Intelligent control system and method for preventing coal mill from being blocked based on multi-dimensional internal measuring points

Abstract

The application discloses a coal mill anti-blocking intelligent control system and method based on a multi-dimensional internal measuring point, and the system comprises an internal sensing layer, a data analysis layer and a hierarchical control execution layer; the internal sensing layer comprises a multi-dimensional sensor network arranged at key positions in the coal mill, and the multi-dimensional sensor network is used for collecting multi-source sensing signals in the coal mill; the multi-dimensional sensor network comprises a back powder cone area temperature / pressure / acoustic sensor group, a multi-point temperature-pressure composite sensor group above a grinding bowl and a grinding disc, a wind ring area differential pressure sensor group and a cylinder outer wall acoustic sensor; the data analysis layer comprises a data acquisition and preprocessing module, an online fluctuation characteristic extraction module, a multi-dimensional deviation calculation module, a blocking risk index fusion module and a historical baseline model library; and the hierarchical control execution layer comprises a hierarchical intervention execution module, an intervention effect monitoring and self-recovery module and a man-machine interaction module.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control technology for pulverizing equipment in thermal power plants. Specifically, it is an intelligent control system and method for preventing clogging in coal mills based on multi-dimensional internal measuring points. It involves online sensing of clogging faults in medium-speed coal mills, multi-source data feature analysis, risk quantification and early warning, and graded automatic closed-loop intervention technology. It is applicable to the intelligent control of clogging prevention in various types of direct-fired medium-speed coal mills such as MPS, HP, and ZGM. Background Technology

[0002] Currently, the anti-clogging mechanism of coal mills in power plants relies solely on fixed threshold alarms from external DCS measuring points (inlet and outlet pressure difference, primary air volume, outlet temperature, and coal mill current), and on operators to make manual judgments and adjust the coal feed rate and separator speed. The few existing early warning models also only use external pipeline parameters for modeling, which only achieves fault early warning and lacks automatic control closed loop.

[0003] The aforementioned existing technologies have the following core defects: (1) Low accuracy of judgment and many false alarms and missed reports: Only the decrease in air volume is used as the core criterion for wear blockage. Air volume fluctuations can be caused by irrelevant factors such as damper adjustment and dust accumulation in the pipeline, and it is impossible to distinguish the true signs of wear blockage.

[0004] (2) Delayed fault detection: All sensors are located in the external pipelines of the coal mill, which cannot monitor the internal coal powder accumulation areas such as the return cone, grinding bowl, and air ring. When the external parameters are obviously abnormal, the mill blockage is already serious.

[0005] (3) Only static values ​​are monitored, ignoring dynamic fluctuation characteristics: In the early stage of wear and tear, pressure / temperature fluctuations, temperature-pressure coupling relationship, and acoustic spectrum changes will occur. The existing system only judges whether the absolute value of the parameter exceeds the limit, and cannot capture the early weak anomalies.

[0006] (4) No fully automatic closed-loop self-healing control: After the anomaly is detected, it relies on manual operation, which has a response delay of tens of seconds to several minutes. The intervention range depends on the experience of the personnel and lacks standardized automatic adjustment and recovery logic. Summary of the Invention

[0007] This invention provides an intelligent control system and method for preventing coal mill blockage based on multi-dimensional internal measuring points. It directly senses the internal flow field, temperature field, and sound field of the coal mill. Through multi-dimensional parameter fluctuation characteristic analysis, it identifies early signs of blockage in advance and realizes fully automatic graded intervention and adaptive recovery closed-loop control according to the risk level. This solves the pain points of existing technologies such as inaccurate judgment, delayed early warning, and inability to automatically handle blockage.

[0008] To achieve the above objectives, the present invention provides an intelligent control system for preventing coal mill blockage based on multi-dimensional internal measuring points, which includes an internal sensing layer, a data analysis layer, and a hierarchical control execution layer. The internal sensing layer comprises a multi-dimensional sensor network deployed at key locations within the coal mill. This network collects multi-source sensing signals from within the mill. The multi-dimensional sensor network includes a temperature / pressure / acoustic sensor group in the return cone region, a multi-point temperature-pressure composite sensor group above the grinding bowl / grinding disc, a differential pressure sensor group in the air ring region, and acoustic sensors on the outer wall of the mill shell. The temperature / pressure / acoustic sensor group in the return cone region is deployed at different heights on the inner wall of the dynamic separator's return cone, collecting coal powder temperature, pressure gradient, and acoustic signals at different height sections within the return cone. The multi-point temperature-pressure composite sensor group above the grinding bowl / grinding disc is deployed along the radial direction of the grinding disc near the center, the middle transition zone, and near the outer edge of the grinding zone to collect the temperature field distribution and pressure pulsation signals of the coal powder accumulation area above the grinding bowl; the differential pressure sensor group in the air ring area includes multiple differential pressure sensors, with several differential pressure sensors evenly distributed circumferentially on the upper and lower sections of the air ring to collect the local ventilation resistance signals of each sector of the air ring; the acoustic sensor on the outer wall of the cylinder is attached to the lower middle part of the outer wall of the coal mill cylinder to collect the structural acoustic signals generated by the movement, collision, and accumulation of coal powder inside the mill. The data analysis layer includes a data acquisition and preprocessing module, an online fluctuation feature extraction module, a multi-dimensional deviation calculation module, a mill blockage risk index fusion module, and a historical baseline model library. The data acquisition and preprocessing module receives output signals from the internal sensing layer and data from the DCS system, including coal feed rate, primary air volume, separator speed, and mill current signals. It performs clock calibration, timestamp alignment, anti-aliasing filtering, resampling, range conversion, and zero-point correction. It also removes spike interference through median filtering and amplitude limiting rules, identifies sensor faults and missing data, interpolates short-term missing data, adds invalid labels to long-term missing channels, and records purging, start-up, shutdown, manual intervention, and operating condition switching statuses. It outputs a synchronous data sequence with quality labels. The online feature extraction module uses a sliding time window to perform time-domain, frequency-domain, and temperature-pressure coupling feature calculations on the synchronous data sequence; it extracts the mean, standard deviation, root mean square value, peak-to-peak value, slope of change, pulsation energy, and differential pressure non-uniformity coefficient of the air ring sector from temperature, pressure, and differential pressure signals; it extracts the preset frequency band energy, spectral centroid, spectral kurtosis, and high-low frequency energy ratio from acoustic signals using fast Fourier transform; it calculates the correlation coefficient, phase difference, and maximum cross-correlation time delay for temperature and pressure signals at the same or adjacent locations; it binds the calculated fluctuation features to the operating condition tags corresponding to the current coal feed rate and primary air volume and outputs them to the multi-dimensional deviation calculation module; the multi-dimensional deviation calculation module reads the baseline mean of each feature under the current operating condition from the historical baseline model library. and standard deviation According to the formula Calculate normalized deviation ;in The feature value of the current window. To allow for fluctuation coefficients, To prevent positive numbers with a denominator of zero, a unilateral deviation is calculated for features that only have abnormal significance in one direction, generating a multidimensional deviation vector. Channel weights for sensor failures, measuring point purging, and data quality issues are reset to zero, and the remaining channel weights are renormalized. The clogging risk index fusion module determines the weights based on the contribution of each feature to historical clogging sample identification and sensor reliability. The original clogging risk index R is calculated by combining the multidimensional deviation vector; when multiple adjacent measuring points deviate in the same direction, the temperature-pressure coupling relationship is abnormal, or the differential pressure non-uniformity coefficient continues to increase, a collaborative correction amount is superimposed on the clogging risk index R; two levels of thresholds are set. , The risk of clogging is divided into three levels: R < Normal level ≤R< For attention level, R≥ Hazard level; threshold , The threshold values ​​for exiting the warning level and the danger level are obtained by optimizing the dual thresholds of historical labeled samples. The hysteresis threshold is lower than the entry threshold. The historical baseline model library stores the baseline mean and standard deviation of parameters and fluctuation characteristics of each measuring point under normal operating conditions of the coal mill. The baseline model is divided into operating condition intervals according to the coal feed rate and primary air volume. The tiered control execution layer includes a tiered intervention execution module, an intervention effect monitoring and self-healing recovery module, and a human-machine interaction module. The tiered intervention execution module receives risk level signals from the data analysis layer and outputs control commands to the feeder frequency converter and separator frequency converter. At the caution level, it performs a slight reduction in coal feed and separator speed; at the danger level, it performs a stronger reduction in coal feed and a decrease in separator speed. The intervention effect monitoring and self-healing recovery module continuously receives the mill blockage risk index and measurement point data from the data analysis layer during intervention execution. Based on the downward trend of the mill blockage risk index and whether the measurement point parameters return to the historical normal operating curve for the corresponding working condition, it decides whether to maintain, increase, or cancel the intervention. Preventive measures; the human-computer interaction module uses green-yellow-red three colors to display the risk level of grinding blockage, and displays the real-time temperature and pressure field distribution heat map and acoustic spectrum of each internal measuring point, the time-domain / frequency-domain analysis results of the fluctuation characteristics of each parameter, the grinding blockage risk index and its changing trend curve fused by multiple parameters, the execution status of the system's automatic graded intervention, the intervention range and parameter recovery trend; warning prompts and intervention suggestions pop up at the attention level, and audible and visual alarms are output at the danger level; manual-automatic mode switching is supported; the system stores the original data of the measuring points, fluctuation characteristics, historical risk index, and intervention event logs into the historical database, and supports historical playback and trend analysis according to time, grinding number, and risk level.

[0009] In one embodiment of the present invention, preferably, in the multidimensional sensor network, each internal measuring point adopts an isolated protective structure in which the wear-resistant sampling end or sensing end is placed inside the mill and the sensitive element and signal conditioning unit are placed outside; the pressure and differential pressure measuring points are equipped with independent low-flow pulse backflushing or continuous micro-purge interfaces, and the purging medium is cleanroom air; the acoustic measuring points use wear-resistant metal acoustic guide rods or high-temperature acoustic waveguides to transmit the acoustic vibration signal inside the mill to the acoustic sensor outside the housing.

[0010] In one embodiment of the present invention, preferably, the threshold , The specific method for dual-threshold optimization is as follows: Samples of stable normal operation windows, early mill blockage windows, and dangerous mill blockage windows were extracted from historical databases, and the sample labels were verified in conjunction with records of coal feed interruption, abnormal mill pressure difference, manual coal reduction, and mill shutdown. The samples are divided into training set, validation set, and test set; in the range 0 < < Within the range of ≤100, use grid search, progressive interval narrowing, or Bayesian optimization to traverse the threshold combination; by The objective function is defined as follows: FAR is the false alarm rate under normal operating conditions, and MAR is the missed alarm rate for clogging samples. The average delay from the confirmed signs of wear and tear to the triggering of the corresponding warning level. To determine the number of frequent back-and-forth switching of risk levels, α, β, γ, and δ are weighted to represent safety priorities. Under the constraints that the false negative rate and maximum warning delay do not exceed preset upper limits, the threshold combination that minimizes the objective function J of the validation set and ensures stable performance on the test set is selected as the [missing information]. , ; The threshold will only be re-optimized and updated after the number of newly confirmed event samples reaches the preset number and offline verification is completed.

[0011] In one embodiment of the present invention, preferably, when the purge valve is activated, the system adds a status indicator to the corresponding measuring point channel data, and the channel data collected at this time is not used to calculate the clogging risk index.

[0012] In one embodiment of the present invention, preferably, the initial construction method of the historical baseline model library is as follows: during the initial stage of unit commissioning, continuous collection of normal operation data for 7-14 days is used to complete the initial construction of the baseline library.

[0013] This invention also provides an intelligent control method for preventing coal mill blockage based on multi-dimensional internal measuring points. This method is implemented using the above-mentioned system and includes the following steps: S1: Multi-dimensional internal data synchronous acquisition and preprocessing; acquires sensing signals from the return cone, the top of the grinding bowl, the air ring, and the outer wall of the cylinder; synchronously acquires DCS coal feed rate, primary air volume, separator speed, and coal mill current parameters; completes data synchronization alignment, filtering, fault identification, and data quality marking; and outputs a synchronous data sequence with quality marking. S2: Operating condition matching and baseline loading; Determine the current operating condition range based on real-time coal feed and primary air volume, and load the normal operating baseline of the fluctuation characteristics of each measuring point under this operating condition from the historical baseline model library; S3: Online extraction of sliding window fluctuation features; using a preset time window and sliding step size, time-domain features, frequency-domain features, and temperature-pressure coupling features are calculated on the synchronous data sequence. S4: Multidimensional deviation vector calculation; compare each current feature with the corresponding working condition baseline, calculate the normalized deviation, and generate a multidimensional deviation vector after removing invalid channel weights; S5: Clogging Risk Index Integration Calculation and Risk Level Classification; Calculate the clogging risk index R and classify it into three levels: normal, caution, and danger; If it is determined to be at the normal level, return to step S1 for continuous monitoring; If it is determined to be at the caution or danger level, proceed to step S6. S6: Graded closed-loop intervention execution; when the risk level is the attention level, the coal feed rate is reduced by the first preset amount at the first preset rate, and the separator speed is reduced by the second preset amount at the second preset rate; when the risk level is the danger level, the coal feed rate is reduced by the third preset amount at the third preset rate, and the separator speed is reduced to the preset safe speed. S7: Intervention effect monitoring and adaptive adjustment; after intervention, steps S1-S5 are executed cyclically; if the mill blockage risk index R continues to decrease and falls back to the normal range within the preset stability observation time, and all internal measuring point parameters return to the corresponding historical normal operation curve, then the intervention is judged to be successful, and the coal feed rate and separator speed are gradually restored to the normal set value at the preset rate, and then step S8 is entered; if R does not fall back or even continues to rise within the stability observation time, the intervention intensity is automatically upgraded and step S6 is executed again; if the risk still cannot be alleviated after reaching the preset maximum number of interventions, or if the coal feed rate and separator speed have reached the safety lower limit but the risk still exists, then a manual mill stop inspection prompt is output; S8: Baseline model adaptive update; Baseline update trigger condition: Clogging risk index R continuously falls below [a certain value]. The preset stabilization time is reached; the coal feed rate, primary air volume, and separator speed remain unchanged within the operating range, and the parameter change rate does not exceed the threshold; all sensors involved in the update are in normal health status and the data validity rate meets the standard; there are no events such as mill start-up / shutdown, coal type switching, measuring point purging, maintenance, manual intervention, or automatic grading intervention during the update window; after the cumulative effective sliding window in the same operating range reaches the preset number, abnormal sample windows are removed, the mean and variance of candidate samples are calculated, and the baseline is dynamically adjusted according to the sample quantity and data quality; smooth interpolation is performed at the boundary between different operating ranges to complete the incremental update of the historical baseline model library, and then the process returns to step S1.

[0014] In one embodiment of the present invention, preferably, in step S3, the preset time window is 60 seconds and the sliding step is 10 seconds.

[0015] In one embodiment of the present invention, preferably, in step S7, the stability observation time is 5 minutes; the preset maximum number of interventions is 3.

[0016] In one embodiment of the present invention, preferably, in step S8, the preset stabilization time is not less than 30 minutes, and the baseline candidate update is performed daily or weekly; abnormal window samples are removed by using the median absolute deviation or interquartile range method; at the boundary of the working condition interval, the baseline is smoothed by linear interpolation or distance weighting to suppress the abrupt deviation caused by the switching of working conditions.

[0017] In one embodiment of the present invention, preferably, in step S6, attention is paid to the level intervention: the coal feed rate is reduced by 5%-8%, and the adjustment rate is 2t / h. For min, reduce the separator speed by 30-50 rpm; for hazard intervention: reduce the coal feed rate by 10%-15%, with an adjustment rate of 3 t / h. The separator speed is reduced by 80-100 rpm compared to the rated speed.

[0018] The present invention has the following beneficial technical effects: This invention sets up internal multi-dimensional measuring points in areas prone to coal dust accumulation, such as the return powder cone, the top of the grinding disc, and the air ring, and combines them with acoustic signals from the outer wall of the cylinder to directly sense coal dust accumulation, increased local ventilation resistance, and abnormalities in the return powder channel. It can identify early signs of mill blockage before external parameters such as inlet and outlet pressure difference and primary air volume significantly exceed the limits. This invention analyzes the time domain, frequency domain, and temperature-pressure coupling characteristics of temperature, pressure, differential pressure, and acoustic signals simultaneously, and matches the normal operating condition baseline according to the coal feed rate and primary air volume. This can eliminate single-parameter disturbances caused by non-clogging factors such as damper operation and load changes, thereby reducing false alarms and missed alarms. This invention drives the closed-loop regulation of coal feed rate and separator speed through a graded risk index of mill blockage, and automatically maintains, strengthens or cancels intervention based on the downward trend of risk, which can shorten the manual response time and avoid the impact of a one-time large adjustment on the stability of the pulverizing system. This invention continuously adapts to changes in coal quality, load, and equipment status through historical baseline adaptive updates and event log playback, improving the consistency of early warnings under different mill types and different operating stages, thereby reducing the risks of severe mill blockage, unplanned mill shutdowns, and related safety risks. Attached Figure Description

[0019] The above and other objects, features, and advantages of the present invention will become more apparent from the more detailed description of the embodiments of the invention in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same parts or steps.

[0020] Figure 1 This is an overall architecture diagram of an intelligent control system for preventing coal mill blockage based on multi-dimensional internal measuring points, according to an embodiment of the present invention. Figure 2 This is an overall flowchart of an embodiment of the intelligent control method for preventing coal mill blockage based on multi-dimensional internal measuring points according to the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] Unless otherwise specifically stated, the technical or scientific terms used in the embodiments of this invention should be understood in their ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains. The terms "comprising" or "including," as used in the embodiments of this invention, do not limit the shapes, numbers, steps, actions, operations, components, elements, and / or groups thereof mentioned, nor do they exclude the appearance or addition of one or more other different shapes, numbers, steps, actions, operations, components, elements, and / or groups thereof, or the inclusion of these.

[0023] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale, and techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail; however, where appropriate, the illustrated techniques, methods, and apparatus should be considered part of the specification. In all the examples shown and discussed herein, any other specific example may have different values. It should be noted that similar symbols and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0024] In the description of the embodiments of the present invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In the embodiments of the present invention, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in the embodiments of the present invention, as well as the features of different embodiments or examples.

[0025] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0026] This invention is applicable to mainstream direct-fired medium-speed coal mills such as MPS, HP, and ZGM series. It can be adapted to coal mills in newly built thermal power units and can also be used for coal mill retrofit projects in existing units. During the on-site retrofit implementation phase, it is necessary to check the location of the measuring point openings based on the structural drawings of the target coal mill, avoiding the internal grinding rollers, guide baffles, and metal components of the return powder channel to prevent damage to the sensor sampling end from impacts by moving parts. The opening work should be completed during the unit's shutdown and maintenance window, strictly adhering to the boiler pulverizing system safety maintenance specifications. After hot work and opening, a sealing pressure test should be performed on the wall penetration points of the coal mill shell to ensure the negative pressure sealing of the coal mill shell and avoid air leakage problems that could interfere with the operating parameters of the pulverizing system.

[0027] In the present invention, the system intuitively displays the risk level with three colors of "green-yellow-red": green indicates normal operation (R<R1), yellow indicates attention level (R1≤R<R2), and red indicates danger level (R≥R2). After entering the yellow level, the system automatically pops up an early warning prompt and suggested intervention strategies; after entering the red level, the system automatically performs enhanced intervention while issuing an audible and visual alarm to the operating personnel. The operating personnel can switch to the manual control mode on the system at any time to customize the intervention strategy.

[0028] The system simultaneously stores all internal measurement point data, fluctuation characteristic data, risk index historical records and intervention event logs into the historical database, supports historical playback and trend analysis by dimensions such as time, mill number, risk level, etc., and provides data support for operation optimization and maintenance decision-making.

[0029] As Figure 1 shown, the present invention provides an intelligent control system for preventing mill blockage of a coal mill based on multi-dimensional internal measurement points, which includes an internal perception layer, a data analysis layer and a hierarchical control execution layer; The internal perception layer includes a multi-dimensional sensor network arranged at key positions inside the coal mill. The multi-dimensional sensor network is used to collect multi-source perception signals inside the coal mill, and the multi-dimensional sensor network includes a temperature / pressure / acoustic sensor group in the powder return cone area, a multi-point temperature-pressure composite sensor group above the grinding bowl / grinding disc, a differential pressure sensor group in the air ring area, and an acoustic sensor on the outer wall of the cylinder; the temperature / pressure / acoustic sensor group in the powder return cone area is arranged at different height positions on the inner wall of the powder return cone of the dynamic separator, collects the pulverized coal temperature, pressure gradient and acoustic signals of different height sections in the powder return cone, and is used for real-time monitoring of the change trend of the circulation ratio and the precursor of blockage of the powder return channel; the multi-point temperature-pressure composite sensor group above the grinding bowl / grinding disc is arranged along the radial direction of the grinding disc in the area close to the center, the middle transition area, and the area close to the outer edge of the grinding area respectively, to collect the temperature field distribution and pressure pulsation signals of the pulverized coal accumulation area above the grinding bowl; the differential pressure sensor group in the air ring area includes a plurality of differential pressure sensors, a number of differential pressure sensors are evenly distributed along the circumferential direction on the upper section and the lower section of the air ring respectively, to collect the local ventilation resistance signals of each sector of the air ring, so as to detect the ventilation imbalance caused by local blockage of the air ring; the acoustic sensor on the outer wall of the cylinder is attached to the middle and lower part of the outer wall of the coal mill cylinder, collects the structural acoustic signals generated by the movement, collision and accumulation of pulverized coal in the mill, and reflects the change of material concentration in the mill; The data analysis layer includes a data acquisition and preprocessing module, an online fluctuation feature extraction module, a multi-dimensional deviation calculation module, a mill blockage risk index fusion module, and a historical baseline model library. The data acquisition and preprocessing module receives output signals from the internal sensing layer and data from the DCS system, including coal feed rate, primary air volume, separator speed, and mill current signals. It performs clock calibration, timestamp alignment, anti-aliasing filtering, resampling, range conversion, and zero-point correction. It removes spike interference through median filtering and amplitude limiting rules (using continuity checks, rate of change checks, and sensor health status checks) to identify sensor faults and missing data. It interpolates short-term missing data and adds invalid labels to long-term missing channels. Simultaneously, it records purging, start / stop, manual intervention, and operating condition switching statuses, and outputs data with quality labels. The system identifies the synchronous data sequence; the online fluctuation feature extraction module uses a sliding time window to calculate the time domain, frequency domain, and temperature-pressure coupling features of the synchronous data sequence; it extracts the mean, standard deviation, root mean square value, peak-to-peak value, slope of change, pulsation energy, and differential pressure non-uniformity coefficient of the wind ring sector from temperature, pressure, and differential pressure signals; it extracts the preset frequency band energy, spectral centroid, spectral kurtosis, and high-low frequency energy ratio from acoustic signals using fast Fourier transform; it calculates the correlation coefficient, phase difference, and maximum cross-correlation time delay for temperature and pressure signals at the same or adjacent locations; it binds the calculated fluctuation features to the operating condition tags corresponding to the current coal feed rate and primary air volume and outputs them to the multi-dimensional deviation calculation module; the multi-dimensional deviation calculation module reads the baseline mean of each feature under the current operating condition from the historical baseline model library. and standard deviation According to the formula Calculate normalized deviation ;in The feature value of the current window. To allow for fluctuation coefficients, To prevent positive numbers with a denominator of zero, a unilateral deviation is calculated for features that only have abnormal significance in one direction, generating a multidimensional deviation vector. Channel weights for sensor failures, measuring point purging, and data quality issues are reset to zero, and the remaining channel weights are renormalized. The clogging risk index fusion module determines the weights based on the contribution of each feature to historical clogging sample identification and sensor reliability. The original clogging risk index R is calculated by combining a multidimensional deviation vector. When multiple adjacent measuring points deviate in the same direction, the temperature-pressure coupling relationship is abnormal, or the differential pressure non-uniformity coefficient continues to increase, a collaborative correction amount is superimposed on the clogging risk index R. This ensures that short-term fluctuations at a single measuring point do not directly trigger high-level alarms, while continuous anomalies in multiple areas and multiple physical quantities can quickly increase the risk level. Two threshold levels are set. , The risk of clogging is divided into three levels: R < Normal level (green) ≤R< Note the level (yellow), R≥ Danger level (red); threshold , The threshold values ​​for exiting the warning level and the danger level are obtained by optimizing the dual thresholds of historical labeled samples. The hysteresis threshold is lower than the entry threshold. The historical baseline model library stores the baseline mean and standard deviation of parameters and fluctuation characteristics of each measuring point under normal operating conditions of the coal mill. The baseline model is divided into operating condition intervals according to the coal feed rate and primary air volume (e.g., one interval is divided according to the coal feed rate of 5t / h and one sub-interval is divided according to the air volume of 10t / h). Among them, weight The weights of features are used to characterize the importance of features in determining clogging and wear faults. The weights are determined based on the feature's contribution to clogging and wear identification and the sensor's reliability. Features with stronger early warning capabilities in historical clogging and wear cases are considered more valuable. The larger the value, the higher the reliability of the sensor measurement point. The larger the value, the more likely the corresponding channel will fail in the event of sensor malfunction, purging action, or invalid data. =0 (weights are reset to zero, this feature is no longer included in risk calculation); all effective feature weights satisfy the normalization constraint ∑ =1. Additionally... .

[0030] The tiered control execution layer includes a tiered intervention execution module, an intervention effect monitoring and self-healing recovery module, and a human-machine interaction module. The tiered intervention execution module receives risk level signals from the data analysis layer and outputs control commands to the feeder frequency converter and separator frequency converter. At the caution level, it performs a slight reduction in coal feed and separator speed; at the danger level, it performs a stronger reduction in coal feed and a decrease in separator speed. The intervention effect monitoring and self-healing recovery module continuously receives the mill blockage risk index and measurement point data from the data analysis layer during intervention execution. Based on the downward trend of the mill blockage risk index and whether the measurement point parameters return to the historical normal operating curve for the corresponding working condition, it decides whether to maintain, increase, or cancel the intervention. Preventive measures; the human-computer interaction module uses green-yellow-red three colors to display the risk level of grinding blockage, and displays the real-time temperature and pressure field distribution heat map and acoustic spectrum of each internal measuring point, the time-domain / frequency-domain analysis results of the fluctuation characteristics of each parameter, the grinding blockage risk index and its changing trend curve fused by multiple parameters, the execution status of the system's automatic graded intervention, the intervention range and parameter recovery trend; warning prompts and intervention suggestions pop up at the attention level, and audible and visual alarms are output at the danger level; manual-automatic mode switching is supported; the system stores the original data of the measuring points, fluctuation characteristics, historical risk index, and intervention event logs into the historical database, and supports historical playback and trend analysis according to time, grinding number, and risk level.

[0031] In this embodiment, to meet the requirements of operating in the high-temperature, high-dust, high-erosion, and high-vibration environment inside the coal mill, each internal measuring point in the multi-dimensional sensor network adopts an isolated protective structure with wear-resistant sampling ends or sensing ends placed inside the mill and sensitive elements and signal conditioning units placed outside. Pressure and differential pressure measuring points are equipped with independent low-flow pulse backflushing or continuous micro-purge interfaces. The purging medium is cleanroom air, and the purging pressure is higher than the maximum pressure of the measuring point, with the purging flow rate limited to a range that does not affect the mainstream field, to prevent coal dust from entering the pressure tapping channel. Acoustic measuring points use wear-resistant metal acoustic guide rods or high-temperature acoustic waveguides to transmit acoustic vibration signals from inside the mill to acoustic sensors outside the shell. Acoustic sensors on the outer wall of the cylinder are fixed with studs or magnetic bases and equipped with heat insulation pads, shielding covers, and anti-loosening structures. All signal lines use high-temperature resistant insulated cables or metal braided shielded armored cables, led out through sealed wall-penetrating joints, and laid away from grinding components, hot surfaces, and high-voltage cables. In addition, the temperature measuring points preferably use insulated armored thermocouples, and the temperature measuring end and the protective sleeve are kept in thermally conductive contact; the pressure and differential pressure measuring points preferably use flush-type wear-resistant pressure taps or short pressure taps with ceramic bushings, and the pressure is transmitted to the pressure sensor outside the coal mill through high-temperature pressure taps.

[0032] In one embodiment of the present invention, preferably, the threshold , The specific method for dual-threshold optimization is as follows: Samples of stable normal operation windows, early mill blockage windows, and dangerous mill blockage windows were extracted from historical databases, and the sample labels were verified in conjunction with records of coal feed interruption, abnormal mill pressure difference, manual coal reduction, and mill shutdown. (Based on time period or coal mill number) Divide the samples into training set, validation set, and test set; in the range 0 < < Within the range of ≤100, use grid search, progressive interval narrowing, or Bayesian optimization to traverse the threshold combination; by The objective function is defined as follows: FAR is the false alarm rate under normal operating conditions, and MAR is the missed alarm rate for clogging samples. The average delay from the confirmed signs of wear and tear to the triggering of the corresponding warning level. To determine the number of frequent back-and-forth switching of risk levels, α, β, γ, and δ are weighted to represent safety priorities. Under the constraints that the false negative rate and maximum warning delay do not exceed preset upper limits, the threshold combination that minimizes the objective function J of the validation set and ensures stable performance on the test set is selected as the [missing information]. , ; To prevent oscillations near the threshold, a hysteresis threshold lower than the entry threshold is used when exiting the attention or danger level. The threshold is only re-optimized and updated after the number of newly confirmed event samples reaches the preset number and offline verification is completed. The threshold is not automatically changed directly using unlabeled online data.

[0033] In one embodiment of the present invention, preferably, when the purge valve is activated, the system adds a status indicator to the corresponding measuring point channel data, and the channel data collected at this time is not used to calculate the clogging risk index.

[0034] In one embodiment of the present invention, preferably, the initial construction method of the historical baseline model library is as follows: during the initial stage of unit commissioning, 7-14 days of continuous normal operation data are collected to complete the initial construction of the baseline library, and subsequent periodic incremental updates are performed.

[0035] like Figure 2 As shown, the present invention also provides an intelligent control method for preventing coal mill blockage based on multi-dimensional internal measuring points. This method is implemented using the above-mentioned system and includes the following steps: S1: Multi-dimensional internal data synchronous acquisition and preprocessing; acquires sensing signals from the return cone, the top of the grinding bowl, the air ring, and the outer wall of the cylinder; synchronously acquires DCS coal feed rate, primary air volume, separator speed, and coal mill current parameters; completes data synchronization alignment, filtering, fault identification, and data quality marking; and outputs a synchronous data sequence with quality marking. S2: Operating condition matching and baseline loading; Determine the current operating condition range based on real-time coal feed and primary air volume, and load the normal operating baseline of the fluctuation characteristics of each measuring point under this operating condition from the historical baseline model library; S3: Online extraction of sliding window fluctuation features; using a preset time window and sliding step size, time-domain features, frequency-domain features, and temperature-pressure coupling features are calculated on the synchronous data sequence. S4: Multidimensional deviation vector calculation; compare each current feature with the corresponding working condition baseline, calculate the normalized deviation, and generate a multidimensional deviation vector after removing invalid channel weights; S5: Clogging Risk Index Integration Calculation and Risk Level Classification; Calculate the clogging risk index R and classify it into three levels: normal, caution, and danger; If it is determined to be at the normal level, return to step S1 for continuous monitoring; If it is determined to be at the caution or danger level, proceed to step S6. S6: Graded closed-loop intervention execution; when the risk level is the attention level, the coal feed rate is reduced by the first preset amount at the first preset rate, and the separator speed is reduced by the second preset amount at the second preset rate; when the risk level is the danger level, the coal feed rate is reduced by the third preset amount at the third preset rate, and the separator speed is reduced to the preset safe speed. S7: Intervention effect monitoring and adaptive adjustment; after intervention, steps S1-S5 are executed cyclically; if the mill blockage risk index R continues to decrease and falls back to the normal range within the preset stability observation time, and all internal measuring point parameters return to the corresponding historical normal operation curve, then the intervention is judged to be successful, and the coal feed rate and separator speed are gradually restored to the normal set value at the preset rate, and then step S8 is entered; if R does not fall back or even continues to rise within the stability observation time, the intervention intensity is automatically upgraded and step S6 is executed again; if the risk still cannot be alleviated after reaching the preset maximum number of interventions, or if the coal feed rate and separator speed have reached the safety lower limit but the risk still exists, then a manual mill stop inspection prompt is output; S8: Baseline model adaptive update; Baseline update trigger condition: Clogging risk index R continuously falls below [a certain value]. The preset stabilization time is reached; the coal feed rate, primary air volume, and separator speed remain unchanged within the operating range, and the parameter change rate does not exceed the threshold; all sensors involved in the update are in normal health status and the data validity rate meets the standard; there are no events such as mill start-up / shutdown, coal type switching, measuring point purging, maintenance, manual intervention, or automatic grading intervention during the update window; after the cumulative effective sliding window in the same operating range reaches the preset number, abnormal sample windows are removed, the mean and variance of candidate samples are calculated, and the baseline is dynamically adjusted according to the sample quantity and data quality; smooth interpolation is performed at the boundary between different operating ranges to complete the incremental update of the historical baseline model library, and then the process returns to step S1.

[0036] In one embodiment of the present invention, preferably, in step S3, the preset time window is 60 seconds and the sliding step is 10 seconds.

[0037] In one embodiment of the present invention, preferably, in step S7, the stability observation time is 5 minutes; the preset maximum number of interventions is 3.

[0038] In one embodiment of the present invention, preferably, in step S8, the preset stabilization time is not less than 30 minutes, and the baseline candidate update is performed daily or weekly; abnormal window samples are removed by using the median absolute deviation or interquartile range method; at the boundary of the working condition interval, the baseline is smoothed by linear interpolation or distance weighting to suppress the abrupt deviation caused by the switching of working conditions.

[0039] In one embodiment of the present invention, preferably, in step S6, attention is paid to the level intervention: the coal feed rate is reduced by 5%-8%, and the adjustment rate is 2t / h. For min, reduce the separator speed by 30-50 rpm; for hazard intervention: reduce the coal feed rate by 10%-15%, with an adjustment rate of 3 t / h. The separator speed is reduced by 80-100 rpm compared to the rated speed.

[0040] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A coal mill anti-clogging intelligent control system based on multi-dimensional internal measuring points, characterized in that, It includes an internal perception layer, a data analysis layer, and a hierarchical control execution layer; The internal sensing layer comprises a multi-dimensional sensor network deployed at key locations within the coal mill. This network collects multi-source sensing signals from within the mill. The multi-dimensional sensor network includes a temperature / pressure / acoustic sensor group in the return cone region, a multi-point temperature-pressure composite sensor group above the grinding bowl / grinding disc, a differential pressure sensor group in the air ring region, and acoustic sensors on the outer wall of the mill shell. The temperature / pressure / acoustic sensor group in the return cone region is deployed at different heights on the inner wall of the dynamic separator's return cone, collecting coal powder temperature, pressure gradient, and acoustic signals at different height sections within the return cone. The multi-point temperature-pressure composite sensor group above the grinding bowl / grinding disc is deployed along the radial direction of the grinding disc near the center, the middle transition zone, and near the outer edge of the grinding zone to collect the temperature field distribution and pressure pulsation signals of the coal powder accumulation area above the grinding bowl; the differential pressure sensor group in the air ring area includes multiple differential pressure sensors, with several differential pressure sensors evenly distributed circumferentially on the upper and lower sections of the air ring to collect the local ventilation resistance signals of each sector of the air ring; the acoustic sensor on the outer wall of the cylinder is attached to the lower middle part of the outer wall of the coal mill cylinder to collect the structural acoustic signals generated by the movement, collision, and accumulation of coal powder inside the mill. The data analysis layer includes a data acquisition and preprocessing module, an online fluctuation feature extraction module, a multidimensional deviation calculation module, a clogging risk index fusion module, and a historical baseline model library; The data acquisition and preprocessing module receives output signals from the internal sensing layer and data from the DCS system, including coal feed rate, primary air volume, separator speed, and mill current. It performs clock calibration, timestamp alignment, anti-aliasing filtering, resampling, range conversion, and zero-point correction. It removes spike interference through median filtering and amplitude limiting rules, identifies sensor faults and missing data, interpolates short-term missing data, and adds invalid labels to long-term missing channels. Simultaneously, it records purging, start / stop, manual intervention, and operating condition switching statuses, outputting a synchronized data sequence with quality indicators. The online fluctuation feature extraction module uses a sliding time window to calculate the time-domain, frequency-domain, and temperature-pressure coupling features of the synchronized data sequence. It extracts the mean, standard deviation, root mean square value, peak-to-peak value, slope of change, pulsating energy, and differential pressure non-uniformity coefficient of the air ring sector from temperature, pressure, and differential pressure signals. For acoustic signals, it extracts preset frequency band energy, spectral centroid, spectral kurtosis, and high-low frequency energy ratio through fast Fourier transform. It calculates the correlation coefficient, phase difference, and maximum cross-correlation time delay for temperature and pressure signals at the same or adjacent locations. The calculated fluctuation characteristics are bound to the operating condition labels corresponding to the current coal feed and primary air volume and output to the multi-dimensional deviation calculation module. The multidimensional deviation calculation module reads the baseline mean of each feature under the current operating condition from the historical baseline model library. and standard deviation According to the formula Calculate normalized deviation ;in The feature value of the current window. To allow for fluctuation coefficients, To prevent positive numbers with a denominator of zero, a unilateral deviation is calculated for features that only have abnormal significance in one direction, generating a multidimensional deviation vector. Channel weights for sensor failures, measuring point purging, and data quality issues are reset to zero, and the remaining channel weights are renormalized. The clogging risk index fusion module determines the weights based on the contribution of each feature to historical clogging sample identification and sensor reliability. The original clogging risk index R is calculated by combining the multidimensional deviation vector; when multiple adjacent measuring points deviate in the same direction, the temperature-pressure coupling relationship is abnormal, or the differential pressure non-uniformity coefficient continues to increase, a collaborative correction amount is superimposed on the clogging risk index R; two levels of thresholds are set. , The risk of clogging is divided into three levels: R < Normal level ≤R< For attention level, R≥ Hazard level; threshold , The threshold values ​​for exiting the warning level and the danger level are obtained by optimizing the dual thresholds of historical labeled samples. The hysteresis threshold is lower than the entry threshold. The historical baseline model library stores the baseline mean and standard deviation of parameters and fluctuation characteristics of each measuring point under normal operating conditions of the coal mill. The baseline model is divided into operating condition intervals according to the coal feed rate and primary air volume. The tiered control execution layer includes a tiered intervention execution module, an intervention effect monitoring and self-healing recovery module, and a human-machine interaction module. The tiered intervention execution module receives risk level signals from the data analysis layer and outputs control commands to the feeder frequency converter and separator frequency converter. At the caution level, it performs a slight reduction in coal feed and separator speed; at the danger level, it performs a stronger reduction in coal feed and a decrease in separator speed. The intervention effect monitoring and self-healing recovery module continuously receives the mill blockage risk index and measurement point data from the data analysis layer during intervention execution. Based on the downward trend of the mill blockage risk index and whether the measurement point parameters return to the historical normal operating curve for the corresponding working condition, it decides whether to maintain, increase, or cancel the intervention. Preventive measures; the human-computer interaction module uses green-yellow-red three colors to display the risk level of grinding blockage, and displays the real-time temperature and pressure field distribution heat map and acoustic spectrum of each internal measuring point, the time-domain / frequency-domain analysis results of the fluctuation characteristics of each parameter, the grinding blockage risk index and its changing trend curve fused by multiple parameters, the execution status of the system's automatic graded intervention, the intervention range and parameter recovery trend; warning prompts and intervention suggestions pop up at the attention level, and audible and visual alarms are output at the danger level; manual-automatic mode switching is supported; the system stores the original data of the measuring points, fluctuation characteristics, historical risk index, and intervention event logs into the historical database, and supports historical playback and trend analysis according to time, grinding number, and risk level.

2. The intelligent control system for preventing coal mill blockage based on multi-dimensional internal measuring points according to claim 1, characterized in that, In the multidimensional sensor network, each internal measuring point adopts an isolated protective structure with wear-resistant sampling ends or sensing ends placed inside the mill and sensitive elements and signal conditioning units placed outside; pressure and differential pressure measuring points are equipped with independent low-flow pulse backflushing or continuous micro-purge interfaces, and the purging medium is cleanroom air; acoustic measuring points use wear-resistant metal acoustic guide rods or high-temperature acoustic waveguides to transmit acoustic vibration signals inside the mill to acoustic sensors outside the housing.

3. The intelligent control system for preventing coal mill blockage based on multi-dimensional internal measuring points according to claim 1, characterized in that, threshold , The specific method for dual-threshold optimization is as follows: Samples of stable normal operation windows, early mill blockage windows, and dangerous mill blockage windows were extracted from historical databases, and the sample labels were verified in conjunction with records of coal feed interruption, abnormal mill pressure difference, manual coal reduction, and mill shutdown. The samples are divided into training set, validation set, and test set; in the range 0 < < Within the range of ≤100, use grid search, progressive interval narrowing, or Bayesian optimization to traverse the threshold combination; by The objective function is defined as follows: FAR is the false alarm rate under normal operating conditions, and MAR is the missed alarm rate for clogging samples. The average delay from the confirmed signs of wear and tear to the triggering of the corresponding warning level. To determine the number of frequent back-and-forth switching of risk levels, α, β, γ, and δ are weighted to represent safety priorities. Under the constraints that the false negative rate and maximum warning delay do not exceed preset upper limits, the threshold combination that minimizes the objective function J of the validation set and ensures stable performance on the test set is selected as the [missing information]. , ; The threshold will only be re-optimized and updated after the number of newly confirmed event samples reaches the preset number and offline verification is completed.

4. The intelligent control system for preventing coal mill blockage based on multi-dimensional internal measuring points as described in claim 1 is characterized in that, When the purge valve is activated, the system adds a status indicator to the corresponding measuring point channel data. At this time, the collected channel data is not included in the calculation of the clogging risk index.

5. The intelligent control system for preventing coal mill blockage based on multi-dimensional internal measuring points according to claim 1, characterized in that, The initial construction method for the historical baseline model library is as follows: during the initial stage of unit commissioning, 7-14 days of continuous normal operation data are collected to complete the initial baseline library construction.

6. A smart control method for preventing coal mill blockage based on multi-dimensional internal measuring points, characterized in that, This method is performed using the system described in any one of claims 1-5, and includes the following steps: S1: Multi-dimensional internal data synchronous acquisition and preprocessing; acquires sensing signals from the return cone, the top of the grinding bowl, the air ring, and the outer wall of the cylinder; synchronously acquires DCS coal feed rate, primary air volume, separator speed, and coal mill current parameters; completes data synchronization alignment, filtering, fault identification, and data quality marking; and outputs a synchronous data sequence with quality marking. S2: Operating condition matching and baseline loading; Determine the current operating condition range based on real-time coal feed and primary air volume, and load the normal operating baseline of the fluctuation characteristics of each measuring point under this operating condition from the historical baseline model library; S3: Online extraction of sliding window fluctuation features; Using a preset time window and sliding step size, the time domain characteristics, frequency domain characteristics, and temperature-pressure coupling characteristics of the synchronous data sequence are calculated. S4: Multidimensional deviation vector calculation; compare each current feature with the corresponding working condition baseline, calculate the normalized deviation, and generate a multidimensional deviation vector after removing invalid channel weights; S5: Integrated calculation and risk level classification of the clogging risk index; Calculate the clogging risk index R and classify it into three levels of risk: normal, caution, and danger; if it is determined to be at the normal level, return to step S1 for continuous monitoring; If the level is determined to be either caution or danger, proceed to step S6; S6: Graded closed-loop intervention execution; when the risk level is the attention level, the coal feed rate is reduced by the first preset amount at the first preset rate, and the separator speed is reduced by the second preset amount at the second preset rate; when the risk level is the danger level, the coal feed rate is reduced by the third preset amount at the third preset rate, and the separator speed is reduced to the preset safe speed. S7: Monitoring and adaptive adjustment of intervention effects; After intervention, steps S1-S5 are executed repeatedly. If the mill blockage risk index R continues to decrease and returns to the normal range within the preset stability observation period, and all internal measuring point parameters return to the historical normal operating curve of the corresponding working condition, the intervention is deemed successful. The coal feed rate and separator speed are gradually restored to the normal set value at the preset rate, and then step S8 is entered. If R does not decrease or even continues to rise within the stability observation period, the intervention intensity is automatically upgraded and step S6 is executed again. If the risk cannot be alleviated after reaching the preset maximum number of interventions, or if the coal feed rate and separator speed have reached the safety lower limit but the risk still exists, a manual mill shutdown and inspection prompt is output. S8: Baseline model adaptive update; Baseline update trigger condition: Clogging risk index R continuously falls below [a certain value]. The preset stabilization time is reached; the coal feed rate, primary air volume, and separator speed remain unchanged within the operating range, and the parameter change rate does not exceed the threshold; all sensors involved in the update are in normal health status and the data validity rate meets the standard; there are no events such as mill start-up / shutdown, coal type switching, measuring point purging, maintenance, manual intervention, or automatic grading intervention during the update window; after the cumulative effective sliding window in the same operating range reaches the preset number, abnormal sample windows are removed, the mean and variance of candidate samples are calculated, and the baseline is dynamically adjusted according to the sample quantity and data quality; smooth interpolation is performed at the boundary between different operating ranges to complete the incremental update of the historical baseline model library, and then the process returns to step S1.

7. The intelligent control method for preventing coal mill blockage based on multi-dimensional internal measuring points according to claim 6, characterized in that, In step S3, the preset time window is 60 seconds and the sliding step is 10 seconds.

8. The intelligent control method for preventing coal mill blockage based on multi-dimensional internal measuring points according to claim 6, characterized in that, In step S7, the observation time for stabilization assessment is 5 minutes; the maximum number of interventions is preset to 3.

9. The intelligent control method for preventing coal mill blockage based on multi-dimensional internal measuring points according to claim 6, characterized in that, In step S8, the preset stabilization time is no less than 30 minutes, and the baseline candidate update is performed daily or weekly; abnormal window samples are removed using the median absolute deviation or interquartile range method. Linear interpolation or distance-weighted methods are used to smooth the baseline at the boundary of the operating condition intervals to suppress abrupt changes in deviation caused by the switching of operating conditions.

10. The intelligent control method for preventing coal mill blockage based on multi-dimensional internal measuring points according to claim 6, characterized in that, In step S6, pay attention to the following intervention levels: the coal feed rate should be reduced by 5%-8%, with an adjustment rate of 2t / h. For min, reduce the separator speed by 30-50 rpm; for hazard intervention: reduce the coal feed rate by 10%-15%, with an adjustment rate of 3 t / h. The separator speed is reduced by 80-100 rpm compared to the rated speed.

Images