Coal washing equipment image processing intelligent monitoring system based on strategy optimization

Abstract

The invention discloses a coal washing equipment image processing intelligent monitoring system based on strategy optimization, relates to the technical field of coal washing equipment image processing, and constructs a closed-loop intelligent optimization system composed of an image acquisition module, a process monitoring module, a simulation modeling module, a multi-objective optimization module, a strategy scoring module, a control execution module, a feedback regulation module and the like. By introducing response difference calculation and a response space redistribution model, sparse processing is effectively carried out on a Pareto solution set, and the strategy distinction degree is improved; multi-factor dynamic scoring is carried out by combining image features, historical performance and response tracks, and it is guaranteed that the selected strategy has the optimal execution effect under the current working condition; the system also has model self-correction and scoring parameter self-adaption capabilities in operation, high-precision, steady-state and self-learning control of the whole process is realized, and the intelligent level and the operation reliability of the coal washing equipment are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing of coal washing equipment, and particularly to an intelligent monitoring system for image processing of coal washing equipment based on policy optimization. Background Art

[0002] Intelligent monitoring of image processing of coal washing equipment based on policy optimization refers to using image processing technology to monitor the operating state of equipment in real time during the coal washing process, and combining policy optimization algorithms to intelligently adjust and control the monitoring system. Its core goal is to improve the monitoring efficiency and accuracy by intelligently identifying and analyzing equipment image data, timely detect equipment abnormalities or failures, and dynamically adjust monitoring parameters according to preset or self-learning optimization strategies, so as to achieve intelligent, stable and efficient management of the operation of coal washing equipment.

[0003] The prior art has the following deficiencies:

[0004] In the prior art, the Pareto multi-objective optimization algorithm is used to train and optimize the simulation model to achieve refined control and intelligent decision-making support for the coal washing process. However, in multi-objective optimization based on the Pareto front algorithm, if there are high-dimensional objective combinations, there may be a phenomenon that the Pareto solution distribution is too dense, which is called the Pareto front being too dense. This problem will cause multiple optimization strategies to be approximately equivalent mathematically, but in actual operation, due to the system being overly sensitive to minor adjustments, it will cause slow response or even unstable operation. Since this deviation is difficult to detect during the training stage, it often only gradually appears after the system is deployed and run for a period of time, posing a hidden and serious challenge to the stability and accuracy of the intelligent monitoring system. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent monitoring system for image processing of coal washing equipment based on policy optimization to solve the deficiencies in the background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: An intelligent monitoring system for image processing of coal washing equipment based on policy optimization, including an image acquisition module, a process parameter monitoring module, a simulation modeling module, a multi-objective optimization module, a policy scoring module, a control execution module, and a feedback adjustment module;

[0007] The image acquisition module is used to obtain real-time image information of different parts of the coal washing equipment, identify and structurally process it to form image feature data;

[0008] The process parameter monitoring module is used to collect several process parameters during the coal washing process;

[0009] The simulation modeling module constructs a dynamic simulation model based on image feature data and process monitoring data, and is used to simulate the response relationship between multiple optimization objectives during the coal washing process;

[0010] The multi-objective optimization module generates multiple control strategies based on the Pareto front algorithm, and establishes a response space redistribution model by calculating the response difference degree between strategies to sparsify the Pareto solution set and obtain a sparsified optimization solution set;

[0011] The strategy scoring module dynamically scores the sparsified optimization solution set based on image features, historical operation data and response trajectories, and selects the optimal strategy according to the scoring results;

[0012] The control execution module adjusts the operating parameters of the coal washing equipment according to the selected optimal strategy;

[0013] The feedback regulation module is used to monitor the control execution result, and automatically correct the simulation model or adjust the strategy scoring parameters based on the equipment response situation and target deviation.

[0014] Preferably, the image acquisition module includes: multiple groups of image sensors installed on the screen surface, sorting tank outlet and conveyor belt discharge outlet of the coal washing equipment, and the image sensors include high-resolution industrial cameras, infrared thermal imagers and laser profile scanners.

[0015] Preferably, the image acquisition module is used to collect the original image data during the operation of the equipment; and perform gray processing, edge enhancement and noise removal on the images, identify the coal block boundary and extract image features through the target recognition algorithm, including particle size distribution, stacking thickness, movement direction and coal-gangue ratio; output the processed image features in the form of a structured vector and input them into the simulation modeling module as input variables.

[0016] Preferably, the process parameter monitoring module is used to collect the process parameters during the coal washing process in real time, including medium concentration, water-coal ratio, feed particle size, vibration frequency, motor current, power consumption, water supply flow rate and system temperature.

[0017] Preferably, the simulation modeling module is used to construct a data-driven model:

[0018] Extract the particle size distribution and material flow velocity image features from the image acquisition module;

[0019] Obtain the water-coal ratio, medium concentration and power consumption process data from the process parameter monitoring module;

[0020] Normalize and process the collected data to form multi-dimensional modeling samples;

[0021] Use neural network, support vector machine or random forest algorithm to establish a data-driven model;

[0022] Output the predicted values of each target parameter under the analog control strategy for transmission to the multi-objective optimization module.

[0023] Preferably, after generating the Pareto solution set, the multi-objective optimization module performs a sparsification processing step:

[0024] Calculate the difference degree between the target responses corresponding to any two sets of control strategies in the solution set, and the difference degree is measured by the Euclidean distance or the Mahalanobis distance;

[0025] Establish a target space redistribution model based on the response difference degree;

[0026] Use a clustering algorithm to cluster the solution set;

[0027] Select the representative solution with the largest response difference in each cluster to form a sparsified optimization solution set;

[0028] Output the sparse optimization result to the policy scoring module.

[0029] Preferably, the policy scoring module includes:

[0030] Calculate the matching degree score between the policy parameters and the current image features: perform matching scoring through the Euclidean distance between the current image features and the mean value of the policy historical image features;

[0031] Calculate the average effect score of the policy under similar working conditions: count the average execution effect of the policy under similar working conditions and take the average value of multiple scores;

[0032] Calculate the deviation control ability score after the policy is executed: fit the actual system response to a first-order response model, extract the response time constant, delay time and fitting error, and construct a scoring function;

[0033] Fuse the matching degree score, average effect score and deviation control ability score according to the weighted formula, and output the comprehensive scoring result;

[0034] Sort the policies according to the score from high to low, and select the one with the highest score as the optimal control policy.

[0035] Preferably, the feedback adjustment module includes:

[0036] Sample the indicators in the control execution result;

[0037] Compare with the prediction result of the simulation model and calculate the target deviation vector;

[0038] If the deviation exceeds the preset threshold, trigger the fine-tuning mechanism of the modeling module, including model incremental training, parameter update or structure reconstruction.

[0039] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0040] 1. The present invention realizes the full-process intelligent control of the coal washing process from perception to decision-making by constructing an intelligent monitoring system composed of modules such as image acquisition, process parameter monitoring, simulation modeling, multi-objective optimization, strategy scoring, control execution, and feedback regulation. The system combines image information and process data, establishes a multi-objective simulation model through a data-driven method, and introduces a Pareto front sparsification mechanism in the process of generating control strategies, effectively solving the problems of slow response and unstable control caused by over-dense Pareto solution sets in the prior art, and improving the accuracy, responsiveness, and strategy differentiation ability of the system.

[0041] 2. The present invention introduces a multi-factor scoring mechanism such as image matching degree, historical performance, and response fitting in strategy scoring and feedback regulation, and is equipped with a first-order system response modeling and adaptive feedback optimization mechanism to ensure the practical feasibility and stable execution ability of the control strategy under the current working conditions. The overall system has the advantages of complete structure, strong learning ability, high real-time performance, and good control accuracy, significantly improving the operation efficiency of coal washing equipment, the coal quality control ability, and the intelligent level of system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0043] Figure 1 It is the system mind map of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0045] Embodiment, please refer to Figure 1 As shown, the intelligent monitoring system for image processing of coal washing equipment based on strategy optimization in this embodiment includes an image acquisition module, a process parameter monitoring module, a simulation modeling module, a multi-objective optimization module, a strategy scoring module, a control execution module, and a feedback regulation module;

[0046] An image acquisition module, which is used to obtain real-time image information of different parts of the coal washing equipment, identify and structurally process it, and form image feature data;

[0047] A process parameter monitoring module, which is used to collect several process parameters during the coal washing process;

[0048] A simulation modeling module, which constructs a dynamic simulation model based on the image feature data and process monitoring data, and is used to simulate the response relationship between multiple optimization objectives during the coal washing process;

[0049] A multi-objective optimization module, which generates multiple control strategies based on the Pareto front algorithm, and establishes a response space redistribution model by calculating the response difference degree between the strategies, sparsifies the Pareto solution set, and obtains a sparsified optimization solution set;

[0050] A strategy scoring module, which dynamically scores the sparsified optimization solution set based on the image features, historical operation data and response trajectory, and selects the optimal strategy according to the scoring results;

[0051] A control execution module, which adjusts the operation parameters of the coal washing equipment according to the selected optimal strategy;

[0052] A feedback regulation module, which is used to monitor the control execution result, and automatically correct the simulation model or adjust the strategy scoring parameters based on the equipment response situation and target deviation.

[0053] The image acquisition module mainly consists of the following sub-components:

[0054] An image sensor assembly: including a high-resolution industrial camera (such as a CCD / CMOS camera); an infrared thermal imager and a laser profile scanner can be optionally configured to collect non-visible light images (such as detecting changes in material temperature and humidity); it has an automatic focusing and dust protection mechanism and is suitable for the coal mine dust environment.

[0055] An image acquisition controller: connected to the main system PLC or an embedded control chip; controls the acquisition frequency (frame rate), exposure time, and shutter parameters; has edge computing capabilities and supports preliminary image preprocessing.

[0056] Image acquisition position and perspective design: installed at key parts of the coal washing equipment, such as: the screening section (detecting the material distribution and blockage situation on the screen surface); the sorting tank outlet (detecting the flow state and concentration); the conveyor belt discharge outlet (judging the particle size and coal-gangue separation effect); multi-perspective deployment supports stereo vision reconstruction.

[0057] A high-resolution visible light camera (CCD / CMOS): mainly used to collect dynamic images of coal blocks and materials on the screen surface, belt, and sorting tank. Combining with image recognition algorithms, the following information can be extracted:

[0058] Coal lump size and particle size distribution (used to evaluate screening and separation effects), material stacking thickness (reflecting feeding uniformity), flow velocity and direction (evaluating fluid state stability), and equipment surface operating status (identifying abnormalities such as blockage and deviation).

[0059] Infrared cameras or thermal imaging sensors: Used to analyze the thermal characteristics of the material surface, which can assist in judging: moisture content changes (identifying wet coal through thermal radiation differences), and the heating conditions of conveyor belts or vibrating screens (used to predict equipment abnormalities or overloads).

[0060] Color image sensors: By analyzing image color and brightness, identify: color differences between coal and gangue to judge coal quality components; material mixing conditions or slime trends (detected through changes in grayscale and saturation).

[0061] The image processing and feature extraction process includes:

[0062] Image preprocessing: Gray conversion, edge enhancement, image denoising; adapting to uneven lighting and particulate dust interference in the coal mine environment.

[0063] Image segmentation and recognition: Apply deep learning algorithms (such as YOLOv5, Mask R-CNN) or traditional algorithms (such as threshold segmentation, contour detection); achieve coal lump boundary recognition, particle size statistics, and distribution area extraction.

[0064] Structured feature extraction: Output vectorized features, such as: average particle size, particle size variance; proportion of coal lumps, proportion of gangue; density distribution of the image area; dynamic flow trajectory features (such as velocity field, flow direction); upload the output results to the modeling module to participate in simulation modeling or optimize the input.

[0065] Docking with the simulation modeling module: Image features are used as one of the input variables to participate in multi-objective modeling; support physical modeling of non-numerical parameters, such as the adhesion state of coal slime, fluid state mode, etc.

[0066] Linkage with the control system: For example, if the particle size is identified as too large through image recognition, the vibration frequency is automatically adjusted; when detecting pile offset, the feeding direction or speed can be adjusted in linkage; image anomaly detection can trigger the safety shutdown logic.

[0067] The process parameter monitoring module obtains physical quantities, energy consumption indicators, and process parameters closely related to the operation status of the coal washing process through a variety of industrial sensors deployed at key positions of the equipment, mainly including the following categories:

[0068] Medium concentration monitoring, purpose: Judge the concentration fluctuations of the separation medium (such as magnetic fluid or heavy medium) to control the separation accuracy; sensor types: conductivity sensors, optical concentration meters, ultrasonic concentration meters; collection locations: medium circulation system, inlet and outlet of the separation tank.

[0069] Incoming material particle size and flow rate monitoring, purpose: to control the uniformity of feeding and prevent the influence of particle size fluctuations on screening efficiency and coal-gangue separation; sensor types: laser particle size analyzer, belt scale, flow meter; acquisition locations: vibrating feeder, belt conveying system.

[0070] Water-coal ratio and water supply volume monitoring, purpose: to regulate the water used in coal washing and affect the sedimentation of coal slime and washing effect; sensor types: electromagnetic flow meter, ultrasonic level gauge, on-line water quality analyzer; acquisition locations: washing water pipeline, water tank, flotation cell.

[0071] Operating state parameters of the screen surface, purpose: to monitor whether the vibrating screen operates stably; sensor types: acceleration sensor, vibration sensor; acquisition locations: both sides of the screen machine body, bottom vibration device.

[0072] Energy consumption and motor state monitoring, purpose: to monitor the energy consumption efficiency of the system and identify abnormal equipment load or fault trends in advance; sensor types: current / voltage transformer, intelligent electricity meter, thermistor; acquisition locations: key driving equipment such as main motor, pump, and fan.

[0073] System temperature and ambient humidity monitoring, purpose: to ensure the safe operation of equipment, especially in harsh environments such as wet coal and high dust; sensor types: temperature and humidity integrated sensor, infrared thermometer; acquisition locations: machine body shell, conveyor belt passage, indoor environment of the equipment.

[0074] Execution state parameters such as valve opening and pump speed, purpose: to record the current state of the actuator in real time and use it as feedback to judge whether the adjustment result takes effect; sensor types: angle encoder, rotational speed measuring instrument; acquisition locations: automatic valve, water supply pump, electromagnetic flow regulating mechanism.

[0075] It should be noted here that the data collected by the process parameter monitoring module are all uploaded to the system central controller in the form of digital signals in real time and processed synchronously with the image feature data. The collected parameters are not only used for dynamic modeling and calculation of the optimization objective function, but also provide the basis for closed-loop control of the system. After the strategy is executed, this module can also be used to compare the actual results with the predicted values to achieve self-correction and continuous optimization of the system.

[0076] The simulation modeling module is the core decision-making support unit in the present invention. Its function is to construct a mathematical model or data-driven model that can simulate the coal washing process according to the collected image feature data and process parameter data, and is used to predict the changes and mutual influences between multiple optimization objectives (such as yield, energy consumption, medium consumption, separation efficiency, etc.) under different control strategies.

[0077] The specific implementation methods of simulation modeling include:

[0078] Model type (selected according to application):

[0079] Physical modeling (such as modeling mass conservation and energy conservation equations): suitable for basic coal washing processes, such as heavy medium separation and flotation;

[0080] Data-driven models (mainstream methods): trained using historical data and real-time data, applicable to systems with complex couplings and difficult to accurately model; typical algorithms: multiple regression, random forest, neural networks (such as DNN, RNN), support vector machines (SVM), etc.;

[0081] Hybrid modeling: use physical models for some processes and machine learning to fill in the remaining parts.

[0082] The specific modeling steps are as follows:

[0083] Step 1: Extract image features such as particle size distribution, bulk density, and material flow rate from the image acquisition module; obtain real-time data such as medium concentration, water-coal ratio, screen frequency, and power consumption from the process parameter monitoring module; clean, normalize, and standardize the above data to form modeling samples.

[0084] Step 2: Define the output objectives of the simulation, such as: Objective 1: Maximize the clean coal recovery rate; Objective 2: Minimize the energy consumption; Objective 3: Minimize the medium loss; Objective 4: Minimize the separation error; Express the objective function in mathematical forms such as linear weighting and sum of squared errors.

[0085] Step 3: Select a suitable algorithm (such as a neural network) to establish a mapping function: Input: X = [image features + process parameters + control strategy variables]; Output: Y = [values of each optimization objective]; Perform cross-validation and error evaluation to prevent overfitting of the model.

[0086] Step 4: Train the model with historical data and update it in real-time using the latest data; Set the training set and validation set; Adjust hyperparameters (such as the number of network layers, learning rate) to improve the simulation accuracy.

[0087] Step 5: Perform simulation predictions for different combinations of control strategies and output corresponding multiple target response values; Analyze the mutual influence and trade-off relationships between the targets.

[0088] Transfer the multi-objective results output by the simulation to the multi-objective optimization module (such as the Pareto algorithm) for generating the optimal control strategy.

[0089] The multi-objective optimization module is used to generate an optimal control strategy combination in the intelligent monitoring system for the image processing of coal washing equipment, based on multiple optimization objectives (such as clean coal recovery rate, energy consumption, medium consumption, separation efficiency, etc.). The system forms an optimization solution set through the Pareto front algorithm, and further uses the response difference degree calculation to establish a response space redistribution model to sparsify the dense solution set, and finally outputs a sparse optimization solution set with good response discrimination, improving the system response efficiency and control stability.

[0090] Receive input data from the simulation modeling module, including: image feature parameters (such as particle size distribution, stacking thickness); process operation parameters (such as medium concentration, water-coal ratio, screen frequency, etc.); control strategy variables (such as adjusting water supply flow rate, screen machine speed, etc.).

[0091] Define multiple objectives to be optimized simultaneously: Objective 1: Maximize the clean coal yield; Objective 2: Minimize the system energy consumption; Objective 3: Minimize the medium consumption; Objective 4: Minimize the separation error; Standardize the objective functions into comparable numerical function forms for unified optimization.

[0092] Construct an initial population containing multiple groups of feasible control strategy parameters, with each group of control parameters being a solution vector; each solution vector is input into the simulation model to output the predicted values of each objective.

[0093] Use NSGA-II, MOEA / D or other Pareto front algorithms to iterate the population; select a group of solutions that have no obvious disadvantages under multiple objectives according to the non-dominated sorting principle to form a preliminary Pareto optimal solution set.

[0094] For any two solutions in the preliminary Pareto solution set, calculate their response difference degrees (such as Euclidean distance, Manhattan distance or Mahalanobis distance) in the output objective space; the difference degree represents the distinguishability of the two groups of strategies in actual operation, and the larger the value, the more obvious the difference in control behavior.

[0095] Use the response difference degree as a metric to construct a response distribution map in the objective space; use clustering algorithms (such as K-means, DBSCAN) to cluster the solution set; merge or extract representative points for redundant solutions with similar responses in each cluster to obtain a sparser solution set with stronger representativeness.

[0096] Output the representative solution set obtained after response redistribution to form the final sparsified optimization solution set; this solution set retains the overall diversity of the solution set while excluding similar, repeated, and non-response-different strategy combinations.

[0097] The strategy scoring module is used to comprehensively evaluate each control strategy in the sparse optimization solution set output by the multi-objective optimization module. By introducing multi-source data such as image features, historical operation data, and system response trajectories, it dynamically scores the actual adaptability and control effectiveness of each strategy, and selects the optimal strategy according to the scoring results to achieve intelligent decision-making and dynamic adjustment of the system.

[0098] Receive the sparse optimization solution set output by the multi-objective optimization module, where each control strategy contains a set of specific control parameters (such as screening frequency, water volume, medium concentration adjustment value, etc.);

[0099] Obtain the image recognition features under the current working condition from the image acquisition module, including coal block particle size, material distribution, movement trajectory, etc.; convert the image features into structured vectors for subsequent model matching and scoring.

[0100] Call the historical operation records stored in the system database. The data includes: the usage records of each control strategy under similar working conditions; the corresponding actual effect evaluations (such as recovery rate, energy consumption, error rate); environmental and equipment conditions (such as season, humidity, material properties);

[0101] Read the actual response trajectory of the system after executing the strategy from the feedback adjustment module, including: system execution delay; process index response time; fluctuation amplitude after adjustment; actual target deviation change;

[0102] Establish a scoring function S(i) based on three factors: image features, historical data, and response trajectories, for comprehensively evaluating each strategy i;

[0103] For example, the calculation expression of the scoring function can be: ; where: is the matching degree score between the strategy parameters and the current image features, is the average effect score of the strategy under similar working conditions, is the deviation control ability score after the strategy is executed; is an adjustable weight, adaptively optimized according to the scenario. Calculate the scoring function for each strategy in the sparse optimization solution set one by one; sort all the scoring results from high to low to generate a strategy priority list.

[0104] Among them, The acquisition method of is: set the image feature vector collected by the current system as: ; where each represents an image feature index (such as average particle size, bulk density, movement directionality, etc.); the average value of the image features applicable to strategy i in history is: ; then can be calculated as: ; in the formula, is the norm of the standard deviation vector of the full-history image feature data, representing the overall fluctuation range, and ε is a very small constant to prevent the denominator from being zero.

[0105] The historical performance factor is used to evaluate the actual execution effect of a certain control strategy under similar working conditions in the past. Its core goal is: if a certain strategy has been frequently applied and has good effects under similar coal quality, equipment status and process conditions, then it still has high adaptability and execution reliability under the current working conditions. This factor can prevent the system from selecting a strategy that is theoretically optimal but has a high historical execution failure rate.

[0106] For example, The acquisition method of is: Let the target effect score of strategy i in the past m similar working conditions be: , then the historical performance score is defined as: ; In the formula, is the comprehensive effect score (normalized, 0-1) of strategy i in the j-th similar working condition, and m is the number of similar working condition samples matched.

[0107] After the system executes a certain control strategy, its key outputs (such as clean coal yield, medium consumption, motor power, etc.) will change dynamically within a certain time range. By regarding this output as a dynamic response curve and fitting a first-order inertial system model, parameters reflecting the dynamic characteristics of the system are extracted, such as: response time constant and delay, to quantify the response stability.

[0108] For example, The acquisition method of is: After the control strategy i is executed, record the time series data of the target response quantity y(t) within the control period; the data sampling period is, for example, sampling once per second, and the duration is recommended to be no less than 3-5 times the expected time constant length. Model the system response as the standard first-order system response form: ; In the formula, y(t) is the actual output of the system (such as clean coal yield, power consumption, medium consumption, etc.), K is the system gain, representing the improvement amplitude of the final stable value relative to the initial value, τ is the time constant, describing the time required for the system to reach 63% of the steady-state response, the smaller the faster, is the response delay time, the lag of the system start change after the control strategy starts, is the response initial value, that is, the initial output value of the system before the control strategy.

[0109] Use the least squares fitting (such as the Levenberg-Marquardt algorithm) to estimate the parameters of the first-order inertial model to obtain the optimal fitting parameters: the response time constant , the delay time ; At the same time, calculate the fitting error (sum of squared residuals) as the goodness-of-fit index. Convert the fitting result into a score, that is, calculate to obtain , and the expression is: ; where, is the sum of squared fitting residuals of strategy i, which measures the error between the model and the true response curve, is the reference maximum response time constant set by the system (such as historical mean or industry standard), is the maximum allowable response delay time (such as 30 seconds), is the maximum value of the residual, which is used to normalize the fitting error. α, β, and γ are weight coefficients, reflecting the influence degree of the three items on the score (it is recommended that the default values are all 1 / 3).

[0110] The control module ranks according to the scores and preferentially selects the strategy with the highest score; if the system configuration allows multiple strategies to be reserved (such as dual-control mode), the top N high-score strategies can be retained for subsequent switching. The control parameters of the strategy with the highest score (such as pump speed, screen frequency, target value of medium concentration, etc.) are sent to the control execution module; the control module executes parameter adjustment.

[0111] The feedback regulation module is used to monitor and analyze the execution results of the control strategy sent by the control execution module in real time, and compare the actual equipment response with the prediction results of the simulation model. When the system detects a deviation between the actual operation result and the expectation, it can automatically correct the parameter model in the simulation modeling module, or dynamically adjust the score weights and factor structures in the strategy scoring module, so as to realize the adaptive learning and closed-loop optimization of the intelligent monitoring system.

[0112] Obtain the equipment response information after control execution from the process parameter monitoring module, such as: clean coal yield, power consumption, medium consumption, separation error; dynamic response time, steady-state value, fluctuation range; the acquisition time range covers the complete action cycle of the control strategy.

[0113] Call the target output value Ypred predicted by the simulation modeling module, compare it with the actual monitoring result Yreal, and calculate the target deviation vector : ; if the deviation exceeds the preset threshold (such as ±5%), the feedback correction mechanism is triggered.

[0114] According to the error fine-tune the model used for prediction in the modeling module, and the following correction methods can be adopted:

[0115] Update the regression model weights or neural network parameters; add the current sample as incremental data to the training set for retraining; reconstruct the local model or add auxiliary input variables when the continuous error is large; the updated model participates in the next round of control strategy simulation and evaluation again.

[0116] If the feedback analysis result shows that the scoring deviation mainly comes from the strategy scoring module (i.e., the scoring result fails to accurately reflect the execution effect): analyze the explanatory power of scoring factors such as image adaptability, historical performance, and response stability on the actual effect; for the weight coefficients in the scoring function

[0117] perform automatic correction; preferentially enhance the factors strongly correlated with the current response result and weaken the ineffective scoring items; the updated scoring model will participate in the subsequent strategy screening and scoring to improve the accuracy of strategy recommendation.

[0118] For example: after the strategy control, the system predicts the clean coal yield to be 86%, but the actual detection is 78%, and the deviation exceeds the allowable threshold of 5%. The feedback module is triggered to retrain the modeling model and adjust the scoring mechanism (such as strengthening the weight of the response stability factor), and preferentially screen the strategies with stronger control stability in the next round of strategy generation, successfully improving the yield to within the target value range.

[0119] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. An intelligent monitoring system for image processing of coal washing equipment based on policy optimization, characterized in that: It includes an image acquisition module, a process parameter monitoring module, a simulation modeling module, a multi-objective optimization module, a strategy scoring module, a control execution module, and a feedback adjustment module; The image acquisition module is used to obtain real-time image information of different parts of the coal washing equipment, identify and structurally process it, and form image feature data; The process parameter monitoring module is used to collect several process parameters during the coal washing process; The simulation modeling module constructs a dynamic simulation model based on the image feature data and process monitoring data, and is used to simulate the response relationship between multiple optimization objectives during the coal washing process; The multi-objective optimization module generates multiple control strategies based on the Pareto front algorithm, and by calculating the response difference degree between the strategies, establishes a response space redistribution model to sparsify the Pareto solution set and obtain a sparsified optimization solution set; The strategy scoring module dynamically scores the sparsified optimization solution set based on image features, historical operation data, and response trajectories, and selects the optimal strategy according to the scoring results; The control execution module adjusts the operating parameters of the coal washing equipment according to the selected optimal strategy; The feedback adjustment module is used to monitor the control execution results, and automatically correct the simulation model or adjust the strategy scoring parameters based on the equipment response situation and target deviation.

2. The intelligent monitoring system for image processing of coal washing equipment based on policy optimization according to claim 1, wherein: The image acquisition module includes: multiple groups of image sensors installed at the screen surface, sorting tank outlet, and conveyor belt discharge outlet of the coal washing equipment, and the image sensors include high-resolution industrial cameras, infrared thermal imagers, and laser profile scanners.

3. The intelligent monitoring system for image processing of coal washing equipment based on policy optimization according to claim 2, wherein: The image acquisition module is used to collect the original image data during the operation of the equipment; perform grayscale processing, edge enhancement, and noise removal on the images, identify the coal block boundaries and extract image features through target recognition algorithms, including particle size distribution, stacking thickness, movement direction, and coal-gangue ratio; output the processed image features in the form of a structured vector as input variables to the simulation modeling module.

4. The intelligent monitoring system for image processing of coal washing equipment based on policy optimization according to claim 1, characterized in that: The process parameter monitoring module is used to collect the process parameters during the coal washing process in real time, including medium concentration, water-coal ratio, feed particle size, vibration frequency, motor current, power consumption, water supply flow rate, and system temperature.

5. The intelligent monitoring system for image processing of coal washing equipment based on policy optimization according to claim 4, wherein: The simulation modeling module is used to construct a data-driven model: Extract the particle size distribution and material flow velocity image features from the image acquisition module; Obtain the water-coal ratio, medium concentration, and power consumption process data from the process parameter monitoring module; Normalize and process the collected data for outliers to form multi-dimensional modeling samples; Use neural network, support vector machine, or random forest algorithm to establish a data-driven model; Output the predicted values of each target parameter under the simulated control strategy for transmission to the multi-objective optimization module.

6. The intelligent monitoring system for image processing of coal washing equipment based on policy optimization according to claim 5, wherein: After generating the Pareto solution set, the multi-objective optimization module performs the sparsification processing steps: Calculate the difference degree of the target responses corresponding to any two groups of control strategies in the solution set, and the difference degree is measured by the Euclidean distance or Mahalanobis distance; Establish a target space redistribution model based on the response difference degree; Use the clustering algorithm to cluster the solution set; Select the representative solution with the largest response difference in each cluster to form the sparsified optimization solution set; Output the sparse optimization result to the strategy scoring module.

7. The intelligent monitoring system for image processing of coal washing equipment based on policy optimization according to claim 6, characterized in that: The strategy scoring module includes: Calculate the matching score between the calculation strategy parameters and the current image features: The matching score is calculated through the Euclidean distance between the current image features and the mean of the historical image features of the strategy; Calculate the average effect score of the strategy under similar working conditions: Statistically calculate the average execution effect of the strategy under similar working conditions and take the average value of multiple scores; Calculate the deviation control ability score after the strategy is executed: Fit the actual system response to a first-order response model, extract the response time constant, delay time, and fitting error, and construct a scoring function; Fuse the matching score, average effect score, and deviation control ability score according to the weighted formula and output the comprehensive score result; Sort the strategies according to the score from high to low, and select the one with the highest score as the optimal control strategy.

8. The intelligent monitoring system for image processing of coal washing equipment based on policy optimization according to claim 7, wherein: The feedback adjustment module includes: Sample the indicators in the control execution result; Compare with the simulation model prediction result and calculate the target deviation vector; If the deviation exceeds the preset threshold, trigger the fine-tuning mechanism of the modeling module, including model incremental training, parameter update, or structure reconstruction.

Images