A blast furnace burden distribution self-adaptive compensation method and system based on chute vibration spectrum

Abstract

The application discloses a blast furnace material distribution adaptive compensation method and system based on a chute vibration spectrum, in the blast furnace material distribution process, the vibration signal of the material chute is collected, pretreated and feature extracted, the vibration intensity feature and the particle size distribution feature are obtained; the vibration intensity feature and the particle size distribution feature are input into the pre-trained machine learning inversion model, the current raw material powder proportion value and the particle size grade are calculated and output in real time; the powder proportion value and the particle size grade are matched with the preset material distribution compensation rule library, the compensation control instruction is generated; the material distribution chute is driven based on the compensation control instruction to complete the in-furnace distribution of the blast furnace raw material; the application solves the problems of the prior art, such as the raw material physical property sensing lag, the disconnection of the control decision and the process target, and the insufficient compensation accuracy, realizes the online real-time sensing and adaptive feedforward compensation of the raw material particle size fluctuation, and improves the uniformity and stability of the blast furnace material distribution.

Description

Technical Field

[0001] This invention belongs to the field of blast furnace ironmaking and automatic control technology, specifically relating to an adaptive compensation method and system for blast furnace charging based on the vibration spectrum of the chute. Background Technology

[0002] In existing technologies, intelligent control of the blast furnace charging process, especially adaptive compensation for fluctuations in raw material properties, typically relies on the aggregation of process data such as the opening of the material flow valve, charging time, and weighing weight. By constructing expert rules, fuzzy self-learning, or physical models, control decisions are made based on these data to achieve control over the uniformity of the charging process. However, existing methods have significant shortcomings in real-time and direct perception of fluctuations in the physical properties of raw materials, as well as in feedforward compensation control based on property perception.

[0003] In practical applications, blast furnace charging systems involve multiple factors such as raw material particle size, hardness, moisture content, and equipment characteristics, and their interrelationships exhibit complex and nonlinear characteristics. Existing adaptive control often takes the form of feedback correction of flow or time deviations, model prediction, or rule matching. Due to the indirectness and limitations of sensing methods, existing technologies struggle to directly and online capture and quantify instantaneous fluctuations in raw material properties, resulting in a lack of crucial information in the prerequisite inputs for control decisions. In existing technologies, the core control quantity remains the opening of the material flow valve, and the sensing sources are weight and time. It is impossible to distinguish whether the cause of flow deviation is a change in the actual flow rate of the raw material or a change in the equivalent flow signal due to increased flowability caused by an increase in raw material powder. This leads to low specificity and consistency of compensation mechanisms under complex operating conditions with frequent fluctuations in raw material properties, and the overall causal chain matching degree from disturbance identification to compensation execution is also weak.

[0004] It is evident that existing technologies in adaptive control of blast furnace charging suffer from problems such as delayed and indirect perception of raw material properties, weak correlation between control decision-making basis and final process objective, and insufficient precision of compensation actions. These are the shortcomings of existing technologies.

[0005] In view of this, it is very necessary to provide an adaptive compensation method and system for blast furnace charging based on the vibration spectrum of the chute, so as to solve the above-mentioned defects in the prior art. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies in adaptive control of blast furnace charging, such as delayed and indirect perception of raw material properties, weak correlation between control decision-making basis and final process objective, and insufficient accuracy of compensation actions. This invention provides a method and system for adaptive compensation of blast furnace charging based on the vibration spectrum of the chute, thereby solving the aforementioned technical problems.

[0007] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, this application provides an adaptive compensation method for blast furnace charging based on the vibration spectrum of the chute, including: Step S1: During the blast furnace charging process, the vibration signal of the charging chute is collected, and preprocessed and feature extracted to obtain vibration intensity characteristics and particle size distribution characteristics. Step S2: Input the vibration intensity characteristics and particle size distribution characteristics into the pre-trained machine learning inversion model to calculate and output the powder ratio and particle size grade of the current raw material in real time. Step S3: Match the powder ratio and particle size with the preset fabric compensation rule library to generate compensation control instructions; Step S4: Based on the compensation control command, drive the charging chute to complete the in-furnace distribution of raw materials in the blast furnace.

[0008] By employing the above technical solution, the vibration signals generated by the impact of raw materials into the chute are captured and the features that characterize their physical properties are extracted. The key powder ratio parameters are then derived in real time using a machine learning model. Based on these parameters, an intelligent rule base is used to generate precise compensation instructions for the material distribution parameters. This achieves closed-loop control from the perception of raw material property disturbances to the feedforward compensation of material distribution parameters. It can effectively offset the influence of raw material particle size fluctuations on the material distribution point, improve the uniformity of material distribution, and thus enhance the stability of blast furnace operation.

[0009] Preferably, the preprocessing in step S1 involves using a bandpass filter to filter the vibration acceleration time-domain signal. The lower passband frequency of the bandpass filter is set to the mechanical vibration frequency generated by the tilting and rotating equipment of the chute, and the upper passband frequency is set to the electromagnetic noise frequency generated by the power grid and electrical equipment.

[0010] By adopting the above technical solution and setting a bandpass filter that meets specific process objectives and physical mechanisms, it is possible to retain the characteristic frequency band components that can effectively reflect the differences in raw material particle size while filtering out the low-frequency mechanical vibration and high-frequency electrical noise interference inherent in the blast furnace site. This significantly improves the signal-to-noise ratio of the vibration signal and the accuracy of subsequent feature extraction and property inversion, providing clear and reliable original sensing information for the entire adaptive compensation system.

[0011] Preferably, the feature extraction specifically involves: analyzing the vibration acceleration time-domain signal obtained after preprocessing using a sliding time window of fixed length to extract vibration intensity features and particle size distribution features; The vibration intensity characteristics are calculated based on the vibration acceleration time-domain signal obtained after preprocessing, and the calculation formula is as follows:

[0012] in, The values ​​are discrete time-domain vibration acceleration signals, where N is the number of sampling points within the sliding time window. The particle size distribution characteristic is the centroid frequency of the spectrum. A fast Fourier transform is performed on the vibration acceleration time-domain signal obtained after preprocessing to obtain a complex spectrum. The power spectral density of the complex spectrum is calculated. Based on the power spectral density, the centroid frequency of the spectrum is calculated using the following formula:

[0013] in, As for particle size distribution characteristics, For frequency, This represents the power spectral density at the corresponding frequency.

[0014] By adopting the above technical solution, the continuous vibration signal can be processed in real time and in segments through sliding time window analysis, and features with clear physical meaning can be extracted from two complementary dimensions, time domain and frequency domain. These two features together constitute a joint characterization of the physical state of the raw material, providing sufficient and necessary input information for the subsequent machine learning model to perform accurate property inversion.

[0015] Preferably, step S1 also includes receiving online moisture detection data of blast furnace raw materials to obtain moisture level; and obtaining information on the type of current fabric raw materials in real time.

[0016] By adopting the above technical solution and integrating online moisture detection data with raw material type information, a more comprehensive raw material status perception system is constructed. Moisture and type are important auxiliary factors affecting fabric behavior. Inputting them as supplementary information into the subsequent decision-making system enables the generated compensation control commands to be more precise and targeted, thereby further improving the robustness and control effect of the adaptive compensation system under complex and variable working conditions.

[0017] Preferably, in step S2, the pre-trained machine learning inversion model is constructed based on a supervised machine learning method, the construction method including: Vibration signals from historical blast furnace charging processes and offline screening test data of corresponding batches of raw materials are collected to form a training sample set; each sample is a set of "feature-label" pairs, where the feature is a feature vector composed of vibration intensity features and particle size distribution features, and the label is the actual powder ratio value of the corresponding raw material; Using feature vectors as input and labels as target output, a machine learning algorithm suitable for regression problems is used to train the model, and the model parameters are adjusted by an optimization algorithm to establish a mapping relationship from the feature vectors to the powder ratio. The trained model is evaluated using an independent validation dataset. When the evaluation results meet the preset accuracy requirements, the model parameters and structure are solidified into an executable model file for online property inversion.

[0018] By employing the above technical solution, supervised learning is carried out using vibration signals from historical production and authoritative offline test data, enabling the machine learning model to automatically fit the complex nonlinear mapping relationship between vibration characteristics and raw material properties. This data-driven approach avoids the difficulties of complex physical modeling, and once the model is trained, it can respond online in milliseconds, achieving rapid and accurate inversion of the proportion of raw material powders that are difficult to detect directly online.

[0019] Preferably, the fabric compensation rule library in step S3 contains multiple "condition-action" type rules; the conditions are set based on powder ratio and particle size. The actions refer to adjusting the inclination angle of the fabric chute and adjusting the rotational speed of the fabric chute by a percentage.

[0020] By adopting the above technical solution, a structured "condition-action" rule base is constructed, which digitizes and solidifies the valuable experience of blast furnace operation experts on how to adjust the charging parameters for different physical properties of raw materials. The rule conditions are directly based on the physical property parameter settings derived from step S2, ensuring the objectivity and real-time nature of the decision-making basis. The actions are directly aimed at the inclination angle and rotation speed that determine the charging landing point, ensuring a high degree of consistency between the compensation actions and the process objectives.

[0021] Preferably, step S3 includes: matching the real-time acquired powder ratio value and particle size class with the rules in the fabric compensation rule library; When the powder ratio or particle size class meets the conditions of a rule, a compensation control command including a specific tilt angle adjustment value and a speed adjustment percentage is generated according to the action of that rule.

[0022] By adopting the above technical solution, the physical property parameters obtained by real-time sensing and inversion are quickly matched and triggered with the rule base, which can instantly transform the quantitative judgment of the raw material state into specific control instructions that can drive the actuator, thus realizing seamless connection and automated operation from sensing and judgment to control output.

[0023] Preferably, when the powder ratio or particle size grade matches at least two rules simultaneously, the final rule to be executed is determined according to a preset conflict resolution strategy. The conflict resolution strategy is as follows: among all matching rules, the rule with the largest absolute value of the specified compensation amount is selected as the execution rule; wherein, the absolute value of the compensation amount is calculated by weighting the absolute value of the tilt angle adjustment amount and the absolute value of the speed adjustment percentage, and the formula is:

[0024] in, For compensation amount, | | is the absolute value of the tilt adjustment amount specified in the rules; | | represents the absolute value of the speed adjustment percentage specified in the rules; and These are the weighting coefficients for the tilt adjustment amount and the speed adjustment percentage, respectively.

[0025] By adopting the above technical solution and by pre-setting a clear conflict resolution strategy, the decision-making conflict problem that may arise when the raw material state simultaneously meets multiple rule conditions is solved. The strategy of prioritizing the larger absolute value of the compensation amount reflects the process logic of taking stronger compensation actions when dealing with significant fluctuations in physical properties, ensuring the uniqueness and certainty of the decision result, and tending to select control actions that can produce more obvious compensation effects, thereby enhancing the system's response strength and reliability.

[0026] Preferably, in step S3, the generation of the compensation control command includes: After selecting a single execution rule, the target tilt angle and target rotation speed are calculated based on the compensation action specified by the rule and the current fabric reference parameters. The target tilt angle is obtained by adding the adjustment amount specified in the rules to the current reference tilt angle, and the target rotational speed is obtained by adjusting the current reference rotational speed according to the percentage specified in the rules. Based on the target tilt angle and target speed, a compensation control command is generated that includes the target tilt angle setting value and the target speed setting value.

[0027] By adopting the above technical solution, the specific generation process of compensation control commands is clarified. This process combines the abstract adjustment amount in the rules with the current real-time reference parameters of the fabric system, and obtains a precise target setpoint that can be directly issued to the actuator through determined mathematical calculations. This ensures the accuracy and repeatability of the compensation action and is a key step in achieving precise closed-loop control.

[0028] Secondly, this application also provides an adaptive compensation system for blast furnace charging based on the vibration spectrum of the chute, comprising: The sensing module is used to collect vibration signals from the charging chute in real time during the blast furnace charging process; The signal processing and analysis module is used to preprocess and extract features from the acquired vibration signals to obtain vibration intensity features and particle size distribution features. The material property inversion and judgment module has a built-in pre-trained machine learning inversion model, which is used to input vibration intensity features and particle size distribution features into the model, and calculate and output the powder ratio and particle size grade of the current raw material in real time. The intelligent decision-making module has a built-in fabric compensation rule library, which is used to match the powder ratio value and particle size grade with the rule library to generate compensation control commands for adjusting the inclination angle of the fabric chute and adjusting the rotation speed of the fabric chute by a percentage. The execution and communication module is used to send the compensation control command to the blast furnace bellless top control system to drive the charging chute to complete the in-furnace distribution of raw materials at the compensated inclination angle and rotation speed.

[0029] The beneficial effects of this invention are as follows: By real-time acquisition and analysis of vibration signals generated by the impact of raw materials on the chute during blast furnace charging, and through preprocessing and feature extraction, these signals are transformed into feature vectors that can quantify the particle size and flow rate of the raw materials. A pre-trained machine learning inversion model is then used to achieve online, real-time, and direct perception of the proportion and particle size of the raw material powder. By constructing a charging compensation rule base based on these real-time physical property parameters as decision triggering conditions, the process knowledge of blast furnace operators is transformed into automatically executable digital intelligent decisions, establishing a strong correlation between the basis of control decisions and the final process objective. After obtaining accurate physical property judgments, quantitative adjustment commands for the inclination angle and rotation speed of the charging chute are generated based on the rule base. These compensation commands are then issued and executed in real-time within a single charging cycle, forming a rapid feedforward compensation closed loop. This allows the compensation action to directly and accurately act on the charging trajectory deviation caused by changes in raw material particle size, thereby achieving a qualitative improvement in mechanism targeting, real-time response, and consistency of effect.

[0030] Furthermore, the design principle of this invention is reliable, the structure is simple, and it has a very wide range of application prospects.

[0031] Therefore, it is evident that the present invention has substantial features and progress compared with the prior art, and the beneficial effects of its implementation are also obvious. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0033] Figure 1 This is a flowchart of an adaptive compensation method for blast furnace charging based on the vibration spectrum of a chute.

[0034] Figure 2 This is a schematic diagram of a blast furnace charging adaptive compensation system based on the vibration spectrum of a chute.

[0035] Among them, 1-perception module, 2-signal processing and analysis module, 3-physical property inversion and judgment module, 4-intelligent decision-making module, and 5-execution and communication module. Detailed Implementation

[0036] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The following embodiments are explanations of the present invention, but the present invention is not limited to the following implementation methods.

[0037] Example 1: This embodiment provides an adaptive compensation method for blast furnace charging based on the vibration spectrum of the chute, such as... Figure 1 As shown, it includes the following steps: Step S1: During the blast furnace charging process, the vibration signal of the charging chute is collected, and preprocessed and feature extracted to obtain vibration intensity characteristics and particle size distribution characteristics. Step S2: Input the vibration intensity characteristics and particle size distribution characteristics into the pre-trained machine learning inversion model to calculate and output the powder ratio and particle size grade of the current raw material in real time. Step S3: Match the powder ratio and particle size with the preset fabric compensation rule library to generate compensation control instructions; Step S4: Based on the compensation control command, drive the charging chute to complete the in-furnace distribution of raw materials in the blast furnace.

[0038] Furthermore, the core of step S1 lies in acquiring vibration signal characteristics that can characterize the physical properties of the raw materials, specifically including: In this embodiment, in order to achieve real-time monitoring of the vibration of the feeding chute, a high-temperature resistant vibration sensor is installed in the non-direct impact area on the back of the chute; the high-temperature resistant vibration sensor is preferably an industrial-grade piezoelectric accelerometer that can withstand the high-temperature environment of the blast furnace top; this installation position can effectively capture the vibration generated by the material flow impacting the chute liner, and avoid direct impact from the material, thereby ensuring clear and stable signals and extending the service life of the sensor in the high-temperature environment of the furnace top. During the blast furnace charging process, industrial-grade piezoelectric accelerometers work continuously to collect the original vibration acceleration time-domain signal generated by the impact of raw materials in real time and output this continuous time-domain waveform data. The acquired raw vibration acceleration time-domain signal is preprocessed and its features are extracted; the preprocessing includes signal filtering and noise reduction. A bandpass filter is used to filter out background interference in the vibration acceleration time-domain signal. The lower passband frequency of the bandpass filter is set to the mechanical vibration frequency generated by the tilting of the chute and the operation of the rotating equipment to filter out low-frequency background mechanical noise. The upper passband frequency of the bandpass filter is set to the electromagnetic noise frequency generated by the power grid and electrical equipment to avoid high-frequency electromagnetic interference. The passband frequency band of the bandpass filter covers the frequency domain range where the impact vibration characteristics of block materials and powder materials differ most significantly, retaining effective key signal components for subsequent material particle size inversion. In this embodiment, a bandpass filter with a passband of 50Hz-500Hz is selected. The lower limit of 50Hz can effectively filter out the background mechanical vibration of conventional equipment below this frequency, and the upper limit of 500Hz can avoid common high-frequency electrical noise on site. Experimental verification shows that the 50-500Hz frequency band can completely retain the vibration characteristic information that characterizes the difference in raw material particle size. After completing the bandpass filtering, the signal is further processed by the moving average method or wavelet threshold denoising method to suppress random impulse noise and improve the signal-to-noise ratio.

[0039] Furthermore, in an embodiment of the present invention, the window length N of the moving average method is determined by the sampling frequency, and its calculation formula is as follows:

[0040] in, The sampling frequency of the signal; The wavelet thresholding denoising method specifically employs a 5-level decomposition using the db4 wavelet basis and selects soft thresholding; in the soft thresholding, the threshold function is:

[0041] Where x is the wavelet coefficient and λ is the threshold set based on the noise statistical characteristics of the signal; The feature extraction specifically involves: analyzing the vibration acceleration time-domain signal obtained after preprocessing using a sliding time window of fixed length, and extracting key time-domain and frequency-domain features for subsequent property inversion, including vibration intensity features and particle size distribution features; The vibration intensity characteristic is the root mean square (RMS) value, which is calculated based on the vibration acceleration time-domain signal obtained after preprocessing. The vibration intensity characteristic reflects the average energy level of the signal and is positively correlated with the instantaneous flow rate of the raw material in the impact chute. The larger the flow rate, the stronger the impact energy, and the higher the RMS value of the vibration signal is usually.

[0042] The calculation formula can be expressed as:

[0043] in, The values ​​are discrete time-domain vibration acceleration signals, where N is the number of sampling points within the sliding time window. The particle size distribution characteristic is the centroid frequency Fc of the spectrum. A fast Fourier transform is performed on the preprocessed vibration acceleration time-domain signal to obtain its complex spectrum. The power spectrum of the complex spectrum is then calculated. In engineering implementation, the power spectrum is typically calculated using the periodogram method, with the following formula:

[0044] in, is the power spectral density value at the i-th frequency component; N is the number of signal sampling points within a time window; Let i be the complex spectrum value of the i-th frequency component obtained after performing a Fast Fourier Transform on the time-domain signal; | indicates The model; The calculated power spectrum In the above calculation, the centroid frequency Fc of the power spectrum is calculated; the centroid frequency is the first moment of the power spectrum, representing the average frequency of the vibration signal energy; its calculation formula is:

[0045] in, For frequency, This represents the power spectral density at the corresponding frequency.

[0046] Where Fc is the extracted particle size distribution feature, i.e., the frequency of the centroid of the spectrum; This represents the actual frequency value corresponding to the k-th frequency component. =k*( ), The sampling frequency of the signal; For the corresponding frequency The power spectral density value; In one embodiment of the present invention, the method accesses online moisture detection data of blast furnace raw materials; this data is provided in real time by an online moisture detector deployed above the raw material batching belt, weighing hopper or transfer belt, and outputs as a moisture level or a specific percentage value; this "moisture level" parameter can serve as an auxiliary raw material property information, and the moisture level includes high, normal, and low. During the blast furnace charging process, the system reads the raw material type code that strictly corresponds to the current charging instruction from the blast furnace charging control program or production execution system in real time. This code clearly identifies the type of raw material falling in the charging chute, including sinter, pellets, lump ore, or coke. This information serves as a raw material type label and is strictly aligned in time with the subsequently extracted vibration signal characteristics, together forming a complete physical property sensing data stream.

[0047] Thus, step S1 completes its function: it transforms the original physical impact signal into two vibration intensity characteristics and particle size distribution characteristics with clear physical meaning, which can be used to quantify the raw material flow rate and particle size distribution; these characteristic values, as standardized data outputs, provide direct input for the physical property parameter inversion in the subsequent step S2.

[0048] Step S2 receives the root mean square value of vibration intensity feature RMS and the centroid frequency Fc of particle size distribution feature spectrum output from step S1, and calculates and outputs key process parameters characterizing the current physical state of the raw material in real time through a pre-trained machine learning inversion model. The key process parameters include the powder ratio of the raw materials and the particle size grade determined accordingly; the powder ratio is the percentage of particles smaller than 5 mm; in one embodiment of the present invention, when the powder ratio calculated by inversion is greater than 20%, the particle size grade is determined to be fine. As another implementation method for particle size classification, it is also possible to make a quick judgment directly based on the centroid frequency Fc of the spectrum: when Fc is at 150Hz±10Hz, it is determined that the material is mainly blocky; when Fc is greater than 200Hz, it is determined that the proportion of powder has increased; when Fc jumps from the reference value to 220Hz, it is determined that the particle size has become significantly finer. The construction of the pre-trained machine learning inversion model is based on a supervised machine learning method, and its construction process includes: Vibration signals from the historical blast furnace charging process and offline screening test data of the corresponding batch of raw materials were collected to form a training sample set consisting of "feature-label" pairs. The feature is a feature vector (RMS, Fc) composed of vibration intensity features and particle size distribution features extracted using the same method as in step S1, and the label is the actual powder ratio value. A machine learning algorithm suitable for regression problems is selected, and the model is trained with the features as input and the labels as the target output. The model parameters are adjusted by the optimization algorithm so that the model can learn the nonlinear mapping relationship from vibration features to powder ratio. For the vibration signals collected from the historical blast furnace charging process, the same signal processing and feature extraction methods as described in step S1 are used to calculate the feature vector (RMS, Fc) composed of vibration intensity features and particle size distribution features corresponding to each time window. These feature values ​​are then paired with the actual powder ratio values ​​of the corresponding batch of raw materials obtained through offline screening and testing with strict time alignment to form an initial sample set. Invalid samples caused by equipment malfunction or unstable operation are removed to complete data cleaning.

[0049] The cleaned sample set is randomly divided into a training set, a validation set, and a test set in a ratio of 7:2:1. The training set is used for model learning, the validation set is used to adjust hyperparameters and monitor the training process, and the test set is used to finally evaluate the model's generalization performance. Using the (RMS, Fc) features in the training set as input and the corresponding true powder ratio as the target output, a feedforward neural network is used to train the model, specifically including: Initialize the weights and bias parameters of the feedforward neural network to prevent gradient vanishing or exploding; for each batch of samples in the training set, perform the following calculations: The input features (RMS, Fc) are fed into the first layer of the feedforward neural network; the signal is passed layer by layer through linear transformation and nonlinear activation functions until the output layer; the output layer generates a predicted powder ratio value; the mean squared error between the predicted values ​​and the corresponding true powder ratio values ​​of all samples in the batch is calculated as the loss value of the current model. The formula for calculating the mean squared error is:

[0050] Where L is the loss and N is the batch size. It is the actual value. These are model predictions; By using the chain rule, starting from the output layer, the gradient of the loss function is calculated with respect to the parameters of each layer in the neural network; this process determines how much error each weight and bias contributes to the total loss. The network parameters are updated using the stochastic gradient descent optimization algorithm based on the calculated gradient. The parameter update formula is as follows:

[0051] Where θ represents the network parameters; α is the learning rate, a preset hyperparameter that controls the update step size; It is the gradient of the loss with respect to the parameters. Repeat the above steps to traverse the entire training set multiple times; each complete traversal of the training set is called a training cycle. During training, the validation set is used periodically to evaluate the model’s performance on unseen data, helping to tune hyperparameters and early stopping strategies. When the training reaches the preset maximum number of training epochs, the loss on the training set decreases below the preset threshold, or the loss on the validation set no longer decreases within several consecutive training epochs, it indicates that the model may be starting to overfit. At this point, training should be stopped and the model should be rolled back to the state with the lowest validation loss. The final model is evaluated using an independent test set; evaluation metrics such as mean absolute error and coefficient of determination are calculated. The formula for calculating the mean absolute error is:

[0052] Where M is the number of samples in the test set. It is the actual value. It is a predicted value; The formula for calculating the coefficient of determination is as follows:

[0053] Where M is the number of samples in the test set. It is the mean of the true values ​​in the test set; The trained model is evaluated using an independent test set. Evaluation metrics include mean absolute error and coefficient of determination. When the mean absolute error of the model on the test set is less than 2.0% and the coefficient of determination is greater than 0.85, the model accuracy is considered to meet the process requirements and the verification is qualified.

[0054] The validated model parameters and structure are solidified into an executable file or a serialized model file and integrated online into the property inversion and judgment module; the online system calls the model, inputs the real-time feature vector (RMS, Fc), and obtains the powder ratio prediction value.

[0055] The mapping relationship learned by the model depends on the inherent physical relationship between the vibration signal and the material properties of the raw material. During the blast furnace charging process, the powder ratio of the raw material directly affects the energy distribution and frequency characteristics of the impact vibration signal: the vibration intensity characteristic RMS mainly reflects the impact momentum of the material flow, and the centroid frequency Fc mainly reflects the main frequency distribution of the impact event. The higher the proportion of fine particles, the higher Fc usually becomes. The feature vector (RMS, Fc) constitutes a joint characterization of the powder ratio of the raw material. Through training, the model learns a continuous function f from a two-dimensional feature space to a one-dimensional target space, namely powder ratio = f(RMS, Fc; θ), where θ is the model parameter; the mapping relationship is established automatically by minimizing the loss function between the model's predicted value and the true value on the training set. When used online, the RMS and Fc feature values ​​extracted in real time in step S1 are input into the model. The model performs forward calculations through its internally fixed function f, and can instantly output the predicted powder ratio of the current raw material. Thus, step S2 transforms the digital features of the signal obtained in step S1 into powder proportion and particle size grade, which directly characterize the physical state of the raw material. These parameters, as standardized physical property judgment results, provide a unique and quantitative input basis for the intelligent decision-making of fabric parameters in the subsequent step S3.

[0056] Step S3 receives the powder ratio and particle size distribution output in step S2; based on the powder ratio and particle size distribution, it automatically makes intelligent decisions by querying a built-in fabric compensation rule library and generates compensation control instructions for the fabric chute to actively offset the adverse effects of raw material property fluctuations on the fabric application effect.

[0057] The fabric compensation rule base is a knowledge set containing multiple "condition-action" (IF-THEN) type rules; the condition part of each rule is based on the physical property parameters output in step S2, while the action part defines quantitative adjustment instructions for the fabric actuator, specifically including the adjustment angle of the fabric chute inclination angle and the adjustment percentage of the fabric chute rotation speed. In actual operation, the real-time powder ratio and particle size are matched one by one with each rule in the rule base. When the physical property parameters meet the conditions of a rule, the rule is triggered. The system then executes the action part of the rule and generates the corresponding compensation control command, which contains the specific tilt angle adjustment value and speed adjustment percentage.

[0058] The rules in this embodiment include: Rule R001: If the powder ratio is >20%, then reduce the inclination angle of the fabric chute by 0.7° and reduce the rotation speed of the fabric chute by 5%. Rule R002: If the powder ratio is >15%, then reduce the inclination angle of the fabric chute by 1.0° and reduce the rotation speed of the fabric chute by 5%. Rule R003: If the particle size class is "fine" and the moisture class is "normal", then reduce the fabric chute inclination angle by 0.7° and reduce the fabric chute rotation speed by 5%. Rule R004: When the particle size class is "normal" and the moisture class is "high", THEN will reduce the inclination angle of the fabric chute by 0.5° and increase the rotation speed of the fabric chute by 3%. Rule R005: When the raw material type is "pellet", THEN increases the reference dip angle by 1.0°.

[0059] When at least two rules are matched simultaneously, the final rule to be executed is determined according to the preset conflict resolution strategy. The preset priority logic in this embodiment is: among all matched rules, the rule with the largest absolute value of compensation is selected as the final execution rule. The compensation amount here can be defined as the weighted comprehensive evaluation value of tilt adjustment amount and speed adjustment percentage, or the main compensation item preset by the system. In this embodiment, the absolute value of the compensation amount is calculated by weighting the absolute value of the tilt angle adjustment and the absolute value of the speed adjustment percentage, and the formula is as follows:

[0060] in, For compensation amount, | | is the absolute value of the tilt adjustment amount specified in the rules; | | represents the absolute value of the speed adjustment percentage specified in the rules; and These are the weighting coefficients for the tilt adjustment amount and the speed adjustment percentage, respectively, and they satisfy... + =1; In this embodiment, considering that the tilt angle adjustment has a more direct and significant impact on the fabric landing point, it is set to 1. =0.7, =0.3.

[0061] After selecting a single execution rule, the system calculates the compensated fabric execution parameters based on the compensation action specified by that rule and the current fabric reference parameters; the calculation follows the general formula below: Compensated caster angle:

[0062] in, To compensate for the inclination angle of the fabric chute, The reference tilt angle for the current fabric operation is Δα, which is the adjustment amount for the tilt angle specified in the execution rules. Speed ​​after compensation:

[0063] in, To compensate for the rotational speed of the fabric chute, The reference rotational speed for the current fabric operation is δ, which is the percentage adjustment of the rotational speed specified in the execution rules. An increase is positive and a decrease is negative, and it is expressed in decimal form. Based on the above calculation results, a precise tilt angle setting value is generated. and precise speed setting value Compensation control commands; At this point, step S3 transforms the physical property parameters into compensation control commands. This step essentially transforms the experience of blast furnace operation experts into automatically executable digital rules, which is the key intelligent hub for realizing a closed loop from perception to execution.

[0064] Step S4 receives the compensation control command generated in step S3; the command contains precise set values: the compensated fabric chute inclination angle and the compensated fabric chute rotation speed. The compensation control command is sent in real time to the blast furnace bell-less top control system through the system communication link; the top control system parses the command to obtain the target tilt angle and target rotation speed. The furnace top control system drives the inclination adjustment mechanism and rotation drive mechanism of the charging chute to perform the following compensation actions: The control system drives the chute to adjust its inclination angle from the current reference angle to the target angle specified by the command; at the same time, the control system adjusts the rotation drive motor of the chute to adjust its speed from the current reference speed to the target speed specified by the command. The aforementioned adjustments to the tilt angle and rotation speed are completed rapidly within a single fabric feeding cycle. Once the adjustments are complete, the fabric chute immediately resumes the remaining fabric feeding operation for the current batch with new, compensated parameters.

[0065] At this point, step S4 completes its function: it accurately and quickly transforms the digital instructions output by the upstream intelligent decision-making module into the actual physical actions of the fabric chute, realizing feedforward real-time compensation for fluctuations in the physical properties of raw materials.

[0066] Example 2: This embodiment provides an adaptive compensation system for blast furnace charging based on the vibration spectrum of the chute, such as... Figure 2 As shown, it includes: Sensing module 1 collects vibration signals from the charging chute in real time during the blast furnace charging process; Signal processing and analysis module 2 preprocesses and extracts features from the collected vibration signals to obtain vibration intensity features and particle size distribution features; The material property inversion and judgment module 3 has a built-in pre-trained machine learning inversion model. The vibration intensity characteristics and particle size distribution characteristics are input into the machine learning inversion model to calculate and output the powder ratio and particle size grade of the current raw material in real time. The intelligent decision-making module 4 has a built-in fabric compensation rule library, which is used to match the powder ratio value and particle size grade with the fabric compensation rule library to generate compensation control commands for adjusting the inclination angle of the fabric chute and adjusting the percentage of the rotation speed of the fabric chute. The execution and communication module 5 sends the compensation control command to the blast furnace bellless top control system, driving the charging chute to complete the in-furnace distribution of raw materials at the compensated inclination angle and rotation speed.

[0067] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the systems disclosed in the embodiments; relevant details can be found in the method section.

[0068] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0069] In the embodiments provided by this invention, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0070] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0071] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit.

[0072] Similarly, in the various embodiments of the present invention, each processing unit can be integrated into a functional module, or each processing unit can exist physically, or two or more processing units can be integrated into a functional module.

[0073] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0074] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0075] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A method for adaptive compensation of blast furnace charging based on the vibration spectrum of a chute, characterized in that, Includes the following steps: Step S1: During the blast furnace charging process, the vibration signal of the charging chute is collected, and preprocessed and feature extracted to obtain vibration intensity characteristics and particle size distribution characteristics. Step S2: Input the vibration intensity characteristics and particle size distribution characteristics into the pre-trained machine learning inversion model to calculate and output the powder ratio and particle size grade of the current raw material in real time. Step S3: Match the powder ratio and particle size with the preset fabric compensation rule library to generate compensation control instructions; Step S4: Based on the compensation control command, drive the charging chute to complete the in-furnace distribution of raw materials in the blast furnace.

2. The method according to claim 1, characterized in that, The preprocessing described in step S1 involves using a bandpass filter to filter the vibration acceleration time-domain signal; The lower passband frequency of the bandpass filter is set to the mechanical vibration frequency generated by the tilting and rotating equipment of the chute, and the upper passband frequency is set to the electromagnetic noise frequency generated by the power grid and electrical equipment.

3. The method according to claim 2, characterized in that, The feature extraction specifically involves analyzing the vibration acceleration time-domain signal obtained after preprocessing using a sliding time window of fixed length to extract vibration intensity features and particle size distribution features. The vibration intensity characteristics are calculated based on the vibration acceleration time-domain signal obtained after preprocessing, and the calculation formula is as follows: in, The values ​​are discrete time-domain vibration acceleration signals, where N is the number of sampling points within the sliding time window. The particle size distribution characteristic is the centroid frequency of the spectrum. A fast Fourier transform is performed on the vibration acceleration time-domain signal obtained after preprocessing to obtain a complex spectrum. The power spectral density of the complex spectrum is calculated. Based on the power spectral density, the centroid frequency of the spectrum is calculated using the following formula: in, As for particle size distribution characteristics, For frequency, This represents the power spectral density at the corresponding frequency.

4. The method according to claim 3, characterized in that, In step S2, the pre-trained machine learning inversion model is constructed based on a supervised machine learning method, which includes: Vibration signals from historical blast furnace charging processes and offline screening test data of corresponding batches of raw materials are collected to form a training sample set; each sample is a "feature-label" pair, where the feature is a feature vector composed of vibration intensity features and particle size distribution features, and the label is the actual powder ratio value of the corresponding raw material; Using feature vectors as input and labels as target output, a machine learning algorithm suitable for regression problems is used to train the model, and the model parameters are adjusted by an optimization algorithm to establish a mapping relationship from the feature vectors to the powder ratio. The trained model is evaluated using an independent validation dataset. When the evaluation results meet the preset accuracy requirements, the model parameters and structure are solidified into an executable model file for online property inversion.

5. The method according to claim 4, characterized in that, The fabric compensation rule library mentioned in step S3 contains multiple "condition-action" type rules; the conditions are set based on powder ratio values ​​and particle size grades. The actions refer to adjusting the inclination angle of the fabric chute and adjusting the rotational speed of the fabric chute by a percentage.

6. The method according to claim 5, characterized in that, Step S3 includes: matching the real-time acquired powder ratio and particle size class with the rules in the fabric compensation rule library; When the powder ratio or particle size class meets the conditions of a rule, a compensation control command including a specific tilt angle adjustment value and a speed adjustment percentage is generated according to the action of that rule.

7. The method according to claim 5, characterized in that, When the powder ratio or particle size class matches at least two rules simultaneously, the final rule to be executed is determined according to a preset conflict resolution strategy. The conflict resolution strategy is as follows: among all matching rules, the rule with the largest absolute value of the specified compensation amount is selected as the execution rule; wherein, the absolute value of the compensation amount is calculated by weighting the absolute value of the tilt angle adjustment amount and the absolute value of the speed adjustment percentage, and the formula is: in, For compensation amount, | | is the absolute value of the tilt adjustment amount specified in the rules; | | represents the absolute value of the speed adjustment percentage specified in the rules; and These are the weighting coefficients for the tilt adjustment amount and the speed adjustment percentage, respectively.

8. The method according to claim 7, characterized in that, In step S3, the generation of the compensation control command includes: After selecting a single execution rule, the target tilt angle and target rotation speed are calculated based on the compensation action specified by the rule and the current fabric reference parameters. The target tilt angle is obtained by adding the adjustment amount specified in the rules to the current reference tilt angle, and the target rotational speed is obtained by adjusting the current reference rotational speed according to the percentage specified in the rules. Based on the target tilt angle and target speed, a compensation control command is generated that includes the target tilt angle setting value and the target speed setting value.

9. The method according to claim 1, characterized in that, Step S1 also includes receiving online moisture detection data of blast furnace raw materials, obtaining moisture level, and obtaining information on the type of current fabric raw materials in real time.

10. A blast furnace charging adaptive compensation system based on chute vibration spectrum, characterized in that, include: The sensing module collects vibration signals from the charging chute in real time during the blast furnace charging process. The signal processing and analysis module preprocesses and extracts features from the acquired vibration signals to obtain vibration intensity features and particle size distribution features. The physical property inversion and judgment module has a built-in pre-trained machine learning inversion model. Vibration intensity characteristics and particle size distribution characteristics are input into the machine learning inversion model to calculate and output the powder ratio and particle size grade of the current raw material in real time. The intelligent decision-making module has a built-in fabric compensation rule library. It matches the powder ratio value and particle size grade with the fabric compensation rule library to generate compensation control commands for adjusting the inclination angle of the fabric chute and adjusting the rotation speed of the fabric chute by a percentage. The execution and communication module sends the compensation control command to the blast furnace bellless top control system, driving the charging chute to complete the in-furnace distribution of raw materials at the compensated inclination angle and rotation speed.

Images