A coal mill hydraulic loading force automatic control method and system

Abstract

The present invention discloses a method and system for automatically controlling the hydraulic loading force of a coal mill, and specifically relates to the field of hydraulic control technology, and is used for quickly adjusting the load of a coal mill, making the operation of the coal mill stable, and effectively reducing the energy consumption of pulverizing. The present invention obtains multidimensional data during the operation of the coal mill, determines the change in the load and operating parameters of the coal mill unit, and corrects the system loading force in real time according to the change in the unit load and the feedback of the operating parameters, and controls the response amplitude of the loading force through an integral equation. It has wide adaptability to different types of coal, and uses the key parameters of the coal mill operation state as the control conditions of the loading force. The control method of the loading force is more reasonable, and the system enters the automatic control mode after use without manual intervention. The operating difficulty of the operating personnel is simplified, and the operating accidents caused by the operator's misjudgment are reduced. The control system has stable performance, fast response speed to changes in operating conditions, and is suitable for long-term continuous work.

Description

Technical Field

[0001] The present invention relates to the field of hydraulic control technology, and more specifically, to a method and system for automatically controlling hydraulic loading force of a coal mill. Background Art

[0002] Coal mill is a mechanical device used to grind coal into coal powder. It is widely used in power, metallurgy, chemical industry, cement and other industries. Its main function is to grind large pieces of coal into fine coal powder particles in coal-fired boilers or other coal-fired devices to improve the combustion efficiency of coal. The performance of coal mill directly affects the boiler combustion efficiency, system energy consumption and emission level.

[0003] With the continuous improvement of industrial automation level, coal mill, as a key energy consumption equipment in power plants and metallurgical industries, has gradually attracted widespread attention for its operating efficiency and energy consumption management. In actual operation, the working state of coal mill is affected by multiple parameters, such as hydraulic loading force, coal load, system pressure, temperature and coal mill vibration value, etc. These parameters are not only related to each other, but also jointly affect the overall operating performance and energy consumption level of coal mill.

[0004] Deficiencies in the existing technology: The existing coal mill hydraulic loading force control system has many deficiencies in terms of energy consumption optimization and operating efficiency. In dealing with the lack of correlation between multiple parameters, usually, the hydraulic loading force, as the main control parameter of the coal mill, is often adjusted through a preset fixed control strategy. However, this fixed strategy lacks comprehensive consideration of real-time working conditions and other related parameters, resulting in the system being unable to dynamically adjust the hydraulic loading force according to the actual operating status. For example, when the coal mill is in a low-load or no-load state, the system still maintains a high hydraulic loading force and fails to make effective adjustments according to actual needs, which leads to an abnormal increase in energy consumption. It is impossible to effectively identify the optimal control strategy under different working conditions, which may cause energy consumption to gradually increase when the equipment operating status deteriorates, affecting overall efficiency. Summary of the invention

[0005] In order to overcome the above defects of the prior art, the present invention provides a coal mill hydraulic loading force automatic control method and system thereof to solve the problem of excessive energy consumption in the coal mill hydraulic loading force control system under low load or no-load conditions in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for automatically controlling hydraulic loading force of a coal mill comprises the following steps:

[0008] Acquire multi-dimensional data during the operation of the coal mill, determine the changes in the load and operating parameters of the coal mill unit, and modify the system loading force in real time according to the changes in the unit load and the feedback of the operating parameters, and control the loading force response amplitude through the integral equation;

[0009] The integral equation is expressed as:

[0010] , where L(t) is the loading force at the current time t, ΔM(t) is the change in the coal feed M(t), ΔQ(t) is the change in coal seam thickness, ΔR(t) is the change in the resistance of the grinding area, and V(t) is the ventilation volume of the coal mill. , , is the integral coefficient, corresponding to the influence weights of coal feed rate, coal seam thickness and grinding zone resistance, , are differential coefficients, corresponding to the weights of the response speed of the coal feed rate and ventilation volume changes to the loading force.

[0011] Acquire multidimensional data during the operation of the coal mill, perform nonlinear transformation and standardization, and generate high-dimensional feature vectors through time delay embedding;

[0012] The kernel density of the extracted multi-dimensional feature vector is estimated to construct the joint probability distribution between the features, and the marginal probability distribution is obtained by normalization as the basis for calculating the mutual information entropy.

[0013] The mutual information entropy between features is calculated based on the obtained joint probability distribution, and a feature dependency network is constructed. The mutual information entropy matrix is ​​dynamically adjusted using an adaptive update algorithm, and the energy consumption sensitivity is determined based on the mutual information entropy difference.

[0014] Based on the mutual information entropy matrix, a variable coupling strength matrix is ​​constructed to screen out key coupling relationships and generate a variable coupling network to identify abnormal energy consumption patterns under low load or no-load conditions;

[0015] The variable coupling relationship is adjusted, the abnormal energy consumption trend is predicted according to the abnormal energy consumption mode, and the hydraulic loading force control parameters are dynamically adjusted in combination with the feedback mechanism.

[0016] In a preferred embodiment, multidimensional data is obtained during the operation of the coal mill, nonlinear transformation and standardization are performed, and a high-dimensional feature vector is generated by time delay embedding. The specific process is as follows:

[0017] Multi-dimensional data collection of time series data of key operating parameters, including coal seam thickness, coal feed rate, grinding zone resistance, coal mill ventilation volume, and coal mill vibration value;

[0018] Temporal alignment of multidimensional data using spline interpolation;

[0019] Use boundary detection and DBSCAN algorithm to remove data points;

[0020] The multidimensional data is transformed nonlinearly using a combined transformation function and standardized using the minimum and maximum normalization method;

[0021] Based on the standardized multidimensional data, time delay embedding expansion is performed, the time delay step and embedding dimension are set, and each standardized data is expanded into a delayed embedding vector and used as a high-dimensional feature vector.

[0022] In a preferred embodiment, kernel density estimation is performed on the extracted multidimensional feature vectors to construct a joint probability distribution between features, and the marginal probability distribution obtained by normalization is used as the basis for calculating the mutual information entropy. The specific process is as follows:

[0023] Dynamic interval normalization of high-dimensional feature vectors;

[0024] Adaptive Gaussian kernel function is used to estimate the probability density of high-dimensional feature vectors processed by dynamic interval normalization;

[0025] Construct a high-dimensional feature matrix based on the high-dimensional feature vector, calculate the joint probability distribution of the high-dimensional feature matrix, and calculate the joint probability distribution between different feature dimensions;

[0026] After the joint probability distribution is calculated, a separate kernel density calculation is performed on the marginal probability distribution of each high-dimensional feature vector.

[0027] In a preferred embodiment, the mutual information entropy between the features is calculated based on the obtained joint probability distribution, and a feature dependency network is constructed at the same time. The mutual information entropy matrix is ​​dynamically adjusted using an adaptive update algorithm, and the energy consumption sensitivity is determined based on the mutual information entropy difference. The specific process is as follows:

[0028] The joint probability distribution and marginal probability distribution between multi-dimensional features are obtained through non-parametric kernel density estimation. The mutual information entropy between each feature variable is calculated based on the joint probability distribution and marginal probability distribution of the probability distribution to determine the information dependence of the features.

[0029] Define the feature dependency matrix, and construct a feature dependency network graph based on the feature dependency matrix. The nodes in the feature dependency network graph represent the features, and the edge weights are the mutual information entropy values ​​between the corresponding features.

[0030] A high-order mutual information entropy matrix is ​​constructed to compare the changes in mutual information entropy under low load or no-load conditions with those under normal load conditions, and the sensitivity of each feature to energy consumption is calculated.

[0031] In a preferred embodiment, a variable coupling strength matrix is ​​constructed based on the mutual information entropy matrix, key coupling relationships are screened out and a variable coupling network is generated. The specific process is as follows:

[0032] The mutual information entropy and high-order mutual information entropy between each characteristic variable are used to perform variable coupling analysis, and the high-order mutual information entropy matrix is ​​normalized;

[0033] Based on the normalized high-order mutual information entropy matrix, a variable coupling strength matrix is ​​constructed to determine the coupling degree between characteristic variables;

[0034] The coupling strength matrix is ​​used to construct a variable coupling network diagram, where the nodes are characteristic variables and the edge weights are the coupling strengths between the corresponding characteristic variables.

[0035] In a preferred embodiment, the abnormal energy consumption pattern under low load or no load state is identified, and the specific process is as follows:

[0036] Determine the initial conditions of nodes and edges, determine the network nodes from the feature matrix, and initialize the non-zero elements in the coupling strength matrix as the initial weights of the edges;

[0037] Filter key coupling relationships, set a threshold based on the edge weight, and retain edges with weights greater than the threshold;

[0038] In the constructed variable coupling network, energy consumption abnormal pattern recognition is performed, and the probability distribution of energy consumption abnormal pattern under low load or no-load state is calculated according to the variable coupling intensity matrix;

[0039] The abnormal energy consumption pattern is determined based on the probability distribution of the variable coupling strength and the abnormal energy consumption pattern.

[0040] In a preferred embodiment, the abnormal energy consumption trend is predicted according to the abnormal energy consumption mode, and the hydraulic loading force control parameters are dynamically adjusted in combination with the feedback mechanism. The specific steps are as follows:

[0041] Identify low load or no load status during operation of the coal mill according to a set load threshold;

[0042] Perform preliminary anomaly detection on the energy consumption in the current state according to the sensitivity coefficient, identify the abnormal energy consumption under low load or no-load state, and define the energy consumption abnormal pattern function after the preliminary anomaly detection, and mark the energy consumption abnormal pattern intensity exceeding the abnormal pattern threshold as significant energy consumption abnormality;

[0043] Build a feedback control mechanism to dynamically adjust the hydraulic loading force of the system.

[0044] A coal mill hydraulic loading force automatic control system, used to implement the above-mentioned coal mill hydraulic loading force automatic control method, comprising:

[0045] The data processing module is used to obtain multidimensional data during the operation of the coal mill, perform nonlinear transformation and standardization, generate high-dimensional feature vectors through time delay embedding, perform kernel density estimation on the extracted multidimensional feature vectors, construct the joint probability distribution between features, and obtain the marginal probability distribution through normalization as the basis for calculating the mutual information entropy;

[0046] Energy consumption analysis module, which is used to calculate the mutual information entropy between each feature based on the obtained joint probability distribution, build a feature dependency network, and use an adaptive update algorithm to dynamically adjust the mutual information entropy matrix to determine the energy consumption sensitivity based on the mutual information entropy difference;

[0047] Energy consumption identification module, which is used to construct a variable coupling strength matrix based on the mutual information entropy matrix, screen out key coupling relationships and generate a variable coupling network, and identify abnormal energy consumption patterns under low load or no-load conditions;

[0048] The control module is used to adjust the variable coupling relationship, predict the abnormal energy consumption trend according to the abnormal energy consumption mode, and dynamically adjust the hydraulic loading force control parameters in combination with the feedback mechanism.

[0049] Technical effects and advantages of the present invention:

[0050] The present invention uses multivariate coupling analysis and non-parametric kernel density estimation methods to accurately capture the complex relationship between various parameters in the hydraulic loading force control system of the coal mill, especially under low load or no-load conditions, and can identify and analyze the key factors that lead to abnormal energy consumption in real time. By constructing a high-dimensional feature matrix and a mutual information entropy matrix, the problem of insufficient multi-parameter correlation in traditional control systems is effectively solved, ensuring the control accuracy and response speed of the system under different working conditions. With the help of an efficient variable coupling intensity matrix and an adaptive update mechanism, the control strategy can be dynamically adjusted and the hydraulic loading force can be optimized in real time, thereby avoiding unnecessary energy waste.

[0051] In addition, the present invention introduces an energy consumption trend prediction and feedback control mechanism based on the LSTM model, which can identify potential energy consumption anomalies in advance and adjust control parameters in real time through a feedback loop to ensure that the system maintains the best energy efficiency during operation. Compared with traditional fixed control strategies, the present invention significantly improves the system's adaptability and intelligence level and reduces the need for manual intervention. Through this innovative technology, the present invention not only improves the overall operating efficiency of the coal mill hydraulic loading force control system, but also significantly reduces energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The present invention is a flow chart of a method for automatically controlling the hydraulic loading force of a coal mill.

[0053] Figure 2The present invention is a structural schematic diagram of a coal mill hydraulic loading force automatic control system.

[0054] Figure 3 The present invention is a process principle diagram of a coal mill hydraulic loading force automatic control system.

[0055] The markings in the figure are: 1-coal feed rate, 2-grinding zone resistance, 3-coal seam thickness, 4-mill ventilation volume, 5-mill vibration value, 6-grinding roller, 7-grinding disc. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] Example 1: Figure 1 As shown, a method for automatically controlling the hydraulic loading force of a coal mill comprises the following steps:

[0058] Acquire multidimensional data during the operation of the coal mill, perform nonlinear transformation and standardization, generate high-dimensional feature vectors through time delay embedding and recursive nonlinear mapping, and extract key features representing the dynamic behavior of the system;

[0059] Perform kernel density estimation on the extracted multi-dimensional feature vectors, construct the joint probability distribution between features, and use normalization processing to obtain the marginal probability distribution as the basis for calculating mutual information entropy;

[0060] The mutual information entropy between each feature is calculated based on the obtained joint probability distribution, and extended to multivariate mutual information. At the same time, a feature dependency network is constructed, and the mutual information entropy matrix is ​​dynamically adjusted using an adaptive update algorithm to determine the changes in information dependency.

[0061] Based on the mutual information entropy matrix, a variable coupling strength matrix is ​​constructed to screen out key coupling relationships and generate a variable coupling network to identify abnormal energy consumption patterns under low load or no-load conditions;

[0062] The variable coupling relationship is adjusted, the abnormal energy consumption trend is predicted according to the abnormal energy consumption mode, and the hydraulic loading force control parameters are dynamically adjusted in combination with the feedback mechanism.

[0063] During the normal operation of the coal mill, the hydraulic loading system outputs the loading force from the hydraulic cylinder, and transmits the force to the grinding roller 6 through the loading transmission mechanism. The grinding disc 7 rotates to drive the grinding roller 6 to rotate. The grinding roller 6 presses down on the grinding disc 7. The grinding roller 6 is in motion with the operation of the coal mill. The grinding disc 7 drives the grinding roller 6 to rotate. As the grinding roller 6 breaks the coal and new coal blocks enter, the grinding roller 6 moves up and down in the vertical direction at the same time.

[0064] The control system needs to collect real-time operation data of the coal mill during operation, including coal seam thickness 3, coal feed 1, grinding zone resistance 2, coal mill ventilation volume 4, coal mill vibration value 5, etc., and adjust the loading force output value according to the changes of the above parameters;

[0065] The coal seam thickness 3 refers to the thickness of the coal ground between the grinding roller 6 and the grinding disc 7 when the coal mill is running, that is, the distance between the grinding roller 6 and the grinding disc 7 during operation. The coal seam thickness 3 is controlled to maintain an appropriate coal seam thickness 3 during operation. A large coal seam thickness 3 will result in large system resistance and insufficient coal mill output. A small coal seam thickness 3 will result in vibration of the coal mill and prone to failure. The corresponding relationship with the loading force is that the greater the loading force, the smaller the coal seam thickness 3;

[0066] Coal feed 1 refers to the weight of raw coal fed into the coal mill by the coal feeder per unit time, i.e., the coal mill output. The maximum coal feed that can continuously and stably operate the coal mill is the maximum output of the coal mill. The larger the coal feed 1, the greater the amount of coal entering the coal mill per unit time, and the greater the loading force that needs to be applied to the grinding roller 6;

[0067] The resistance 2 of the grinding area refers to the pressure difference between the inlet of the nozzle ring and the middle and upper part of the grinding roller 6. The factor affecting the resistance of the grinding area is the coal powder concentration in this area. The greater the coal powder concentration, the greater the resistance, and the lower the coal powder concentration, the lower the resistance. The coal powder concentration in the grinding area decreases with the increase of the loading force.

[0068] Coal mill ventilation volume 4 refers to the mass flow of desiccant when the coal mill is running. The desiccant can be hot air or a mixture of flue gas and hot air from the tail of the boiler. The air-coal ratio of the pulverizing system is calculated based on the coal mill ventilation volume 4, that is, the coal mill ventilation volume 4 divided by the coal feed volume 1. The larger the air-coal ratio, the smaller the loading force output;

[0069] The vibration value 5 of the coal mill refers to the vibration amplitude of the lower part of the coal mill frame. Long-term high vibration value will cause fatigue of the mechanical parts of the coal mill, causing equipment damage and affecting the continuous stability of the equipment. Excessive loading force will cause the vibration value 5 of the coal mill to increase;

[0070] Step 1: Extract and preliminarily construct multivariate features. Obtain time series data of multiple key operating parameters from the coal mill hydraulic loading force automatic control system (referred to as the system) as multidimensional data. Each key operating parameter is collected at a high frequency and saved as time series data, which is set as ,in, represents the nth operating parameter at time t. The multiple key operating parameters involved in this step may include coal seam thickness 3, coal feed rate 1, grinding zone resistance 2, coal mill ventilation rate 4, coal mill vibration value 5, etc.;

[0071] During the operation of the coal mill, multiple key parameters of the hydraulic system interact with each other, and the changes between the parameters directly affect the energy consumption. Therefore, by collecting the time series data of these parameters at a high frequency, the operating status of the system can be fully reflected, which is convenient for subsequent feature extraction and analysis.

[0072] The coal pulverizing process is usually an auxiliary system in the power, chemical, steel and other industries. It needs to be adjusted according to the real-time operating parameters of the main engine. The coal feed rate 1 changes with the load of the main engine. The physical properties of the raw coal also change with the different coal quality entering the mill. In addition, the wind system generally uses one machine to share the wind for several coal mills at the same time. The start and stop of the coal mill under different working conditions, the adjustment of the cold and hot air doors will affect the resistance of the coal mill system and the changes in the wind system. The hydraulic loading control system adjusts the loading force of the coal mill in real time according to the changes in the above five operating parameters;

[0073] The operating parameters affect each other. According to the different parameter variable amplitudes, the system calculates the amplitude of the increase and decrease of the loading force through the integral equation. The response speed of the operating parameters after the parameter change is calculated through the differential equation, and then the loading force assignment is corrected to ensure that the given value of the load quickly reaches the most ideal set value. At the same time, according to the operating characteristics of the coal mill, the performance of the coal mill is limited at high load, and the operating stability of the equipment is limited at low load. Different calculation methods are used for the increasing control and decreasing control of the loading force, and the adjustment of the loading force is more scientific and more in line with the actual operating conditions of the coal mill. For example, the expression formula is:

[0074] , where L(t) is the loading force at the current time t, ΔM(t) is the change in coal feed 1, ΔQ(t) is the change in coal seam thickness 3M(t), ΔR(t) is the change in grinding zone resistance 2, V(t) is the coal mill ventilation volume 4, , , is the integral coefficient, corresponding to the influence weights of coal feed 1, coal seam thickness 3, and grinding zone resistance 2, respectively. , are differential coefficients, corresponding to the weights of the response speed of the loading force to the changes in the coal feed rate 1 and the coal mill ventilation rate 4;

[0075] It should be noted that the relevant weights and coefficients are not fixed, but are adjusted and determined according to the actual scenario.

[0076] like Figure 3As shown in the figure, the adjustment range of the loading force is calculated according to the changes in the coal feed rate 1, the coal seam thickness 3, the grinding area resistance 2, the coal mill ventilation volume 4, and the coal mill vibration value 5. The change of each parameter is accumulated through the integral equation to calculate its long-term influence on the loading force; the response speed after the parameter change is calculated through the differential equation, and the loading force assignment is corrected to ensure that the adjustment of the loading force is more timely and the response is faster under different working conditions;

[0077] The amplitude of the loading force is increased through integral calculation to cope with the load increase. At the same time, the adjustment speed of the loading force is corrected through the differential equation to ensure that the coal mill can quickly reach the ideal loading state when the load increases without overloading or delayed response.

[0078] Since the data acquisition frequencies of different sensors may be different, it is necessary to perform time synchronization on the collected multi-dimensional data. Set a unified time step Δt, interpolate and align all parameter data, and use spline interpolation for time alignment. to Time alignment is performed within the time period, and the formula is as follows:

[0079] ,in, represents parameter data that has been interpolated and aligned to a uniform time step, is the interpolation function, Represents the kth moment. Time synchronization ensures the consistency of the time dimension between different parameters, providing a basis for subsequent data analysis;

[0080] The difference in sensor data collection frequency will lead to misalignment of time series data. Time alignment of data through spline interpolation can ensure the consistency of data of different parameters at the same time node, thereby avoiding time deviation problems in subsequent analysis.

[0081] Data cleaning and outlier processing,There may be noise or outliers in the aligned time series data. These outliers may come from sensor failure, data transmission errors or external environmental interference. First, perform boundary detection on each parameter data and set the threshold range. , to exclude physically unreasonable data points. Then, the density-based outlier detection (DBSCAN) algorithm is used to identify and remove outliers. The formula is as follows:

[0082] ,in, is the density threshold, is the cluster center, represents the i-th operating parameter at time t. This method is used to remove abnormal data that deviates greatly from other data points;

[0083] Through boundary detection and DBSCAN algorithm, unreasonable data points can be effectively eliminated to ensure the accuracy and consistency of data;

[0084] To enhance the nonlinear feature expression of the data, the cleaned multidimensional data is transformed nonlinearly, and a combined transformation function is used. The nonlinear transformation is to enhance the ability to express complex relationships in the data so that it can better capture the nonlinear dynamic behavior of the system in subsequent analysis. By combining the transformation function, the original data is mapped to a new nonlinear space, the feature expression ability of the data is improved, and the data is mapped to a higher-dimensional nonlinear space. The formula is:

[0085] ,in, Expressed as a logarithmic smoothing transformation, is used to capture the periodic characteristics of the data. A nonlinear suppression factor is introduced, α is the adjustment coefficient, and is used to adjust the size of the data suppression factor;

[0086] The magnitude of data with or without different parameters may vary greatly. Direct use of these data will cause certain parameters to dominate the analysis and mask the influence of other parameters. Through standardization, the data of all parameters are scaled to the same magnitude range, which can eliminate the influence of such magnitude differences and ensure that each parameter has equal weight in subsequent analysis.

[0087] Standardize the multidimensional data to avoid the impact of the magnitude difference of different parameter data, and standardize the transformed data. Use the minimum and maximum normalization method to scale the data of each parameter to the interval [0, 1]. The standardization process ensures that the data of each parameter is within the same magnitude range, which is helpful for data processing and calculation in subsequent analysis.

[0088] To extract dynamic features from multidimensional data, we use sliding window technology to extract dynamic features based on standardized multidimensional data. We set the window size W and calculate the accumulation and change rate of the data in the window at each time step t. The formula is as follows: , where this dynamic feature It reflects the cumulative trend and change range of the parameters in the sliding window, capturing the short-term dynamic changes of the system. represents the standardized data, which is used to eliminate the magnitude difference; k represents the time step increment in the sliding window, which is used to traverse the data points in the window;

[0089] Through the above steps, the operation parameters can be effectively collected and preprocessed. This process ensures the quality, accuracy and nonlinear feature expression ability of the data.

[0090] After preprocessing and standardization, the multidimensional data Perform time delay embedding expansion. The main idea of ​​time delay embedding is to expand the original time series data into a high-dimensional feature vector and form a high-dimensional feature space by introducing historical information.

[0091] Set the time delay step τ and embedding dimension d, and expand the standardized data of each parameter into a delayed embedding vector as follows: , where τ controls the time step of the delay and d controls the dimension of the embedding. The delayed embedding vector By introducing the data of the previous d-1 moments, the temporal variation pattern of parameter data can be captured;

[0092] Time delay embedding is an important method in time series analysis. It aims to expand the original one-dimensional time series into a high-dimensional feature vector by introducing historical information, so as to better reflect the dynamic behavior of the system. This step ensures that the law of parameter evolution over time can be captured in subsequent analysis. Especially under low load or no-load conditions, the changes in system characteristics are more hidden, and delay embedding can effectively reveal these changes.

[0093] Similarly, to enhance the nonlinear expression ability of features, recursive nonlinear mapping can also be performed. This mapping generates new high-dimensional features in a recursive way to capture more complex system dynamic characteristics. The mapping function can be designed as a combination of quadratic mapping and periodic function.

[0094] The recursive nonlinear mapping is designed to enhance the nonlinear expression ability of delayed embedding features so that the generated features can better reflect the complex dynamic behavior of the system. The secondary mapping introduces high-order feature interactions, and the periodic function captures the periodic changes in the system.

[0095] Since the feature dimension after expansion is high, sparse coding method can be used to reduce the feature dimension. Sparse coding is an effective feature selection method. By introducing sparsity constraints, it ensures that only the most representative features for energy consumption analysis are retained. This step compresses the feature dimension to a controllable range, reduces the computational complexity, and provides for subsequent entropy calculation and coupling analysis.

[0096] If the feature dimension is still high (not meeting the dimension requirement) after feature selection, in order to reduce the computational complexity, a dimensionality reduction method based on local preserving projection (LPP) is used to perform feature dimensionality reduction processing. The formula is:

[0097] ,in, Represents the local neighborhood weight matrix between feature points. By reducing the feature dimension by maintaining the local structure, the feature representation after dimensionality reduction is obtained. , , They represent the high-dimensional feature values ​​of the i-th and j-th features at time t respectively.

[0098] In the automatic control system of hydraulic loading force of coal mill, feature dimension expansion and generation have the following key benefits:

[0099] Capturing the dynamic evolution characteristics of the system, the operating characteristics of the hydraulic loading force system of the coal mill have significant time dependence. Through feature dimension expansion, especially time delay embedding, the dynamic evolution characteristics of system parameters in time can be captured. This expansion method enables the model to not only process the system state at the current moment, but also consider the impact of historical data on the current state, so as to better predict and adjust the hydraulic loading force;

[0100] Revealing hidden nonlinear relationships: In hydraulic loading systems, the relationships between multiple parameters often have complex nonlinear characteristics, and simple linear analysis is difficult to reveal these hidden nonlinear relationships. Through recursive nonlinear mapping and high-dimensional interactive feature generation, the nonlinear expression ability of features can be enhanced, so that the model can better identify and process nonlinear dependencies in the system, thereby improving the accuracy and robustness of the control system.

[0101] In summary, the application of feature dimension expansion and generation in the automatic control system of hydraulic loading force of coal mill not only improves the dynamic response capability and nonlinear recognition capability of the system, but also improves the analysis accuracy and energy consumption optimization effect of the system by increasing feature complexity and multi-parameter interaction information, which has significant practical value.

[0102] Step 2: Based on the joint probability distribution calculation of non-parametric kernel density estimation, a high-dimensional feature matrix has been obtained through feature dimension expansion and generation , perform nonparametric kernel density estimation, further normalize the high-dimensional feature vector to ensure that each feature has the same dimension, and for the feature matrix , using the dynamic interval normalization method, the data is scaled to the [a, b] interval, usually set to [0, 1] to ensure the operability of the data, and the normalized feature data is obtained ;

[0103] The adaptive Gaussian kernel function is selected to estimate the probability density of the dynamic interval normalization process. The Gaussian kernel has good smoothness, and the adaptive kernel width can be dynamically adjusted according to the local characteristics of the data. The kernel width can be adjusted in different data density areas to ensure accurate estimation of the probability density in the high-dimensional feature space.

[0104] Perform kernel density estimation of joint probability distribution. After determining the kernel function, estimate the joint probability distribution of the high-dimensional feature matrix. Suppose there are N sample points , then the joint probability distribution The estimation formula is as follows:

[0105] ,in, Represents the kernel function of kernel density estimation. The adaptive Gaussian kernel is used to smooth and estimate the probability density of multidimensional features.

[0106] Furthermore, the joint probability distribution between different feature dimensions is estimated, specifically:

[0107] ;

[0108] in, represents the probability density of the nth feature, is the time step during data acquisition, is the feature data collected at the jth time point, representing the jth time sample. This formula is used to estimate the joint probability density between all feature dimensions, ensuring that the joint distribution between features can be accurately described in the multidimensional feature space;

[0109] The kernel density estimation of joint probability distribution aims to construct the joint distribution of all features in the multidimensional feature space, so that these joint probability densities can be used in subsequent analysis to identify coupling relationships and detect anomalies. Through kernel density estimation, a relatively smooth and accurate probability distribution can be obtained in high-dimensional space, laying the foundation for subsequent mutual information entropy calculation.

[0110] After completing the estimation of the joint probability distribution, it is necessary to perform a separate kernel density estimation on the marginal probability distribution of each high-dimensional feature vector. The estimation formula for the marginal probability distribution is set as follows:

[0111] ;

[0112] The estimation of marginal probability distribution is to properly normalize the dependencies between features when calculating mutual information entropy. By performing kernel density estimation on the marginal probability of each feature, it is ensured that each variable involved in the calculation of mutual information entropy can be accurately described by probability.

[0113] After completing the joint and marginal probability distribution estimation, in order to ensure the consistency and operability of the probability distribution, the estimated probability density function needs to be normalized. The normalization formula is as follows:

[0114] , where the denominator is the integral value of the probability density function, which is used to ensure that all estimated probability density functions meet the normalization condition, that is, the integral is 1.

[0115] The normalization of the probability density function is to ensure that the distribution function estimated by the kernel density meets the basic requirements of probability theory, that is, the integral of the probability density function is 1. Through normalization, the estimated joint and marginal probability distributions can be more standardized, which is convenient for the subsequent calculation and analysis of mutual information entropy;

[0116] The hydraulic loading force control system of the coal mill may have excessive energy consumption when running at low load or no load. However, this phenomenon is often hidden in the complex multidimensional data characteristics, and it is difficult to identify by traditional linear or simple statistical analysis methods. The joint probability distribution obtained by kernel density estimation can effectively reveal abnormal patterns under complex working conditions, and detect early signs of abnormal energy consumption by comparing the probability distribution under actual working conditions. This role is helpful to improve the system's early warning capability and reduce energy waste.

[0117] Step 3: Calculation of mutual information entropy. In step 2, the joint probability distribution and marginal probability distribution between multidimensional features are obtained by nonparametric kernel density estimation. Based on these probability distributions, the mutual information entropy between each feature variable is calculated. The mutual information entropy I(X; Y) is a measure of the degree of information dependence between two random variables X and Y. Specifically, the mutual information entropy is defined as follows: , this formula calculates the i-th feature and the jth feature The degree of information sharing between them. The integration operation here ensures that the mutual information entropy can capture the nonlinear dependencies between multi-dimensional features.

[0118] The basic definition of mutual information entropy reveals the degree of association between feature variables through the ratio of joint probability distribution and marginal probability distribution. In the automatic control system of hydraulic loading force of coal mill, the interaction between different parameters is complex. Mutual information entropy is used to identify complex relationships. This step quantitatively describes the information dependence between features by calculating mutual information entropy.

[0119] In the multidimensional feature space, it is necessary not only to calculate the mutual information entropy between two variables, but also to consider the joint mutual information between multiple variables. By extending the basic mutual information entropy formula, the mutual information entropy of three variables is defined, such as three variables in the form of I(X; Y; Z).

[0120] The feature dependency network is constructed based on the high-order mutual information entropy, and the feature dependency matrix M is defined, where the elements represents the mutual information entropy between the i-th feature and the j-th feature, , this matrix is ​​used to construct the feature dependency network graph G(V, E), where the node V represents the feature and the weight of the edge E is the mutual information entropy value between the corresponding features ;

[0121] Define the high-order mutual information entropy matrix , calculated by the following formula:

[0122] ,in, represents the kth feature. Similarly, , They represent the i-th and j-th features respectively. This high-order matrix is ​​used to capture the deep dependencies between multi-dimensional features and reflects the weight changes of nodes and edges in the feature dependency network. The feature dependency network is used for the interdependent structure between multiple features through visualization. Especially in high-dimensional feature space, the weights of nodes and edges can show the strong and weak dependencies between features. Introducing high-order mutual information entropy into the feature dependency network is helpful to identify the core variables and key interactions in the system.

[0123] After obtaining the feature dependency network, the sensitivity of each feature to energy consumption is calculated for the abnormal energy consumption under low load or no load conditions. Sensitivity analysis is based on the difference in mutual information entropy. By comparing the changes in mutual information entropy under low load or no load conditions with those under normal load conditions, the sensitivity coefficient is defined: ,in, and Represents the mutual information entropy and sensitivity coefficient under low load and normal load conditions respectively. Indicates the sensitivity of a feature to changes in energy consumption;

[0124] Sensitivity analysis quantifies the impact of each feature on energy consumption through the difference in mutual information entropy. Under low load or no-load conditions, the impact of certain features on energy consumption may be more significant. The sensitivity coefficient can help identify these key features and provide a basis for optimizing control strategies.

[0125] Step 4: Variable coupling analysis based on mutual information entropy: perform variable coupling analysis on the mutual information entropy and high-order mutual information entropy between each feature variable, and normalize the high-order mutual information entropy matrix to ensure the comparability of mutual information entropy between different features, and obtain the normalized high-order mutual information entropy. ;

[0126] Based on the normalized high-order mutual information entropy matrix, the variable coupling strength matrix C is constructed. The coupling strength matrix is ​​used to describe the degree of coupling between different characteristic variables and is defined as: ,in, represents the coupling strength between the i-th feature and the j-th feature, is the regularization coefficient, which controls the smoothness of the coupling strength; , Respectively represent the mutual information entropy between the i-th feature and the k-th feature and the mutual information entropy between the k-th feature and the j-th feature. Through the above formula, the high-order mutual information entropy between feature variables is converted into coupling strength to further analyze the interaction between variables;

[0127] The coupling strength matrix quantifies the interaction strength between different variables by aggregating high-order mutual information entropy. The coupling strength matrix can reveal the strong coupling relationship between variables in the system and help identify the key variable combinations that may cause abnormal energy consumption under low load or no-load conditions.

[0128] Using the coupling strength matrix C, construct the variable coupling network diagram , where node V represents the characteristic variable in the system and edge The weight is the coupling strength between the corresponding variables The specific steps are as follows:

[0129] Determine the initial conditions of nodes and edges from the feature matrix Determine the network nodes, initialize the non-zero elements in the coupling strength matrix C as the initial weights of the edges;

[0130] Filter key coupling relationships according to edge weights , set the threshold , only retain the edges with weights greater than the threshold, and build a streamlined variable coupling network. The filtered network graph focuses more on the variable coupling relationship that has a greater impact on energy consumption; the formula is: ;

[0131] By screening key coupling relationships, it can help analyze which variable combinations have a significant impact on system energy consumption under low load or no-load conditions, providing a reference for optimal control;

[0132] In the constructed variable coupling network, energy consumption abnormal pattern recognition is performed. The variable coupling intensity matrix is ​​used to calculate the probability distribution of energy consumption abnormal pattern under low load or no load conditions. The formula is as follows: ,in, It indicates the probability that the variable combination (i, j) will lead to abnormal energy consumption under low load or no load conditions. In order to adjust the parameters and control the sensitivity of abnormal energy consumption patterns, this probability distribution is used to identify the combination of variables in the system that is most likely to cause abnormal energy consumption;

[0133] Abnormal energy consumption pattern recognition quantifies the contribution of different variable combinations in the system to abnormal energy consumption by calculating the probability distribution relationship between variable coupling intensity and abnormal energy consumption pattern, so as to detect possible abnormal energy consumption in advance and take corresponding measures for optimization control.

[0134] After identifying the coupling relationship between key variables and abnormal energy consumption patterns, the variable coupling model is optimized and the feedback mechanism is designed. The Bayesian optimization algorithm can be used to optimize the coupling intensity matrix, build a feedback loop, feed the identified abnormal energy consumption pattern back to the control system, and adjust the hydraulic loading force in real time to reduce the occurrence of abnormal energy consumption. By optimizing the coupling intensity matrix, the model's ability to predict abnormal energy consumption is improved, and the control strategy can be adjusted in time through the feedback mechanism.

[0135] Step 5: Abnormal energy consumption detection under low load or no load state; In step 4, for the variable coupling model that has been built and the coupling relationship between key variables has been identified, it is first necessary to identify the low load or no load state in real time during system operation. Set the load threshold , by real-time monitoring of the system load L(t), it is determined whether the system is in a low-load or no-load state, and the following calibration function is defined: ,in, Indicates whether the system is in a low-load or no-load state at time t;

[0136] Identify low load or no-load status during operation and provide calibration signals for subsequent energy consumption anomaly detection. By setting the load threshold, it can be determined in real time whether the system has entered a potential energy consumption anomaly state;

[0137] After identifying the low load or no-load state, use the sensitivity coefficient calculated in step 3 to perform preliminary abnormal detection on the energy consumption in the current state. The sensitivity coefficient reflects the degree of influence of different variables on energy consumption changes. Therefore, by monitoring the changes in the sensitivity coefficient, it is possible to preliminarily determine whether there is an energy consumption anomaly and set the energy consumption anomaly threshold. , when the sensitivity coefficient of a pair of variables When the threshold is exceeded, it is marked as a preliminary detection signal of abnormal energy consumption : ; Preliminary detection based on sensitivity coefficient can quickly identify abnormal energy consumption under low load or no-load conditions. This process uses the change of sensitivity coefficient to provide a preliminary judgment basis, which can detect potential energy consumption problems in advance;

[0138] Based on the preliminary detection, the abnormal energy consumption pattern is further identified through the variable coupling model. , build an energy consumption abnormal pattern recognition model and define the energy consumption abnormal pattern function: , where A(t) represents the intensity of abnormal energy consumption mode at time t. If A(t) exceeds the set abnormal mode threshold , it is marked as a significant energy consumption anomaly and triggers an alarm signal;

[0139] Multivariate coupling abnormal pattern recognition integrates the initial detection signal and the abnormal probability distribution in the coupling model to deeply analyze the energy consumption under low load or no-load conditions. This process combines the interaction relationship and sensitivity analysis of multi-dimensional variables, which can more accurately determine the severity of energy consumption anomalies and issue corresponding warnings, thereby controlling the hydraulic loading force of the coal mill.

[0140] After identifying abnormal energy consumption, analyze and predict the abnormal energy consumption trend, and use time series prediction models (such as long short-term memory network, LSTM) to model the future evolution trend of abnormal energy consumption. Analysis and prediction of abnormal energy consumption trends can help foresee serious energy consumption problems that may occur in the system in advance, and provide data support for preventive maintenance and optimization control. Through the time series prediction model, the intensity of abnormal energy consumption in the future can be predicted to help operators prepare for response;

[0141] After completing the detection and prediction of energy consumption anomalies, a feedback control mechanism is built to dynamically adjust the hydraulic loading force of the system. According to the abnormal pattern recognition and trend prediction results, the hydraulic loading force control parameters are adjusted in real time to optimize energy consumption performance;

[0142] The energy consumption abnormality feedback control mechanism reduces the impact of energy consumption abnormalities by dynamically adjusting the hydraulic loading force. This process combines real-time detection and prediction results, enabling the system to make adaptive adjustments based on current and future operating conditions, improve energy efficiency, and reduce unnecessary energy consumption.

[0143] Through the above detailed steps, the abnormal energy consumption detection under low load or no-load state can effectively identify and predict the energy consumption problems in the system, and realize the dynamic optimization control of the system through the feedback control mechanism, ensuring that the coal mill can maintain efficient operation under different working conditions.

[0144] The dynamic optimization control of the system is achieved through the feedback control mechanism. The specific optimization control measures can include the following aspects:

[0145] The system automatically adjusts the hydraulic loading force setting value according to the load and energy consumption status monitored in real time. For example, when it is detected that the coal mill is in a low load state and the energy consumption is abnormal, the system will reduce the hydraulic loading force to avoid unnecessary energy consumption;

[0146] For example, the hydraulic system of a coal mill usually maintains a certain pressure to ensure the stability of the coal grinding process. However, if the system detects that the current coal load is low and the energy consumption is too high, the feedback control mechanism will reduce the energy consumption by lowering the hydraulic loading force set value. This adjustment is based on learning from historical data and predicting energy consumption trends to ensure that the appropriate hydraulic loading force is maintained under low load conditions;

[0147] Automatically adjust the valve opening. According to the load changes and energy consumption analysis results, the valve opening in the hydraulic system is automatically adjusted to control the flow rate and flow of the hydraulic oil. Through precise adjustment, it ensures that the hydraulic system can supply hydraulic oil according to actual needs and avoid waste of resources.

[0148] Adaptively adjust the response speed of the equipment, dynamically adjust the response speed of the system equipment according to the changes in working conditions, and ensure that the system reduces energy consumption without affecting production efficiency. For example, reduce the start-up frequency of the hydraulic pump in the system or extend the response time of the equipment. When the system detects that the load changes slowly and there is no significant energy consumption risk, the feedback mechanism will appropriately reduce the start-up frequency of the hydraulic pump, thereby reducing the energy loss caused by frequent equipment startup.

[0149] It should be noted that the thresholds involved in the embodiments may be determined according to specific scenarios and requirements, and are usually adjusted and optimized based on factors such as historical data and real-time data.

[0150] The present invention uses multivariate coupling analysis and non-parametric kernel density estimation methods to accurately capture the complex relationship between various parameters in the hydraulic loading force control system of the coal mill, especially under low load or no-load conditions, and can identify and analyze the key factors that lead to abnormal energy consumption in real time. By constructing a high-dimensional feature matrix and a mutual information entropy matrix, the problem of insufficient multi-parameter correlation in traditional control systems is effectively solved, ensuring the control accuracy and response speed of the system under different working conditions. With the help of an efficient variable coupling intensity matrix and an adaptive update mechanism, the control strategy can be dynamically adjusted and the hydraulic loading force can be optimized in real time, thereby avoiding unnecessary energy waste.

[0151] In addition, the present invention introduces an energy consumption trend prediction and feedback control mechanism based on the LSTM model, which can identify potential energy consumption anomalies in advance and adjust control parameters in real time through a feedback loop to ensure that the system maintains the best energy efficiency during operation. Compared with traditional fixed control strategies, the present invention significantly improves the system's adaptability and intelligence level and reduces the need for manual intervention. Through this innovative technology, the present invention not only improves the overall operating efficiency of the coal mill hydraulic loading force control system, but also significantly reduces energy consumption.

[0152] Embodiment 2: This embodiment is a system embodiment of embodiment 1, which is used to implement a coal mill hydraulic loading force automatic control method introduced in embodiment 1, such as Figure 2 As shown, specifically including:

[0153] The data processing module is used to obtain multidimensional data during the operation of the coal mill, perform nonlinear transformation and standardization, generate high-dimensional feature vectors through time delay embedding, perform kernel density estimation on the extracted multidimensional feature vectors, construct the joint probability distribution between features, and obtain the marginal probability distribution through normalization as the basis for calculating the mutual information entropy;

[0154] Energy consumption analysis module, which is used to calculate the mutual information entropy between each feature based on the obtained joint probability distribution, build a feature dependency network, and use an adaptive update algorithm to dynamically adjust the mutual information entropy matrix to determine the energy consumption sensitivity based on the mutual information entropy difference;

[0155] Energy consumption identification module, which is used to construct a variable coupling strength matrix based on the mutual information entropy matrix, screen out key coupling relationships and generate a variable coupling network, and identify abnormal energy consumption patterns under low load or no-load conditions;

[0156] The control module is used to adjust the variable coupling relationship, predict the abnormal energy consumption trend according to the abnormal energy consumption mode, and dynamically adjust the hydraulic loading force control parameters in combination with the feedback mechanism.

[0157] The above formulas are all dimensionless and calculated numerically. Specific dimension removal can be achieved by various means such as standardization, which will not be elaborated here. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0158] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, an ATA hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state ATA hard disk.

[0159] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0160] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0161] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0162] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0163] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0164] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for automatically controlling the hydraulic loading force of a coal mill, characterized in that: The steps include: Acquire multi-dimensional data during the operation of the coal mill, determine the changes in the load and operating parameters of the coal mill unit, and modify the system loading force in real time according to the changes in the unit load and the feedback of the operating parameters, and control the loading force response amplitude through the integral equation; The integral equation is expressed as: , where L(t) is the loading force at the current time t, ΔM(t) is the change in the coal feed M(t), ΔQ(t) is the change in coal seam thickness, ΔR(t) is the change in the resistance of the grinding area, and V(t) is the ventilation volume of the coal mill. is the integral coefficient, corresponding to the influence weights of coal feed rate, coal seam thickness and grinding zone resistance, are differential coefficients, corresponding to the weights of the response speed of the coal feed rate and ventilation volume changes to the loading force.

2. A method for automatically controlling hydraulic loading force of a coal mill according to claim 1, characterized in that: Multi-dimensional data acquisition collects time series data of key operating parameters, including coal seam thickness, coal feed rate, grinding zone resistance, coal mill ventilation volume, and coal mill vibration value.

3. The method for automatically controlling the hydraulic loading force of a coal mill according to claim 1, characterized in that: The following steps are also included: Acquire multidimensional data during the operation of the coal mill, perform nonlinear transformation and standardization, and generate high-dimensional feature vectors through time delay embedding; Perform kernel density estimation on the extracted high-dimensional feature vectors, construct the joint probability distribution between features, and obtain the marginal probability distribution through normalization as the basis for calculating mutual information entropy. The mutual information entropy between features is calculated based on the obtained joint probability distribution, and a feature dependency network is constructed. The mutual information entropy matrix is ​​dynamically adjusted using an adaptive update algorithm, and the energy consumption sensitivity is determined based on the mutual information entropy difference. Based on the mutual information entropy matrix, a variable coupling strength matrix is ​​constructed to screen out key coupling relationships and generate a variable coupling network to identify abnormal energy consumption patterns under low load or no-load conditions; Adjust the variable coupling relationship, predict the abnormal energy consumption trend according to the abnormal energy consumption mode, and dynamically adjust the hydraulic loading force control parameters in combination with the feedback mechanism; The multi-dimensional data during the operation of the coal mill is obtained, nonlinear transformation and standardization are performed, and high-dimensional feature vectors are generated through time delay embedding. The specific process is as follows: Temporal alignment of multidimensional data using spline interpolation; Use boundary detection and DBSCAN algorithm to remove data points; The multidimensional data is transformed nonlinearly using a combined transformation function and standardized using the minimum and maximum normalization method; Based on the standardized multidimensional data, time delay embedding expansion is performed, the time delay step and embedding dimension are set, and each standardized data is expanded into a delayed embedding vector and used as a high-dimensional feature vector.

4. A method for automatically controlling hydraulic loading force of a coal mill according to claim 3, characterized in that: The kernel density of the extracted high-dimensional feature vector is estimated to construct the joint probability distribution between the features, and the marginal probability distribution is obtained by normalization as the basis for calculating the mutual information entropy. The specific process is as follows: Dynamic interval normalization of high-dimensional feature vectors; Adaptive Gaussian kernel function is used to estimate the probability density of high-dimensional feature vectors processed by dynamic interval normalization; Construct a high-dimensional feature matrix based on the high-dimensional feature vector, calculate the joint probability distribution of the high-dimensional feature matrix, and calculate the joint probability distribution between different feature dimensions; After the joint probability distribution is calculated, a separate kernel density calculation is performed on the marginal probability distribution of each high-dimensional feature vector.

5. The method for automatically controlling the hydraulic loading force of a coal mill according to claim 3, characterized in that: The mutual information entropy between the features is calculated based on the obtained joint probability distribution, and a feature dependency network is constructed. The mutual information entropy matrix is ​​dynamically adjusted using an adaptive update algorithm, and the energy consumption sensitivity is determined based on the mutual information entropy difference. The specific process is as follows: The joint probability distribution and marginal probability distribution between high-dimensional features are obtained through non-parametric kernel density estimation. The mutual information entropy between each feature variable is calculated based on the joint probability distribution and marginal probability distribution of the probability distribution to determine the information dependency of the features. Define the feature dependency matrix, and construct a feature dependency network graph based on the feature dependency matrix. The nodes in the feature dependency network graph represent the features, and the edge weights are the mutual information entropy values ​​between the corresponding features. A high-order mutual information entropy matrix is ​​constructed to compare the changes in mutual information entropy under low load or no-load conditions with those under normal load conditions, and the sensitivity of each feature to energy consumption is calculated.

6. A coal mill hydraulic loading force automatic control system, used to implement a coal mill hydraulic loading force automatic control method according to any one of claims 1 to 5, characterized in that: include: The data processing module is used to obtain multidimensional data during the operation of the coal mill, perform nonlinear transformation and standardization, generate high-dimensional feature vectors through time delay embedding, perform kernel density estimation on the extracted high-dimensional feature vectors, construct the joint probability distribution between features, and obtain the marginal probability distribution obtained by normalization as the basis for calculating the mutual information entropy; Energy consumption analysis module, which is used to calculate the mutual information entropy between each feature based on the obtained joint probability distribution, build a feature dependency network, and use an adaptive update algorithm to dynamically adjust the mutual information entropy matrix to determine the energy consumption sensitivity based on the mutual information entropy difference; Energy consumption identification module, which is used to construct a variable coupling strength matrix based on the mutual information entropy matrix, screen out key coupling relationships and generate a variable coupling network, and identify abnormal energy consumption patterns under low load or no-load conditions; The control module is used to adjust the variable coupling relationship, predict the abnormal energy consumption trend according to the abnormal energy consumption mode, and dynamically adjust the hydraulic loading force control parameters in combination with the feedback mechanism.

Images