Dynamic dust concentration and air network balance optimization method and system in dust removal equipment pipeline

Abstract

The invention relates to the technical field of industrial dust removal and dust monitoring, in particular to a dynamic dust concentration and air network balance optimization method and system in a dust removal equipment pipeline. According to the state estimation method, state vectors containing dust concentration, wind speed, wind pressure and other parameters are constructed, Kalman filtering physical constraint and pre-training GRU network time sequence analysis are combined, and optimal estimation is obtained through predicted value correction weighted fusion; the operation risk judgment method is based on the estimation, and the risk level is determined through three-level threshold value, concentration trend quantification and material blockage confidence coefficient calculation; according to the opening degree prediction method, a relation model of the branch pipe wind speed, the valve opening degree and the total wind pressure is established, and the valve opening degree is optimized with wind network balance and energy saving as targets. The technical problems of inaccurate state estimation, lagging risk prejudgment, unbalanced wind network and high energy consumption of the dust removal system are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of industrial dust removal and dust monitoring technology, specifically a method and system for optimizing dynamic dust concentration and air network balance within the pipeline of a dust removal equipment. Background Technology

[0002] In industrial settings such as port bulk grain unloading, grain transportation, and mining and metallurgy, dust removal systems are core equipment for ensuring production safety and controlling dust pollution. Their operational status directly affects the continuity of operations and the safety of personnel and equipment. In these scenarios, dust removal systems operate continuously under high dust concentrations and high airflow loads. Key parameters such as dust concentration, wind speed, and wind pressure within the pipelines are prone to nonlinear changes with fluctuations in production load. If these parameter changes are not monitored in a timely manner, serious risks such as pipeline blockage, decreased dust removal efficiency, and even dust explosions can easily occur.

[0003] Existing dust removal systems typically monitor status by deploying sensors at key locations and employing a single-point monitoring and alarm mechanism based on fixed thresholds. This involves comparing sensor data with preset thresholds to trigger an alarm. Control at the control level relies on manual experience to adjust fan frequency, valve opening, etc., or uses feedback control logic based on current / historical states. This approach is essentially a reactive mode, only alerting to risks when an anomaly occurs or approaches a critical state. Furthermore, the monitoring, alarm, and actuator control logic are disconnected, lacking a holistic understanding of the overall situation and the ability to proactively adjust control measures.

[0004] Although existing technologies have achieved basic condition monitoring and control, the following technical problems still exist: First, sensor data is susceptible to noise and drift caused by dust interference and airflow fluctuations. The reliability of single monitoring data is insufficient, and there is a lack of a fusion mechanism that combines physical constraints with temporal feature learning, resulting in inaccurate estimation of dynamic dust concentration and system state, which makes it difficult to support subsequent accurate decision-making.

[0005] Secondly, risk assessment relies on a single threshold comparison, focusing only on whether the current parameters exceed the standard, without combining multi-dimensional information such as concentration change trends and wind speed attenuation characteristics for comprehensive judgment. This makes it impossible to identify the initial risk of material blockage and the cumulative risk of dust explosion in advance, resulting in a significant lag in early warning.

[0006] Third, the wind network regulation adopts a single-point independent control mode and has not established a global state assessment and multi-objective optimization model, which easily leads to the phenomenon of "wind grabbing" and "wind stagnation" in branch pipes, poor wind network balance, difficulty in accurately controlling the opening of pipeline valves, and failure to achieve energy consumption optimization under the premise of ensuring safety, resulting in energy waste.

[0007] This shows that the current dust removal system still needs further improvement. Summary of the Invention

[0008] To address the technical problem of inaccurate state estimation of dust removal equipment pipelines due to dust interference and airflow fluctuations in sensor data, this invention provides a state estimation method for dust removal equipment pipelines. To address the technical problem of delayed pipeline safety warnings caused by reliance on single threshold comparisons for risk assessment and focusing only on whether current parameters exceed limits, this invention provides an operational risk assessment method for dust removal equipment pipelines based on the state estimation method. To address the technical problem of difficulty in accurately controlling the opening of various valves in pipelines due to the use of a single-point independent control mode in wind network regulation without considering risk balance, this invention provides an opening prediction method for dust removal equipment pipelines based on the operational risk assessment method. Combining the above three methods, this invention also provides a control system for dust removal equipment pipelines.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A method for estimating the state of a dust removal equipment pipeline includes the following estimation steps: Construct a state vector containing dust concentration, wind speed, wind pressure, and the rate of change of these three parameters, and load the observation noise covariance matrix; Based on the optimal state estimation vector of the previous moment, the state is predicted under the constraints of physical laws through the state transition matrix, and the predicted state vector and its prediction error covariance matrix at the current moment are obtained. Using a pre-trained gated recurrent unit network, temporal feature analysis is performed on historical observation sequences and real-time observation state vectors at the current moment to distinguish between real state changes and abnormal interference, and to output corrected observation state vectors. Based on the prediction error covariance matrix and the observation noise covariance matrix, the dynamic Kalman gain is calculated; using the Kalman gain, the predicted state vector and the corrected observation state vector are weighted and fused to obtain the optimal state estimation vector at the current time.

[0010] As a further improvement to the above scheme, state prediction is achieved through the following state prediction equation and covariance prediction equation: State prediction equation: ; Covariance prediction equation: ; In the formula, Let A be the predicted state vector at current time t based on the optimal estimation vector of the state at the previous time t-1; A is the state transition matrix, A T This is the transpose of A; U is the optimal state estimation vector at the previous time t-1; B is the control input matrix; u t-1 P is the control quantity at the previous time t-1; t|t-1P is the prediction error covariance matrix at current time t based on the optimal estimation vector of the state at the previous time t-1; t-1|t-1 Q is the prediction error covariance matrix of the optimal estimation vector of the state at the previous time t-1; Q is the process noise covariance matrix.

[0011] As a further improvement to the above scheme, the optimal state estimation vector at the current moment is represented as follows: ; ; In the formula, K is the optimal state estimation vector at time t; t Z is the Kalman gain at time t; t H is the corrected observation state vector at the current time t; H is the observation matrix. T R is the transpose of H; R is the observation noise covariance matrix.

[0012] A method for determining the operational risk of dust removal equipment pipelines includes the following determination steps: Real-time dust concentration data, real-time wind speed data, and dust concentration prediction sequence are obtained from each monitoring point; based on the above-mentioned state estimation method for dust removal equipment pipelines, rolling prediction is performed to output the dust concentration prediction sequence for multiple future time steps; The real-time dust concentration at each monitoring point is compared with the preset three-level concentration threshold to generate the first alarm level signal; For each monitoring point, based on real-time dust concentration data from multiple historical moments, a linear regression method is used to calculate the instantaneous slope of concentration change as a quantitative indicator of concentration accumulation trend; simultaneously, for each monitoring point, based on the dust concentration prediction sequence, the average predicted concentration change rate for multiple future moments is calculated. By integrating the instantaneous slope and the average predicted concentration change rate, and combining them with the real-time wind speed attenuation rate at the corresponding monitoring point, the comprehensive confidence level of the blockage risk at the monitoring point is calculated using a preset confidence level calculation model. Based on the combined confidence level of the first alarm level signal and the risk of material blockage, the final risk level is determined through predefined risk mapping rules.

[0013] As a further improvement to the above scheme, the formula for calculating the instantaneous slope is as follows: ; In the formula, k i Let be the instantaneous slope of the dust concentration change at the i-th monitoring point; M is the total number of historical time points; t t-M+m This indicates that starting from the current time t, we can trace back Mm time steps to the previous historical moment; Let C be the average time of M historical moments;i,t-M+m For the i-th monitoring point at t t-M+m The dust concentration; Let be the average dust concentration at the i-th monitoring point over M historical time periods.

[0014] As a further improvement to the above scheme, the confidence calculation model is expressed as follows: ; ; In the formula, Conf i,block Let be the comprehensive confidence level of the blockage risk at the i-th monitoring point; min(·) is the minimum value operation; C represents the average predicted concentration change rate at the i-th monitoring point; i,t Let be the dust concentration at the i-th monitoring point at time t; Δt is the change over time. Let be the real-time wind speed attenuation rate at the i-th monitoring point; , These are the threshold values ​​for predicting concentration changes and wind speed decreases, respectively.

[0015] As a further improvement to the above scheme, the risk mapping rules are as follows: If the first alarm level signal at the current monitoring point is an emergency alarm or the overall confidence level of the material blockage risk is greater than the preset high confidence threshold, then the current monitoring point is determined to be high risk. If the first alarm level signal of the current monitoring point is a warning or the comprehensive confidence level of the material blockage risk is between the preset high confidence threshold and low confidence threshold, then the current monitoring point is determined to be of medium risk. If the first alarm level signal at the current monitoring point is normal or the overall confidence level of the material blockage risk is lower than the preset low confidence threshold, then the current monitoring point is determined to be risk-free.

[0016] A method for predicting the opening degree of a dust removal equipment duct includes the following prediction steps: The system acquires real-time wind speed data, real-time wind pressure data, real-time valve opening data, and the final risk level and comprehensive confidence level of material blockage risk output by the above-mentioned method for determining the operational risk of dust removal equipment pipelines. Based on real-time wind speed, real-time wind pressure, and real-time valve opening data, a quantitative relationship model is established between the wind speed of each branch pipe, the valve opening, and the total wind pressure. The first optimization objective is to minimize the wind speed imbalance between the branches, and the second optimization objective is to minimize the total energy consumption. A multi-objective optimization problem is then constructed. The final risk level and the combined confidence level of the blockage risk are transformed into the minimum wind speed constraint of the corresponding branch pipe, and this minimum wind speed constraint is used as the basic wind speed constraint, which together constitutes the constraint conditions of the multi-objective optimization problem. Solve the multi-objective optimization problem to obtain the optimal opening command for each branch valve in the next control cycle, and control the valve opening according to the optimal opening command.

[0017] As a further improvement to the above scheme, the quantitative relationship model is expressed as follows: ; In the formula, v j Let $\frac{j}{j}$ be the real-time wind speed of the $j$-th branch pipe; j η represents the real-time valve opening degree of the j-th branch pipe; j Let be the real-time resistance coefficient of the j-th branch pipe; P is the total air pressure. And / or, update the real-time drag coefficient using recursive least squares with a forgetting factor: ; In the formula, , are the real-time resistance coefficients of the j-th branch pipe in the nth and n-1th iterations, respectively; α is the forgetting factor. , These represent the real-time wind speed and valve opening of the j-th branch pipe during the nth iteration, respectively. The total wind pressure in the nth iteration; And / or, the first optimization objective is to minimize the wind speed imbalance index U, which is calculated as follows: ; In the formula, J represents the total number of branches; This represents the average wind speed across all branches; And / or, the minimum wind speed constraint is expressed as follows: ; In the formula, v j,min v is the minimum wind speed of the j-th branch pipe; min,base Basic process requires wind speed; L j,final Let be the final risk level of the j-th branch pipe, and its value is the average of the final risk levels of all monitoring points in the j-th branch pipe; Conf j,block β represents the overall confidence level of the blockage risk in the j-th branch pipe, and its value is the mean of the overall confidence levels of the blockage risk at all monitoring points in the j-th branch pipe; L β C These are the corresponding calibration coefficients with dimensions.

[0018] A control system for a dust removal equipment pipeline includes: The state estimation unit is configured to execute the state estimation method of the dust removal equipment pipeline described above, so as to output the real-time optimal state estimation vector and dust concentration prediction sequence of each monitoring point after cleaning and fusion. The operation risk assessment unit is connected to the state estimation unit and is configured to execute the above-mentioned method for assessing the operation risk of a dust removal equipment pipeline. It receives the real-time optimal state estimation vector and the dust concentration prediction sequence, and outputs a comprehensive risk assessment result that includes the risk level and the comprehensive confidence level of the material blockage risk. The dynamic balancing unit is communicatively connected to the state estimation unit and the operation risk assessment unit, respectively. It is configured to execute the above-mentioned method for predicting the opening of a dust removal equipment pipeline, receive real-time wind speed, wind pressure, valve opening data and comprehensive risk assessment results, and output the optimal opening command for each branch valve. The arbitration and execution unit, which is connected to the operation risk assessment unit and the dynamic balancing unit, is configured to arbitrate the optimal opening command from the dynamic balancing unit based on the risk level in the comprehensive risk assessment result, generate the final control command, and send it to the corresponding valve actuator.

[0019] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention achieves accurate state estimation through a three-layer logic of "physical model constraints + temporal feature learning + dynamic weighted fusion". First, a state vector containing dust concentration, wind speed, wind pressure, and their rates of change is constructed, and an observation noise covariance matrix is ​​loaded to clarify the core monitoring dimensions and quantify measurement errors. Second, based on the previous optimal estimate and state transition matrix, prediction is performed under physical constraints. The state prediction equation and covariance prediction equation using Kalman filtering ensure that the prediction results conform to fluid dynamics and filter out sudden abnormal fluctuations without physical basis. Then, a pre-trained gated recurrent unit (GRU) network analyzes historical observation sequences and current real-time data to accurately distinguish between "real operating condition changes" and "abnormal interference", outputting a corrected observation vector. Finally, the dynamic Kalman gain is calculated using the prediction error covariance matrix and the observation noise covariance matrix, and the predicted state vector and the corrected observation vector are weighted and fused. This ensures both the physical rationality of the data and conformity to the actual operating condition temporal patterns, ultimately outputting a high signal-to-noise ratio optimal state estimation vector, effectively solving the problems of unreliable single data and large state estimation deviations.

[0020] 2. This invention addresses the problems of existing risk assessment methods that rely on single threshold comparisons, focus only on current parameters, and fail to integrate multi-dimensional information, leading to delayed early warnings. Based on accurate state estimation, it constructs a multi-dimensional assessment system integrating "basic threshold alarm + trend feature fusion + confidence quantification." First, it obtains dust concentration prediction sequences for multiple future moments through rolling forecasts, providing data support for forward-looking predictions. Second, it retains high, medium, and low concentration threshold comparisons to generate a first alarm level signal, ensuring the core safety baseline. It introduces trend quantification indicators: using linear regression to calculate the instantaneous slope of historical concentration data, it captures the cumulative concentration trend, while simultaneously calculating the average predicted concentration change rate based on the dust concentration prediction sequence, allowing for early perception of future risk evolution directions. Then, it integrates real-time wind speed attenuation rates and calculates the comprehensive confidence level of material blockage risk through a pre-defined confidence model, deeply binding "current state" with "future trend" and "related parameters." Finally, through predefined risk mapping rules, it combines the first alarm level signal and the comprehensive confidence level to determine the final risk level, achieving a shift from "alarm after exceeding the standard" to "early warning," effectively solving the problem of delayed prediction of material blockage and dust explosion risks.

[0021] 3. This invention addresses the problem of imbalance in wind network regulation caused by the use of single-point independent control and lack of global optimization in existing wind network regulation, and constructs a regulation logic of "global state modeling + multi-objective optimization + safety constraint embedding". First, based on real-time wind speed, wind pressure, and valve opening data, a quantitative relationship model is established between branch pipe wind speed, valve opening, and total wind pressure. The branch pipe resistance coefficient is updated in real-time using a recursive least squares method with a forgetting factor to ensure the model adapts to changes in operating conditions such as pipe dust accumulation and equipment aging, accurately reflecting the overall air network status. Second, dual optimization objectives are defined: minimizing branch pipe wind speed imbalance is the core objective to solve the problems of "wind grabbing" and "wind stagnation," while minimizing total energy consumption is the secondary objective to achieve energy-saving requirements. The key lies in transforming the final risk level and the comprehensive confidence level of material blockage risk into the minimum wind speed constraint of the corresponding branch pipe. Combined with the basic process wind speed requirements, constraints for a multi-objective optimization problem are constructed to ensure targeted protection of risk points during control. Finally, by solving the multi-objective optimization problem, the optimal valve opening command for the next control cycle is output, achieving a unified approach of "global air network balance," "safety control," and "energy saving and consumption reduction," completely overcoming the limitations of single-point control. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method of the present invention.

[0023] Figure 2 This is a dust concentration prediction curve for this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figure 1 In this embodiment of the invention, addressing the technical problems of inaccurate state estimation, delayed risk prediction, wind network imbalance, and safety-energy efficiency conflicts in dust removal systems in high-dust industrial scenarios such as grain transportation, bulk grain transportation at ports, and mining and metallurgy, a complete implementation solution for dust removal equipment pipeline state estimation, operational risk assessment, opening degree prediction, and control system is elaborated based on a cloud-edge collaborative architecture of "real-time response at the edge + deep optimization at the cloud". This solution utilizes the KF-GRU fusion algorithm to achieve accurate state perception, multi-dimensional indicator fusion for forward-looking risk assessment, a multi-objective optimization model to achieve wind network balance and energy-saving synergy, and a reinforcement learning arbitration mechanism to achieve a balance between safety and efficiency. A closed-loop process of "data acquisition - state estimation - risk assessment - optimization and control - arbitration execution - model evolution" is constructed to ensure the completeness and effectiveness of the technical solution.

[0026] I. Overall Architecture

[0027] 1. Cloud-edge collaborative closed-loop framework

[0028] This invention uses a core architecture of "real-time edge response + deep cloud optimization" to achieve a closed-loop process encompassing "data acquisition - state estimation - risk assessment - optimization and control - arbitration execution - model evolution". The architecture design balances the real-time requirements (millisecond-level risk response) and self-evolution needs of industrial scenarios (adapting to different grain types and seasonal conditions), solving the technical problems of "lagging response and poor adaptability" in traditional dust removal systems.

[0029] 2. Software and hardware deployment

[0030] (1) Sensor

[0031] At key monitoring points such as dust collector branch pipes, dust collector inlets and outlets, and main pipelines, laser dust concentration sensors and ultrasonic wind speed and pressure sensors (adapted to high dust environments, anti-adhesion, and anti-interference) are deployed. The sensor sampling frequency is matched with the control cycle, and the data is accessed to the edge gateway via the Modbus RTU protocol to ensure timestamp alignment.

[0032] (2) Edge side

[0033] Deploy an industrial edge computing gateway (running the KF-GRU lightweight model and risk assessment logic), a PLC controller (connecting to electric regulating butterfly valves and variable frequency fans), an operating system adapted to the industrial environment, and integrate TensorFlow Lite (lightweight model running) and pymodbus (data acquisition) tools to ensure real-time performance.

[0034] (3) Cloud

[0035] Deploy cloud servers (running LSTM deep learning models, Stacking ensemble learners, and reinforcement learning arbitration strategies), store historical data (operational data for different grain types, seasons, and workloads), and communicate bidirectionally with the edge via the MQTT protocol to achieve parameter distribution and data upload.

[0036] (4) Human-computer interaction layer

[0037] It provides an industrial-grade HMI interface that supports real-time status monitoring, manual intervention, and fault alarms. Manual commands have higher priority than automatic commands, adapting to on-site operation and maintenance needs.

[0038] II. State Estimation

[0039] This section performs state estimation based on the state estimation unit, which is the core of accurate data in the dust removal system. Based on the KF-GRU fusion algorithm, it achieves high signal-to-noise ratio estimation of dust concentration, wind speed, wind pressure, and rate of change. The specific implementation is as follows: 1. Core parameter initialization (1) Constructing the state vector The state vector is X=[C,V,P,ΔC,ΔV,ΔP] T Where C is the dust concentration, V is the wind speed, and P is the wind pressure, and ΔC / ΔV / ΔP are the rates of change of these three values. The state vector design is adapted to the nonlinear fluctuation characteristics of parameters in grain transportation scenarios, and can comprehensively capture the dynamic correlation between airflow and dust.

[0040] (2) Define the noise covariance matrix

[0041] The initial value of the process noise covariance matrix Q (used to describe the uncertainty of the dust removal system, such as small fluctuations in airflow) is derived from the statistical analysis of parameter fluctuations under normal operating conditions of grain transportation, and is a diagonal matrix.

[0042] The initial value of the observation noise covariance matrix R (describing sensor error) is set according to the accuracy specifications in the sensor manual and is a diagonal matrix.

[0043] (3) State transition matrix

[0044] The state transition matrix A is constructed based on the principles of fluid mechanics and the "wind speed / concentration inertial characteristics" of the grain transportation scenario. The matrix elements reflect the time evolution of each parameter (such as the concentration maintaining inertia for a short period of time), ensuring that the prediction results conform to physical common sense and filtering out unfounded abnormal fluctuations.

[0045] (4) Control input matrix and control quantity

[0046] The control input matrix B quantizes the effect of fan frequency adjustment on wind speed, u t-1 This is the fan frequency adjustment amount at the previous time step t-1, ensuring the correlation between control actions and state evolution.

[0047] 2. GRU network

[0048] (1) Dataset construction

[0049] Data on normal operating conditions (covering different grain types such as wheat and corn, and different loads such as empty and full load) of bulk grain unloading and grain conveyor belts in ports are collected and divided into training set, validation set and test set according to a set ratio (such as 7:2:1). The data dimension is consistent with the state vector to ensure that the model is adapted to the characteristics of grain transportation scenarios.

[0050] (2) GRU network structure design

[0051] The GRU network consists of an input layer (6-dimensional, matching the state vector), two hidden layers (the number of neurons in each layer is set based on the complexity of the temporal data), and an output layer (6-dimensional, outputting the corrected observation vector). The activation function is ReLU, the optimizer is Adam, and the loss function is MSE (minimizing the deviation between the predicted and the true values). After training until the loss function converges, it is solidified into a lightweight .tflite model and deployed to the edge.

[0052] (3) Function

[0053] By extracting the temporal features of historical observation sequences through a gating mechanism, the system distinguishes between "real operating condition changes" (such as a sudden increase in grain unloading leading to an increase in concentration) and "abnormal interference" (such as dust adhering to the sensor causing reading drift), and outputs the corrected observation state vector Z. t .

[0054] 3. State fusion

[0055] (1) State prediction

[0056] Within each control cycle (e.g., 1 second), based on the optimal estimation vector from the previous moment... By using the state prediction equation and the covariance prediction equation, the predicted state vector and the prediction error covariance matrix at the current moment are obtained. Based on these two processes, the distortion effect of single sensor data is avoided by using physical laws to constrain the data.

[0057] State prediction equation: ; Covariance prediction equation: ; In the formula, Let A be the predicted state vector at current time t based on the optimal estimation vector of the state at the previous time t-1; A is the state transition matrix, A T This is the transpose of A; B is the optimal state estimation vector at the previous time t-1; B is the control input matrix; u t-1 P is the control quantity at the previous time t-1; t|t-1 P is the prediction error covariance matrix at current time t based on the optimal estimation vector of the state at the previous time t-1; t-1|t-1 Let Q be the prediction error covariance matrix of the optimal estimation vector of the state at the previous time t-1; Q is the process noise covariance matrix.

[0058] (2) Calculate the Kalman gain

[0059] The Kalman gain is calculated using the following formula: ; In the formula, K t H represents the Kalman gain at the current time t; H is the observation matrix, which is the identity matrix, and its elements are the state parameters directly measured by the sensor; H T R is the transpose of H; R is the observation noise covariance matrix.

[0060] (3) Fusion optimal estimation

[0061] The predicted state vector and the corrected observed state vector are weighted and fused by the state update formula to output the optimal state estimation vector with high signal-to-noise ratio, providing accurate data support for subsequent processing.

[0062] The state update formula is: ; In the formula, Z is the optimal state estimation vector at time t; t R is the corrected observation state vector at the current time t; R is the observation noise covariance matrix.

[0063] 4. Online parameter calibration and cloud-edge collaboration

[0064] (1) Parameter calibration

[0065] The deviation between the optimal state estimation vector and the actual observation data is statistically analyzed on the edge side per unit time. Q and R are updated using the exponential weighted moving average method to ensure that the noise matrix adapts to changes in operating conditions.

[0066] (2) Cloud-edge collaboration

[0067] The edge side uploads the optimal state estimation data to the cloud as input to the cloud LSTM model for long-term trend prediction; the cloud optimizes the GRU network parameters and Q / R initial values ​​based on massive historical data and sends them down to the edge side for updates, forming a closed loop of "real-time edge processing + deep cloud optimization".

[0068] III. Operational Risk Assessment

[0069] This section establishes a communication connection between the risk assessment unit and the state estimation unit. Based on the state estimation output, it constructs a multi-dimensional assessment system that combines threshold alarms, trend quantification, and confidence level fusion. This system is adapted to the characteristics of grain transportation scenarios, such as "gradual material blockage and non-linear concentration changes." The specific implementation is as follows: 1. Input Data and Preprocessing (1) Input data Receive three types of data from the state estimation output: Real-time data after cleaning: Dust concentration C t Wind speed V t Wind pressure P t Noise has been filtered out and timestamps are aligned.

[0070] Short-term forecast data: Dust concentration prediction sequences for multiple future times obtained through rolling forecasts (e.g., Figure 2 (as shown) and prediction confidence levels. From Figure 2 As can be seen, the predicted value can closely follow the overall trajectory of the actual dust concentration. Whether it is the fluctuation and adjustment in the early stage, the rapid rise and peak in the middle stage, or the decline and fall in the later stage, it can basically match the rhythm of the actual value. In particular, for the concentration peak that appears around 30 hours (h), the model not only accurately captures the time point of the peak, but also closely matches the actual value in terms of magnitude. This shows that it has a strong perception and fitting ability for the main dynamic changes of dust concentration, and can provide trend predictions with reference value for relevant scenarios.

[0071] Equipment status data: Branch valve opening degree t , Fan operating frequency.

[0072] (2) Pretreatment

[0073] Data validity is verified (outliers exceeding the sensor's range are removed), and the format is standardized to ensure the reliability of risk assessment.

[0074] 2. Generate the first alarm level signal

[0075] (1) Threshold setting

[0076] Based on the lower explosion limit of grain dust (wheat / corn), three - level concentration thresholds are set: Warning threshold T1: 1 / 3 of the lower explosion limit.

[0077] Alarm threshold T2: 2 / 3 of the lower explosion limit.

[0078] Emergency threshold T3: 83% of the lower explosion limit.

[0079] (2)Alarm signal generation

[0080] When C t < T1 and V t ≥ V min,base (basic process air velocity), the output is normal.

[0081] When T1 ≤ C t < T3, the warning is output.

[0082] When C t ≥ T3 or V t < V min,base , the emergency alarm is output.

[0083] 3. Calculate the trend quantification index

[0084] (1)Calculate the instantaneous slope

[0085] The instantaneous slope k is calculated by linear regression through the following formula i , quantifying the cumulative trend of concentration.

[0086] ; In the formula, k i is the instantaneous slope of the change in dust concentration at the i - th monitoring point; M is the total number of historical moments; t t-M+m represents the historical moment m time steps back from the current moment t; is the average moment of M historical moments; C i,t-M+m is the dust concentration at the i - th monitoring point at t t-M+m ; is the average dust concentration of M historical moments at the i - th monitoring point.

[0087] (2)Average predicted concentration change rate

[0088] The average predicted concentration change rate is calculated according to the following formula to perceive the future concentration evolution direction in advance.

[0089] ; In the formula, is the average predicted concentration change rate at the i - th monitoring point; C i,t is the dust concentration at the i - th monitoring point at time t; Δt is the time change amount.

[0090] 4. Calculate the overall confidence level of the risk of material blockage.

[0091] The comprehensive confidence level of the corresponding blockage risk is calculated using a confidence level calculation model, which is expressed as follows: ; In the formula, Conf i,block Let be the comprehensive confidence level of the blockage risk at the i-th monitoring point; min(·) is the minimum value operation; Let be the real-time wind speed attenuation rate at the i-th monitoring point; , These are the threshold values ​​for predicting concentration changes and wind speed decline, respectively, which can be calibrated on-site.

[0092] The closer the overall confidence level of the material blockage risk is to 1, the higher the material blockage risk.

[0093] 5. Final Risk Level Determination and Model Optimization

[0094] (1) Risk mapping

[0095] Based on the combined confidence level of the first alarm level signal and the risk of material blockage, the final risk level is determined through predefined risk mapping rules.

[0096] The risk mapping rules are as follows: If the first alarm level signal at the current monitoring point is an emergency alarm or the overall confidence level of the material blockage risk is greater than the preset high confidence threshold (e.g., 0.7), then the current monitoring point is determined to be high risk.

[0097] If the first alarm level signal of the current monitoring point is a warning or the comprehensive confidence level of the material blockage risk is between the preset high confidence threshold and low confidence threshold (such as 0.3), then the current monitoring point is determined to be of medium risk.

[0098] If the first alarm level signal at the current monitoring point is normal or the overall confidence level of the material blockage risk is lower than the preset low confidence threshold, then the current monitoring point is determined to be risk-free.

[0099] (2) Stacking Ensemble Learning Optimization

[0100] Extracting features from state estimation data (C t V t k i Conf i,blockThe model was labeled with tags for blockage, fan failure, and sensor failure. A Stacking ensemble learner (based on XGBoost, Random Forest, and Rule Baseline) was used. The data was divided into 60% training set, 20% validation set, and 20% test set. The model parameters were iteratively updated with the false positive rate and false negative rate as optimization objectives.

[0101] IV. Opening Degree Prediction

[0102] This section establishes a communication connection between the dynamic balancing unit and the operational risk assessment unit. Based on the risk assessment results, it constructs a global multi-objective optimization model for the wind network, achieving synergy between "explosion prevention and suppression + wind network balancing + energy conservation and consumption reduction." The specific implementation is as follows: 1. Quantitative Relationship Model (1) Input data It receives real-time wind speed, wind pressure, and valve opening data for status estimation, as well as the final risk level and overall confidence level of material blockage risk for risk assessment.

[0103] (2) Constructing a quantitative relationship model

[0104] Construct a quantitative relationship model using the following formula: ; In the formula, v j Let be the real-time wind speed of the j-th branch pipe. j Let η be the real-time valve opening degree of the j-th branch pipe, with a value ranging from [0,1]. j is the real-time resistance coefficient of the j-th branch pipe, determined by the pipe length, diameter, and grain dust deposition characteristics. P is the total air pressure.

[0105] (3) Online update of drag coefficient

[0106] Update the real-time drag coefficient using recursive least squares with a forgetting factor: ; In the formula, , are the real-time resistance coefficients of the j-th branch pipe in the nth and n-1th iterations, respectively; α is the forgetting factor. , These represent the real-time wind speed and valve opening of the j-th branch pipe during the nth iteration, respectively. This represents the total wind pressure during the nth iteration.

[0107] 2. Construct a multi-objective optimization problem

[0108] (1) Optimization objective

[0109] Primary optimization objective: Minimize wind speed imbalance according to the following formula:

[0110] In the formula, U is the index for minimizing wind speed imbalance; J represents the total number of branches; This represents the average wind speed across all branches.

[0111] Second optimization objective (sub-objective): Minimize the total energy consumption using the following formula: ; In the formula, G represents the total energy consumption; a j The energy consumption coefficient of the j-th branch pipe can be determined based on the fan characteristic curve and the grain conveying load.

[0112] (2) Constraints

[0113] Safety constraints: , Let C be the predicted dust concentration of the j-th branch pipe at time t+1. safe,j Let be the safe concentration threshold for the j-th branch pipe.

[0114] Wind speed constraint: Calculate the wind speed using the following formula: ; In the formula, v j,min v is the minimum wind speed of the j-th branch pipe; min,base Basic process requires wind speed; L j,final Let be the final risk level of the j-th branch pipe, and its value is the average of the final risk levels of all monitoring points in the j-th branch pipe; Conf j,block β represents the overall confidence level of the blockage risk in the j-th branch pipe, and its value is the mean of the overall confidence levels of the blockage risk at all monitoring points in the j-th branch pipe; L β C These are the corresponding calibration coefficients, whose dimensions are adapted to the corresponding parameters so that the unit of the entire formula is m / s.

[0115] Valve opening constraint: 0.1 ≤ o j ≤0.9.

[0116] 3. Optimize the solution and command generation

[0117] The NSGA-II multi-objective optimization algorithm is used to solve for the Pareto optimal solution set. The algorithm parameters are: population size 100, number of iterations 50, crossover probability 0.8, and mutation probability 0.1. Based on engineering experience, a compromise solution of "wind grid balance + energy saving" is selected, and the valve opening vector O for the next control cycle is output. t+1 For example, high-risk branch pipes are opened wider (to enhance cleaning), while other branch pipes are finely adjusted to maintain the balance of the air network.

[0118] V. Implementation of Arbitration and Enforcement

[0119] The arbitration and enforcement unit in this section, acting as the "central dispatch center" of the dust removal system, enables module collaboration and safety-energy efficiency coordination. The specific implementation is as follows: 1. Module communication and data synchronization (1) Communication connection The status estimation unit communicates with the operation risk assessment unit and the dynamic balancing unit via industrial Ethernet; the operation risk assessment unit, the dynamic balancing unit, and the arbitration and execution unit transmit data through standardized protocols to ensure real-time performance.

[0120] (2) Data flow

[0121] State estimation unit outputs: real-time optimal estimation vector and dust concentration prediction sequence.

[0122] Risk assessment unit output: Comprehensive risk assessment result.

[0123] Dynamic balancing unit output: Optimal valve opening command; Arbitration and Enforcement Unit Output: Final Control Command.

[0124] 2. State Machine Management and Instruction Arbitration

[0125] (1) System state division: Normal operation: No high-risk points, dynamic balancing unit instructions are executed directly.

[0126] Emergency handling: At any high-risk point, interrupt the optimization command and execute the preset safety command (branch valve opening → 0.9, fan frequency → 80% of rated value, to achieve "safe dilution / enhanced cleaning").

[0127] Recovery in progress: After the high risk is resolved, the system will linearly transition to optimized instructions within a set time.

[0128] Manual mode: The operator issues a command (highest priority) through the HMI to pause automatic control.

[0129] (2) Instruction queue priority

[0130] Manual commands > Emergency safety commands > Medium risk control commands > Normal optimization commands (executed according to timestamp for the same priority).

[0131] 3. Health monitoring and reinforcement learning arbitration strategy

[0132] (1) Health monitoring

[0133] The watchdog timer is set to a response threshold of 200ms. If the state estimation / risk assessment unit does not provide feedback after the timeout, it is determined to be a module failure. The system immediately enters a safe state (the valve maintains its current opening, and the fan maintains the minimum safe frequency) and pops up an alarm.

[0134] (2) Strengthen learning arbitration strategy

[0135] Reinforcement learning optimization utilizes a triplet containing the dust removal system state (composed of wind speed imbalance, average wind speed, final risk level, and comprehensive confidence level of material blockage risk), arbitration decision, and execution effect. A corresponding reward mechanism is established: the reward value is +1 when the system is safe, wind speed imbalance < 0.5, and energy consumption is reduced by 5%; the reward value is 0 when the system is safe and wind speed imbalance U < 1.0; and the reward value is -1 when high risk occurs or wind speed imbalance U ≥ 1.5. Simultaneously, the DQN algorithm is employed, with a discount factor of 0.9, an experience replay pool capacity of 10000, and a target network update cycle of 100 steps. By maximizing long-term cumulative rewards, the state switching logic is continuously optimized, thereby resolving the technical problem of conflict between safety and energy efficiency instructions.

[0136] 4. Instruction execution and closed-loop feedback

[0137] (1) Issuance of instructions

[0138] The final control command is sent to the PLC via the Profinet protocol to drive the electric regulating butterfly valve and the variable frequency fan (valve response time ≤ 0.5s, adapting to real-time control requirements).

[0139] Closed-loop feedback: After the action stabilizes, the sensor collects new state data and feeds it back to the state estimation unit to start the next round of "estimation-determination-optimization-arbitration" closed-loop operation.

[0140] VI. System Collaboration and Cloud-Edge Collaboration

[0141] 1. Edge side

[0142] The system utilizes a lightweight KF-GRU model, risk assessment logic, and dynamic balance optimization algorithm to achieve millisecond-level data processing and command issuance (meeting the real-time control requirements for grain transportation); it only uploads high-value data (abnormal operating conditions, decision-making effects, and standardized normal data) to reduce network pressure.

[0143] 2. Cloud

[0144] Run an LSTM deep learning model (for long-term trend prediction), a Stacking ensemble learner (for risk model optimization), and a reinforcement learning arbitration strategy (for global strategy optimization). Based on historical data of different grain types, seasons, and workloads, optimize parameters such as Q, R, threshold, and resistance coefficient on the edge side, and send updates via the MQTT protocol to achieve system self-evolution.

[0145] 3. Resumable download and redundancy design

[0146] The edge side and the cloud adopt 5G+fiber dual links, which support breakpoint resume and avoid data loss caused by network interruption; the edge side has built-in basic control logic (fixed threshold early warning, basic wind network balancing) and runs independently in the event of cloud failure (ensuring uninterrupted grain transportation).

[0147] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for estimating the state of a dust removal equipment pipeline, characterized in that, The estimation steps include the following: Construct a state vector containing dust concentration, wind speed, wind pressure, and the rate of change of these three parameters, and load the observation noise covariance matrix; Based on the optimal state estimation vector of the previous moment, the state is predicted under the constraints of physical laws through the state transition matrix, and the predicted state vector and its prediction error covariance matrix at the current moment are obtained. Using a pre-trained gated recurrent unit network, temporal feature analysis is performed on historical observation sequences and real-time observation state vectors at the current moment to distinguish between real state changes and abnormal interference, and to output corrected observation state vectors. Based on the prediction error covariance matrix and the observation noise covariance matrix, the dynamic Kalman gain is calculated; using the Kalman gain, the predicted state vector and the corrected observation state vector are weighted and fused to obtain the optimal state estimation vector at the current time.

2. The method for estimating the state of a dust removal equipment pipeline according to claim 1, characterized in that, State prediction is achieved through the following state prediction equation and covariance prediction equation: State prediction equation: ； Covariance prediction equation: ； In the formula, Let A be the predicted state vector at current time t based on the optimal estimation vector of the state at the previous time t-1; A is the state transition matrix, A T This is the transpose of A; U is the optimal state estimation vector at the previous time t-1; B is the control input matrix; u t-1 P is the control quantity at the previous time t-1; t|t-1 P is the prediction error covariance matrix at current time t based on the optimal estimation vector of the state at the previous time t-1; t-1|t-1 Q is the prediction error covariance matrix of the optimal estimation vector of the state at the previous time t-1; Q is the process noise covariance matrix.

3. The method for estimating the state of a dust removal equipment pipeline according to claim 2, characterized in that, The optimal state estimation vector at the current moment is represented as follows: ； ； In the formula, K is the optimal state estimation vector at time t; t Z is the Kalman gain at time t; t H is the corrected observation state vector at the current time t; H is the observation matrix. T R is the transpose of H; R is the observation noise covariance matrix.

4. A method for determining the operational risk of a dust removal equipment pipeline, characterized in that, The following determination steps are included: Acquire real-time dust concentration data, real-time wind speed data, and dust concentration prediction sequence at each monitoring point; based on the state estimation method for dust removal equipment pipelines according to any one of claims 1-3, perform rolling prediction to output a dust concentration prediction sequence for multiple future time steps; The real-time dust concentration at each monitoring point is compared with the preset three-level concentration threshold to generate the first alarm level signal; For each monitoring point, based on real-time dust concentration data from multiple historical moments, a linear regression method is used to calculate the instantaneous slope of concentration change as a quantitative indicator of concentration accumulation trend; simultaneously, for each monitoring point, based on the dust concentration prediction sequence, the average predicted concentration change rate for multiple future moments is calculated. By integrating the instantaneous slope and the average predicted concentration change rate, and combining them with the real-time wind speed attenuation rate at the corresponding monitoring point, the comprehensive confidence level of the blockage risk at the monitoring point is calculated using a preset confidence level calculation model. Based on the combined confidence level of the first alarm level signal and the risk of material blockage, the final risk level is determined through predefined risk mapping rules.

5. The method for determining the operational risk of a dust removal equipment pipeline according to claim 4, characterized in that, The formula for calculating the instantaneous slope is as follows: ； In the formula, k i Let be the instantaneous slope of the dust concentration change at the i-th monitoring point; M is the total number of historical time points; t t-M+m This indicates that starting from the current time t, we can trace back Mm time steps to the previous historical moment; Let C be the average time of M historical moments; i,t-M+m For the i-th monitoring point at t t-M+m The dust concentration; Let be the average dust concentration at the i-th monitoring point over M historical time periods.

6. The method for determining the operational risk of a dust removal equipment pipeline according to claim 5, characterized in that, The confidence level calculation model is represented as follows: ； ； In the formula, Conf i,block The overall confidence level of the blockage risk at the i-th monitoring point; min(·) is the minimum value operation; C represents the average predicted concentration change rate at the i-th monitoring point; i,t Let be the dust concentration at the i-th monitoring point at time t; Δt is the change over time. Let be the real-time wind speed attenuation rate at the i-th monitoring point; , These are the threshold values ​​for predicting concentration changes and wind speed decreases, respectively.

7. The method for determining the operational risk of a dust removal equipment pipeline according to claim 6, characterized in that, The risk mapping rules are as follows: If the first alarm level signal at the current monitoring point is an emergency alarm or the overall confidence level of the material blockage risk is greater than the preset high confidence threshold, then the current monitoring point is determined to be high risk. If the first alarm level signal of the current monitoring point is a warning or the comprehensive confidence level of the material blockage risk is between the preset high confidence threshold and low confidence threshold, then the current monitoring point is determined to be of medium risk. If the first alarm level signal at the current monitoring point is normal or the overall confidence level of the material blockage risk is lower than the preset low confidence threshold, then the current monitoring point is determined to be risk-free.

8. A method for predicting the opening degree of a dust removal equipment pipeline, characterized in that, The prediction steps include the following: The method acquires real-time wind speed data, real-time wind pressure data, real-time valve opening data, and the final risk level and comprehensive confidence level of material blockage risk output by the method for determining the operational risk of a dust removal equipment pipeline as described in any one of claims 4-7. Based on real-time wind speed, real-time wind pressure, and real-time valve opening data, a quantitative relationship model is established between the wind speed of each branch pipe, the valve opening, and the total wind pressure. The first optimization objective is to minimize the wind speed imbalance between the branches, and the second optimization objective is to minimize the total energy consumption. A multi-objective optimization problem is then constructed. The final risk level and the combined confidence level of the blockage risk are transformed into the minimum wind speed constraint of the corresponding branch pipe, and this minimum wind speed constraint is used as the basic wind speed constraint, which together constitutes the constraint conditions of the multi-objective optimization problem. Solve the multi-objective optimization problem to obtain the optimal opening command for each branch valve in the next control cycle, and control the valve opening according to the optimal opening command.

9. The method for predicting the opening degree of a dust removal equipment pipeline according to claim 8, characterized in that, The quantitative relationship model is represented as follows: ； In the formula, v j Let $\frac{j}{j}$ be the real-time wind speed of the $j$-th branch pipe; j η represents the real-time valve opening degree of the j-th branch pipe; j Let be the real-time resistance coefficient of the j-th branch pipe; P is the total air pressure. And / or, update the real-time drag coefficient using recursive least squares with a forgetting factor: ； In the formula, , are the real-time resistance coefficients of the j-th branch pipe in the nth and n-1th iterations, respectively; α is the forgetting factor. , These represent the real-time wind speed and valve opening of the j-th branch pipe during the nth iteration, respectively. The total wind pressure in the nth iteration; And / or, the first optimization objective is to minimize the wind speed imbalance index U, which is calculated as follows: ； In the formula, J represents the total number of branches; This represents the average wind speed across all branches; And / or, the minimum wind speed constraint is expressed as follows: ； In the formula, v j,min v is the minimum wind speed of the j-th branch pipe; min,base Basic process requires wind speed; L j,final Let be the final risk level of the j-th branch pipe, and its value is the average of the final risk levels of all monitoring points in the j-th branch pipe; Conf j,block β represents the overall confidence level of the blockage risk in the j-th branch pipe, and its value is the mean of the overall confidence levels of the blockage risk at all monitoring points in the j-th branch pipe; L β C These are the corresponding calibration coefficients.

10. A control system for a dust removal equipment pipeline, characterized in that, include: The state estimation unit is configured to execute a state estimation method for a dust removal equipment pipeline as described in any one of claims 1-3, so as to output the real-time optimal state estimation vector and dust concentration prediction sequence of each monitoring point after cleaning and fusion. The operation risk assessment unit is connected to the state estimation unit and is configured to perform the operation risk assessment method for the dust removal equipment pipeline as described in any one of claims 4-7. It receives the real-time state optimal estimation vector and dust concentration prediction sequence, and outputs a comprehensive risk assessment result including the risk level and the comprehensive confidence level of the material blockage risk. The dynamic balancing unit is communicatively connected to the state estimation unit and the operation risk assessment unit, respectively. It is configured to execute the opening prediction method of the dust removal equipment pipeline as described in claim 8 or 9, receive real-time wind speed, wind pressure, valve opening data and comprehensive risk assessment results, and output the optimal opening command for each branch valve. The arbitration and execution unit, which is connected to the operation risk assessment unit and the dynamic balancing unit, is configured to arbitrate the optimal opening command from the dynamic balancing unit based on the risk level in the comprehensive risk assessment result, generate the final control command, and send it to the corresponding valve actuator.

Images