Anti-corona recognition based rapping system adaptive control method and device

Abstract

The application relates to the technical field of industrial data analysis, and discloses a rapping system adaptive control method and device based on anti-corona identification, the method comprising the following steps: acquiring current data, voltage data and dust concentration data; performing feature extraction according to the current data, the voltage data and the dust concentration data to obtain an operation feature vector; performing abnormality detection according to the operation feature vector to obtain an abnormality occurrence time point; acquiring equipment operation parameters corresponding to the abnormality occurrence time point; performing data correlation according to the equipment operation parameters and the operation feature vector to obtain an abnormality correlation data set; performing parameter optimization according to the abnormality correlation data set to obtain optimal control parameters, and outputting corresponding control instructions. The method has the following effects: the method can improve the abnormal state recognition accuracy in the working process of the rapping system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial data analysis, and in particular to a rapping system adaptive control method and device based on reverse corona identification. BACKGROUND

[0002] Currently, in the application of electric dust collectors, the rapping system, as one of the key components, its main function is to remove dust particles deposited on the dust collecting plate through mechanical vibration, so as to maintain the efficient operation of the equipment. However, due to the change of operating conditions and the wear caused by long-time operation, the efficiency of the rapping system may decrease, and even cause reverse corona phenomenon, that is, under the action of high voltage, the local area of the dust collecting plate surface appears discharge phenomenon, which not only reduces the dust removal effect, but also may cause secondary pollution. Therefore, how to accurately identify the reverse corona phenomenon and adjust the rapping strategy accordingly becomes the key to improve the performance of the electric dust collector.

[0003] In the prior art, a rapping system adaptive control method based on reverse corona identification includes the following steps: first, real-time monitoring of voltage, current and other parameters through sensors installed inside the electric dust collector; second, using signal processing algorithm to analyze these data to identify whether there is reverse corona phenomenon; third, once the reverse corona phenomenon is detected, the system will automatically adjust the rapping frequency and intensity to quickly remove the dust on the dust collecting plate and restore normal electrical characteristics; finally, continuously monitor the effect of the adjustment and further optimize the rapping strategy according to the feedback information.

[0004] However, the identification of reverse corona phenomenon is based on the monitoring of electrical parameters such as voltage and current, and by comparing with the pre-set fixed threshold to judge whether reverse corona occurs. But under different working conditions, such as temperature, humidity and the change of dust composition, these electrical parameters will show significant fluctuations. For example, in a high humidity environment, the moisture in the air will increase the conductivity of the dust collecting plate surface, thereby causing the current to increase; and when dealing with a specific type of dust, due to its different resistance characteristics from the normal situation, it will also cause the voltage and current readings to deviate from the normal range. Therefore, when using fixed threshold for judgment, such changes will cause the system to misjudge or miss the occurrence of reverse corona phenomenon. Ultimately, the accuracy of identifying abnormal state in the working process of the rapping system is low. SUMMARY

[0005] The present application provides a rapping system adaptive control method and device based on reverse corona identification to achieve the effect of improving the accuracy of identifying abnormal state in the working process of the rapping system.

[0006] In the first aspect, in order to solve the above technical problems, the present application provides a rapping system adaptive control method based on reverse corona identification, comprising:

[0007] obtain current data, voltage data and dust concentration data;

[0008] perform feature extraction according to the current data, the voltage data and the dust concentration data to obtain an operation feature vector;

[0009] perform anomaly detection according to the operation feature vector to obtain an anomaly occurrence time point;

[0010] obtain a device operation parameter corresponding to the anomaly occurrence time point;

[0011] perform data correlation according to the device operation parameter and the operation feature vector to obtain an anomaly correlation dataset;

[0012] perform parameter optimization according to the anomaly correlation dataset to obtain an optimal control parameter, and output a corresponding control instruction.

[0013] In an optional implementation, the performing feature extraction according to the current data, the voltage data and the dust concentration data to obtain an operation feature vector comprises:

[0014] perform amplitude normalization on the current data to obtain normalized current data;

[0015] perform amplitude normalization on the voltage data to obtain normalized voltage data;

[0016] perform wavelet transform according to the normalized current data and the normalized voltage data to obtain electrical wavelet features;

[0017] perform time series statistical analysis according to the normalized current data and the normalized voltage data to obtain electrical statistical features;

[0018] perform gradient calculation according to the dust concentration data to obtain a concentration change gradient;

[0019] perform normalization on the concentration change gradient to obtain a normalized concentration gradient;

[0020] perform horizontal splicing according to the electrical wavelet features, the electrical statistical features and the normalized concentration gradient to obtain an operation feature vector.

[0021] In an optional implementation, the performing anomaly detection according to the operation feature vector to obtain an anomaly occurrence time point comprises:

[0022] input the operation feature vector into a pre-trained anomaly detection model to output an anomaly determination result;

[0023] when the anomaly determination result is normal, do not perform subsequent processing;

[0024] When the abnormality determination result is abnormal, a timestamp corresponding to the operation feature vector is obtained as an abnormality occurrence time point.

[0025] In an optional implementation, the training process of the abnormality detection model comprises:

[0026] obtaining historical operation data and historical abnormality records;

[0027] performing feature extraction according to the historical operation data to obtain a historical feature vector;

[0028] extracting an abnormality type of a time point corresponding to the historical feature vector according to the historical abnormality records, and generating a historical abnormality data set in combination with the historical feature vector;

[0029] building an abnormality detection model based on a support vector machine;

[0030] training by taking the historical feature vector as an input of the model, the corresponding abnormality type as an output of the model, and a mean square error as a loss function;

[0031] completing the training of the model when a training number reaches a preset upper limit or the loss function is detected to be less than a preset threshold.

[0032] In an optional implementation, the obtaining of the device operation parameter corresponding to the abnormality occurrence time point comprises:

[0033] the device operation parameter comprises a beating frequency and a beating strength.

[0034] In an optional implementation, the data correlation according to the device operation parameter and the operation feature vector to obtain an abnormality correlation data set comprises:

[0035] performing time sequence alignment according to the device operation parameter and the operation feature vector to obtain aligned operation parameters and aligned operation features;

[0036] performing correlation analysis according to the aligned operation parameters and the aligned operation features to obtain a parameter-feature correlation degree;

[0037] eliminating data with a parameter-feature correlation degree less than a preset correlation degree threshold;

[0038] constructing an abnormality correlation data set according to data with a parameter-feature correlation degree greater than or equal to the correlation degree threshold.

[0039] In an optional implementation, the parameter optimization according to the abnormality correlation data set to obtain an optimal control parameter and output a corresponding control instruction comprises:

[0040] initializing a reinforcement learning environment, defining a state space, an action space and a reward function according to the abnormal association dataset;

[0041] wherein the action space comprises a rapping force and a rapping frequency;

[0042] in each iteration, updating the rapping force and the rapping frequency according to a preset parameter update rule according to the reward function;

[0043] after each iteration, calculating an average anti-corona suppression rate;

[0044] when the average anti-corona suppression rate is less than or equal to a preset suppression rate threshold, completing the training, obtaining the optimal control parameters, and outputting the corresponding control instructions;

[0045] when the average anti-corona suppression rate is greater than the suppression rate threshold, continuing the iteration;

[0046] when the learning round reaches a preset upper limit of learning rounds, outputting the current result as the optimal control parameters and making a note.

[0047] In a second aspect, the present application provides a rapping system adaptive control device based on anti-corona identification, comprising:

[0048] a data acquisition module for acquiring current data, voltage data and dust concentration data;

[0049] a feature extraction module for performing feature extraction according to the current data, the voltage data and the dust concentration data to obtain a running feature vector;

[0050] an anomaly detection module for performing anomaly detection according to the running feature vector to obtain an abnormal occurrence time point;

[0051] a running parameter module for acquiring device running parameters corresponding to the abnormal occurrence time point;

[0052] a data association module for performing data association according to the device running parameters and the running feature vector to obtain an abnormal association dataset;

[0053] a parameter optimization module for performing parameter optimization according to the abnormal association dataset to obtain optimal control parameters and output corresponding control instructions.

[0054] In a third aspect, the present application further provides an electronic device comprising a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the rapping system adaptive control method based on anti-corona identification according to any one of the above aspects when executing the computer program.

[0055] In a fourth aspect, the present application also provides a computer readable storage medium comprising a stored computer program, wherein the computer program, when executed, controls a device in which the computer readable storage medium is located to perform the adaptive control method of the anti-corona based beating system according to any one of the preceding aspects.

[0056] Compared with the prior art, the present application has the following beneficial effects:

[0057] (1) The process of obtaining current data, voltage data and dust concentration data ensures comprehensive perception of the running state of the beating system. Real-time collection of key parameters closely related to the anti-corona phenomenon by high-precision sensors provides a complete and accurate data basis for subsequent feature extraction and anomaly identification. This step not only improves the reliability of the original data, but also provides high-quality input for local intelligent analysis, enhancing the environmental adaptability and response capability of the system.

[0058] (2) Feature extraction is performed according to the current data, voltage data and dust concentration data to obtain a running feature vector. Time domain statistics, wavelet transform and other methods are used to fuse and process multi-dimensional data to construct a structured running feature vector. This process effectively improves the compactness and representativeness of data expression, reduces redundant information interference, enhances the identifiability of key features, and significantly improves data processing efficiency and model input quality.

[0059] (3) Abnormality detection is performed according to the running feature vector to obtain an abnormality occurrence time point. Machine learning-based abnormality detection is used to analyze the running feature vector in real time to accurately identify the occurrence time of anti-corona and other abnormal events. This step realizes rapid response to abnormal behavior under complex working conditions, improves the sensitivity and accuracy of system identification, and provides accurate time positioning basis for subsequent in-depth analysis.

[0060] (4) The device running parameters corresponding to the abnormality occurrence time point are obtained. After identifying the abnormality occurrence time point, the system automatically traces back and extracts the related device running parameters before and after this time point to form complete abnormal context information. This method ensures the strong correlation between abnormal events and running states, providing sufficient data support for abnormal cause analysis and control strategy optimization.

[0061] (5) Data correlation is performed according to the device running parameters and the running feature vector to obtain an abnormality associated data set. Through correlation analysis between running parameters and feature vectors, highly correlated variable combinations are selected to construct an abnormality associated data set. This step strengthens the understanding of the causes of abnormality, helps to establish a more accurate data set, and improves the scientificity and robustness of system diagnosis.

[0062] (6) According to the abnormal association data set, the parameter optimization is carried out to obtain the optimal control parameter, and the corresponding control instruction is output. Based on the historical abnormal data and the current associated characteristics, the control parameter of the vibration system is dynamically adjusted by using the reinforcement learning algorithm, the adaptive control instruction is generated and executed. The process realizes the closed-loop adjustment mechanism from abnormal identification to control response, which not only improves the ability of the system to cope with complex phenomena such as anti-cathode, but also significantly improves the timeliness and accuracy of control decision. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 is a kind of vibration system adaptive control method flow chart based on anti-cathode identification provided by the first embodiment of the application;

[0064] Figure 2 is a kind of vibration system adaptive control device structure schematic diagram based on anti-cathode identification provided by the second embodiment of the application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0066] Referring to Figure 1 , the first embodiment of the application provides a kind of vibration system adaptive control method based on anti-cathode identification, comprising the following steps:

[0067] S11, obtains current data, voltage data and dust concentration data;

[0068] S12, according to the current data, the voltage data and the dust concentration data are extracted features, to obtain running feature vector;

[0069] S13, according to the running feature vector, abnormal detection is carried out, to obtain abnormal occurrence time point;

[0070] S14, obtains the equipment running parameter corresponding to the abnormal occurrence time point;

[0071] S15, according to the equipment running parameter and the running feature vector, data association is carried out, to obtain abnormal association data set;

[0072] S16, according to the abnormal association data set, parameter optimization is carried out to obtain the optimal control parameter, and the corresponding control instruction is output.

[0073] In step S11, current data, voltage data and dust concentration data are acquired.

[0074] In an embodiment, in the adaptive control method of the rapping system based on the recognition of the back corona, the data acquisition mode is realized by the sensor network deployed in the industrial field: the current data (unit: A) and voltage data (unit: V) are collected in real time by high-precision current transformers and voltage sensors, and the dust concentration data (unit: mg / m³) is obtained by a laser scattering dust sensor or a capacitive particulate matter detection module. The sensor transmits the raw data to the data acquisition terminal (such as PLC or edge computing device) through RS485, Modbus or industrial Ethernet protocol, and stores it in a time-stamped format (such as ISO 8601 standard). The storage form can adopt a time series database (such as InfluxDB or TimescaleDB) to record time series data in the form of key-value pairs or tables (for example: {"timestamp":"2025-06-01T19:00:00Z","current": 5.3A,"voltage": 6.8kV,"dust_concentration": 250mg / m³}). This scheme is suitable for real-time monitoring of industrial dust removal equipment and can meet the requirements of data timeliness and integrity for abnormal detection and parameter optimization.

[0075] In step S12, feature extraction is performed according to the current data, voltage data and dust concentration data to obtain an operating feature vector.

[0076] In an embodiment, the current data is amplitude-normalized to obtain normalized current data;

[0077] The voltage data is amplitude-normalized to obtain normalized voltage data;

[0078] Wavelet transform is performed according to the normalized current data and the normalized voltage data to obtain electrical wavelet features;

[0079] Time series statistical analysis is performed according to the normalized current data and the normalized voltage data to obtain electrical statistical features;

[0080] Gradient calculation is performed according to the dust concentration data to obtain a concentration change gradient;

[0081] The concentration change gradient is normalized to obtain a normalized concentration gradient;

[0082] The electrical wavelet features, the electrical statistical features and the normalized concentration gradient are horizontally spliced to obtain an operating feature vector.

[0083] It is worth mentioning that in the amplitude normalization process, the reference value of current and voltage data is determined first, for example, the maximum instantaneous value during operation is selected as the reference for current data, and the peak value of the rated operating voltage is selected as the reference for voltage data. Then, the original value of each sampling point is calculated by ratio with the corresponding reference value, and the original data is mapped into the normalized interval of 0-1, for example, if the current value at a certain time is 5A and the reference value is 10A, the normalized result is 0.5. This process eliminates the influence of sensor range difference and physical dimension, so that the current and voltage data are in the same numerical scale. The core purpose of amplitude normalization is to eliminate the numerical difference between features and interfere with subsequent wavelet transform and statistical analysis, for example, to avoid the excessive dominance of high-amplitude voltage features in the calculation of model weights, thereby improving the accuracy of feature fusion and providing a more balanced input dimension for anti-corona anomaly detection.

[0084] It is worth mentioning that the implementation process of wavelet transform is based on the principle of multi-scale analysis, which extracts features by decomposing the low-frequency and high-frequency components of the signal at each level. In specific operation, the standardized current and voltage data are first decomposed using the db4 wavelet basis function (Daubechies 4th order) as the decomposition tool, with a 3-level decomposition depth to adapt to the frequency spectrum range of typical industrial signals. In the first layer processing, the original data with a sampling frequency of 1024Hz is input into a low-pass filter (coefficient length 8) to extract the low-frequency approximation coefficient (such as decomposing the 5.3A mean value of the current signal into a 1.2A low-frequency part), while the high-frequency detail coefficient is obtained through a high-pass filter (also 8-order coefficient) (such as capturing the peak noise in voltage fluctuation as a 0.8kV instantaneous change). The second layer further separates the low-frequency information (such as 0.3A change of current trend) and corresponding high-frequency features (such as 0.2kV component of voltage harmonic distortion) by filtering and downsampling the low-frequency approximation coefficient of the first layer. The third layer repeats the above steps, and finally forms 3 scales of approximation coefficients (corresponding to 256Hz, 128Hz, and 64Hz frequency bands) and 3 scales of detail coefficients (corresponding to 128Hz-256Hz, 64Hz-128Hz, and 32Hz-64Hz frequency bands). The electrical wavelet feature is a three-dimensional array structure, including the approximation coefficients (such as 128-point length of the 3rd layer approximation coefficient) and detail coefficients (such as 256-point length of the 2nd layer detail coefficient) of each layer decomposition. These coefficients represent the multi-scale characteristics of the signal through amplitude value (such as 0.1-2.5A / kV range) and time sequence distribution, for example, the high-frequency transient disturbance corresponding to the anti-corona phenomenon will concentrate in the 1st layer detail coefficient and show a peak value, while the device running trend is reflected in the smooth change of the 3rd layer approximation coefficient.

[0085] It is worth noting that when performing time series statistical analysis on standardized current and voltage data, first calculate the local statistical features through sliding window (such as every 100 sampling points as a group). For example, for the normalized current data (range 0-1), calculate the mean (such as 0.45), variance (such as 0.05), maximum (such as 0.92), minimum (such as 0.31), skewness (such as 0.15, reflecting the symmetry of the distribution) and kurtosis (such as 2.8, representing the sharpness of the distribution) of each group. At the same time, the same operation is performed on the voltage data (standardized range 0-1) to obtain its mean (such as 0.63), variance (such as 0.03), maximum (such as 0.95), minimum (such as 0.51), skewness (such as -0.08) and kurtosis (such as 3.1). In addition, in combination with the coordinated change of current and voltage, the correlation coefficient (such as 0.72) is calculated, reflecting the synchronization of the two. Finally, the above features are spliced according to the time sequence to form a multi-dimensional feature vector containing mean, variance, extreme value, skewness, kurtosis and covariance, for example: [0.45, 0.05, 0.92, 0.31, 0.15, 2.8, 0.63, 0.03, 0.95, 0.51, -0.08, 3.1, 0.72]. This process quantifies the distribution form and dynamic fluctuation characteristics of the signal, providing a statistical basis for subsequent anomaly detection, for example, when the current mean drops below 0.3 and the variance increases to 0.15, it indicates that the equipment is running abnormally.

[0086] It is worth noting that in the calculation of the dust concentration gradient, first calculate the difference between the original concentration data (for example, the concentration value at a certain time is 300 mg / m³) and its value at the previous time (for example, 280 mg / m³) to obtain the concentration change gradient at that time (for example, 300-280=20 mg / m³), and then generate a complete gradient sequence by traversing all time points. Then standardize the gradient data, determine the maximum value (such as 150 mg / m³) and minimum value (such as -100 mg / m³) of the gradient, and subtract the minimum value from each gradient value and divide by the range (150-(-100)=250), for example, the original gradient is 20 mg / m³, the standardized result is (20-(-100)) / 250=0.48.

[0087] It is worth noting that finally, the electrical wavelet features, electrical statistical features (such as 12 statistical indicators) and standardized concentration gradients are spliced in the column direction to form a running feature vector containing three dimensions of total sum. This process quantifies the dynamic change trend of dust concentration and integrates it with electrical features to build a multi-dimensional device running state representation system.

[0088] In step S13, abnormality detection is performed according to the running feature vector to obtain the time point of abnormality occurrence.

[0089] In an embodiment, the operation feature vector is input into a pre-trained anomaly detection model, and an anomaly determination result is output;

[0090] When the anomaly determination result is normal, no subsequent processing is performed;

[0091] When the anomaly determination result is abnormal, the timestamp corresponding to the operation feature vector is obtained as an abnormal occurrence time point.

[0092] It is worth noting that when the operation feature vector is input into the trained anomaly detection model, the model outputs a binary determination result: "normal" or "abnormal". If the determination result is "normal", the system directly skips the subsequent processing procedure and only maintains the regular record of the device operation state; if the determination result is "abnormal", the system extracts the timestamp corresponding to the operation feature vector (for example, 2023-10-05 14:32:17), marks it as an abnormal occurrence time point and stores it in the abnormal event log. This process directly associates the model output with the corresponding relationship of time series data, achieving precise positioning of abnormal events, where "normal" determination indicates that the device operation parameters meet the expected range, and "abnormal" determination indicates that there are fluctuations or fault characteristics beyond the normal mode, such as sudden increase in current, voltage harmonic distortion, or abnormal change in dust concentration gradient.

[0093] In an embodiment, historical operation data and historical abnormal records are obtained;

[0094] According to the historical operation data, feature extraction is performed to obtain a historical feature vector;

[0095] According to the historical abnormal records, the abnormal types of the time points corresponding to the historical feature vector are extracted, and a historical abnormal data set is generated in combination with the historical feature vector;

[0096] An anomaly detection model is built based on a support vector machine;

[0097] The historical feature vector is used as the input of the model, the corresponding abnormal type is used as the output of the model, and the mean square error is used as the loss function for training;

[0098] When the number of training times reaches a preset upper limit or the loss function is detected to be less than a preset threshold, the training of the model is completed.

[0099] It is worth noting that in the model training stage, first, two types of data are retrieved from the historical database: historical operation data (containing past collected current, voltage, dust concentration and other raw data, which has been standardized, wavelet transformed and statistically analyzed to generate historical feature vectors) and historical abnormal records (records of abnormal events that have occurred during equipment operation, such as reverse corona, short circuit, dust exceeding standard, etc., each event contains a time stamp and an abnormal type label). Subsequently, the historical operation data is subjected to the same feature extraction process as the real-time data to generate corresponding historical feature vectors (such as the feature vector at the time of occurrence of a certain abnormal event is [approximate coefficient value 1, detail coefficient value 2, …, standardized gradient value]), and according to the abnormal type (such as "reverse corona" or "normal") in the historical abnormal record with the same time stamp, each feature vector is assigned a label. Finally, all feature vectors and corresponding labels are aligned in time sequence to construct a structured historical abnormal data set in the form of [[feature vector 1, label 1], [feature vector 2, label 2], …]. This process binds the physical features in the historical data with the actual occurrence of abnormal types, providing a clear input-output mapping relationship for the support vector machine model, enabling the model to learn the feature distribution rules corresponding to different abnormal patterns.

[0100] It is worth noting that in the process of building an abnormality detection model based on support vector machine, first, the historical feature vectors are taken as the input of the model, and the corresponding abnormal type labels ("normal" or "abnormal") are taken as the output target to build a binary classification task. The model uses a radial basis function kernel (RBF kernel), and its hyperparameters include: penalty coefficient C=10 (controls the tolerance of the classifier to misjudgment, the larger the value, the stricter), kernel function parameter γ=0.1 (adjusts the influence range of RBF kernel, smaller γ value makes the decision boundary smoother). During training, the optimal parameter combination is found in the range of C∈{1, 10, 100} and γ∈{0.01, 0.1, 1} through grid search method, and the mean square error (MSE) is taken as the loss function to measure the deviation between the model's predicted value and the true label (for example, if the true label of a sample is "normal" which is 0, but the model predicts 0.8, then the loss value is (0.8-0)^2=0.64). The upper limit of the number of training iterations is set to 500, and the training is stopped when the MSE value of 5 consecutive iterations decreases by less than 0.001 or the total loss is lower than the preset threshold 0.05. The final model outputs a continuous value (such as -2.3 to +1.8) through the decision function, and sets the threshold rule: if the output value is ≥0.5, it is determined as "abnormal", otherwise as "normal", thereby realizing binary classification. This process balances the model complexity and generalization ability, ensuring high accuracy in identifying abnormal events such as reverse corona in industrial scenarios.

[0101] In step S14, the device operating parameters corresponding to the abnormal occurrence time point are obtained.

[0102] In an embodiment, the device operation parameter includes a rapping frequency and a rapping intensity.

[0103] It is worth noting that after detecting the abnormal occurrence time point, the system will backtrack and extract the device operation parameters corresponding to that time point, among which the rapping frequency and the rapping intensity are the core concerns. The rapping frequency refers to the number of times the mechanical rapping device applies vibration to the filter bag per unit time (such as 20-30 times per second), which directly affects the dust shedding efficiency: too low frequency will lead to insufficient dust removal, and the accumulation of dust layer will increase the filter bag resistance; too high frequency will cause excessive vibration and lead to fatigue damage of the filter bag. The rapping intensity is a quantitative indicator of the vibration amplitude (such as amplitude 20-50 mm), which determines the strength of the dust layer being destroyed: insufficient intensity makes it difficult for dust with strong adhesion to fall off; excessive intensity will tear the filter bag fibers or accelerate the structure aging. During the dust removal process, these two parameters jointly regulate the vibration mode of the filter bag to ensure efficient peeling of the dust layer with minimal damage. For example, when the abnormality is determined to be "reverse corona" or "dust clumping", the system can temporarily increase the rapping frequency (such as from 25 times / s to 30 times / s) or moderately increase the rapping intensity (such as amplitude from 30 mm to 45 mm) to strengthen the dust removal effect and restore the filter bag air permeability, while preventing excessive rapping by monitoring the resistance changes in real time, thereby balancing the dust removal efficiency and the device life.

[0104] In step S15, data correlation is performed according to the device operation parameters and the operation feature vectors to obtain an abnormal correlation data set.

[0105] In an embodiment, time sequence alignment is performed according to the device operation parameters and the operation feature vectors to obtain aligned operation parameters and aligned operation features.

[0106] Correlation analysis is performed according to the aligned operation parameters and the aligned operation features to obtain a parameter feature correlation degree.

[0107] Data with a parameter feature correlation degree less than a preset correlation degree threshold is removed.

[0108] An abnormal correlation data set is constructed according to data with a parameter feature correlation degree greater than or equal to the correlation degree threshold.

[0109] It is worth noting that the core goal of time alignment is to eliminate the misalignment of device operating parameters (such as beating frequency, beating force) and operating feature vectors caused by differences in sampling frequency or asynchronous collection through a unified time axis. The specific implementation process includes: first, selecting a high-frequency sampling sensor (such as a 1Hz beating parameter record controlled by PLC) or merging multiple sensor timestamps to generate a master timeline through the least common multiple method (LCM); then, linear interpolation, spline interpolation and other algorithms are used to complete the missing values (for example, mapping the 5-second granularity dust concentration to the 1-second time point) for low-frequency or asynchronous data; for non-linear offset scenarios (such as time sequence misalignment when the device starts), the dynamic time warping (DTW) algorithm is introduced to construct a distance matrix and dynamically plan the minimum cumulative error to achieve flexible alignment of parameters and features; finally, the aligned data is reorganized into a two-dimensional matrix ([timestamp, frequency, force, resistance, concentration]).

[0110] It is worth noting that after completing the time alignment of the aligned operating parameters (such as beating frequency, beating force) and the aligned operating features, the correlation between the parameters and the features needs to be quantified through correlation analysis. The specific process is as follows: first, the Spearman rank correlation coefficient (suitable for non-linear but monotonic relationships) is used to calculate the correlation between each pair of parameters and features. For example, if the beating frequency and voltage variance show strong positive correlation after time alignment, it indicates that there is a significant direct influence between them; if the correlation coefficient between the beating force and the dust concentration gradient is small, the correlation is weak. Then, set a correlation threshold (such as 0.7) to eliminate data pairs with a correlation coefficient below the threshold to eliminate noise interference; for example, if the correlation coefficient between a parameter and a feature is 0.65, it is filtered out, and only data pairs with a correlation coefficient greater than or equal to 0.7 (such as the correlation coefficient between the beating force and the dust concentration gradient is 0.75) are retained. Finally, based on the high-correlation data pairs selected, an abnormal correlation data set is constructed to form a structured matrix, providing accurate input for subsequent anomaly pattern recognition.

[0111] In step S16, parameter optimization is performed according to the abnormal correlation data set to obtain optimal control parameters and output corresponding control instructions.

[0112] In one embodiment, the reinforcement learning environment is initialized, and the state space, action space and reward function are defined according to the abnormal correlation data set;

[0113] The action space includes beating force and beating frequency.

[0114] In each iteration, the beating force and beating frequency are updated according to the reward function and a preset parameter update rule;

[0115] At the end of each iteration, the average anti-corona suppression rate is calculated;

[0116] When the average anti-corona suppression rate is less than or equal to the preset suppression rate threshold, the training is completed, the optimal control parameter is obtained, and the corresponding control instruction is output;

[0117] When the average anti-corona suppression rate is greater than the suppression rate threshold, the iteration continues;

[0118] When the learning round reaches the preset upper limit of the learning round, the current result is output as the optimal control parameter, and a note is made.

[0119] It is worth noting that in the parameter optimization process based on reinforcement learning, first, the reinforcement learning environment is initialized, the key variables in the abnormal correlation data set (such as the vibration frequency, the vibration intensity, the dust concentration gradient, etc.) are defined as the state space, and the action space is composed of the adjustable vibration intensity (range: 0-100N) and the vibration frequency (range: 10-50Hz). In each iteration, the agent selects an action (such as adjusting the vibration intensity to 75N and the frequency to 30Hz) according to the current state, observes the system response (such as the change in the anti-corona suppression rate) after execution, and updates the policy parameters (such as adjusting the neural network weights by the gradient ascent method) according to the reward function. After the iteration is completed, the average anti-corona suppression rate under the current policy (for example, 92%) is calculated, and if the value is less than or equal to the preset suppression rate threshold (for example, 90%), it is determined that the training goal is achieved, and the current optimal control parameter (such as vibration intensity 80N, frequency 35Hz) and the corresponding control instruction are output; if the suppression rate is higher than the suppression rate threshold (such as 92%>90%), the iteration optimization continues. When the learning round reaches the preset upper limit (such as 1000 rounds), if the suppression rate threshold requirement is not met, the current optimal solution is output as an approximate solution, and a note of "not fully converged" is made. For example, in the dust removal system training, if the suppression rate is always lower than 90% in the first 950 rounds, the parameter combination of vibration intensity 70N and frequency 28Hz is output in the 1000th round, and the result is recorded as a suboptimal solution. This process combines dynamic policy updating with threshold constraints to ensure efficient optimization of anti-corona suppression within limited resources.

[0120] It is worth noting that in calculating the average anti-cathode suppression rate under the current strategy, first of all, it is necessary to execute the control instructions based on the adjusted vibration force and vibration frequency in each round of iteration (for example, force 75N, frequency 30Hz), and record the running data of the system under this parameter combination. The specific process is as follows: in each round of iteration, the anti-cathode intensity during the operation of the equipment is monitored (such as through the value of current or voltage fluctuation), and the baseline value before processing (such as the anti-cathode intensity before adjusting the parameters is 120A) and the actual value after processing (such as the intensity decreases to 30A after adjusting the parameters) are recorded respectively. The calculation of the suppression rate is based on the comparison of the two measured values, that is, (baseline value - actual value) / baseline value x 100% (for example, (120A - 30A) / 120A x 100% = 75%). Then, the suppression rate obtained in each round of iteration (such as 75% in the first round, 80% in the second round, and 78% in the third round) is accumulated, and finally divided by the total number of iterations (such as 100 rounds) to obtain the average suppression rate. For example, if the cumulative suppression rate in 100 rounds of iteration is 7500%, the average suppression rate is 75%. If this value is lower than the preset suppression rate threshold (such as 90%), the training termination condition is triggered, and the current optimal parameters are output; if it is higher than the threshold, optimization continues. This calculation process quantifies the actual change in anti-cathode intensity before and after parameter adjustment, and combines the statistical results of multiple iterations to ensure that the average suppression rate can objectively reflect the effectiveness of the strategy.

[0121] It is worth noting that in designing the reward function, the following rules need to be followed: first of all, it is necessary to clarify the core goal of the task as suppressing the anti-cathode phenomenon, and to decompose it into quantifiable sub-goals (such as reducing current fluctuation, reducing energy consumption, maintaining equipment stability). For each sub-goal, a corresponding reward item needs to be designed: 1. Positive reward: when the anti-cathode suppression rate increases after adjusting the parameters (such as from 75% to 85%), give +0.5 reward for every 1% increase; if the suppression rate exceeds the preset threshold (such as 90%), give an additional high reward of +2; 2. Negative punishment: if the vibration force or frequency exceeds the safe range (such as force > 100N or frequency > 50Hz), deduct -1 for each over-limit; if the energy consumption of each round exceeds 5kW, deduct -0.2 / kW for the excess; 3. Action smoothness constraint: for actions with large parameter changes between adjacent rounds (such as vibration force fluctuation > 10N), apply a penalty of -0.3 each time to avoid system oscillation; 4. Long-term goal guidance: if the suppression rate has been stable above the threshold for 3 consecutive rounds (such as 90% ± 2%), give an additional reward of +1.5 to encourage the strategy to converge. The weight distribution needs to balance the priority of each sub-goal: suppression rate increase accounts for 60% of the total reward, energy consumption penalty accounts for 20%, action smoothness accounts for 15%, and long-term stability accounts for 5%; Dynamic adjustment mechanism: if the convergence is slow at the beginning of training, the positive reward weight can be increased in stages (such as the suppression rate increase reward from +0.5 to +0.8), and gradually restored to balance in the later stage.

[0122] It is worth noting that in each round of iteration, the parameter update rule dynamically adjusts the shaking force (range 0-100N) and shaking frequency (range 10-50Hz) based on the feedback signal of the reward function, the specific process is as follows: first, execute the control instruction according to the shaking parameters generated by the current strategy, and record the reward value of this round of iteration (such as +1.2). If the reward value is positive (such as +0.5 due to the increase of suppression rate), adjust the current parameters in the optimal direction by a preset step (such as +2N / time, +1Hz / time) respectively: if the shaking force is 70N and the reward value comes from the increase of suppression rate, increase to 72N; if the shaking frequency is 30Hz and the reward value comes from the reduction of energy consumption, keep or fine-tune to 31Hz. If the reward value is negative (such as -0.8 due to over-limit penalty), adjust the parameters in the opposite direction: if the shaking force is 95N and points are deducted due to over-limit, reduce to 90N; if the shaking frequency is 48Hz and points are deducted due to high energy consumption, reduce to 45Hz. At the same time, an adaptive step size mechanism is introduced: when the reward value increases steadily for 3 consecutive rounds, gradually reduce the adjustment step size (such as from +2N to +1N) to avoid overshoot; if the reward value fluctuates sharply, expand the step size (such as from +1N to +3N) to quickly explore new strategies. In addition, boundary constraints are set: if the adjusted parameters exceed the safe range (such as force >100N or frequency <10Hz), force them to return to the last valid value. For example, in a certain round of iteration, the shaking force is 98N, and it needs to be increased by 2N to 100N due to the reward value +0.7, but in the next round, if the suppression rate does not increase and the energy consumption increases, a negative reward is triggered, then the force is returned to 95N and the step size is reduced to 1N. This rule combines positive feedback amplification and negative feedback correction driven by reward signals to balance exploration and convergence, and finally approaches the optimal solution.

[0123] In summary, the present application discloses a kind of based on anti-corona identification's shaking system adaptive control method, to pass through electrical characteristics and dust concentration characteristics, utilize machine learning algorithm to carry out abnormal detection, improve identification accuracy.And through reinforcement learning algorithm is carried out equipment operating parameter optimization, improve dust removal effect.

[0124] With reference Figure 2 The second embodiment of the present application provides a kind of based on anti-corona identification's shaking system adaptive control device, comprising:

[0125] Data acquisition module is used to obtain current data, voltage data and dust concentration data;

[0126] Feature extraction module is used to carry out feature extraction according to the current data, the voltage data and the dust concentration data, and obtain operating feature vector;

[0127] Abnormal detection module is used to carry out abnormal detection according to the operating feature vector, and obtain abnormal occurrence time point;

[0128] a running parameter module, configured to acquire a device running parameter corresponding to the time point of the anomaly occurrence;

[0129] a data association module, configured to perform data association according to the device running parameter and the running feature vector to obtain an anomaly association data set;

[0130] a parameter optimization module, configured to perform parameter optimization according to the anomaly association data set to obtain an optimal control parameter and output a corresponding control instruction.

[0131] It should be noted that the adaptive control device of the vibrating system based on the anti-corona identification provided by the embodiment of the present application is used to execute all process steps of the adaptive control method of the vibrating system based on the anti-corona identification provided by the above embodiment, and the working principles and beneficial effects of the two are one-to-one correspondence, thus not being repeated.

[0132] The embodiment of the present application further provides an electronic device. The electronic device comprises a processor, a memory and a computer program, such as a data acquisition program, stored in the memory and executable on the processor. The processor implements the steps in the above various adaptive control methods of the vibrating system based on the anti-corona identification when executing the computer program, such as the step S11 shown in the figure. Figure 1 Alternatively, the processor implements the functions of the modules / units in the above various device embodiments when executing the computer program, such as the data acquisition module.

[0133] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0134] The electronic device can be a desktop computer, a notebook, a palm computer and a smart tablet, etc. The electronic device can include, but is not limited to, a processor, a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device, and can include more or fewer components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, etc.

[0135] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is a control center of the electronic device, and connects various parts of the electronic device through various interfaces and lines.

[0136] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory device.

[0137] The modules / units integrated in the electronic device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. that can carry the computer program code. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0138] It should be noted that the above-described device embodiments are only schematic, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment. In addition, the connection relationship between the modules in the device embodiment provided by the present application indicates that there is a communication connection between them, which can be realized as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0139] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are only for the specific embodiments of the present application and are not used to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for adaptive control of a rapping system based on recognition of back corona, characterized in that, The method comprises the following steps: obtaining current data, voltage data and dust concentration data; performing feature extraction according to the current data, the voltage data and the dust concentration data to obtain an operating feature vector; performing anomaly detection according to the operating feature vector to obtain an anomaly occurrence time point; obtaining device operating parameters corresponding to the anomaly occurrence time point; performing data correlation according to the device operating parameters and the operating feature vector to obtain an anomaly correlation data set; performing parameter optimization according to the anomaly correlation data set to obtain optimal control parameters and output corresponding control instructions; wherein performing parameter optimization according to the anomaly correlation data set to obtain optimal control parameters and output corresponding control instructions comprises: initializing a reinforcement learning environment, defining a state space, an action space and a reward function according to the anomaly correlation data set; wherein the state space refers to key variables in the anomaly correlation data set, including rapping frequency, rapping intensity and dust concentration gradient; the action space includes rapping intensity and rapping frequency; in each iteration, updating the rapping intensity and the rapping frequency according to the reward function and a preset parameter update rule; after each iteration, the suppression rate obtained in each iteration is accumulated, and the average anti-corona suppression rate is finally obtained by dividing the total number of iterations; wherein in each iteration, the anti-corona intensity during device operation is monitored, and the baseline value before processing and the actual value after processing are recorded respectively; the calculation process of the suppression rate obtained in each iteration is based on the comparison of the two measurement values, that is, (baseline value - actual value) / baseline value × 100%; when the average anti-corona suppression rate is less than or equal to a preset suppression rate threshold, the training is completed, the optimal control parameters are obtained, and the corresponding control instructions are outputted; when the average anti-corona suppression rate is greater than the suppression rate threshold, the iteration is continued; when the learning round reaches a preset upper limit of learning rounds, the current result is outputted as the optimal control parameters, and a note is made.

2. The anti-cornoid-based rapping system adaptive control method according to claim 1, characterized in that, The feature extraction according to the current data, the voltage data and the dust concentration data to obtain an operating feature vector comprises: performing amplitude normalization on the current data to obtain normalized current data; performing amplitude normalization on the voltage data to obtain normalized voltage data; performing wavelet transform on the normalized current data and the normalized voltage data to obtain electrical wavelet features; performing time series statistical analysis on the normalized current data and the normalized voltage data to obtain electrical statistical features; performing gradient calculation on the dust concentration data to obtain concentration change gradient; performing standardization on the concentration change gradient to obtain normalized concentration gradient; performing horizontal splicing on the electrical wavelet features, the electrical statistical features and the normalized concentration gradient to obtain an operating feature vector.

3. The anti-cornoid-based rapping system adaptive control method according to claim 1, wherein, The anomaly detection according to the operating feature vector to obtain an anomaly occurrence time point comprises: inputting the operating feature vector into a pre-trained anomaly detection model to output an anomaly judgment result; when the anomaly judgment result is normal, no subsequent processing is performed. When the abnormality determination result is abnormal, a timestamp corresponding to the operation feature vector is obtained as an abnormality occurrence time point.

4. The anti-cornoid-based rapping system adaptive control method according to claim 3, characterized in that, The training process of the abnormality detection model comprises: obtaining historical operation data and historical abnormality records; performing feature extraction according to the historical operation data to obtain historical feature vectors; extracting abnormality types of time points corresponding to the historical feature vectors according to the historical abnormality records, and generating a historical abnormality data set in combination with the historical feature vectors; constructing an abnormality detection model based on a support vector machine; training by taking the historical feature vectors as inputs of the model, corresponding abnormality types as outputs of the model, and a mean square error as a loss function; completing the training of the model when the number of training times reaches a preset upper limit or the loss function is detected to be less than a preset threshold.

5. The anti-cornoid-based rapping system adaptive control method according to claim 1, wherein, The device operation parameters corresponding to the abnormality occurrence time point comprise: The device operation parameters comprise a frequency of vibration and a degree of vibration.

6. The anti-cornoid-based rapping system adaptive control method according to claim 1, wherein, The data correlation based on the device operation parameters and the operation feature vectors comprises: performing time sequence alignment based on the device operation parameters and the operation feature vectors to obtain aligned operation parameters and aligned operation features; performing correlation analysis based on the aligned operation parameters and the aligned operation features to obtain parameter-feature correlation degrees; eliminating data with a parameter-feature correlation degree less than a preset correlation degree threshold; constructing an abnormality correlation data set based on data with a parameter-feature correlation degree greater than or equal to the correlation degree threshold.

7. A device for adaptive control of a rapping system based on recognition of a counter corona, characterized in that A method for implementing the adaptive control of a vibration system based on the recognition of reverse corona as claimed in any one of claims 1 to 6 comprises: a data acquisition module for acquiring current data, voltage data and dust concentration data; a feature extraction module for performing feature extraction based on the current data, the voltage data and the dust concentration data to obtain an operation feature vector; an abnormality detection module for performing abnormality detection based on the operation feature vector to obtain an abnormality occurrence time point; an operation parameter module for acquiring device operation parameters corresponding to the abnormality occurrence time point; a data correlation module for performing data correlation based on the device operation parameters and the operation feature vector to obtain an abnormality correlation data set; a parameter optimization module for performing parameter optimization based on the abnormality correlation data set to obtain optimal control parameters and output corresponding control instructions.

8. An electronic device, comprising: A computer program product comprising a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the adaptive control method of a vibration system based on the recognition of reverse corona as claimed in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer readable storage medium controls a device in which the computer readable storage medium is located to implement the adaptive control method of a vibration system based on the recognition of reverse corona as claimed in any one of claims 1 to 6 when the computer program is executed.

Images