Real-time temperature control system, method and storage medium of lime rotary kiln

Abstract

The application discloses a real-time temperature control system and method of a lime rotary kiln and a storage medium, relates to the technical field of lime rotary kiln production control, and has the technical points that granularity distribution is extracted from material feeding data and is divided into multiple granularity clusters, time sequence offset correction is performed on temperature data of different measuring points in combination with axial movement characteristics of materials of the clusters, the corrected temperature data can accurately match the complete calcination process of the same batch of materials of the same granularity from feeding to discharging, the real calcination heating time sequence is restored, and data support that conforms to actual production states is provided for subsequent control decisions; on the basis, through a temperature-activity mapping model of the clusters, product lime activity pre-estimates of the granularity clusters are obtained first, and then, with activity uniformity among different granularities as a core judgment criterion, in combination with deviation directions of the clusters, output power of segmented combustion devices of corresponding calcination stages is adjusted in a differentiated manner.

Description

Technical Field

[0001] This invention relates to the field of lime rotary kiln production control technology, and in particular to a real-time temperature control system, method and storage medium for lime rotary kilns. Background Technology

[0002] The rotary kiln is the core industrial equipment for producing quicklime by high-temperature calcination of limestone. The precision of temperature control within the kiln directly determines the activity level and quality consistency of the finished quicklime.

[0003] The temperature control scheme commonly used in the industry involves first evenly distributing several fixed temperature measuring points along different sections of the rotary kiln axis to collect real-time temperature data of the kiln wall or inside the kiln. The control system then compares the collected real-time temperature with the preset temperature range and directly adjusts the overall output power of the combustion device. The entire control logic aims to maintain a stable kiln temperature, ultimately ensuring that the average activity of the entire batch of finished lime meets production standards.

[0004] During data acquisition, since the temperature measuring points are fixed on the kiln structure, and the material entering the kiln is continuously conveyed forward along the axial direction by relying on the rotation and inclination of the kiln, the temperature data collected from different measuring points actually correspond to the instantaneous calcination state of different batches of material when they pass through that measuring point. The temperature data of the upstream kiln section and the downstream kiln section cannot match the heating process of the same batch of material. For example, the material temperature collected at the kiln tail feed end needs to be conveyed for tens of minutes to reach the kiln head discharge end. The two belong to completely different feed batches and cannot be pieced together to form a complete heating trajectory of the same batch of material from feed to discharge. The calcination state deduced based on this has a natural deviation, which directly limits the upper limit of the accuracy of temperature control. Using only the average activity of the entire batch of products as the evaluation index, the industry generally believes that the difference in calcination quality of materials with different particle sizes fed into the kiln is an inherent phenomenon caused by the particle size characteristics of the raw materials. Therefore, the uniformity of activity of finished products with different particle sizes has never been included in the control dimension. However, in actual production, fine-particle limestone heats up quickly and requires a short high-temperature calcination time, while coarse-particle limestone heats up slowly and requires a longer high-temperature residence time. The control method of uniformly adjusting the combustion power cannot take into account the calcination needs of materials with different particle sizes. In the end, the problem of over-burning of fine particles and under-burning of coarse particles is likely to occur, resulting in large fluctuations in the activity of the entire batch of products. This makes it difficult to meet the requirements of downstream high-end scenarios such as metallurgy and chemical industry for lime quality stability. To this end, we propose a real-time temperature control system, method and storage medium for lime rotary kilns. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a real-time temperature control system, method, and storage medium for a lime rotary kiln. The core control is reversed from the traditional kiln temperature standard achievement to achieving the standard of uniform activity of finished lime across all particle sizes. Through a multi-layered nested technical solution involving particle size grouping time-series correction, diffusion accuracy optimization, group-specific mapping model, and deviation-driven adaptive update, the activity difference of finished products with different particle sizes is reduced, thereby improving the quality stability of the entire batch of products.

[0006] To solve the above problems, the technical solution adopted by the present invention is as follows: A real-time temperature control method for a lime rotary kiln, the method comprising the following steps: Acquire multiple sets of operational source data in the lime rotary kiln operation scenario. The operational source data includes multi-point temperature data of the kiln body, combustion condition data, material feed data, and flue gas emission data. Based on the source data, differentiated data acquisition cycle information is generated and multiple batches of samples are taken to construct multiple batches of temperature sample sets. The temperature sample sets are integrated and analyzed to obtain the temperature acquisition quality assessment value. The particle size distribution parameters of the material entering the kiln are extracted from the material feed data. The material is divided into multiple particle size groups according to the particle size range. Combined with the axial movement characteristics of the material corresponding to each particle size group, the time offset correction is performed on the temperature data of multiple points in the kiln body to generate the calcination heating time sequence that is independent for each particle size group. Using the temperature acquisition quality assessment value as the time series data validity correction coefficient, and combining the calcination heating time series of each particle size group, the estimated value of the finished lime activity corresponding to each particle size group is obtained by converting the temperature-activity mapping model corresponding to the group. The activity dispersion statistic of the estimated activity of finished lime in all particle size groups is statistically analyzed and used as the core judgment benchmark for temperature control. The activity dispersion statistic is used to characterize the activity uniformity of finished lime of different particle sizes. If the activity dispersion statistic exceeds the preset qualified threshold, the output power of the segmented combustion device set along the kiln axis corresponding to each calcination stage is adjusted differently according to the activity deviation direction of each particle size group. Otherwise, if the threshold is not exceeded, the current operating state of the segmented combustion device is maintained.

[0007] Preferably, when acquiring the source data, multiple sets of raw operating data are collected simultaneously by locating multiple data access nodes throughout the entire production process of the lime rotary kiln, and the collected raw operating data is then classified. Specifically, multi-point temperature data of the kiln body is identified from the original operating data by the temperature field distribution characteristics; combustion condition data is extracted by classifying from the set of combustion equipment operating parameters; material feeding data is generated by sorting based on material conveying metering parameters and inlet particle size detection results; and flue gas emission data is obtained by screening and separating based on flue gas component concentration and flue gas flow parameters.

[0008] Preferably, the specific steps for generating the calcination heating sequence are as follows: For each particle size group that has been divided, the axial travel rate of the material and the reference value of the total residence time in the kiln are matched for the corresponding particle size range. The axial travel rate of the material is obtained by combining the kiln rotation speed and kiln inclination in the combustion condition data with the average particle size of the corresponding particle size group in the material feed data. The distribution coordinates of each temperature measuring point along the kiln axis are determined according to the multi-point temperature data of the kiln body, and the theoretical travel time difference of a single particle size group of material between adjacent temperature measuring points is calculated. Using the first temperature measuring point at the kiln tail feed end as the time reference, the temperature data of each downstream temperature measuring point are sequentially time-shifted and corrected along the material travel direction, so that the data of all temperature measuring points correspond to the same particle size material segment of the same batch of feed. The corrected temperature measurement data are sorted according to the calcination stage to generate a preliminary calcination heating sequence of a single particle size group from feed to discharge.

[0009] Preferably, after obtaining the preliminary calcination heating sequence of individual particle size groups, material motion parameters are introduced to perform diffusion-residence time coupling correction. The specific steps are as follows: Based on the average particle size of the corresponding particle size group, the material friction coefficient in the material feed data, and the kiln rotation parameters in the combustion condition data, the axial diffusion coefficient and radial tumbling frequency of the material in this particle size group are calculated. By combining the axial diffusion coefficient and radial tumbling frequency of the material, the probability distribution range of the residence time of the material at each temperature measuring point is derived. According to the residence time probability distribution, the corresponding time nodes of the preliminary calcination heating time sequence are weighted and superimposed for correction, and the corrected single-particle-size cluster calcination heating time sequence is output as the input data of the temperature-activity mapping model.

[0010] Preferably, the temperature-activity mapping model corresponding to the cluster is constructed by fitting historical production data, and the construction process is as follows: Retrieve calcination heating sequence data for each particle size group in historical production batches, actual activity test results of finished lime after sieving for the corresponding particle size group, and flue gas emission data for the corresponding batches. The calcination stage division boundary is corrected by combining the flue gas component concentration in the flue gas emission data. The calcination process of each particle size group is divided into four stages along the kiln axis: preheating zone, decomposition zone, calcination zone, and cooling zone. The four calcination stages correspond one-to-one with the axial setting position of the segmented combustion device. Correlation analysis was used to fit the mapping relationship between the average temperature of each stage and the activity of the finished lime in the corresponding particle size group, and the temperature contribution weight of each stage for each particle size group was obtained. The temperature contribution weights of all granularity groups together constitute the core parameters of the temperature-activity mapping model.

[0011] Preferably, the temperature-activity mapping model employs a differential update mechanism driven by discreteness bias, and the update process is as follows: After completing the particle size test of each batch of finished lime, the absolute deviation between the estimated value and the actual value of the activity of finished lime for each particle size group is calculated. Calculate the proportion of the deviation of each particle size group to the total deviation of all particle size groups. For particle size groups whose deviation proportion exceeds the preset deviation threshold, adjust the corresponding temperature contribution weight of each stage according to the deviation proportion. For other particle size groups whose deviation proportion is lower than the preset deviation threshold, adjust the weights slightly according to the total deviation amplitude. The updated temperature contribution weight is directly used to calculate the estimated activity of finished lime in subsequent production batches, iteratively reducing the overall level of the activity dispersion statistic.

[0012] A real-time temperature control system for a lime rotary kiln, the control system comprising: The data acquisition and preprocessing unit acquires multiple sets of operational source data for the lime rotary kiln operation scenario. The operational source data includes multi-point temperature data of the kiln body, combustion condition data, material feed data, and flue gas emission data. It generates differentiated data acquisition cycles and completes multiple batches of sampling. After constructing a temperature sample set, it outputs the temperature acquisition quality assessment value. The grouping timing correction unit receives multi-point temperature data, combustion condition data and material feed data of the kiln body, extracts the particle size distribution of the material entering the kiln from the material feed data and divides it into multiple particle size groups. It then performs timing offset correction of the multi-point temperature data of the kiln body in combination with the axial motion characteristics of each particle size group and outputs an independent calcination heating timing sequence for each particle size group. The grouping quality prediction unit has a built-in temperature-activity mapping model corresponding to the group. It receives the temperature acquisition quality assessment value and the calcination heating time sequence, and uses the temperature acquisition quality assessment value as a correction coefficient to calculate the finished lime activity prediction value for each particle size group. The dispersion control unit receives the estimated activity of finished lime from all particle size groups, calculates the activity dispersion statistic as the control benchmark, compares it with the preset qualified threshold, and outputs control commands to differentiate the output power of the segmented combustion devices set along the axial direction of the rotary kiln for each calcination stage.

[0013] Preferably, the cluster timing correction unit includes: The diffusion correction sub-unit calculates the axial diffusion coefficient and radial tumbling frequency for a single particle size group, derives the residence time probability distribution range, performs weighted superposition correction on the preliminary calcination heating time sequence, and outputs the final calcination heating time sequence.

[0014] Preferably, the cluster quality prediction unit comprises: The weighted fitting subunits, based on the calcination heating time series data, finished product testing data and flue gas emission data of historical production batches, obtain the initial temperature contribution weight of each particle size group, and construct the initial parameters of the temperature-activity mapping model. The sub-units are dynamically updated, and the weight parameters of the high-deviation particle size group are adjusted first, taking into account the deviation ratio of each batch of finished product testing, to iteratively update the temperature-activity mapping model. The activity calculation subunit calls the current effective weights, temperature acquisition quality assessment values, and real-time calcination heating time sequence data to output the estimated activity of finished lime for each particle size group.

[0015] A computer-readable storage medium storing a computer program, wherein when executed by a processor, the computer program implements the steps of a real-time temperature control method for a lime rotary kiln.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: Particle size distribution is extracted from material feed data and divided into multiple particle size groups. Combined with the axial movement characteristics of each group, time-series offset correction is applied to temperature data at different measuring points. The corrected temperature data accurately matches the complete calcination process of the same batch of material with the same particle size from feed to discharge, reconstructing the true calcination heating sequence and providing data support for subsequent control decisions that align with actual production conditions. Based on this, a temperature-activity mapping model for each particle size group is used to first obtain the estimated activity of the finished product. Then, using the uniformity of activity among different particle sizes as the core judgment criterion, and combined with the deviation direction of each group, the output power of the segmented combustion device at the corresponding calcination stage is adjusted differentially. When the activity of the coarse particle group is low, the calcination zone power is increased; conversely, when the activity of the fine particle group is high, the decomposition zone power is decreased. This specifically compensates for the differences in calcination requirements of different particle sizes, reduces the activity gap between finished products of different particle sizes, and improves the overall quality stability of the lime product.

[0017] Secondly, the accompanying temperature-activity mapping model adopts a deviation-driven update mechanism. After each batch of finished products is tested, the model parameters of the particle size group with a higher deviation ratio are adjusted first. The model can automatically iterate and adapt to the fluctuations of the characteristics of the material entering the kiln and the kiln conditions, without the need for maintenance personnel to frequently manually calibrate the parameters, thus reducing the maintenance cost of long-term operation. Moreover, the segmented combustion device arranged along the kiln axis can independently adjust the power for different calcination stages without the need to change the overall combustion conditions. The control process has less impact on production stability and further refines the control particle size of the calcination process. Attached Figure Description

[0018] Figure 1 This is a flowchart of the entire process of the real-time temperature control method for lime rotary kilns in this invention. Figure 2 This is a detailed internal logic block diagram of the diffusion-residence time coupling correction in this invention; Figure 3 This is a block diagram of the architecture of the real-time temperature control system for the lime rotary kiln in this invention. Figure 4 This is a temperature distribution curve of various measuring points along the axial direction of the kiln body in this invention. Detailed Implementation

[0019] The present invention will be further described below with reference to specific embodiments.

[0020] Taking a metallurgical-grade active lime rotary kiln production line as an example, K-type armored thermocouples inserted into the material layer are evenly arranged along the kiln body axis as temperature measuring points, covering the entire kiln section from the kiln tail feeding to the kiln head discharging; the combustion system is equipped with four sets of independent adjustable pulverized coal burners, i.e., segmented combustion devices, corresponding to the four calcination stages along the axis; an electronic belt scale and an online laser particle size analyzer are configured at the feeding end, and an electrochemical flue gas online analyzer is configured at the kiln tail. All production data are uniformly connected to the Siemens S7-400 series DCS system.

[0021] Example 1: Real-time temperature control method for a basic type of lime rotary kiln, such as Figure 1 As shown, to address the technical issues of misaligned timing of fixed measurement points and the control target focusing only on average activity in existing technologies, and to avoid impacting the operation of existing production lines and reduce production inputs, this approach aims to minimize modifications to existing hardware and adapt to conventional quicklime production scenarios. The specific details are as follows: First, source data is acquired and preprocessed. Data from four categories covering the entire production process—kiln temperature measuring points, burner control modules, feed metering devices, and flue gas analyzers—is connected to the nodes. Then, raw operational data is synchronously acquired through an industrial data acquisition gateway and categorized according to business characteristics to obtain the following four types of core operational source data: The first category is multi-point temperature data of the kiln body, which is identified from the original operating data through the temperature field gradient distribution characteristics. It is a real-time temperature set of 16 axial measuring points. The reason for collecting this temperature is to obtain the calcination temperature distribution of each kiln section in a timely manner, and also to provide basic support for subsequent time-series correction and activity prediction. The second category is combustion condition data, which is extracted from the set of operating parameters of combustion equipment. It includes all operating parameters of the combustion system and the kiln, such as the output power of the segmented combustion device, the kiln speed, the kiln inclination, the kiln rotation frequency, the pulverized coal flow rate, and the primary air volume. It is the core basis for calculating the material movement characteristics. The third category is material feeding data, which is generated by sorting based on material conveying metering parameters and kiln particle size detection results. It includes feeding rate, kiln material particle size distribution, limestone friction coefficient, raw material moisture content, CaO grade and other feeding end metering and raw material characteristic parameters, which are the core basis for particle size grouping. The fourth category is flue gas emission data, which is obtained by screening and separating flue gas component concentration and flue gas flow parameters. It includes emission end detection parameters such as CO concentration, CO2 concentration, O2 concentration, flue gas flow rate, and flue gas temperature of the kiln tail flue gas, which are used for subsequent calcination stage division and atmosphere correction.

[0022] Differentiated data acquisition cycle information is generated based on the fluctuation rate and business weight of the source data. This differs from the traditional coarse-grained mode of uniform sampling frequency, which can reduce the system's computational load while ensuring data accuracy. The baseline cycle is set according to data characteristics: the kiln multi-point temperature data, as the core process variable, has the fastest fluctuation rate and is acquired in 5-second cycles; the combustion condition data, as the actuator parameter changes at a low frequency, is acquired in 10-second cycles; the material feed data has a low fluctuation frequency and is acquired in 30-second cycles; and the flue gas emission data has a strong lag in changes and is acquired in 90-second cycles. Furthermore, the acquisition cycle can be dynamically adjusted according to the operating conditions to cope with various situations. For example, during the heating and commissioning phase or when the kiln temperature fluctuation exceeds 50℃ / h, the temperature acquisition cycle is automatically shortened to 2 seconds to capture rapid changes; when the temperature acquisition quality assessment value is below 0.8, the sampling cycle is shortened to 3 seconds to increase data redundancy and offset the impact of single-point data failure; and sampling is automatically suspended during fault shutdown to avoid invalid data entering the database.

[0023] After completing multiple batches of sampling according to the corresponding cycle, the 3σ principle was used to remove abnormal data during kiln start-up and shutdown, fault stages, and invalid jump points where the temperature jump of a single measuring point exceeded 200°C. The 3σ principle is a common outlier removal rule in industrial data processing, which is based on the characteristics of normal distribution. Data that deviates from the mean by three times the standard deviation is judged as outliers. The filtered data was used to construct multiple batches of temperature sample sets. Multi-dimensional integrated analysis was carried out on the temperature sample sets, and scores were calculated based on three dimensions: data completeness, fluctuation rationality, and sensor calibration status, with weights of 40%, 30%, and 30%, respectively. Among them, data completeness is the ratio of the effective data volume to the theoretically required data volume, directly normalized to the 0-1 range; fluctuation rationality is deducted based on the percentage of jumps exceeding the limit, with no jumps resulting in full marks, and each jump exceeding the limit deducted points. Each instance of exceeding the limit results in a 5% deduction in points; sensor calibration status is penalized based on the duration of overdue calibration, with full marks awarded within the calibration cycle and a 10% deduction per month for overdue calibration; the three weighted sums are normalized to the 0-1 range to obtain the temperature acquisition quality assessment value; under normal operating conditions, this assessment value is stable in the 0.92-0.97 range, and the system automatically triggers a sensor anomaly alarm when it falls below 0.8; this assessment value serves as the data confidence parameter throughout the entire process in subsequent control stages, with the quantification rule being: single power adjustment amplitude = baseline step size × temperature acquisition quality assessment value, dynamic control dead zone width = basic dead zone × (1 - temperature acquisition quality assessment value). The lower the data quality, the smaller the adjustment amplitude and the larger the dead zone, and it only affects the confidence weight of the control stage, without directly correcting the calculated activity value.

[0024] Subsequently, particle size grouping and time sequence offset correction are performed, corresponding to the calcination heating time sequence generation process. Specifically, the particle size distribution parameters of the material entering the kiln are extracted from the material feed data. The moving average of three consecutive measurements by the online particle size analyzer is used to reduce detection fluctuations. Based on the critical particle size values ​​of limestone calcination kinetics, the material is divided into three particle size groups. The division is based on the differences in calcination rate control mechanisms for different particle sizes. For the coarse particle group (25-40mm), the calcination process is controlled by internal diffusion, requiring a longer high-temperature residence time. For the medium particle group (10-25mm), the calcination rate is between the two. For the fine particle group (0-10mm), the calcination process is controlled by external diffusion, with a rapid heating rate and a tendency to over-burn. Each batch of material entering the kiln is assigned a unique batch number to bind the entire data chain. In practical applications, the number of groups can be adjusted to 2-6 based on the raw material particle size distribution, balancing control accuracy and system computational load. Based on the axial movement characteristics of materials corresponding to each particle size group, time-series offset correction is performed on the multi-point temperature data of the kiln body. This is applicable to conventional production conditions with particle sizes of 0-40mm and filling rates of 10%-20%. Larger particle sizes and higher kiln speeds result in faster axial movement of the material and shorter residence time within the kiln. The axial movement rate of the material is calculated using a semi-empirical formula commonly used in the rotary kiln industry: v = k × n × D × sinα, where v is the axial movement rate of the material (in m / min), and k is an empirical coefficient for material movement, typically ranging from 0.02 to 0.0 for limestone raw materials. 5. The larger the particle size and the smaller the friction coefficient, the higher the value should be; n is the kiln rotation speed, in r / min; D is the kiln diameter, in m; α is the kiln installation inclination; taking the rated rotation speed condition of this embodiment as an example, the kiln diameter is 4m, the empirical coefficient for coarse particle grouping is taken as 0.045, and the derived axial travel speed is approximately 0.95m / min; the empirical coefficient for medium particle grouping is taken as 0.035, and the speed is approximately 0.75m / min; the empirical coefficient for fine particle grouping is taken as 0.026, and the speed is approximately 0.55m / min; calibrate the distribution coordinates of 16 temperature measuring points along the kiln axis, such as... Figure 4 As shown, the distance between adjacent measuring points is 2.5m. The theoretical travel time difference between adjacent measuring points for a single particle size group is calculated. Then, taking the first measuring point at the kiln tail feed end as the time reference, the temperature data of each downstream measuring point are sequentially corrected by cumulative time shift. That is, the second measuring point is shifted backward by 1 time difference, the third measuring point is shifted backward by 2 time differences, and so on, so that the temperature data of all measuring points are aligned with the feed time point of the same batch of materials, corresponding to the actual heating process of the same batch of materials with the same particle size. Finally, the corrected temperatures of each measuring point are sorted according to the calcination stage to generate a preliminary calcination heating time sequence of a single particle size group from feed to discharge. The sequence is bound to the batch number and stored in the historical database for subsequent activity prediction and model iteration.

[0025] Finally, the activity prediction and dispersion cyclic control of finished lime were carried out. Combining the calcination heating sequence of each group, the activity prediction value of finished lime corresponding to each particle size group was obtained by conversion through the group basic mapping model. In the calculation, the average temperature of each calcination stage of a single group was first counted, multiplied by the empirical weight of the corresponding stage and summed to obtain the initial calculated value of activity. Then, a flue gas atmosphere correction coefficient was introduced for secondary correction. The correction coefficient was calibrated based on the ratio of O2 to CO concentration at the kiln tail. The coefficient was less than 1 under the reducing atmosphere and approached 1 under the oxidizing atmosphere to offset the error of calcination efficiency reduction caused by the reducing atmosphere. After correction, the final activity prediction value of finished lime was obtained. The activity dispersion statistics of the activity prediction values ​​of finished lime of all particle size groups were statistically analyzed, which is the core control benchmark. In this embodiment, the overall standard deviation is used to characterize the activity uniformity of finished lime of different particle sizes. The overall standard deviation is a statistical measure of the dispersion of a set of data. The smaller the value, the closer the activity of each group is and the better the product uniformity. The preset qualified threshold is taken as the 90th percentile of the dispersion of normal production batches in the past six months, calibrated to 12mL, to ensure that more than 90% of normal batches fall within the qualified range. A dynamic control dead zone is set in combination with the temperature-collected quality assessment value to avoid frequent adjustments that disturb the kiln condition. When the activity dispersion statistics exceed the preset qualified threshold, the output power of the segmented combustion devices set along the kiln axis corresponding to each calcination stage is adjusted differently according to the deviation direction of each group relative to the 320mL target activity. Specifically, if the activity of the coarse particle group is too low, the power of the burner in the later section of the calcination zone is increased to solve the problem of insufficient calcination of coarse particles; if the activity of the medium particle group deviates, the power of the burner in the middle section of the calcination zone is adjusted; if the activity of the fine particle group is too high, the power of the burner in the decomposition zone is decreased to reserve over-burning margin. Under normal operating conditions, the single adjustment step is 1% of the rated power, and at least 10 minutes are reserved after each adjustment before the next judgment is executed to match the large lag characteristics of the rotary kiln and avoid large power fluctuations that cause kiln instability.

[0026] Example 2: This example, based on all the basic schemes of Example 1, adds diffusion-residence time coupling correction, such as... Figure 2 As shown, this embodiment addresses the timing deviations caused by the ideal piston flow assumption, making it suitable for production scenarios with higher control accuracy requirements. The data acquisition, particle size grouping, preliminary timing correction, and closed-loop control processes in this embodiment are completely consistent with Embodiment 1, except that a secondary accuracy correction is nested after the initial calcination heating timing sequence is generated. Since the material inside the rotary kiln is not an ideal piston flow, particles will undergo radial tumbling and axial diffusion as the kiln rotates. Material segments from the same batch will be dispersed within a certain length of the kiln segment. Correcting only based on the average travel rate will ignore the heating deviations caused by particle dispersion. The diffusion deviation of fine particle grouping can reach over 10%. Therefore, diffusion-residence time coupling correction further improves timing accuracy, as detailed below: The correction process is based on a rotary kiln particle motion and diffusion model. First, based on the average particle size of the corresponding particle group, the material friction coefficient in the material feed data, and the kiln rotation parameters in the combustion condition data, the axial diffusion coefficient and radial tumbling frequency of that group are calculated. The axial diffusion coefficient is calculated using the Sahil empirical formula, which is proportional to the -0.5 power of the particle size and 1 / 2 power of the kiln rotation speed. The power is directly proportional to the power; coarse particles have strong tumbling inertia and a higher radial tumbling frequency, while fine particles are more likely to accumulate with the material layer and have a larger axial diffusion coefficient. Combining these two parameters, the probability distribution range of the residence time of the material at each temperature measuring point is derived. A normal distribution is used to represent the residence time probability, with the average travel time point as the distribution center and the standard deviation taken as one sampling step. The sampling window covers five time points, two sampling steps before and after the center. According to the residence time probability distribution, the corresponding time nodes of the preliminary calcination heating sequence are weighted and corrected. The temperature data of the five time points within the window are weighted and summed according to the normal distribution weights, with the center time point accounting for 40% of the weight, and the weights of adjacent points decreasing sequentially. Finally, the calcination heating sequence corrected for particle diffusion deviation is output as the input for subsequent activity prediction.

[0027] Example 3: Based on all the basic schemes in Example 2, this example adds a group-specific temperature-activity mapping model and a differentiated update mechanism driven by dispersion deviation to solve the problems of poor adaptability of fixed models and the need for manual calibration, and adapts to high-end production scenarios with large raw material fluctuations and high quality requirements. The data acquisition, timing correction, diffusion correction and closed-loop control process of this example are completely consistent with Example 2. Unlike the scheme that uses a fixed empirical model, this example uses a group-specific mapping model and sets an automatic iteration mechanism.

[0028] The model was initially built by fitting historical production data. Selected historical production batches were retrieved, with the selection criteria being normal batches that had been running continuously and stably for over 72 hours without failures. The number of valid batches was no less than 30. Leave-one-out cross-validation was used to avoid overfitting. Leave-one-out cross-validation is a common model validation method in machine learning, where one set of samples is reserved as the validation set and the rest as the training set, and iterative validation ensures the model's generalization ability. The training data included three categories: calcination heating time series data for each particle size group, actual activity test results of the finished lime after screening for the corresponding group, and flue gas emission data for the corresponding batch. The activity of the finished product was tested according to GB / T20855 "Determination of Lime Activity". The calcination stage division boundary was corrected by incorporating the flue gas component concentration from the emission data, replacing the simple... The traditional method of temperature-based classification uses the following rules: the starting point where the CO concentration rises by more than 500 ppm for three consecutive sampling points is the decomposition zone; the interval where the O2 concentration is consistently below 2% and the kiln temperature is above 1100℃ is the calcination zone; and the section where the kiln head temperature continues to decrease is the cooling zone. Along the kiln's axial direction, the calcination process of each group is divided into four stages: preheating zone, decomposition zone, calcination zone, and cooling zone. These four stages correspond one-to-one with the axial installation positions of four groups of segmented combustion devices. Each group of burners can independently control the calcination intensity of its corresponding stage, providing a precise basis for subsequent segmented control. Through Pearson correlation analysis combined with least squares regression, the mapping relationship between the average temperature of each stage and the activity of the finished lime in the corresponding group is fitted. After normalizing the temperature of each stage to its rated value, the correlation coefficient is calculated to obtain the specific temperature contribution weight for each stage for each particle size group.

[0029] The typical weights obtained from the fitting in this embodiment are as follows: coarse particle group has the highest weight in the calcination zone due to its short residence time and controlled internal diffusion rate, namely 15% in the preheating zone, 25% in the decomposition zone, 50% in the calcination zone, and 10% in the cooling zone; fine particle group has the highest weight in the decomposition zone due to its long residence time and controlled external diffusion rate, namely 10% in the preheating zone, 40% in the decomposition zone, 35% in the calcination zone, and 15% in the cooling zone; the weight distribution of medium particle group is relatively balanced; the temperature contribution weights of all groups together constitute the core parameters of the temperature-activity mapping model. In this scenario, the target activity of the finished lime is preset to 320mL, which serves as a unified judgment benchmark for the deviation direction of each group. The deviation judgment is assumed to be in the range where the activity increases with the calcination intensity. The purpose of the adjustment is to reserve the margin for over-burning / under-burning in advance.

[0030] The model's differential update employs a dispersion-bias-driven differential update mechanism, prioritizing the convergence of errors in high-bias clusters based on the Pareto principle, resulting in an iteration efficiency improvement of over 40% compared to the overall correction mode. An update is triggered after each batch of finished lime is processed for particle size analysis. Before updating, independent batch matching for each cluster is performed, which is an extended supporting logic: starting with the feeding time of the batch entering the kiln, and adding the average residence time of the corresponding particle size cluster, the finished product test results for the corresponding discharge time period of that cluster are locked, and a one-to-one correspondence is established with the batch number. A ±5-minute tolerance window is set to avoid batch misalignment caused by differences in discharge time between different clusters. After matching, the activity of each cluster is statistically analyzed. The absolute deviation between the estimated value and the actual measured value is calculated as the proportion of the deviation of a single particle size group to the total deviation of all particle size groups. The preset deviation threshold is 30%, which is set based on the industrial adaptation value of deviation contribution, i.e., a few groups contribute the majority of the deviation. For particle size groups whose deviation proportion exceeds the preset deviation threshold, the corresponding temperature contribution weight of each stage is adjusted by amplifying the deviation proportion, and the larger the deviation, the larger the adjustment. For other groups whose deviation proportion is lower than the threshold, the weight is adjusted slightly in sync with the total deviation. The single-dimensional weight adjustment range does not exceed 5% to prevent parameter jumps from affecting system stability. When the operating condition fluctuation exceeds 10%, the update is automatically paused to avoid abnormal operating conditions from biasing the model parameters.

[0031] The updated temperature contribution weight directly overwrites the original parameters and is used to calculate the estimated activity of finished lime in subsequent production batches. As production continues to iterate, the overall level of the activity dispersion statistic is gradually reduced. When the activity dispersion statistic of three consecutive production batches cannot be reduced to within the qualified threshold by adjusting the combustion power, the feed particle size distribution can be linked to optimization. This generates instructions to adjust the discharge particle size ratio of the crushing and screening system, appropriately reducing the proportion of the feed group with the largest deviation. This reduces quality differences from the source of feed, forming a dual-cycle optimization system that combines combustion control and feed control, adapting to complex production scenarios with large raw material fluctuations.

[0032] Example 4: This example is a system architecture implementation of the aforementioned method, such as... Figure 3 As shown, all unit functions and method steps correspond one-to-one, and can be directly deployed in the industrial field; the entire system is deployed on the processor and industrial control host computer, and communicates with the field PLC devices through industrial Ethernet. Each unit forms a data flow loop through the internal data bus, and it contains four core functional units; as detailed below: The first unit is the data acquisition and preprocessing unit, which connects to four types of data access nodes. It uses an industrial data acquisition gateway to achieve hardware connection and is responsible for multi-source data synchronous acquisition, differentiated sampling period configuration, 3σ abnormal data filtering, temperature sample set construction, and finally outputs standardized operating source data and temperature acquisition quality assessment values, which are simultaneously sent to the cluster timing correction unit and the cluster quality prediction unit. The second unit is the clustering timing correction unit, which receives multi-point temperature data, combustion condition data and material feed data of the kiln body output by the acquisition and preprocessing unit. It is responsible for particle size clustering, axial travel rate derivation and preliminary calcination heating timing sequence generation. It has an embedded diffusion correction subunit, which is batch-triggered and responsible for performing diffusion-residence time coupling correction and outputting the final high-precision calcination heating timing sequence to the downstream clustering quality prediction unit. The third unit is the cluster quality prediction unit, which has a built-in temperature-activity mapping model specific to each cluster. It contains three sub-units: the weight fitting sub-unit completes the initial model construction based on historical production data; the dynamic update sub-unit receives the test data of each batch of finished products, iterates the model parameters after completing batch matching; and the activity calculation sub-unit calls the current weight and time series data in real time and outputs the predicted value of the activity of finished lime to the dispersion control unit. The fourth unit is the dispersion control unit, which receives the estimated activity of finished lime from all particle size groups, calculates the activity dispersion statistics, and outputs power control commands based on the temperature acquisition quality assessment value and the qualified threshold to differentiate the output power of the segmented combustion device. This unit has a reserved expansion interface that can be connected to the control module of the feeding crushing and screening system to realize the linkage optimization of the feeding particle size distribution.

[0033] Example 5: This example implements the control method using a program carrier. The control program is stored in an industrial-grade solid-state storage medium and installed in the host computer of the rotary kiln control cabinet. The computer-readable storage medium stores the computer program. When the computer program is executed by the processor, it can implement all the steps of any of the real-time temperature control methods for lime rotary kilns in Examples 1 to 3. The storage medium supports power-off data protection and a dual backup mechanism. It can automatically resume operation after abnormal program interruption, adapting to the harsh environment of industrial sites and the requirements for long-term stable operation. In addition to the core control program, it also stores historical production databases, model parameter libraries, threshold configuration files, batch matching ledgers, operation logs, and other supporting content in partitions. Each module can directly call the parameters and historical data stored in the storage, ensuring the independent and complete operation of the system.

[0034] Comparative Example: This comparative example uses the same production line hardware, raw material conditions and production conditions as the aforementioned embodiments, only the control logic is replaced with the conventional fixed measuring point control scheme commonly used in the lime rotary kiln industry to ensure that the comparison benchmark is unique; this scheme is the mainstream mature technology of domestic lime production lines, with simple logic, low deployment cost and strong operational stability, which can fully meet the needs of ordinary production scenarios with stable raw materials and conventional quality requirements. The system adopts direct temperature control at fixed measuring points. During the data acquisition phase, all production data are collected using a uniform 10-second sampling frequency. Only basic zero-value rejection is performed. There is no multi-dimensional data quality assessment mechanism. Kiln temperature is the only core control input, and other parameters are only used for auxiliary monitoring. The temperature data processing stage directly uses the instantaneous temperature value of fixed measuring points without any time-series offset correction. It does not perform particle size grouping processing on the material entering the kiln, and assumes that the material in the kiln is in an ideal piston flow state. All particle sizes share the same set of temperature judgment standards. The temperature-activity mapping adopts a unified manual experience model for all particle sizes. During the calcination stage, it is only divided according to a fixed temperature threshold. There is no automatic update mechanism. The parameters are manually calibrated by the operation and maintenance personnel every month based on the average test results of the finished product. The primary goal of the cyclic control is to keep the temperature of each kiln section stable within the preset range. The control priority is that the kiln temperature meets the standard > the average activity meets the standard. The activity uniformity of finished products with different particle sizes is not included in the control dimension. The total combustion power is adjusted as a whole, and only a fixed temperature deviation dead zone is set. The operation and maintenance iteration relies entirely on manual labor, with about 8 man-hours invested per month to calibrate parameters. The adjustment effect depends on the experience level of the operation and maintenance personnel.

[0035] The only difference is in the control logic. Excluding interference from other variables, the test methods all follow industry-standard practices. The timing error is obtained by comparing the actual residence time with the calculated value using the tracer particle method. The activity error is calculated based on the test value and the estimated value according to the GB / T20855 standard. The activity dispersion is the overall standard deviation of the particle size test results after sieving.

[0036] From the perspective of applicable scenarios, the conventional solution represented by the comparative example has simple logic, low deployment cost, and strong operational stability, and is fully compatible with the production needs of ordinary building material grade lime, which is the mainstream mature technology in the industry. The progressive embodiments of this invention correspond to the production needs of different precision levels. The basic solution can meet the quality improvement needs of conventional active lime, while the high-precision and self-iterative solutions can adapt to the consistency requirements of high-end scenarios. Users can choose the corresponding level of solution according to their own production needs, which has strong implementation flexibility and does not require any new hardware equipment throughout the process. Performance improvement can be achieved only by upgrading the control logic, with low transformation cost and clear benefits.

[0037] Finally, it should be noted that the above embodiments are merely examples for clearly illustrating the present invention and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A real-time temperature control method for a lime rotary kiln, characterized in that, The control method includes the following steps: Acquire multiple sets of operational source data in the lime rotary kiln operation scenario. The operational source data includes multi-point temperature data of the kiln body, combustion condition data, material feed data, and flue gas emission data. Based on the source data, differentiated data acquisition cycle information is generated and multiple batches of samples are taken to construct multiple batches of temperature sample sets. The temperature sample sets are integrated and analyzed to obtain the temperature acquisition quality assessment value. The particle size distribution parameters of the material entering the kiln are extracted from the material feed data. The material is divided into multiple particle size groups according to the particle size range. Combined with the axial movement characteristics of the material corresponding to each particle size group, the time offset correction is performed on the temperature data of multiple points in the kiln body to generate the calcination heating time sequence that is independent for each particle size group. Using the temperature acquisition quality assessment value as the time series data validity correction coefficient, and combining the calcination heating time series of each particle size group, the estimated value of the finished lime activity corresponding to each particle size group is obtained by converting the temperature-activity mapping model corresponding to the group. The activity dispersion statistic of the estimated activity of finished lime in all particle size groups is statistically analyzed and used as the core judgment benchmark for temperature control. The activity dispersion statistic is used to characterize the activity uniformity of finished lime of different particle sizes. If the activity dispersion statistic exceeds the preset qualified threshold, the output power of the segmented combustion device set along the kiln axis corresponding to each calcination stage is adjusted differently according to the activity deviation direction of each particle size group. Otherwise, if the threshold is not exceeded, the current operating state of the segmented combustion device is maintained.

2. The real-time temperature control method for a lime rotary kiln according to claim 1, characterized in that, When acquiring the source data, multiple sets of raw operating data are collected simultaneously by locating multiple data access nodes throughout the entire production process of the lime rotary kiln, and the collected raw operating data is then classified. Specifically, multi-point temperature data of the kiln body is identified from the original operating data by the temperature field distribution characteristics; combustion condition data is extracted by classifying from the set of combustion equipment operating parameters; material feeding data is generated by sorting based on material conveying metering parameters and inlet particle size detection results; and flue gas emission data is obtained by screening and separating based on flue gas component concentration and flue gas flow parameters.

3. The real-time temperature control method for a lime rotary kiln according to claim 1, characterized in that, The specific steps for generating the calcination heating time sequence are as follows: For each particle size group that has been divided, the axial travel rate of the material and the reference value of the total residence time in the kiln are matched for the corresponding particle size range. The axial travel rate of the material is obtained by combining the kiln rotation speed and kiln inclination in the combustion condition data with the average particle size of the corresponding particle size group in the material feed data. The distribution coordinates of each temperature measuring point along the kiln axis are determined according to the multi-point temperature data of the kiln body, and the theoretical travel time difference of a single particle size group of material between adjacent temperature measuring points is calculated. Using the first temperature measuring point at the kiln tail feed end as the time reference, the temperature data of each downstream temperature measuring point are sequentially time-shifted and corrected along the material travel direction, so that the data of all temperature measuring points correspond to the same particle size material segment of the same batch of feed. The corrected temperature measurement data are sorted according to the calcination stage to generate a preliminary calcination heating sequence of a single particle size group from feed to discharge.

4. The real-time temperature control method for a lime rotary kiln according to claim 3, characterized in that, After obtaining the preliminary calcination heating sequence of individual particle size groups, material motion parameters are introduced to perform diffusion-residence time coupling correction. The specific steps are as follows: Based on the average particle size of the corresponding particle size group, the material friction coefficient in the material feed data, and the kiln rotation parameters in the combustion condition data, the axial diffusion coefficient and radial tumbling frequency of the material in this particle size group are calculated. By combining the axial diffusion coefficient and radial tumbling frequency of the material, the probability distribution range of the residence time of the material at each temperature measuring point is derived. According to the residence time probability distribution, the corresponding time nodes of the preliminary calcination heating time sequence are weighted and superimposed for correction, and the corrected single-particle-size cluster calcination heating time sequence is output as the input data of the temperature-activity mapping model.

5. The real-time temperature control method for a lime rotary kiln according to claim 1, characterized in that, The temperature-activity mapping model corresponding to each cluster is constructed by fitting historical production data. The construction process is as follows: Retrieve calcination heating sequence data for each particle size group in historical production batches, actual activity test results of finished lime after sieving for the corresponding particle size group, and flue gas emission data for the corresponding batches. The calcination stage division boundary is corrected by combining the flue gas component concentration in the flue gas emission data. The calcination process of each particle size group is divided into four stages along the kiln axis: preheating zone, decomposition zone, calcination zone, and cooling zone. The four calcination stages correspond one-to-one with the axial setting position of the segmented combustion device. Correlation analysis was used to fit the mapping relationship between the average temperature of each stage and the activity of the finished lime in the corresponding particle size group, and the temperature contribution weight of each stage for each particle size group was obtained. The temperature contribution weights of all granularity groups together constitute the core parameters of the temperature-activity mapping model.

6. The real-time temperature control method for a lime rotary kiln according to claim 5, characterized in that, The temperature-activity mapping model employs a differential update mechanism driven by discrete bias. The update process is as follows: After completing the particle size test of each batch of finished lime, the absolute deviation between the estimated value and the actual value of the activity of finished lime for each particle size group is calculated. Calculate the proportion of the deviation of each particle size group to the total deviation of all particle size groups. For particle size groups whose deviation proportion exceeds the preset deviation threshold, adjust the corresponding temperature contribution weight of each stage according to the deviation proportion. For the remaining granular groups with a deviation percentage lower than the preset deviation threshold, the weights are adjusted slightly in sync with the total deviation magnitude. The updated temperature contribution weight is directly used to calculate the estimated activity of finished lime in subsequent production batches, iteratively reducing the overall level of the activity dispersion statistic.

7. A real-time temperature control system for a lime rotary kiln, characterized in that, The control system includes: The data acquisition and preprocessing unit acquires multiple sets of operational source data for the lime rotary kiln operation scenario. The operational source data includes multi-point temperature data of the kiln body, combustion condition data, material feed data, and flue gas emission data. It generates differentiated data acquisition cycles and completes multiple batches of sampling. After constructing a temperature sample set, it outputs the temperature acquisition quality assessment value. The grouping timing correction unit receives multi-point temperature data, combustion condition data and material feed data of the kiln body, extracts the particle size distribution of the material entering the kiln from the material feed data and divides it into multiple particle size groups. It then performs timing offset correction of the multi-point temperature data of the kiln body in combination with the axial motion characteristics of each particle size group and outputs an independent calcination heating timing sequence for each particle size group. The grouping quality prediction unit has a built-in temperature-activity mapping model corresponding to the group. It receives the temperature acquisition quality assessment value and the calcination heating time sequence, and uses the temperature acquisition quality assessment value as a correction coefficient to calculate the finished lime activity prediction value for each particle size group. The dispersion control unit receives the estimated activity of finished lime from all particle size groups, calculates the activity dispersion statistic as the control benchmark, compares it with the preset qualified threshold, and outputs control commands to differentiate the output power of the segmented combustion devices set along the axial direction of the rotary kiln for each calcination stage.

8. The real-time temperature control system for a lime rotary kiln according to claim 7, characterized in that, The cluster timing correction unit includes: The diffusion correction sub-unit calculates the axial diffusion coefficient and radial tumbling frequency for a single particle size group, derives the residence time probability distribution range, performs weighted superposition correction on the preliminary calcination heating time sequence, and outputs the final calcination heating time sequence.

9. The real-time temperature control system for a lime rotary kiln according to claim 7, characterized in that, The cluster quality prediction unit includes: The weighted fitting subunits, based on the calcination heating time series data, finished product testing data and flue gas emission data of historical production batches, obtain the initial temperature contribution weight of each particle size group, and construct the initial parameters of the temperature-activity mapping model. The sub-units are dynamically updated, and the weight parameters of the high-deviation particle size group are adjusted first, taking into account the deviation ratio of each batch of finished product testing, to iteratively update the temperature-activity mapping model. The activity calculation subunit calls the current effective weights, temperature acquisition quality assessment values, and real-time calcination heating time sequence data to output the estimated activity of finished lime for each particle size group.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when executed by a processor, the computer program implements the steps of the real-time temperature control method for a lime rotary kiln as described in any one of claims 1 to 6.

Images