System and method for operation monitoring and early warning control of coal mill

Abstract

The invention relates to the field of data monitoring, in particular to a system and a method for operation monitoring and early warning control of a coal mill, and discloses the method for operation monitoring and early warning control of the coal mill, which comprises the following steps of: preprocessing acquired data; multi-modal data feature extraction is carried out on the preprocessed data, and mutual information is calculated through a cross-parameter correlation analysis method. And classifying operating conditions of the coal mill according to the extracted feature set, performing reasoning prediction on risks of different working conditions by constructing a fuzzy reasoning system, and finally performing dynamic weight distribution according to risk levels and mutual information. And generating a dynamic control instruction set for operation of the coal mill according to the dynamic weight vector and the feature vector set, performing verification and comparison, and outputting a predicted equipment simulation state. According to the method, feature region and correlation analysis is performed on the multi-modal data, so that synchronous detection of the data sent by different sensors can be realized.

Description

Technical Field

[0001] This invention relates to the field of data monitoring, and in particular to a system and method for monitoring and early warning control of coal mill operation. Background Technology

[0002] In daily operation, thermal power plants typically use coal mills to grind coal lumps into fine pulverized coal. This pulverized coal is then blown into the combustion chamber through pipes via primary and secondary air blowing. Coal mills are usually fed through a feed inlet, and the coal lumps are ground by grinding components, typically using ball bearings or grinding mills. During the grinding process, air blowing is also performed. Because the airflow is relatively weak, only pulverized coal meeting the required size is blown up and fed into the combustion chamber for combustion, while larger particles are circulated and ground internally until the entire processing is complete.

[0003] During the operation of a coal mill, various sensors are typically installed to perform online data monitoring, and different monitoring systems are used for different equipment. However, existing detection and early warning systems do not adequately identify the interactions between different parameters, which leads to incomplete risk warnings in some cases and the possibility of false warnings. Therefore, it is necessary to design a system and method for monitoring and controlling the operation of a coal mill to address this problem. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides the following technical solution:

[0005] The method for monitoring and early warning control of coal mill operation includes the following steps:

[0006] S10: Preprocess the collected data.

[0007] S20: Extract multimodal data features from the preprocessed data and calculate mutual information using cross-parameter correlation analysis.

[0008] S30: Classify the operating conditions of the coal mill based on the extracted feature set, and predict the risks of different operating conditions by constructing a fuzzy inference system. Finally, perform dynamic weight allocation based on risk level and mutual information.

[0009] S40: Generate a set of dynamic control commands for the operation of the coal mill based on the dynamic weight vector and the set of feature vectors, verify and compare them, output the predicted equipment simulation status, and perform automatic adjustment or early warning based on the equipment simulation status.

[0010] As an improvement to the above technical solution, step S10 includes the following steps:

[0011] S11: Uses the IEEE 1588 precision time protocol to synchronize the network clock, and converts the input data into floating-point values ​​through the Modbus protocol.

[0012] S12: 3 based on sliding window The criteria filter out outliers in the transformed data.

[0013] S13: Use the improved Z-score normalization algorithm to normalize the filtered data and output the normalized data.

[0014] As an improvement to the above technical solution, step S20 includes the following steps:

[0015] S21: Calculate the statistics of the data in the input window and extract the time-domain feature vector.

[0016] S22: Perform a fast Fourier transform on the time-domain eigenvectors and then perform wavelet packet energy decomposition to obtain the frequency-domain eigenvectors.

[0017] S23: Calculate mutual information for various parameters during the operation of the coal mill.

[0018] As an improvement to the above technical solution, step S30 includes the following steps:

[0019] S31: Based on the set of input feature vectors, the operating states are divided using an improved K-means++ algorithm to obtain the working condition classification data for each parameter.

[0020] S32: Construct a fuzzy inference system and use this fuzzy inference system to assess the risk level of the set of input feature vectors.

[0021] S33: Based on the risk level assessment results, the dynamic weight allocation of each parameter is obtained by using a nonlinear mapping method.

[0022] As an improvement to the above technical solution, step S40 includes the following steps:

[0023] S41: Construct a multi-model collaborative controller and obtain control commands for various parameters based on the geometry of the input feature vectors and the dynamic weight allocation.

[0024] S42: Perform physical simulation based on control commands to obtain prediction results, and use an LSTM model to predict the remaining lifespan. Output the predicted equipment simulation status. If the equipment simulation is in a normal state, automatically adjust or do not modify according to the control commands. If the equipment simulation is in a risky state, automatically adjust or issue a risk warning according to the control commands.

[0025] As an improvement to the above technical solution, the multi-model collaborative controller includes at least: a temperature controller, a pressure controller, a vibration controller, and a gas controller. The temperature controller is used to dynamically adjust the temperature of the primary hot air based on the pressure and vibration conditions of the equipment. The pressure controller is used to dynamically adjust the internal air pressure based on the equipment vibration and the internal gas filling status. The gas controller is used to determine the filling gas status based on the temperature and pressure conditions. The vibration controller is used to adjust the injected gas volume or issue an early warning based on the equipment vibration.

[0026] The system for monitoring and early warning control of coal mill operation adopts the method for monitoring and early warning control of coal mill operation as described in any of the aforementioned technical solutions, including: a data processing module, a feature extraction module, a dynamic parameter identification module, a coordinated control module, and a data verification and prediction module.

[0027] Specifically, the data processing module preprocesses the input multimodal data; the feature extraction module extracts features from the processed standardized data and constructs a cross-parameter correlation matrix to obtain a set of feature vectors for the multimodal data; the dynamic parameter identification module classifies operating conditions based on the set of feature vectors for the multimodal data, assesses the risk level of each operating condition, and obtains the dynamic weight vector corresponding to the risk level; the coordination control module generates dynamic control commands for the operation of the coal mill based on the dynamic weight vectors and the set of feature vectors; and the data verification and prediction module verifies and compares the input dynamic control commands with the equipment operating status and outputs the predicted equipment simulation status.

[0028] As an improvement to the above technical solution, the data processing module preprocesses multimodal data in at least the following ways: data format standardization, outlier filtering, and normalization.

[0029] As an improvement to the above technical solution, the feature extraction module includes a time-domain feature extraction module, a frequency-domain feature extraction module, and a cross-parameter correlation analysis module.

[0030] The time-domain feature extraction module is used to extract the time-domain features of the standardized data, the frequency-domain feature extraction module is used to extract the frequency-domain features of the standardized data, and the cross-parameter correlation analysis module is used to calculate the mutual information between various parameters during the operation of the coal mill.

[0031] As an improvement to the above technical solution, the dynamic parameter identification module includes a working condition classification module, a risk assessment module, and a dynamic weight allocation module.

[0032] The operating condition classification module is used to classify operating condition parameters based on the set of input feature vectors. The risk assessment module is used to assess the risk level based on the classified operating condition parameters. The dynamic weight allocation module is used to quickly locate other parameters with the strongest correlation to abnormal parameters based on mutual information and risk level, and dynamically adjust the control weights.

[0033] The beneficial effects of this invention are:

[0034] By performing feature region and correlation analysis on multimodal data, the implicit coupling mechanism between parameters can be revealed by quantifying the nonlinear dependencies between multiple parameters. Based on this implicit relationship, a dynamic weight redistribution can be constructed, thereby achieving synchronous detection of data from different sensors. The operating status of the coal mill can be dynamically adjusted according to the relationship between the detected data, and the service life of the coal mill can be predicted by combining the mutual influence between the data. Compared with existing monitoring systems, the monitoring data is more accurate and can prevent problems in advance, achieving early diagnosis of faults. Attached Figure Description

[0035] Figure 1 This is a flowchart of the online monitoring and early warning method for coal mill operation safety of the present invention;

[0036] Figure 2 This is a schematic diagram of the system for monitoring and early warning control of coal mill operation according to the present invention. Detailed Implementation

[0037] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0038] During the operation of a coal mill, various sensors are typically installed to perform online data monitoring, and different monitoring systems are used for different equipment. However, existing detection and early warning systems do not adequately identify the interactions between different parameters, which leads to incomplete risk warnings in some cases and the possibility of false warnings. Therefore, it is necessary to design a system and method for monitoring and controlling the operation of a coal mill to address this problem.

[0039] To address the above problems, the following implementation method is provided:

[0040] Example 1

[0041] Please see Figure 1A method for monitoring and early warning control of coal mill operation is provided, comprising the following steps:

[0042] S10: Preprocess the collected data.

[0043] Data acquisition typically involves data transmitted from various sensors, such as temperature sensors, pressure sensors, vibration frequency sensors, and gas sensors.

[0044] The specific processing method for sensor data is as follows:

[0045] S11: Uses the IEEE 1588 precision time protocol to synchronize the network clock, and converts the input data into floating-point values ​​through the Modbus protocol.

[0046] IEEE 1588 is a precision time protocol used to synchronize clock information in computer networks. It can reduce time synchronization errors when multiple sensors transmit data and ensure data consistency in the time dimension. The Modbus protocol is a serial communication protocol widely used in industrial control. This protocol is characterized by high reliability, high efficiency and simplicity.

[0047] After the basic information of the data is unified, the unified data is filtered, specifically in step S12.

[0048] S12: 3 based on sliding window The criteria filter out outliers in the transformed data.

[0049] Because this algorithm introduces a variable sliding window during the denoising process, it performs step-by-step continuous signal processing, while using 3D filtering to filter noise. The criterion algorithm solves the problems of noise omission and over-denoising.

[0050] 3 based on sliding window The criterion algorithm depends on the following formula:

[0051]

[0052] in, The original measurement value of the i-th data point. This represents the arithmetic mean of the data within the sliding window. This represents the standard deviation of the data within the sliding window.

[0053] S13: Use the improved Z-score normalization algorithm to normalize the filtered data and output the normalized data.

[0054] The implementation of the improved Z-score normalization algorithm depends on the following formula:

[0055]

[0056] in, For a single data point in the original measurement, For the original dataset containing all data points, This indicates taking the median value. This represents the absolute deviation of the median. These are the standardized values.

[0057] To further illustrate the solution in this embodiment, the following example is provided. The execution of the example depends on Python code, as detailed below:

[0058] import numpy as np

[0059] from scipy.stats import median_abs_deviation

[0060] class DataPreprocessor:

[0061] def __init__(self, window_size=60):

[0062] self.window = np.zeros(window_size)

[0063] def process(self, raw_data):

[0064] # Sliding window update

[0065] self.window = np.roll(self.window, -1)

[0066] self.window[-1] = raw_data

[0067] # Outlier Detection and Repair

[0068] median = np.median(self.window)

[0069] mad = median_abs_deviation(self.window, scale='normal')

[0070] if abs(raw_data - median) > 3 * mad:

[0071] return np.mean(self.window[:-1]) # Use the window mean to replace outliers

[0072] else:

[0073] return raw_data

[0074] def normalize(self, x):

[0075] return (x - np.median(self.window)) / (1.4826 * mad)

[0076] After filtering outliers and normalizing the input data using the above code, step S20 is executed.

[0077] S20: Extract multimodal data features from the preprocessed data and calculate mutual information using cross-parameter correlation analysis.

[0078] Specifically, step S20 includes the following steps:

[0079] S21: Calculate the statistics of the data in the input window and extract the time-domain feature vector.

[0080] The statistical methods for time-domain feature vectors are typically as follows:

[0081]

[0082] After the extraction of time-domain features is completed, the parameters of the time-domain feature vector are converted to time-frequency, specifically step S22.

[0083] S22: Perform a fast Fourier transform on the time-domain eigenvectors and then perform wavelet packet energy decomposition to obtain the frequency-domain eigenvectors.

[0084] After undergoing Fast Fourier Transform, the data is further refined by wavelet packet energy decomposition to obtain more detailed feature vectors, providing higher time-frequency resolution. In subsequent analysis, if the energy of a certain frequency band is significantly higher than historical data, a warning is issued indicating a potential fault risk at the location corresponding to that frequency band.

[0085] To further improve this embodiment, the following example is provided, which relies on Python code, as follows:

[0086] def fft_features(signal, fs=1000):

[0087] n = len(signal)

[0088] freq = np.fft.rfftfreq(n, d=1 / fs)

[0089] amp = np.abs(np.fft.rfft(signal))

[0090] # Extract dominant frequency components

[0091] top3_freq = freq[np.argsort(amp)[-3:]]

[0092] return top3_freq, amp.max()

[0093] The code above is used to perform frequency domain transformation on the input information. The final transformed data is then subjected to wavelet packet energy decomposition, which depends on the following equation:

[0094]

[0095] in, It represents frequency band energy, used to quantify the energy intensity of a signal within a specific frequency range, and is used to analyze signal characteristics. Represents the first frequency band after decomposition. There are 1 coefficient, where N represents the total number of components within the frequency band, used to determine the range of summation, covering all components in that frequency band.

[0096] S23: Calculate mutual information for various parameters during the operation of the coal mill.

[0097] The calculation of mutual information refers to calculating the mutual information of parameters such as temperature, pressure, and vibration during the operation of a coal mill. Mutual information can capture the complex physical coupling relationships between parameters such as temperature, pressure, and vibration during the operation of the coal mill (for example, thermal expansion leads to changes in gaps → increased vibration). The specific calculation relies on the following formula:

[0098]

[0099] in, Indicates parameters and Mutual information, express and The joint probability, express The marginal probability represents ignoring... hour, The probability distribution, express The marginal probability represents ignoring... hour, The probability distribution, Represents point mutual information, if ,but and Positive correlation under this combination of values, if If so, then there is a negative correlation.

[0100] The format of the parameters in the above scheme is as follows:

[0101] {

[0102] "temperature": {"mean": 72.3, "gradient": 0.5, "fft_peak": 45.2},

[0103] "pressure": {"variance": 1.2, "dominant_freq": [12.5, 25.0]},

[0104] "vibration": {"kurtosis": 4.1, "wavelet_energy": [0.3, 0.5, 0.2]},

[0105] "gas": {"CO_slope": 0.8, "O2_fft_corr": 0.75}

[0106] }

[0107] This includes various parameters such as temperature, pressure, vibration, and gas. These parameters are presented as vector sets, and the sets of information need to be classified according to operating conditions and subject to risk assessment.

[0108] S30: Classify the operating conditions of the coal mill based on the extracted feature set, and predict the risks of different operating conditions by constructing a fuzzy inference system. Finally, perform dynamic weight allocation based on risk level and mutual information.

[0109] Specifically, step S30 includes the following steps:

[0110] S31: Based on the set of input feature vectors, the operating states are divided using an improved K-means++ algorithm to obtain the working condition classification data for each parameter.

[0111] The improved K-means++ algorithm is a clustering algorithm that treats all points as a single cluster and then repeatedly splits the cluster into two sub-clusters until a predetermined number of clusters is reached. The implementation of this algorithm in this embodiment relies on the following example:

[0112] from sklearn.cluster import KMeans

[0113] class DynamicCluster:

[0114] def __init__(self, n_clusters=3):

[0115] self.model = KMeans(n_clusters=n_clusters, init='k-means++')

[0116] def predict(self, features):

[0117] # Feature dimensions: [Temperature gradient, pressure variance, vibration kurtosis, CO change rate]

[0118] return self.model.predict(features.reshape(1,-1))

[0119] By calling the Kmeans method, the working conditions are classified into vector sets with multiple feature dimensions. After the classification is completed, the risk level of each state needs to be assessed. The specific execution step is S32.

[0120] S32: Construct a fuzzy inference system and use this fuzzy inference system to assess the risk level of the set of input feature vectors.

[0121] For a fuzzy inference system, the first step is to define a fuzzy set, then specify the fuzzy rules for the fuzzy set, estimate the results through defuzzification, and finally output the predicted inference structure. The execution of this embodiment depends on the following example:

[0122] import skfuzzy as fuzz

[0123] def risk_assessment(temp, vibration):

[0124] # Define fuzzy set

[0125] temp_risk = fuzz.trimf(temp, [70, 85, 100])

[0126] vib_risk = fuzz.gbellmf(vibration, 2, 3, 10)

[0127] # Fuzzy rules

[0128] rule1 = np.fmax(temp_risk, vib_risk)

[0129] # Deblurring (Center of Gravity Method)

[0130] return fuzz.defuzz(np.linspace(0,10), rule1, 'centroid')

[0131] After the above fuzzy inference system, risk level assessments for different working conditions and corresponding parameters can be obtained. Based on the risk level assessment, the allocation of dynamic weights is determined, specifically in step S33.

[0132] S33: Based on the risk level assessment results, the dynamic weight allocation of each parameter is obtained by using a nonlinear mapping method.

[0133] The nonlinear mapping based on risk level includes the following formula:

[0134]

[0135] in, This represents the control weight of the i-th parameter. This represents the risk score of the i-th parameter. This is a preset constant value used as an adjustment factor to control the degree of "sharpening" in weight allocation. This indicates that the risk score is weighted by an index. The variable is used to calculate the sum of the weighted scores by iterating through the four parameters.

[0136] The dynamic weights are redistributed by inputting the inference results, and the control instructions are dynamically adjusted based on the redistribution results, specifically in step S40.

[0137] S40: Generate a set of dynamic control commands for the operation of the coal mill based on the dynamic weight vector and the set of feature vectors, verify and compare them, output the predicted equipment simulation status, and perform automatic adjustment or early warning based on the equipment simulation status.

[0138] Specifically, step S40 includes the following steps:

[0139] S41: Construct a multi-model collaborative controller and obtain control commands for various parameters based on the geometry of the input feature vectors and the dynamic weight allocation.

[0140] The multi-model collaborative controller includes at least: a temperature controller, a pressure controller, a vibration controller, and a gas controller. The temperature controller dynamically adjusts the temperature of the primary hot air based on pressure and vibration conditions. The pressure controller dynamically adjusts the internal gas pressure based on equipment vibration and internal gas filling status. The gas controller determines the filling gas status based on temperature and pressure conditions. The vibration controller adjusts the injected gas volume or issues an early warning based on equipment vibration.

[0141] This is because temperature changes cause synchronous changes in the internal pressure and vibration of the equipment. Pressure changes, in turn, affect the amount of gas injected. For example, if the internal pressure is too high, the gas injection needs to be paused or slowed down. If the gas controller continues to pressurize when the internal temperature is too high, it may cause the gas to rupture. Furthermore, the amount and speed of gas injection also need to be considered when the internal pressure is too high. Similarly, the vibration controller adjusts the amount of gas injected based on the vibration of the equipment. If the problem cannot be solved by control commands, an alarm will be issued directly.

[0142] The specific execution instructions need to be implemented through the coordination between various controllers. Active control of temperature and vibration can be achieved through an MPC controller, while active control of pressure can be achieved through a fuzzy PID controller, and gas control needs to be achieved through a DRC dynamic range controller.

[0143] The control of the MPC controller depends on the following formula:

[0144]

[0145] in, This indicates that the optimization is performed by minimizing the control input sequence u. This represents the predicted temperature value at time t. This indicates the set temperature value. This indicates the amount of change in the control input. This is the regularization coefficient, used to balance the weights of tracking error and changes in control input.

[0146] An example of a fuzzy PID controller is shown below:

[0147] class FuzzyPID:

[0148] def __init__(self):

[0149] self.Kp = 1.0

[0150] self.Ki = 0.1

[0151] def update(self, e, de):

[0152] # Adjusting parameters for fuzzy rule tables

[0153] if abs(e) > 5:

[0154] self.Kp *= 1.2

[0155] return self.Kp*e + self.Ki*integral(e)

[0156] Threshold state control is achieved by adjusting the adjustment parameters in the fuzzy rule table. When the threshold is exceeded, the dynamic adjustment function is activated. Temperature correction control for abnormal vibration depends on the following formula:

[0157]

[0158] in, This indicates the corrected temperature value. This indicates the original temperature setting value. This represents the vibration temperature compensation coefficient. The effective value of vibration velocity is a key indicator for measuring vibration intensity. This indicates that the vibration has exceeded the limit. If the temperature exceeds the limit, the temperature setting needs to be lowered; otherwise, no adjustment is needed.

[0159] For gas control, the design of the DRC dynamic range controller is as follows:

[0160]

[0161] in, This indicates the amount of nitrogen injected. This represents the baseline nitrogen flow rate under medium risk conditions. This indicates the maximum design flow rate of the nitrogen system and the risk factor. .

[0162] The above formula indicates that during analysis, if the risk coefficient is below 0.5, it indicates a low-risk state where no inert gas injection is needed; if the risk coefficient is between 0.5 and 1, it indicates a medium-risk state where an appropriate amount of inert gas needs to be injected; and if the risk coefficient is above 1, the maximum flow injection method should be activated to quickly suppress combustion and explosion conditions. The risk coefficient R is calculated by dividing the oxygen content by 12% and the carbon monoxide content by 300 ppm, and then multiplying the two values.

[0163] S42: Perform physical simulation based on control commands to obtain prediction results, and use an LSTM model to predict the remaining lifespan. Output the predicted equipment simulation status. If the equipment simulation is in a normal state, automatically adjust or do not modify according to the control commands. If the equipment simulation is in a risky state, automatically adjust or issue a risk warning according to the control commands.

[0164] The control commands generated by the aforementioned scheme are used to perform physical simulations, and the lifetime is predicted based on the simulation results. An example of an LSTM model is shown below:

[0165] class LSTM_Predictor(nn.Module):

[0166] def __init__(self):

[0167] super().__init__()

[0168] self.lstm = nn.LSTM(input_size=4, hidden_size=32)

[0169] self.fc = nn.Linear(32, 1)

[0170] def forward(self, x):

[0171] out, _ = self.lstm(x)

[0172] return self.fc(out[:, -1, :])

[0173] The above code is used to predict the lifetime based on the simulation results. In this embodiment, the system's input and output states are as follows:

[0174] When the input disturbance is: the inlet temperature jumps from 300℃ to 340℃, the effective value of the vibration velocity increases from 5.0mm / s to 8.5mm / s.

[0175] The system's response output may be: Timeline Control Action Parameter changes t=0s Temperature gradient exceeded limit detected Start MPC control optimization t=2s The opening of the air conditioning valve has been increased by 25%. The rate of temperature increase decreased by 60%. t=5s Vibration correlation analysis triggers load reduction Coal supply reduced to 80% t=8s Predicting temperature overshoot risk Inject inert gas in advance

[0176] Example 2

[0177] For compatibility with Example 1, please refer to... Figure 2Furthermore, a system for monitoring and early warning control of coal mill operation is provided, which employs the method for monitoring and early warning control of coal mill operation as described in any one of the embodiments described in Example 1, including: a data processing module, a feature extraction module, a dynamic parameter identification module, a coordinated control module, and a data verification and prediction module.

[0178] Specifically, the data processing module preprocesses the input multimodal data; the feature extraction module extracts features from the processed standardized data and constructs a cross-parameter correlation matrix to obtain a set of feature vectors for the multimodal data; the dynamic parameter identification module classifies operating conditions based on the set of feature vectors for the multimodal data, assesses the risk level of each operating condition, and obtains the dynamic weight vector corresponding to the risk level; the coordination control module generates dynamic control commands for the operation of the coal mill based on the dynamic weight vectors and the set of feature vectors; and the data verification and prediction module verifies and compares the input dynamic control commands with the equipment operating status and outputs the predicted equipment simulation status.

[0179] By standardizing the input data, different types of data can be processed in a standardized and unified form. The data processing module preprocesses multimodal data in at least the following ways: data format standardization, outlier filtering, and normalization.

[0180] After data processing, time-frequency domain features and mutual information are extracted using a feature extraction module. Specifically, the feature extraction module includes a time-domain feature extraction module, a frequency-domain feature extraction module, and a cross-parameter correlation analysis module. The time-domain feature extraction module extracts the time-domain features of the standardized data, the frequency-domain feature extraction module extracts the frequency-domain features of the standardized data, and the cross-parameter correlation analysis module calculates the mutual information between various parameters during the operation of the coal mill. The characteristics of the signal are reflected through the feature vectors in the time-frequency domain, and the implicit relationships between these feature vectors are obtained by calculating the mutual information between them.

[0181] After obtaining the implicit relationships, dynamic weights are allocated based on the influence between these relationships. Specifically, the dynamic parameter identification module includes a working condition classification module, a risk assessment module, and a dynamic weight allocation module. The working condition classification module is used to classify the working condition parameters according to the set of input feature vectors. The risk assessment module is used to assess the risk level based on the classified working condition parameters. The dynamic weight allocation module is used to quickly locate other parameters with the strongest correlation to abnormal parameters based on mutual information and risk level, and dynamically adjust the control weights.

[0182] First, the operating conditions of the input feature vectors need to be classified, and a risk assessment needs to be performed on the parameters for each operating condition. Finally, dynamic weights are allocated based on the mutual influences and feature parameters. After the dynamic weight allocation is completed, corresponding control commands are generated according to the importance of the weights and the corresponding parameters. These control commands need to be physically simulated internally, and life prediction is performed based on the simulation results to avoid problems such as incorrect estimation. During the prediction, the system can perform early fault diagnosis based on the prediction results and generate corresponding command schemes. Users can choose to execute automatically or after review. When a fault cannot be adjusted by modifying parameters, an alarm can be issued directly to promptly remind operators to handle the situation manually.

[0183] The above embodiments are merely illustrative of the technical solutions of the present invention and are not intended to limit it. Anyone skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for monitoring and early warning control of coal mill operation, characterized in that, Includes the following steps: S10: Preprocess the collected data; S20: Extract multimodal data features from the preprocessed data and calculate mutual information using cross-parameter correlation analysis. S30: Classify the operating conditions of the coal mill based on the extracted feature set, and infer and predict the risks of different operating conditions by constructing a fuzzy inference system. Finally, perform dynamic weight allocation based on risk level and mutual information. S40: Generate a set of dynamic control commands for the operation of the coal mill based on the dynamic weight vector and the set of feature vectors, verify and compare them, output the predicted equipment simulation status, and perform automatic adjustment or early warning based on the equipment simulation status.

2. The method for monitoring and early warning control of coal mill operation according to claim 1, characterized in that: Step S10 includes the following steps: S11: Uses the IEEE 1588 precision time protocol to synchronize the network clock, and converts the input data into floating-point values ​​through the Modbus protocol; S12: 3 based on sliding window The criteria filter out outliers in the transformed data; S13: Use the improved Z-score normalization algorithm to normalize the filtered data and output the normalized data.

3. The method for monitoring and early warning control of coal mill operation according to claim 1, characterized in that: Step S20 includes the following steps: S21: Calculate the statistics of the data in the input window and extract the time-domain feature vector; S22: Perform a fast Fourier transform on the time-domain eigenvectors and then perform wavelet packet energy decomposition to obtain the frequency-domain eigenvectors; S23: Calculate mutual information for various parameters during the operation of the coal mill.

4. The method for monitoring and early warning control of coal mill operation according to claim 1, characterized in that: Step S30 includes the following steps: S31: Based on the set of input feature vectors, the operating states are divided using the improved K-means++ algorithm to obtain the working condition classification data of each parameter; S32: Construct a fuzzy inference system and use this fuzzy inference system to assess the risk level of the set of input feature vectors; S33: Based on the risk level assessment results, the dynamic weight allocation of each parameter is obtained by using a nonlinear mapping method.

5. The method for monitoring and early warning control of coal mill operation according to claim 1, characterized in that: Step S40 includes the following steps: S41: Construct a multi-model collaborative controller and obtain control commands for various parameters based on the geometry of the input feature vectors and the dynamic weight allocation. S42: Perform physical simulation based on control commands to obtain prediction results, and use an LSTM model to predict the remaining lifespan. Output the predicted equipment simulation status. If the equipment simulation is in a normal state, automatically adjust or do not modify according to the control commands. If the equipment simulation is in a risky state, automatically adjust or issue a risk warning according to the control commands.

6. The method for monitoring and early warning control of coal mill operation according to claim 5, characterized in that: The multi-model collaborative controller includes at least: a temperature controller, a pressure controller, a vibration controller, and a gas controller; The temperature controller is used to dynamically adjust the temperature of the primary hot air based on pressure and vibration conditions. The pressure controller is used to dynamically adjust the internal air pressure based on equipment vibration and internal gas filling conditions. The gas controller is used to determine the filling gas condition based on temperature and pressure conditions. The vibration controller is used to adjust the injected gas volume or issue an early warning based on equipment vibration conditions.

7. A system for monitoring and early warning control of coal mill operation, wherein the method for monitoring and early warning control of coal mill operation as described in any one of claims 1-6 is used for control, characterized in that, include: The data processing module is used to preprocess the input multimodal data; The feature extraction module is used to extract features from the processed standardized data and construct a cross-parameter correlation matrix to obtain a set of feature vectors for multimodal data. The dynamic parameter identification module is used to classify operating conditions based on the feature vector set of multimodal data, assess the risk level of operating conditions, and obtain the dynamic weight vector corresponding to the risk level. The coordination control module is used to generate dynamic control commands for the operation of the coal mill based on the dynamic weight vector and the feature vector set. The data verification and prediction module verifies and compares the input dynamic control commands and the equipment operating status, and outputs the predicted equipment simulation status.

8. The system for monitoring and early warning control of coal mill operation according to claim 7, characterized in that: The data processing module preprocesses multimodal data in at least the following ways: data format standardization, outlier filtering, and normalization.

9. The system for monitoring and early warning control of coal mill operation according to claim 7, characterized in that: The feature extraction module includes a time-domain feature extraction module, a frequency-domain feature extraction module, and a cross-parameter correlation analysis module; The time-domain feature extraction module is used to extract the time-domain features of the standardized data, the frequency-domain feature extraction module is used to extract the frequency-domain features of the standardized data, and the cross-parameter correlation analysis module is used to calculate the mutual information between various parameters during the operation of the coal mill.

10. The system for monitoring and early warning control of coal mill operation according to claim 7, characterized in that: The dynamic parameter identification module includes a working condition classification module, a risk assessment module, and a dynamic weight allocation module; The operating condition classification module is used to classify operating condition parameters based on the set of input feature vectors. The risk assessment module is used to assess the risk level based on the classified operating condition parameters. The dynamic weight allocation module is used to quickly locate other parameters with the strongest correlation to abnormal parameters based on mutual information and risk level, and dynamically adjust the control weights.

Images