Intelligent sorting parameter dynamic adaptation method for hot multi-stage impurity removal of calcined petroleum coke

Abstract

The application discloses a kind of intelligent sorting parameter dynamic adaptation methods of calcined petroleum coke hot multi-stage impurity removal, belong to magnetic separation purification and cross technical field of intelligent manufacturing.This method is aimed at the industry pain point that the fluctuation of feed condition is large in hot impurity removal process, fixed parameter adaptability is poor, impurity removal efficiency and magnetic medium life are difficult to optimize, constructs material characteristics, working condition temperature and equipment parameter multidimensional coupling adaptation model.This method can adapt to the fluctuation of feed temperature, particle size and impurity content, while ensuring that the metal impurity of discharge is not more than 20ppm, the core temperature of magnetic medium is accurately controlled in the safety interval of 100-120℃, the annual attenuation rate of magnetic medium is not more than 2%, the comprehensive energy consumption is reduced by more than 15%, solve the problem of poor stability and high energy consumption of traditional fixed parameter mode, and adapt to large-scale continuous production of calcined petroleum coke for high-end lithium battery negative electrode.

Description

Technical Field

[0001] This invention relates to the technical field of carbon material purification and intelligent manufacturing, and in particular to a method for dynamically adapting intelligent sorting parameters for multi-stage impurity removal of calcined petroleum coke in a hot state. Background Technology

[0002] Calcined petroleum coke is a core precursor for artificial graphite anode materials. The content of magnetic metal impurities such as iron, nickel, cobalt, and copper in it directly determines the cycle life, rate performance, and safety performance of lithium batteries. As power lithium batteries iterate towards higher energy density and higher safety, the total metal impurity requirement for calcined petroleum coke used in high-end anodes has been reduced to below 30 ppm, and batch fluctuations are required to be controlled within ±3 ppm, which places extremely high demands on the stability and intelligent level of the purification process.

[0003] High-gradient magnetic separation is the mainstream technology for removing magnetic metallic impurities from calcined petroleum coke. However, in actual continuous industrial production, hot magnetic separation always faces severe challenges due to fluctuations in operating conditions. Currently, industrial production generally adopts a fixed-parameter operation mode, that is, preset fixed magnetic field strength, material flow rate, and cooling water flow rate, which remain unchanged throughout the process. This mode has revealed three major technical defects that are difficult to overcome in practical applications.

[0004] First, the product quality is unstable and its adaptability to operating conditions is insufficient. Fixed parameters can only achieve optimal impurity removal for a specific benchmark condition. When the impurities in the feed increase or the particle size becomes finer, the impurity removal capacity is insufficient, resulting in excessive impurities in the finished product. When the impurities in the feed decrease or the particle size becomes coarser, the impurity removal capacity is excessive, leading to waste of magnetic field and energy consumption. According to industry production data, under the traditional fixed parameter mode, the fluctuation range of metal impurities in the finished product can reach ±10ppm or more, and the batch qualification rate is only about 70%, which is difficult to meet the stringent requirements for high consistency of power battery raw materials.

[0005] Secondly, the lack of control over the lifespan of the magnetic media leads to uncontrollable equipment wear and tear. The core problem with hot magnetic separation lies in the fact that the heat conduction of the high-temperature material and the heating of the excitation coil work together on the magnetic media, while a fixed cooling water flow rate cannot adapt to the dynamic fluctuations in the feed temperature. When the feed temperature rises, insufficient cooling causes the core temperature of the magnetic media to exceed the 120°C safety threshold, resulting in irreversible decay of magnetic properties and a significant reduction in equipment lifespan.

[0006] Third, the adjustment methods are simplistic and lack synergistic effects across the entire process. Existing parameter adjustment schemes are mostly limited to the magnetic separation process itself, simply adjusting the magnetic field strength without linking the operating parameters of the preceding dispersion and screening processes. However, the agglomeration state of hot materials and the proportion of fine powder directly determine the degree of exposure of metal impurities, which are key pre-existing factors affecting magnetic separation efficiency.

[0007] In summary, there is currently no mature technical solution that can achieve multi-dimensional collaborative dynamic adaptation of sorting parameters, temperature control parameters, and pre-process parameters in hot impurity removal scenarios, which seriously restricts the stable industrial application of hot magnetic separation technology and the intelligent manufacturing upgrade of the calcined petroleum coke industry. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of existing technologies and systematically solve three core technical problems in the hot-state calcined petroleum coke impurity removal process. First, fixed parameter modes have poor adaptability to operating conditions, resulting in large fluctuations in product quality. Second, sorting and cooling are independently controlled, failing to coordinate impurity removal efficiency and magnetic media lifespan. Third, adjustment methods are limited and lack overall process optimization. This invention provides an intelligent sorting parameter dynamic adaptation method for hot-state multi-stage impurity removal of calcined petroleum coke. Through a multi-dimensional coupling model and dual closed-loop control, it achieves intelligent collaborative matching of parameters across all processes under fluctuating operating conditions, simultaneously ensuring product purity, equipment lifespan, and energy efficiency.

[0009] The technical solution adopted by this invention to solve its technical problem is: a method for intelligent sorting parameter dynamic adaptation for multi-stage impurity removal of calcined petroleum coke in a hot state, comprising the following steps: S1: Real-time acquisition of multi-source working condition data: Sensor detection units are installed at the feed end, two-stage magnetic separation unit and discharge end of the hot multi-stage impurity removal production line to simultaneously collect five types of working condition data: feed temperature, feed particle size distribution, feed metal impurity content, magnetic medium core temperature and discharge metal impurity content. The sampling frequency is 1s to 10s. S2: Data preprocessing and feature extraction: The raw collected data is subjected to outlier removal, smoothing filtering and dimensional feature extraction in sequence to obtain a standardized working condition feature vector; S3: Multi-dimensional coupling model dynamic solution: The standardized operating condition feature vector is input into the pre-trained multi-dimensional coupling adaptation model of material properties, operating temperature and equipment parameters, and the separation operation parameters and magnetic medium cooling temperature control parameters of the two-stage magnetic separation unit are solved and output simultaneously; the separation operation parameters include the first-stage coarsening magnetic field strength, the second-stage cleaning magnetic field strength, the first-stage material flow rate and the second-stage material flow rate, and the cooling temperature control parameters include the first-stage magnetic separation cooling water flow rate and the second-stage magnetic separation cooling water flow rate; S4: Multi-parameter collaborative execution control: The calculated parameter commands are synchronously sent to the excitation power supply, the feeding actuator and the cooling water regulating valve to synchronously adjust the magnetic field strength, material flow rate and cooling water flow rate, so as to achieve coordinated matching of sorting intensity and temperature control intensity. S5: Dual closed-loop feedback iterative optimization: The content of metal impurities in the discharged material is used as the quality feedback quantity, and the core temperature of the magnetic medium is used as the life feedback quantity. The feature weights and adjustment coefficients of the coupled adaptation model are corrected in reverse respectively, forming a dual-objective closed-loop control of quality indicators and equipment life, and the model adaptation accuracy is continuously optimized through incremental training.

[0010] As a preferred embodiment of the present invention, in step S1: the feed temperature is detected using an armored K-type thermocouple, with the thermocouple inserted into the center of the feed chute to a depth of not less than 150 mm, the detection range covering 300℃ to 600℃, and the measurement accuracy being ±2℃; the feed particle size distribution is detected using an online laser particle size analyzer, with the detection range covering 0.05 mm to 10 mm, outputting characteristic particle sizes D10, D50, and D90 and the percentage of fine powder below 0.075 mm; the content of metal impurities in both the feed and discharge is detected using an online X-ray fluorescence spectrometer, with the detected elements including iron, nickel, and copper, and the detection accuracy being ±1 ppm; the core temperature of the magnetic medium is detected using an embedded PT100 platinum resistance thermometer, with 3 to 5 detection points evenly distributed along the material flow direction within the magnetic medium filling layer of each stage of the magnetic separator, and the average value of multiple point detections is taken as the core temperature of the magnetic medium.

[0011] As a preferred embodiment of the present invention, in step S3, the multi-dimensional coupling adaptation model is constructed based on the random forest regression algorithm. The construction process is as follows: collect the optimal parameter dataset under different working conditions in historical production, with a sample size of no less than 5,000 groups, and divide it into training set and test set in an 8:2 ratio; take feed temperature, particle size characteristic value, feed metal impurity content, and initial temperature of magnetic medium as input features, and take two-level magnetic field strength, two-level material flow rate, and two-level cooling water flow rate as output targets. The comprehensive optimization objective function is set as F = w1 × (1 - discharge impurity excess rate) + w2 × (1 - magnetic medium over-temperature rate) + w3 × (1 - unit energy consumption), where w1, w2, and w3 are configurable weight coefficients. The hyperparameters of the model are optimized by grid search and five-fold cross-validation. The coefficient of determination R² ≥ 0.97 and the mean absolute error MAE ≤ 5% are obtained.

[0012] As a preferred embodiment of the present invention, in step S3, the dynamic adaptation of the magnetic field strength follows the following rules: the magnetic field strength of the first-stage coarse selection is always 5000Gs to 9000Gs lower than that of the second-stage fine selection magnetic field strength, forming a gradient impurity removal system; the adjustment amount of the magnetic field strength of the second-stage fine selection satisfies the relationship ΔB2=k1×ΔC-k2×ΔT, where ΔC is the change in the total metal impurities in the feed relative to the reference value, ΔT is the change in the core temperature of the magnetic medium relative to the reference value, and k1 and k2 are pre-calibrated proportional coefficients; when the feed impurities increase, the magnetic field strength is increased, and when the magnetic medium temperature increases, the increase in the magnetic field strength is narrowed to avoid further aggravation of temperature rise due to excitation heating.

[0013] As a preferred embodiment of the present invention, in step S3, the dynamic adaptation of the material flow rate follows the following rules: the material flow rate is positively correlated with the feed particle size D50 and negatively correlated with the total metal impurity content of the feed; the single-step flow rate adjustment amount does not exceed 10% of the current set value, and a step-by-step adjustment is adopted to avoid large fluctuations in the material layer thickness that could cause oscillations in the working condition.

[0014] As a preferred embodiment of the present invention, in step S3, the coordinated adaptation of cooling temperature control parameters follows the following rules: the cooling water flow rate is adjusted synchronously with the feed temperature and magnetic field strength: for every 20°C increase in feed temperature, the corresponding cooling water flow rate of the magnetic separation unit increases by 8% to 12%; for every 1000Gs increase in magnetic field strength, the cooling water flow rate increases synchronously by 5% to 7%; a forced protection threshold is set: when the core temperature of the magnetic medium is higher than 115°C, regardless of how other operating parameters change, the corresponding cooling water flow rate automatically increases to the rated maximum value until the magnetic medium temperature drops below 110°C.

[0015] As a preferred embodiment of the present invention, in step S3, the coupling adaptation model also simultaneously calculates the operating parameters of the preceding process, including the high-speed dispersing roller speed and the ultrasonic cleaning power of the secondary vibrating screen; when the proportion of fine powder below 0.075mm in the feed increases, the dispersing roller speed and ultrasonic cleaning power are increased proportionally to remove the effect of fine powder agglomeration on metal impurities, and form a full-process coordinated adaptation with the magnetic separation parameters.

[0016] As a preferred embodiment of the present invention, in step S2, the data preprocessing specifically includes: using the 3σ criterion to identify and remove abnormal jump data exceeding three times the standard deviation, and replacing them with the moving average of the preceding valid data; using a moving average filtering algorithm with a window size of 5 to 10 sampling periods to smooth the valid data; and finally extracting a 10-dimensional standardized operating condition feature vector including feed temperature, D10, D50, D90, fine powder ratio, iron content, nickel content, copper content, total metal content, and average temperature of the magnetic medium.

[0017] As a preferred embodiment of the present invention, step S5 further includes a graded abnormal working condition early warning mechanism: when the content of impurities in the output material exceeds the upper limit of the threshold for three consecutive sampling cycles, a second-level early warning is triggered, the parameter adjustment step size is automatically narrowed and manual verification is prompted; when the core temperature of the magnetic medium continues to rise and the cooling water flow reaches the maximum value but still cannot cool down, a first-level early warning is triggered, the production line processing capacity is automatically reduced to 70% of the rated value, and the safety of equipment operation is prioritized.

[0018] As a preferred embodiment of the present invention, when applied to the two-stage high-gradient magnetic separation and impurity removal process of petroleum coke after hot calcination at 300℃ to 500℃, the fluctuation range of the total metal impurity content in the output is no greater than ±2ppm, the annual magnetic intensity attenuation rate of the magnetic medium is no greater than 2%, and the comprehensive energy consumption of excitation and cooling per unit processing volume is reduced by more than 15% compared with the fixed parameter mode.

[0019] The beneficial effects of this invention are: 1. Overcoming the bottleneck of coupling effect and achieving synergistic optimization of quality and lifespan. Through a collaborative calculation mechanism of sorting parameters and cooling temperature control, the problem of temperature and magnetic field coupling effect in hot magnetic separation is solved. Under dynamic operating conditions, it can simultaneously ensure stable compliance of output impurities and safe magnetic medium temperature. Compared with the traditional fixed parameter mode, the fluctuation range of finished product metal impurities is narrowed from ±10ppm to within ±2ppm, consistency is improved by 80%, and batch qualification rate is increased from 70% to 100%. The annual decay rate of magnetic medium is reduced from over 5% to within 2%, and the equipment lifespan is more than doubled.

[0020] 2. Seamless integration across all processes, simultaneously improving adjustment efficiency and energy efficiency. Parameter adjustment is extended from the magnetic separation process to the entire process of dispersing and screening. De-agglomeration and quality improvement in the pre-process reduces the adjustment pressure on the magnetic separation stage. Compared to solutions that only adjust the magnetic field strength, under the same impurity removal effect, the combined energy consumption for excitation and cooling is reduced by more than 15%, and the adjustment response speed after operating condition fluctuations is improved by 40%, achieving the dual benefits of quality improvement and cost reduction.

[0021] 3. Dual closed-loop autonomous iteration, high adaptability and intelligence. Based on a closed-loop feedback and incremental training mechanism with dual objectives of quality and lifespan, the model can continuously and autonomously optimize with production, eliminating the need for repeated manual parameter adjustments and adapting to slow variables such as raw material characteristic drift and equipment performance degradation. A supporting graded anomaly early warning mechanism enables stable unattended operation, meeting the needs of intelligent manufacturing production line upgrades.

[0022] 4. Low modification cost, easy implementation, and significant industrial application value. This method can be directly upgraded to intelligent systems based on existing hot-state impurity removal equipment. Only sensors and control systems need to be added, without modifying the core equipment structure. The modification cost is low and the results are quick. Detailed Implementation

[0023] Example 1: A method for dynamically adapting intelligent sorting parameters in hot-state multi-stage impurity removal of calcined petroleum coke includes the following steps: S1 Real-time acquisition of multi-source operating conditions: Multiple types of sensing and detection units are deployed at key nodes of the hot-state multi-stage impurity removal production line to construct a full-dimensional operating condition perception system, and five types of basic data are collected in real time. The sampling frequency can be configured in the range of 1s to 10s according to the fluctuation characteristics of the production line.

[0024] In the feed temperature detection stage, a sheathed K-type thermocouple is installed at the center of the insulated chute at the outlet of the feed buffer silo. The insertion depth is no less than 150mm, allowing direct contact with the material to detect the core temperature and avoiding interference from the chute wall temperature. The detection range covers 300℃ to 600℃, with a measurement accuracy of ±2℃, used to characterize the heat load level brought in by the hot material.

[0025] In the feed particle size distribution detection stage, an online laser particle size analyzer is installed at the feed end of the primary vibrating screen. It employs diffuse reflection detection, eliminating the need for material contact and adapting to high-temperature conditions. The detection range covers 0.05mm to 10mm, providing real-time output of characteristic particle sizes D10, D50, and D90, as well as the proportion of fine powder below 0.075mm, reflecting the particle size uniformity and agglomeration risk of the material.

[0026] In the process of detecting metal impurities in the feed, an online X-ray fluorescence spectrometer is installed at the discharge end of the secondary vibrating screen to perform in-situ real-time detection. The detected elements include three core magnetic metals: iron, nickel, and copper. The detection accuracy for a single component is ±1 ppm, and the detection accuracy for the total content is ±2 ppm, which serves as the core input basis for setting the sorting intensity.

[0027] In the core temperature detection stage of the magnetic media, three to five PT100 platinum resistance thermometers are evenly distributed along the material flow direction inside the magnetic media filling layer of the two-stage magnetic separator. The temperature probes are directly embedded in the magnetically conductive steel wool filling layer to detect the actual operating temperature of the magnetic media, rather than the wall temperature. The average value of multiple points is taken as the core temperature of the magnetic media to characterize the demagnetization risk level of the magnetic media.

[0028] In the process of detecting metal impurities in the finished product, an online X-ray fluorescence spectrometer of the same precision is installed in the finished product discharge chute to detect the level of impurities in the finished product in real time. This serves as a quality indicator for closed-loop feedback, verifying the impurity removal effect and driving parameter correction.

[0029] S2 Data Preprocessing and Feature Extraction: Raw data collected from industrial sites suffers from electromagnetic interference noise, abnormal jump values, and dimensional redundancy. It needs to be standardized before it can be input into the model. The specific processing flow is as follows.

[0030] The first step is outlier removal. The 3σ criterion is used to identify continuous sampling data. The mean μ and standard deviation σ of the most recent N data points are calculated. Data points exceeding three times the standard deviation are identified as outliers, removed, and replaced with the moving average of the previous 5 valid data points. This avoids parameter misadjustment caused by factors such as sensor interference and accidental material agglomeration.

[0031] The second step is smoothing filtering. A moving average filtering algorithm with a window size of 5 to 10 sampling periods is used to smooth the effective data, suppressing small random fluctuations, avoiding frequent start-ups and adjustments of the actuators, and extending the service life of the equipment. For operating conditions with drastic fluctuations, the filtering window can be adaptively reduced to improve response speed.

[0032] The third step is feature extraction and standardization. The original multi-dimensional data undergoes feature reduction and normalization. For particle size, four features are extracted: D10, D50, D90, and the proportion of fine powder smaller than 0.075mm. For impurities, four features are extracted: the content of individual iron, nickel, and copper components, and the total metal content. For temperature, two features are extracted: feed temperature and average temperature of the magnetic medium. Finally, a 10-dimensional standardized operating condition feature vector is formed, with all feature values ​​normalized to the 0-1 range, improving the model's solution efficiency and accuracy.

[0033] The S3 multi-dimensional coupling model dynamically solves by inputting the standardized operating condition feature vector into the pre-trained multi-dimensional coupling adaptation model, and simultaneously solves to obtain two main categories of outputs: sorting operation parameters and cooling temperature control parameters, thereby achieving multi-parameter collaborative matching.

[0034] 3.1 Construction of Coupling Adaptation Model The multi-dimensional coupling adaptation model is built based on the random forest regression algorithm, which has the advantages of anti-overfitting, strong nonlinear fitting ability, and high training efficiency, making it suitable for complex working conditions in industrial sites. The specific model construction and training process is as follows.

[0035] First, dataset construction. Optimal parameter datasets were collected under different operating conditions during historical production, covering optimal magnetic field strength, material flow rate, and cooling water flow rate combinations corresponding to different feed temperatures, particle size distributions, and impurity levels, with a sample size of no less than 5000 sets. Data from periods of equipment failure, abnormal operating conditions, and manual intervention were removed to ensure that all samples represent steady-state optimal operating conditions.

[0036] Second, the dataset is divided randomly into a training set and a test set in an 8:2 ratio. The training set is used for model fitting, and the test set is used for model performance evaluation.

[0037] Third, model training. The model uses feed temperature, particle size characteristics, feed impurity characteristics, and initial temperature of the magnetic medium as input features, and six parameters as output targets: primary magnetic field strength, secondary magnetic field strength, primary material flow rate, secondary material flow rate, primary cooling water flow rate, and secondary cooling water flow rate.

[0038] The overall optimization objective function is set as F = w1 × (1 - discharge impurity excess rate) + w2 × (1 - magnetic medium over-temperature rate) + w3 × (1 - unit energy consumption), where w1, w2, and w3 are configurable weight coefficients with a default ratio of 5:3:2, which can be dynamically adjusted according to production priorities. In quality-priority mode, the proportion of w1 is increased; in energy-priority mode, the proportion of w3 is increased.

[0039] Fourth, hyperparameter optimization. Grid search and five-fold cross-validation were used to optimize the model's hyperparameters. The number of base learners was set to 80 to 120, and the maximum depth of the decision tree was set to 6 to 10 layers. The final model achieved a coefficient of determination R² ≥ 0.97 and a mean absolute error (MAE) ≤ 5% on the test set, meeting the accuracy requirements for industrial control.

[0040] 3.2 Parameter Adaptation Rules The various parameters output by the model follow the following quantization adaptation rules to ensure that the adjustment logic conforms to the process mechanism and avoids unreasonable results from the black-box output of the model.

[0041] First, the magnetic field strength gradient adaptation rule. A gradient impurity removal system is formed by primary coarsening and secondary cleaning. The magnetic field strength of the primary stage is always 5000 Gs to 9000 Gs lower than that of the secondary stage. The primary stage is responsible for removing large, strongly magnetic particles, while the secondary stage is responsible for deeply removing fine, weakly magnetic particles. The adjustment of the magnetic field strength in the secondary cleaning stage satisfies the quantitative relationship ΔB2 = k1 × ΔC - k2 × ΔT, where ΔC is the change in total metal impurities in the feed relative to the reference value, ΔT is the change in the core temperature of the magnetic medium relative to the reference value, and k1 and k2 are proportionality coefficients obtained through calibration. When the feed impurities increase, the magnetic field strength is increased to ensure the impurity removal effect; when the magnetic medium temperature increases, the increase in magnetic field strength is narrowed to avoid further exacerbating the temperature rise due to excitation heating, thus achieving a balance between quality and lifespan.

[0042] Second, the dynamic matching rule for material flow rate. Material flow rate is bidirectionally correlated with particle size distribution and impurity level. The larger the feed particle size (D50) and the higher the proportion of coarse particles, the faster the material flow rate, making it easier to capture coarse impurities without requiring excessive residence time. The higher the impurity content in the feed, the slower the secondary cleaning flow rate, extending the material's residence time in the high-gradient magnetic field and improving the removal depth. To avoid operational fluctuations, the single-step flow rate adjustment should not exceed 10% of the current set value, using a stepped adjustment mechanism.

[0043] Third, the cooling and temperature control are coordinated. The cooling water flow rate, feed temperature, and magnetic field strength work in tandem to offset the temperature rise caused by both material heat conduction and excitation heating. For every 20°C increase in feed temperature, the cooling water flow rate of the magnetic separator increases by 8% to 12%, offsetting the additional heat brought in by the material. For every 1000 Gs increase in magnetic field strength, the cooling water flow rate increases by 5% to 7%, offsetting the increased heat generated by the excitation coil. A forced protection threshold is set; when the core temperature of the magnetic medium exceeds 115°C, a forced cooling mode is triggered. Regardless of changes in other parameters, the corresponding cooling water flow rate automatically increases to its rated maximum until the magnetic medium temperature drops below 110°C, prioritizing equipment safety.

[0044] Fourth, the pre-process linkage adaptation rules. The model synchronously outputs two pre-process parameters: the high-speed dispersing roller speed and the ultrasonic cleaning power of the secondary screen. When the proportion of fine powder below 0.075mm in the feed increases, the dispersing roller speed is increased proportionally to enhance the mechanical de-agglomeration effect, break up coke powder agglomerates, and release the encapsulated metal impurities. When the proportion of fine powder increases, the ultrasonic cleaning power is increased simultaneously to prevent fine coke powder from clogging the screen and to ensure screening efficiency and material particle size stability. Through the pre-treatment and quality improvement of the pre-process, the adjustment pressure of the magnetic separation stage is reduced, achieving synergistic efficiency improvement throughout the entire process.

[0045] The parameter commands obtained by the S4 multi-parameter collaborative execution control model are sent to each actuator through the Profinet industrial bus to achieve synchronous adjustment of multiple parameters, with an adjustment response time of no more than 5 seconds.

[0046] In the magnetic field strength adjustment stage, the magnetic field strength is continuously adjusted by adjusting the DC output current of the magnetic separator excitation power supply. The current adjustment resolution is no greater than 0.1A, and the magnetic field strength control accuracy is ±200Gs.

[0047] In the material flow rate regulation stage, the material volume flow rate and material layer thickness are precisely controlled by adjusting the opening of the electric regulating valve of the magnetic separator feed pipe and the speed of the variable frequency feeder. The flow rate regulation accuracy is ±0.05m / s.

[0048] In the cooling water flow regulation stage, the flow rate of the cooling water circuit is adjusted by an electric proportional regulating valve. The adjustment range is 30% to 100% of the rated flow rate, and the control accuracy is ±2%.

[0049] All parameter adjustments are triggered and implemented synchronously to avoid intermediate operating condition oscillations caused by asynchronous adjustments of different parameters.

[0050] The S5 dual-loop feedback iterative optimization constructs a dual-loop feedback system to continuously optimize model adaptation accuracy and achieve autonomous system evolution.

[0051] First, a quality closed-loop process is implemented. The metal impurity content detected in real time at the discharge end is compared with the target value. When the content deviates from the target value for three consecutive sampling cycles, a small-step gradient descent method is used to reverse-correct the weight coefficients of impurity-related features in the coupled model. The step size is set to 0.01, gradually improving the impurity removal accuracy under subsequent operating conditions.

[0052] Second, a closed-loop lifespan mechanism. By comparing the core temperature of the magnetic medium with a safe threshold, and adjusting the cooling parameter coefficients when the temperature remains consistently too high or too low, the temperature control accuracy is optimized to ensure that the magnetic medium always operates within its optimal temperature range.

[0053] Third, incremental model training. The system automatically aggregates high-quality steady-state data generated each day at midnight and performs incremental training on the coupled model. Each incremental training iteration involves at least 200 sets of samples, continuously enriching the model's coverage of operating conditions. The longer the running time, the higher the model's adaptation accuracy and the less manual intervention is required.

[0054] The method also includes a graded abnormal operating condition early warning and protection mechanism. The level-two early warning is triggered when the impurity content in the output exceeds the upper threshold for three consecutive sampling cycles. Upon triggering, the system automatically narrows the parameter adjustment step size and simultaneously sends an early warning message to the central control room, prompting manual verification of the raw material and equipment status. The level-one early warning is triggered when the core temperature of the magnetic medium continues to rise and the cooling water flow reaches its maximum value but still cannot cool the material. Upon triggering, the system automatically reduces the production line's throughput to 70% of the rated value, reducing heat load input and prioritizing equipment safety. Production capacity is gradually increased after the operating conditions recover.

[0055] Example 2: Basic Adaptation to Conventional Fluctuating Operating Conditions This embodiment verifies the basic adaptability and steady-state operation capability of the method under normal fluctuating operating conditions during normal production.

[0056] The operating conditions are as follows: domestic conventional delayed petroleum coke is used for calcination and discharge; the feed temperature fluctuates between 400℃ and 440℃; the total metal impurities in the feed are between 110ppm and 140ppm; the particle size D50 is between 2.5mm and 3.5mm; and the proportion of fine powder below 0.075mm is between 18% and 22%. This is a typical continuous production conventional fluctuating operating condition.

[0057] The system is configured to enable the intelligent dynamic adaptation method of this invention, with a sampling period of 5 seconds. The coupling model uses a pre-trained general model, and the weight coefficient ratio w1:w2:w3 is 5:3:2, prioritizing product quality.

[0058] The dynamic parameter range is calculated in real time by the system based on operating conditions, and the dynamic adjustment range of the parameters is as follows: Primary coarsening magnetic field strength: 9500 Gs to 11500 Gs; Secondary cleaning magnetic field strength: 17500 Gs to 20500 Gs. Primary material flow rate: 0.7 m / s to 0.9 m / s; Secondary material flow rate: 0.4 m / s to 0.5 m / s. Primary cooling water flow rate: 14 cubic meters per hour to 20 cubic meters per hour; Secondary cooling water flow rate: 16 cubic meters per hour to 22 cubic meters per hour. High-speed dispersing roller speed: 400 rpm to 500 rpm; Secondary screen ultrasonic cleaning power: 400 W to 600 W.

[0059] The operating results show that the system ran continuously and stably for 72 hours. During this period, the feed parameters fluctuated naturally with the calcination conditions, and the system automatically adjusted throughout the process without manual intervention. Test results showed that the total metal impurities in the feed remained stable between 16 ppm and 20 ppm, with a fluctuation range of ±2 ppm, and the impurity compliance rate was 100%. The core temperature of the magnetic medium remained stable between 102℃ and 115℃, with no overheating and a temperature fluctuation range not exceeding 13℃. The combined energy consumption for excitation and cooling per unit processing capacity was reduced by 16.2% compared to the original fixed parameter mode of the production line, demonstrating significant operational stability and energy-saving effects.

[0060] Example 3: Enhanced Adaptation for High Temperature and High Impurity Operating Conditions This embodiment verifies the enhanced adaptability of the method and its protective effect under extreme conditions caused by high-temperature and high-impurity raw materials.

[0061] The operating conditions are as follows: high-sulfur and high-ash delayed petroleum coke feedstock is used, the calcination discharge temperature fluctuates between 460℃ and 500℃, the total metal impurities in the feed are 150ppm to 180ppm, the particle size is relatively fine, D50 is 1.8mm to 2.5mm, and the proportion of fine powder is 25% to 30%, which is a high-load and harsh operating condition.

[0062] The system is configured to use the intelligent dynamic adaptation method, adjusting the weight ratio w1:w2:w3 to 6:3:1 to prioritize both noise removal effectiveness and equipment safety. The model automatically switches to high-load adaptation mode.

[0063] The dynamic parameter ranges are as follows: Primary magnetic field strength: 11000 Gs to 13000 Gs; Secondary magnetic field strength: 20000 Gs to 22000 Gs. Primary flow velocity: 0.6 m / s to 0.7 m / s; Secondary flow velocity: 0.3 m / s to 0.4 m / s. Primary cooling water flow rate: 20 cubic meters per hour to 26 cubic meters per hour; Secondary cooling water flow rate: 22 cubic meters per hour to 28 cubic meters per hour. Dispersing roller speed: 500 rpm to 600 rpm; Ultrasonic power: 700 W to 900 W.

[0064] The results showed that after 48 hours of continuous operation, the total metal impurities in the output remained stable between 15 ppm and 19 ppm, with a fluctuation range of ±2 ppm, fully meeting the quality requirements of high-end products. The core temperature of the magnetic medium remained stable between 108℃ and 118℃, only briefly reaching 118℃ before the system quickly returned to normal by increasing the cooling flow rate, with no risk of irreversible demagnetization.

[0065] Compared to the original fixed-parameter mode of the production line, the impurity removal compliance rate under this operating condition is only 62%, and the magnetic medium frequently overheats, with an estimated annual attenuation rate of 8%. This method increases the compliance rate to 100% and reduces the duration of magnetic medium overheating by more than 95%, fully demonstrating the technical advantages of collaborative adaptation.

[0066] Example 4: Energy-saving optimization and adaptation under low-impurity operating conditions This embodiment verifies the energy-saving optimization capability and energy efficiency improvement effect of the method under low-impurity conditions with high-quality, low-sulfur raw materials.

[0067] The operating conditions are as follows: imported low-sulfur high-quality delayed petroleum coke is used, the calcination discharge temperature fluctuates between 300℃ and 340℃, the total metal impurities in the feed are 80ppm to 100ppm, the particle size is relatively coarse, the D50 is 3.5mm to 4.5mm, and the fine powder accounts for 12% to 15%, which is a low-load high-quality raw material operating condition.

[0068] The system is configured to enable the intelligent dynamic adaptation method, and the weight coefficient ratio w1:w2:w3 is adjusted to 3:2:5 to prioritize energy efficiency and maximize energy saving while meeting quality standards.

[0069] The dynamic parameter ranges are as follows: Primary magnetic field strength: 9000 Gs to 10000 Gs; Secondary magnetic field strength: 16000 Gs to 18000 Gs. Primary flow velocity: 0.9 m / s to 1.0 m / s; Secondary flow velocity: 0.5 m / s to 0.6 m / s. Primary cooling water flow rate: 10 cubic meters per hour to 14 cubic meters per hour; Secondary cooling water flow rate: 12 cubic meters per hour to 16 cubic meters per hour. Dispersing roller speed: 300 rpm to 400 rpm; Ultrasonic power: 300 W to 400 W.

[0070] The results show that after 72 hours of continuous operation, the total metal impurities in the output remained stable between 17 ppm and 20 ppm, fully meeting product quality standards. The core temperature of the magnetic medium remained stable between 95℃ and 105℃, within a safe temperature range. The combined energy consumption for excitation and cooling per unit throughput was reduced by 22.8% compared to the fixed parameter mode, demonstrating significant energy savings. In contrast, the traditional fixed parameter mode, to accommodate high impurity conditions, maintains consistently high magnetic field and cooling parameters, resulting in substantial energy waste under these conditions. The dynamic adaptation of this method fully demonstrates the energy efficiency advantages of on-demand control.

[0071] Comparative Example 1: Traditional Fixed Parameter Operation Mode The production line was operated with conventional fixed parameters. The magnetic field strength, material flow rate, and cooling water flow rate were all set at medium operating conditions, and the line was operated under the same conventional fluctuating operating conditions as in Example 1.

[0072] Test results showed that the total metal impurities in the output ranged from 12 ppm to 28 ppm, with a fluctuation range of ±8 ppm, and the impurity exceedance rate was approximately 28%. The magnetic medium temperature ranged from 95℃ to 135℃, with frequent overheating, and the estimated annual magnetic intensity decay rate was 5.8%. The unit energy consumption was higher than that of Example 1, and it could not adapt to high-temperature, high-impurity and low-impurity operating conditions, exhibiting extremely poor adaptability.

[0073] Comparative Example 2: Single Magnetic Field Feedback Adjustment Mode The magnetic field strength is adjusted based solely on feedback from the discharge impurity data, while the cooling water flow rate and parameters of the preceding process remain fixed, with all other conditions being the same as in Example 1.

[0074] Test results showed that the discharge impurities fluctuated within ±4ppm, with the compliance rate increasing to 85%. However, the magnetic medium temperature fluctuated greatly, frequently exceeding the limit, and the estimated annual attenuation rate was 4.5%. Furthermore, the energy consumption reduction was not significant, decreasing by only 6%. It was impossible to balance equipment lifespan and energy efficiency optimization, representing a single-point improvement with limited system optimization effects.

[0075] In summary, by comparing the above embodiments with the comparative examples, the following conclusions can be clearly drawn.

[0076] First, the method of this invention can simultaneously and stably ensure product quality, magnetic medium safety and energy efficiency under various working conditions such as normal, harsh, low load and large fluctuation, solving the common industry problem that existing technologies cannot simultaneously take into account the three major goals of quality, lifespan and energy efficiency.

[0077] Second, the multi-dimensional coupling and synergistic mechanism of this invention is not a simple superposition of conventional technical means in the field. Simply adjusting the magnetic field strength cannot solve the problem of magnetic medium overheating, and simply adjusting the cooling cannot improve the impurity removal effect. When the two are linked with the pre-dispersing and screening process, a significant synergistic effect is produced. The overall effect is far better than the simple addition of the effects of each step, and unexpected technical progress has been achieved.

[0078] Third, the dual-closed-loop incremental iteration mechanism enables the system to have autonomous learning and optimization capabilities, breaking through the limitation that the accuracy of traditional static control schemes gradually decreases with the drift of operating conditions, and has outstanding substantive characteristics.

[0079] The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for dynamically adapting intelligent sorting parameters for multi-stage impurity removal of calcined petroleum coke in a hot state, characterized in that, Includes the following steps: S1: Real-time acquisition of multi-source working condition data: Sensor detection units are installed at the feed end, two-stage magnetic separation unit and discharge end of the hot multi-stage impurity removal production line to simultaneously collect five types of working condition data: feed temperature, feed particle size distribution, feed metal impurity content, magnetic medium core temperature and discharge metal impurity content. The sampling frequency is 1s to 10s. S2: Data preprocessing and feature extraction: The raw collected data is subjected to outlier removal, smoothing filtering and dimensional feature extraction in sequence to obtain a standardized working condition feature vector; S3: Multi-dimensional coupling model dynamic solution: The standardized operating condition feature vector is input into the pre-trained multi-dimensional coupling adaptation model of material properties, operating temperature and equipment parameters, and the separation operation parameters and magnetic medium cooling temperature control parameters of the two-stage magnetic separation unit are solved and output simultaneously; the separation operation parameters include the first-stage coarsening magnetic field strength, the second-stage cleaning magnetic field strength, the first-stage material flow rate and the second-stage material flow rate, and the cooling temperature control parameters include the first-stage magnetic separation cooling water flow rate and the second-stage magnetic separation cooling water flow rate; S4: Multi-parameter collaborative execution control: The calculated parameter commands are synchronously sent to the excitation power supply, the feeding actuator and the cooling water regulating valve to synchronously adjust the magnetic field strength, material flow rate and cooling water flow rate, so as to achieve coordinated matching of sorting intensity and temperature control intensity. S5: Dual closed-loop feedback iterative optimization: The content of metal impurities in the discharged material is used as the quality feedback quantity, and the core temperature of the magnetic medium is used as the life feedback quantity. The feature weights and adjustment coefficients of the coupled adaptation model are corrected in reverse respectively, forming a dual-objective closed-loop control of quality indicators and equipment life, and the model adaptation accuracy is continuously optimized through incremental training.

2. The intelligent sorting parameter dynamic adaptation method for multi-stage impurity removal of calcined petroleum coke in a hot state as described in claim 1, characterized in that, In step S1: the feed temperature is detected by armored K-type thermocouples. The thermocouple is inserted into the center of the feed chute to a depth of not less than 150 mm. The detection range covers 300℃ to 600℃ and the measurement accuracy is ±2℃. The feed particle size distribution is detected using an online laser particle size analyzer, covering a range from 0.05 mm to 10 mm, and outputs data on characteristic particle sizes D10, D50, and D90, as well as the proportion of fine powder below 0.075 mm. The content of metal impurities in both the feed and discharge is detected using an online X-ray fluorescence spectrometer, which detects iron, nickel, and copper, with a detection accuracy of ±1 ppm. The core temperature of the magnetic media is detected using an embedded PT100 platinum resistance thermometer. Three to five detection points are evenly distributed along the material flow direction within the magnetic media filling layer of each stage of the magnetic separator, and the average value of multiple point detections is taken as the core temperature of the magnetic media.

3. The intelligent sorting parameter dynamic adaptation method for multi-stage impurity removal of calcined petroleum coke in a hot state as described in claim 1, characterized in that, In step S3, the multi-dimensional coupling adaptation model is constructed based on the random forest regression algorithm. The construction process is as follows: collect the optimal parameter dataset under different working conditions in historical production, with a sample size of no less than 5,000 groups, and divide it into training set and test set in an 8:2 ratio; take feed temperature, particle size feature value, feed metal impurity content, and initial temperature of magnetic medium as input features, and take two-level magnetic field strength, two-level material flow rate, and two-level cooling water flow rate as output targets. The comprehensive optimization objective function is set as F = w1 × (1 - discharge impurity excess rate) + w2 × (1 - magnetic medium over-temperature rate) + w3 × (1 - unit energy consumption), where w1, w2, and w3 are configurable weight coefficients. The hyperparameters of the model are optimized by grid search and five-fold cross-validation. The coefficient of determination R² ≥ 0.97 and the mean absolute error MAE ≤ 5% are obtained.

4. The intelligent sorting parameter dynamic adaptation method for multi-stage impurity removal of calcined petroleum coke in a hot state as described in claim 1 or 3, characterized in that, In step S3, the dynamic adaptation of the magnetic field strength follows these rules: the magnetic field strength of the primary coarse selection is always 5000 Gs to 9000 Gs lower than that of the secondary fine selection magnetic field strength, forming a gradient impurity removal system; the adjustment of the magnetic field strength of the secondary fine selection satisfies the relationship ΔB2=k1×ΔC-k2×ΔT, where ΔC is the change in the total metal impurities in the feed relative to the reference value, ΔT is the change in the core temperature of the magnetic medium relative to the reference value, and k1 and k2 are pre-calibrated proportional coefficients; when the feed impurities increase, the magnetic field strength is increased, and when the magnetic medium temperature increases, the increase in the magnetic field strength is narrowed to avoid further aggravation of temperature rise due to excitation heating.

5. The intelligent sorting parameter dynamic adaptation method for multi-stage impurity removal of calcined petroleum coke in a hot state as described in claim 1, characterized in that, In step S3, the dynamic adaptation of material flow rate follows the following rules: the material flow rate is positively correlated with the feed particle size D50 and negatively correlated with the total metal impurity content of the feed; the single-step flow rate adjustment amount does not exceed 10% of the current set value, and a step-by-step adjustment is adopted to avoid large fluctuations in the material layer thickness that could cause oscillations in the working condition.

6. The intelligent sorting parameter dynamic adaptation method for multi-stage impurity removal of calcined petroleum coke in a hot state as described in claim 1, characterized in that, In step S3, the coordinated adaptation of cooling temperature control parameters follows these rules: the cooling water flow rate is adjusted synchronously with the feed temperature and magnetic field strength: for every 20°C increase in feed temperature, the corresponding cooling water flow rate of the magnetic separation unit increases by 8% to 12%; for every 1000 Gs increase in magnetic field strength, the cooling water flow rate increases synchronously by 5% to 7%. Set a mandatory protection threshold: When the core temperature of the magnetic medium is higher than 115℃, regardless of how other operating parameters change, the corresponding cooling water flow rate will automatically increase to the rated maximum value until the temperature of the magnetic medium drops below 110℃.

7. The intelligent sorting parameter dynamic adaptation method for multi-stage impurity removal of calcined petroleum coke in a hot state as described in claim 1, characterized in that, In step S3, the coupling adaptation model also simultaneously calculates the operating parameters of the preceding process, including the speed of the high-speed dispersing roller and the ultrasonic cleaning power of the secondary vibrating screen. When the proportion of fine powder below 0.075mm in the feed increases, the speed of the dispersing roller and the ultrasonic cleaning power are increased proportionally to remove the effect of fine powder agglomeration on metal impurities, thus forming a full-process synergistic adaptation with the magnetic separation parameters.

8. The intelligent sorting parameter dynamic adaptation method for multi-stage impurity removal of calcined petroleum coke in a hot state as described in claim 1, characterized in that, In step S2, data preprocessing specifically includes: using the 3σ criterion to identify and remove abnormal jump data exceeding three times the standard deviation, and replacing them with the moving average of the preceding valid data; using a moving average filtering algorithm with a window size of 5 to 10 sampling periods to smooth the valid data; and finally extracting a 10-dimensional standardized operating condition feature vector including feed temperature, D10, D50, D90, fine powder ratio, iron content, nickel content, copper content, total metal content, and average temperature of the magnetic medium.

9. The intelligent sorting parameter dynamic adaptation method for multi-stage impurity removal of calcined petroleum coke in a hot state as described in claim 1, characterized in that, Step S5 also includes a graded abnormal operating condition early warning mechanism: when the content of impurities in the output exceeds the upper limit of the threshold for three consecutive sampling cycles, a level two early warning is triggered, the parameter adjustment step size is automatically narrowed and manual verification is prompted; when the core temperature of the magnetic medium continues to rise and the cooling water flow reaches the maximum value but still cannot cool down, a level one early warning is triggered, the production line processing capacity is automatically reduced to 70% of the rated value, and the safety of equipment operation is prioritized.

10. The intelligent sorting parameter dynamic adaptation method for multi-stage impurity removal of calcined petroleum coke in a hot state as described in claim 1, characterized in that, When applied to the two-stage high-gradient magnetic separation and impurity removal process of petroleum coke after hot calcination at 300℃ to 500℃, the fluctuation range of the total metal impurity content in the output is no greater than ±2ppm, the annual magnetic intensity attenuation rate of the magnetic medium is no greater than 2%, and the comprehensive energy consumption of excitation and cooling per unit processing volume is reduced by more than 15% compared with the fixed parameter mode.