level dust risk prediction model is a time-series prediction model based on a long short-term memory network. It takes as input the local historical monitoring data sequence of the previous 10 consecutive sampling periods, each period being 30 seconds, and outputs the predicted value of the local particulate matter concentration and its confidence level for the next 2 sampling periods. The first-level control decision logic is used to compare the predicted particulate matter concentration value with a preset first-level concentration threshold. When the predicted concentration exceeds the first-level concentration threshold and the confidence level is higher than 85%, the main control computing unit immediately generates a first-level control command and directly controls the local control actuator to start through the electrical drive circuit, executing a predefined baseline intensity dust suppression operation. The total delay of this process is controlled within 200 milliseconds.

[0009] The regional collaborative controller maintains a bidirectional data connection with all edge intelligent nodes within the coverage area via a second wireless communication module. The regional collaborative controller includes a collaborative computing unit and a strategy management unit. The collaborative computing unit runs a second-level dust diffusion prediction model and a collaborative optimization algorithm. The second-level dust diffusion prediction model is a gridded model based on simplified computational fluid dynamics principles. It uses real-time monitoring data reported by all edge intelligent nodes within the area as boundary conditions, combined with digital three-dimensional geographic information of the construction site, to simulate the spatial diffusion trend of dust within the next 5 minutes, generating a dynamic risk concentration distribution heat map. The collaborative optimization algorithm is used to calculate the optimal collaborative working parameters of the control actuators of each edge intelligent node based on the risk concentration distribution heat map. This algorithm takes maximizing the overall regional dust suppression effect and minimizing total energy consumption as its dual objectives, and solves the problem using a multi-objective particle swarm optimization algorithm, outputting the control intensity correction coefficient and start / stop timing scheme for each edge intelligent node.

[0010] The strategy management unit is used to dynamically manage decision thresholds and strategies. It receives long-term strategy parameters from the cloud management platform and, in conjunction with the regional risk profile output by the collaborative computing unit, dynamically adjusts the first-level concentration threshold and baseline control intensity issued to each edge intelligent node. Specifically, when the regional risk concentration distribution heatmap shows an overall increase in risk level, the strategy management unit will issue a reduced first-level concentration threshold and an enhanced baseline control intensity parameter to the relevant edge intelligent nodes.

[0011] The cloud management platform connects to all regional collaborative controllers via a wide area network. The cloud management platform includes a data warehouse, a model training engine, and a panoramic command center. The data warehouse persistently stores raw monitoring data uploaded by all edge intelligent nodes, event logs, and collaborative decision records reported by regional collaborative controllers. Based on historical big data accumulated in the data warehouse, the model training engine periodically performs incremental training and version iteration on the first-level dust risk prediction model deployed on edge intelligent nodes and the second-level dust diffusion prediction model deployed on regional collaborative controllers, and distributes the trained model parameters to the corresponding devices for updates. The panoramic command center provides a visual human-computer interaction interface, displaying the overall network monitoring situation, equipment status, alarm information, and supporting managers in macro-strategy formulation, manual intervention of model parameters, and historical data analysis.

[0012] Furthermore, the main control computing unit of the edge intelligent node also performs the following operations: continuously monitoring environmental response data after the local control actuator is activated. Within 3 minutes of the local control actuator's activation, the main control computing unit collects particulate matter concentration data at a higher frequency, i.e., once per second, and calculates the concentration decrease rate. If the concentration decrease rate is lower than a preset efficiency threshold, it is determined that the single-point control efficiency is insufficient, and the main control computing unit immediately sends a collaborative support request to the regional collaborative controller through the first wireless communication module. This request includes the current node identifier and real-time monitoring data.

[0013] Furthermore, in response to a collaborative support request from an edge intelligent node, the collaborative computing unit of the regional collaborative controller executes the following collaborative decision-making process: First, based on the second-level dust diffusion prediction model, it quickly analyzes the dust sources and diffusion paths around the requesting node. Second, from the spatially adjacent upwind edge intelligent nodes, it selects nodes that are not currently in a high-intensity operation state and have redundant control resources, marking them as candidate collaborative nodes. Finally, through a collaborative optimization algorithm, it recalculates and generates a temporary collaborative group containing the requesting node and all candidate collaborative nodes, assigning an optimized combination of control parameters to each node in the collaborative group, instructing the nodes in the group to synchronously start or adjust their operations, forming a regional joint prevention and control effect.

[0014] Furthermore, the local control actuator of the edge intelligent node adopts a multimodal operation design, specifically including a high-pressure fine water mist nozzle and an intelligent dry fog dust suppression device. The high-pressure fine water mist nozzle is used to deal with rapidly rising, localized peak particulate matter concentrations, with a fast response speed and relatively concentrated coverage. The intelligent dry fog dust suppression device is used to deal with continuous, large-scale dust diffusion, with low water consumption and long dust suppression duration. Based on the predicted concentration change trend output by the first-level dust risk prediction model, the main control computing unit specifies the actuator mode to be activated and the initial operating parameters in the first-level control command.

[0015] Furthermore, the first and second wireless communication modules exchange data using a dedicated wireless communication protocol based on a hybrid time division multiple access and frequency hopping mechanism. This protocol allocates the highest priority transmission time slots and the most stable communication frequencies to control commands and emergency alarm data, ensuring that the transmission delay of critical commands remains stable below 50 milliseconds, while routine monitoring data is transmitted using the remaining communication resources.

[0016] Furthermore, the system operates within a hierarchical decision-making framework. This framework clearly defines the decision-making authority and response time targets for each level. The edge intelligent node layer is responsible for rapid, autonomous local responses at the millisecond to second level. The regional collaborative controller layer is responsible for regional situational assessment and multi-node collaborative optimization at the minute level. The cloud management platform layer is responsible for model optimization, strategy adjustment, and global monitoring at the hourly to daily level. Information exchange and command transmission between each level are achieved through standardized data interfaces and event-driven mechanisms.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a three-layer distributed architecture consisting of edge intelligent nodes, regional collaborative controllers, and a cloud management platform. This architecture moves the core prediction and decision-making capabilities from the central cloud to the network edge. The edge intelligent nodes have local data processing and real-time decision-making capabilities, and can complete the closed loop from perception to execution within 200 milliseconds. This fundamentally eliminates the inherent delays caused by remote data transmission and centralized processing, and enables an ultra-fast response to sudden construction dust events, effectively curbing the spread of pollution.

[0018] 2. This invention introduces a two-level prediction and coordination mechanism based on spatiotemporal correlation. Edge nodes perform local microscale predictions based on time-series data to achieve rapid self-control; the regional controller performs mesoscale diffusion predictions and multi-objective optimization based on spatial multi-point data and fluid dynamics models to achieve intelligent coordination. This mechanism enables the system not only to respond to pollution that has already occurred, but also to predict risks in advance and optimize resource allocation, transforming passive governance into proactive prevention and precise control, significantly improving overall dust suppression efficiency and energy utilization efficiency.

[0019] 3. This invention designs a dynamic strategy management and adaptive learning system. The regional collaborative controller can dynamically adjust the decision thresholds of edge nodes based on the global situation, and the cloud platform can continuously optimize the prediction model using historical data. This enables the system to adapt to the environmental characteristics and changes of different construction sites, and to continuously improve itself as the operating time increases, maintaining long-term effectiveness and advancement, overcoming the shortcomings of poor adaptability of traditional static threshold systems.

[0020] 4. This invention achieves flexible scheduling and fault mitigation of system resources through equipment status monitoring and collaborative support request mechanisms. When the efficiency of single-point control is insufficient, regional collaboration can be automatically triggered to mobilize surrounding resources to form joint prevention and control, ensuring that the overall control capability of the system does not diminish when local equipment performance deteriorates or encounters abnormally strong dust sources, thus enhancing the robustness and reliability of the system. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the three-layer distributed collaborative processing and hierarchical decision-making framework in this invention; Figure 3 This is a millisecond-level closed-loop logic flowchart of the local perception-prediction-execution process of edge intelligent nodes in this invention; Figure 4 This is a flowchart illustrating the logical flow of diffusion prediction and multi-node collaborative optimization performed by the regional collaborative controller in this invention. Figure 5 This is a schematic diagram of the information interaction and collaborative support data flow between the three layers of edge nodes, regional controllers and cloud platforms in this invention. Detailed Implementation

[0022] Example 1: The overall technical architecture of the real-time monitoring and control system for construction dust based on intelligent sensors proposed in this invention is shown in the attached figure. Figure 1 As shown, a three-tiered distributed collaborative processing system is adopted, consisting of edge intelligent nodes, regional collaborative controllers, and a cloud management platform. This system strictly adheres to the principles of layered decoupling, proximity-based response, and global optimization in its physical deployment, data flow, and decision-making logic. This ensures that the system can meet millisecond-level local rapid response requirements while possessing minute-level regional collaborative control capabilities and hour-level or longer global policy evolution capabilities. The following will combine the attached diagram... Figure 1 To be continued Figure 5 It provides a detailed explanation of the structure, function implementation, interaction mechanism, and operation process of each component of the system.

[0023] First, edge intelligent nodes, serving as the system's front-end sensing and execution units, are densely deployed at key dust source locations on the construction site, such as high-risk areas like earthwork excavation areas, material storage areas, vehicle entrances and exits, and crushing operation areas. Each edge intelligent node is a highly integrated embedded device, containing four core subsystems: a micro-meteorological sensor array, a main control computing unit, a local control actuator, and a first wireless communication module. The micro-meteorological sensor array consists of a high-precision laser scattering particulate matter concentration sensor, a digital temperature and humidity sensor, and an ultrasonic anemometer and wind direction sensor. It continuously collects five-dimensional environmental parameters at the deployment points, including aerosol particulate matter concentration (primarily PM10 and PM2.5), ambient temperature, relative humidity, three-dimensional wind speed vector, and wind direction angle, with a fixed sampling period of 30 seconds. All raw sensor data undergoes hardware-level signal conditioning circuitry for filtering, amplification, and analog-to-digital conversion before entering the main control computing unit, ensuring a signal-to-noise ratio of no less than 40 dB.

[0024] The main control computing unit adopts a low-power heterogeneous computing architecture, integrating an ARM Cortex-A series application processor and a dedicated neural network acceleration coprocessor. This unit contains two core software modules: a first-level dust risk prediction model and a first-level control decision logic. The first-level dust risk prediction model is a pruned and optimized Long Short-Term Memory (LSTM) time-series prediction model. Its input dimension is 5 (corresponding to the aforementioned five-dimensional environmental parameters), and its time step is 10, meaning it uses the historical monitoring data sequence of the previous 10 consecutive sampling periods (a total of 300 seconds) as model input. The model output is the predicted value of local particulate matter concentration for the next two sampling periods (i.e., within the next 60 seconds) and its corresponding confidence probability. This model has completed initial training before leaving the factory and is continuously optimized during system operation through incremental update packages distributed from the cloud. The model inference process is entirely completed locally, with a single inference time controlled within 15 milliseconds.

[0025] The first-level control decision logic is implemented in the form of a state machine. Its core judgment condition is: when the predicted future particulate matter concentration output by the first-level dust risk prediction model exceeds the preset first-level concentration threshold (default value is 150 micrograms per cubic meter, dynamically adjustable), and the corresponding confidence level is higher than 85%, the main control computing unit immediately generates a first-level control command. This command includes fields such as actuator start flag, operating mode selection code, initial spray pressure setpoint, and basic operating duration. After the command is generated, the main control computing unit triggers the electrical drive circuit through a high-speed GPIO interface, directly controlling the local control actuator to start. The end-to-end latency from data acquisition to actuator action is strictly limited to within 200 milliseconds. This millisecond-level closed-loop response mechanism is shown in the attached figure. Figure 3 As shown, this constitutes the first line of defense for the system in responding to sudden dust storms.

[0026] The local control actuator adopts a multi-modal operation design, specifically including two physical actuators: a high-pressure fine water mist nozzle and an intelligent dry fog dust suppression device. The high-pressure fine water mist nozzle operates at a pressure range of 7 to 10 MPa, with a nozzle orifice diameter of 0.1 mm, producing fine water mist with particle sizes between 10 and 50 micrometers. It is suitable for handling localized peak scenarios where particulate matter concentration rises sharply in a short period, with a start-up response time of less than 100 milliseconds and a coverage radius of approximately 5 meters. The intelligent dry fog dust suppression device utilizes the Venturi effect principle, atomizing water into ultrafine dry fog particles of 1 to 10 micrometers using compressed air. It operates at a pressure of 0.5 to 0.7 MPa, with a coverage range of up to 15 meters, suitable for long-term suppression of large-scale, continuous dust pollution. Based on the slope of the concentration change trend output by the first-level dust risk prediction model, the main control calculation unit intelligently selects the actuator mode to be activated in the first-level control command: if the predicted concentration shows a steep upward trend (the absolute value of the slope is greater than 5 micrograms / cubic meter / second), the high-pressure fine water mist nozzle is activated first; if the predicted concentration shows a slow but continuous upward trend, the intelligent dry fog dust suppression device is activated.

[0027] In addition, the main control computing unit also undertakes the task of evaluating local control effectiveness. Within 3 minutes of the local control actuator being activated, the main control computing unit automatically switches to high-frequency monitoring mode, increasing the particulate matter concentration sampling period from 30 seconds to 1 second. During this period, the system calculates the rate of decrease in particulate matter concentration in real time. The calculation method is to perform linear fitting on the concentration data within the most recent 30 seconds and take the slope of the fitted line as the rate of decrease value. If the rate of decrease is lower than the preset effectiveness threshold (default value is -2 micrograms / cubic meter / second), the system determines that the current single-point control effectiveness is insufficient. At this time, the main control computing unit immediately encapsulates a collaborative support request data packet. This data packet contains the unique identifier of the current edge intelligent node, the current five-dimensional environmental parameters, the activated actuator mode, operating parameters, and effectiveness evaluation results, and sends it to the corresponding regional collaborative controller with the highest priority through the first wireless communication module.

[0028] The first wireless communication module employs a dedicated wireless communication protocol based on a hybrid time division multiple access (TDMA) and frequency hopping mechanism. This protocol divides the communication cycle into fixed-length time-slot frames, each containing 16 time slots. The first two time slots are permanently reserved for transmitting control commands and emergency alarm data, while the remaining 14 time slots are used for polling and uploading routine monitoring data. Simultaneously, the system pre-sets eight frequency hopping points within the 2.4 GHz ISM band, and the protocol dynamically selects the optimal frequency point for hopping based on the Channel Quality Indicator (CQI). For critical data such as collaborative support requests and control commands, the system mandates the use of reserved time slots and the frequency point with the best channel quality for transmission, ensuring that the end-to-end transmission delay remains stable below 50 milliseconds and the packet loss rate is less than 0.1%.

[0029] The regional collaborative controller, serving as the system's intermediate coordination layer, is typically deployed in a temporary control room or signal tower at the construction site. Its wireless coverage radius is no less than 200 meters, effectively managing all edge intelligent nodes within that radius. The regional collaborative controller establishes a bidirectional, reliable data connection with all edge intelligent nodes within its coverage area via a second wireless communication module. Internally, the controller comprises two main functional modules: a collaborative computing unit and a strategy management unit. The core task of the collaborative computing unit is to run the second-level dust diffusion prediction model and collaborative optimization algorithm. The second-level dust diffusion prediction model is a two-dimensional gridded diffusion model constructed based on simplified principles of computational fluid dynamics, dividing the construction site into several regular 5m × 5m grid units. The model uses the five-dimensional environmental parameters reported in real-time by all edge intelligent nodes within the area as dynamic boundary conditions, combined with digital three-dimensional geographic information of the construction site (including static obstacle data such as building outlines, material stockpile heights, and fence locations) synchronously acquired from the cloud management platform. It solves the simplified Navier-Stokes equations and particulate transport equations to simulate the spatial diffusion path and concentration distribution of dust within the next 5 minutes, ultimately generating a dynamically updated risk concentration distribution heat map. The heatmap is updated every 30 seconds, with a resolution accurate to a single grid cell.

[0030] The collaborative optimization algorithm takes the heatmap of the risk concentration distribution as input and constructs a multi-objective optimization problem with the dual objective functions of maximizing the overall dust suppression effect and minimizing total energy consumption. The dust suppression effect is quantified by the reduction in the total area of ​​grid cells in the heatmap whose concentration exceeds the warning threshold (e.g., 200 micrograms per cubic meter); total energy consumption is measured by the cumulative power consumption of the control actuators of all edge intelligent nodes within the region. This optimization problem is solved using a multi-objective particle swarm optimization algorithm (MOPSO). The algorithm initializes the particle swarm size to 50, the maximum number of iterations to 100, and uses a linearly decreasing inertia weight strategy. The position vector of each particle encodes the control intensity correction coefficient (ranging from 0.5 to 2.0) and start / stop timing offset (ranging from -30 seconds to +30 seconds) of all edge intelligent nodes within the region. The algorithm ultimately outputs a set of Pareto optimal solutions. The strategy management unit selects a compromise solution that balances effect and energy efficiency from these solutions, generates a personalized control parameter combination for each edge intelligent node, and distributes it for execution via a second wireless communication module.

[0031] The strategy management unit is responsible for dynamically managing the decision thresholds and baseline control intensity of edge intelligent nodes. This unit continuously receives long-term strategy parameters (such as seasonal dust characteristic coefficients and construction phase weighting factors) from the cloud management platform and integrates the regional risk situation information output by the collaborative computing unit in real time. Specifically, the strategy management unit defines a regional risk level index R, the calculation formula of which is as follows: Where N is the total number of edge smart nodes in the region. Let i be the current measured particulate matter concentration at the i-th node. The warning threshold is set at 200 micrograms per cubic meter (default). This is the critical threshold (default 300 micrograms per cubic meter). When the regional risk level index R is greater than 0.6, the strategy management unit determines that the overall risk level has increased, and then issues a reduced first-level concentration threshold (e.g., from 150 micrograms per cubic meter to 120 micrograms per cubic meter) and an enhanced baseline control intensity (e.g., increasing the base pressure of the high-pressure fine water mist nozzle from 8 MPa to 9 MPa) to all edge intelligent nodes in the region. These parameter adjustment instructions are sent in broadcast form, with a version number and time tag to ensure the consistency of the edge node status.

[0032] When the regional collaborative controller receives a collaborative support request from an edge intelligent node, the collaborative computing unit immediately initiates an emergency collaborative decision-making process. This process is as follows: Figure 4 As shown, firstly, based on the second-level dust diffusion prediction model, a high-precision re-simulation is performed on the grid cells within a 50-meter radius of the requesting node to identify potential upwind dust sources and dominant diffusion paths. Secondly, the system traverses all edge intelligent nodes located within the upwind sector (within ±30 degrees of the wind direction angle centered on the requesting node), selecting nodes that are not currently in a high-intensity operation state (defined as having a current control intensity correction coefficient less than 1.2) and have sufficient water tank capacity (greater than 30% of the total capacity), marking them as candidate collaborative nodes. Finally, the collaborative optimization algorithm constructs a temporary collaborative group with the requesting node and all candidate collaborative nodes as objects, and re-runs the multi-objective particle swarm optimization algorithm to generate a unique combination of control parameters for each node within the collaborative group.

[0033] The new parameter combination not only includes the control intensity correction coefficient and the timing offset, but may also specify specific actuator mode combinations (for example, requiring upwind nodes to activate high-pressure fine water mist to intercept the source, and downwind nodes to activate dry fog to prolong the suppression time). After the command is issued, all nodes in the group execute synchronously, forming a spatially continuous and temporally coordinated joint prevention and control effect, thereby effectively making up for the inadequacy of single-point control capabilities.

[0034] The cloud management platform serves as the system's top-level monitoring and evolution hub, maintaining connectivity with all regional collaborative controllers via 4G / 5G wide area networks or fiber optic leased lines. This platform comprises three main subsystems: a data warehouse, a model training engine, and a panoramic command center. The data warehouse employs a distributed time-series database architecture, using dual indexes based on time partitioning and device identifiers. It persistently stores all raw monitoring data uploaded by edge intelligent nodes (sampling frequency of one record every 30 seconds), event logs (including control commands, performance evaluation results, equipment fault alarms, etc.), and collaborative decision-making records reported by regional collaborative controllers (including risk heatmap snapshots, optimization parameter schemes, etc.). The data retention strategy is to retain hot data for 30 days, lukewarm data for 180 days, and cold data archived to object storage. The total storage capacity is planned based on approximately 1.2 gigabytes of data generated daily per square kilometer of construction site.

[0035] The model training engine, based on historical big data in the data warehouse, regularly performs incremental model training tasks. For the first-level dust risk prediction model deployed on edge intelligent nodes, the training engine extracts high-quality time-series samples (requiring continuous, complete samples with no missing data and containing complete "prediction-actual" pairs) from the same construction phase and similar meteorological conditions weekly. Using transfer learning, it fine-tunes the original model, focusing on optimizing the ability to capture concentration abrupt changes within the next 60 seconds. For the second-level dust diffusion prediction model deployed on the regional collaborative controller, the training engine uses monthly accumulated multi-node synchronous observation data to calibrate key physical parameters such as turbulent diffusion coefficient and settling rate in the model through inversion methods, improving the model's prediction accuracy under complex terrain. After training, the new model is generated in the form of differential update packages and distributed to the corresponding devices through a secure encrypted channel. The devices automatically load the new model during low-load periods at night and perform A / B testing to verify performance improvements before officially switching to the new model.

[0036] The panoramic command center provides a web-based visual human-computer interaction interface, supporting multi-level drill-down data display. Administrators can view a real-time monitoring map of the entire network, where color block density and shade intuitively reflect the current dust risk level of each area. It can monitor the online status, remaining resources (such as water tank volume and battery power), and historical operation records of all edge intelligent nodes and regional collaborative controllers. It can receive and process automatically generated alarm information (such as insufficient single-point efficiency, excessive regional risk, and communication interruption). Furthermore, the command center supports manual intervention in macro-level strategies, such as manually adjusting long-term strategy parameters for a specific area, forcibly freezing the automatic updates of a certain type of model, or issuing temporary ultra-high priority commands to specific edge nodes. All manual operations are fully recorded in the audit log to ensure traceability.

[0037] The entire system operates within a strict hierarchical decision-making framework, as shown in the attached document. Figure 2As shown in the diagram, the framework clearly defines the following: the decision-making authority of the edge intelligent node layer focuses on millisecond-level autonomous response to local emergencies, with a target response time of less than 200 milliseconds; the decision-making authority of the regional collaborative controller layer focuses on minute-level collaborative optimization of multi-node resources within the region, with a target response time of less than 2 minutes; and the decision-making authority of the cloud management platform layer focuses on model evolution, strategy evolution, and global supervision, with a target response time of hours or longer. Information exchange between the layers occurs through standardized JSON-formatted data interfaces and an event-driven mechanism. For example, when an edge node triggers a collaborative support request, the event is published to a message queue with a specific topic. The regional collaborative controller, as a subscriber, immediately consumes the event and initiates an emergency response process. After the regional controller completes a collaborative optimization, its decision result is packaged into a "collaborative decision event" and reported to the cloud for display by the panoramic command center and analysis by the model training engine. This clear division of responsibilities and efficient information flow mechanism is illustrated in the attached diagram. Figure 5 As shown, this ensures that the system can respond quickly and coordinate globally in complex and ever-changing construction site environments, combining agility and intelligence.

[0038] In summary, this embodiment, by constructing a three-layer distributed architecture and deeply integrating advanced technologies such as edge computing, spatiotemporal prediction, multi-objective optimization, and adaptive learning, achieves a fundamental transformation in construction dust control, from passive monitoring to proactive prevention, from single-point control to regional collaboration, and from static thresholds to dynamic evolution. While ensuring ultra-low latency response, the system significantly improves the utilization efficiency of dust suppression resources and the overall control effect, providing solid technical support for smart construction site construction.

[0039] Example 2: In certain special construction scenarios, such as ultra-large infrastructure projects (e.g., airports, ports, high-speed rail hubs) or mountainous projects with complex terrain, the coverage capability of a single regional collaborative controller may be insufficient to support the needs of full-domain collaboration. Therefore, based on Example 1, this example introduces a cascaded architecture of multi-level regional collaborative controllers to expand the spatial coverage and collaborative granularity of the system.

[0040] Specifically, the system adds a converged regional collaborative controller on top of the existing regional collaborative controller. The original regional collaborative controller is redefined as an access-type regional collaborative controller, responsible for managing edge intelligent nodes within a 200-meter diameter range and performing the same local collaborative optimization tasks as in Example 1. Multiple access-type regional collaborative controllers (typically 4 to 8) are connected to the converged regional collaborative controller via wired or high-bandwidth wireless links. The converged regional collaborative controller has a coverage range of 1 to 2 kilometers and also includes a collaborative computing unit and a policy management unit, but its function focuses on larger-scale spatial collaboration.

[0041] The collaborative computing unit of the converged regional collaborative controller runs the third-level dust dispersion prediction model. This model, based on the second-level model, further incorporates atmospheric boundary layer stability parameters and regional background pollution concentrations as inputs, and widens the grid resolution to 20m × 20m to accommodate simulations at a larger spatial scale. The model uses the risk concentration distribution heatmaps reported by all subordinate connected regional collaborative controllers as the initial field, combined with regional wind field forecast data provided by the meteorological department, to predict the migration and accumulation trend of dust across the entire project area within the next 15 minutes.

[0042] The objective function of its collaborative optimization algorithm has also been upgraded accordingly. In addition to the suppression effect and energy consumption, a third objective, "minimizing cross-regional pollution transfer," has been added. The algorithm coordinates the control intensity of the sub-regions under the jurisdiction of different access-type regional collaborative controllers to avoid airflow disturbances caused by excessive local dust suppression, which could push pollutants to neighboring areas. The optimization results are distributed to each access-type regional collaborative controller in the form of a control intensity upper limit, which the latter must not exceed during its internal optimization. This two-level regional collaborative mechanism retains the fine-grained control capability within small areas while achieving macro-level balance between large areas, effectively solving the problem of collaborative dust control at ultra-large-scale construction sites.

[0043] Furthermore, in terms of communication, the access-type and aggregation-type regional collaborative controllers adopt a wireless mesh network protocol based on the IEEE 802.11s standard to ensure a highly reliable and low-latency backbone communication link even in complex construction site environments. All cross-level data interactions undergo two-way authentication and AES-256 encryption to ensure system information security.

[0044] Through the above extensions, this embodiment seamlessly extends the system's applicability from small and medium-sized construction sites to ultra-large and complex projects, while maintaining the core millisecond-level response and intelligent collaboration characteristics, further enhancing the universality and engineering practical value of the invention.

Claims

1. A real-time monitoring and control system for construction dust based on intelligent sensors, characterized in that, include: Multiple edge intelligent nodes are deployed at the construction site; each edge intelligent node includes a micro-meteorological sensor array, a main control computing unit, a local control actuator, and a first wireless communication module; the micro-meteorological sensor array is used to collect data on aerosol particulate matter concentration, temperature, humidity, wind speed, and wind direction at the deployment points; the main control computing unit has a built-in first-level dust risk prediction model and first-level control decision logic; At least one regional collaborative controller; the regional collaborative controller maintains a bidirectional data connection with all edge intelligent nodes within the coverage area through a second wireless communication module; A cloud management platform; the cloud management platform is connected to all regional collaborative controllers via a wide area network.

2. The real-time monitoring and control system for construction dust based on intelligent sensors according to claim 1, characterized in that, The first-level dust risk prediction model is a time-series prediction model based on long short-term memory network. It takes the local historical monitoring data sequence of 10 consecutive sampling periods, each period being 30 seconds, as input and outputs the predicted value of local particulate matter concentration and its confidence level for the next 2 sampling periods. The first-level control decision logic is used to compare the predicted particulate matter concentration value with the preset first-level concentration threshold. When the predicted concentration exceeds the first-level concentration threshold and the confidence level is higher than 85%, a first-level control command is generated and the local control actuator is started through the electrical drive circuit.

3. The real-time monitoring and control system for construction dust based on intelligent sensors according to claim 2, characterized in that, The regional collaborative controller includes a collaborative computing unit and a policy management unit; The collaborative computing unit runs the second-level dust diffusion prediction model and collaborative optimization algorithm; The second-level dust diffusion prediction model is a gridded model based on the simplified principle of computational fluid dynamics. It uses the real-time monitoring data reported by all edge intelligent nodes in the area as boundary conditions, combined with the digital three-dimensional geographic information of the construction site, to simulate the spatial diffusion trend of dust in the next 5 minutes and generate a dynamic risk concentration distribution heat map. The collaborative optimization algorithm is used to solve the problem based on the risk concentration distribution heat map, with the dual objectives of maximizing the overall regional dust suppression effect and minimizing the total energy consumption, through a multi-objective particle swarm optimization algorithm, and outputs the regulation intensity correction coefficient and start-stop timing scheme for each edge intelligent node. The strategy management unit is used to receive long-term strategy parameters from the cloud management platform and, in conjunction with the regional risk situation output by the collaborative computing unit, dynamically adjust the first-level concentration threshold and baseline control intensity sent to each edge intelligent node.

4. The real-time monitoring and control system for construction dust based on intelligent sensors according to claim 3, characterized in that, The cloud management platform includes a data warehouse, a model training engine, and a panoramic command center; The data warehouse is used to persistently store the raw monitoring data, event logs, and collaborative decision records reported by the regional collaborative controller uploaded by all edge intelligent nodes. The model training engine is based on historical big data accumulated in the data warehouse. It regularly performs incremental training and version iteration on the first-level dust risk prediction model deployed on the edge intelligent node and the second-level dust diffusion prediction model deployed on the regional collaborative controller, and sends the trained model parameters to the corresponding devices for updating. The panoramic command center provides a visual human-computer interaction interface to display the overall network monitoring situation, equipment status, alarm information, and supports managers in formulating macro strategies, manually intervening in model parameters, and analyzing historical data.

5. The real-time monitoring and control system for construction dust based on intelligent sensors according to claim 4, characterized in that, The main control computing unit of the edge intelligent node is also used to: collect particulate matter concentration data and calculate the concentration decrease rate at a frequency of once per second within 3 minutes after the local control actuator is started; when the concentration decrease rate is lower than the preset efficiency threshold, it is determined that the single-point control efficiency is insufficient, and a collaborative support request containing the current node identifier and real-time monitoring data is sent to the regional collaborative controller through the first wireless communication module.

6. The real-time monitoring and control system for construction dust based on intelligent sensors according to claim 5, characterized in that, In response to a collaborative support request, the collaborative computing unit of the regional collaborative controller executes the following collaborative decision-making process: based on the second-level dust diffusion prediction model, it analyzes the dust sources and diffusion paths around the requesting node; and selects nodes that are not currently in a high-intensity operation state and have redundant control resources from the upwind edge intelligent nodes in the spatially adjacent area as candidate collaborative nodes. The collaborative optimization algorithm is used to recalculate and generate a temporary collaborative group containing the requesting node and all candidate collaborative nodes. An optimized combination of control parameters is assigned to each node in the collaborative group, and the nodes in the group are instructed to start or adjust their jobs synchronously.

7. The real-time monitoring and control system for construction dust based on intelligent sensors according to claim 6, characterized in that, The local control actuator of the edge intelligent node adopts a multimodal operation design, including a high-pressure fine water mist nozzle and an intelligent dry fog dust suppression device; the main control computing unit specifies the actuator mode to be started and the initial operation parameters in the first-level control command based on the predicted concentration change trend output by the first-level dust risk prediction model.

8. The real-time monitoring and control system for construction dust based on intelligent sensors according to claim 7, characterized in that, The first wireless communication module and the second wireless communication module use a dedicated wireless communication protocol based on a hybrid time division multiple access and frequency hopping mechanism for data exchange. This protocol allocates the highest priority transmission time slots and the most stable communication frequencies to control commands and emergency alarm data, ensuring that the transmission delay of critical commands is stable below 50 milliseconds.

9. The real-time monitoring and control system for construction dust based on intelligent sensors according to claim 8, characterized in that, The system operates within a hierarchical decision-making framework. The hierarchical decision-making framework clearly defines the decision-making authority and response time targets for each level, including: the edge intelligent node layer responsible for rapid and autonomous local responses at the millisecond to second level; The system consists of a regional collaborative controller layer responsible for minute-level regional situation assessment and multi-node collaborative optimization; a cloud management platform layer responsible for hourly to day-level model optimization, strategy adjustment, and global monitoring; and information exchange and command transmission between the various layers through standardized data interfaces and event-driven mechanisms.

10. The real-time monitoring and control system for construction dust based on intelligent sensors according to claim 9, characterized in that, The process by which the strategy management unit dynamically adjusts the first-level concentration threshold and the benchmark control intensity is as follows: The strategy management unit defines the regional risk level index R, and its calculation formula is... ; When the regional risk level index R is greater than 0.6, the strategy management unit determines that the overall risk level has increased and sends the reduced first-level concentration threshold and the enhanced baseline control intensity parameters to the relevant edge intelligent nodes.

Images