A dust concentration space-time prediction and hierarchical early warning method and system for dry bulk cargo wharf

Abstract

The embodiment of the application provides a kind of dry bulk cargo wharf dust concentration space-time prediction and hierarchical early warning method and system, please involve big data technical field, by according to the time series fluctuation form of dust concentration signal and space position mark, construct dust concentration spatial distribution field evolving along time axis, and deduce dust diffusion potential distribution based on concentration change gradient between different space points;The dust concentration space-time prediction sequence is generated by inputting the two into the cross-space diffusion state deduction network along the preset future time window and recursively generating the dust concentration space-time prediction sequence;The hierarchical early warning distribution map of the space position set consisting of different early warning level marks is generated by comparing the predicted distribution map of each time node with early warning threshold;It is matched with the preset disposal strategy to generate dust hierarchical early warning instruction carrying space coordinate chain and early warning response scheme.The present application realizes the accurate prediction and hierarchical early warning of wharf dust.

Description

Technical Field

[0001] This application relates to the field of big data technology, and more specifically, to a method and system for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals. Background Technology

[0002] As the core location for loading, unloading, and transshipment of bulk cargo, dry bulk cargo terminals inevitably generate a large amount of dust during loading and unloading operations. Exceeding the dust concentration limit not only seriously endangers the health of workers and the quality of the surrounding air environment, but may also cause safety accidents such as dust explosions. Therefore, it is necessary to implement accurate prediction and timely early warning of dust concentration in the storage yard.

[0003] Existing approaches typically rely on threshold-triggered alarms triggered by single concentration sensors at fixed locations. This means that an alarm is triggered when the dust concentration at a monitoring point exceeds a preset threshold. However, this approach has significant limitations when dealing with the large-scale, dynamically changing dust diffusion processes in storage yards. Firstly, current technology lacks the ability to perform spatiotemporal correlation modeling of dust concentration response signals from multiple spatial locations within the storage yard to construct a spatial distribution field of dust concentration. It cannot characterize the concentration gradient and diffusion potential energy distribution of dust at different spatial locations, resulting in a severe deficiency in predicting dust diffusion trends. Secondly, current technology lacks intelligent means to recursively predict the future evolution of the dust concentration field using cross-spatiotemporal extrapolation networks. It cannot generate spatiotemporal prediction sequences of dust concentration for future timeframes, making early warning decisions lack foresight. Furthermore, at the early warning response level, current technology can only provide simple alarms at a single threshold level, lacking the refined control capability to generate tiered early warnings and match differentiated handling strategies based on the comparison results of predicted concentration values ​​at different spatial locations with the early warning threshold. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a method and system for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals.

[0005] In conjunction with the first aspect of this application, a method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals is provided, applied to a system for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals. The method includes: By deploying an array of environmental sensing devices in the dry bulk cargo terminal yard area, a multi-dimensional environmental monitoring data stream carrying timestamp and spatial location markers is collected. The multi-dimensional environmental monitoring data stream includes yard dust concentration response signals, yard wind speed response signals, yard wind direction response signals and yard atmospheric humidity response signals that are synchronously acquired at different spatial measurement points at different time nodes. Based on the time series fluctuation pattern and spatial location markers of the dust concentration response signal in the storage yard, a spatial distribution field of dust concentration evolving over time in the dry bulk cargo terminal storage yard area is constructed. Based on the concentration change gradient between different spatial measurement points in the spatial distribution field of dust concentration, the dust diffusion potential energy distribution in the dry bulk cargo terminal storage yard area is derived. The dust concentration spatial distribution field and the dust diffusion potential energy distribution are input into the spatiotemporal diffusion state inference network. The spatiotemporal diffusion state inference network recursively infers the diffusion evolution process of the dust concentration field in the dry bulk cargo terminal yard area along the preset future time window, generating a spatiotemporal prediction sequence of dust concentration composed of dust concentration prediction distribution maps corresponding to each future time node. Extract the dust concentration prediction distribution map of each future time node in the dust concentration spatiotemporal prediction sequence, compare the dust concentration value marked at each spatial location in the dust concentration prediction distribution map with the dust concentration warning threshold, and obtain a dust concentration graded warning distribution map composed of spatial location sets marked with different warning levels. The spatial location set corresponding to the warning level markers in the dust concentration classification warning distribution map is mapped and matched with the preset warning handling strategy to generate a dust classification warning instruction carrying a spatial location coordinate chain and a corresponding warning response scheme.

[0006] In conjunction with the second aspect of this application, a spatiotemporal prediction and classification early warning system for dust concentration at a dry bulk cargo terminal is provided. The system includes a machine-readable storage medium and a processor. The machine-readable storage medium stores machine-executable instructions. When the processor executes the machine-executable instructions, the system implements the aforementioned spatiotemporal prediction and classification early warning method for dust concentration at a dry bulk cargo terminal.

[0007] In conjunction with a third aspect of this application, a computer-readable storage medium is provided, wherein computer-executable instructions are stored therein, and when the computer-executable instructions are executed, the aforementioned method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals is implemented.

[0008] Combining any of the above aspects, firstly, by deploying an array of environmental sensing devices in the storage yard area, multi-dimensional environmental monitoring data streams containing dust concentration, wind speed, wind direction, and atmospheric humidity, carrying timestamps and spatial location markers, are simultaneously collected. Then, based on the time-series fluctuation pattern of dust concentration response signals and spatial location markers, a spatial distribution field of dust concentration evolving with the time axis is constructed. Based on the concentration change gradient between different spatial measurement points, the dust diffusion potential energy distribution is deduced. On this basis, the spatial distribution field of dust concentration and the diffusion potential energy distribution are input into a cross-temporal diffusion state extrapolation network to recursively generate a spatiotemporal prediction sequence of dust concentration along a preset future time window. The dust concentration prediction distribution map of each time node in the prediction sequence is extracted and compared with the warning threshold to generate a hierarchical warning distribution map composed of spatial location sets marked with different warning levels. This map is then mapped and matched with a preset warning response strategy to generate a dust hierarchical warning instruction carrying a spatial location coordinate chain and a corresponding warning response scheme. This significantly improves the spatiotemporal foresight, hierarchical precision, and targeted response strategy of dust concentration warnings in dry bulk cargo terminal storage yards. Attached Figure Description

[0009] Figure 1 A flowchart illustrating the method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals provided in this application embodiment. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] In the following description, the terms "first, second, third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0012] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0013] Figure 1This document illustrates a flowchart of a method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals, provided in an embodiment of this application. This method is applied to dust pollution monitoring and early warning scenarios in the storage yard area of ​​dry bulk cargo terminals. The storage yard area of ​​dry bulk cargo terminals stores bulk materials such as coal and ore, which are prone to dust generation under wind conditions. An array of environmental sensing devices is deployed in the storage yard area, including dust concentration sensors, ultrasonic wind speed and direction sensors, and atmospheric humidity sensors. The deployment locations of these sensors cover key spatial measurement points such as the storage yard boundary, work passages, and the top of the stockpiles. Each sensor transmits the collected data to an edge computing node in real time via a LoRa wireless communication module. The deployment and data collection of the environmental sensing device array have obtained compliant authorization from the port management department. The collected environmental parameters are used only for dust diffusion prediction and early warning purposes and do not involve the collection of any personally identifiable information or privacy data. The dust concentration sensor uses the light scattering principle, the wind speed and direction sensor uses the ultrasonic time-of-flight method, and the atmospheric humidity sensor uses a capacitive polymer humidity-sensitive element. All sensors have undergone metrological calibration before deployment.

[0014] Step S110: Collect multi-dimensional environmental monitoring data streams carrying timestamps and spatial location markers by deploying an array of environmental sensing devices in the dry bulk cargo terminal yard area. The multi-dimensional environmental monitoring data streams include yard dust concentration response signals, yard wind speed response signals, yard wind direction response signals, and yard atmospheric humidity response signals that are synchronously acquired at different spatial measurement points at different time nodes.

[0015] In this embodiment, the environmental sensing device array includes N1 dust concentration sensors, N2 wind speed and direction sensors, and N3 atmospheric humidity sensors. Each sensor has a unique device identifier and known three-dimensional spatial coordinates. The sensors synchronously sample at a fixed sampling interval dt, generating a data frame for each sampling. The data frame includes a sampling time timestamp t, the device identifier, spatial location coordinates (x, y, z), and the sensor response signal value. The dust concentration response signal is measured in micrograms per cubic meter, the wind speed response signal in meters per second, the wind direction response signal in degrees, and the atmospheric humidity response signal in percentage. All data frames are aggregated to the edge computing node via the MQTT protocol. The edge computing node aligns the data frames according to their timestamps, aggregating data frames from different sensors at the same timestamp into time-slice data blocks. A time-slice data block contains four arrays: array D1 stores the response signal values ​​and corresponding spatial coordinates of each dust concentration sensor; array D2 stores the response signal values ​​and corresponding spatial coordinates of each wind speed sensor; array D3 stores the response signal values ​​and corresponding spatial coordinates of each wind direction sensor; and array D4 stores the response signal values ​​and corresponding spatial coordinates of each atmospheric humidity sensor. The sequence of time-slice data blocks arranged in ascending order of timestamps constitutes the multi-dimensional environmental monitoring data stream S1.

[0016] Step S120: Based on the time series fluctuation pattern and spatial location markers of the dust concentration response signal in the dry bulk cargo terminal yard area, construct the dust concentration spatial distribution field that evolves with the time axis in the dry bulk cargo terminal yard area, and deduce the dust diffusion potential energy distribution in the dry bulk cargo terminal yard area based on the concentration change gradient between different spatial measurement points in the dust concentration spatial distribution field.

[0017] Step S121: Extract the dust concentration time-series waveform formed by arranging the dust concentration response signal along the time axis at each spatial measurement point in the stockpile, and perform wave morphology decomposition on the dust concentration time-series waveform to obtain the first wave morphology component characterizing the rate of change of dust concentration and the second wave morphology component characterizing the oscillation period of dust concentration.

[0018] The response signal values ​​of each dust concentration sensor at continuous time nodes are extracted from the multi-dimensional environmental monitoring data stream S1 and arranged in ascending order of timestamps to form the dust concentration time-series waveform C(t) for that spatial measurement point. Empirical Mode Decomposition (EMD) is used to decompose C(t) into a wave pattern. The decomposition process is achieved by screening intrinsic mode functions (IMFs). The screening iteration steps of EMD are as follows: identify all local maxima and local minima of C(t), construct the upper envelope U(t) and lower envelope L(t) respectively using cubic spline interpolation, calculate the mean envelope M(t) = 0.5 × (U(t) + L(t)), and subtract M(t) from C(t) to obtain the candidate IMF h(t). If h(t) does not meet the IMF condition, repeat the above steps until the condition is met to obtain the first IMF1(t). Using C(t) - IMF1(t) as the new signal to be decomposed, the decomposition process is repeated to obtain the second IMF2(t). The first wave morphology component is denoted as IMF1(t), which characterizes the high-frequency rate of change of dust concentration, with dimensions in micrograms per cubic meter. The second wave morphology component is the sum of IMF2(t) and the subsequent low-frequency intrinsic mode functions, which characterizes the low-frequency oscillation period of dust concentration, with dimensions in micrograms per cubic meter.

[0019] Step S122: Spatiotemporally bind the first and second wave morphology components with the spatial location markers of the corresponding spatial measurement points to generate dust concentration wave morphology features carrying spatial location markers.

[0020] For the i-th spatial measurement point, its spatial coordinates (xi, yi, zi), the first fluctuation morphology component IMF1i(t), and the second fluctuation morphology component IMF2i(t) are combined into a data structure. This data structure contains spatial coordinate attributes, high-frequency component time series attributes, and low-frequency component time series attributes. The spatiotemporal binding method is to construct a key-value pair mapping relationship with spatial coordinates as the primary key and the two fluctuation morphology components as subordinate attributes. The dust concentration fluctuation morphology features of all spatial measurement points constitute the feature set F1.

[0021] Step S123: Input the dust concentration fluctuation morphology characteristics corresponding to all spatial measurement points into the spatial interpolation grid constructor. The spatial interpolation grid constructor outputs the dust concentration characterization values ​​on the continuous spatial grid nodes covering the dry bulk cargo terminal yard area, forming a dust concentration spatial distribution field.

[0022] The spatial interpolation grid constructor employs the Kriging interpolation algorithm. First, the storage area is divided into a regular rectangular grid with Xnum rows and Ynum columns, and a grid resolution (the distance between adjacent nodes) of dxy. For each grid node g, its spatial coordinates are (xg, yg, zg), where zg is obtained from a table in the digital elevation grid data. Using the spatial coordinates and fluctuation morphology components of each spatial measurement point in feature set F1 as known sampling points, ordinary Kriging interpolation is used to estimate the dust concentration characterization value of node g. The Kriging interpolation weights are obtained by solving a system of linear equations with a semi-variance function as the kernel. The semi-variance function γ(h) is calculated based on the spatial distance h between sampling point pairs and the concentration difference, and is the mean of γ(h) = 0.5 × (C(location a) - C(location b))², where the spatial distance between location a and location b is h. For each time node t, Kriging interpolation is performed on each grid node g to obtain the concentration characterization value Cg(t). The concentration characterization values ​​of all grid nodes are arranged in rows and columns to form a two-dimensional array. This two-dimensional array is the dust concentration spatial distribution field Mt at the current time node t, with Ynum as the number of rows and Xnum as the number of columns.

[0023] Step S124: Along the time axis, calculate the concentration difference of the dust concentration spatial distribution field at adjacent time nodes for each grid node, and divide it by the time interval between adjacent time nodes to obtain the concentration change rate field per unit time at each spatial grid node.

[0024] For adjacent time nodes t and t+dt, the spatial distribution fields of dust concentration, Mt and Mt+dt, are read. For each grid node g, the concentration difference ΔCg = Cg(t+dt) - Cg(t). The concentration change rate Rg = ΔCg / dt, with dimensions in micrograms per cubic meter per second. The Rg values ​​of all grid nodes are arranged in rows and columns to form the concentration change rate field Rt.

[0025] Step S125: Extract the direction vector and magnitude of the concentration change rate from the concentration change rate field, and perform directional consistency fusion with the wind speed response signal and wind direction response signal of the corresponding spatial grid node to generate a dust diffusion direction field carrying the wind field driving direction and the concentration decay direction.

[0026] For each grid node g, the direction vector of the concentration change rate is obtained by calculating the spatial gradient direction of that node in the concentration distribution field. The Sobel operator is used to calculate the gradient components Gx and Gy of the concentration distribution field Mt in the x and y directions, respectively. The direction vector direction angle θg = arctan(Gy / Gx), and the modulus Mg = √(Gx² + Gy²). The wind direction angle φg and wind speed value Wg at time t are extracted from the spatially nearest wind speed and direction sensor of grid node g in the multi-dimensional environmental monitoring data stream S1. The direction vector direction angle θg and the wind direction angle φg are fused for directional consistency. The fusion method is to calculate the direction difference Δθ = |θg - φg|. If Δθ is less than the preset direction tolerance angle θtol, the direction vector is corrected to the wind direction; otherwise, the direction vector direction remains unchanged. The fused direction angle is denoted as θg_fused, which together with the corrected modulus Mg constitutes the dust diffusion direction field Dg. Dg is a vector field, and the vector at each grid node g is (Mg×cos(θg_fused), Mg×sin(θg_fused)).

[0027] Step S126: Based on the modulus of the concentration change rate field, normalize it with the preset spatial grid scale factor to generate a dimensionless diffusion driving intensity coefficient. Then, use the diffusion driving intensity coefficient as a weighting factor to assign it to each component of the dust diffusion direction field to synthesize a dust diffusion situation field that characterizes the dust migration trend and relative intensity.

[0028] For each grid node g, the diffusion driving intensity coefficient Kg = Mg / Mmax, where Mmax is the maximum value of all Mg in the concentration change rate field Rt at the current time step. Kg is a dimensionless coefficient. Multiplying Kg by the x and y components of the dust diffusion direction field Dg yields the dust diffusion potential field Sg = (Kg × Mg × cos(θg_fused), Kg × Mg × sin(θg_fused)). The modulus of Sg represents the relative intensity of dust migration at that grid node, and the direction represents the dominant direction of dust migration.

[0029] Step S127: Map the modulus of the concentration change rate field to the diffusion driving intensity coefficient, and synthesize the diffusion driving intensity coefficient with the dust diffusion direction field to generate the dust diffusion potential energy distribution with spatial grid nodes as carriers.

[0030] The potential energy field is synthesized as follows: For each grid node g, the dust diffusion potential energy is defined as Eg = Kg × Mg² × 0.5. The dimension of Eg is (micrograms per cubic meter per second)², which characterizes the magnitude of the potential driving force for dust diffusion at that grid node. The Eg values ​​of all grid nodes are arranged in the grid rows and columns to form the dust diffusion potential energy distribution Et.

[0031] Step S130: Input the spatial distribution field of dust concentration and the distribution of dust diffusion potential energy into the spatiotemporal diffusion state inference network. The spatiotemporal diffusion state inference network recursively infers the diffusion evolution process of dust concentration field in the dry bulk cargo terminal yard area along the preset future time window, and generates a spatiotemporal prediction sequence of dust concentration composed of dust concentration prediction distribution maps corresponding to each future time node.

[0032] Step S131: Encode the spatial distribution field of dust concentration at the current time node into an initial concentration state tensor, and encode the dust diffusion potential energy distribution at the current time node into a diffusion driving force tensor.

[0033] The dust concentration spatial distribution field Mtc at the current time node tc is a two-dimensional array with dimensions Ynum rows and Xnum columns. Mtc is encoded as an initial concentration state tensor T0, with a shape of (1, Ynum, Xnum), i.e., a two-dimensional matrix of one channel. The dust diffusion potential energy distribution Etc at the current time node tc is a two-dimensional array with dimensions Ynum rows and Xnum columns. Etc is encoded as a diffusion driving force tensor F0, with a shape of (1, Ynum, Xnum).

[0034] Step S132: The concentration characterization values ​​of each spatial grid node in the initial concentration state tensor are combined with the diffusion driving intensity coefficient and dust diffusion direction field of the corresponding spatial grid node in the diffusion driving force tensor to perform a first fusion transformation, and scale conversion is performed in combination with the preset extrapolation time step to generate a diffusion flux tensor characterizing the amount of dust transport per unit time.

[0035] The simulation time step is denoted as dtp. For each grid node g, the diffusion driving force coefficient Kg is extracted from its diffusion driving force tensor F0, and the direction vector is extracted from the dust diffusion direction field Dg in step S125. The first fusion transformation calculates the diffusion flux Φg = Kg × Cg × (cos(θg_fused), sin(θg_fused)) × dtp, where Cg is the concentration characterization value of node g in the initial concentration state tensor T0. The dimension of Φg is micrograms per cubic meter multiplied by meters. The diffusion flux Φg is a two-dimensional vector, representing the amount of dust transported along the x and y directions at node g within the simulation time step dtp. The diffusion fluxes of all grid nodes are arranged in a grid to form the diffusion flux tensor Φ, with the shape of Φ being (2, Ynum, Xnum), and the two channels corresponding to the diffusion flux components in the x and y directions, respectively.

[0036] Step S133: Integrate the diffusion flux tensor along the extrapolation time step to obtain the cumulative concentration change of each grid node within the time step, and superimpose the cumulative concentration change onto the concentration characterization value of the corresponding grid node of the initial concentration state tensor to generate the dust concentration spatial distribution prediction field for the next time node.

[0037] For each grid node g, the cumulative concentration change ΔCg is calculated from the net inflow of diffusion flux from the surrounding adjacent grid nodes to that node. The finite volume method is used to discretize the convection-diffusion equation: ΔCg = -dtp × ((Φx_g_east - Φx_g_west) / dx + (Φy_g_north - Φy_g_south) / dy), where Φx_g_east is the x-direction diffusion flux at the eastern boundary of node g, Φx_g_west is the x-direction diffusion flux at the western boundary of node g, Φy_g_north is the y-direction diffusion flux at the northern boundary of node g, Φy_g_south is the y-direction diffusion flux at the southern boundary of node g, and dx and dy are the spacing between grid nodes in the x and y directions, respectively. The predicted concentration value at the next time node tc + dtp is Cg_new = Cg + ΔCg. The Cg_new values ​​of all grid nodes constitute the dust concentration spatial distribution prediction field Mpred.

[0038] Step S134: Input the dust concentration spatial distribution prediction field of the next time node into the concentration attenuation modulation layer. The concentration attenuation modulation layer uses the atmospheric humidity response signal and wind speed response signal of the stockpile at the corresponding time node to attenuate and correct the concentration characterization value of each grid node in the dust concentration spatial distribution prediction field, and outputs the dust concentration spatial distribution correction field jointly modulated by humidity and wind speed.

[0039] The concentration attenuation modulation layer performs attenuation correction on each grid node g: Cg_corrected = Cg_new × exp(-η × Wg × Hg × dtp), where η is a constant attenuation coefficient, Wg is the wind speed response signal value at node g, and Hg is the atmospheric humidity response signal value at node g. exp(-η × Wg × Hg × dtp) is the attenuation correction factor, ranging from 0 to 1. The attenuation correction factor decreases with increasing wind speed and humidity, characterizing the physical mechanism that wind speed accelerates dust diffusion and dilution, while humidity promotes dust settling. The Cg_corrected values ​​of all grid nodes constitute the dust concentration spatial distribution correction field Mcorrected.

[0040] Step S135: Use the dust concentration spatial distribution correction field as the new initial concentration state tensor, and repeatedly execute the diffusion increment tensor generation step, the grid node superposition step, and the concentration decay modulation step until all time nodes in the preset future time pane are covered.

[0041] Using Mcorrected as the new initial concentration state tensor Tk, the dust diffusion direction field of the corrected concentration field is recalculated according to the method in step S125, the diffusion driving intensity coefficient is recalculated according to the method in step S126, and the diffusion potential energy distribution is recalculated according to the method in step S127. Then, the next time step is extrapolated according to the methods in steps S131 to S134. The process is repeated K times, where K = Tpred / dtp, and Tpred is the preset total length of the future time window. Each iteration outputs a corrected dust concentration spatial distribution field.

[0042] Step S136: Arrange the dust concentration spatial distribution correction field output in each cycle in chronological order to generate a dust concentration spatiotemporal prediction sequence composed of dust concentration prediction distribution maps corresponding to each future time node.

[0043] The Mcorrected outputs from the Kth iterations are arranged sequentially by time step number from 1 to K to form the dust concentration spatiotemporal prediction sequence Pseq. The kth element in Pseq is the dust concentration prediction distribution map for the future time node tc+k×dtp.

[0044] Step S140: Extract the dust concentration prediction distribution map for each future time node in the dust concentration spatiotemporal prediction sequence, compare the dust concentration values ​​marked at each spatial location in the dust concentration prediction distribution map with the dust concentration warning threshold, and obtain a dust concentration graded warning distribution map composed of spatial location sets marked with different warning levels.

[0045] Step S141: Traverse the dust concentration prediction distribution map for each future time node in the dust concentration spatiotemporal prediction sequence, and read the dust concentration prediction value for each spatial grid node in the dust concentration prediction distribution map.

[0046] For the k-th dust concentration prediction distribution map in the spatiotemporal prediction sequence Pseq, traverse all spatial grid nodes g in the map and read their dust concentration prediction values ​​Cg_pred.

[0047] Step S142: Obtain a pre-configured set of multi-level dust concentration warning thresholds. The set of multi-level dust concentration warning thresholds contains multiple dust concentration warning boundary values ​​arranged in ascending order of concentration value. Each dust concentration warning boundary value corresponds to a unique warning level label.

[0048] The preconfigured multi-level dust concentration early warning threshold set includes L early warning levels. The early warning boundary value array B=[b1, b2, ..., bL] is arranged in ascending order of concentration values, where b1 is the lowest early warning boundary value, and bL is the highest early warning boundary value. The early warning level label array Lvl=[L1, L2, ..., LL] corresponds to B one-to-one, where L1 is the lowest early warning level, and LL is the highest early warning level. The early warning level label L0 indicates no early warning, corresponding to the scenario where the concentration value is less than b1.

[0049] Step S143: Compare the predicted dust concentration value of each spatial grid node with the dust concentration early warning boundary values in the multi-level dust concentration early warning threshold set level by level, and determine the early warning level label to which the spatial grid node belongs in the concentration interval corresponding to the dust concentration early warning boundary value.

[0050] For each grid node g, compare Cg_pred with each element in the early warning boundary value array B in sequence. If Cg_pred<b1, the early warning level label of g is L0. If b1≤Cg_pred<b2, the early warning level label of g is L1. By analogy, if bL≤Cg_pred, the early warning level label of g is LL.

[0051] Step S144: Perform spatial connected domain aggregation on the spatial grid nodes belonging to the same early warning level label to obtain an independent spatial connected domain set corresponding to each early warning level label, where each independent spatial connected domain is composed of mutually adjacent spatial grid nodes.

[0052] For each early warning level label Lj (j ranges from 1 to L), extract the grid node set Sj with all early warning level labels being Lj. Within Sj, the four-neighborhood connectivity criterion is adopted for connected domain analysis: if two grid nodes are adjacent in the up, down, left, and right directions in the grid, they belong to the same connected domain. Traverse all nodes in Sj through breadth-first search, and aggregate mutually connected nodes into one independent spatial connected domain. For each early warning level label Lj, several independent spatial connected domains Cj1, Cj2, ..., Cjm are output.

[0053] Step S145: Extract the boundary contour of each independent spatial connected domain, and generate an early warning area boundary polygon represented by a sequence of spatial position coordinates.

[0054] For each independent spatial connected domain Cjm, the Moore neighborhood boundary tracing algorithm is used to extract its contour. Starting from the top-leftmost node of the connected domain, the boundary nodes are traced in the clockwise direction, and the spatial coordinates of the traced boundary nodes are connected in sequence to form a closed polygon. The polygon is thinned and simplified using the Douglas-Peucker algorithm to remove redundant collinear points. The simplified sequence of polygon vertex coordinates is the early warning area boundary polygon Bjm.

[0055] Step S146: Perform layer superposition on all warning level labels, warning area boundary polygons, and the spatial grid node sets contained within the warning area boundary polygons corresponding to each future time node, so as to generate a dust concentration graded warning distribution map composed of spatial position sets marked with different warning levels.

[0056] For the k-th future time node, create a layer object. The layer object includes three fields: field L stores the warning level label list of each warning area boundary polygon, field P stores the vertex coordinate sequence list of each warning area boundary polygon, and field N stores the grid node index list in each warning area. The layer object is serialized into JSON format, which is the dust concentration graded warning distribution map Wk of this time node. The dust concentration graded warning distribution maps of all time nodes constitute a set W.

[0057] Step S150: Map and match the spatial position set corresponding to the warning level label in the dust concentration graded warning distribution map with a preset warning disposal strategy, so as to generate a dust graded warning instruction carrying a spatial position coordinate chain and a corresponding warning response scheme.

[0058] Step S151: Extract the warning area boundary polygon corresponding to each warning level label under each future time node from the dust concentration graded warning distribution map, and convert the warning area boundary polygon into a spatial position coordinate chain composed of continuous spatial position coordinates.

[0059] For the dust concentration graded warning distribution map Wk of the k-th time node, traverse each warning area boundary polygon Bjm corresponding to each warning level label Lj therein. Expand the vertex coordinate sequence of Bjm into a one-dimensional coordinate chain according to the boundary tracking sequence, and each element in the coordinate chain is an (x, y, z) triplet. The start point and the end point of the coordinate chain are the same vertex, forming a closed chain.

[0060] Step S152: Call a preset warning disposal strategy mapping table, wherein the warning disposal strategy mapping table stores the correspondence between warning level labels and warning response schemes, and the warning response scheme includes a warning notification object, a warning notification content template, and a warning response linkage device identifier.

[0061] The preset warning disposal strategy mapping table is stored in the edge computing node in the form of a relational database table. The table includes four columns: column Lvl stores warning level labels, column Target stores a warning notification object list, column Template stores warning notification content template texts, and column Device stores a warning response linkage device identifier list. The warning notification object is the communication terminal identifier of the person in charge of the port safety and environmental protection department and the team leader of the storage yard operation, and the warning response linkage device is the device control address of the storage yard spray dust reduction device.

[0062] Step S153: Match the warning level marker associated with the spatial location coordinate chain with the warning level marker in the warning handling strategy mapping table, and extract the successfully matched warning response scheme.

[0063] For each spatial coordinate chain, the warning level marker Lj is used to perform an equal-value match in the warning response strategy mapping table using Lj as the query key. Upon successful matching, the Target, Template, and Device field values ​​of the corresponding row are extracted as the warning response plan.

[0064] Step S154: Encapsulate the spatial location coordinate chain, warning level marker, warning notification object, warning notification content template, and warning response linkage device identifier to generate a hierarchical warning information package for a single time node.

[0065] Construct a JSON object as a tiered early warning information package. The JSON object contains the following fields: Chain (stores the spatial location coordinate chain), Level (stores the early warning level marker), Target (stores the list of early warning notification objects), Msg (stores the complete notification text after filling the early warning area coordinate chain into the early warning notification content template), Device (stores the list of early warning response and linkage device identifiers), and Time (stores the timestamp of the future time node corresponding to this early warning).

[0066] Step S155: Arrange the graded early warning information packets of each future time node in chronological order along the time axis of the dust concentration spatiotemporal prediction sequence to generate a graded early warning time sequence information stream.

[0067] All tiered early warning information packets for future time points are arranged in ascending order of timestamps. Each time point may contain multiple tiered early warning information packets. The arranged sequence constitutes the tiered early warning time-series information stream Falert.

[0068] Step S156: Perform spatiotemporal evolution correlation of spatial location coordinate chains on adjacent time nodes of hierarchical early warning information packets in the hierarchical early warning time sequence information stream, and connect the spatial location coordinate chains of the same early warning area boundary polygon on consecutive time nodes to form the early warning area migration trajectory.

[0069] For adjacent time nodes t and t+dtp, their respective hierarchical early warning information packet lists are extracted from Falert. For each early warning area at time node t, the spatial distance between the center coordinates of its boundary polygon and the center coordinates of the boundary polygons of each early warning area at time node t+dtp is calculated. The Hungarian algorithm is then used for optimal matching, and two successfully matched early warning areas are determined to be the same early warning area at different time nodes. The spatial coordinate chains of the same early warning area that are successfully matched at consecutive time nodes are connected end-to-end in chronological order to form the early warning area migration trajectory.

[0070] Step S157: Bind the migration trajectory of the warning area with the warning response plan for each time node to generate a dust classification warning instruction carrying a spatial location coordinate chain and the corresponding warning response plan.

[0071] Each warning area migration trajectory and its corresponding warning response scheme at each time point are encapsulated into a dust classification warning instruction object. The instruction object contains a `Traj` field storing the coordinate sequence of the warning area migration trajectory, and a `Scheme` field storing the list of warning response schemes for each time point. The dust classification warning instruction is sent to the port safety and environmental monitoring platform and the PLC controller of the dust suppression spray device via the MQTT protocol.

[0072] Step S210: Extract continuous time segments of the wind speed response signal and wind direction response signal of the stockyard from the multi-dimensional environmental monitoring data stream, and synthesize the wind speed response signal and wind direction response signal of the stockyard into a wind field vector time stream.

[0073] Extract the wind speed values ​​W and wind direction values ​​θ from each wind speed and direction sensor at consecutive time nodes from S1. For the i-th sensor location, synthesize the wind field vector Vx = W × cos(θ) and Vy = W × sin(θ). The dimensions of Vx and Vy are meters per second. Arrange the Vx and Vy values ​​of all sensors in time sequence to form the wind field vector time-series flow F1.

[0074] Step S220: Perform wind direction disturbance pattern identification on the wind field vector time series flow to separate the persistent basic wind field component and the intermittent gust disturbance component.

[0075] Wavelet transform was used to perform time-frequency decomposition on the Vx and Vy time series of each sensor in F1. Morlet wavelet basis functions were selected to perform continuous wavelet transforms on Vx and Vy respectively, obtaining wavelet coefficient matrices. A scale threshold sa was set; wavelet coefficients below sa were reconstructed as persistent fundamental wind field components Vx1 and Vy1, while wavelet coefficients above sa were reconstructed as intermittent gust disturbance components Vx2 and Vy2. The dimensions of Vx1 and Vx2 are both meters per second.

[0076] Step S230: Decompose the persistent basic wind field component into a transport wind vector along the dominant wind direction of the storage yard and a diffusion wind vector perpendicular to the dominant wind direction of the storage yard.

[0077] Calculate the vector average direction angle αm for the time series of Vx1 and Vy1: αm = arctan(mean(Vy1) / mean(Vx1)). The transport wind vector Vx3 = Vx1 × cos²(αm) + Vy1 × sin(αm) × cos(αm), Vy3 = Vx1 × sin(αm) × cos(αm) + Vy1 × sin²(αm). The diffusion wind vector Vx4 = Vx1 - Vx3, Vy4 = Vy1 - Vy3. The dimensions of Vx3 and Vx4 are meters per second.

[0078] Step S240: Map the intermittent gust disturbance component as a transient wind field pulse superimposed on the transport wind vector and the diffusion wind vector.

[0079] The transient wind field pulses are directly taken as Vx2 and Vy2. The dimensions of Vx2 and Vy2 are both meters per second.

[0080] Step S250: Align the conveying wind vector with the stockpile dust concentration response signal, and extract the dust concentration change sequence along the conveying path driven by the conveying wind vector from the stockpile dust concentration response signal.

[0081] For each dust concentration sensor, extract its concentration time series C. Concentration spatial gradient. and Calculated by dividing the concentration difference between adjacent sensors by the distance between them. Delivery concentration change. The dimension of C1 is micrograms per cubic meter per second.

[0082] Step S260: Align the direction of the diffusion wind vector with the dust concentration response signal of the stockpile, and extract the dust lateral diffusion concentration change sequence driven by the diffusion wind vector in the dust concentration response signal of the stockpile.

[0083] Diffusion concentration change The dimension of C2 is micrograms per cubic meter per second.

[0084] Step S270: Time-series synchronization matching is performed between the transient wind field pulse and the instantaneous concentration peak of the stockpile dust concentration response signal to extract the gust dust pulse response concentration sequence.

[0085] Peak detection is performed on the C-time series to extract concentration spikes (ts) exceeding b times the baseline concentration. The baseline concentration is the average of the sliding window of the C-time series. At time ts, if the square root of Vx²² + Vy²² is greater than a preset pulse threshold p1, the concentration increment Cp obtained by subtracting the baseline concentration from the C value at that time is added to the gust dust pulse response concentration sequence. The dimension of Cp is micrograms per cubic meter.

[0086] Step S280: Feed back the dust concentration change sequence along the conveying path, the dust lateral diffusion concentration change sequence, and the gust dust pulse response concentration sequence to the cross-temporal diffusion state inference network, and perform wind field component weighted correction on the diffusion driving force tensor.

[0087] The absolute value of the ratio of the mean of sequence C1 to the mean of sequence C2 is denoted as λ. The ratio of the maximum value of sequence Cp to the standard deviation of sequence C is denoted as γ. The diffusion driving intensity coefficient Kg in step S126 is corrected as follows: Kg2 = Kg × (1 + λ × C1n + γ × Cpn), where C1n is the value of C1 after Min-Max normalization, and Cpn is the value of Cp after Min-Max normalization. Kg2 is substituted into step S132 to replace Kg.

[0088] Step S310: Obtain the basic digital elevation grid data of the dry bulk cargo terminal yard area, and spatially register the basic digital elevation grid data with the spatial grid nodes of the dust concentration spatial distribution field to obtain the dust concentration spatial distribution field carrying elevation attributes.

[0089] The basic digital elevation grid data is a DEM raster file of the stockpile area, with the grid resolution consistent with the spatial distribution field of dust concentration, denoted as dxy. For each grid node g, the elevation value Hg of the corresponding cell is extracted from the DEM raster, in meters. Hg is assigned to the corresponding grid node g in Mt, forming Mtdem. Each node g in Mtdem contains two attributes: Cg and Hg.

[0090] Step S320: Perform a concentration stratification profile analysis along the elevation direction on the spatial distribution field of dust concentration carrying elevation attributes, and extract horizontal distribution slices of dust concentration on different elevation layers.

[0091] The elevation range from H1 to H2 is divided into Z elevation layers according to the vertical layer spacing dz. H1 is the lowest elevation of the stockpile, and H2 is the highest elevation of the stockpile. The elevation range of the k-th elevation layer is [H1 + (k-1) × dz, H1 + k × dz]. All grid nodes whose elevation values ​​fall within this range are extracted from Mtdem, and their concentration values ​​are arranged according to their horizontal grid positions to form the k-th layer slice Sk, where Sk is a two-dimensional array of Ynum rows and Xnum columns.

[0092] Step S330: Calculate the concentration difference by dividing the horizontal distribution slices of dust concentration in adjacent elevation layers to obtain the vertical decreasing gradient distribution of dust concentration.

[0093] For adjacent k-th and k+1-th elevation layers, calculate the concentration difference array ΔSk = Sk - Sk+1, where each element in ΔSk is the concentration difference. The vertical decreasing gradient Gk = ΔSk / dz, where each element in Gk has the dimension of micrograms per cubic meter per meter. The Gk values ​​of all elevation layers constitute the vertical decreasing gradient distribution Gz of dust concentration.

[0094] Step S340: Extract the elevation critical layer with zero vertical gradient from the vertically decreasing gradient distribution of dust concentration, and determine the elevation critical layer as the vertical boundary height of dust diffusion.

[0095] Iterate through all grid nodes of each elevation level in Gk and calculate the mean Gk value of all grid nodes in that elevation level. When the mean Gk value of the k-th level is less than the preset gradient threshold g1, record the elevation value of the k-th level as Hz. Hz is the vertical boundary height of dust diffusion, in meters.

[0096] Step S350: Apply three-dimensional spatial constraints to the dust diffusion vertical boundary height and the dust concentration prediction distribution map generated by the cross-temporal diffusion state inference network, and set the dust concentration prediction values ​​of the elevation layer that exceed the dust diffusion vertical boundary height in the dust concentration prediction distribution map to zero.

[0097] For each dust concentration prediction distribution map in Pseq, traverse all grid nodes g. If its Hg is greater than Hz, then set its Cg_pred to zero.

[0098] Step S360: Perform vertical integration on the dust concentration prediction distribution map after three-dimensional spatial constraints to compress the three-dimensional spatial concentration field into a two-dimensional horizontal concentration distribution field on the ground surface.

[0099] For each constrained grid node g, its equivalent two-dimensional concentration C2D = (∑(k=1 to z1)Ck×dz) / Hz, where z1 is the total number of elevation layers satisfying Hg≤Hz, and Ck is the concentration value at position g in the k-th slice. The dimension of C2D is micrograms per cubic meter.

[0100] Step S370: Replace the corresponding dust concentration prediction distribution map in the dust concentration spatiotemporal prediction sequence with the horizontal two-dimensional concentration distribution field on the ground surface, and update the dust concentration spatiotemporal prediction sequence.

[0101] Replace the dust concentration prediction distribution map for the corresponding time node in Pseq with a two-dimensional array constructed using C2D. In the updated Pseq, each map is a two-dimensional array with Ynum rows and Xnum columns.

[0102] Step S410: Extract the predicted dust concentration values ​​of the same spatial location marker at different future time nodes from the dust concentration spatiotemporal prediction sequence to form a spatial location concentration prediction time series curve.

[0103] For each spatial grid node g, extract the concentration prediction values ​​of that node at all future time nodes from the updated Pseq, and arrange them in chronological order to form a sequence Cg of length K, where K is the total number of future time nodes.

[0104] Step S420: Identify the trend inflection point of the spatial location concentration prediction time series curve and extract the peak concentration time node when the dust concentration changes from an upward trend to a downward trend.

[0105] Calculate the first-order difference sequence Dg for Cg, where Dg(i) = Cg(i+1) - Cg(i). Iterate through Dg, and when Dg(i-1) is greater than 0 and Dg(i) is less than 0, denote the i-th time node as the concentration peak time node tp. tp is the time step index from 1 to K.

[0106] Step S430: Spatially aggregate the concentration peak time nodes marked at all spatial locations to generate a concentration peak arrival time distribution map.

[0107] Arrange the tp values ​​of all grid nodes g according to the grid rows and columns to form a concentration peak arrival time distribution map Tp. Tp is a two-dimensional array of Ynum rows and Xnum columns, and the array elements are time step index values.

[0108] Step S440: Perform spatial clustering on the concentration peak arrival time distribution map, and classify the spatial locations where the concentration peak arrival times are in the same time period into the same concentration peak synchronization region.

[0109] The K future time nodes are divided into several time periods according to a preset time period length ΔK. The j-th time period contains the time step index range [(j-1)×ΔK+1, j×ΔK]. For each time period, the set of grid nodes whose tp values ​​fall within this range is extracted from Tp, and the above nodes are spatially clustered using the four-neighbor connected component aggregation method. Each connected component obtained from the clustering is recorded as a concentration peak synchronization region.

[0110] Step S450: Extract the boundary contour of each concentration peak synchronization region to generate a polygon set of concentration peak synchronization regions.

[0111] For each concentration peak synchronization region, the Moore neighborhood boundary tracing algorithm is used to extract its contour boundary. The boundary vertex sequence is then simplified using Douglas-Puk thinning to obtain the vertex coordinate sequence of the boundary polygon. The boundary polygons of all concentration peak synchronization regions constitute a set Bp.

[0112] Step S460: Overlay the polygon set of the concentration peak synchronization area with the warning level markers of the corresponding time nodes in the dust concentration classification warning distribution map, and associate the warning level markers with the concentration peak synchronization area.

[0113] For each concentration peak synchronization region, the corresponding time period is [t1, t2]. Extract all Wk values ​​whose time step indices fall within the range [t1, t2] from the dust concentration graded early warning distribution map set W. For each grid node g within the concentration peak synchronization region, take the highest early warning level marker among the aforementioned Wk values ​​as the early warning level for that grid node. Take the mode of all early warning levels for all grid nodes within the region as the associated early warning level Lp for that concentration peak synchronization region.

[0114] Step S470: Mark the spatial location within each concentration peak synchronization area, and use the time period when the predicted dust concentration value corresponding to the spatial location mark reaches the concentration peak time node earlier than the concentration peak time node as the early warning lead time period.

[0115] For each grid node g within the concentration peak synchronization region, its concentration peak time node is tp. The time step index ta is used to detect the first time step in the Cg sequence when the concentration value reaches the lowest warning boundary value b1 corresponding to the associated warning level Lp. The warning lead time Δt = tp - ta, in time steps, is calculated. The average Δt of all grid nodes within the region is taken as the warning lead time ΔT for that region.

[0116] Step S480: Integrate the warning lead time period with the warning response scheme of the corresponding concentration peak synchronization area in the dust classification warning instruction to generate a dust classification warning instruction carrying the warning lead time period.

[0117] In the dust classification early warning instruction object, an Ad field is added to the early warning response scheme corresponding to each concentration peak synchronization area, storing ΔT. The updated instruction object is sent to the port safety and environmental monitoring platform via the MQTT protocol.

[0118] Step S510: Extract the numerical pairs of the dust concentration response signal and the atmospheric humidity response signal at the same time point in the stockpile, and construct a set of concentration and humidity numerical pairs.

[0119] Extract the concentration value C of each dust concentration sensor at time t from S1, and the humidity value H of the nearest atmospheric humidity sensor at the same time t, to form a numerical pair (C, H). All numerical pairs at all time points constitute a set Q, which contains R pairs of values.

[0120] Step S520: Fit the scatter distribution trend of the concentration and humidity values ​​to extract the concentration and humidity correlation trend line that shows the dust concentration response signal decreasing as the atmospheric humidity response signal of the stockpile increases.

[0121] Linear regression was performed using the least squares method to fit the numerical pairs in Q. The regression line equation is C = a1 × H + b1, where a1 = (R × ∑(H × C) - ∑H × ∑C) / (R × ∑(H²) - (∑H)²), and b1 = (∑C - a1 × ∑H) / R. The units of a1 and b1 are micrograms per cubic meter per percentage.

[0122] Step S530: Convert the concentration-humidity correlation trend line into a humidity-concentration inhibition coefficient curve. The humidity-concentration inhibition coefficient curve represents the dimensionless dust settling probability factor corresponding to different atmospheric humidity values.

[0123] The humidity concentration suppression coefficient S = (b1 - (a1 × H + b1)) / b1 = -a1 × H / b1. Since a1 is negative, S is positive and increases with increasing H. S is a dimensionless value between 0 and 1.

[0124] Step S540: Map the diffusion driving intensity coefficient to the atmospheric humidity response signal at the corresponding spatial grid node to obtain the corresponding dimensionless sedimentation probability factor, and multiply the diffusion driving intensity coefficient with the sedimentation probability factor to obtain the effective diffusion driving intensity coefficient after humidity attenuation effect correction.

[0125] For grid node g, obtain its atmospheric humidity value Hg, and substitute it into the formula Sg=-a1×Hg / b1 to obtain the sedimentation probability factor. The effective diffusion driving intensity coefficient Kg3=Kg×(1-Sg). Kg3 is a dimensionless value.

[0126] Step S550: Replace the original diffusion driving intensity coefficient with the effective diffusion driving intensity coefficient, and re-execute the potential energy field synthesis step to generate a humidity-corrected dust diffusion potential field.

[0127] In step S127, Kg is replaced with Kg3, and the dust diffusion potential energy Eg2 = Kg3 × Mg² × 0.5 is recalculated. The dimensions of Eg2 are the same as those of Eg.

[0128] Step S560: Input the humidity-corrected dust diffusion state field into the spatiotemporal diffusion state inference network, modulate the diffusion driving force tensor with a humidity suppression component, and use the humidity-suppressed diffusion driving force tensor modulated by the humidity suppression component to recursively deduce the diffusion evolution process of the dust concentration field and update the spatiotemporal prediction sequence of dust concentration.

[0129] In step S132, Kg is replaced by Kg3 to calculate the diffusion flux Φg = Kg3 × Cg × (cos(θg_fused), sin(θg_fused)) × dtp. Subsequent steps S133 to S136 are executed sequentially to generate the updated Pseq.

[0130] For example, in step S610: extract the concentration change trend of each spatial grid node in the dust concentration prediction distribution map of adjacent future time nodes in the spatiotemporal prediction sequence, mark the spatial grid nodes with increasing concentration prediction values ​​as concentration growth nodes, and mark the spatial grid nodes with decreasing concentration prediction values ​​as concentration decline nodes.

[0131] For the prediction maps of adjacent time nodes, calculate the concentration difference ΔC = Cg(t + dtp) - Cg(t) for each grid node g. If ΔC is greater than 0, mark g as a growth node; if ΔC is less than 0, mark g as a decline node.

[0132] Step S620: Aggregate spatially adjacent concentration growth nodes into concentration growth regions, and aggregate spatially adjacent concentration decline nodes into concentration decline regions.

[0133] The four-neighbor connected component aggregation method is used to perform connected component aggregation on the growing node set and the fading node set respectively, to obtain the growing region set Ginc and the fading region set Gdec.

[0134] Step S630: Extract the geometric center coordinates of the concentration increase region and extract the geometric center coordinates of the concentration decrease region adjacent to the concentration increase region.

[0135] For each growing region, its geometric center coordinates (xc1, yc1) are the arithmetic mean of the x and y coordinates of all nodes within the region. For each fading region spatially adjacent to the growing region, its geometric center coordinates (xc2, yc2) are calculated similarly.

[0136] Step S640: Construct a concentration migration direction vector from the geometric center coordinates of the concentration increase region to the geometric center coordinates of the concentration decrease region, and use the concentration migration direction vector as the spatial migration direction of the dust concentration field.

[0137] The concentration migration direction vectors are Vx5 = xc2 - xc1 and Vy5 = yc2 - yc1. The vector magnitude is √(Vx5² + Vy5²). The dimensions of Vx5 and Vy5 are meters.

[0138] Step S650: Calculate the sum of the products of the predicted concentration increment of each grid node in the concentration growth region and the area of ​​its corresponding grid cell to obtain the total mass of dust growth; calculate the sum of the products of the predicted concentration decrease of each grid node in the concentration decline region and the area of ​​its corresponding grid cell to obtain the total mass of dust decline; take the difference between the total mass of dust growth and the total mass of dust decline as the net mass of dust remaining in the field.

[0139] The area of ​​a single grid cell, Acell, is calculated as dxy × dxy. The total mass of dust growth in the growing region, Minc, is calculated as ∑(g∈Ginc)ΔCg_pos × Acell, where ΔCg_pos is the positive concentration increment at the growing node g. The total mass of dust regressing in the fading region, Mdec, is calculated as ∑(g∈Gdec)|ΔCg_neg| × Acell, where ΔCg_neg is the negative concentration increment at the fading node g. The net retained mass, Mnet, is calculated as Minc - Mdec. The dimensions of Minc, Mdec, and Mnet are all micrograms.

[0140] Step S660: Bind the concentration migration direction vector, the net retention mass of dust in the field, and the spatial coverage of the concentration growth area into a set of dust migration trend feature descriptions.

[0141] For each growth region, construct a migration trend feature description F2={direction vector (Vx5, Vy5), net retention quality Mnet, and the vertex coordinate sequence of the growth region boundary polygon}.

[0142] Step S670: Compare the migration direction vector carried in the dust migration trend feature description set with the migration trajectory of the warning area at the corresponding time node in the dust classification warning instruction to ensure directional consistency. Correct the direction of the migration trajectory of the warning area that deviates from the tolerance range in the comparison results.

[0143] For each growth region's migration direction vector (Vx5, Vy5), calculate its direction angle α1 = arctan(Vy5 / Vx5). Extract the direction angle α2 of the warning region's migration trajectory for the corresponding time node from the dust classification warning command. If |α1 - α2| is greater than the preset deviation tolerance angle Δα, replace α2 with α1 and regenerate the warning region's migration trajectory along the α1 direction.

[0144] Step S680: Rebind the direction-corrected warning area migration trajectory to the corresponding warning response scheme in the dust classification warning instruction, and update the dust classification warning instruction.

[0145] In the dust classification early warning instruction object, the corrected early warning area migration trajectory coordinate sequence is updated to the Traj field of the corresponding early warning response scheme. The updated instruction is sent to the port safety and environmental monitoring platform via the MQTT protocol.

[0146] In some embodiments, the spatiotemporal prediction and classification early warning system for dry bulk terminal dust concentration used to perform the above-described method can be any electronic device with data calculation, processing, and storage functions. This spatiotemporal prediction and classification early warning system for dry bulk terminal dust concentration can be used to implement the spatiotemporal prediction and classification early warning method for dry bulk terminal dust concentration provided in the above embodiments.

[0147] Typically, a dry bulk terminal dust concentration spatiotemporal prediction and classification early warning system includes a processor and memory. The processor may include one or more processing cores, such as a 4-core processor or an 8-core processor. The processor can be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field Programmable Gate Array), or PLA (Programmable Logic Array). The processor may also include a main processor and coprocessors. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0148] The memory may include one or more computer-readable storage media, which may be non-transitory. The memory may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory is used to store a computer program configured to be executed by one or more processors to implement the above-described method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals.

[0149] In an illustrative embodiment, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor of a computer device, implements the aforementioned method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals. Optionally, the aforementioned computer-readable storage medium may be ROM (Read-Only Memory), RAM (Random Access Memory), CD-ROM (CompactDisc Read-Only Memory), magnetic tape, floppy disk, and optical data storage devices, etc.

[0150] This application provides a computer program product, which includes computer-executable instructions or computer programs. When the computer-executable instructions or computer programs are executed by a processor, the processor will execute the spatiotemporal prediction and classification early warning method for dust concentration at dry bulk cargo terminals provided in this application.

[0151] This application provides a computer-readable storage medium storing computer-executable instructions or computer programs. When the computer-executable instructions or computer programs are executed by a processor, the processor will execute the spatiotemporal prediction and classification early warning method for dust concentration at dry bulk cargo terminals provided in this application.

[0152] In some embodiments, the computer-readable storage medium may be a read-only memory (ROM), random access memory (RAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, magnetic surface memory, optical disk, or CD-ROM, etc.; or it may be a device that includes one or any combination of the above-mentioned memories.

[0153] In some embodiments, computer-executable instructions may take the form of programs, software, software modules, scripts, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as stand-alone programs or as modules, components, subroutines, or other units suitable for use in a computing environment.

[0154] As an example, computer-executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple co-located files (e.g., files that store one or more modules, subroutines, or code sections).

[0155] As an example, computer-executable instructions can be deployed to execute on a single electronic device, or on multiple electronic devices located at one location, or on multiple electronic devices distributed across multiple locations and interconnected via a communication network.

[0156] Finally, it should be noted that the above-disclosed embodiments are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals, characterized in that, The method includes: By deploying an array of environmental sensing devices in the dry bulk cargo terminal yard area, a multi-dimensional environmental monitoring data stream carrying timestamp and spatial location markers is collected. The multi-dimensional environmental monitoring data stream includes yard dust concentration response signals, yard wind speed response signals, yard wind direction response signals and yard atmospheric humidity response signals that are synchronously acquired at different spatial measurement points at different time nodes. Based on the time-series fluctuation pattern and spatial location markers of the dust concentration response signal in the storage yard, a spatial distribution field of dust concentration evolving over time in the dry bulk cargo terminal storage yard area is constructed. Based on the concentration change gradient between different spatial measurement points in the spatial distribution field of dust concentration, the dust diffusion potential energy distribution in the dry bulk cargo terminal storage yard area is derived. The dust concentration spatial distribution field and the dust diffusion potential energy distribution are input into the spatiotemporal diffusion state inference network. The spatiotemporal diffusion state inference network recursively infers the diffusion evolution process of the dust concentration field in the dry bulk cargo terminal yard area along the preset future time window, generating a spatiotemporal prediction sequence of dust concentration composed of dust concentration prediction distribution maps corresponding to each future time node. Extract the dust concentration prediction distribution map of each future time node in the dust concentration spatiotemporal prediction sequence, compare the dust concentration value marked at each spatial location in the dust concentration prediction distribution map with the dust concentration warning threshold, and obtain a dust concentration graded warning distribution map composed of the set of spatial locations marked with different warning levels. The spatial location set corresponding to the warning level markers in the dust concentration classification warning distribution map is mapped and matched with the preset warning handling strategy to generate a dust classification warning instruction carrying a spatial location coordinate chain and a corresponding warning response scheme.

2. The method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals according to claim 1, characterized in that, Based on the time-series fluctuation pattern and spatial location markers of the dust concentration response signal in the storage yard, a spatial distribution field of dust concentration evolving over time in the dry bulk cargo terminal storage yard area is constructed. Based on the concentration change gradient of this spatial distribution field between different spatial measurement points, the dust diffusion potential energy distribution in the dry bulk cargo terminal storage yard area is derived, including: The dust concentration response signal of the stockpile is extracted and arranged along the time axis at each spatial measurement point to form a dust concentration time series waveform. The dust concentration time series waveform is decomposed into a wave pattern to obtain a first wave pattern component that characterizes the rate of change of dust concentration and a second wave pattern component that characterizes the oscillation period of dust concentration. The first and second wave morphology components are spatiotemporally bound to the spatial location markers of the corresponding spatial measurement points to generate dust concentration wave morphology features carrying spatial location markers. The dust concentration fluctuation morphology characteristics corresponding to all spatial measurement points are input into the spatial interpolation grid constructor, and the spatial interpolation grid constructor outputs the dust concentration characterization values ​​on the continuous spatial grid nodes covering the dry bulk cargo terminal yard area, forming a spatial distribution field of dust concentration. Along the time axis, the concentration difference of the spatial distribution field of dust concentration at adjacent time nodes is calculated for each grid node, and then divided by the time interval between the adjacent time nodes to obtain the concentration change rate field per unit time at each spatial grid node. The direction vector and magnitude of the concentration change rate are extracted from the concentration change rate field. The direction vector is then fused with the wind speed response signal and wind direction response signal of the corresponding spatial grid node to generate a dust diffusion direction field carrying the wind field driving direction and the concentration decay direction. Based on the modulus of the concentration change rate field, a preset spatial grid scale factor is used for normalization to generate a dimensionless diffusion driving intensity coefficient. The diffusion driving intensity coefficient is then used as a weighting factor and assigned to each component of the dust diffusion direction field to synthesize a dust diffusion situation field that characterizes the dust migration trend and relative intensity. The modulus of the concentration change rate field is mapped to the diffusion driving intensity coefficient, and the diffusion driving intensity coefficient is combined with the dust diffusion direction field to generate a dust diffusion potential energy distribution with spatial grid nodes as carriers.

3. The method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals according to claim 1, characterized in that, The process involves inputting the spatial distribution field of dust concentration and the distribution of dust diffusion potential energy into a spatiotemporal diffusion state extrapolation network. This network then recursively extrapolates the diffusion evolution of the dust concentration field in the dry bulk cargo terminal yard area along preset future time windows, generating a spatiotemporal prediction sequence of dust concentration composed of predicted dust concentration distribution maps corresponding to each future time node. This includes: The spatial distribution field of dust concentration at the current time node is encoded as the initial concentration state tensor, and the dust diffusion potential energy distribution at the current time node is encoded as the diffusion driving force tensor. The concentration characterization values ​​of each spatial grid node in the initial concentration state tensor are combined with the diffusion driving intensity coefficient and dust diffusion direction field of the corresponding spatial grid node in the diffusion driving force tensor to perform a first fusion transformation, and scale conversion is performed in combination with a preset extrapolation time step to generate a diffusion flux tensor characterizing the amount of dust transport per unit time. The diffusion flux tensor is integrated along the extrapolation time step to obtain the cumulative concentration change of each grid node within that time step. This cumulative concentration change is then superimposed onto the concentration characterization value of the grid node corresponding to the initial concentration state tensor to generate the dust concentration spatial distribution prediction field for the next time step. The dust concentration spatial distribution prediction field at the next time node is input into the concentration attenuation modulation layer. The concentration attenuation modulation layer uses the atmospheric humidity response signal and wind speed response signal of the stockpile at the corresponding time node to attenuate and correct the concentration characterization value of each grid node in the dust concentration spatial distribution prediction field, and outputs the dust concentration spatial distribution correction field jointly modulated by humidity and wind speed. The dust concentration spatial distribution correction field is used as a new initial concentration state tensor. The diffusion increment tensor generation step, the grid node superposition step, and the concentration decay modulation step are executed cyclically until all time nodes in the preset future time window are covered. The dust concentration spatial distribution correction field output in each cycle is arranged in chronological order to generate a dust concentration spatiotemporal prediction sequence composed of dust concentration prediction distribution maps corresponding to each future time node.

4. The method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals according to claim 1, characterized in that, The process involves extracting dust concentration prediction distribution maps for each future time node in the spatiotemporal prediction sequence of dust concentration, comparing the dust concentration values ​​marked at each spatial location in the dust concentration prediction distribution map with the dust concentration warning threshold, and obtaining a dust concentration graded warning distribution map composed of spatial locations marked with different warning levels, including: Traverse the dust concentration prediction distribution map for each future time node in the dust concentration spatiotemporal prediction sequence, and read the dust concentration prediction value for each spatial grid node in the dust concentration prediction distribution map. Obtain a pre-configured set of multi-level dust concentration warning thresholds, which includes multiple dust concentration warning boundary values ​​arranged in ascending order of concentration value, and each dust concentration warning boundary value corresponds to a unique warning level label; The predicted dust concentration value of each spatial grid node is compared step by step with the dust concentration warning boundary value in the multi-level dust concentration warning threshold set to determine the warning level label of the spatial grid node in the concentration range corresponding to the dust concentration warning boundary value. Spatial grid nodes belonging to the same warning level are aggregated into spatial connected components to obtain an independent set of spatial connected components corresponding to each warning level, wherein each independent spatial connected component is composed of mutually adjacent spatial grid nodes; For each independent spatially connected domain, the boundary contour is extracted to generate a warning area boundary polygon represented by a sequence of spatial location coordinates; By overlaying all warning level markers, warning area boundary polygons, and the set of spatial grid nodes contained within the warning area boundary polygons corresponding to each future time node, a dust concentration graded warning distribution map composed of the spatial location set of different warning level markers is generated.

5. The method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals according to claim 1, characterized in that, The step of mapping and matching the set of spatial locations corresponding to the warning level markers in the dust concentration classification warning distribution map with a preset warning response strategy to generate a dust classification warning instruction carrying a spatial location coordinate chain and a corresponding warning response scheme includes: Extract the warning area boundary polygon corresponding to each warning level marker at each future time node from the dust concentration classification warning distribution map, and convert the warning area boundary polygon into a spatial coordinate chain composed of continuous spatial coordinates. Retrieve a preset early warning and response strategy mapping table. The early warning and response strategy mapping table stores the correspondence between early warning level markers and early warning response schemes. The early warning response scheme includes the early warning notification object, the early warning notification content template, and the early warning response linkage device identifier. Match the warning level marker associated with the spatial location coordinate chain with the warning level marker in the warning handling strategy mapping table, and extract the successfully matched warning response scheme; The spatial location coordinate chain, the warning level marker, the warning notification object, the warning notification content template, and the warning response linkage device identifier are encapsulated to generate a hierarchical warning information package for a single time node. Along the time axis of the dust concentration spatiotemporal prediction sequence, the hierarchical early warning information packets of each future time node are arranged in chronological order to generate a hierarchical early warning time sequence information stream. The spatiotemporal evolution correlation of spatial location coordinate chains is performed on the hierarchical early warning information packets of adjacent time nodes in the hierarchical early warning time sequence information stream, and the spatial location coordinate chains of the boundary polygon of the same early warning area at consecutive time nodes are connected to form the migration trajectory of the early warning area. The migration trajectory of the warning area is bound to the warning response scheme at each time point to generate a dust classification warning instruction carrying a spatial location coordinate chain and the corresponding warning response scheme.

6. The method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals according to claim 3, characterized in that, The method further includes: Continuous time-series segments of the wind speed response signal and wind direction response signal of the stockyard are extracted from the multi-dimensional environmental monitoring data stream, and the wind speed response signal and wind direction response signal of the stockyard are synthesized into a wind field vector time-series stream. Wind direction disturbance pattern identification is performed on the wind field vector time series flow to separate the persistent basic wind field component and the intermittent gust disturbance component; The persistent basic wind field components are decomposed into a transport wind vector along the dominant wind direction of the stockyard and a diffusion wind vector perpendicular to the dominant wind direction of the stockyard. The intermittent gust disturbance component is mapped as a transient wind field pulse superimposed on the transport wind vector and the diffusion wind vector; Align the conveying wind vector with the dust concentration response signal of the stockpile, and extract the dust concentration change sequence along the conveying path driven by the conveying wind vector from the dust concentration response signal of the stockpile. Align the direction of the diffusion wind vector with the dust concentration response signal of the stockpile, and extract the dust lateral diffusion concentration change sequence driven by the diffusion wind vector from the dust concentration response signal of the stockpile. The transient wind field pulse is time-synchronized with the instantaneous concentration spike of the stockpile dust concentration response signal to extract the gust dust pulse response concentration sequence. The dust concentration variation sequence along the transport path, the dust concentration variation sequence in the lateral diffusion, and the dust pulse response concentration sequence in the gust are fed back to the spatiotemporal diffusion state inference network, and the diffusion driving force tensor is corrected by wind field component weighting.

7. The method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals according to claim 1, characterized in that, The method further includes: Obtain the basic digital elevation grid data of the dry bulk cargo terminal yard area, and spatially register the basic digital elevation grid data with the spatial grid nodes of the dust concentration spatial distribution field to obtain the dust concentration spatial distribution field carrying elevation attributes. A concentration stratification profile analysis along the elevation direction is performed on the spatial distribution field of dust concentration carrying elevation attributes to extract horizontal distribution slices of dust concentration at different elevation layers. The concentration difference is calculated by slicing the horizontal distribution of dust concentration in adjacent elevation layers to obtain the vertical decreasing gradient distribution of dust concentration. The elevation critical layer with zero vertical gradient is extracted from the vertically decreasing gradient distribution of dust concentration, and the elevation critical layer is determined as the vertical boundary height of dust diffusion. The vertical boundary height of dust diffusion is constrained in three dimensions by the dust concentration prediction distribution map generated by the cross-temporal diffusion state inference network, and the predicted dust concentration values ​​of the elevation layer that exceed the vertical boundary height of dust diffusion in the dust concentration prediction distribution map are set to zero. Vertical integration is performed on the dust concentration prediction distribution map after three-dimensional spatial constraints to compress the three-dimensional spatial concentration field into a two-dimensional horizontal concentration distribution field on the ground surface; Replace the corresponding dust concentration prediction distribution map in the dust concentration spatiotemporal prediction sequence with the horizontal two-dimensional concentration distribution field at the ground surface, and update the dust concentration spatiotemporal prediction sequence.

8. The method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals according to claim 1, characterized in that, The method further includes: The dust concentration prediction values ​​of the same spatial location marker at different future time nodes are extracted from the dust concentration spatiotemporal prediction sequence to form a spatial location concentration prediction time series curve. The trend inflection point of the spatial location concentration prediction time series curve is identified, and the peak concentration time node when the dust concentration changes from an upward trend to a downward trend is extracted. Spatially aggregate the concentration peak time nodes marked at all spatial locations to generate a concentration peak arrival time distribution map; Spatial clustering is performed on the concentration peak arrival time distribution map, and spatial locations where the concentration peak arrival times are in the same time period are marked and grouped into the same concentration peak synchronization region. Extract the boundary contour of each concentration peak synchronization region to generate a polygon set of concentration peak synchronization regions; The polygon set of the concentration peak synchronization area is overlaid with the warning level marker of the corresponding time node in the dust concentration classification warning distribution map, and the warning level marker is associated with the concentration peak synchronization area. For each concentration peak synchronization region, the time period during which the predicted dust concentration value corresponding to the spatial location mark reaches the concentration peak time node is earlier than the concentration peak time node is taken as the early warning lead time period. The warning response scheme is integrated with the corresponding concentration peak synchronization area in the dust classification warning instruction to generate a dust classification warning instruction carrying the warning lead time period.

9. The method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals according to claim 3, characterized in that, The method further includes: Extract the numerical pairs of the dust concentration response signal and the atmospheric humidity response signal at the same time point in the stockpile, and construct a set of concentration and humidity numerical pairs. The concentration and humidity values ​​were fitted with a scatter distribution trend to extract the concentration and humidity correlation trend line that shows the decrease of dust concentration response signal as the atmospheric humidity response signal of the stockpile increases. The concentration-humidity correlation trend line is converted into a humidity-concentration inhibition coefficient curve, which represents the dimensionless dust settling probability factor corresponding to different atmospheric humidity values. The diffusion driving intensity coefficient is mapped by looking up the atmospheric humidity response signal at the corresponding spatial grid node to obtain the corresponding dimensionless sedimentation probability factor. The diffusion driving intensity coefficient is then multiplied by the sedimentation probability factor to obtain the effective diffusion driving intensity coefficient after humidity attenuation effect correction. Replace the original diffusion driving intensity coefficient with the effective diffusion driving intensity coefficient, and re-execute the potential energy field synthesis step to generate a humidity-corrected dust diffusion potential field. The humidity-corrected dust diffusion state field is input into the spatiotemporal diffusion state inference network, and the diffusion driving force tensor is modulated with a humidity suppression component. The spatiotemporal diffusion state inference network uses the diffusion driving force tensor modulated with the humidity suppression component to recursively deduce the diffusion evolution process of the dust concentration field and update the spatiotemporal prediction sequence of dust concentration.

10. A spatiotemporal prediction and hierarchical early warning system for dust concentration at a dry bulk cargo terminal, characterized in that, The method includes a processor and a computer-readable storage medium storing machine-executable instructions, which, when executed by a computer, implement the method for spatiotemporal prediction and graded early warning of dust concentration at dry bulk cargo terminals as described in any one of claims 1-9.

Images