Intelligent control method and system for screening recarburizer

Abstract

The present application belongs to the field of intelligent control, and particularly relates to a kind of carbon additive screening intelligent control method and system, the method comprises: first, the screening efficiency is weighted with dimensionless particle size distribution penalty term, and is quantified as unified comprehensive performance index;Then the nonlinear relationship between the index and control input is probabilistically modeled using a deep Gaussian process model, and implicit constraints such as blockage and overload are identified using a Gaussian process classifier to define a dynamic feasible region;Within the dynamic feasible region, efficient Bayesian optimization is performed through the construction of an asymmetric information gain acquisition function to search for the optimal operating condition;The particle filtering framework is integrated with the acquisition function to achieve dynamic tracking of time-varying targets;Finally, the system operates according to the recommended input and updates all models online using new data to form an adaptive closed loop.The present application solves the optimization problem of dynamic, nonlinear and implicit constraint screening process, significantly improving the intelligent level, safety and economic benefit of control.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control, and in particular relates to an intelligent control method and system for carbon raiser sieving. Background Technology

[0002] Carbon raisers are a commonly used auxiliary material in the metallurgical industry, used to adjust the carbon content in molten steel or iron. The uniformity of the particle size distribution and the screening efficiency of carbon raisers directly affect the carbon raising effect, melting rate, yield, and production cost in subsequent smelting processes. Screening, as a key step in the carbon raiser production process, is typically measured by two indicators: screening efficiency and product particle size distribution. These two indicators are in turn influenced by a complex coupling of various process parameters such as vibration frequency, screen inclination angle, and feed rate.

[0003] In actual production, the physical properties of the carbon raiser raw materials (such as moisture content, particle shape, and hardness) fluctuate from batch to batch, and the operating status of the screening equipment (such as screen wear and tension changes) also evolves over time. These uncertainties often cause the optimal operating conditions of the screening process to drift. Current industrial practices mostly rely on manual adjustments based on the operator's experience, which is not only slow to respond but also difficult to maintain optimal operating conditions for a long time, often resulting in decreased screening efficiency, unstable product particle size, and increased energy consumption.

[0004] To address these issues, researchers have attempted to employ traditional optimization methods such as mechanistic modeling and response surface methodology. However, these methods are typically based on static assumptions, requiring extensive offline experiments to build models, and once determined, they struggle to adapt to dynamic operating conditions. Furthermore, traditional methods do not adequately consider process uncertainties, particularly implicit constraints such as screen clogging and equipment overload, which are difficult to model explicitly, thus impacting the safety and stability of the screening process.

[0005] Therefore, existing technologies still lack an intelligent control method that can model screening performance in real time, dynamically track optimal operating conditions, and update the model online while ensuring equipment safety, in order to adapt to the complexity, time-varying nature, and uncertainty of the carbon raiser screening process. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent control method for carbonizer sieving, which addresses the complexity, time-varying nature, and uncertainty of the carbonizer sieving process, and includes the following steps:

[0007] Vibration frequency, screen inclination angle, and feed rate are obtained as control inputs, and screening efficiency and product particle size distribution are quantified into a unified comprehensive performance index. A deep Gaussian process model is obtained by probabilistically modeling the relationship between the comprehensive performance index and the control inputs. The deep Gaussian process model uses a deep kernel function to identify nonlinear characteristics.

[0008] A Gaussian process classifier is constructed to probabilistically model implicit constraints including screen blockage and / or equipment overload. The dynamic feasible region is determined based on the prediction results of the Gaussian process classifier. Within the dynamic feasible region, an acquisition function is constructed using asymmetric information gain based on the posterior prediction distribution output by the deep Gaussian process model.

[0009] A state-space model describing the dynamic evolution of the optimal operating condition is constructed. A particle filter framework is adopted to propagate the posterior distribution of the optimal operating condition at the previous moment through the state-space model to obtain the prior distribution at the current moment. The acquisition function is used to guide importance sampling and resampling to generate the next set of recommended control inputs.

[0010] The operation of the screening system is adjusted according to the recommended control input, and the deep Gaussian process model and the Gaussian process classifier are updated online using the newly collected performance data and constraint states.

[0011] Preferably, quantifying screening efficiency and product particle size distribution into a unified comprehensive performance index includes:

[0012] Obtain screening efficiency and product particle size ;

[0013] Set the target range for product granularity as , and d max >d min ;

[0014] Define a dimensionless product granularity distribution penalty term. :

[0015] ;

[0016] Calculate comprehensive performance indicators ,in and These are preset positive weighting coefficients. This is the weighting coefficient for screening efficiency. This is the product granularity weighting coefficient.

[0017] By directly obtaining screening efficiency and product particle size These two crucial physical quantities provide raw and accurate data input for performance evaluation; by setting the target range for product granularity... This study clarifies the "acceptable" range for product quality, providing a benchmark for subsequent calculations of particle size distribution deviations, thus making quality control objectives more specific and quantifiable. By defining a dimensionless product particle size distribution penalty term P(d), a penalty value is only generated when the particle size d exceeds the target range, and the magnitude of the penalty value is proportional to the degree of deviation. Normalization of the denominator makes it dimensionless, facilitating combination with other indicators of different units, such as screening efficiency, greatly simplifying the complexity of multi-objective trade-offs. Finally, a unified performance index J is constructed through linear weighted summation (efficiency as a positive incentive, penalty term as a negative incentive). The introduction of weighting coefficients w1 and w2 allows for flexible adjustment of the relative importance of the two objectives based on actual production needs (e.g., prioritizing quality or quantity), providing an effective tool for achieving customized, economically efficient optimization strategies.

[0018] Preferably, the step of using a deep Gaussian process to probabilistically model the relationship between the comprehensive performance index and the control input to obtain a deep Gaussian process model includes:

[0019] The control input is used as the input to the first-level Gaussian process to generate latent variables;

[0020] The latent variables are used as input to the second-level Gaussian process to output the posterior prediction distribution of the comprehensive performance index.

[0021] The deep Gaussian process model with the aforementioned two-layer structure can simulate the intrinsic logical chain from "control action" to "intermediate physical state" and then to "final performance." The first layer maps the control input to a set of abstract latent variables, and the second layer links these latent variables to the final performance index. This structure not only endows the model with powerful nonlinear fitting capabilities to improve prediction accuracy, but its complete probability distribution (rather than a single predicted value) provides crucial uncertainty information for subsequent Bayesian optimization, risk assessment, and exploratory decision-making, thereby significantly enhancing the robustness and intelligence of the control strategy.

[0022] Preferably, the prediction results based on the Gaussian process classifier determine the dynamic feasible region, including:

[0023] For any candidate point in the control input space, the probability that the candidate point will not experience screen clogging or equipment overload is predicted using the Gaussian process classifier. ;

[0024] All satisfied The set of control input points that are greater than or equal to the safety probability threshold is defined as the dynamic feasible region at the current moment.

[0025] By using a Gaussian process classifier to predict the safety probability P(safe|x) of any candidate control point, and combining all points with safety probabilities greater than or equal to a preset threshold, a dynamic feasible region is formed. This design achieves proactive risk avoidance, moving from "post-event alarm" to "pre-event prediction." More importantly, this safety boundary can be updated online based on newly collected data, achieving an intelligent dynamic balance between safety and performance.

[0026] Preferably, the particle filter framework is used to propagate the posterior distribution of the optimal operating condition at the previous time step through the state-space model to obtain the prior distribution at the current time step. The acquisition function is then used to guide importance sampling and resampling to generate the next set of recommended control inputs, including:

[0027] a) The set of particles representing the posterior distribution of the optimal working condition in the previous moment is propagated through the state space model to obtain the prior particle set at the current moment.

[0028] b) Calculate the acquisition function value of the control input for each particle in the prior particle set, and use the acquisition function value as the importance weight of the particle;

[0029] c) Based on the importance weights calculated in step b), resample from the prior particle set to form the posterior particle set at the current moment;

[0030] d) From the posterior particle set, select the control input corresponding to the particle with the highest predicted mean of comprehensive performance index as the next set of recommended control inputs.

[0031] This particle filtering framework uses the acquisition function value to guide particle weight allocation and resampling, enabling the particle cloud to intelligently focus on the most valuable areas to explore, rather than blindly tracking. Ultimately, it makes optimal decisions from a pool of "elite" particles selected through a process of natural selection, ensuring that the system not only tracks the time-varying optimal target quickly and accurately, but also that each recommended operation balances the wisdom of exploration with the efficiency of production. In other words, it deeply integrates the "efficient exploration" capability of Bayesian optimization with the "dynamic tracking" capability of particle filtering to solve the problem of dynamic drift of the optimal operating point caused by changes in raw material characteristics or equipment status.

[0032] Preferably, the step of updating the deep Gaussian process model and the Gaussian process classifier online using newly acquired performance data and constraint states includes:

[0033] The newly collected control inputs, corresponding comprehensive performance indicators, and constraint states are added to the historical database.

[0034] The hyperparameters of the deep Gaussian process model and the Gaussian process classifier are updated using a gradient optimization method.

[0035] By adding valuable data generated from each online run to the database and using gradient optimization algorithms to update model hyperparameters online, the system ensures that its two core models (deep Gaussian process model and Gaussian process classifier) ​​can always keep up with the slow changes in process characteristics. This online update mechanism guarantees the long-term effectiveness and advancement of the control strategy, enabling the model to continuously approximate the real production process characteristics. This maximizes the value of data and continuously optimizes system performance, overcoming the limitation of offline models gradually becoming ineffective due to "model-reality mismatch" during long-term operation, and endowing the entire control system with the ability to continuously learn and adapt.

[0036] Furthermore, the present invention also provides an intelligent control system for carbon raiser sieving, comprising the following units:

[0037] The control input acquisition unit acquires the vibration frequency, screen inclination angle, and feeding rate as control inputs, and quantifies the screening efficiency and product particle size distribution into a unified comprehensive performance index. A deep Gaussian process model is obtained by probabilistically modeling the relationship between the comprehensive performance index and the control inputs. The deep Gaussian process model uses a deep kernel function to identify nonlinear features.

[0038] The probabilistic modeling unit constructs a Gaussian process classifier to probabilistically model implicit constraints including screen blockage and / or equipment overload, and determines the dynamic feasible region based on the prediction results of the Gaussian process classifier; within the dynamic feasible region, an acquisition function is constructed using asymmetric information gain based on the posterior prediction distribution output by the deep Gaussian process model.

[0039] The model building unit constructs a state-space model describing the dynamic evolution of the optimal working condition. Using a particle filter framework, the posterior distribution of the optimal working condition at the previous moment is propagated through the state-space model to obtain the prior distribution at the current moment. The acquisition function is used to guide importance sampling and resampling to generate the next set of recommended control inputs.

[0040] The control and update unit adjusts the operation of the screening system according to the recommended control input, and updates the deep Gaussian process model and the Gaussian process classifier online using newly acquired performance data and constraint states.

[0041] Preferably, quantifying screening efficiency and product particle size distribution into a unified comprehensive performance index includes:

[0042] Obtain screening efficiency and product particle size ;

[0043] Set the target range for product granularity as , and d max >d min ;

[0044] Define a dimensionless product granularity distribution penalty term. :

[0045] ;

[0046] Calculate comprehensive performance indicators ,in and These are preset positive weighting coefficients. This is the weighting coefficient for screening efficiency. This is the product granularity weighting coefficient.

[0047] Preferably, the step of using a deep Gaussian process to probabilistically model the relationship between the comprehensive performance index and the control input to obtain a deep Gaussian process model includes:

[0048] The control input is used as the input to the first-level Gaussian process to generate latent variables;

[0049] The latent variables are used as input to the second-level Gaussian process to output the posterior prediction distribution of the comprehensive performance index.

[0050] Preferably, the prediction results based on the Gaussian process classifier determine the dynamic feasible region, including:

[0051] For any candidate point in the control input space, the probability that the candidate point will not experience screen clogging or equipment overload is predicted using the Gaussian process classifier. ;

[0052] All satisfied The set of control input points that are greater than or equal to the safety probability threshold is defined as the dynamic feasible region at the current moment.

[0053] Preferably, the particle filter framework is used to propagate the posterior distribution of the optimal operating condition at the previous time step through the state-space model to obtain the prior distribution at the current time step. The acquisition function is then used to guide importance sampling and resampling to generate the next set of recommended control inputs, including:

[0054] a) The set of particles representing the posterior distribution of the optimal working condition in the previous moment is propagated through the state space model to obtain the prior particle set at the current moment.

[0055] b) Calculate the acquisition function value of the control input for each particle in the prior particle set, and use the acquisition function value as the importance weight of the particle;

[0056] c) Based on the importance weights calculated in step b), resample from the prior particle set to form the posterior particle set at the current moment;

[0057] d) From the posterior particle set, select the control input corresponding to the particle with the highest predicted mean of comprehensive performance index as the next set of recommended control inputs.

[0058] Preferably, the step of updating the deep Gaussian process model and the Gaussian process classifier online using newly acquired performance data and constraint states includes:

[0059] The newly collected control inputs, corresponding comprehensive performance indicators, and constraint states are added to the historical database.

[0060] The hyperparameters of the deep Gaussian process model and the Gaussian process classifier are updated using a gradient optimization method.

[0061] In summary, compared with the prior art, the present invention has at least the following beneficial effects:

[0062] This invention constructs a deep Gaussian process model, which captures the complex nonlinear relationship between control inputs and comprehensive performance indicators during the carbonizer screening process, quantifies the uncertainty of the model, and improves the reliability of prediction and control. Furthermore, this invention utilizes a Gaussian process classifier to probabilistically identify implicit constraints that are difficult to model, such as screen blockage and equipment overload, defining the dynamic feasible region for safe operation, reducing potential production risks, and ensuring the stability and safety of equipment operation. Based on this, by constructing an asymmetric information gain acquisition function, efficient search and convergence of the optimal operating condition are achieved. This invention can effectively track changes in the optimal operating point caused by fluctuations in raw material characteristics or evolution of equipment status, enabling the screening process to continuously approach and maintain the optimal operating point, improving screening efficiency and the uniformity of product particle size distribution without relying on human experience. Attached Figure Description

[0063] Figure 1 A flowchart of steps S1-S4 in an intelligent control method for carbon raiser sieving according to an embodiment of the present invention;

[0064] Figure 2 This is a schematic diagram of the deep Gaussian process model structure in an embodiment of the present invention;

[0065] Figure 3 This is a schematic diagram of the dynamic feasible region at a certain moment in an embodiment of the present invention;

[0066] Figure 4 This is a schematic diagram of the asymmetric information gain acquisition function in an embodiment of the present invention;

[0067] Figure 5 This is a schematic diagram of the optimal working condition for particle filter tracking in an embodiment of the present invention;

[0068] Figure 6 This is a structural block diagram of an intelligent control system for carbon raiser sieving according to an embodiment of the present invention. Detailed Implementation

[0069] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0070] The control method of this invention is implemented on a typical carbon raiser screening production line, which mainly includes a vibrating screening device with adjustable parameters, a frequency converter for controlling the vibration frequency, a servo motor and angle sensor for controlling the screen surface inclination angle, a material conveyor for controlling the feeding rate, an online particle size analyzer for measuring the particle size distribution of the product, a weighing sensor for calculating the screening efficiency, a motor current sensor and a material blockage sensor for monitoring the constraint state, and an industrial computer; wherein, the industrial computer communicates with the above-mentioned actuators and sensing systems via a bus, and the feeding speed of the material conveyor is controlled by an independent frequency converter.

[0071] Reference Figure 1 A method for intelligent control of carbon raiser sieving, specifically including the following steps:

[0072] S1, obtain the vibration frequency, screen inclination angle and feeding rate as control inputs, and quantify the screening efficiency and product particle size distribution into a unified comprehensive performance index; use a deep Gaussian process to probabilistically model the relationship between the comprehensive performance index and the control inputs to obtain a deep Gaussian process model, and use a deep kernel function to identify nonlinear features in the deep Gaussian process model;

[0073] The vibration frequency is set and read by a frequency converter, and the screen inclination angle is controlled and obtained by an angle sensor and servo motor. The feeding rate is controlled by adjusting the speed or amplitude of the belt conveyor or electromagnetic vibrating feeder. At the system output, the particle size distribution of the product is measured by image analysis or an online particle size analyzer. The proportion of particles within the preset target particle size range or the Wasserstein distance from the target distribution is calculated as the particle size distribution index. The mass of the material under the screen is measured by a weighing sensor to calculate the screening efficiency. A weighted summation method is used, for example, multiplying the screening efficiency by the negative particle size distribution deviation by their respective economic weight coefficients and then adding them together to form a scalar comprehensive performance index. .

[0074] A multi-layered deep Gaussian process model is constructed. The first layer maps a three-dimensional input vector consisting of vibration frequency, screen inclination angle, and feed rate to a high-dimensional latent space. The input of each subsequent Gaussian process layer is the output of the previous layer. The output of the final Gaussian process layer represents the comprehensive performance index. The predicted mean and variance, and a schematic diagram of the deep Gaussian process model are shown below. Figure 2 As shown, the hierarchical structure is equivalent to a deep kernel function that is automatically learned from the data, which can obtain highly nonlinear and non-stationary relationships between input and output.

[0075] S2, Construct a Gaussian process classifier to probabilistically model implicit constraints including screen blockage and / or equipment overload, and determine the dynamic feasible region based on the prediction results of the Gaussian process classifier; Within the dynamic feasible region, construct an acquisition function based on the posterior prediction distribution output by the deep Gaussian process model using asymmetric information gain.

[0076] During operation, material blockage sensors, motor current monitoring, or acoustic signal analysis are used to determine whether screen blockage or equipment overload has occurred. These are categorized as Category 1 and normal as Category 0. The control inputs and their corresponding category labels form a training dataset for training a Gaussian process classifier. This classifier can predict the probability of a constraint violation event for any new set of control inputs. A safety probability threshold, such as 95%, is set. All combinations of control inputs that result in a predicted constraint violation probability below this threshold constitute the dynamic feasible region at the current moment. Figure 3 As shown.

[0077] Within the dynamically feasible region defined by a Gaussian process classifier, a series of candidate control input points are evaluated; for each candidate point, the posterior mean and variance of the comprehensive performance index of the point are predicted using a deep Gaussian process model; the acquisition function is preferably in the form of asymmetric information gain, such as... Figure 4 As shown, the value function of the acquisition function not only considers the reduction in uncertainty about the optimal value location brought about by the sampling candidate points, i.e., entropy reduction, but also gives extra weight to regions whose posterior mean is higher than the current optimal value. That is, it gives higher weight to exploring high-potential regions and lower weight to exploring low-performance regions or explorations that are purely for reducing uncertainty. The candidate point that maximizes the asymmetric information gain is selected as the test case for the next step.

[0078] In a more specific embodiment, within the dynamic feasible region, an acquisition function is constructed using asymmetric information gain based on the posterior prediction distribution output by the deep Gaussian process model. Specifically, a safety probability-weighted expected improvement function is constructed. This function first uses the deep Gaussian process model to calculate the expected improvement in system performance at a candidate control input point compared to the current optimal value. Then, it uses a Gaussian process classifier or a new model to predict the safety probability that blockage or overload will not occur at that point. The value of the acquisition function is the product of these two factors, i.e., safety probability × expected improvement. Only those points predicted to be both safe and have high improvement potential receive high scores, thereby guiding the system to find the optimal operating condition most efficiently while avoiding risks.

[0079] S3. Construct a state-space model describing the dynamic evolution of the optimal working condition. Using a particle filter framework, the posterior distribution of the optimal working condition at the previous moment is propagated through the state-space model to obtain the prior distribution at the current moment. The acquisition function is used to guide importance sampling and resampling to generate the next set of recommended control inputs.

[0080] Assuming the optimal operating point evolves over time according to a Gaussian random walk model, meaning the current optimal operating condition is based on the previous optimal operating condition with a small random perturbation; using particle filtering, a set of particles representing the probability distribution of the optimal operating condition position in the previous time step is propagated through this random walk model to obtain the prior particle set of the current optimal operating condition; using the asymmetric information gain acquisition function constructed in the previous step as the importance density function, the propagated particles are weighted, meaning particles with larger acquisition function values ​​are assigned higher weights; resampling is performed based on the weights to generate a new generation of particle sets, from which the particle that maximizes the acquisition function value is selected as the recommended control input for the current time step.

[0081] In one embodiment, the step of using the acquisition function to guide importance sampling and resampling to generate the next set of recommended control inputs specifically involves: acquiring a priori particle set representing the possible locations of optimal operating conditions, where each particle is a specific set of control inputs, such as vibration frequency, screen inclination angle, and feed rate. For each particle, an acquisition function value is calculated, i.e., safety probability × expected improvement, and this value is directly used as the importance weight of that particle. Particles located in more promising regions (safe and with high performance improvement potential) will receive higher weights. Resampling is performed based on these weights: high-weight particles are replicated multiple times, while low-weight particles are eliminated. To provide a definite operating instruction, the particle with the highest average comprehensive performance index predicted by the deep Gaussian process model is selected from this refined posterior particle set. Its corresponding control input combination is used as the next set of optimal operating conditions recommended to the screening system for execution.

[0082] S4. Adjust the operation of the screening system according to the recommended control input, and update the deep Gaussian process model and the Gaussian process classifier online using the newly collected performance data and constraint status.

[0083] The vibration frequency, screen inclination angle, and feed rate values ​​recommended by the particle filter are sent to the underlying PID controller or equipment drive unit for execution. After a stable operation cycle, the corresponding measured values ​​of comprehensive performance indicators and the labels indicating whether constraints are triggered are collected. This new set of data points, namely control inputs, performance indicators, and constraint labels, is added to their respective historical databases. The hyperparameters of the deep Gaussian process model and the Gaussian process classifier are then retrained or updated online, enabling both models to absorb the latest information in real time and adjust their predictions accordingly to reflect changes in the process.

[0084] In one embodiment, quantifying screening efficiency and product particle size distribution into a unified comprehensive performance index includes:

[0085] Obtain screening efficiency and product particle size ;

[0086] Set the target range for product granularity as , and d max >d min ;

[0087] Define a dimensionless product granularity distribution penalty term. :

[0088] ;

[0089] Calculate comprehensive performance indicators ,in and These are preset positive weighting coefficients. This is the weighting coefficient for screening efficiency. This is the product granularity weighting coefficient.

[0090] In screening, screening efficiency The ratio of the mass of material passing through the sieve to the mass of the feed material, with a value ranging from [0% to 100%]. Product particle size. The unit is millimeters, d min and d max These represent the lower and upper limits of the product particle size, respectively. Assume the target product particle size range is set to 0.1 mm to 0.5 mm. If the screening efficiency is measured at a certain moment... The product particle size is 92%. The particle size is 0.6 mm, which exceeds the target upper limit. A penalty term is calculated according to the formula. The value is 0.25. Assume the weight... Set to 1, Setting it to 2 to emphasize the importance of granularity control, the overall performance index is... The calculated value is 0.42.

[0091] Under another operating condition, if the screening efficiency 90%, product particle size The particle size is 0.4 mm, which falls within the ideal range, and a penalty term is applied. The calculated value is 0. At this point, the overall performance index... The value is 0.90. By comparing the two operating conditions... The value of 0.90 is much higher than 0.42, indicating that even though the screening efficiency is slightly lower, the latter performs better because the product particle size meets the requirements. This allows the optimization algorithm to clearly distinguish the advantages and disadvantages of different control strategies and find the optimal balance point to achieve the overall production goal.

[0092] In one embodiment, the step of using a deep Gaussian process to probabilistically model the relationship between the comprehensive performance index and the control input to obtain a deep Gaussian process model includes:

[0093] The control input is used as the input to the first-level Gaussian process to generate latent variables;

[0094] The latent variables are used as input to the second-level Gaussian process to output the posterior prediction distribution of the comprehensive performance index.

[0095] Understanding complex production processes can be achieved through a hierarchical structure. For example, the vibration frequency and amplitude of a vibrating screen are fed as control inputs into a first-level Gaussian process. This first level does not directly predict the final performance but transforms these physical inputs into a set of abstract latent variables. These latent variables may correspond to intermediate process states that cannot be directly measured, such as the stratification effect of the material on the screen surface or the throwing intensity.

[0096] The latent variables representing intermediate states are used as input to the second-level Gaussian process, which is responsible for modeling the intermediate states and the final comprehensive performance index. The mapping relationship between them. Its output is not a fixed value. It's not a single value, but a probability distribution, such as a predicted mean of 0.85 and a variance of 0.05. Variance represents the degree of uncertainty the model has about the current prediction.

[0097] In one embodiment, the prediction results based on the Gaussian process classifier determine the dynamic feasible region, including:

[0098] For any candidate point in the control input space, the probability that the candidate point will not experience screen clogging or equipment overload is predicted using the Gaussian process classifier. ;

[0099] All satisfied The set of control input points that are greater than or equal to the safety probability threshold is defined as the dynamic feasible region at the current moment.

[0100] To ensure the stability and safety of the production process, a dynamic safety boundary definition mechanism is introduced. A Gaussian process classifier trained on historical data is used to assess the risk of equipment failure that might result from any set of potential control inputs, such as a feed rate of 50 tons per hour and a screen inclination angle of 15 degrees. The Gaussian process classifier outputs a safety probability; for example, for the above combination, the predicted probability of safe operation is 98%.

[0101] A pre-set safety probability threshold, such as 95%, is established. This threshold represents the acceptable risk level for production management, and is typically set between 90% and 99%. All control input points with a safety probability of at least 95% after evaluation by the Gaussian process classifier collectively constitute a safe operating region, i.e., the dynamic feasible region. If another set of candidate control inputs, such as a feed rate of 80 tons per hour and a screen inclination angle of 12 degrees, has a predicted safety probability of only 60%, then this input point will be excluded from the dynamic feasible region. As new data is continuously collected, such as when the motor current approaches the alarm value during an operation, the Gaussian process classifier will be updated online, and the boundary of the dynamic feasible region will be adjusted accordingly, ensuring that the optimization algorithm always seeks the optimal operating condition within a dynamically updated safety range.

[0102] The acquisition function quantifies the expected reduction in posterior distribution entropy at the optimal operating position due to sampling at candidate points. Asymmetric weights are introduced to distinguish the information value of performance-enhancing and degrading regions. The acquisition function is the core of the Bayesian optimization algorithm; its main function is to guide the algorithm in each iteration to select the next most valuable sample point, i.e., to control the input for evaluation, i.e., to run on the actual system and observe the results. The information gain acquisition function selects a point x, and after evaluation, obtains y, which allows for a more accurate assessment of the optimal solution position. posterior probability distribution The entropy decreases the most. Asymmetric information gain assumes that different types of information have different values. The process of constructing the asymmetric information gain acquisition function is as follows:

[0103] Define the value function The value function measures the value or utility derived from observing a certain outcome y, i.e., the value of the overall performance index J. It is asymmetric. For example, when the y-value is higher than the current optimal value... At that time, its value Rapid growth; while when the y-value is lower than At this point, the value is very small or zero. In constrained problems, if a point x leads to failure such as blockage, the resulting value y is negative or extremely low. This invention does not limit the specific form of the value function.

[0104] Calculating the expected information value: Acquisition function The value of is the value V(y) for all possible observations y, given its predicted probability. The expectations below. It is given by the deep Gaussian process model, representing the posterior predicted distribution of y observed at point x. Mathematically, it can be expressed as:

[0105]

[0106] Where x is the candidate control input vector (e.g., vibration frequency, screen inclination angle, feed rate, etc.); D is the current historical dataset; The posterior predicted distribution at point x is obtained based on the Deep Gaussian Process Model (DGP), and the random variable y represents the possible observed value of the comprehensive performance index J; the above integral calculation is the expected value that can be obtained by conducting an experiment at point x.

[0107] More complex asymmetric information gains can modify how information entropy is calculated. For example, when calculating KL divergence, different weights can be assigned to different result regions such as "regions with significant performance improvements", "regions with minor performance improvements", "regions with performance degradation", and "regions with system failures".

[0108] The data acquisition function guides the next steps of exploration to find the optimal operating condition most efficiently. Not all unknown areas deserve equal exploration. Assume the currently known optimal overall performance index... The model performance is 0.8. Now we evaluate two candidate sampling points A and B. Candidate point A is located in a region where the model prediction performance may reach 0.9 but with high uncertainty, while candidate point B is located in a region where the model prediction performance is approximately 0.6 and also with high uncertainty.

[0109] Acquisition functions might assign similar scores to both points because they significantly reduce uncertainty. However, using asymmetric information gain rewards regions that may offer performance improvements. For candidate point A, its information value is amplified by a larger weighting coefficient because it promises to discover a better operating condition than the current optimum of 0.8. For candidate point B, although information is also obtained, it concerns a known suboptimal region, and its value is diminished by a smaller weighting coefficient. Therefore, assigning a higher acquisition function score to point A and prioritizing it for the next trial accelerates convergence to higher-performance regions and avoids wasting valuable trial opportunities in inefficient areas.

[0110] In one embodiment, the particle filter framework is used to propagate the posterior distribution of the optimal operating condition from the previous time step to the prior distribution of the current time step through the state-space model. The acquisition function is then used to guide importance sampling and resampling to generate the next set of recommended control inputs, including:

[0111] a) The set of particles representing the posterior distribution of the optimal working condition in the previous moment is propagated through the state space model to obtain the prior particle set at the current moment.

[0112] b) Calculate the acquisition function value of the control input for each particle in the prior particle set, and use the acquisition function value as the importance weight of the particle;

[0113] c) Based on the importance weights calculated in step b), resample from the prior particle set to form the posterior particle set at the current moment;

[0114] d) From the posterior particle set, select the control input corresponding to the particle with the highest predicted mean of comprehensive performance index as the next set of recommended control inputs.

[0115] A set of particles, say a thousand, is used to track the optimal operating condition that dynamically changes over time. In the first propagation phase, the thousand particles representing the optimal operating condition position at the previous moment move according to a state-space model describing the system's drift characteristics to predict the possible new position of the optimal operating condition at the current moment, forming a priori particle cloud, such as... Figure 5 As shown.

[0116] In the second step, weight calculation and resampling, the aforementioned asymmetric information gain acquisition function is used to evaluate each particle in the prior particle cloud. Particles falling into regions with high information value are assigned high weights, while those falling into less valuable regions are assigned low weights. Resampling is then performed based on these weights, with high-weight particles having a greater probability of being copied and retained, while low-weight particles may be eliminated. This causes the distribution of the particle cloud to gradually concentrate towards the most worthwhile regions to explore, forming the posterior particle set.

[0117] During the decision-making phase, for each particle in the posterior particle set, a deep Gaussian process model is used to predict the corresponding comprehensive performance index of the particle. The mean of the control inputs is used. For example, among 1,000 posterior particles after resampling, particle number 372, with a control input of 5 mm amplitude and 900 rpm frequency, achieves the highest predictive performance mean of 0.92. This set of control inputs is then selected as the recommended operation command for the next cycle and issued to the control system for execution. In an alternative embodiment, the point that maximizes the acquisition function is found within the entire dynamic feasible region and used as the next recommended control input, where the entire dynamic feasible region is all possible combinations of control inputs under safety constraints.

[0118] In one embodiment, updating the deep Gaussian process model and the Gaussian process classifier online using newly acquired performance data and constraint states includes:

[0119] The newly collected control inputs, corresponding comprehensive performance indicators, and constraint states are added to the historical database.

[0120] The hyperparameters of the deep Gaussian process model and the Gaussian process classifier are updated using a gradient optimization method.

[0121] After executing the previously recommended control commands, such as running for ten minutes at an amplitude of 5 mm and a frequency of 900 rpm, it will collect the actual operating data for that period. This data constitutes a complete data point, including the input control variables and the actual calculated comprehensive performance indicators. For example, 0.91, and the recorded constraint status, such as normal motor current and no material blockage, which is a safe state.

[0122] New data points are immediately added to the system's historical database. The updated dataset is then used to retrain or fine-tune the model. Gradient-based optimization algorithms allow adjustment of the internal hyperparameters of the deep Gaussian process model and the Gaussian process classifier, such as the length scale of the kernel function. For example, new data might reveal that performance is more sensitive to amplitude under current humidity conditions than the model initially anticipated; the hyperparameter updates will reflect this new information. This enables the model to continuously approximate the characteristics of the actual production process, adapting to slow changes such as raw material variations and equipment wear, ensuring the long-term effectiveness and accuracy of optimization recommendations.

[0123] Reference Figure 6 The present invention also provides an intelligent control system for carbon raiser sieving, which is used to execute any of the above-described embodiments of the intelligent control method for carbon raiser sieving; specifically, the system is an industrial computer configured with dedicated software, or an integrated control device; in this embodiment, the control system includes the following units:

[0124] The control input acquisition unit acquires the vibration frequency, screen inclination angle, and feeding rate as control inputs, and quantifies the screening efficiency and product particle size distribution into a unified comprehensive performance index. A deep Gaussian process model is obtained by probabilistically modeling the relationship between the comprehensive performance index and the control inputs. The deep Gaussian process model uses a deep kernel function to identify nonlinear features.

[0125] The probabilistic modeling unit constructs a Gaussian process classifier to probabilistically model implicit constraints including screen blockage and / or equipment overload, and determines the dynamic feasible region based on the prediction results of the Gaussian process classifier; within the dynamic feasible region, an acquisition function is constructed using asymmetric information gain based on the posterior prediction distribution output by the deep Gaussian process model.

[0126] The model building unit constructs a state-space model describing the dynamic evolution of the optimal working condition. Using a particle filter framework, the posterior distribution of the optimal working condition at the previous moment is propagated through the state-space model to obtain the prior distribution at the current moment. The acquisition function is used to guide importance sampling and resampling to generate the next set of recommended control inputs.

[0127] The control and update unit adjusts the operation of the screening system according to the recommended control input, and updates the deep Gaussian process model and the Gaussian process classifier online using newly acquired performance data and constraint states.

[0128] The above provides a detailed description of the intelligent control method and system for carbon raiser sieving provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and its core ideas. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for intelligent control of carbon raiser sieving, characterized in that, Includes the following steps: Vibration frequency, screen inclination angle, and feed rate are obtained as control inputs to quantify screening efficiency and product particle size distribution into a unified comprehensive performance index, including: Obtain screening efficiency and product particle size ; Set the target range for product granularity as , and d max >d min ; Define a dimensionless product particle size distribution penalty term. : ； Calculate comprehensive performance indicators ,in and These are preset positive weighting coefficients. This is the weighting coefficient for screening efficiency. This is the product granularity weighting coefficient; A deep Gaussian process model is obtained by probabilistically modeling the relationship between the comprehensive performance index and the control input using a deep Gaussian process model. The deep Gaussian process model uses a deep kernel function to identify nonlinear features. A Gaussian process classifier is constructed to probabilistically model implicit constraints including screen blockage and / or equipment overload. The dynamic feasible region is determined based on the prediction results of the Gaussian process classifier. Within the dynamic feasible region, an acquisition function is constructed using asymmetric information gain based on the posterior prediction distribution output by the deep Gaussian process model. A state-space model describing the dynamic evolution of the optimal operating condition is constructed. A particle filter framework is adopted to propagate the posterior distribution of the optimal operating condition at the previous moment through the state-space model to obtain the prior distribution at the current moment. The acquisition function is used to guide importance sampling and resampling to generate the next set of recommended control inputs. The operation of the screening system is adjusted according to the recommended control input, and the deep Gaussian process model and the Gaussian process classifier are updated online using the newly collected performance data and constraint states.

2. The intelligent control method for carbon raiser sieving according to claim 1, characterized in that, The process of using a deep Gaussian process to probabilistically model the relationship between the comprehensive performance index and the control input to obtain a deep Gaussian process model includes: The control input is used as the input to the first-level Gaussian process to generate latent variables; The latent variables are used as input to the second-level Gaussian process to output the posterior prediction distribution of the comprehensive performance index.

3. The intelligent control method for carbon raiser sieving according to claim 1, characterized in that, The prediction results based on the Gaussian process classifier determine the dynamic feasible region, including: For any candidate point in the control input space, the probability that the candidate point will not experience screen blockage or equipment overload is predicted using the Gaussian process classifier. ; All satisfied The set of control input points that are greater than or equal to the safety probability threshold is defined as the dynamic feasible region at the current moment.

4. The intelligent control method for carbon raiser sieving according to claim 1, characterized in that, The particle filter framework is used to propagate the posterior distribution of the optimal operating condition from the previous time step to the prior distribution of the current time step through the state space model. The acquisition function guides importance sampling and resampling to generate the next set of recommended control inputs, including: a) The set of particles representing the posterior distribution of the optimal working condition in the previous moment is propagated through the state space model to obtain the prior particle set at the current moment. b) Calculate the acquisition function value of the control input for each particle in the prior particle set, and use the acquisition function value as the importance weight of the particle; c) Based on the importance weights calculated in step b), resample from the prior particle set to form the posterior particle set at the current moment; d) From the posterior particle set, select the control input corresponding to the particle with the highest predicted mean of comprehensive performance index as the next set of recommended control inputs.

5. The intelligent control method for carbon raiser sieving according to claim 1, characterized in that, The online updating of the deep Gaussian process model and the Gaussian process classifier using newly acquired performance data and constraint states includes: The newly collected control inputs, corresponding comprehensive performance indicators, and constraint states are added to the historical database. The hyperparameters of the deep Gaussian process model and the Gaussian process classifier are updated using a gradient optimization method.

6. A smart control system for carbon raiser sieving, characterized in that, Includes the following units: The control input acquisition unit acquires vibration frequency, screen inclination angle, and feed rate as control inputs, quantifying screening efficiency and product particle size distribution into a unified comprehensive performance index, including: Obtain screening efficiency and product particle size ; Set the target range for product granularity as , and d max >d min ; Define a dimensionless product particle size distribution penalty term. : ； Calculate comprehensive performance indicators ,in and These are preset positive weighting coefficients. This is the weighting coefficient for screening efficiency. This is the product granularity weighting coefficient; A deep Gaussian process model is obtained by probabilistically modeling the relationship between the comprehensive performance index and the control input using a deep Gaussian process model. The deep Gaussian process model uses a deep kernel function to identify nonlinear features. The probabilistic modeling unit constructs a Gaussian process classifier to probabilistically model implicit constraints including screen blockage and / or equipment overload, and determines the dynamic feasible region based on the prediction results of the Gaussian process classifier; within the dynamic feasible region, an acquisition function is constructed using asymmetric information gain based on the posterior prediction distribution output by the deep Gaussian process model. The model building unit constructs a state-space model describing the dynamic evolution of the optimal working condition. Using a particle filter framework, the posterior distribution of the optimal working condition at the previous moment is propagated through the state-space model to obtain the prior distribution at the current moment. The acquisition function is used to guide importance sampling and resampling to generate the next set of recommended control inputs. The control and update unit adjusts the operation of the screening system according to the recommended control input, and updates the deep Gaussian process model and the Gaussian process classifier online using newly acquired performance data and constraint states.

7. The intelligent control system for carbon raiser sieving according to claim 6, characterized in that, The process of using a deep Gaussian process to probabilistically model the relationship between the comprehensive performance index and the control input to obtain a deep Gaussian process model includes: The control input is used as the input to the first-level Gaussian process to generate latent variables; The latent variables are used as input to the second-level Gaussian process to output the posterior prediction distribution of the comprehensive performance index.

8. The intelligent control system for carbon raiser sieving according to claim 6, characterized in that, The prediction results based on the Gaussian process classifier determine the dynamic feasible region, including: For any candidate point in the control input space, the probability that the candidate point will not experience screen blockage or equipment overload is predicted using the Gaussian process classifier. ; All satisfied The set of control input points that are greater than or equal to the safety probability threshold is defined as the dynamic feasible region at the current moment.

Images