Electronic greenhouse closed system based on multi-dimensional full-time management and control of port bulk cargo yard

Abstract

The present application relates to the technical field of dust pollution control, and particularly relates to a port dry bulk cargo yard electronic greenhouse closed system based on multi-dimensional full-time control. The system establishes a time reference by identifying operation events, maps multi-source sensor data to a unified time axis, and realizes space-time registration by combining Bayesian probability inference; a physical information neural network is constructed based on the registered data to predict a continuous dust concentration field; a virtual closed boundary is generated according to the concentration threshold and the closed effectiveness is evaluated; the physical parameters are calibrated by using a CFD model combined with a parameter inversion method; in the rolling prediction time domain, an optimization model weighted by spraying cost, environmental penalty and control cost is constructed to solve the optimal spraying control strategy, and the execution effect is fed back to the model to form a closed-loop optimization. The system can realize accurate management of yard dust, and can also ensure that the dust is enclosed in the electronic greenhouse in the open-air scene.

Description

Technical Field

[0001] This invention relates to the field of dust pollution control technology, and in particular to an electronic enclosure system for port dry bulk cargo yards based on multi-dimensional, real-time control. Background Technology

[0002] Port dry bulk cargo yards generate significant amounts of dust during the loading, unloading, and storage of bulk cargo such as coal and ore, impacting the surrounding environment and residents' health. Enclosed management of bulk cargo yards is necessary to prevent dust spillage. Traditional enclosed management methods primarily include constructing physical enclosures and configuring sprinkler dust suppression systems. Physical enclosure solutions create an enclosed space above the yard using steel or air-supported membrane structures, but these methods suffer from high construction costs, disruption of large equipment operations, and higher internal dust concentrations.

[0003] Existing technologies for dust control mainly suffer from the following problems: First, monitoring methods are limited, typically only 3-5 discrete monitoring points are set up around the storage yard, which cannot fully grasp the dust concentration distribution of the entire storage yard, resulting in a large number of monitoring blind spots; Second, there is a lack of accurate prediction capabilities. Although some solutions use computational fluid dynamics (CFD) simulation, the model parameters are fixed and the boundary conditions cannot be updated in real time, resulting in low prediction accuracy and long processing time, which cannot meet the needs of real-time decision-making; Third, the control strategies are crude, adopting a passive response mode of "spraying when exceeding the standard," failing to proactively optimize according to the dust diffusion trend, leading to water waste or failure to meet environmental protection standards. Summary of the Invention

[0004] To overcome the above deficiencies, this invention provides an electronic enclosure system for port dry bulk cargo yards based on multi-dimensional, real-time control. The system aims to achieve visualized, predictable, and controllable electronic enclosure of dust in open-air storage yards, effectively confining dust within the target range, thereby achieving an enclosure effect similar to that of a physical enclosure.

[0005] This invention provides the following technical solution: a port dry bulk cargo yard electronic enclosure system based on multi-dimensional, real-time control, comprising:

[0006] The multidimensional sensing and registration module is used to establish a time reference by identifying work events. Using the occurrence time of the work event as the anchor point, it maps sensor data with different sampling frequencies to a unified time axis and updates the spatial location distribution of each sensor by combining the Bayesian probabilistic inference method to obtain spatiotemporal registration data.

[0007] The environmental field reconstruction module is used to construct a physical information neural network based on spatiotemporal registration data, and introduces a loss function that includes data fitting loss and physical residual loss to predict the continuous concentration field under each spatiotemporal coordinate.

[0008] The virtual closed boundary generation module is used to extract isosurfaces from the continuous concentration field based on the environmental protection concentration threshold, generate dynamic virtual closed boundaries, determine whether they completely surround the work area, and evaluate the closed effectiveness of the electronic greenhouse.

[0009] The simulation model calibration module is used to predict dust diffusion in future periods using CFD models. After the prediction period ends, actual observation data is collected, and the boundary conditions and physical parameters of the CFD model are initially calibrated using the parameter inversion method.

[0010] The predictive optimization control module is used to establish a rolling time-domain optimization framework, with the weighted sum of operation cost, environmental penalty and control cost as the objective function, the concentration of the virtual closed boundary as the constraint, and the prediction results of the CFD model to solve the objective optimization problem to obtain the optimal operation control strategy.

[0011] The closed-loop feedback and continuous iteration module is used to feed back the actual effect of the optimal operation control strategy to the CFD model and continuously update the boundary conditions and physical parameters.

[0012] Preferably, the steps for obtaining spatiotemporal registration data include:

[0013] Identify job events through video streams to determine the time of the job event as a time anchor.

[0014] Using the aforementioned time anchor point as a reference, the multi-source sensor data are mapped and interpolated on the time axis to achieve time alignment of data with different sampling frequencies;

[0015] The spatial location distribution of each sensor is updated using a Bayesian probabilistic inference method to correct sensor spatial layout errors and establish a unified spatial coordinate system.

[0016] Preferably, the Bayesian probabilistic inference method includes the following steps:

[0017] Using the initial installation location of each sensor as the prior distribution, a prior probability model of the spatial distribution of the sensors is established by combining historical operation data.

[0018] The likelihood function is calculated based on the multi-source sensing data collected at the time of the operation event to reflect the correlation between sensor observations and spatial location;

[0019] Based on Bayes' theorem, the spatial location distribution of each sensor is updated posteriorly to obtain the optimal location estimate and confidence interval for each sensor.

[0020] Preferably, the steps of constructing a physical information neural network include:

[0021] Using the spatiotemporal coordinates of the stockpile area as input variables and dust concentration as output variables, a neural network model was established, and spatiotemporal registration data was used as supervisory samples for network training.

[0022] A composite loss function containing data fitting error terms and physical equation residual terms is introduced during network training to ensure that the prediction results simultaneously satisfy observation consistency and diffusion equation constraints.

[0023] The gradient descent method is used to optimize the network parameters until the loss function converges, thus obtaining a continuous concentration field prediction model that satisfies the constraints of physical laws.

[0024] Preferably, the formula for calculating the composite loss function is:

[0025] ;

[0026] in, This represents the data fitting loss. Represents the loss in the physical equations. Indicates boundary condition loss. This represents the loss due to the conservation of mass. , , and These are the weighting coefficients corresponding to each type of loss.

[0027] Preferably, the step of generating a dynamic virtual closed boundary includes:

[0028] A numerical isosurface extraction method is used to extract isosurfaces corresponding to a preset threshold in a continuous concentration field to form a preliminary virtual boundary geometry.

[0029] The connectivity and closure of the preliminary virtual boundary geometry are checked to determine the integrity of the boundary.

[0030] In cases where the boundary is not closed, the boundary is repaired or extended based on morphological closure to obtain a closed virtual boundary.

[0031] Preferably, the steps for evaluating the sealing effectiveness of the electronic greenhouse include:

[0032] Perform integrity checks on the virtual closed boundary to determine whether it forms a closed structure and completely surrounds the target work area;

[0033] Calculate the escape index of the virtual closed boundary to assess the degree of dust escape;

[0034] Based on the deviation relationship between the aforementioned escaping indicators and the environmental protection concentration threshold, determine whether the sealing effect of the electronic greenhouse meets the standard, and output the sealing effectiveness evaluation result.

[0035] Preferably, the preliminary calibration step of the CFD model using the parameter inversion method includes:

[0036] Construct a parameter vector to be optimized, containing the physical parameters of the CFD model;

[0037] Based on the deviation between the CFD model's predicted output and the actual observed data, an error function is defined.

[0038] The error function is solved by the Nelder-Mead optimization method to obtain the parameter combination that makes the CFD model output closest to the observed data.

[0039] Preferably, the steps to obtain the optimal job control strategy include:

[0040] Within the rolling prediction time domain, a predicted sequence of dust concentration evolution is generated based on a calibrated CFD model;

[0041] An optimization model is constructed with the weighted sum of spraying operation costs, environmental penalties, and control implementation costs as the objective function;

[0042] The concentration threshold determined by the virtual closed boundary is used as an optimization constraint to form an objective-constrained optimization problem;

[0043] The optimization problem is solved iteratively using the particle swarm optimization algorithm to obtain the optimal spray control strategy that satisfies environmental constraints.

[0044] The present invention has the following beneficial effects:

[0045] 1. This invention constructs a physical information neural network, directly embedding the partial differential equations of dust diffusion, boundary conditions, and mass conservation constraints into the network training process, enabling the model to obtain high-precision, continuous, and physically consistent three-dimensional concentration field reconstruction results with only a small amount of sensor data.

[0046] 2. This invention employs a parameter inversion method for online calibration of the CFD model. By integrating actual observation data with model predictions, it dynamically corrects key physical quantities such as wind field, turbulence parameters, and dust source intensity, enabling the CFD model to accurately reflect real-time operations and meteorological changes at the stockpile. This method effectively reduces the deviation between the model and the actual environment, improving the accuracy and stability of the predicted concentration field.

[0047] 3. This invention constructs a target optimization model within a rolling time-domain framework, taking into account the cost of spraying operations, environmental compliance deviations, and control execution costs. Through comprehensive analysis of the predicted concentration field and the virtual closed boundary, it solves for the optimal spraying control strategy that satisfies environmental constraints. Attached Figure Description

[0048] Figure 1This is a structural diagram of the electronic enclosure system for port dry bulk cargo yards based on multi-dimensional, real-time control proposed in this invention.

[0049] Figure 2 This is a flowchart illustrating the implementation of the electronic enclosure system for port dry bulk cargo yards based on multi-dimensional, real-time control proposed in this invention. Detailed Implementation

[0050] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example 1

[0052] In a first embodiment of the present invention, the present invention provides an electronic enclosure system for port dry bulk cargo yards based on multi-dimensional, real-time control, such as... Figure 1 As shown, it includes:

[0053] The multidimensional sensing and registration module is used to establish a time reference by identifying work events. Using the occurrence time of the work event as the anchor point, it maps sensor data with different sampling frequencies to a unified time axis and updates the spatial location distribution of each sensor by combining the Bayesian probabilistic inference method to obtain spatiotemporal registration data.

[0054] Preferably, the steps for obtaining spatiotemporal registration data include:

[0055] Identify job events through video streams to determine the time of the job event as a time anchor.

[0056] Using the aforementioned time anchor point as a reference, the multi-source sensor data are mapped and interpolated on the time axis to achieve time alignment of data with different sampling frequencies;

[0057] The spatial location distribution of each sensor is updated using a Bayesian probabilistic inference method to correct sensor spatial layout errors and establish a unified spatial coordinate system.

[0058] Preferably, the Bayesian probabilistic inference method includes the following steps:

[0059] Using the initial installation location of each sensor as the prior distribution, a prior probability model of the spatial distribution of the sensors is established by combining historical operation data.

[0060] The likelihood function is calculated based on the multi-source sensing data collected at the time of the operation event to reflect the correlation between sensor observations and spatial location;

[0061] Based on Bayes' theorem, the spatial location distribution of each sensor is updated posteriorly to obtain the optimal location estimate and confidence interval for each sensor.

[0062] Specifically, intelligent cameras installed high up in the stockyard continuously collect operational video streams, and a video event recognition model based on convolutional neural networks is used to detect typical dust-related events such as grab bucket unloading, belt conveyor startup, and vehicle unloading. When an operational event is detected, the system automatically extracts its occurrence time and records it as the time anchor point for the current time period.

[0063] Using a time anchor point as a reference, multi-source sensor data such as dust concentration, wind speed and direction, temperature and humidity, and equipment status are mapped to the same time axis. For data with a sampling frequency higher than the video sampling frequency, the system uses a sliding window mean to downsample and smooth the data; for data with a low sampling frequency, a linear interpolation method is used to pad the data to the same time interval, thereby completing the time alignment of multi-source data.

[0064] The three-dimensional coordinates recorded during sensor installation This serves as the initial prior distribution for spatial location. To describe the uncertainty caused by slight displacements, the prior can be modeled as a Gaussian distribution:

[0065] ;

[0066] in, This represents the covariance matrix obtained based on installation errors and historical observation statistics. Represented in three-dimensional coordinates As the mean, with The covariance follows a Gaussian distribution. The system further incorporates historical concentration gradient distributions, consistent wind direction fields, and distance stability between neighboring sensors during operation periods to construct a prior probability model of the sensor spatial distribution, which constrains its reasonable location range.

[0067] Once a work event is detected and a time anchor point is determined, the system extracts multi-source sensing data near the time anchor point, including concentration values, wind direction vectors, and covariance signals from adjacent sensors. To characterize the relationship between sensor observations and the actual spatial location, this embodiment constructs the following likelihood function:

[0068] ;

[0069] in, This represents the actual observation vector of the sensor at time t. This indicates the concentration gradient or wind field consistency response predicted based on the spatial location of the sensor. Represents the observation noise covariance matrix. Indicates As the mean, with The covariance is a Gaussian distribution. This likelihood is used to reflect the degree of agreement between the sensor's true location and the neighborhood field distribution.

[0070] Based on Bayes' theorem, this invention updates the posterior distribution of spatial location:

[0071] ;

[0072] in, Indicates a prior Gaussian distribution, Represents the likelihood function. This represents the updated posterior distribution. This represents the normalization term, which can be eliminated through integration.

[0073] Under the assumptions of Gaussian prior and Gaussian likelihood, the posterior distribution remains Gaussian, and its mean and covariance can be obtained through the following closed-form solution:

[0074] ;

[0075] ;

[0076] in, This represents the optimal spatial location estimate for each sensor. This represents the posterior covariance reflecting the confidence interval of the estimate. This represents the observation model matrix.

[0077] By following the steps above, we can obtain spatiotemporal registration data that is consistent in time and space, has unified coordinates, and has high environmental reliability.

[0078] The environmental field reconstruction module is used to construct a physical information neural network based on spatiotemporal registration data, and introduces a loss function that includes data fitting loss and physical residual loss to predict the continuous concentration field under each spatiotemporal coordinate.

[0079] Preferably, the steps of constructing a physical information neural network include:

[0080] Using the spatiotemporal coordinates of the stockpile area as input variables and dust concentration as output variables, a neural network model was established, and spatiotemporal registration data was used as supervisory samples for network training.

[0081] A composite loss function containing data fitting error terms and physical equation residual terms is introduced during network training to ensure that the prediction results simultaneously satisfy observation consistency and diffusion equation constraints.

[0082] The gradient descent method is used to optimize the network parameters until the loss function converges, thus obtaining a continuous concentration field prediction model that satisfies the constraints of physical laws.

[0083] Preferably, the formula for calculating the composite loss function is:

[0084] ;

[0085] in, This represents the data fitting loss. Represents the loss in the physical equations. Indicates boundary condition loss. This represents the loss due to the conservation of mass. , , and These are the weighting coefficients corresponding to each type of loss.

[0086] Specifically, a set of spatiotemporal coordinates is constructed within the storage yard area. ,in Corresponding to three-dimensional spatial coordinates, Corresponding to the time dimension. This four-dimensional coordinate system is used as network input, along with the dust concentration value at the corresponding time point. As the network output, the framework of the physical information neural network can be any common machine learning approach; in this embodiment, a fully connected feedforward neural network is preferred.

[0087] In training a physical information neural network, a composite loss function is used for training, specifically for data fitting loss:

[0088] ;

[0089] in, Indicates the number of sensors. Indicates the first The network predicts concentration values ​​from a single sensor. Indicates the first The actual concentration values ​​of each sensor. For the loss in the physical equation:

[0090] ;

[0091] ;

[0092] in, This indicates the number of PDE residuals, which is the number of sampling points randomly selected within the region. Represents the physical residual term. This represents the partial derivative of concentration with respect to time. Represents the wind speed vector. Indicates the dust diffusion coefficient. Represents the concentration gradient. This represents the second-order partial derivative of the concentration. For the boundary condition loss:

[0093] ;

[0094] in, Indicates the number of boundary points. This represents the concentration gradient along the boundary normal. For mass conservation loss:

[0095] ;

[0096] in, Indicates the target spatial region. Indicates the boundary of the region. This represents the partial derivative of concentration with respect to time. Represents the boundary normal vector. , Represents infinitesimal elements of volume and area.

[0097] During training, the network parameters are initialized, and the total loss L, consisting of four types of losses, is calculated in each training round. The Adam optimizer is used for gradient descent until the loss converges or reaches a preset threshold, and finally a PINN continuous concentration field prediction model that satisfies physical constraints is obtained.

[0098] Through the above steps, the physical information neural network constructed by the system can not only learn the concentration distribution characteristics based on the observation data, but also achieve physical consistency through diffusion equations, boundary conditions and mass conservation constraints, thereby obtaining high-precision, full-space, and continuous dust concentration field reconstruction results.

[0099] The virtual closed boundary generation module is used to extract isosurfaces from the continuous concentration field based on the environmental protection concentration threshold, generate dynamic virtual closed boundaries, determine whether they completely surround the work area, and evaluate the closed effectiveness of the electronic greenhouse.

[0100] Preferably, the step of generating a dynamic virtual closed boundary includes:

[0101] A numerical isosurface extraction method is used to extract isosurfaces corresponding to a preset threshold in a continuous concentration field to form a preliminary virtual boundary geometry.

[0102] The connectivity and closure of the preliminary virtual boundary geometry are checked to determine the integrity of the boundary.

[0103] In cases where the boundary is not closed, the boundary is repaired or extended based on morphological closure to obtain a closed virtual boundary.

[0104] Specifically, based on the continuous concentration distribution output by the physical information neural network The storage yard operation area is divided into regular three-dimensional meshes, and the corresponding concentration value is obtained at each spatial mesh node. Then, based on the environmental concentration threshold set by local environmental protection requirements, locations with concentrations equal to the threshold are searched in the three-dimensional concentration field, and corresponding triangular mesh structures are constructed. Finally, the vertex coordinates of the mesh structures are extracted. and triangular facet set .

[0105] A topological graph is constructed based on the adjacency relationships between vertices and faces. The topology graph is analyzed using depth-first search to determine if multiple disconnected boundary segments exist. Simultaneously, the Euler characteristic of the topology is calculated to determine if the boundaries form a closed three-dimensional envelope.

[0106] When an unclosed boundary is detected, the system performs geometric repair on the initial virtual boundary. For small-scale gaps caused by local fluctuations in the concentration field or mesh discretization, the system performs expansion and contraction processing on the boundary region to fill local voids and form a natural closure. For large-scale fractures caused by factors such as wind direction changes, the system first extracts the edge points of all openings and generates patch surfaces based on this set of edge points. A smooth connection is formed at the fracture point through numerical fitting. The generated patch is merged with the original boundary mesh to obtain a continuous and complete closed virtual boundary.

[0107] Through the above steps, a dynamic, closed, and physically plausible three-dimensional virtual closed boundary can be successfully generated based on a continuous concentration field, thereby achieving accurate depiction of the outer contour of dust diffusion.

[0108] Preferably, the steps for evaluating the sealing effectiveness of the electronic greenhouse include:

[0109] Perform integrity checks on the virtual closed boundary to determine whether it forms a closed structure and completely surrounds the target work area;

[0110] Calculate the escape index of the virtual closed boundary to assess the degree of dust escape;

[0111] Based on the deviation relationship between the aforementioned escaping indicators and the environmental protection concentration threshold, determine whether the sealing effect of the electronic greenhouse meets the standard, and output the sealing effectiveness evaluation result.

[0112] Specifically, the integrity of the virtual closed boundary is again determined using depth-first search and Eulerian characteristic. Further, several representative points within the ground projection range or three-dimensional spatial domain of the stockpile are selected, and each point is assessed using ray casting to determine if it is within a polyhedron. The proportion of points determined to be inside the boundary is then counted. If all representative points are inside the virtual closed boundary, the stockpile is determined to be completely enclosed by the boundary; if any points are not enclosed, the closed structure is determined to be incomplete.

[0113] To further assess whether dust has crossed the virtual closed boundary, the system calculates an escape index based on a continuous concentration field. In one feasible implementation, an annular region with a grid thickness of 1–3 layers is extracted outside the virtual closed boundary as an escape monitoring zone. The concentration field within this monitoring zone is statistically analyzed, and its average or maximum concentration is defined as the escape index to quantify the degree of dust escape.

[0114] The effectiveness of the electronic greenhouse enclosure is determined by analyzing the deviation between the emission index and the environmental protection concentration threshold. If the emission index is within the allowable range, the enclosure is deemed to be effective; if the emission index exceeds the threshold, the enclosure is deemed to be ineffective.

[0115] Through the above steps, a quantitative evaluation mechanism for the sealing effect of virtual closed boundaries can be formed, enabling real-time monitoring of the sealing performance of electronic greenhouses.

[0116] The simulation model calibration module is used to predict dust diffusion in future periods using CFD models. After the prediction period ends, actual observation data is collected, and the boundary conditions and physical parameters of the CFD model are initially calibrated using the parameter inversion method.

[0117] Preferably, the preliminary calibration step of the CFD model using the parameter inversion method includes:

[0118] Construct a parameter vector to be optimized, containing the physical parameters of the CFD model;

[0119] Based on the deviation between the CFD model's predicted output and the actual observed data, an error function is defined.

[0120] The error function is solved by the Nelder-Mead optimization method to obtain the parameter combination that makes the CFD model output closest to the observed data.

[0121] Specifically, based on the characteristics of the dust diffusion process at the stockpile, physical parameters of the CFD model that have a significant impact on prediction accuracy are selected as the parameters to be inverted. In one feasible implementation, the selected parameters to be inverted include the surface roughness coefficient. Dust source intensity Humidity-related diffusion coefficient and near-ground wind speed profile parameters The above parameters are combined to form the parameter vector to be optimized. And set its value range, and the initial parameters can be given based on empirical values ​​or historical simulation results.

[0122] After the CFD model completes one prediction cycle, actual observed concentration data from multiple monitoring points within the stockpile are acquired. Meanwhile, the CFD model outputs the predicted concentration at the same location. Define the error function used for parameter inversion, for example, using the mean squared error form:

[0123] ;

[0124] in, Indicates the number of monitoring points. Indicates parameters The discrepancy between the CFD model predictions and the observed data.

[0125] The Nelder-Mead simplex optimization method was used to optimize the parameter vector. Perform iterative updates. Specifically, using the initial parameters... Centered on a set step size, multiple parameter combinations are generated to form an initial simplex. In each iteration, the error function values ​​of each parameter combination are sorted, and reflection is performed on the point with the largest error. If the error can be further reduced, expansion is performed; if reflection fails, contraction or compression is performed, gradually approaching the optimal solution. The iteration terminates when the error change is below a threshold or the simplex size is below a set range. Reflection, expansion, contraction, and compression are all basic operations in the Nelder-Mead method: reflection is used to mirror the exploration of the outer space of the simplex; expansion is used to increase the step size in the direction of error reduction; contraction is used to move the search range closer to the optimal region; and compression is used to reduce the overall size of the simplex when the search stalls.

[0126] Through the above steps, the CFD model can automatically adjust key physical parameters, gradually aligning model predictions with actual observation data and improving the reliability of environmental field predictions.

[0127] The predictive optimization control module is used to establish a rolling time-domain optimization framework, with the weighted sum of operation cost, environmental penalty and control cost as the objective function, the concentration of the virtual closed boundary as the constraint, and the prediction results of the CFD model to solve the objective optimization problem to obtain the optimal operation control strategy.

[0128] Preferably, the steps to obtain the optimal job control strategy include:

[0129] Within the rolling prediction time domain, a predicted sequence of dust concentration evolution is generated based on a calibrated CFD model;

[0130] An optimization model is constructed with the weighted sum of spraying operation costs, environmental penalties, and control implementation costs as the objective function;

[0131] The concentration threshold determined by the virtual closed boundary is used as an optimization constraint to form an objective-constrained optimization problem;

[0132] The optimization problem is solved iteratively using the particle swarm optimization algorithm to obtain the optimal spray control strategy that satisfies environmental constraints.

[0133] Specifically, within the rolling prediction time domain, a CFD model calibrated using the parametric inversion method is used as the prediction engine. Inputting the current sensor data, the CFD model is run to obtain the dust concentration distribution sequence over multiple future prediction steps: .

[0134] To achieve a comprehensive optimization of both the economic efficiency and environmental compliance of sprinkler systems, the system constructs a weighted objective function that incorporates sprinkler operation costs, environmental penalties, and control implementation costs:

[0135] ;

[0136] In one feasible implementation, the predicted intra-step water consumption for spraying is used as a cost item for spraying operations. The difference between the predicted dust concentration and the regionally stipulated concentration threshold will be used as an environmental penalty item. The predicted number of spray start-stop cycles within a step is used as the control execution cost item. After normalizing the above indicators, a weighted average is calculated, with the weighting coefficients being as follows: , and .

[0137] By incorporating the isosurface structure of the virtual closed boundary module, the system uses the boundary concentration limit as an optimization constraint to ensure that the optimization strategy keeps dust within the virtual closed boundary. In addition, other constraints can be added to construct a multi-objective constrained optimization problem, such as the amount of water used for spraying or the number of spray start-stop cycles.

[0138] The above optimization problem is solved using the particle swarm optimization algorithm. The specific execution process is as follows:

[0139] Construct an initial particle swarm, where each particle represents a complete spray control strategy, for example... Continuous sequence of step size ;

[0140] Based on the controllable range of the spraying equipment, multiple candidate control strategies are randomly generated. For each particle, the strategy is input into the CFD model to predict the concentration field, the objective function is calculated, and the boundary constraints are checked.

[0141] Update the strategy value according to the standard rules of particle swarm optimization, so that the particles are close to both the local optimum and the global optimum at the same time.

[0142] Repeat the update steps until the set number of iterations is reached or the objective function converges. Ultimately, the optimal spray control strategy that satisfies environmental constraints, is economical in terms of spraying, and ensures smooth equipment start-up and shutdown is obtained.

[0143] Through the above steps, an executable, economical, and environmentally compliant spray control strategy is generated based on future dust diffusion predictions. At the same time, the stability and service life of the equipment are improved by reducing the number of spray start-ups and shutdowns.

[0144] The closed-loop feedback and continuous iteration module is used to feed back the actual effect of the optimal operation control strategy to the CFD model and continuously update the boundary conditions and physical parameters.

[0145] Example 2

[0146] In port dry bulk cargo yards, a large amount of dust is generated during the loading, unloading, and stacking of materials. This dust, carried by the wind, not only affects the air quality in the yard's operating area but also poses a potential hazard to the surrounding environment and residents' health. Traditional dust control methods mainly rely on fixed spray equipment or simple shielding measures. However, in open-air yards, wind speed and direction change frequently, making it difficult to confine dust within the yard and easily leading to dust concentrations exceeding standards.

[0147] To address the aforementioned issues, the electronic enclosure system for port dry bulk cargo yards based on multi-dimensional, real-time control, provided by this invention, was adopted. The implementation process of this system is as follows: Figure 2 As shown, it includes:

[0148] By identifying work events to establish a time reference, and using the occurrence time of the work events as the anchor point, sensor data with different sampling frequencies are mapped to a unified time axis. The spatial location distribution of each sensor is updated by combining the Bayesian probabilistic inference method to obtain spatiotemporal registration data.

[0149] A physical information neural network is constructed based on spatiotemporal registration data, and a loss function including data fitting loss and physical residual loss is introduced to predict the continuous concentration field under each spatiotemporal coordinate.

[0150] Based on the environmental protection concentration threshold, isosurfaces are extracted from the continuous concentration field to generate dynamic virtual closed boundaries and determine whether they completely surround the work area, thereby evaluating the sealing effectiveness of the electronic greenhouse.

[0151] The CFD model was used to predict dust dispersion in future periods. After the prediction period ended, actual observation data were collected, and the physical parameters of the CFD model were initially calibrated using the parameter inversion method.

[0152] A rolling time-domain optimization framework is established, with the weighted sum of operation cost, environmental penalties and control costs as the objective function, the concentration of the virtual closed boundary as the constraint, and the prediction results of the CFD model combined to solve the objective optimization problem to obtain the optimal operation control strategy.

[0153] The actual effects of the optimal operation control strategy are fed back to the CFD model, and the boundary conditions and physical parameters are continuously updated.

[0154] Through the implementation of this system, dust from the storage yard can be effectively confined within a virtual closed boundary, achieving a similar enclosed effect to an electronic greenhouse. At the same time, the spray control strategy based on rolling optimization optimizes the spray water consumption and start-stop frequency while ensuring environmental compliance. The system calibrates the CFD model through parameter inversion and combines it with physical information neural network prediction to achieve precise control and closed-loop adaptive optimization of dust diffusion, thereby achieving a comprehensive effect of precise dust control, economical and efficient operation, and long-term stable management.

[0155] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A port dry bulk cargo yard electronic enclosure system based on multi-dimensional, real-time control, characterized in that: include: The multidimensional sensing and registration module is used to establish a time reference by identifying work events. Using the occurrence time of the work event as the anchor point, it maps sensor data with different sampling frequencies to a unified time axis and updates the spatial location distribution of each sensor by combining the Bayesian probabilistic inference method to obtain spatiotemporal registration data. The environmental field reconstruction module is used to construct a physical information neural network based on spatiotemporal registration data, and introduces a loss function that includes data fitting loss and physical residual loss to predict the continuous concentration field under each spatiotemporal coordinate. The virtual closed boundary generation module is used to extract isosurfaces from the continuous concentration field based on the environmental protection concentration threshold, generate dynamic virtual closed boundaries, determine whether they completely surround the work area, and evaluate the closed effectiveness of the electronic greenhouse. The simulation model calibration module is used to predict dust diffusion in future periods using CFD models, collect actual observation data after the prediction period ends, and preliminarily calibrate the physical parameters of the CFD model using the parameter inversion method. The predictive optimization control module is used to establish a rolling time-domain optimization framework, with the weighted sum of operation cost, environmental penalty and control cost as the objective function, the concentration of the virtual closed boundary as the constraint, and the prediction results of the CFD model to solve the objective optimization problem to obtain the optimal operation control strategy. The closed-loop feedback and continuous iteration module is used to feed back the actual effect of the optimal operation control strategy to the CFD model and continuously update the boundary conditions and physical parameters. The steps for constructing a physical information neural network include: Using the spatiotemporal coordinates of the stockpile area as input variables and dust concentration as output variables, a neural network model was established, and spatiotemporal registration data was used as supervisory samples for network training. A composite loss function containing data fitting error terms and physical equation residual terms is introduced during network training to ensure that the prediction results simultaneously satisfy observation consistency and diffusion equation constraints. The gradient descent method is used to optimize the network parameters until the loss function converges, thus obtaining a continuous concentration field prediction model that satisfies the constraints of physical laws. The formula for calculating the composite loss function is as follows: ； in, This represents the data fitting loss. Represents the loss in the physical equations. Indicates boundary condition loss. This represents the loss due to the conservation of mass. , , and These are the weighting coefficients corresponding to each type of loss; In training a physical information neural network, a composite loss function is used for training, specifically for data fitting loss: ； in, Indicates the number of sensors. Indicates the first The network predicts concentration values ​​from a single sensor. Corresponding to three-dimensional spatial coordinates, Corresponding to the time dimension, Indicates the first The actual concentration value of each sensor; For the loss in the physical equations: ； ； in, This indicates the number of PDE residuals, which is the number of sampling points randomly selected within the region. Represents the physical residual term. This represents the partial derivative of concentration with respect to time. Represents the wind speed vector. Indicates the dust diffusion coefficient. Represents the concentration gradient. This represents the second-order partial derivative of concentration; For boundary condition loss: ； in, Indicates the number of boundary points. This represents the concentration gradient along the boundary normal. Regarding the loss due to mass conservation: ； in, Indicates the target spatial region. Indicates the boundary of the region. This represents the partial derivative of concentration with respect to time. Represents the boundary normal vector. , Represents infinitesimal elements of volume and area.

2. The port dry bulk cargo yard electronic enclosure system based on multi-dimensional all-time control as described in claim 1, characterized in that, The steps to obtain spatiotemporal registration data include: Identify job events through video streams to determine the time of the job event as a time anchor. Using the aforementioned time anchor point as a reference, the multi-source sensor data are mapped and interpolated on the time axis to achieve time alignment of data with different sampling frequencies; The spatial location distribution of each sensor is updated using a Bayesian probabilistic inference method to correct sensor spatial layout errors and establish a unified spatial coordinate system.

3. The port dry bulk cargo yard electronic enclosure system based on multi-dimensional all-time control as described in claim 2, characterized in that, The Bayesian probabilistic inference method includes the following steps: Using the initial installation location of each sensor as the prior distribution, a prior probability model of the spatial distribution of the sensors is established by combining historical operation data. The likelihood function is calculated based on the multi-source sensing data collected at the time of the operation event to reflect the correlation between sensor observations and spatial location; Based on Bayes' theorem, the spatial location distribution of each sensor is updated posteriorly to obtain the optimal location estimate and confidence interval for each sensor.

4. The port dry bulk cargo yard electronic enclosure system based on multi-dimensional all-time control as described in claim 1, characterized in that, The steps for generating dynamic virtual closed boundaries include: A numerical isosurface extraction method is used to extract isosurfaces corresponding to a preset threshold in a continuous concentration field to form a preliminary virtual boundary geometry. The connectivity and closure of the preliminary virtual boundary geometry are checked to determine the integrity of the boundary. In cases where the boundary is not closed, the boundary is repaired or extended based on morphological closure to obtain a closed virtual boundary.

5. The port dry bulk cargo yard electronic enclosure system based on multi-dimensional all-time control as described in claim 1, characterized in that, The steps for assessing the sealing effectiveness of an electronic greenhouse include: Perform integrity checks on the virtual closed boundary to determine whether it forms a closed structure and completely surrounds the target work area; Calculate the escape index of the virtual closed boundary to assess the degree of dust escape; Based on the deviation relationship between the aforementioned escaping indicators and the environmental protection concentration threshold, determine whether the sealing effect of the electronic greenhouse meets the standard, and output the sealing effectiveness evaluation result.

6. The port dry bulk cargo yard electronic enclosure system based on multi-dimensional all-time control as described in claim 1, characterized in that, The preliminary calibration steps for the CFD model using the parameter inversion method include: Construct a parameter vector to be optimized, containing the physical parameters of the CFD model; Based on the deviation between the CFD model's predicted output and the actual observed data, an error function is defined. The error function is solved by the Nelder-Mead optimization method to obtain the parameter combination that makes the CFD model output closest to the observed data.

7. The port dry bulk cargo yard electronic enclosure system based on multi-dimensional all-time control as described in claim 1, characterized in that, The steps to obtain the optimal job control strategy include: Within the rolling prediction time domain, a predicted sequence of dust concentration evolution is generated based on a calibrated CFD model; An optimization model is constructed with the weighted sum of spraying operation costs, environmental penalties, and control implementation costs as the objective function; The concentration threshold determined by the virtual closed boundary is used as an optimization constraint to form an objective-constrained optimization problem; The optimization problem is solved iteratively using a numerical optimization algorithm to obtain the optimal sprinkler control strategy that meets environmental constraints.

Images