Oil mist particle trajectory control method and system in oil mist separation

Abstract

The invention discloses an oil mist particle trajectory control method and system in oil mist separation, and relates to the technical field of intelligent control, and the method comprises the steps: collecting the three-dimensional position and velocity vector of oil mist particles in real time, and carrying out the three-dimensional modeling of an oil mist trajectory; taking the deviation between the actually measured track and the ideal track as a training sample, inputting the training sample into a multi-layer perceptron, and outputting a current error vector and a compensation increment; constructing a multi-target evaluation function based on the compensated position target and the safe energy consumption factor; optimizing the control parameters by adopting a gray wolf optimization algorithm with improved congestion degree; and issuing the optimized parameters to an execution system to automatically adjust corresponding values. By means of the method, high-precision online control over oil mist particle movement is achieved; the separation efficiency and the energy consumption are effectively balanced; the system has self-adaptive closed-loop capacity, parameters can be quickly adjusted when working conditions change, and separation stability and equipment safety are guaranteed; manual intervention is reduced through automatic execution, and operation reliability and production efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a method and system for controlling oil mist particle trajectories in oil mist separation. Background Art

[0002] Oil mist separation technology has been widely used in industrial production and environmental protection. Early oil mist separation devices mainly relied on gravity sedimentation and inertial collision. Representative equipment included gravity settling chambers, cyclone separators, and fiber filter separators. With the increasing requirements for separation efficiency and energy consumption, researchers have gradually combined static separation structures with computational fluid dynamics (CFD), optimizing the cavity geometry and baffle arrangement through numerical simulation; at the same time, particle image velocimetry (PIV) and laser Doppler velocimetry (LDA) are used to obtain the trajectory of oil droplets, providing experimental data support for structural design. In recent years, machine learning and intelligent optimization algorithms (such as genetic algorithms and particle swarm optimization) have begun to be applied in multi-objective design to balance separation efficiency and pressure loss, but they are mostly limited to parameter optimization under static conditions and have not yet deeply solved the problem of online adaptive control under dynamic conditions.

[0003] Although the above technologies have achieved certain results in separation efficiency and structural optimization, there are still several key deficiencies: First, existing methods for predicting oil mist particle trajectories mostly rely on offline CFD or experimental data, which makes it difficult to reflect the motion deviation caused by turbulence and boundary condition changes during operation in real time; second, multi-objective optimization is often limited to static parameter scanning in the design phase, lacking feedback loop and online adaptive capabilities, resulting in high energy consumption or unstable separation efficiency during actual operation; third, traditional separation systems often focus on a single performance indicator, making it difficult to take into account the comprehensive needs of safety, energy consumption and efficient separation. These limitations are particularly prominent when there are load fluctuations or sudden changes in operating conditions. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed. The technical problem to be solved by the present invention is that the existing oil mist separation technology has problems such as insufficient real-time online adaptive control capability and inability to track the trajectory of oil mist particles with high precision.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] A method for controlling oil mist particle trajectories in oil mist separation, comprising:

[0007] Collect the three-dimensional position and velocity vector of oil mist particles in real time and perform three-dimensional modeling of the oil mist trajectory;

[0008] The deviation between the measured trajectory and the ideal trajectory is input into the multi-layer perceptron as a training sample, and the current error vector and compensation increment are output;

[0009] Based on the compensated position target and safety energy consumption factors, a multi-objective evaluation function is constructed;

[0010] The multi-objective evaluation function is converted into a composite fitness function, and the control parameters are optimized using the crowding-modified grey wolf optimization algorithm;

[0011] The optimized parameters are sent to the execution system to automatically adjust the corresponding values.

[0012] As a preferred embodiment of the method for controlling the trajectory of oil mist particles in oil mist separation according to the present invention, the three-dimensional modeling of the oil mist trajectory includes arranging two synchronously triggered high-speed stereo cameras and calibrating their internal and external parameters; performing background subtraction and connected domain segmentation on the left and right images during each sampling period to extract the pixel center of gravity of the oil mist particles at each viewing angle;

[0013] After stereo rectification, the pixel centroids are mapped to three-dimensional points in the global coordinate system using triangulation. The three-dimensional point sequences of consecutive frames are temporally correlated and smoothed using a Kalman filter, and the velocity vector is calculated based on the difference between two adjacent frames. Cubic spline interpolation is applied to each smoothed discrete trajectory to construct a continuous three-dimensional motion curve model.

[0014] As a preferred embodiment of the oil mist particle trajectory control method in oil mist separation of the present invention, the outputting of the current error vector and compensation increment includes recording the current three-dimensional position and velocity of the oil mist particles, as well as the control parameters and local flow field properties at the corresponding moment during sampling, to form a continuous sequence of measured trajectory and flow field property data;

[0015] The deviation between the measured trajectory and the ideal trajectory in spatial coordinates is used as the error sample, and the spatial deviation at the current moment is used as the target output of the multi-layer perceptron (MLP). The control parameters, flow field properties, and historical deviations are combined into an input vector and input into the MLP model.

[0016] MLP obtains the prediction error through forward propagation and inverts the prediction error to obtain the compensation increment.

[0017] As a preferred embodiment of the oil mist particle trajectory control method in oil mist separation of the present invention, the position target after compensation and the safety energy consumption factor include a comprehensive consideration of trajectory fit and operational safety energy consumption based on the difference between the target position after compensation and the actual position, combined with high wear areas, dead zones, and blockage risks in the local flow field;

[0018] The trajectory fit is a measure of the distance between the predicted position under the control parameters and the compensated target position; the operational safety energy consumption includes an estimate based on flow field risk mapping and equipment power consumption.

[0019] As a preferred solution of the oil mist particle trajectory control method in oil mist separation of the present invention, the composite fitness function includes maintaining a batch of candidate control parameters in each iteration, assuming that the i-th group of candidate control parameters includes wind speed v i , wind direction α i , baffle angle β i , denoted as vector Θ i =[v i ,α i ,β i ];

[0020] For the i-th group of parameters Θ i ,Based on the current actual position and sampling period, the candidate control parameters and local flow field properties are used to calculate the predicted position at the next moment in the simulation mapping model;

[0021] The current particle point moves in the velocity direction and reaches the predicted position after one sampling period;

[0022] The square of the spatial distance between the predicted position and the current compensation target position is used as the fit index f1; the risk value is obtained by querying the pre-established empirical model using the flow field properties at the predicted position to obtain the risk index f2;

[0023] The fit index f1 and the risk index f2 are linearly combined according to the preset weights to obtain the composite fitness function of the parameter group Θ.

[0024] As a preferred solution of the oil mist particle trajectory control method in oil mist separation of the present invention, the gray wolf optimization algorithm with improved crowding degree includes initializing a number of wolf groups in a custom parameter space, with N wolves in a group; using archive Save the set of non-dominated solutions of the previous Pareto frontiers;

[0025] Associate the current group with the archive Merge, perform Pareto stratification on all binary indicators f1 and f2 corresponding to Θ, and divide multiple individual levels Front;

[0026] Any solution in level Front1 is not dominated by other solutions on f1 and f2 simultaneously;

[0027] For each solution in the Front, sort it in ascending order of f1 and f2 respectively, and determine the difference between the neighbors of each solution in the two sortings; add the differences in the two directions after standardization to obtain the crowding value of each solution; assign the maximum crowding value to the two solutions on the Front boundary;

[0028] From Front1, the three solutions with the highest crowding degree are selected as leaders and named as the best wolf, the second best wolf, and the third best wolf, respectively, representing the three sets of parameters with the best fit and risk balance in the current group;

[0029] For each wolf i , calculate the new parameter vector Θ i '; Use the known number of iterations and the maximum number of times to calculate the decay factor, and then randomly generate two coefficients to calculate the position update factor and position scaling factor respectively; use the position scaling factor to calculate the distance vectors of the three wolves respectively, and take the absolute value to get the distance from the leader; the leader, the distance, and the position update factor together form three sets of candidate update positions; finally, take the average of the three sets of candidate positions as the position of the new generation individual;

[0030] According to the position of the candidate new generation individuals, merge and update Θ i ′, for Θ i Each component of ′ is clipped to the upper and lower bounds to ensure that it falls within the allowed range of the device;

[0031] Use the updated group Θ′ and archive Repeat the non-dominated sorting and congestion calculation, select the new top N non-dominated solutions, and refresh the file When the number of iterations reaches the preset upper limit and the frontier solution set no longer changes, the parameter Θ with the largest congestion in the archive Front1 is taken. * The optimization results are sent to the execution system to automatically adjust the wind speed, wind direction and baffle angle;

[0032] The Front1 solution in this round of archives is used as the initial group and leader wolf candidate for the next iteration.

[0033] As a preferred solution of the oil mist particle trajectory control method in oil mist separation of the present invention, wherein: the sending of the optimized parameters includes sending the optimal parameter θ corresponding to the optimal wolf * =[v * , α * , β * ] is sent to the execution system, where the fan control unit and servo-driven partition automatically adjust the wind speed, wind direction and baffle angle; after each sampling cycle, the new samples are stored in the incremental data pool; when the number of samples reaches the specified batch, the MLP is fine-tuned using the learning rate; based on the latest collected oil mist trajectory data, the risk indicator mapping is regularly re-evaluated;

[0034] The termination condition is defined. When the arc length parameter is greater than 1 and the positioning error is less than ε, the closed-loop control is terminated, completing the high-precision trajectory guidance and oil mist separation tasks.

[0035] As a preferred solution of the oil mist particle trajectory control system in the oil mist separation of the present invention, it includes: a data acquisition module, an error compensation module, a multi-objective optimization module, and an execution control module;

[0036] The data acquisition module is used to integrate a laser point cloud sensor to obtain the three-dimensional motion trajectory of oil mist particles and local flow field properties in real time and perform preprocessing;

[0037] The error compensation module is used to use the deviation between the measured trajectory and the ideal trajectory as a label, adopt an online fine-tuning MLP model to model historical features, and output a current error compensation vector;

[0038] The multi-objective optimization module is used to construct a composite fitness, use the crowding-modified grey wolf optimization algorithm, and dynamically search for the optimal control combination;

[0039] The execution control module is used to send the optimal parameters to the fan and partition control system for real-time adjustment; and send a new round of measured data back to the data acquisition module.

[0040] A computer device includes a memory and a processor. The memory stores a computer program. The processor executes the computer program to implement a method for controlling oil mist particle trajectories in oil mist separation.

[0041] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for controlling oil mist particle trajectories in oil mist separation.

[0042] Beneficial effects of the present invention: The oil mist particle trajectory control method in oil mist separation provided by the present invention realizes high-precision online control of the oil mist particle movement through real-time three-dimensional trajectory modeling and error compensation; combines multi-objective evaluation and improved gray wolf optimization to effectively balance separation efficiency and energy consumption; the system has adaptive closed-loop capability and can quickly adjust parameters when working conditions change to ensure separation stability and equipment safety; automated execution reduces manual intervention and improves operational reliability and production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 This is an overall flow chart of the oil mist particle trajectory control method in oil mist separation provided by the first embodiment of the present invention. DETAILED DESCRIPTION

[0045] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0046] Example 1, with reference to Figure 1 , as one embodiment of the present invention, provides a method for controlling the trajectory of oil mist particles in oil mist separation, comprising:

[0047] S1: Collect the three-dimensional position and velocity vector of oil mist particles in real time and perform three-dimensional modeling of the oil mist trajectory.

[0048] Install two or more synchronously triggered high-speed cameras or laser lighting systems at the key sections of the separator to ensure that oil mist particles can be captured simultaneously from different perspectives.

[0049] All sensors are calibrated with internal parameters (focal length, principal point, distortion coefficient) and external parameters (position, attitude) and unified into the same global coordinate system to ensure the accuracy of subsequent 3D reconstruction.

[0050] Each sampling cycle captures multi-view images or point cloud frames. Grayscale conversion, background modeling, and foreground segmentation are performed on the images to extract candidate pixel clusters of oil mist particles. Alternatively, straight-through filtering and voxel downsampling are performed on the point cloud to remove redundancy.

[0051] In each frame, the foreground cluster is subjected to connected domain analysis or cluster segmentation to obtain the pixel centroid (u i ,v i ).

[0052] Record the timestamp t and save the corresponding pixel coordinate list {(u i ,v i )} k Using the projection model and calibration parameters of multi-view imaging, the center of gravity of each pixel (u i ,v i )The corresponding beam vector is projected into space.

[0053] Perform the least squares intersection or the closest point pair calculation on the beams of the same particle from different cameras to obtain the three-dimensional coordinates P of the particle at time t. real (t).

[0054] The Kalman filter is used to track the 3D points identified in the continuous frames and to perform spatiotemporal correlation to form a time sequence of particle motion points. real (t), P real(t-Δt)}.

[0055] Calculate the velocity vector V(t), the formula is expressed as:

[0056]

[0057] Perform sliding window averaging or Kalman filtering on the velocity to filter out noise. real (t) Splice into a curve according to time.

[0058] Perform aggregate analysis on the curves of multiple particles, or build a sparse point cloud time series diagram in three-dimensional space, and then use polynomial or cubic spline fitting to obtain a continuous three-dimensional trajectory function T(s), s∈[0,1].

[0059] The processed three-dimensional position and velocity vectors are imported into the visualization module to display the oil mist movement trend in real time in the form of arrows or streamlines on the digital twin interface.

[0060] At the same time, {P real (t), V(t)} are stored in the data buffer for subsequent error modeling and optimization algorithms.

[0061] It should be noted that millisecond-level synchronous multi-view measurement generates centimeter-level spatial coordinates of oil mist particles; continuous time series correlation and filtering can accurately obtain the speed and direction of particle movement; provide stable and rich space-time samples for subsequent error compensation and optimized control; accurately adjust control parameters based on the real trajectory model to effectively enhance oil mist shedding and deposition.

[0062] S2: The deviation between the measured trajectory and the ideal trajectory is input into the multilayer perceptron as a training sample, and the current error vector and compensation increment are output.

[0063] Furthermore, the output of the current error vector and compensation increment includes collecting the three-dimensional position, velocity, and local flow field characteristics of the oil mist particles; recording the current control parameters and local flow field properties for each sampling to form a continuous sequence of measured trajectory and flow field property data; performing MLP offline training and collecting the sampling time t according to historical experiments or simulation records:

[0064] x(t)=[v(t―k:t),α(t―k:t),β(t―k:t),g(P real (t)),E(t―k:t―1)]

[0065] Among them, v, α, β are the historical wind speed, wind direction, and baffle angle sequences respectively, E is the historical displacement deviation vector; the target is the current positioning error E(t), which is expressed as follows:

[0066] E(t)=P real (t)-Ptarget (s(t))

[0067] Where k represents the length of the historical step window used for error modeling.

[0068] Construct an MLP with input dimension D, two hidden layers (size can be set to H1, H2), ReLU activation; output dimension 3, corresponding to the error Use Adam to optimize with mean square error and train until the loss converges.

[0069] Short-term dynamics prediction uses a simplified flow field model or a pre-calibrated local velocity map to calculate the predicted position of the particle under the control period Δt:

[0070]

[0071] Among them, u(·) is the unit velocity direction function, which is determined by the current flow field attribute g(P real (t)) decision.

[0072] The trajectory fit term calculates the deviation between the predicted position and the compensated target position:

[0073]

[0074] The risk formula is expressed as:

[0075]

[0076] The composite fitness is linearly weighted by weights w1 and w2:

[0077]

[0078] Among them, P adj (t) represents the target coordinate after compensation. A smaller F value indicates that the parameter combination can better fit the target trajectory and reduce the separation risk.

[0079] The spatial deviation between the measured and ideal trajectories is used as an error sample and fed into a multilayer perceptron (MLP). The model MLP inputs include historical control parameter data, contemporaneous flow field properties, turbulence intensity, temperature and humidity sensor data, and the trajectory deviation vector from the previous step. The current error vector is output and then negated to form a compensation increment.

[0080] It should be noted that existing methods typically use simple linear regression or PID control to compensate for deviations, which makes it difficult to capture nonlinear and time-varying characteristics. This paper employs a multi-layer perceptron (MLP) to model oil mist trajectory deviations. Its inputs include multi-dimensional features such as historical control parameter sequences, local flow field properties, and previous deviation vectors. Through Reluctant Unit (ReLU) activation and a multi-layer hidden layer architecture, it accurately approximates complex three-dimensional trajectory error mappings. Combining Adam optimization with an online fine-tuning strategy, this approach achieves rapid response to sudden changes in operating conditions.

[0081] S3: Based on the compensated position target and safety energy consumption factors, a multi-objective evaluation function is constructed.

[0082] Furthermore, the constructing of the multi-objective evaluation function includes defining a separation risk cost objective function based on the difference between the compensated target position and the actual position, combined with high wear areas, dead zones and blockage risks in the local flow field.

[0083] A multi-objective evaluation function is constructed by comprehensively considering trajectory fit and operational safety energy consumption. The trajectory fit measures the distance between the predicted position under control parameters and the compensated target position. The operational safety energy consumption is estimated based on flow field risk mapping and equipment power consumption. Weights are assigned to trajectory fit and operational safety energy consumption to obtain a composite fitness function with a consistent structure.

[0084] The composite fitness function includes: candidate parameters are defined in each iteration, and a group of candidate control parameters are maintained, each group of parameters includes wind speed, wind direction, and baffle angle; denoted as vector Θ i =[v i ,α i ,β i ], there are N groups in total.

[0085] For the i-th group of parameters Θ i ,Based on the current actual position and sampling period, the candidate control parameters and local flow field properties are used to calculate the predicted position at the next moment in the simulation mapping model.

[0086] The current particle point moves in the velocity direction at a rate of one sampling period and reaches the predicted position.

[0087] The trajectory fit evaluation calculates the spatial distance between the predicted position and the current compensation target position, which is recorded as the deviation. The fit index f1 is the sum of squared deviations. The smaller the value, the better the i-th parameter can guide the particle along the target trajectory. At the same time, the risk cost evaluation uses the flow field properties at the predicted position to query a risk value in the pre-established empirical model to obtain the risk index f2.

[0088] The fit index f1 and the risk index f2 are linearly combined according to the preset weights to obtain the composite fitness function of the parameter group Θ.

[0089] Comprehensiveness gives equal weight to trajectory accuracy, safety, and energy consumption, eliminating the need to sacrifice other performance indicators for a single indicator; flexibility weights can be dynamically adjusted based on on-site priorities (such as safety priority or energy saving priority); interpretability separates risk and energy consumption items, each with a clear physical source, making it easier for engineers to understand and optimize.

[0090] It should be noted that previous multi-objective evaluations often focused on a single metric or simple weighting, ignoring the coordination between risk and energy consumption. Based on "trajectory fit," combined with flow field risk mapping (high wear areas, dead zones, and blockage risks) and equipment power consumption estimation, a composite fitness function with consistent structure and adjustable weights was constructed. This function not only quantifies the degree to which particles deviate from the target but also assesses safety and energy costs, making the optimization objectives more aligned with industrial application needs.

[0091] S4: The multi-objective evaluation function is converted into a composite fitness function, and the control parameters are optimized using the crowding-modified grey wolf optimization algorithm. The optimized parameters are sent to the execution system to automatically adjust the corresponding values.

[0092] Furthermore, the crowding-modified gray wolf optimization algorithm includes initializing a number of wolf pack individuals in a custom parameter space, where the group currently has N wolves; Save the set of non-dominated solutions of the previous Pareto frontiers.

[0093] Associate the current group with the archive Merge and perform Pareto stratification on all binary indicators f1 and f2 corresponding to Θ to divide the hierarchy Front into multiple individual levels.

[0094] Any solution from level Front1 is not dominated by any other solution on both f1 and f2.

[0095] For each solution within the Front, first sort it in ascending order of f1 and f2 respectively, and determine the difference between the neighbors of each solution in the two sortings; the difference in the two directions is normalized and added to obtain the congestion value of the solution; the two solutions on the Front boundary are artificially assigned the maximum congestion.

[0096] Every L upd A sampling period is used to perform a CDGWO iteration to find the new optimal control parameters. If the population and archive initialization are updated for the first time, then in the parameter range [Θ min ,Θ max ] randomly generates N candidate vectors {Θ i}. Create external archives at the same time Used to store all previous non-dominated solutions. Non-dominated sorting will be the current group {Θ i}(and files ) are merged, and Pareto stratification is performed based on the dual objectives (f1, f2), marking the level (Front) of each individual. Crowding calculation For each individual in the Front, sort them in ascending order by f1 and f2 respectively, and calculate the crowding degree:

[0097]

[0098] Indicates that the boundary individuals are assigned maximum values ​​to ensure diversity.

[0099] The leader wolf is selected from level 1 (optimal Front), and the first three individuals are selected in descending order of crowding degree CD, and are set as α0 (optimal), β0 (second optimal), and δ0 (third optimal) respectively.

[0100] For each wolf i , calculate the new parameter vector Θ i '; Use the known number of iterations and the maximum number of times to calculate the attenuation factor, and then randomly generate two coefficients, which are used to calculate the position update factor and the position scaling factor respectively; use the position scaling factor to calculate the distance vectors of the three wolves respectively, and take the absolute value to get the distance from the leader; the leader, the distance and the position update factor together form three sets of candidate update positions; finally, take the average of the three sets of candidate positions as the new generation individual position.

[0101] For each "wolf" Θ i The attenuation coefficient is calculated by parallel iteration according to the following formula:

[0102] A=2ar1―a,C=2r2

[0103] Calculate the distance vector for each of the three wolves:

[0104] D α =|Cα―Θ i |

[0105] D β =|Cβ―Θ i |

[0106] D δ =|Cδ―Θ i |

[0107] Calculate three candidate positions:

[0108] X α =α―AD α

[0109] X β =β-AD β

[0110] Xδ =δ―AD δ

[0111] General Update:

[0112]

[0113] And clip each component to [Θ min ,Θ max ]. Among them, D α ,D β ,D δ The distance vector to the three leading wolves, X α ,X β ,X δ Three sets of candidate update positions are calculated based on the leader wolf's position, distance vector and coefficients. External archive for storing non-dominated solutions to maintain multi-objective diversity. i Represents the degree of crowding, which is used to maintain the diversity of individuals at the same level. ε is the termination decision threshold, at which the expected accuracy is considered to have been achieved.

[0114] Merge the new group with the old archive, repeat the steps of non-dominated sorting and congestion calculation, filter out the new top N non-dominated solutions, and update the external archive In the archive that ends with this iteration, the individual with the best crowding degree will be selected as the candidate for the next initial leader wolf.

[0115] According to the position of the candidate new generation individuals, merge and update Θ i ', for Θ i Each component of ' is clipped to the upper and lower bounds to ensure that it falls within the range allowed by the device.

[0116] Use the updated group Θ′ and archive Repeat the non-dominated sorting and congestion calculation, select the new top N non-dominated solutions, and refresh the file When the number of iterations reaches the preset upper limit or the frontier solution set no longer changes, the parameter Θ with the largest congestion in the archive Front1 is taken. * , as the optimization result, it is sent to the execution system to complete the automatic adjustment of wind speed, wind direction and baffle angle.

[0117] The Front1 solution in this round of archives is used as the initial group and leader wolf candidate for the next iteration.

[0118] The sending of the optimized parameters includes sending the optimal parameter Θ corresponding to the optimal wolf * =[v * , α * , β * ] is sent to the execution system, and the fan control unit and servo drive partition automatically adjust the wind speed, wind direction and baffle angle.

[0119] After each sampling cycle, new samples are stored in the incremental data pool. When the number of samples reaches the specified batch, the MLP is fine-tuned using the learning rate. Based on the latest collected oil mist trajectory data, the risk indicator mapping is regularly reassessed to reflect changes in the flow field and equipment status.

[0120] The termination condition is defined. When the arc length parameter is greater than 1 and the positioning error is less than ε, the closed-loop control is terminated, completing the high-precision trajectory guidance and oil mist separation tasks.

[0121] The improved wolf pack algorithm is essentially to find the optimal wolf, use Pareto stratification, and merge groups and archives, that is, compare the current group of wolves with the old wolves in the memory box.

[0122] The first layer (Front 1) of the Pareto stratification is the group where no wolf is inferior to it in both objectives. The second layer (Front 2) deletes the first layer and then finds a new group, and so on. This step is to find the most comprehensive group that is both on track and safe.

[0123] Maintaining diversity and defining crowding, in order to prevent all wolves from being crowded into the same "excellent but similar" position, also depends on how "lonely" each wolf is in its layer:

[0124] Sort the two goals separately: sort the wolves in this layer from small to large according to "deviation" and from small to large according to "risk".

[0125] Measure the neighbor gap: see how far apart the two wolves in front and behind you are in the ranking; add up the gaps between the two targets to get the congestion degree.

[0126] Border wolves (front and back of the row) are given a higher crowding degree, making them more likely to be retained. This way, "various excellent solutions" can be retained at the same time, without all wolves being gathered at one point at once.

[0127] Select the leader wolf α0, β0, δ0. Wolf α0 represents the most crowded one in the first layer (the best and most representative one).

[0128] Wolf β0 and wolf δ0 represent the second- and third-best representatives, equivalent to the leader of the hunting team. Each wolf simultaneously references the three leaders α, β, and δ, and moves closer to them, but with a touch of randomness and a "step-down" mechanism. The decay coefficient a: As the number of iterations increases, the amplitude of each step gradually decreases, ensuring accurate fine-tuning later. The random coefficients r1 and r2: Add a bit of randomness to each movement to prevent getting stuck in local blind spots. The distance vector: Calculates the distance from the wolf to the three wolves α, β, and δ. Three sets of candidate new positions: Averaging these three sets of new positions yields the next generation's position. The clipping range means that if a wolf strays outside the allowed range, it will be pulled back. This way, the wolf pack can both converge on the best direction while maintaining exploration diversity due to randomness and multiple leaders.

[0129] The new generation of wolves is merged with the old archive, and the Pareto stratification and congestion degree are re-performed to update the non-dominated solutions in the archive. The optimal wolf in the latest archive is used as the next round of α, β, and δ, and the update continues. This cycle continues until the "frontier solution" stops changing or the number of iterations reaches the upper limit.

[0130] It should be noted that traditional gray wolf optimization struggles to simultaneously address multiple objectives and population diversity, and is prone to falling into local optima. This paper introduces a Pareto front non-dominated sorting and crowding degree maintenance strategy (CD-GWO) into the classic gray wolf algorithm. This strategy ensures that historically excellent solutions are not lost through archive merging. It also maintains population diversity through stratification of binary indicators and crowding degree calculation. Furthermore, it combines a decay factor with a random coefficient to achieve adaptive attraction / repulsion of leader wolves, enhancing global search capabilities. Ultimately, the optimal solution is distributed to the execution system, enabling precise and automated adjustment of wind speed, direction, and baffle angle.

[0131] Example 2 is an embodiment of the present invention, which provides a method for controlling the trajectory of oil mist particles in oil mist separation. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0132] First, this example constructed a digital oil mist separation simulation platform on an indoor test bench. A 3D laser point cloud scan of the separator interior was performed. Physical parameters such as turbulence intensity, oil droplet concentration, and temperature and humidity were derived from fluid simulation and acoustic tomography inversion to generate a digital separation cavity model with physical properties. Based on the simulation results, the ideal separation trajectory of the oil mist particles was determined and discretized into 100 target trajectory points. A high-frequency PIV system and a high-speed imaging camera were then deployed at key sections of the cavity, along with programmable servo actuators for wind speed (0–5 m / s), wind direction (0–360°), and baffle angle (0–30°). At the beginning of the experiment, a uniform spray device was used to generate oil mist particles with a size distribution of 5–10 μm. A pre-scan was performed at a wind speed of 1.0 m / s, a wind direction of 90°, and a baffle angle of 10°. 3D position and velocity data were recorded for the initial 20 seconds for offline MLP pre-training with a batch size of 32, 50 epochs, and a learning rate of 0.001.

[0133] During the online phase, the system collects the control parameter history (previous five steps), current local turbulence intensity and temperature and humidity sensor data, and the previous three-dimensional deviation vector with a sampling period of 0.5 seconds. This is used to construct an input vector, which is then predicted by the MLP to output a real-time error vector. This error is inverted to obtain a compensation increment, which is then applied to the target position. Every 2 seconds, the crowding-degree modified grey wolf optimization (CD-GWO) algorithm is activated to perform a multi-objective fitness evaluation on six candidate control parameter sets (wind speed, wind direction, and baffle angle) within the current range. Weights w1 = 0.7 (trajectory fit) and w2 = 0.3 (separation risk / energy consumption) are used. The algorithm then performs non-dominated sorting, crowding calculation, leader selection, and iterative position updates. The group size N = 20, with a maximum number of iterations of 50, is used. After the optimization is complete, the optimal parameters are distributed to the servo drive for automatic adjustment. The entire experiment lasted 200 seconds, encompassing 100 MLP predictions and 100 GWO optimizations. Real-time correction reduced the average positioning error from an initial 12.5 mm to 3.2 mm, boosted separation efficiency from 82% to 94%, and reduced energy consumption per unit time from 1.20 kW·h to 1.05 kW·h. This example fully demonstrates the synergistic effects of digital modeling, neural network error compensation, and multi-objective swarm intelligent optimization, validating the significant advantages of the present invention in improving separation accuracy, efficiency, and energy conservation.

[0134] All methods significantly reduced the positioning error and improved the separation efficiency, while reducing the unit energy consumption. Taking Experiment 1 as an example, the initial prediction error of 12.5mm was reduced to 3.5mm after MLP compensation, the separation efficiency was improved from 82% (compared with the traditional constant parameter method) to 91.2%, and the energy consumption was reduced from 1.20kW·h to 1.10kW·h; in Experiment 6, the initial error of 12.0mm was reduced to 3.3mm, the separation efficiency reached 95.0%, and the energy consumption was only 1.00kW·h. The error values ​​of each group of experiments were stabilized in the range of 3.3–3.8mm. Compared with the 5–8mm level of the existing technology, the error was reduced by more than 40%, reflecting the high-precision characteristics of three-dimensional error modeling and compensation. The separation efficiency generally exceeded 90%, which is about 10% higher than the traditional fixed parameter separation (≤85%), indicating that the multi-objective optimization of the present invention effectively takes into account both capture efficiency and resource consumption. The energy consumption index is coupled with risk mapping, and high-risk flow fields are automatically avoided in the GWO optimization, striving for low-speed and high-efficiency separation, resulting in an energy consumption reduction of about 13%.

[0135] Example 3 is an embodiment of the present invention, which provides an oil mist particle trajectory control system in oil mist separation, including a data acquisition module, an error compensation module, a multi-objective optimization module, and an execution control module.

[0136] The data acquisition module is used to integrate a laser point cloud sensor to obtain the three-dimensional motion trajectory of oil mist particles and local flow field properties in real time and perform preprocessing.

[0137] The error compensation module is used to use the deviation between the measured trajectory and the ideal trajectory as a label, adopt an online fine-tuning MLP model to model historical features, and output a current error compensation vector.

[0138] The multi-objective optimization module is used to construct a composite fitness, use the crowding-modified grey wolf optimization algorithm, and dynamically search for the optimal control combination.

[0139] The execution control module is used to send the optimal parameters to the fan and partition control system for real-time adjustment; and send a new round of measured data back to the data acquisition module.

[0140] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0141] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0142] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0143] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, it can be implemented using a combination of any of the following technologies known in the art: a discrete logic circuit having logic gate circuits for implementing logical functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc. It should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention and are not intended to be limiting. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications should be encompassed by the scope of the claims of the present invention.

[0144] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for controlling the trajectory of oil mist particles in oil mist separation, characterized in that: include: Collect the three-dimensional position and velocity vector of oil mist particles in real time and perform three-dimensional modeling of the oil mist trajectory; The deviation between the measured trajectory and the ideal trajectory is input into the multi-layer perceptron as a training sample, and the current error vector and compensation increment are output; Based on the compensated position target and safety energy consumption factors, a multi-objective evaluation function is constructed; The multi-objective evaluation function is converted into a composite fitness function, and the control parameters are optimized using the crowding-modified grey wolf optimization algorithm; The optimized parameters are sent to the execution system to automatically adjust the corresponding values.

2. The oil mist particle trajectory control method in oil mist separation according to claim 1, characterized in that: The three-dimensional modeling of the oil mist trajectory includes arranging two synchronously triggered high-speed stereo cameras and calibrating their internal and external parameters; performing background subtraction and connected domain segmentation on the left and right images in each sampling period to extract the pixel center of gravity of the oil mist particles at each viewing angle; After stereo rectification, the pixel centroids are mapped to three-dimensional points in the global coordinate system using triangulation. The three-dimensional point sequences of consecutive frames are temporally correlated and smoothed using a Kalman filter, and the velocity vector is calculated based on the difference between two adjacent frames. Cubic spline interpolation is applied to each smoothed discrete trajectory to construct a continuous three-dimensional motion curve model.

3. The oil mist particle trajectory control method in oil mist separation according to claim 2, characterized in that: Outputting the current error vector and compensation increment includes recording the three-dimensional position and velocity of the current oil mist particles and the control parameters and local flow field properties at the corresponding moment during sampling to form a continuous sequence of measured trajectory and flow field property data; The deviation between the measured trajectory and the ideal trajectory in spatial coordinates is used as the error sample, and the spatial deviation at the current moment is used as the target output of the multi-layer perceptron (MLP). The control parameters, flow field properties, and historical deviations are combined into an input vector and input into the MLP model. MLP obtains the prediction error through forward propagation and inverts the prediction error to obtain the compensation increment.

4. The oil mist particle trajectory control method in oil mist separation according to claim 3, characterized in that: The compensated position target and safety energy consumption factors include, based on the difference between the compensated target position and the actual position, combined with the high wear area, dead zone and blockage risk in the local flow field, and a comprehensive consideration of the trajectory fit and operational safety energy consumption; The trajectory fit is a measure of the distance between the predicted position under the control parameters and the target position after compensation; The operational safety energy consumption includes estimation based on flow field risk mapping and equipment power consumption.

5. The oil mist particle trajectory control method in oil mist separation according to claim 4, characterized in that: The composite fitness function includes, in each iteration, maintaining a batch of candidate control parameters, assuming that the i-th group of candidate control parameters includes wind speed v i , wind direction α i , baffle angle β i , denoted as vector Θ i =[v i ,α i ,β i ]; For the i-th group of parameters Θ i ,Based on the current actual position and sampling period, the candidate control parameters and local flow field properties are used to calculate the predicted position at the next moment in the simulation mapping model; The current particle point moves in the velocity direction and reaches the predicted position after one sampling period; The square of the spatial distance between the predicted position and the current compensation target position is used as the fit index f1; the risk value is obtained by querying the pre-established empirical model using the flow field properties at the predicted position to obtain the risk index f2; The fit index f1 and the risk index f2 are linearly combined according to the preset weights to obtain the composite fitness function of the parameter group Θ.

6. The oil mist particle trajectory control method in oil mist separation according to claim 5, characterized in that: The gray wolf optimization algorithm with improved crowding includes initializing a number of wolf packs in a custom parameter space, with N wolves in a pack; using archive Save the set of non-dominated solutions of the previous Pareto frontiers; Associate the current group with the archive Merge, perform Pareto stratification on all binary indicators f1 and f2 corresponding to Θ, and divide multiple individual levels Front; Any solution in level Front1 is not dominated by other solutions on f1 and f2 simultaneously; For each solution in the Front, sort it in ascending order of f1 and f2 respectively, and determine the difference between the neighbors of each solution in the two sortings; add the differences in the two directions after standardization to obtain the crowding value of each solution; assign the maximum crowding value to the two solutions on the Front boundary; From Front1, the three solutions with the highest crowding degree are selected as leaders and named as the best wolf, the second best wolf, and the third best wolf, respectively, representing the three sets of parameters with the best fit and risk balance in the current group; For each wolf i , calculate the new parameter vector Θ i '; Calculate the attenuation factor using the known number of iterations and the maximum number of times, and then randomly generate two coefficients, which are used to calculate the position update factor and the position scaling factor respectively; The distance vectors of the three wolves are calculated using the position scaling factor, and the absolute value is taken to get the distance from the leader. The leader, the distance, and the position update factor together form three sets of candidate update positions. Finally, the three sets of candidate positions are averaged as the position of the new generation individual. According to the position of the candidate new generation individuals, merge and update Θ i ′, for Θ i Each component of ′ is clipped to the upper and lower bounds to ensure that it falls within the allowed range of the device; Use the updated group Θ′ and archive Repeat the non-dominated sorting and congestion calculation, select the new top N non-dominated solutions, and refresh the file When the number of iterations reaches the preset upper limit and the frontier solution set no longer changes, the parameter Θ with the largest congestion in the archive Front1 is taken. * The optimization results are sent to the execution system to automatically adjust the wind speed, wind direction and baffle angle; The Front1 solution in this round of archives is used as the initial group and leader wolf candidate for the next iteration.

7. The oil mist particle trajectory control method in oil mist separation according to claim 6, characterized in that: The sending of the optimized parameters includes sending the optimal parameter Θ corresponding to the optimal wolf * =[v * , α * , β * ] is sent to the execution system, and the fan control unit and servo drive partition automatically adjust the wind speed, wind direction and baffle angle; after each sampling cycle, the new samples are stored in the incremental data pool; When the number of samples reaches the specified batch, the MLP is fine-tuned using the learning rate; Regularly reassess risk indicator mapping based on the latest collected oil mist trajectory data; The termination condition is defined. When the arc length parameter is greater than 1 and the positioning error is less than ε, the closed-loop control is terminated, completing the high-precision trajectory guidance and oil mist separation tasks.

8. A system using the oil mist particle trajectory control method in oil mist separation according to any one of claims 1 to 7, characterized in that: Including data acquisition module, error compensation module, multi-objective optimization module, and execution control module; The data acquisition module is used to integrate a laser point cloud sensor to obtain the three-dimensional motion trajectory of oil mist particles and local flow field properties in real time and perform preprocessing; The error compensation module is used to use the deviation between the measured trajectory and the ideal trajectory as a label, adopt an online fine-tuned MLP model to model historical features, and output a current error compensation vector; The multi-objective optimization module is used to construct a composite fitness and dynamically search for the optimal control combination using the crowding-modified grey wolf optimization algorithm; The execution control module is used to send the optimal parameters to the fan and partition control system, adjust them in real time, and send a new round of measured data back to the data acquisition module.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for controlling the trajectory of oil mist particles in oil mist separation according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for controlling the trajectory of oil mist particles in oil mist separation according to any one of claims 1 to 7 are implemented.

Images