A Digital Twin-Based Optimization Method for Bulk Material Handling Scheduling
By constructing a digital twin model of pneumatic conveying and performing multi-objective optimization, the problems of pipeline blockage and high energy consumption in pneumatic conveying systems were solved, and dynamic adaptation and stable conveying of gas-solid two-phase flow were achieved.
Patent Information
- Application Number
- CN202611611910.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing intelligent real-time control systems for production lines based on digital twins have failed to effectively adapt to positive pressure dilute phase and negative pressure dilute phase pneumatic conveying processes, resulting in frequent pipeline blockages, severe particle wear, and high fan energy consumption.
A digital twin model of pneumatic conveying is constructed, adjustable parameter nodes are set, and joint control is carried out through multi-objective optimization algorithm. The initial mechanism parameter set is dynamically corrected, the pipeline virtual state tensor is output, the pipeline conveying capacity margin and processing quality index are verified in real time, and global rolling re-optimization is triggered.
It achieves dynamic adaptation of gas-solid two-phase flow characteristics, reduces pipeline blockage frequency and particle wear, optimizes fan energy consumption, and improves the operational stability and energy efficiency of pneumatic conveying systems.
Smart Images

Figure CN123113402A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of bulk material conveying control technology, and more specifically to a bulk material handling scheduling optimization method based on digital twins. Background Technology
[0002] Particles such as pulverized coal, fly ash, and plastic particles are widely used in chemical and power industries. Pneumatic conveying uses compressed air as a medium to transport materials through closed pipelines. It includes both positive and negative pressure conveying structures and is divided into dilute and dense phase operating conditions based on the material-to-air ratio. Along the pipeline, integrated processing such as mixing, metering, dust removal, and unloading can be completed simultaneously. Processing aids can be added during conveying. Positive pressure is suitable for long-distance, high-volume transport, while negative pressure is suitable for multi-point feeding. Dilute phase conveying has high energy consumption and wear. Traditional solutions rely on manually fixing air pressure, air volume, and material-to-air ratio, which cannot match the gas-solid two-phase flow state inside the pipeline in real time. This easily leads to pipe blockage, particle breakage, and blower wastage, and also poses risks of dust leakage and powder explosion. By using digital twins to build a virtual mirror of the pipeline, the material flow inside the pipeline can be mapped in real time, the risk of pipe blockage can be predicted, and the conveying operating parameters can be optimized through simulation, realizing dynamic intelligent scheduling of the pneumatic conveying system.
[0003] A patent application with publication number CN121857553A discloses an intelligent real-time control system and method for production lines based on digital twins. The system includes: a state perception and acquisition module for acquiring real-time operating parameters of production equipment, material flow status information, and environmental disturbance characteristics; a digital twin modeling module for fusing multi-source data and constructing a precise mapping between the physical entity and virtual model of the production line through a dynamic data-driven twin model, predicting in real-time equipment performance degradation trends and disturbance propagation paths; an intelligent decision-making module for generating real-time scheduling strategies for the production line through a distributed collaborative reinforcement learning algorithm, outputting a set of production line control instructions; and a real-time control feedback module for executing production line control actions through a multi-objective dynamic trade-off algorithm, automatically generating a production line optimization report with equipment health compensation. This addresses the problems of static model construction, lagging decision-making mechanisms, and lack of control feedback in existing technologies.
[0004] Although existing intelligent real-time control systems and methods for production lines based on digital twins have achieved real-time perception and digital twin modeling of the operating status of production equipment, material flow and environmental disturbances, and have carried out scheduling decisions and multi-objective dynamic trade-off control through distributed collaborative reinforcement learning, they have not modeled and designed collaborative scheduling logic for pneumatic conveying processes under multiple working conditions such as "positive pressure dilute phase" and "negative pressure dilute phase". They have not combined the fan air volume, pipeline material-to-air ratio, conveying mode, pipe blockage risk and workstation material flow rate to achieve joint control, making it difficult to adapt to the flow characteristics of gas-solid two-phase flow in the pipe, thus causing problems such as frequent pipe blockage, severe particle wear and high fan energy consumption. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention aims to provide a digital twin-based method for optimizing the scheduling of bulk material handling. This method involves constructing a pneumatic conveying digital twin model and setting adjustable parameter nodes. It calculates a comprehensive prediction error index, dynamically corrects the initial mechanism parameter set, outputs the pipeline virtual state tensor, and calculates the conveying capacity margin of each pipe segment. The positive pressure dilute phase conveying mode and the negative pressure dilute phase conveying mode are used as decision variables for multi-objective optimization. Combined with fan airflow, pipeline material-to-air ratio, workstation material flow rate, and pipe blockage risk level, joint control is implemented to achieve synergistic optimization of conveying modes and control parameters. Simultaneously, the conveying capacity margin and processing quality indicators of the pipe segments are verified in real time based on the pipeline virtual state tensor, dynamically adapting to the gas-solid two-phase flow characteristics, reducing pipeline blockage frequency, minimizing particle wear, and optimizing fan energy consumption.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A digital twin-based method for optimizing the scheduling of bulk material handling includes: Multi-source heterogeneous data is acquired, and a spatiotemporal adaptive interpolation alignment algorithm is used with five physical constraints to form a multi-dimensional time-series data stream. A digital twin model of pneumatic conveying is constructed, which abstracts the closed conveying pipeline into a gas-solid two-phase flow unit, sets adjustable parameter nodes, calculates the comprehensive prediction error index, dynamically corrects the initial mechanism parameter set, and outputs the pipeline virtual state tensor. Calculate the transport capacity margin of each pipe section, mark abnormal pipe sections and perform gas-solid two-phase flow simulation in parallel, set multi-objective optimization criteria, select positive pressure dilute phase transport mode or negative pressure dilute phase transport mode as needed for global optimization, and output control command set. The control command set is sent to each execution device to update the pipeline virtual state tensor and verify the conveying capacity margin, blockage risk level and processing quality indicators of the pipeline section; when the differential pressure, material flow rate or processing parameters of any pipeline section deviate from the safety threshold, global rolling re-optimization is triggered.
[0007] Specifically, the steps for constructing a digital twin model of pneumatic conveying include: The current operating condition features are extracted from the multidimensional time-series data stream, and after normalization, the dimensionality is reduced by the principal component analysis algorithm with physical constraint regularization to obtain the operating condition time window feature vector. Historical operating condition features are obtained, and the operating condition time window feature vector is matched with the historical operating condition features on a unified pipeline mileage coordinate and time scale using weighted similarity to obtain a comprehensive weighted similarity. After sorting in descending order, the top N historical working condition samples with the highest comprehensive weighted similarity are selected. The initial working condition prior label is determined by similarity-weighted voting, and the corresponding historical mechanism parameter values are obtained. The initial mechanism parameter set is generated by local weighted averaging. Based on the initial mechanism parameter set and the characteristic vector of the working condition time window, combined with the hydrodynamic mechanism parameters of gas-solid two-phase flow, a digital twin model of pneumatic conveying is constructed, and the closed conveying pipeline is abstracted as a gas-solid two-phase flow unit and adjustable parameter nodes are set.
[0008] Specifically, the steps for outputting the virtual state tensor of the pipeline include: The comprehensive prediction error index is calculated by continuously comparing the pipeline differential pressure, material conveying rate, pipeline flow velocity, and field measured data predicted by the pneumatic conveying digital twin model with online ensemble learning algorithms. Based on the comprehensive prediction error index, the initial mechanism parameters within the mechanism parameter set are dynamically corrected, and the virtual state tensor of the pipeline is output.
[0009] Specifically, the steps for dynamically correcting the initial mechanism parameter set include: With the goal of minimizing the comprehensive prediction error index, the contribution of each mechanism parameter in the initial mechanism parameter set to the total error is solved by back-calculating the online ensemble learner, and an adaptive correction step size is set to incrementally update each mechanism parameter. Using the sensitivity of the comprehensive prediction error index to the current mechanism parameters as the parameter gradient, the corresponding gradient direction and gradient magnitude are calculated through partial derivatives, and the three mechanism parameters with the largest gradient magnitude are updated first. After updating, recalculate the comprehensive prediction error index. If the comprehensive prediction error index decreases, retain the updated result; otherwise, repeat the iteration until the maximum number of iterations is reached. When the comprehensive prediction error index is continuously lower than the preset steady-state threshold, the correction step size is reduced and the steady-state fine-tuning mode is entered; when the comprehensive prediction error index continuously exceeds the preset steady-state threshold and the operating condition characteristics are deviated, the weighted similarity matching is re-performed, the initial mechanism parameter set is updated, and then the correction is performed.
[0010] Specifically, the steps for calculating the transport capacity margin of each pipeline segment include: Based on unified pipeline mileage coordinates, the entire closed conveying pipeline is discretized into independent pipe segments and formed by dividing nodes such as the inlet, pipeline diameter change section, mixing bin access point, and the inlet and outlet of the sifting and powder removal station. Extract the material conveying rate, upper limit of material mass flow rate, and material consumption rate of downstream gravity mixing bin and sifting and de-powdering station for each pipe section from the pipeline virtual state tensor. The supply and demand deviation of the pipe section is calculated by mass flow rate conservation according to the pipe section topology relationship, with the material consumption rate of the downstream gravity mixing bin, the material consumption rate of the elution and de-powdering station, and the material demand of the adjacent downstream pipe section as the demand. Based on the upper limit of material mass flow rate and the current actual conveying rate, the conveying capacity margin of each pipe section is calculated, and the current conveying status of the pipe section is characterized by the supply and demand deviation of the pipe section.
[0011] Specifically, the steps for parallel simulation and deduction of gas-solid two-phase flow include: Using the unified pipeline mileage coordinates and timestamps of each pipeline segment as spatiotemporal positioning information, and the conveying capacity margin, pipeline supply and demand deviation, material conveying rate, material flow velocity, pipeline pressure difference, and pipeline pressure difference gradient as state features, the spatiotemporal state trajectory of each pipeline segment is formed within a sliding time window. After normalization, a density-based clustering method is used to identify state clusters. Based on the conveying capacity margin, pipeline pressure difference, supply and demand deviation of pipeline section and material flow rate, the state clusters are marked as high blockage risk pipeline sections, pipeline sections with excessive material flow rate and pipeline sections with insufficient material flow rate, and integrated into an abnormal pipeline section set; For abnormal pipe sections, the pneumatic conveying digital twin model is used to perform parallel gas-solid two-phase flow simulation. The simulation pipe section is constructed and the simulation boundary conditions are set. The gas-solid two-phase flow dynamics simulation under positive pressure dilute phase conveying mode and negative pressure dilute phase conveying mode is performed in parallel, and the simulation results of each simulation pipe section are output.
[0012] Specifically, the steps for outputting the control instruction set include: The optimization criteria are based on minimizing the risk of pipe blockage, minimizing the wear rate of particulate materials, achieving the required uniformity of mixing, ensuring compliance with the dust content of discharged materials, and minimizing the energy consumption of the blower. Positive pressure dilute phase transport mode and negative pressure dilute phase transport mode are used as discrete transport mode decision variables, and the control parameters corresponding to the adjustable parameter nodes are used as continuous control variables. The system performs global optimization by using the pipeline's allowable pressure range, the equipment's rated operating range, the minimum stable conveying speed, the maximum allowable material concentration, the lower limit of mixing uniformity, and the upper limit of discharge dust content as optimization constraints, and outputs a set of control instructions.
[0013] Specifically, the steps for global optimization include: A non-dominated sorting genetic algorithm with an elite strategy is used for global optimization. An initial population is generated based on the current operating parameters, positive pressure dilute phase transport mode and negative pressure dilute phase transport mode. Discrete transport mode decision variables and continuous control variables are mixed and encoded. The pneumatic conveying digital twin model is used to solve for each optimization objective value and constraint violation. Elite individuals are selected through non-dominated sorting and crowding calculation. The next generation population is generated through selection, crossover, and mutation operations. The Pareto optimal solution set is obtained by iterating to the maximum number of iterations. The comprehensive optimal solution is selected based on the entropy weight TOPSIS method to determine the optimal conveying mode and corresponding control parameter combination.
[0014] Specifically, the steps for updating the pipeline virtual state tensor and verifying the pipeline segment's transport capacity margin, blockage risk level, and processing quality indicators include: Based on the control characteristics of the execution equipment and the impact characteristics of pneumatic conveying operation, the control instruction set is decomposed into a three-level priority instruction subset. Combined with the current pipeline blockage risk level, pipeline topology and equipment executable status, an operation condition adaptation timing arbitration rule is constructed. Arbitration generates equipment instruction sequences based on the working condition adaptation timing arbitration rules. Before the control instructions are officially issued, a pneumatic conveying digital twin model pre-test is performed and verified. After outputting an executable control instruction package, it is sent to the corresponding execution equipment. Acquire actual equipment operation feedback, pipe segment sensor measured data and command settings, construct triplet time series data to calculate execution deviation, mark abnormal execution equipment in combination with preset allowable deviation range, and output execution feedback dataset; Acquire real-time data of the entire pipeline operating conditions from sensors, combine it with the execution feedback dataset, update the pipeline virtual state tensor, and verify the pipeline section's transport capacity margin, blockage risk level, and processing quality indicators.
[0015] Specifically, the steps to trigger global scroll re-optimization include: When the differential pressure of any pipe section, material flow rate, or processing parameters of any workstation continuously deviates from the preset safety threshold, a global rolling re-optimization is triggered. Using the full pipeline operating condition sensor measured time series data, the updated pipeline virtual state tensor and equipment execution status as the initial optimization benchmark, the over-limit pipe sections and abnormal work positions are located, and the upstream and downstream related pipe sections are linked to form a global optimization boundary. The non-dominated sorting genetic algorithm is used again for global optimization. The optimization criteria are based on multiple objectives: minimizing pipeline blockage risk, minimizing particulate material wear rate, achieving mixing uniformity, ensuring compliant discharge dust content, and minimizing fan air volume energy consumption. The preset safety threshold serves as a hard constraint. The pneumatic conveying digital twin model is used to calculate the optimization objective values and constraint violations. After iterative convergence, the entropy weight TOPSIS method is used to select the comprehensive optimal solution from the Pareto optimal solution set and generate the updated control instruction set.
[0016] The beneficial effects of this invention are: 1. This invention constructs a digital twin model of pneumatic conveying, abstracting the closed conveying pipeline into a gas-solid two-phase flow unit and setting adjustable parameter nodes. Combining the hydrodynamic mechanism parameters of the gas-solid two-phase flow, it uses an online ensemble learning algorithm to dynamically correct the initial mechanism parameter set and outputs the pipeline virtual state tensor. Using positive pressure dilute phase conveying mode and negative pressure dilute phase conveying mode as discrete conveying mode decision variables, and the control parameters corresponding to the adjustable parameter nodes as continuous control variables, it performs global optimization through a non-dominated sorting genetic algorithm with an elite strategy, realizing the joint optimization of conveying mode and control parameters. It solves the problem of lack of unified modeling and collaborative scheduling logic for multiple working conditions such as positive pressure, negative pressure, dilute phase, and dense phase in traditional schemes, enabling the bulk material handling system to adaptively select the optimal conveying mode and parameter combination according to real-time working conditions to adapt to the flow characteristics of gas-solid two-phase flow in the pipe.
[0017] 2. This invention calculates the supply-demand deviation of each pipe segment by performing mass flow conservation calculations. It then calculates the conveying capacity margin of each pipe segment by combining the upper limit of material mass flow rate with the current actual conveying rate. Global optimization is performed using multiple objectives: minimizing pipeline blockage risk, minimizing particulate material wear rate, and minimizing fan airflow energy consumption. By verifying the conveying capacity margin, blockage risk level, and processing quality indicators of each pipe segment, global rolling re-optimization is triggered when any pipe segment pressure difference, material flow rate, or workstation processing parameters deviate from the safety threshold. This achieves joint control and closed-loop optimization of multiple variables such as fan airflow, pipeline material-to-air ratio, blockage risk, and workstation material flow rate. This fundamentally solves the technical problems of frequent pipeline blockage, severe particulate wear, and high fan energy consumption, improving the operational stability and energy efficiency of the pneumatic conveying system. Attached Figure Description
[0018] Figure 1 A schematic diagram of a bulk material handling scheduling optimization method based on digital twins; Figure 2 This is a flowchart illustrating the construction of a digital twin model for pneumatic conveying in this invention; Figure 3 This is a flowchart illustrating the dynamic correction of the initial mechanism parameter set in this invention; Figure 4 This is a flowchart of the parallel gas-solid two-phase flow simulation deduction in this invention; Figure 5 This is a flowchart of the output control instruction set in this invention. Detailed Implementation
[0019] 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.
[0020] Example refer to Figures 1 to 5 As shown in the figure, this embodiment introduces a method for optimizing the scheduling of bulk material handling based on digital twins, including the following steps: Step 1: Multi-source heterogeneous data is acquired by deploying gas flow meters, pipeline differential pressure measuring points, fan pressure measuring points, pipeline temperature measuring points, cooler temperature measuring points, blending bin level sensors, powder removal station flow detectors, and weighing modules in the positive pressure dilute phase feed pipeline, multi-bin blending pipeline, powder removal processing pipeline, and discharge conveying pipeline. Addressing the issues of asynchronous timestamps, inconsistent spatial references, and airflow pressure noise interference from the multi-source heterogeneous data, a spatiotemporal adaptive interpolation alignment algorithm is employed. Five physical constraints are set, including pipeline pressure, fan airflow, material mass flow rate, particulate material breakage rate, and pipeline temperature. This continuousizes the discrete material monitoring signals and maps them to a unified pipeline mileage coordinate system. Kalman filtering is used to eliminate pressure and material mass flow signal noise caused by air pressure shocks, forming a comprehensive air supply system covering the Roots blower cooling. The system includes a multi-dimensional time-series data stream with consistent temporal and spatial dimensions, encompassing feeding buffer, multi-compartment gravity mixing, pipeline flow rate and diameter adjustment, vertical-horizontal sifting and dust removal, and unloading throughout the entire pipeline. This multi-source heterogeneous data includes, but is not limited to, fan volume, supply air pressure, pipeline temperature, cooler outlet air temperature, pressure difference across different pipeline sections, fan operating status indicators, instantaneous material mass flow rate, material particle size and breakage rate, process additive dosage, reversing valve status in each mixing compartment, dust concentration at the dust removal station, and flow rate parameters in the pipeline diameter adjustment section. The positive or negative pressure conveying mode is determined by the selected fan type; in this embodiment, a magnetic levitation fan is used for positive pressure conveying, and a vacuum pump is used for negative pressure conveying. The instantaneous material mass flow rate of the solid material is calculated by the weighing module in conjunction with the feeding time. The material particle breakage rate is obtained through offline sampling and detection. The specific process of the spatiotemporal adaptive interpolation alignment algorithm includes: using the system master clock as a unified time reference, performing linear interpolation to complete the raw data asynchronously collected by various sensors, and unifying the sampling time step to eliminate timestamp asynchrony; then mapping the measuring points distributed in different locations to a unified pipeline mileage coordinate to complete spatial reference normalization; substituting five physical constraints such as pipeline pressure, fan air volume, material mass flow rate, particulate material breakage rate, and pipeline temperature to perform piecewise adaptive interpolation, converting discrete material monitoring signals into continuous signals and mapping them to a unified pipeline mileage coordinate to complete spatiotemporal dimension data fusion; subsequently, combining Kalman filtering to remove pressure and material mass flow rate signal noise caused by air pressure impact, and finally generating a spatiotemporally consistent multidimensional time series data stream; Step 2: Obtain historical operating condition characteristics from the historical pneumatic conveying database, and construct a digital twin model of pneumatic conveying by combining multi-dimensional time-series data streams and gas-solid two-phase flow hydrodynamic mechanism parameters. Abstract the closed conveying pipeline into a gas-solid two-phase flow unit with particle settling, pipe wall friction loss, differentiation between positive pressure pushing and negative pressure suction flow directions, and compatibility with pipeline diameter change and deceleration characteristics. Set the fan, feeding device, cooler, multi-compartment gravity mixing station, pipeline flow velocity control node, vertical and horizontal sifting and powder removal processing station, and unloading mechanism as adjustable parameter nodes; utilize online data collection... The learning algorithm continuously compares the pipeline differential pressure, material conveying rate, pipeline flow velocity predicted by the pneumatic conveying digital twin model with the actual measured data to calculate the comprehensive prediction error index, dynamically corrects the initial mechanism parameter set, and outputs the pipeline virtual state tensor to characterize the gas-material distribution of the entire pipeline, the pneumatic conveying operating condition type, the pipe blockage risk level, the material balance in the mixing bin, and the dust content at the dust removal station. The initial mechanism parameter set includes the pipe wall friction resistance coefficient, particle settling coefficient, pipeline gas-material loss coefficient, pipeline pressure loss under different conveying modes, and flow velocity loss coefficient of variable diameter pipeline. Step 3: Based on the pipeline virtual state tensor, extract the material conveying rate, upper limit of material mass flow rate, downstream gravity mixing bin, and material consumption rate of the sifting and de-dusting station for each pipeline segment. Establish the material mass flow supply and demand balance equation for each pipeline segment and calculate the conveying capacity margin of each pipeline segment. Automatically mark abnormal pipeline segments with high blockage risk, excessive or insufficient material flow rate using a spatiotemporal trajectory density clustering algorithm. For abnormal pipeline segments, call the pneumatic conveying digital twin model to perform parallel gas-solid two-phase flow simulation and deduction, with the pipeline having the lowest blockage risk. The optimization criteria are: minimizing the wear rate of particulate materials, achieving the standard of mixing uniformity, complying with the dust content of the discharged material, and minimizing the energy consumption of the blower. Positive pressure dilute phase conveying mode or negative pressure dilute phase conveying mode is selected as needed. Global optimization is performed on adjustable parameters such as blower output air volume, air supply pressure, pipeline material-to-air ratio, feed rate, cooler working power, pipeline diameter change opening, and start-stop sequence of the mixing chamber reversing valve. The output includes a set of control instructions for blower air volume, pipeline material-to-air ratio, conveying mode switching, and operating parameters of each bulk material processing station. Step 4: Send the control command set to the execution equipment such as fans, coolers, rotary valves, mixing bin reversing valves, and dust removal devices. In real time, acquire the differential pressure, material flow rate, fan status indicators, material level in the bin, discharge dust concentration, and pipeline flow rate data of the entire pipeline after executing each control command. Simultaneously update the pipeline virtual state tensor of the pneumatic conveying digital twin model, and verify the conveying capacity margin, blockage risk level, and processing quality indicators of the mixing and dust removal processes of the pipeline section. When the differential pressure, material flow rate, or processing parameters of any pipeline section deviate from the safety threshold, trigger global rolling re-optimization.
[0021] Specifically, the steps for outputting the pipeline virtual state tensor in step 2 include: Extract current operating condition features from multidimensional time-series data streams using a fixed-length sliding time window, including but not limited to pipeline pressure change rate, pipeline differential pressure gradient, fan airflow fluctuation amplitude, material mass flow rate stability, particle breakage rate change trend, pipeline temperature gradient, cooler temperature change, blending bin material level change rate, and dust concentration change features at the dust removal station. After normalizing the current operating condition characteristics, a principal component analysis algorithm with physical constraints regularization is used for dimensionality reduction. The physical constraints include five factors: pipeline pressure, fan airflow, material mass flow rate, breakage rate of particulate material, and pipeline temperature. The cumulative variance contribution rate is calculated based on the ratio of the eigenvalue of each principal component to the sum of all eigenvalues. Principal components with a cumulative variance contribution rate greater than or equal to a contribution rate threshold are retained, resulting in the operating condition time window feature vector. The fixed length is determined jointly based on the pressure fluctuation period of the gas-solid two-phase flow and the material conveying response time, i.e., 2~ of the pressure fluctuation period. The lower limit of the sliding time window is 3 times the maximum length of the sliding time window, and the upper limit is the longest conveying response time of the material from the feed point to the discharge point. The value is a typical value within the intersection of the upper and lower limits, with a default value of 30 seconds. It can be adaptively adjusted according to the actual pipeline conveying cycle. That is, when the pressure fluctuation cycle or the material conveying response time changes, the length of the sliding time window is scaled proportionally according to the change ratio of the two, and limited to between the upper and lower limits. The contribution rate threshold is determined based on the principal component cumulative variance contribution rate distribution of the normal fluctuation range of historical operating conditions, and is set to 95%. Historical operating condition features are obtained from the historical pneumatic conveying operating condition database. The operating condition time window feature vector is then matched with the historical operating condition features using a weighted similarity algorithm on a unified pipeline mileage coordinate and time scale to obtain the comprehensive weighted similarity between each historical operating condition feature and the operating condition time window feature vector. Historical operating condition feature vectors are extracted from the historical pneumatic conveying operating condition database using the same sliding time window extraction, normalization, and physical constraint regularization principal component analysis dimensionality reduction process as the current operating condition time window feature vector. The weighted similarity matching process includes: firstly, mapping all historical operating condition features to a unified pipeline mileage coordinate and time step consistent with the current time window; secondly, splitting the current operating condition time window feature vector and each historical operating condition feature vector into four dimensional sub-vectors according to pressure, flow rate, temperature, and material state; and thirdly, for each category, taking the current operating condition time window feature vector... The cosine similarity between the subvectors of the eigenvectors under this class and the subvectors of the historical operating condition feature vectors under this class is calculated to obtain the cosine similarity values of the historical operating condition sample in four dimensions: pressure, flow, temperature, and material state. Then, the importance of the five physical constraints is combined to assign dimension weights, and the cosine similarity values of the four dimensions are multiplied by the corresponding dimension weights and summed to obtain the comprehensive weighted similarity of the historical operating condition sample. This process is repeated for all historical operating condition samples to obtain the comprehensive weighted similarity between each historical operating condition feature and the operating condition time window feature vector. Among them, the historical pneumatic conveying operating condition database is constructed by acquiring multi-dimensional time-series data during historical operation, performing spatiotemporal adaptive interpolation alignment and Kalman filtering noise reduction, and then associating it with the corresponding operating condition type identifier and the gas-solid two-phase flow mechanism parameters after online correction. The historical operating condition samples are sorted in descending order based on their comprehensive weighted similarity. The top N historical operating condition samples with the highest comprehensive weighted similarity are selected. Initial operating condition prior labels are determined through similarity-weighted voting. These labels serve only as initialization constraints for the digital twin model and not as final output constraints. The initial operating condition prior labels include positive pressure dilute phase push, negative pressure suction, positive pressure dense phase push, variable diameter speed regulation, multi-compartment gravity mixing, vertical precipitator dust removal, and horizontal precipitator dust removal. Here, N is the configurable number of nearest neighbor historical operating condition samples, determined based on the total number of samples and the number of operating condition categories in the historical pneumatic conveying operating condition database. The ratio of the total number of samples to the number of work condition categories is used to obtain the basic nearest neighbor number by taking the square root of the ratio and rounding it to the nearest integer. The basic nearest neighbor number is then limited to a preset lower limit and an upper limit, with the lower limit being 5 and the upper limit being 20. If the basic nearest neighbor number is less than 5, then N is 5; if it is greater than 20, then N is 20; otherwise, N is the basic nearest neighbor number. The preset lower limit is based on the minimum number of nearest neighbors required to ensure that the prior label of a single work condition receives at least a majority of support in similarity-weighted voting, thus avoiding random fluctuations in voting results due to too few nearest neighbors. The preset upper limit is based on the decay characteristics of the comprehensive weighted similarity of historical work condition samples and the computational cost of the digital twin initialization constraints. After determining the initial working condition prior label, the corresponding historical mechanism parameter values are obtained from the historical pneumatic conveying working condition database of the high similarity historical samples of the corresponding prior label. The local weighted average is performed with the comprehensive weighted similarity as the weighting coefficient to generate an initial mechanism parameter set including the pipe wall friction resistance coefficient, particle settling coefficient, pipeline material and gas loss coefficient, positive pressure push mode pressure loss coefficient, negative pressure suction mode pressure loss coefficient, and variable diameter pipeline flow velocity loss coefficient. Based on the initial mechanism parameter set and the characteristic vector of the working condition time window, combined with the hydrodynamic mechanism parameters of gas-solid two-phase flow, a digital twin model of pneumatic conveying is constructed. According to the prior labels of the initial working condition, the boundary conditions and initial parameter values of the gas-solid two-phase flow model under the corresponding conveying mode are called. The closed conveying pipeline is abstracted into a gas-solid two-phase flow unit with particle settling, pipe wall friction loss, distinguishing between positive pressure pushing and negative pressure suction flow directions, and compatible with pipeline diameter change and deceleration characteristics. The fan, feeding device, cooler, multi-compartment gravity mixing station, pipeline flow velocity control node, vertical and horizontal sifting and powder removal processing station, and unloading mechanism are set as adjustable parameter nodes. The comprehensive prediction error index is calculated by continuously comparing the pipeline differential pressure, material conveying rate, pipeline flow velocity, and field measured data predicted by the pneumatic conveying digital twin model with those of the online ensemble learning algorithm. The calculation process is as follows: For each observed variable, namely pipeline differential pressure, material conveying rate, and pipeline flow velocity, the difference between the predicted value of each observed variable and the field measured data is calculated. The difference is divided by the measurement uncertainty of each observed variable, normalized, and squared to obtain the univariate normalized error square term. The univariate normalized error square term is multiplied by the weight of the influence of each observed variable on the pipe blockage state to obtain the univariate weighted error term. The sum of the univariate weighted error terms of all observed variables is divided by the total number of observed variables to obtain the average weighted error term. The product of the parameter adjustment norm and the regularization coefficient is calculated. Finally, the average weighted error term is summed with the product to obtain the comprehensive prediction error index of the pneumatic conveying digital twin model. The online ensemble learning algorithm employs an online ensemble learner, which is an ensemble model that updates the weights of the base learners sequentially using the data stream: multiple base learners are trained using the feature vector of the working condition window, the initial mechanism parameter set, the residuals between the digital twin model predictions and the actual measured values as online samples; the fusion weights are updated exponentially based on the squared recent residuals of each base learner, and the residuals of the observed variables are output; the comprehensive prediction error index is calculated based on the residuals of the observed variables; and the contribution of each mechanism parameter is solved in reverse through online gradient descent using this comprehensive prediction error index as the loss. The initial mechanism parameter set is dynamically corrected based on the comprehensive prediction error index. The initial mechanism parameter set includes the pipe wall friction resistance coefficient, particle settling coefficient, pipeline material and gas loss coefficient, positive pressure push mode pressure loss coefficient, negative pressure suction mode pressure loss coefficient, and variable diameter pipeline flow velocity loss coefficient. The pipeline virtual state tensor is output to characterize the gas and material distribution of the entire pipeline, the pneumatic conveying working condition type, the pipe blockage risk level, the material balance in the mixing bin, and the dust content at the dust removal station. The process of dynamically correcting the initial mechanism parameter set is as follows: With the goal of minimizing the comprehensive prediction error index, an online ensemble learner is used to inversely solve for the contribution of each mechanism parameter in the initial mechanism parameter set to the total error. An adaptive correction step size is set based on the fluctuation level of the current operating conditions, and incremental iterative updates are performed on each mechanism parameter. Specifically, the incremental iterative update process is as follows: using the sensitivity of the comprehensive prediction error index to each current mechanism parameter as the parameter gradient, that is, calculating the partial derivatives of the comprehensive prediction error index with respect to the pipe wall friction resistance coefficient, particle settling coefficient, pipeline material-gas loss coefficient, positive pressure push mode pressure loss coefficient, negative pressure suction mode pressure loss coefficient, and variable diameter pipeline flow velocity loss coefficient, respectively, to obtain the gradient direction and gradient magnitude corresponding to each mechanism parameter; and then applying these parameters... The gradient magnitudes at the current moment are sorted in descending order. The three mechanism parameters with the largest gradient magnitudes are updated first. Mechanism parameters with gradient magnitudes less than a small threshold are not updated temporarily. The small threshold is determined based on the difference between the mean and standard deviation of the gradient magnitudes of each mechanism parameter, and is set to the difference between the mean gradient magnitude and 0.5 times the standard deviation. During each update, the adjustment amount of the mechanism parameter is equal to the current gradient direction multiplied by the adaptive correction step size. The constraint condition is that the adjusted mechanism parameter value does not exceed the upper and lower limits allowed by the physical mechanism of gas-solid two-phase flow. After the update is completed, the comprehensive prediction error index is recalculated. If the comprehensive prediction error index decreases, the update result is retained; otherwise, it reverts to the mechanism parameter value before the update and halves the correction step size. The above process is repeated until the maximum number of iterations is reached. When the comprehensive prediction error index for three consecutive sliding time windows is lower than the steady-state threshold, the correction step size is reduced to enter the steady-state fine-tuning mode. The steady-state fine-tuning mode reduces the correction step size to 10%~30% of the step size in normal operation mode, decreases the parameter update frequency from once per sliding time window to once every 3~5 sliding time windows, and limits the upper limit of each parameter adjustment to 50% of that in normal operation mode to suppress high-frequency parameter oscillations. When the comprehensive prediction error index continuously exceeds the steady-state threshold and the operating condition characteristics shift, weighted similarity matching is performed again, the initial mechanism parameter set is updated, and then correction is executed. The steady-state threshold is determined based on the statistical distribution of the comprehensive prediction error under normal system operation conditions, and is set to the 90th percentile of the comprehensive prediction error under normal operation conditions. The shift in operating condition characteristics refers to the shift in the operating condition window of the current sliding time window. The feature vector and the baseline operating condition feature corresponding to the most recent steady-state confirmation or successful matching, under the unified pipeline mileage coordinates and time scale, have a comprehensive weighted similarity of the four dimensions of sub-vectors (pressure, flow, temperature, and material state) that is lower than the offset similarity threshold. Alternatively, any one of the following features—pipeline pressure change rate, pipeline differential pressure gradient, fan airflow fluctuation amplitude, material mass flow stability, particle breakage rate change trend, pipeline temperature gradient, cooler temperature change, blending bin material level change rate, and dust concentration change at the dust removal station—exceeds the normal fluctuation range for 3 to 5 consecutive sliding time windows and reaches the offset amplitude threshold. The offset similarity threshold is determined based on the statistical distribution of the comprehensive weighted similarity of historical normal operating conditions, and can be taken as the mean of the comprehensive weighted similarity of normal operating conditions minus 2 times the standard deviation. The offset amplitude threshold is taken as 1.2 to 1.5 times the upper or lower limit of the normal fluctuation range.
[0022] Specifically, step 3, which outputs a set of control instructions including fan air volume, pipeline material-to-air ratio, conveying mode switching, and operating parameters of each bulk material processing station, includes the following steps: Based on unified pipeline mileage coordinates, the feed inlet, pipeline diameter change section, mixing bin access point, and the inlet and outlet of the sifting and de-powdering station are used as boundary nodes. At the same time, local densification processing is carried out at elbows, diameter change sections, and before and after the de-powdering station. The entire closed conveying pipeline is discretized into multiple continuous independent pipe sections. A one-to-one mapping relationship between pipe section identifiers and unified pipeline mileage coordinates is established. The geometric parameters, upstream and downstream connection relationships, and corresponding processing stations of each pipe section are associated with the pipe section identifiers to form a pipe section topology relationship for state calculation and control positioning. The material conveying rate, upper limit of material mass flow rate, and material consumption rate of downstream gravity mixing bin and sludge removal station are extracted from the pipeline virtual state tensor. The upper limit of material mass flow rate can be calculated from the effective conveying cross-sectional area of the pipeline section, the maximum allowable apparent gas velocity, and the particle phase volume fraction. The maximum allowable apparent gas velocity is selected as the smaller value between the blockage critical gas velocity and the wear critical gas velocity. That is, the upper limit of material mass flow rate is obtained by multiplying the particle density, particle phase volume fraction, effective conveying cross-sectional area of the pipeline section, and maximum allowable apparent gas velocity. A material mass flow rate supply and demand balance equation for each pipe segment is established. The process is as follows: the material mass flow rate at the inlet is taken as the supply, and the material consumption rate of the downstream gravity mixing bin, the material consumption rate of the sifting and de-powdering station, and the material demand of adjacent downstream pipe segments are taken as the demand. The supply and demand are aligned on a unified time scale, and mass flow rate conservation calculations are performed according to the pipe segment topology to obtain the material supply and demand for each pipe segment. The supply and demand deviation of each pipe segment is obtained by subtracting the corresponding material demand from the supply. If the supply and demand deviation is positive, it indicates that the material supply entering the pipe segment or downstream node exceeds the current demand; if the supply and demand deviation is negative, it indicates that the material supply cannot meet the current demand. During the calculation, the equation is iterated segment by segment according to the topological hierarchy, eliminating overlapping statistical items between station consumption and downstream pipe segment flow to ensure the uniqueness of the mass conservation boundary. The difference between the upper limit of material mass flow rate of the pipeline segment and the current actual conveying rate is calculated. The difference is divided by the upper limit of material mass flow rate to obtain the conveying capacity margin of each pipeline segment. The conveying capacity margin and the supply-demand deviation of the pipeline segment are used to characterize the current conveying status of the pipeline segment. The conveying capacity margin is used to characterize the remaining space of the pipeline segment from the maximum stable conveying capacity, and the supply-demand deviation is used to characterize the degree of matching between the material supply of the pipeline segment and the material demand of the downstream. The spatiotemporal trajectory density clustering algorithm automatically identifies abnormal pipe sections with high blockage risk and excessive or insufficient material flow velocity. The specific process of the algorithm is as follows: using unified pipeline mileage coordinates and timestamps for each pipe section as spatiotemporal positioning information, and using conveying capacity margin, supply-demand deviation of the pipe section, material conveying rate, material flow velocity, pipeline pressure difference, and pipeline pressure difference gradient as state features; setting a continuous sliding time window with a fixed duration and step size, arranging the state features of the pipe section at each sampling moment in chronological order within each sliding time window to form a state trajectory segment of the pipe section within the current sliding time window; and splicing the state trajectory segments of adjacent sliding time windows in chronological order to obtain the spatiotemporal state trajectory of the pipe section; and then processing the spatiotemporal state... The trajectory is normalized, and a density-based clustering method is used to identify state clusters and outlier state points with similar spatiotemporal evolution characteristics. Based on the state characteristics corresponding to each state cluster, state clusters with continuously decreasing conveying capacity margin and abnormally increasing pipeline pressure differential are marked as high-blockage-risk pipe sections; state clusters with continuously positive supply-demand deviation and material flow rate exceeding the allowable range of the corresponding operating condition are marked as material flow rate excess pipe sections; state clusters with continuously negative supply-demand deviation and material flow rate below the allowable range of the corresponding operating condition are marked as material flow rate insufficient pipe sections, and these are integrated to form an abnormal pipe section set; at the same time, single outlier state points are integrated into an early warning observation sequence, and are not directly judged as abnormal pipe sections to avoid misjudgment due to instantaneous fluctuations; For abnormal pipe sections, a pneumatic conveying digital twin model is used to perform parallel gas-solid two-phase flow simulation. The abnormal pipe section is the core, and the upstream and downstream related pipe sections and corresponding workstations are combined to construct the simulation pipe section. The current material conveying rate, pipeline pressure, pipeline pressure difference, particle phase volume fraction, pipeline temperature and the real-time operating parameters of the abnormal pipe section and its upstream and downstream related pipe sections are used as simulation boundary conditions. The gas-solid two-phase flow dynamics simulations under positive pressure dilute phase conveying mode and negative pressure dilute phase conveying mode are performed in parallel. The simulation results of each simulation pipe section are output, including but not limited to pipeline pressure difference, pipeline pressure difference gradient, blockage risk index, particle material breakage and wear rate, mixing uniformity index, outlet dust content, fan air volume and energy consumption, equipment operating parameters, minimum stable conveying speed, maximum allowable material concentration, lower limit of mixing uniformity and upper limit of outlet dust content. Using the lowest pipeline blockage risk, lowest particulate material wear rate, qualified blending uniformity, compliant discharge dust content, and lowest fan air volume energy consumption as multi-objective optimization criteria, positive pressure dilute phase conveying mode and negative pressure dilute phase conveying mode are used as discrete conveying mode decision variables. The control parameters corresponding to the adjustable parameter nodes, including fan output air volume, air supply pressure, pipeline material-to-air ratio, feed rate, cooler working power, pipeline diameter change opening, and start-stop sequence of the mixing chamber reversing valve, are used as continuous control variables. The allowable pressure range of the pipeline, the rated operating range of the equipment, the minimum stable conveying speed, the maximum allowable material concentration, the lower limit of blending uniformity, and the upper limit of discharge dust content are used as optimization constraints for global optimization. The global optimization process is as follows: A non-dominated sorting genetic algorithm with an elitist strategy is used for global optimization. First, an initial population is generated based on the current operating parameters, positive pressure dilute phase transport mode, and negative pressure dilute phase transport mode. Discrete transport mode decision variables and continuous control variables are mixed and encoded. The individual parameters of each generation of the population are mapped to simulation input conditions. The pneumatic transport digital twin model is called to solve for each optimization objective value and constraint violation quantity. Specifically, the pipeline pressure differential gradient and blockage risk index in the simulation results are mapped to the pipeline blockage risk target value, the particulate material crushing and wear rate is mapped to the particulate material crushing and wear target value, and the blending uniformity index is mapped to the blending uniformity target value. The dust content at the outlet is mapped to the target value of the dust content at the discharge, and the fan air volume and energy consumption are mapped to the target values of the fan air volume and energy consumption. At the same time, the pipeline pressure, equipment operating parameters, minimum stable conveying speed, maximum allowable material concentration, lower limit of mixing uniformity, and upper limit of discharge dust content in the simulation results are compared with the optimization constraints to obtain the constraint violation amount. Elite individuals are screened through non-dominated sorting and crowding calculation. The next generation population is generated through selection, crossover, and mutation operations. After iterating to the maximum number of iterations, the Pareto optimal solution set is obtained. The comprehensive optimal solution is screened from the Pareto optimal solution set based on the entropy weight TOPSIS method to determine the optimal conveying mode and corresponding control parameter combination. The final output includes a set of control instructions for fan air volume, pipeline material-to-air ratio, conveying mode switching, and operating parameters of each bulk material processing station. The control instruction set is mapped one-to-one with the pipe segment topology and station number, and includes parameter adjustment step size and execution timing constraints to ensure a smooth control process without operational shocks.
[0023] Specifically, the steps in step 4 that trigger global scroll re-optimization include: Based on the control characteristics of all process execution equipment such as fans, coolers, rotary valves, mixing bin reversing valves, and sifting and dust removal devices, as well as the impact characteristics of pneumatic conveying conditions, the control command set is subjected to hierarchical adaptation processing to clarify the command execution rules for each execution equipment and avoid the risk of disturbance under strong impact conditions. The control command set is broken down into three priority command subsets: a safety-level discrete command subset, a process-level continuous command subset, and an energy efficiency-level adjustment command subset. The safety-level discrete command subset includes on / off commands that are prone to causing pipeline pressure surges, such as switching conveying conditions, switching the reversing valve in the blending bin, and switching the sludge removal station. The process-level continuous command subset includes core process analog commands that can be smoothly adjusted, such as adjusting fan airflow and pipeline material-to-air ratio. The energy efficiency-level adjustment command subset includes quality and energy efficiency adjustment commands that do not affect pipeline safety, such as adjusting the staged wind speed of the sludge removal device and adjusting the sludge removal intensity. Based on a three-level priority instruction subset, and combined with the current pipeline blockage risk level, pipeline topology and equipment executable status, a working condition adaptation timing arbitration rule is constructed. For high-risk blockage working conditions, pressure relief and material reduction safety instructions are executed first, while process adjustment instructions such as production increase and increasing pipeline material-to-gas ratio are temporarily suspended to avoid pressure shocks and blockage failures during working condition switching. Arbitration generates equipment instruction sequences based on the working condition adaptation timing arbitration rules. Before the official issuance of control instructions, a pre-simulation verification of the pneumatic conveying digital twin model is performed: using single-equipment control instructions as simulation input, the virtual state tensor of the current pneumatic conveying pipeline is linked with the operating status of each executing device. The simulation predicts the working condition parameters such as pressure difference in each pipe section, material flow rate, pipeline fan operating status indicators, mixing bin material level, and discharge dust concentration throughout the entire execution cycle. The safe variation range is determined comprehensively based on the rated operating parameters of the executing devices, pipeline safety design parameters, and on-site steady-state operating conditions. If all... If all predicted operating condition variables are below the corresponding safe change range, the operation is deemed safe and an executable control instruction package with a unique identifier is generated. If any predicted operating condition variable is greater than or equal to the corresponding safe change range, the execution timing of the control instructions is adjusted first, the single-step adjustment range is reduced, and the simulation is re-verified. The process is iteratively optimized until the safety requirements are fully met. Finally, an executable control instruction package containing the equipment number, action type, instruction target value, action start time, transition time, execution sequence, and equipment interlocking conditions is output to avoid pipeline safety disturbances caused by the issuance of control instructions from the source. The industrial control network distributes executable control command packages to corresponding execution devices such as fans, coolers, rotary valves, mixing chamber reversing valves, and dust removal devices. Using the execution interval of a single control command as a fixed sampling period, it simultaneously acquires three core types of information: actual equipment operation feedback, pipeline segment sensor measurement data, and command setpoints. This constructs a unified, aligned, and standardized time-series data set of command setting, equipment execution, and sensor measurement, enabling comprehensive dynamic monitoring of the execution process. For each execution device, the execution deviation between the actual executed value and the command target value is calculated. The actual executed value is acquired in real-time via industrial fieldbus and built-in sensors, while the command target value is directly retrieved from the executable control command package. The allowable deviation range is determined based on equipment control accuracy and the allowable fluctuation range of the process. Execution devices whose execution deviation exceeds the allowable deviation range and fails to converge to the allowable deviation range for multiple consecutive sampling periods are marked as abnormal execution devices. Finally, the execution feedback dataset is output, including the actual execution parameters, execution deviation, response time, and abnormal execution devices for each device. Acquire real-time data of the entire pipeline operating conditions from sensors, and synchronously update the pipeline virtual state tensor of the pneumatic conveying digital twin model by combining it with the execution feedback dataset. Iterate and update virtual parameters such as gas-solid two-phase distribution, material flow rate, positive or negative pressure status of the pipeline, material balance in the silo, and dust concentration for each pipeline segment. Based on the updated pipeline virtual state tensor, verify the conveying capacity margin, blockage risk level, and processing quality indicators of the pipeline segment. Among them, the processing quality indicators include, but are not limited to, the mixing uniformity of the mixing silo and the discharge dust concentration of the sifting and dust removal device. When any pipe section pressure difference, material flow rate, or station processing parameter is detected to continuously deviate from the safety threshold, global rolling re-optimization is immediately triggered. The safety threshold is determined based on equipment design limits, process safety regulations, and historical safe operating ranges. The safety threshold for pipe section pressure difference is the smaller value between the maximum allowable working pressure of the pipe section and the critical pressure difference for blockage. The safety threshold for material flow rate is the allowable range between the minimum stable conveying speed and the critical wear gas speed. The safety threshold for station processing parameters is the allowable deviation range of the process regulations. The global rolling re-optimization process is as follows: using the actual measured time-series data of the entire pipeline operating conditions, the updated pipeline virtual state tensor, and the equipment execution state as the initial optimization benchmark, the out-of-limit pipe sections and abnormal work positions are located, and the upstream and downstream related pipe sections are linked to form a global optimization boundary. The non-dominated sorting genetic algorithm is used again for global optimization. The optimization criteria are the lowest pipeline blockage risk, the lowest particulate material wear rate, the standard of mixing uniformity, the compliance of discharge dust content, and the lowest fan air volume energy consumption. The safety threshold is used as a hard constraint. The pneumatic conveying digital twin model is used to simultaneously calculate the optimization objective values and constraint violations. The initial population prioritizes the current optimal solution and the previous cycle's optimal solution. The remaining individuals are generated by perturbation around the current optimal solution. After iterative convergence, the entropy weight TOPSIS method is used to select the comprehensive optimal solution from the Pareto optimal solution set to generate an updated control instruction set. The updated control instruction set is accompanied by execution time constraints.
[0024] In summary, this invention aims to address the technical problems of existing bulk material handling system scheduling optimization methods that fail to model and design collaborative scheduling logic for pneumatic conveying processes under multiple operating conditions such as positive pressure, negative pressure, dilute phase, and dense phase. These methods also fail to integrate fan airflow, pipeline material-to-air ratio, conveying mode, pipe blockage risk, and workstation material flow rate for joint control, making it difficult to adapt to the flow characteristics of gas-solid two-phase flow within the pipe. This results in frequent pipeline blockage, severe particle wear, and high fan energy consumption. The invention constructs a spatiotemporally consistent multidimensional time-series data stream through spatiotemporal adaptive interpolation alignment and Kalman filtering. It then builds a digital twin model of pneumatic conveying based on the hydrodynamic mechanism parameters of gas-solid two-phase flow. Based on the virtual state tensor of the pipeline, it extracts the material conveying rate of each pipe segment and calculates the conveying capacity margin. Spatiotemporal trajectory density clustering is used to automatically mark abnormal pipe segments with high blockage risk, excessive or insufficient material flow rate. The goal is to achieve multiple optimization objectives, including minimizing pipeline blockage risk, minimizing particle breakage and wear rate, meeting mixing uniformity standards, ensuring compliant discharge dust content, and minimizing fan energy consumption. The invention employs a non-dominated sorting genetic algorithm with an elite strategy and the entropy-weighted TOPSIS method for global optimization, outputting a set of control instructions including fan airflow, pipeline material-to-air ratio, conveying condition switching, and operating parameters of each bulk material processing station. Execution safety is ensured through hierarchical instruction adaptation and digital twin pre-simulation verification. The pipeline virtual state tensor is synchronously updated based on the execution feedback dataset and real-time data from full-pipeline sensing. When the differential pressure, material velocity, or processing parameters of any pipe section continuously deviate from the safety threshold, global rolling re-optimization is triggered. A closed-loop evaluation of the accuracy of the pneumatic conveying digital twin model and adaptive correction of mechanism parameters are conducted using a comprehensive prediction error index, forming a complete iterative mechanism of modeling, scheduling, optimization, and rolling re-optimization. This invention is adaptable to non-stationary and multi-type coupled interference scenarios in multi-condition pneumatic conveying systems, improving the ability to suppress pipe blockage risks and the fidelity of particulate material quality, reducing fan energy consumption, and enhancing the operational stability and control accuracy of bulk material handling systems under varying operating conditions.
[0025] 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 optimizing the scheduling of bulk material handling based on digital twins, characterized in that, include: Multi-source heterogeneous data is acquired, and a spatiotemporal adaptive interpolation alignment algorithm is used with five physical constraints to form a multi-dimensional time-series data stream. A digital twin model of pneumatic conveying is constructed, which abstracts the closed conveying pipeline into a gas-solid two-phase flow unit, sets adjustable parameter nodes, calculates the comprehensive prediction error index, dynamically corrects the initial mechanism parameter set, and outputs the pipeline virtual state tensor. Calculate the transport capacity margin of each pipe section, mark abnormal pipe sections and perform gas-solid two-phase flow simulation in parallel, set multi-objective optimization criteria, select positive pressure dilute phase transport mode or negative pressure dilute phase transport mode as needed for global optimization, and output control command set. The control command set is sent to each execution device to update the pipeline virtual state tensor and verify the conveying capacity margin, blockage risk level and processing quality indicators of the pipeline section; when the differential pressure, material flow rate or processing parameters of any pipeline section deviate from the safety threshold, global rolling re-optimization is triggered.
2. The bulk material handling scheduling optimization method based on digital twins according to claim 1, characterized in that, The specific steps for constructing a digital twin model of pneumatic conveying include: The current operating condition features are extracted from the multidimensional time-series data stream, and after normalization, the dimensionality is reduced by the principal component analysis algorithm with physical constraint regularization to obtain the operating condition time window feature vector. Historical operating condition features are obtained, and the operating condition time window feature vector is matched with the historical operating condition features on a unified pipeline mileage coordinate and time scale using weighted similarity to obtain a comprehensive weighted similarity. After sorting in descending order, the top N historical working condition samples with the highest comprehensive weighted similarity are selected. The initial working condition prior label is determined by similarity-weighted voting, and the corresponding historical mechanism parameter values are obtained. The initial mechanism parameter set is generated by local weighted averaging. Based on the initial mechanism parameter set and the characteristic vector of the operating time window, combined with the hydrodynamic mechanism parameters of gas-solid two-phase flow, a digital twin model of pneumatic conveying is constructed, and the closed conveying pipeline is abstracted as a gas-solid two-phase flow unit and adjustable parameter nodes are set.
3. The bulk material handling scheduling optimization method based on digital twins according to claim 2, characterized in that, The specific steps for generating the virtual state tensor of the output pipeline include: The comprehensive prediction error index is calculated by continuously comparing the pipeline differential pressure, material conveying rate, pipeline flow velocity, and field measured data predicted by the pneumatic conveying digital twin model with online ensemble learning algorithms. Based on the comprehensive prediction error index, the initial mechanism parameters within the mechanism parameter set are dynamically corrected, and the virtual state tensor of the pipeline is output.
4. The bulk material handling scheduling optimization method based on digital twins according to claim 3, characterized in that, The specific steps for dynamically correcting the initial mechanism parameter set include: With the goal of minimizing the comprehensive prediction error index, the contribution of each mechanism parameter in the initial mechanism parameter set to the total error is solved by back-calculating the online ensemble learner, and an adaptive correction step size is set to incrementally update each mechanism parameter. Using the sensitivity of the comprehensive prediction error index to the current mechanism parameters as the parameter gradient, the corresponding gradient direction and gradient magnitude are calculated through partial derivatives, and the three mechanism parameters with the largest gradient magnitudes are updated first. After updating, recalculate the comprehensive prediction error index. If the comprehensive prediction error index decreases, retain the updated result; otherwise, repeat the iteration until the maximum number of iterations is reached. When the comprehensive prediction error index is continuously lower than the preset steady-state threshold, the correction step size is reduced and the steady-state fine-tuning mode is entered; when the comprehensive prediction error index continuously exceeds the preset steady-state threshold and the operating condition characteristics are deviated, the weighted similarity matching is re-performed, the initial mechanism parameter set is updated, and then the correction is performed.
5. The bulk material handling scheduling optimization method based on digital twins according to claim 4, characterized in that, The specific steps for calculating the transport capacity margin of each pipeline segment include: Based on unified pipeline mileage coordinates, the entire closed conveying pipeline is discretized into independent pipe segments and formed by dividing nodes such as the inlet, pipeline diameter change section, mixing bin access point, and the inlet and outlet of the sifting and powder removal station. Extract the material conveying rate, upper limit of material mass flow rate, and material consumption rate of downstream gravity mixing bin and sifting and de-powdering station for each pipe section from the pipeline virtual state tensor. The supply and demand deviation of the pipe section is calculated by mass flow rate conservation according to the pipe section topology relationship, with the material consumption rate of the downstream gravity mixing bin, the material consumption rate of the elution and de-powdering station, and the material demand of the adjacent downstream pipe section as the demand. Based on the upper limit of material mass flow rate and the current actual conveying rate, the conveying capacity margin of each pipe section is calculated, and the current conveying status of the pipe section is characterized by the supply and demand deviation of the pipe section.
6. The bulk material handling scheduling optimization method based on digital twins according to claim 5, characterized in that, The specific steps for parallel gas-solid two-phase flow simulation include: Using the unified pipeline mileage coordinates and timestamps of each pipeline segment as spatiotemporal positioning information, and the conveying capacity margin, pipeline supply and demand deviation, material conveying rate, material flow velocity, pipeline pressure difference, and pipeline pressure difference gradient as state features, the spatiotemporal state trajectory of each pipeline segment is formed within a sliding time window. After normalization, a density-based clustering method is used to identify state clusters. Based on the conveying capacity margin, pipeline pressure difference, supply and demand deviation of pipeline section and material flow rate, the state clusters are marked as high blockage risk pipeline sections, pipeline sections with excessive material flow rate and pipeline sections with insufficient material flow rate, and integrated into an abnormal pipeline section set; For abnormal pipe sections, the pneumatic conveying digital twin model is used to perform parallel gas-solid two-phase flow simulation. The simulation pipe section is constructed and the simulation boundary conditions are set. The gas-solid two-phase flow dynamics simulation under positive pressure dilute phase conveying mode and negative pressure dilute phase conveying mode is performed in parallel, and the simulation results of each simulation pipe section are output.
7. The bulk material handling scheduling optimization method based on digital twins according to claim 6, characterized in that, The specific steps for outputting the control instruction set include: The optimization criteria are based on minimizing the risk of pipe blockage, minimizing the wear rate of particulate materials, achieving the required uniformity of mixing, ensuring compliance with the dust content of discharged materials, and minimizing the energy consumption of the blower. Positive pressure dilute phase transport mode and negative pressure dilute phase transport mode are used as discrete transport mode decision variables, and the control parameters corresponding to the adjustable parameter nodes are used as continuous control variables. The system performs global optimization by using the pipeline's allowable pressure range, the equipment's rated operating range, the minimum stable conveying speed, the maximum allowable material concentration, the lower limit of mixing uniformity, and the upper limit of discharge dust content as optimization constraints, and outputs a set of control instructions.
8. The bulk material handling scheduling optimization method based on digital twins according to claim 7, characterized in that, The specific steps for global optimization include: A non-dominated sorting genetic algorithm with an elite strategy is used for global optimization. An initial population is generated based on the current operating parameters, positive pressure dilute phase transport mode and negative pressure dilute phase transport mode. Discrete transport mode decision variables and continuous control variables are mixed and encoded. The pneumatic conveying digital twin model is used to solve for each optimization objective value and constraint violation. Elite individuals are selected through non-dominated sorting and crowding calculation. The next generation population is generated through selection, crossover, and mutation operations. The Pareto optimal solution set is obtained by iterating to the maximum number of iterations. The comprehensive optimal solution is selected based on the entropy weight TOPSIS method to determine the optimal conveying mode and corresponding control parameter combination.
9. The bulk material handling scheduling optimization method based on digital twins according to claim 8, characterized in that, The specific steps for updating the pipeline virtual state tensor and verifying the pipeline segment's transport capacity margin, blockage risk level, and processing quality indicators include: Based on the control characteristics of the execution equipment and the impact characteristics of pneumatic conveying operation, the control instruction set is decomposed into a three-level priority instruction subset. Combined with the current pipeline blockage risk level, pipeline topology and equipment executable status, an operation condition adaptation timing arbitration rule is constructed. Arbitration generates equipment instruction sequences based on the working condition adaptation timing arbitration rules. Before the control instructions are officially issued, a pneumatic conveying digital twin model pre-test is performed and verified. After outputting an executable control instruction package, it is sent to the corresponding execution equipment. Acquire actual equipment operation feedback, pipe segment sensor measured data and command settings, construct triplet time series data to calculate execution deviation, mark abnormal execution equipment in combination with preset allowable deviation range, and output execution feedback dataset; Acquire real-time data of the entire pipeline operating conditions from sensors, combine it with the execution feedback dataset, update the pipeline virtual state tensor, and verify the pipeline section's transport capacity margin, blockage risk level, and processing quality indicators.
10. The bulk material handling scheduling optimization method based on digital twins according to claim 9, characterized in that, The specific steps to trigger global scroll re-optimization include: When the differential pressure of any pipe section, material flow rate, or processing parameters of any workstation continuously deviates from the preset safety threshold, a global rolling re-optimization is triggered. Using the full pipeline operating condition sensor measured time series data, the updated pipeline virtual state tensor and equipment execution status as the initial optimization benchmark, the over-limit pipe sections and abnormal work positions are located, and the upstream and downstream related pipe sections are linked to form a global optimization boundary. The non-dominated sorting genetic algorithm is used again for global optimization. The optimization criteria are based on multiple objectives: minimizing pipeline blockage risk, minimizing particulate material wear rate, achieving mixing uniformity, ensuring compliant discharge dust content, and minimizing fan air volume energy consumption. The preset safety threshold serves as a hard constraint. The pneumatic conveying digital twin model is used to calculate the optimization objective values and constraint violations. After iterative convergence, the entropy weight TOPSIS method is used to select the comprehensive optimal solution from the Pareto optimal solution set and generate the updated control instruction set.
Citation Information
Patent Citations
Intelligent real-time production line regulation and control system and method based on digital twinning
CN121857553A
