A method and system for intelligent anti-clogging control of dense phase pneumatic conveying
By constructing a gas-solid fluid dynamic characteristic matrix and a blockage risk prediction model, and combining it with a gradient depressurization-gas replenishment coordinated control logic, real-time blockage risk perception and adaptive anti-blockage control of the dense phase pneumatic conveying system were realized. This solved the problem that blockage could not be predicted in advance in traditional systems, and improved the continuity and safety of the conveying process.
Patent Information
- Application Number
- CN202611618062.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing dense phase pneumatic conveying systems struggle to identify blockage risks in real time when conveying materials with poor flowability over long distances, leading to frequent pipe blockages. Traditional control systems rely on fixed thresholds or empirical rules, making it impossible to achieve adaptive anti-blockage control.
A gas-solid-fluid state feature matrix is constructed using multi-source time-series data. A blockage risk prediction model is then developed by combining graph attention and temporal self-attention. This model outputs the blockage risk probability and classifies risk levels. A gradient depressurization-gas replenishment coordinated control logic is established, generating a blockage prevention control parameter sequence to achieve adaptive blockage prevention control.
It realizes real-time blockage risk perception and adaptive anti-blockage control of dense phase pneumatic conveying system, improves the continuity and safety of conveying, and overcomes the defect of blockage that cannot be predicted in advance in traditional system.
Smart Images

Figure CN123113405A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of dense phase pneumatic conveying control technology, specifically to an intelligent anti-blocking control method and system for dense phase pneumatic conveying. Background Technology
[0002] Pneumatic conveying is a key method for bulk material transfer in fields such as plastics, chemicals, and fly ash. Dense phase conveying relies on pressure differential to transport materials, featuring low conveying speed, low overall energy consumption, less material breakage, and less pipeline wear. It can reduce particle crushing degradation and pipeline erosion problems, and its application is becoming increasingly widespread. However, when conveying materials with poor flowability over long distances, the complex flow pattern inside the pipe makes it prone to blockage, affecting production continuity and safety. Traditional control systems rely on manual experience or simple threshold judgments, and can only provide alarms after blockage occurs, making it difficult to identify early signs of blockage and resulting in a significant lag in response.
[0003] A patent application with publication number CN120397737A discloses an anti-clogging control method and system, comprising: identifying the pressure imbalance location of the conveying pipeline in response to a pressure change rate triggering a blockage risk; determining at least one air replenishment node in the downstream region of the pressure imbalance location; determining the pneumatic dynamic parameters for each air replenishment node to balance the pressure of the conveying pipeline based on the physical properties of the material; and replenishing air to each air replenishment node according to the pneumatic dynamic parameters. By dynamically capturing transient pressure fluctuations in the conveying pipeline and monitoring the pressure change rate, blockage risks can be identified quickly and accurately; by identifying the pressure imbalance location of the conveying pipeline, the specific location of the blockage can be accurately determined, providing a clear target for subsequent unblocking measures and reducing blind operations; determining the air replenishment node in the downstream region of the pressure imbalance location allows for targeted unblocking at the blockage location, avoiding unnecessary impact on unblocked upstream areas and improving unblocking efficiency.
[0004] Although there is an existing anti-blocking control method and system that identifies blockage risks based on pipeline pressure change rate and performs targeted purging and unblocking based on material properties matching gas supply parameters, the gas-solid two-phase flow state during dense phase pneumatic conveying is difficult to perceive in real time and accurately, and the blockage risk and evolution trend cannot be predicted in advance. Existing gas supply control mainly relies on fixed thresholds or empirical rules, which makes it impossible to achieve adaptive anti-blocking regulation in response to changes in conveying state. Summary of the Invention
[0005] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent anti-blocking control method and system for dense-phase pneumatic conveying. This method achieves real-time perception of the conveying flow state through spatiotemporal alignment of multi-source time-series data and construction of a gas-solid flow characteristic matrix. Based on a blockage risk prediction model fused with graph attention and temporal self-attention, it outputs the probability of pipe segment blockage risk in advance and classifies the risk levels of pre-blockage precursors, breaking through the dependence on fixed thresholds. By combining gradient depressurization-gas replenishment coordinated control logic and rolling optimization, an anti-blocking control parameter sequence is generated, and the control response score is calculated to evaluate the actual effect of the current gradient anti-blocking control. This achieves adaptive anti-blocking control oriented towards changes in the conveying state, improving the continuity and safety of dense-phase pneumatic conveying.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent anti-clogging control of dense phase pneumatic conveying includes: Multi-source time-series data is acquired for timestamp synchronization and pipe segment spatial mapping. Material mass flow rate and instantaneous solid-gas ratio are calculated. Pressure gradient, material velocity increase rate and instantaneous solid-gas ratio deviation of each pipe segment are extracted. Material flow rate change curve is fitted in real time with material property parameters to construct gas-solid flow state characteristic matrix. Obtain historical blockage condition annotation dataset and pipeline topology connection relationship, construct pipe segment adjacency topology map, combine gas-solid flow state feature matrix to construct blockage risk prediction model, output blockage risk probability, and classify pipe blockage precursor risk level; Based on the risk level of pipe blockage precursors, a gradient depressurization-gas replenishment coordinated control logic is established and the target control range is determined. Control parameters are calculated through rolling optimization to generate a blockage prevention control parameter sequence. The anti-blocking control parameter sequence is issued, gas replenishment adjustment and control parameter optimization are performed, feedback data is obtained in real time to calculate the control response score, and the actual effect of the current gradient anti-blocking control is evaluated.
[0007] Specifically, the steps for calculating the material mass flow rate and instantaneous solid-to-gas ratio include: Acquire multi-source time-series data of dense phase transport pipelines, perform timestamp synchronization and pipe segment spatial mapping to obtain time-synchronized multi-source time-series data; Based on the pipeline topology and the installation mileage of each sensor, the time-synchronized multi-source time series data are bound to the corresponding pipe segment. The time delay of the material flow passing through the adjacent measuring point is estimated by the material conveying speed. The pressure and flow signals of the downstream pipe segment are compensated by reverse translation to obtain the pipe segment-level standardized time series data sequence after spatiotemporal alignment. The material mass flow rate is calculated based on the material feeding position and feeding time, and the error is corrected by combining the feeding efficiency coefficient to obtain the material mass flow rate sequence. Obtain the measured gas volumetric flow rate and calculate the instantaneous solid-gas ratio by combining it with the material mass flow rate at the corresponding feeding time in the material mass flow rate sequence.
[0008] Specifically, the steps for constructing the gas-solid flow characteristic matrix include: Set a fixed-length sliding time window, extract three core features of each pipe section: pressure gradient, material rate increase rate, and instantaneous solid-gas ratio deviation. Normalize and weight the material property parameters to obtain the property reference weight coefficient. Within the same sliding time window, the three types of core features are subjected to dimensionless normalization and linear weighted fusion according to preset weight coefficients. The initial flowability index is obtained by combining the physical property benchmark weight coefficients. After sliding average, amplitude standardization and smoothing and noise reduction, the material flowability change curve is output. Based on the pipe section-level standardized time-series data sequence, material mass flow rate sequence, instantaneous solid-gas ratio, and three core features, a gas-solid flow state feature matrix is constructed.
[0009] Specifically, the steps for constructing the pipe segment adjacency topology graph include: Obtain a historical congestion condition labeled dataset, extract historical congestion events for precursor back-introduction labeling, and generate a labeled sample set; Obtain the pipeline topology connection relationship, take each pipe segment of the dense phase conveying pipeline as a graph node, establish directed edges according to the pipeline topology connection relationship, the positive edge is determined by the material conveying direction, and the negative edge is determined by the pressure feedback direction, and construct the pipe segment adjacency topology graph.
[0010] Specifically, the steps for constructing a congestion risk prediction model include: Based on the labeled sample set, the pipe segment adjacency topology graph, and the gas-solid flow state feature matrix, a labeled spatiotemporal graph sample set is constructed. Construct a congestion risk prediction model, including an input layer, a graph attention convolutional layer, a temporal self-attention layer, a feature mapping layer, and an output layer; The input layer receives the pipe segment adjacency topology graph, the gas-solid fluid state feature matrix, and the pipe segment node feature vectors of each pipe segment in the current sliding time window. These are organized into a spatiotemporal graph input tensor according to the pipe segment number and the sliding time window sequence number. The graph attention convolutional layer receives the spatiotemporal graph input tensor, extracts spatial features from the feature vectors of pipe segment nodes within each sliding time window, concatenates the feature vectors of the central pipe segment nodes with the feature vectors of the neighboring pipe segments, calculates the attention score, normalizes it using softmax to obtain the attention weights of each neighbor, and then sums the weighted feature vectors of the neighboring pipe segments to obtain the spatial enhancement features.
[0011] Specifically, the steps for constructing a congestion risk prediction model also include: The temporal self-attention layer arranges the spatial enhancement features of multiple consecutive sliding time windows in temporal order, calculates the attention weight between different sliding time windows through a multi-head self-attention mechanism, automatically focuses on the dynamic patterns of blockage precursors such as pressure gradient rise, continuous decrease of flow index, and expansion of instantaneous solid-gas ratio deviation, and outputs temporal enhancement features. The feature mapping layer flattens the time-series enhancement features through multi-layer feature mapping, extracts high-dimensional combination features between the pressure difference gradient, material rate increase rate, instantaneous solid-gas ratio deviation, and flowability index of each pipe section layer by layer, and maps them into an abstract representation related to blockage precursors. The output layer will abstractly represent the probability of blockage risk for each pipe segment at a future preset time step, which will be mapped to the final fully connected layer.
[0012] Specifically, the steps for outputting the probability of congestion risk include: The labeled spatiotemporal graph sample set is divided into training set and validation set according to a preset ratio. The weighted focus loss function is used as the loss function. The congestion risk prediction model is trained in combination with Adam optimizer. The training is completed when the loss function value on the validation set no longer decreases for a preset number of consecutive rounds or reaches the preset maximum number of training rounds. The real-time gas-solid fluid dynamics feature matrix, pipe segment adjacency topology graph, and pipe segment node feature vectors are input into the trained blockage risk prediction model, which outputs the blockage risk probability.
[0013] Specifically, the steps for classifying the risk levels of pipe blockage precursors include: Calculate the risk escalation rate of each pipe segment under a preset time step in the future, screen the maximum blockage risk probability, and classify the risk level of pipe blockage precursors; The system uses a priority-based approach to determine whether the conditions for a high-risk level are met. When the maximum probability of congestion risk is greater than or equal to the preset third risk threshold and the rate of risk increase is greater than or equal to the preset rapid increase threshold, the system is classified as a high-risk level. If the high-risk level conditions are not met, then it is determined whether the early warning level conditions are met. When the maximum congestion risk probability is greater than or equal to the preset second risk threshold, or the risk rise rate is greater than or equal to the preset rapid rise threshold, it is determined to be an early warning level. If the conditions for the warning level are not met, then it is determined whether the conditions for the attention level are met. When the maximum probability of congestion risk is greater than or equal to the preset first risk threshold, or the rate of risk increase is greater than or equal to the preset slow increase threshold, it is determined to be at the attention level. Otherwise, it is classified as a safe level.
[0014] Specifically, the steps for generating the anti-blocking control parameter sequence include: Based on the probability of blockage risk and the rate of risk increase within a future preset time step, combined with the real-time acquisition of differential pressure gradient, material rate increase rate, instantaneous solid-gas ratio deviation, gas replenishment node pressure, material conveying speed, and the current actual opening degree of each gas replenishment valve, actual discharge speed of the silo pump, and status of pressure relief valve and blockage bypass valve, the risk level of pipe blockage precursors is mapped to a gradient control mode, and a list of target control ranges is formed. Based on the list of risk levels of pipe blockage precursors and target control ranges, a gradient depressurization-gas replenishment coordinated control logic is established to determine the priority order, corresponding adjustment direction, and adjustment range of each control variable. Historical adjustment response data is obtained, and combined with the differential pressure gradient, material velocity increase rate and instantaneous solid-gas ratio deviation of each pipe section, the control parameters of the target opening degree of each air replenishment valve, air replenishment flow rate and silo pump discharge speed under the current conveying state are calculated through rolling optimization to generate an anti-blocking control parameter sequence.
[0015] Specifically, the steps for evaluating the actual effectiveness of the current gradient-based blocking control include: The anti-blockage control parameter sequence is broken down into a single-cycle execution instruction set according to the control cycle. The execution timing offset is set according to the pipe segment adjacency topology and sent to the actuator of each pipe segment to perform gas replenishment adjustment and control parameter optimization. The feedback data window is extracted according to the lag time of the flow response. The feedback data is acquired and aligned in real time. The improvement rate of pressure gradient, the improvement rate of material rate, and the improvement rate of instantaneous solid-gas ratio deviation are calculated. The weighted sum is used to obtain the control response score. To evaluate the actual effect of the current gradient anti-blocking control, when the probability of blockage risk is higher than the preset safety limit and the control response score is lower than the preset response benchmark value of the current gradient for two consecutive control cycles, switch to the next control gradient and call the corresponding anti-blocking control parameters to perform enhanced anti-blocking intervention. When two consecutive control cycles meet opposite conditions, the control gradient is gradually lowered until the maintenance mode of the safety level is restored.
[0016] A dense-phase pneumatic conveying intelligent anti-blocking control system includes: a flow state sensing module, a risk prediction module, an anti-blocking control module, and a control execution module; The flow perception module is used to acquire multi-source time-series data for timestamp synchronization and pipe segment spatial mapping, calculate material mass flow rate and instantaneous solid-gas ratio, extract pressure gradient, material velocity boost rate and instantaneous solid-gas ratio deviation of each pipe segment, and combine material property parameters to fit the material flow rate change curve in real time to construct a gas-solid flow characteristic matrix. The risk prediction module is used to obtain historical blockage condition annotation dataset and pipeline topology connection relationship, construct pipe segment adjacency topology map, construct blockage risk prediction model by combining gas-solid flow state feature matrix, output blockage risk probability, and classify pipe blockage precursor risk level. The anti-blocking control module is used to establish a gradient pressure reduction-gas replenishment coordinated control logic based on the risk level of pipe blockage precursors and determine the target control range. It generates an anti-blocking control parameter sequence by calculating control parameters through rolling optimization. The control execution module is used to issue anti-blocking control parameter sequences, perform gas replenishment adjustment and control parameter optimization, obtain feedback data in real time to calculate the control response score, and evaluate the actual effect of the current gradient anti-blocking control.
[0017] The beneficial effects of this invention are: 1. This invention uses multi-source time-series data timestamp synchronization and pipe segment spatial mapping to extract three core features: pressure gradient, material velocity increase rate, and instantaneous solid-gas ratio deviation. It then fits the material flowability change curve to construct a gas-solid flow state feature matrix, enabling real-time perception of the gas-solid two-phase flow state in dense-phase pneumatic conveying. Based on the pipe segment adjacency topology graph, it builds a blockage risk prediction model including a graph attention convolutional layer and a temporal self-attention layer, outputting the blockage risk probability of each pipe segment at a future preset time step, and classifying the risk level of pipe blockage precursors. This overcomes the shortcomings of gas-solid flow state being difficult to perceive and blockage being impossible to predict in advance.
[0018] 2. This invention establishes a gradient-based pressure reduction-gas replenishment coordinated control logic based on the risk level of pipe blockage precursors. It utilizes rolling optimization to generate an anti-blockage control parameter sequence, eliminating reliance on fixed thresholds and manual experience rules. The anti-blockage control parameter sequence is distributed to the actuators of each pipe section to perform gas replenishment adjustment and control parameter optimization. Feedback data is acquired in real time to calculate the control response score, evaluate the actual effect of the current gradient anti-blockage control, and adaptively switch and adjust the control gradient according to changes in the transport flow state. This achieves closed-loop adaptive anti-blockage control for real-time transport conditions, improving the dynamic adaptability of dense phase pneumatic transport anti-blockage. Attached Figure Description
[0019] Figure 1 This is a schematic diagram of an intelligent anti-blocking control method for dense phase pneumatic conveying. Figure 2 This is a flowchart of the congestion risk prediction model constructed in this invention; Figure 3 This is a flowchart illustrating the risk levels of pipe blockage precursors in this invention; Figure 4 This is a flowchart illustrating the generation of the anti-blocking control parameter sequence in this invention; Figure 5 This is a structural diagram of an intelligent anti-blocking control system for dense phase pneumatic conveying. Detailed Implementation
[0020] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.
[0021] Example 1 refer to Figures 1 to 4 As shown in the figure, this embodiment introduces an intelligent anti-clogging control method for dense phase pneumatic conveying, including the following steps: Step 1: Acquire multi-source time-series data, including but not limited to inlet pressure, outlet pressure, differential pressure, gas supply node pressure, material conveying speed, material level, material feeding time, gas flow rate, and conveying time, using pipeline pressure sensors, level gauges, and gas flow meters installed along the dense phase conveying pipeline. The material conveying speed is calculated from the measured airflow velocity of the gas flow meter. Timestamp alignment and pipeline spatial mapping methods are used to synchronize the timestamps of the multi-source time-series data at different sampling frequencies and map them to the pipeline spatial map. The material mass flow rate is calculated based on the material level and material feeding time, and then calculated based on the measured gas volumetric flow rate and material mass flow rate. The instantaneous solid-gas ratio is calculated by measuring flow rate. Three core features—pressure gradient, material velocity increase rate, and instantaneous solid-gas ratio deviation—are extracted from each pipe section through a sliding time window. These features are then combined with material properties such as friction coefficient, angle of repose, particle shape, and particle size to fit the material flowability curve in real time, constructing a gas-solid flow characteristic matrix. For silo pump discharge, a weighing module can be used to calculate the material mass flow rate. This involves acquiring the material weight signal within the silo pump in real time, performing time difference analysis on the weight signal, and calculating the weight change per unit time to obtain the material mass flow rate. Silo discharge can be directly controlled by a rotary valve to manage the material mass flow rate. Step 2: Obtain the historical blockage condition annotation dataset and pipeline topology connection relationship, construct the pipe segment adjacency topology map to represent the upstream and downstream blockage transmission relationship, combine the gas-solid flow state feature matrix to construct the blockage risk prediction model, mine the inherent correlation between blockage precursors and gas-solid two-phase flow parameters through multi-layer feature mapping, output the blockage risk probability of each pipe segment under the future preset time step, and classify the blockage precursor risk level. Step 3: Based on the risk level of pipe blockage precursors, establish a gradient pressure reduction-gas replenishment coordinated control logic. Determine the corresponding target control range according to the risk level of pipe blockage precursors, including: maintaining the current conveying parameters and continuously monitoring at the safety level; fine-tuning the gas replenishment flow rate to stabilize the flow state in the pipe section corresponding to the monitoring level; increasing the gas replenishment flow rate and simultaneously reducing the discharge speed of the silo pump in the pipe section corresponding to the early warning level; implementing graded pressure relief, multi-point gas replenishment and speed reduction linkage in the pipe section corresponding to the high-risk level, and activating the anti-blockage bypass when necessary; based on the pressure differential gradient, material velocity increase rate, instantaneous solid-gas ratio deviation and historical adjustment response data of each pipe section, calculate the control parameters of the target opening degree of each gas replenishment valve, gas replenishment flow rate and silo pump discharge speed under the current conveying state through rolling optimization, and generate an anti-blockage control parameter sequence. Step 4: The anti-blocking control parameter sequence is sent to the actuators of each pipe section to perform gas replenishment adjustment and control parameter optimization. The feedback data after adjustment and optimization is obtained in real time to calculate the control response score and evaluate the actual effect of the current gradient anti-blocking control. If the blockage risk probability is higher than the preset safety limit and the control response score is lower than the preset response benchmark value, the next control gradient is switched and the corresponding control parameters are called to perform enhanced anti-blocking intervention. The feedback data includes pipeline pressure, pipe section pressure difference, material conveying speed and instantaneous solid-gas ratio change data.
[0022] Specifically, the steps for constructing the gas-solid flow regime characteristic matrix in step 1 include: Multi-source time-series data is acquired by installing pipeline pressure sensors, level gauges, and gas flow meters along the dense phase conveying pipeline. This includes, but is not limited to, inlet pressure, outlet pressure, pipeline pressure difference, gas replenishment node pressure, material conveying speed, material level, material feeding time, gas flow rate, and conveying time for each pipeline section. A timestamp alignment and pipe segment spatial mapping method is used to synchronize the timestamps and map the pipe segments of multi-source time-series data with different sampling frequencies. The specific process includes: using the clock of the high-frequency sampled pipeline pressure sensor as the main time reference; interpolating the low-frequency sampled material level, material feeding time, and gas flow data; aligning the material feeding time to a unified time grid according to the event occurrence time; and resampling all multi-source time-series data to the unified time grid to obtain time-synchronized multi-source time-series data. In this embodiment, the high-frequency sampling is 50Hz and the low-frequency sampling is 1Hz. Based on the pipeline topology and the installation mileage of each sensor, a three-dimensional mapping table of sensor number, installation location, and corresponding pipe segment number is established. The time-synchronized multi-source time series data is bound to the corresponding pipe segment. The time delay of the material flow passing through adjacent measuring points is estimated by using the material conveying speed. The pressure and flow signals of the downstream pipe segment are compensated by reverse translation. That is, the sampling time of the pressure and flow signals of the downstream pipe segment is determined, and the time delay is subtracted from the sampling time to obtain the compensated timestamp. The compensated timestamp is mapped to the time axis of the upstream measuring point, and then a unified time grid is resampled to obtain a pipe segment-level standardized time series data sequence after spatiotemporal alignment, thus eliminating the spatial phase lag caused by material transmission along the pipeline. The material mass flow rate is calculated based on the material level and feeding time. Specifically, within a unit time interval, the product of the material level drop height and the effective cross-sectional area of the feeding tank during the corresponding feeding time is calculated to obtain the volume change. This volume change is then multiplied by the material bulk density to obtain the original material mass flow rate. An error correction is then applied to the original material mass flow rate using a feeding efficiency coefficient. The specific error correction process is as follows: the actual material mass flow rate equals the original material mass flow rate multiplied by the feeding efficiency coefficient. The feeding efficiency coefficient is calibrated by the ratio of the actual weighed feeding mass to the theoretically calculated feeding mass under the same working conditions from historical weighing data, thus calculating a continuous and stable material mass flow rate sequence. The effective cross-sectional area of the feeding tank is calibrated based on the tank's geometric dimensions and the level gauge installation location. The material bulk density is calibrated based on the material type, particle size distribution, and historical test data. Historical weighing data is obtained from the system's historical conveying conditions and on-site weighing calibration records. The measured gas volumetric flow rate is obtained by a gas flow meter, and the instantaneous solid-gas ratio is calculated by combining the material mass flow rate at the corresponding feeding time in the material mass flow rate sequence. The calculation process is as follows: divide the material mass flow rate at the corresponding feeding time by the gas mass flow rate to obtain the instantaneous solid-gas ratio; wherein, the gas mass flow rate is the product of the measured gas volumetric flow rate and the gas density, and the gas density is determined by temperature and pressure compensation conversion based on the ideal gas state equation according to the real-time operating pressure, operating temperature and the standard reference density of the corresponding pipe section. A fixed-length sliding time window is set, the fixed length of which is determined by the average residence time of the material in the pipe section and the pressure signal fluctuation period, and is set to 5 uniform sampling steps, covering the average flow time of the material in a single pipe section and the typical flow pattern fluctuation period. Three core features of each pipe section are extracted through the sliding time window: pressure gradient, material velocity increase rate, and instantaneous solid-gas ratio deviation. These are then combined with material properties such as friction coefficient, angle of repose, particle shape, and particle size to fit the material flowability curve in real time. Specifically, the pressure gradient of each pipe section is obtained by dividing the difference between the inlet pressure and the outlet pressure by the pipe section length. The velocity difference between the average material velocity in the current sliding time window and the average material velocity in the previous sliding time window is calculated, and the velocity increase rate is obtained by dividing the velocity difference by the average material velocity in the previous sliding time window. The instantaneous solid-gas ratio deviation is equal to the difference between the measured instantaneous solid-gas ratio and the set solid-gas ratio divided by the set solid-gas ratio, which is determined based on the historical stable operating average. The fitting process for the material flowability change curve includes: calculating the reference weight coefficients reflecting the inherent flowability of the material based on its physical property parameters. The calculation process includes: normalizing the material physical property parameters such as friction coefficient, angle of repose, particle shape, and particle size to obtain normalized values for friction coefficient, angle of repose, particle shape, and particle size; then weighting and summing these values according to experimentally calibrated weight coefficients to obtain the reference weight coefficients. The friction coefficient is the sliding friction coefficient between the particles and the inner wall of the conveying pipeline, obtained through friction tests between the material sample and the pipeline material sample. The angle of repose is the angle between the inclined surface of the cone formed by the natural accumulation of the material and the horizontal base, measured on-site using an angle of repose testing device. The particle shape is characterized by sphericity. The morphological parameters are defined as follows: the closer the sphericity is to 1, the closer the particle shape is to a standard sphere; the particle size is the volume average particle size, obtained by detecting the material sample on site using a laser particle size analyzer; the experimentally calibrated weighting coefficients include the friction coefficient weight, the angle of repose weight, the particle shape weight, and the particle size weight, and the sum of the weights of the four is 1; since wall friction directly determines the conveying resistance, the friction coefficient has the largest weight, with a value of 0.35; since the angle of repose reflects the internal friction and interlocking between particles, and its influence is second only to wall friction, the angle of repose has the second largest weight, with a value of 0.25; since particle sphericity affects the interaction between particles and airflow, and particle size affects the particle settling and suspension characteristics, and the two have similar influence on the pneumatic conveying process, the particle shape weight and the particle size weight are equal, each with a value of 0.20; Within the same sliding time window, the extracted three core features—pressure gradient, material velocity increase rate, and instantaneous solid-gas ratio deviation—are subjected to dimensionless normalization to obtain normalized values for pressure gradient, material velocity increase rate, and instantaneous solid-gas ratio deviation. Here, pressure gradient represents flow resistance, material velocity increase rate represents pneumatic efficiency, and instantaneous solid-gas ratio deviation represents flow stability. The normalized values for pressure gradient, material velocity increase rate, and instantaneous solid-gas ratio deviation are linearly weighted and fused according to preset weighting coefficients, and then multiplied by a physical property benchmark weighting coefficient for correction, yielding an initial flowability index. This index is then mapped to the 0-1 range through moving average and amplitude standardization, forming an initial flowability index sequence. A filtering algorithm is then used to refine the initial flowability index. The exponential sequence undergoes smoothing and noise reduction processing to suppress jitter caused by sensor noise and operating condition disturbances, outputting a continuous and stable material flowability change curve. The preset weighting coefficients include those for the normalized pressure gradient, the normalized material velocity increase rate, and the normalized instantaneous solid-gas ratio deviation. The weighting coefficient for the normalized pressure gradient is determined based on the core characterizing effect of flow resistance on material flowability in dense-phase conveying, and is set to 0.6. The weighting coefficient for the normalized material velocity increase rate is determined based on the material's sensitivity to pneumatic drive, and is set to 0.15. Since the sum of the weighting coefficients for the normalized pressure gradient, the normalized material velocity increase rate, and the normalized instantaneous solid-gas ratio deviation is 1, the weighting coefficient for the normalized instantaneous solid-gas ratio deviation is set to 0.25. Based on the pipe segment-level standardized time-series data sequence, material mass flow rate sequence, instantaneous solid-gas ratio, and three core features, a gas-solid flow state feature matrix is constructed. The gas-solid flow state feature matrix is arranged with pipe segment number as the row and sliding time window number as the column. The matrix element is the flow index value on the material flow change curve obtained by weighting the initial flow index of the corresponding pipe segment within the corresponding sliding time window with pressure gradient, material velocity increase rate, and instantaneous solid-gas ratio deviation, and then performing moving average, amplitude standardization, and filtering smoothing.
[0023] Specifically, the steps for classifying the risk level of pipe blockage precursors in step 2 include: Historical blockage condition annotation datasets are obtained from historical blockage records, on-site operation logs, and blockage clearing records. Historical blockage events are extracted and labeled with precursors to generate a labeled sample set. Each sample contains a pipe section number, sliding time window number, blockage label, and precursor urgency. The blockage label is 0 for normal window, 1 for precursor window, and 2 for blockage window. The precursor urgency is the number of remaining sliding time windows between the precursor window and the blockage window. The remainder is calculated by adding the number of remaining sliding time windows to the value 1 and taking the reciprocal. The value ranges from 0 to 1. The higher the precursor urgency value, the closer to blockage. Based on the pipeline design drawings, sensor installation records, and material conveying direction markings, the pipeline topology connection relationship is obtained. Each pipe segment of the dense phase conveying pipeline is used as a graph node. Directed edges are established according to the pipeline topology connection relationship. The positive edge is determined by the material conveying direction, i.e., from the upstream pipe segment to the downstream pipe segment. Since downstream blockage will cause upstream pressure buildup, the reverse edge is determined by the pressure feedback direction, i.e., from the downstream pipe segment to the upstream pipe segment. Thus, a pipe segment adjacency topology graph is constructed to represent the upstream and downstream blockage transmission relationship. The nodes in the pipe segment adjacency topology graph are each pipe segment, and the directed edges represent the possible transmission path of blockage along the material flow direction or the pressure feedback direction. Based on the labeled sample set, the pipe segment adjacency topology graph, and the gas-solid flow characteristic matrix, a labeled spatiotemporal graph sample set is constructed. For each sliding time window, a spatiotemporal graph sample is formed by taking a snapshot of that sliding time window and several consecutive historical sliding time windows. Each snapshot is based on a pipe segment as a node, and the node features are the pipe segment node feature vectors within that sliding time window, including the initial flowability index, pressure gradient normalization value, material rate increase rate normalization value, instantaneous solid-gas ratio deviation normalization value, material mass flow rate normalization value, instantaneous solid-gas ratio normalization value, and pressure fluctuation standard deviation. Multiple consecutive snapshots are stacked in chronological order and combined with the pipe segment adjacency topology graph to form a complete spatiotemporal graph sample. The labels of the spatiotemporal graph samples are the blockage labels and precursor urgency corresponding to the sliding time window in the labeled sample set, thus obtaining the labeled spatiotemporal graph sample set, where each sample contains the node feature tensor of consecutive historical sliding time windows, the pipe segment adjacency topology graph, the blockage label, and the precursor urgency. Based on the pipe segment adjacency topology graph and labeled spatiotemporal graph sample set, a blockage risk prediction model is constructed by combining the gas-solid fluid state feature matrix, including an input layer, a graph attention convolutional layer, a temporal self-attention layer, a feature mapping layer and an output layer. The input layer is used to receive the pipe segment adjacency topology graph, the gas-solid fluid state feature matrix, and the pipe segment node feature vectors of each pipe segment in the current sliding time window. These are organized into a spatiotemporal graph input tensor according to the pipe segment number and the sliding time window number. The graph attention convolutional layer receives the spatiotemporal graph input tensor and extracts spatial features from the feature vectors of pipe segments within each sliding time window. It concatenates the feature vectors of the central pipe segment with those of neighboring pipe segments to calculate the attention score. The calculation process is as follows: concatenating the feature vectors of the central pipe segment with those of neighboring pipe segments, performing a linear transformation using a learnable weight matrix, and then applying the LeakyReLU activation function to obtain the attention score. Softmax normalization is then applied to the attention scores of all neighbors of the central pipe segment to obtain the attention weights of each neighbor. The attention weights of each neighbor are then weighted and summed directly onto the feature vectors of the neighboring pipe segments to obtain spatial enhancement features, thereby automatically learning the spatial pattern of blockage transmission between upstream and downstream pipe segments. The learnable weight matrix is a parameter matrix to be optimized within the blockage risk prediction model, used to map the concatenated feature vectors of the central pipe segment and those of neighboring pipe segments to the attention score space. The temporal self-attention layer is used to arrange the spatial enhancement features of multiple consecutive sliding time windows in temporal order. It calculates the attention weight between different sliding time windows through a multi-head self-attention mechanism, automatically focuses on the dynamic patterns of blockage precursors such as pressure gradient rise, continuous decrease of flow index, and expansion of instantaneous solid-gas ratio deviation, and outputs temporal enhancement features. The feature mapping layer is used to mine the intrinsic correlation between blockage precursors and gas-solid two-phase flow parameters through multi-layer feature mapping. After flattening the time-series enhanced features, it sequentially passes them through multiple fully connected layers and nonlinear activation functions to extract high-dimensional combination features between parameters such as pressure gradient, material rate increase rate, instantaneous solid-gas ratio deviation, and flow index of each pipe section. These high-dimensional combination features are then mapped into abstract representations directly related to blockage precursors. The output layer is used to map the abstract representation output by the feature mapping layer to the blockage risk probability of each pipe segment at a future preset time step through the final fully connected layer. The blockage risk probability is mapped to the 0~1 interval by the activation function. In this embodiment, the future preset time step is determined based on the average number of sliding time windows from the first stable appearance of the precursor features to the actual occurrence of the blockage in historical blockage events, and is set to 5 time steps. The labeled spatiotemporal graph sample set is divided into training and validation sets according to a preset ratio. A weighted focus loss function is used as the loss function, and the Adam optimizer is used to train the congestion risk prediction model. Training is complete when the loss function value on the validation set no longer decreases for a preset number of consecutive training rounds or reaches the preset maximum training rounds. The weighted focus loss function is a loss function that determines the category weights based on the number of samples in the normal window, precursor window, and congestion window categories and the urgency of the precursors, weights the cross-entropy loss, and reduces the loss weight of easily classified samples and increases the loss weight of difficult-to-classify samples through focus parameters. The preset ratio is 7:3, determined based on the total number of samples in the labeled spatiotemporal graph sample set, the distribution of positive and negative samples, and the sufficiency requirements of the validation set for evaluating the model's generalization ability. The preset number of training rounds is determined based on the number of consecutive rounds the validation set loss can tolerate without decreasing, and is set to 10 rounds. The preset maximum training rounds are determined based on the empirical upper limit required for model convergence, and are set to 200 rounds. The real-time gas-solid fluid dynamics feature matrix, pipe segment adjacency topology graph, and pipe segment node feature vectors are input to the trained blockage risk prediction model. The input layer organizes the gas-solid fluid dynamics feature matrix, pipe segment adjacency topology graph, and pipe segment node feature vectors into a spatiotemporal graph input tensor according to the pipe segment number and sliding time window sequence. The graph attention convolutional layer extracts spatial enhancement features, and the temporal attention layer extracts temporal enhancement features. The feature mapping layer maps the temporal enhancement features into an abstract representation. The output layer then maps this abstract representation into a blockage risk probability in the range of 0 to 1 through the final fully connected layer and activation function. Through the above layers, the blockage risk probability of each pipe segment at the preset time step in the future is output. Subtract the probability of blockage risk in the first time step from the probability of blockage risk in the fifth time step, and then divide by the interval of four time steps to calculate the risk increase rate of each pipe segment under the future preset time step. Screen the maximum probability of blockage risk under the future preset time step and classify the pipe blockage precursor risk level, including safety level, attention level, warning level and high risk level. The risk level classification for pipe blockage precursors is determined by prioritizing from highest to lowest: First, it is assessed whether the high-risk level conditions are met: if the maximum probability of blockage risk within the next 5 time steps is greater than or equal to the third risk threshold, and the rate of risk increase is greater than or equal to the rapid increase threshold, it is classified as high-risk. If the high-risk level conditions are not met, then it is assessed whether the early warning level conditions are met: if the maximum probability of blockage risk within the next 5 time steps is greater than or equal to the second risk threshold, or the rate of risk increase is greater than or equal to the rapid increase threshold, it is classified as early warning. If the early warning level conditions are not met, then it is assessed whether the attention level conditions are met: if the maximum probability of blockage risk within the next 5 time steps is greater than or equal to the first risk threshold, or the rate of risk increase is greater than or equal to the slow increase threshold, it is classified as attention level. If none of the above high-risk, early warning, and attention level conditions are met, it is classified as safe. The first risk threshold is determined based on the boundary between the risk probability distributions of historical normal operating conditions and congestion conditions, and is set to 0.3. The second risk threshold is determined based on the boundary between the risk probability distributions of historical pre-congestion conditions and obvious abnormal conditions, and is set to 0.6. The third risk threshold is determined based on the risk probability distribution of historical high-probability congestion conditions and the actual congestion occurrence rate, and is set to 0.85. The slow rise threshold is determined based on the maximum natural fluctuation rate of the congestion risk probability under historical stable operating conditions, and is set to 0.02 / time step. The rapid rise threshold is determined based on the minimum risk growth rate during the rapid growth phase of the risk probability before a historical congestion event, and is set to 0.05 / time step.
[0024] For example, suppose the labeled spatiotemporal graph sample set contains 10,000 samples. The training set and validation set are divided in a 7:3 ratio, i.e., 7,000 samples in the training set and 3,000 samples in the validation set. The weighted focus loss function is used as the loss function. The congestion risk prediction model is trained using the Adam optimizer with an initial learning rate of 0.001, a first moment decay coefficient of 0.9, and a second moment decay coefficient of 0.999. The training is complete when the loss function value on the validation set no longer decreases for 10 consecutive rounds or reaches 200 rounds. The real-time gas-solid fluid dynamics feature matrix, pipe segment adjacency topology graph, and pipe segment node feature vectors are input to the trained blockage risk prediction model. The input layer organizes the gas-solid fluid dynamics feature matrix, pipe segment adjacency topology graph, and pipe segment node feature vectors into a spatiotemporal graph input tensor according to the pipe segment number and sliding time window sequence. The graph attention convolutional layer extracts spatial enhancement features, and the temporal attention layer extracts temporal enhancement features. The feature mapping layer maps the temporal enhancement features into an abstract representation. The output layer then maps this abstract representation into a blockage risk probability in the range of 0 to 1 through the final fully connected layer and activation function. Through the above layers, the blockage risk probability of each pipe segment in the next 5 time steps is output. For example, if the blockage risk probabilities of a certain pipe segment in the first to fifth time steps are 0.42, 0.51, 0.58, 0.67, and 0.74 respectively, then subtracting the blockage risk probability of the first time step from the blockage risk probability of the fifth time step (0.74), and then dividing by the interval of four time steps, we obtain the risk escalation rate of the pipe segment as 0.08 / time step. Simultaneously, the maximum blockage risk probability within the next five time steps is identified as 0.74. Since the maximum blockage risk probability of 0.74 is greater than the second risk threshold of 0.6, and the risk escalation rate of 0.08 / time step is greater than the rapid escalation threshold of 0.05 / time step, the pipe segment meets the warning level criteria and is determined to be at the warning level for the precursory blockage risk. For example, if the probability of blockage risk for another pipe segment in the next 5 time steps is 0.68, 0.73, 0.79, 0.84 and 0.88 respectively, then the maximum probability of blockage risk is 0.88. The difference between 0.88 and 0.68 is 0.20. Dividing the difference by the interval of 4 time steps, we get the risk escalation rate as 0.05 / time step. Since the maximum probability of blockage risk 0.88 is greater than the third risk threshold of 0.85, the risk escalation rate reaches the rapid escalation threshold of 0.05 / time step, so the pipe segment is determined to be of high risk level.
[0025] Specifically, the steps for generating the anti-blocking control parameter sequence in step 3 include: Based on the probability of blockage risk and the rate of risk escalation within a future preset time step, combined with the real-time acquisition of differential pressure gradient, material rate increase rate, instantaneous solid-gas ratio deviation, gas replenishment node pressure, material conveying speed, and the current actual opening of each gas replenishment valve, actual discharge speed of the silo pump, and status of pressure relief valve and blockage bypass valve, the risk level of pipe blockage precursors is mapped to a gradient control mode. Among them, the safety level corresponds to the maintenance mode, the attention level corresponds to the fine-tuning mode, the early warning level corresponds to the collaborative adjustment mode, and the high-risk level corresponds to the emergency intervention mode. The control variables and target control ranges that can be adjusted differ under each control mode. The maintenance mode does not adjust any control variables, only maintaining the current delivery parameters and continuously monitoring them. The fine-tuning mode indicates that the supplementary air flow rate of the corresponding pipe section is fine-tuned to stabilize the flow state. Only the opening of the upstream supplementary air valve of the corresponding pipe section is allowed to be adjusted, ensuring that the adjustment range of the supplementary air flow rate does not exceed 10% of the current supplementary air flow rate. 10% is the maximum allowable adjustment ratio of the supplementary air flow rate under the fine-tuning mode. The coordinated control mode indicates that the supplementary air flow rate of the corresponding pipe section is increased while the discharge speed of the silo pump is simultaneously reduced. This means that the opening of the supplementary air valve of the corresponding pipe section can be adjusted to change the supplementary air flow rate, but the increase in the supplementary air flow rate does not exceed 3% of the current supplementary air flow rate. 0%, and the reduction in the discharge speed of the silo pump does not exceed 20% of the current speed. 30% is the maximum allowable increase ratio of the supplementary air flow rate under the coordinated adjustment mode, and 20% is the maximum allowable reduction ratio of the discharge speed of the silo pump under the coordinated adjustment mode. Among them, the opening degree of the supplementary air valve includes the opening degree of the upstream supplementary air valve and the opening degree of the downstream supplementary air valve. The emergency intervention mode indicates that the corresponding pipe section performs graded pressure relief, multi-point air supply and speed reduction linkage, and starts the blockage bypass when necessary. At the same time, for pipe sections with upstream and downstream blockage transmission relationship, when the downstream adjacent pipe section is at the warning level or high risk level, the control mode of the upstream pipe section will be automatically upgraded by one level to form a target control range list. Based on the list of pre-blockage risk levels and target control ranges, a gradient-based pressure reduction and gas replenishment coordinated control logic is established. The priority order, corresponding adjustment direction, and adjustment range of each control variable are determined. Specifically, this includes: determining the priority order of control variable adjustments, in the following order: silo pump discharge rate, upstream and downstream gas replenishment valve openings for the corresponding pipe section, pressure relief valve opening, and blockage bypass valve status; determining the adjustment direction based on the deviation direction between the real-time pressure differential gradient, material rate increase rate, and their respective setpoints for each pipe section, and further determining the adjustment direction based on the sign of the instantaneous solid-gas ratio deviation; and establishing multi-point gas replenishment coordinated control rules, prioritizing adjustments. The opening of the upstream air supply valve near the pipe section inlet is adjusted such that when multiple air supply valves are adjusted simultaneously, the adjustment amount of the downstream air supply valve opening is less than that of the upstream air supply valve opening. A rule for coordinated pressure reduction and air supply is established. For pipe sections with a warning level, the air supply valve opening and the discharge speed of the silo pump are adjusted first. After one control cycle, if the pressure differential gradient is still higher than 50% of the set upper limit, the pressure relief valve is opened. For pipe sections with a high risk level, air supply, pressure reduction and speed reduction are executed simultaneously, and the opening of the pressure relief valve is gradually increased according to the level. The set upper limit is determined by adding three times the standard deviation to the average pressure differential gradient of the pipe section under historical stable operating conditions, and is set to 3.0 kPa / m. The specific process for determining the adjustment direction includes: comparing the real-time differential pressure gradient and material velocity increase rate of each pipe segment within the current sliding time window with their respective set values, and determining the adjustment direction based on the sign of the instantaneous solid-gas ratio deviation; when the differential pressure gradient is higher than the corresponding set value, determining the adjustment direction of decreasing the discharge speed of the silo pump and increasing the opening of the air supply valve; when the material velocity increase rate is lower than the corresponding set value, determining the adjustment direction of increasing the opening of the air supply valve; when the instantaneous solid-gas ratio deviation is positive, determining the adjustment direction of increasing the opening of the air supply valve and decreasing the discharge speed of the silo pump, and when the instantaneous solid-gas ratio deviation is negative, determining the opposite adjustment direction; when multiple characteristics deviate simultaneously, determining the priority control variable based on the risk level of pipe blockage precursors, and outputting the adjustment direction of increasing, decreasing, or keeping the control variables unchanged. Historical adjustment response data is obtained from on-site operation logs and historical adjustment test records. Combined with the differential pressure gradient, material rate increase rate and instantaneous solid-gas ratio deviation of each pipe section, the control parameters of the target opening degree of each air replenishment valve, air replenishment flow rate and silo pump discharge speed under the current conveying state are calculated by rolling optimization. The anti-blocking control parameter sequence is generated, including the target opening degree of each air replenishment valve, air replenishment flow rate and silo pump discharge speed, pressure relief valve opening degree and anti-blocking bypass valve status. The specific process of rolling optimization includes: starting from the current control cycle, setting the rolling time domain length to 5 control cycles; optimizing variables including the target opening degree of each air replenishment valve, the discharge speed of the silo pump, and the target opening degree of the pressure relief valve and the target state of the blockage bypass valve in high-risk pipe sections; the optimization objective function is composed of three weighted terms: blockage risk suppression, flow stability maintenance, and control action smoothing. The blockage risk suppression term minimizes the weighted sum of the predicted blockage risk probabilities for each pipe section over the next 5 control cycles, weighted by time decay; the flow stability maintenance term minimizes the weighted sum of the squares of the predicted instantaneous solid-gas ratio deviation and the squares of the material rate increase rate; and the control action smoothing term minimizes the changes in the air replenishment valve opening degree and the silo pump discharge speed between adjacent control cycles. The weights of each term are determined based on the current control mode of each pipe section, taking into account safety, etc. The flow stability term has the highest weight at the lowest level. The weight of the blockage risk suppression term is increased at the attention level and the warning level. At the high-risk level, the blockage risk suppression term dominates, and the weight of the allowable smooth control action term decreases. Under the constraints of the air supply valve opening being 0-100%, the silo pump discharge speed being between the minimum safe speed and the rated maximum speed, the pressure relief valve opening not exceeding the maximum allowable opening, the blockage bypass valve being a switching quantity, the air supply node pressure not exceeding the pipeline design pressure, and the pressure difference gradient of each pipe section not exceeding the maximum allowable pressure difference gradient, the participation of each control variable in the regulation and its regulation range are determined according to the gradient pressure reduction-air supply coordinated control logic. The optimal control sequence for the next 5 control cycles is obtained using a sequential quadratic programming algorithm, and the optimal control sequence is output as the anti-blockage control parameter sequence.
[0026] Specifically, the steps in step 4 for evaluating the actual effectiveness of the current gradient-based anti-blocking control include: The anti-blockage control parameter sequence is broken down into single-cycle execution instruction sets according to the control cycle. These single-cycle instruction sets include instructions for adjusting the air supply valve, fluidizing gas flow valve, silo pump valve, high-risk level, and air supply and pressure relief adjustments. Execution timing offsets are set based on the upstream and downstream blockage transmission relationships in the pipe segment adjacency topology: for pipe segments with direct upstream and downstream connections, the air supply valve adjustment instruction for the upstream segment is issued one control cycle in advance, and the air supply and pressure relief adjustment instructions for the downstream segment are aligned with the flow state activation time of the upstream adjustment to avoid localized pressure buildup or sudden changes in material velocity caused by asynchrony. Priority is allocated according to a rule that prioritizes high-risk levels, followed by warning levels. In the execution channel, high-risk commands directly preempt execution resources, interrupting current low-priority adjustment actions and marking them as "interrupted and incomplete." They are then re-queued for execution after the high-priority actions are completed. The disassembled single-cycle execution command set is sent to the actuators of each pipe section via the industrial control bus, and gas replenishment adjustment and control parameter optimization are performed. After each actuator completes its action, it provides feedback on the actual execution opening, operating speed, and action completion time. The actuators include a gas replenishment valve actuator, a fluidizing gas flow regulating valve for the silo pump, a silo pump pressure control unit, a pressure relief valve actuator, and a bypass solenoid valve for unblocking. The silo pump discharge speed is controlled by adjusting the fluidizing gas flow and monitoring the internal pressure of the silo pump. Starting from the moment the actuator completes its action, a preset flow response lag time is used to extract the feedback data window. The length of the feedback data window is the same as the length of the sliding time window. Within the feedback data window, the adjusted and optimized feedback data is acquired in real time and timestamped. The feedback data includes pipeline pressure, pipeline pressure difference, material conveying speed, and instantaneous solid-gas ratio change data. The flow response lag time is calibrated based on the corresponding pipeline length and the average material conveying speed, and is set to two uniform sampling steps. The average pipeline pressure fluctuation rate of each pipe segment within the feedback data window is extracted as an auxiliary verification indicator. The average differential pressure gradient, average material rate increase rate, and average instantaneous solid-gas ratio deviation of each pipe segment within the feedback data window are also extracted and used as the adjusted measured values of differential pressure gradient, material rate increase rate, and instantaneous solid-gas ratio deviation, respectively. The differential pressure gradient, material rate increase rate, and instantaneous solid-gas ratio deviation within the previous feedback data window are used as the baseline values for differential pressure gradient, material rate increase rate, and instantaneous solid-gas ratio deviation, respectively. Three improvement rates are calculated: differential pressure gradient improvement rate, material rate increase rate improvement rate, and instantaneous solid-gas ratio deviation improvement rate. If the absolute value of any baseline value is less than... If so, the corresponding improvement rate is set to 0 to avoid division by zero anomalies; The calculation process for the three improvement rates includes: calculating the difference between the measured pressure gradient value and the reference value of the pressure gradient, and dividing the difference by the absolute value of the reference value to obtain the pressure gradient improvement rate; calculating the difference between the measured material rate increase rate and the reference value of the material rate increase rate, and dividing the difference by the absolute value of the reference value to obtain the material rate increase rate improvement rate; calculating the difference between the absolute value of the instantaneous solid-gas ratio deviation reference value and the absolute value of the instantaneous solid-gas ratio deviation, and dividing the difference by the absolute value of the instantaneous solid-gas ratio deviation reference value to obtain the instantaneous solid-gas ratio deviation improvement rate; and then calculating the improvement rates of the pressure gradient, material rate increase rate, and instantaneous solid-gas ratio. The deviation improvement rate is truncated to the 0~1 interval, and a weighted sum is obtained for the 0~1 interval according to the pressure difference gradient weight, material rate improvement rate weight, and instantaneous solid-gas ratio deviation weight. The average pipeline pressure fluctuation rate is not included in the weighted calculation of the control response score, but is only used to assist in verifying the probability of blockage risk. If the pipeline pressure fluctuation amplitude exceeds 20% of the corresponding pipeline section pressure safety threshold, a pressure anomaly flag is triggered, and the current blockage risk probability is increased to no less than the preset safety limit. The pipeline pressure fluctuation amplitude is the difference between the maximum and minimum pipeline pressure within a single feedback data window. Specifically, the differential pressure gradient weight is determined based on the current blockage risk level of the pipeline segment: 0.35 for the monitoring level, 0.45 for the early warning level, and 0.50 for the high-risk level. The material rate increase rate weight is determined based on the material rate sensitivity under the same blockage risk level: 0.30 for the monitoring level, 0.25 for the early warning level, and 0.20 for the high-risk level. The instantaneous solid-gas ratio deviation weight is determined based on the instantaneous solid-gas ratio fluctuation sensitivity under the same blockage risk level: 0.35 for the monitoring level, 0.30 for the early warning level, and 0.30 for the high-risk level. The pressure safety threshold is determined based on the upper limit of the pipeline's design working pressure for the corresponding pipeline segment and the upper limit of fluctuation of the average pipeline pressure plus three standard deviations under historical stable operating conditions, with a value of 0.65 MPa. Based on the control response score, the actual effect of the current gradient-based anti-blocking control is evaluated. The preset safety limit, consistent with the first risk threshold, is set to 0.3 according to the threshold for classifying blockage risk safety levels. Different control gradients correspond to different preset response benchmark values: the preset response benchmark value for the "attention level" is 0.3, based on the lowest response score of historical effective cases of fine-tuning control under the "attention level"; the preset response benchmark value for the "warning level" is 0.4, based on the lowest response score of historical effective cases of coordinated regulation under the "warning level"; and the preset response benchmark value for the "high-risk level" is 0.5, based on the lowest response score of historical effective cases of emergency intervention under the "high-risk level". A two-cycle consecutive control determination mechanism is adopted to avoid operational fluctuations leading to... Frequent switching of control modes: When two consecutive control cycles simultaneously meet the conditions that the probability of congestion risk is higher than the preset safety limit and the control response score is lower than the preset response benchmark value of the current gradient, the current gradient control is deemed to be ineffective, triggering the gradient upgrade logic: switching from the current control mode to the control mode of the next control gradient, and calling the anti-congestion control parameters corresponding to the next control gradient to perform enhanced anti-congestion intervention, re-triggering rolling optimization to generate a new anti-congestion control parameter sequence and issuing it for execution; when two consecutive control cycles meet the opposite conditions, that is, the probability of congestion risk is continuously lower than the preset safety limit and the control response score is higher than the preset response benchmark value, the current gradient control is deemed to be effective, and the control gradient is gradually lowered until the maintenance mode of the safety level is restored.
[0027] Example 2 Please see Figure 5 Another embodiment of the present invention provides: a dense phase pneumatic conveying intelligent anti-blocking control system, comprising: a flow state sensing module, a risk prediction module, an anti-blocking control module, and a control execution module; The flow perception module is used to acquire multi-source time-series data through pipeline pressure sensors, level gauges, and gas flow meters. It performs timestamp synchronization and pipe segment spatial mapping on multi-source time-series data with different sampling frequencies, calculates material mass flow rate and instantaneous solid-gas ratio, and extracts three core features of each pipe segment through a sliding time window: pressure gradient, material velocity increase rate, and instantaneous solid-gas ratio deviation. It combines material physical property parameters to fit the material flowability change curve in real time and constructs a gas-solid flow characteristic matrix. The risk prediction module is used to obtain historical blockage condition annotation datasets and pipeline topology connection relationships, construct pipe segment adjacency topology maps, combine gas-solid flow characteristic matrix to construct blockage risk prediction model, explore the inherent correlation between blockage precursors and gas-solid two-phase flow parameters, output the blockage risk probability of each pipe segment at a future preset time step, and classify the risk level of pipe blockage precursors. The anti-blocking control module is used to establish a gradient pressure reduction-gas replenishment coordinated control logic based on the risk level of pipe blockage precursors. It determines the corresponding target control range according to the risk level of pipe blockage precursors. Based on the pressure differential gradient, material rate increase rate, instantaneous solid-gas ratio deviation and historical adjustment response data of each pipe section, it calculates the control parameters of the target opening degree of each gas replenishment valve, gas replenishment flow rate and silo pump discharge speed through rolling optimization, and generates an anti-blocking control parameter sequence. The control execution module is used to send the anti-blocking control parameter sequence to the actuators of each pipe section, perform gas replenishment adjustment and control parameter optimization, obtain feedback data in real time to calculate the control response score, evaluate the actual effect of the current gradient anti-blocking control, and if the blockage risk probability is higher than the preset safety limit and the control response score is lower than the preset response benchmark value, it will automatically switch to the next control gradient and perform enhanced anti-blocking intervention.
[0028] Working principle and effects: Through the flow perception module, multi-source time-series data is acquired using pipeline pressure sensors, level gauges, and gas flow meters. This enables timestamp synchronization of data from different sampling frequencies and spatial mapping of pipe sections. Material mass flow rate and instantaneous solid-gas ratio are calculated. Three core features—pressure gradient, material velocity increase rate, and instantaneous solid-gas ratio deviation—are extracted via a sliding time window. Combined with material property parameters, the material flowability curve is fitted in real time to construct a gas-solid flow characteristic matrix. This achieves real-time, refined perception of the gas-solid two-phase flow state in dense-phase transport, overcoming the difficulty in accurately characterizing gas-solid two-phase flow and solving the problem that traditional monitoring methods cannot fully capture the dynamic evolution of the flow state. This provides a unified data foundation for subsequent pipe blockage risk prediction and adaptive anti-blockage control decisions. Based on the gas-solid flow characteristic matrix and pipeline topology connection relationship, a pipe segment adjacency topology map and blockage risk prediction model are constructed. The inherent correlation between blockage precursors and gas-solid two-phase flow parameters is explored, and the blockage risk probability of each pipe segment at a preset time step is output, and the blockage precursor risk level is classified. This enables advanced quantitative prediction and graded early warning of blockage precursor risks and their evolution trends, solving the problem that relying solely on on-site operating conditions cannot identify the blockage development trend in advance, and moving away from the passive handling mode that relies solely on on-site alarms. This provides a reliable risk basis for differentiated blockage prevention and control. Based on the risk level of pipe blockage precursors, a gradient pressure reduction-gas replenishment coordinated control logic is matched. The pressure differential gradient, material velocity increase rate, instantaneous solid-gas ratio deviation, and historical adjustment response data of each pipe section are used as optimization basis. The control parameters of the target opening degree of each gas replenishment valve, gas replenishment flow rate, and silo pump discharge speed are solved by rolling optimization to generate an anti-blockage control parameter sequence. The traditional fixed threshold and empirical rule control mode is abandoned, and the anti-blockage control strategy is intelligently matched and generated on demand according to the dynamic changes of the conveying flow. This solves the problem of poor adaptability of fixed control parameters to variable working conditions and improves the working condition adaptability of pipe section anti-blockage measures. The anti-blocking control parameter sequence is sent to the air supply valves and silo pump discharge control units of each pipe section to perform air supply adjustment and control parameter optimization; feedback data is acquired in real time to calculate the control response score and evaluate the actual effect of the current gradient anti-blocking control; when the blockage risk probability is higher than the safety limit and the control response score is lower than the response benchmark value, the system automatically switches to the next control gradient to perform enhanced anti-blocking intervention; a closed-loop adaptive control mechanism of command issuance, coordinated execution, effect feedback and gradient upgrade is formed to solve the problem of control strategy rigidity and execution deviation causing control failure when operating conditions change, and ensure the robustness and dynamic adaptability of multi-equipment linkage control.
[0029] Overall, an intelligent anti-blocking control system for dense-phase pneumatic conveying was constructed through a progressive architecture that integrates flow pattern perception, advance prediction of blockage risk, dynamic generation of gradient control strategies, and closed-loop adaptive execution. This system achieves a closed-loop process encompassing real-time perception of the conveying flow pattern, advance warning of the evolution trend of blockage risk, dynamic generation of anti-blockage strategies, and multi-node collaborative control. It addresses the problems in existing dense-phase pneumatic conveying systems, such as inaccurate perception of gas-solid two-phase flow, difficulty in predicting blockage risk in advance, and the inability to achieve adaptive anti-blockage control based on fixed thresholds and empirical rules.
[0030] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.
Claims
1. A method for intelligent anti-clogging control of dense-phase pneumatic conveying, characterized in that, include: Multi-source time-series data is acquired for timestamp synchronization and pipe segment spatial mapping. Material mass flow rate and instantaneous solid-gas ratio are calculated. Pressure gradient, material velocity increase rate and instantaneous solid-gas ratio deviation of each pipe segment are extracted. Material flow rate change curve is fitted in real time with material property parameters to construct gas-solid flow state characteristic matrix. Obtain historical blockage condition annotation dataset and pipeline topology connection relationship, construct pipe segment adjacency topology map, combine gas-solid flow state feature matrix to construct blockage risk prediction model, output blockage risk probability, and classify pipe blockage precursor risk level; Based on the risk level of pipe blockage precursors, a gradient depressurization-gas replenishment coordinated control logic is established and the target control range is determined. Control parameters are calculated through rolling optimization to generate a blockage prevention control parameter sequence. The anti-blocking control parameter sequence is issued, gas replenishment adjustment and control parameter optimization are performed, feedback data is obtained in real time to calculate the control response score, and the actual effect of the current gradient anti-blocking control is evaluated.
2. The intelligent anti-blocking control method for dense-phase pneumatic conveying according to claim 1, characterized in that, The specific steps for calculating the material mass flow rate and instantaneous solid-gas ratio include: Acquire multi-source time-series data of dense phase transport pipelines, perform timestamp synchronization and pipe segment spatial mapping to obtain time-synchronized multi-source time-series data; Based on the pipeline topology and the installation mileage of each sensor, the time-synchronized multi-source time series data are bound to the corresponding pipe segment. The time delay of the material flow passing through the adjacent measuring point is estimated by the material conveying speed. The pressure and flow signals of the downstream pipe segment are compensated by reverse translation to obtain the pipe segment-level standardized time series data sequence after spatiotemporal alignment. The material mass flow rate is calculated based on the material feeding position and feeding time, and the error is corrected by combining the feeding efficiency coefficient to obtain the material mass flow rate sequence. Obtain the measured gas volumetric flow rate and calculate the instantaneous solid-gas ratio by combining it with the material mass flow rate at the corresponding feeding time in the material mass flow rate sequence.
3. The intelligent anti-blocking control method for dense-phase pneumatic conveying according to claim 2, characterized in that, The specific steps for constructing the gas-solid flow characteristic matrix include: Set a fixed-length sliding time window, extract three core features of each pipe section: pressure gradient, material rate increase rate, and instantaneous solid-gas ratio deviation. Normalize and weight the material property parameters to obtain the property reference weight coefficient. Within the same sliding time window, the three types of core features are subjected to dimensionless normalization and linear weighted fusion according to preset weight coefficients. The initial flowability index is obtained by combining the physical property benchmark weight coefficients. After sliding average, amplitude standardization and smoothing and noise reduction, the material flowability change curve is output. Based on the pipe section-level standardized time-series data sequence, material mass flow rate sequence, instantaneous solid-gas ratio, and three core features, a gas-solid flow state feature matrix is constructed.
4. The intelligent anti-blocking control method for dense-phase pneumatic conveying according to claim 3, characterized in that, The specific steps for constructing a pipe segment adjacency topology graph include: Obtain a historical congestion condition labeled dataset, extract historical congestion events for precursor back-introduction labeling, and generate a labeled sample set; Obtain the pipeline topology connection relationship, take each pipe segment of the dense phase conveying pipeline as a graph node, establish directed edges according to the pipeline topology connection relationship, the positive edge is determined by the material conveying direction, and the negative edge is determined by the pressure feedback direction, and construct the pipe segment adjacency topology graph.
5. The intelligent anti-blocking control method for dense-phase pneumatic conveying according to claim 4, characterized in that, The specific steps for constructing a congestion risk prediction model include: Based on the labeled sample set, the pipe segment adjacency topology graph, and the gas-solid flow state feature matrix, a labeled spatiotemporal graph sample set is constructed. Construct a congestion risk prediction model, including an input layer, a graph attention convolutional layer, a temporal self-attention layer, a feature mapping layer, and an output layer; The input layer receives the pipe segment adjacency topology graph, the gas-solid fluid state feature matrix, and the pipe segment node feature vectors of each pipe segment in the current sliding time window. These are organized into a spatiotemporal graph input tensor according to the pipe segment number and the sliding time window sequence number. The graph attention convolutional layer receives the spatiotemporal graph input tensor, extracts spatial features from the feature vectors of pipe segment nodes within each sliding time window, concatenates the feature vectors of the central pipe segment nodes with the feature vectors of the neighboring pipe segments, calculates the attention score, normalizes it using softmax to obtain the attention weights of each neighbor, and then sums the weighted feature vectors of the neighboring pipe segments to obtain the spatial enhancement features.
6. The intelligent anti-blocking control method for dense-phase pneumatic conveying according to claim 5, characterized in that, The specific steps for constructing a congestion risk prediction model also include: The temporal self-attention layer arranges the spatial enhancement features of multiple consecutive sliding time windows in temporal order, calculates the attention weight between different sliding time windows through a multi-head self-attention mechanism, automatically focuses on the dynamic patterns of blockage precursors such as pressure gradient rise, continuous decrease of flow index, and expansion of instantaneous solid-gas ratio deviation, and outputs temporal enhancement features. The feature mapping layer flattens the time-series enhancement features through multi-layer feature mapping, extracts high-dimensional combination features between the pressure difference gradient, material rate increase rate, instantaneous solid-gas ratio deviation, and flowability index of each pipe section layer by layer, and maps them into an abstract representation related to blockage precursors. The output layer will abstractly represent the probability of blockage risk for each pipe segment at a future preset time step, which will be mapped to the final fully connected layer.
7. The intelligent anti-blocking control method for dense-phase pneumatic conveying according to claim 6, characterized in that, The specific steps for outputting the probability of congestion risk include: The labeled spatiotemporal graph sample set is divided into training set and validation set according to a preset ratio. The weighted focus loss function is used as the loss function. The congestion risk prediction model is trained in combination with Adam optimizer. The training is completed when the loss function value on the validation set no longer decreases for a preset number of consecutive rounds or reaches the preset maximum number of training rounds. The real-time gas-solid fluid dynamics feature matrix, pipe segment adjacency topology graph, and pipe segment node feature vectors are input into the trained blockage risk prediction model, which outputs the blockage risk probability.
8. The intelligent anti-blocking control method for dense-phase pneumatic conveying according to claim 7, characterized in that, The specific steps for classifying the risk levels of pipe blockage precursors include: Calculate the risk escalation rate of each pipe segment under a preset time step in the future, screen the maximum blockage risk probability, and classify the risk level of pipe blockage precursors; The system uses a priority-based approach to determine whether the conditions for a high-risk level are met. When the maximum probability of congestion risk is greater than or equal to the preset third risk threshold and the rate of risk increase is greater than or equal to the preset rapid increase threshold, the system is classified as a high-risk level. If the high-risk level conditions are not met, then it is determined whether the early warning level conditions are met. When the maximum congestion risk probability is greater than or equal to the preset second risk threshold, or the risk rise rate is greater than or equal to the preset rapid rise threshold, it is determined to be an early warning level. If the conditions for the warning level are not met, then it is determined whether the conditions for the attention level are met. When the maximum probability of congestion risk is greater than or equal to the preset first risk threshold, or the rate of risk increase is greater than or equal to the preset slow increase threshold, it is determined to be at the attention level. Otherwise, it is classified as a safe level.
9. The intelligent anti-blocking control method for dense-phase pneumatic conveying according to claim 8, characterized in that, The specific steps for generating the anti-blocking control parameter sequence include: Based on the probability of blockage risk and the rate of risk increase within a future preset time step, combined with the real-time acquisition of differential pressure gradient, material rate increase rate, instantaneous solid-gas ratio deviation, gas replenishment node pressure, material conveying speed, and the current actual opening degree of each gas replenishment valve, actual discharge speed of the silo pump, and status of pressure relief valve and blockage bypass valve, the risk level of pipe blockage precursors is mapped to a gradient control mode, and a list of target control ranges is formed. Based on the list of risk levels of pipe blockage precursors and target control ranges, a gradient depressurization-gas replenishment coordinated control logic is established to determine the priority order, corresponding adjustment direction, and adjustment range of each control variable. Historical adjustment response data is obtained, and combined with the differential pressure gradient, material velocity increase rate and instantaneous solid-gas ratio deviation of each pipe section, the control parameters of the target opening degree of each air replenishment valve, air replenishment flow rate and silo pump discharge speed under the current conveying state are calculated through rolling optimization to generate an anti-blocking control parameter sequence.
10. The intelligent anti-blocking control method for dense-phase pneumatic conveying according to claim 9, characterized in that, The specific steps for evaluating the actual effectiveness of the current gradient-based blocking control include: The anti-blockage control parameter sequence is broken down into a single-cycle execution instruction set according to the control cycle. The execution timing offset is set according to the pipe segment adjacency topology and sent to the actuator of each pipe segment to perform gas replenishment adjustment and control parameter optimization. The feedback data window is extracted according to the lag time of the flow response. The feedback data is acquired and aligned in real time. The improvement rate of pressure gradient, the improvement rate of material rate, and the improvement rate of instantaneous solid-gas ratio deviation are calculated. The weighted sum is used to obtain the control response score. To evaluate the actual effect of the current gradient anti-blocking control, when the probability of blockage risk is higher than the preset safety limit and the control response score is lower than the preset response benchmark value of the current gradient for two consecutive control cycles, switch to the next control gradient and call the corresponding anti-blocking control parameters to perform enhanced anti-blocking intervention. When two consecutive control cycles meet opposite conditions, the control gradient is gradually lowered until the maintenance mode of the safety level is restored.
11. A dense-phase pneumatic conveying intelligent anti-blocking control system, used to implement the dense-phase pneumatic conveying intelligent anti-blocking control method as described in any one of claims 1-10, characterized in that, It includes a flow perception module, a risk prediction module, a blockage prevention and control module, and a control execution module; The flow perception module is used to acquire multi-source time-series data for timestamp synchronization and pipe segment spatial mapping, calculate material mass flow rate and instantaneous solid-gas ratio, extract pressure gradient, material velocity boost rate and instantaneous solid-gas ratio deviation of each pipe segment, and combine material property parameters to fit the material flow rate change curve in real time to construct a gas-solid flow characteristic matrix. The risk prediction module is used to obtain historical blockage condition annotation dataset and pipeline topology connection relationship, construct pipe segment adjacency topology map, construct blockage risk prediction model by combining gas-solid flow state feature matrix, output blockage risk probability, and classify pipe blockage precursor risk level. The anti-blocking control module is used to establish a gradient pressure reduction-gas replenishment coordinated control logic based on the risk level of pipe blockage precursors and determine the target control range. It generates an anti-blocking control parameter sequence by calculating control parameters through rolling optimization. The control execution module is used to issue anti-blocking control parameter sequences, perform gas replenishment adjustment and control parameter optimization, obtain feedback data in real time to calculate the control response score, and evaluate the actual effect of the current gradient anti-blocking control.
Citation Information
Patent Citations
Anti-blocking control method and system
CN120397737A
