A hydraulic support center distance management linkage control system for steeply inclined working face

Abstract

The present application belongs to the technical field of fully mechanized coal mining equipment control, and relates to a hydraulic support center distance management linkage control system for steeply inclined working face. When the kth control cycle starts, the system successively performs: data acquisition and preprocessing, online calibration based on the prediction error of the k-1 prediction interval to obtain a smooth calibration item available for the kth control cycle; construction of a discrete time linear time-varying model based on the working condition characteristic vector and the smooth calibration item; mapping of the master frame and the follow-up frame according to the adjacent grouping principle; generation of in-group collaborative input and follow-up frame compensation based on state difference and center distance deviation, synthesis of collaborative control decision; linkage control is performed within the kth control cycle, and the state of the k+1 control cycle is predicted within the same cycle. Thus, the single-frame prediction calibration capability is improved, the mutual disturbance caused by the hydraulic coupling between adjacent supports and the surrounding rock transmission is weakened, and the risk of in-group chain instability is reduced.

Description

Technical Field

[0001] This invention belongs to the field of automatic control technology, specifically relating to a program control system for hydraulic support groups facing steeply inclined working faces, and more particularly to a hydraulic support center distance management linkage control system based on multi-source state acquisition, working condition feature construction, online calibration, linear time-varying model construction, group collaborative decision-making, and cross-cycle state prediction. Background Technology

[0002] In automatic control systems, the coordinated control of multiple controlled objects typically requires the control system to generate the control input for the next control cycle based on the real-time status of each object, changes in external operating conditions, and the execution results of the previous control cycle. However, in application scenarios with strong disturbances, multivariate coupling, and continuously changing operating conditions, if the control system still uses fixed control parameters, offline gain tables, or independent control logic for a single object, problems such as mismatch between control parameters and real-time operating conditions, failure to promptly correct prediction results from the previous control cycle, and lack of coordinated constraints among multiple objects can easily arise. This can lead to control output lag, overshoot, or instability in the coordinated operation between objects.

[0003] Taking the control of a hydraulic support group on a steeply inclined working face as an example, the hydraulic support, as multiple controlled actuators, needs to be linked and controlled during the working face advancement process based on state and working condition variables such as column stroke, support resistance, roll attitude angle, slippage, pump station pressure supply, and dynamic tilt angle of the working face. Existing control methods mostly employ fixed-gain control or multi-dimensional working condition-gain mapping table control. Fixed-gain control typically involves offline tuning of control parameters for typical working conditions, and these parameters remain unchanged during system operation. Multi-dimensional working condition-gain mapping table control, on the other hand, pre-calibrates control parameters corresponding to different working condition combinations, and the corresponding parameters are retrieved by looking up a table during system operation.

[0004] The above control methods have the following shortcomings: First, the control parameters mainly rely on offline tuning or preset mapping tables, and cannot be calibrated online based on the prediction error of the previous control cycle, resulting in insufficient adaptive capability of the control system to real-time operating condition changes; Second, multiple hydraulic supports are usually controlled as independent execution objects, and the control system has not established a programmed collaborative decision-making logic between the main control frame and the follower frame, making it difficult to uniformly constrain the state difference and center distance deviation between adjacent supports; Third, existing control systems usually only focus on the control output of the current control cycle, lacking a cross-cycle closed-loop mechanism that uses the current cycle control input, collaborative input, and calibration items to predict the state of the next control cycle, resulting in a lack of reliable predictive state basis for the control decision of the next cycle.

[0005] In steeply inclined working faces, the above problems are amplified. Due to the hydraulic coupling and surrounding rock transmission relationship between adjacent hydraulic supports, the control over-response of a single support may be transmitted and amplified through adjacent supports, thereby triggering a chain reaction of instability in the supports within the group. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: A linkage control system for managing the center distance of hydraulic supports on steeply inclined working faces is proposed, comprising: The data acquisition module is used to acquire the measured state vector of each support and the global working condition vector of the working face at the beginning of the k-1 control cycle. The data preprocessing module is used to perform timestamp alignment, outlier removal, and signal filtering on the measured state vector and global operating condition vector output by the data acquisition module, and to construct operating condition feature vectors based on the preprocessed data. The online calibration module compares the measured state vector with the predicted state vector of the kth control period obtained from the (k-1)th control period to obtain the prediction error. Using the vector error in the prediction error as the driving signal, the calibration parameters are updated posteriorly using an online recursive learning algorithm. After amplitude constraint and rate of change constraint processing, a smooth calibration term usable in the kth control period is obtained. The model building module is used to calculate the time-varying coefficient matrix in real time based on the working condition feature vector, and to construct a discrete-time linear time-varying model by constraining the range of values ​​of each time-varying coefficient in the time-varying coefficient matrix through a limiting function. The control group division module is used to divide the hydraulic supports of the working face into multiple control groups according to the principle of adjacent grouping; the odd-numbered supports in each group are the main control frames and the even-numbered supports are the follower frames. The decision generation module is used to generate the control input of the master frame based on the discrete-time linear time-varying model and the smoothing calibration term available in the k-th control cycle; generate the intra-group cooperative input based on the state vector difference and center distance deviation between the master frame and the follower frame; generate the follower frame compensation control input according to the follower frame compensation amount; and generate a cooperative control decision that includes the master frame control command, the intra-group cooperative input, and the follower frame compensation control input. The collaborative control module is used to perform coordinated control based on collaborative control decisions during the k-th control cycle. The state prediction module is used to predict the state vector of the single frame in the (k+1)th control cycle by using the time-varying coefficient matrix of the kth control cycle, the single frame control input, the group cooperative input, and the smoothing calibration terms available in the kth control cycle. The predicted state vector of the (k+1)th control cycle is then used for comparison between the predicted state vector at the beginning of the (k+1)th control cycle and the measured state vector formed after the end of the kth control cycle.

[0007] Compared with the prior art, the present invention has the following advantages and beneficial effects: By setting up a data acquisition module, a data preprocessing module, an online calibration module, a model building module, a control group division module, a decision generation module, a collaborative control module, and a state prediction module, the present invention enables the control system to complete the acquisition of measured state vectors and global operating condition vectors, construction of operating condition feature vectors, calculation of prediction errors, a posteriori update of calibration parameters, construction of discrete-time linear time-varying models, control group division, collaborative control decision generation, linkage control execution, and state prediction for the next control cycle in sequence according to a predetermined program control flow, thereby forming a cross-cycle closed-loop program control link for multiple controlled execution objects. Specifically, the online calibration module compares the measured state vector obtained in the current control cycle with the predicted state vector obtained in the previous control cycle, and uses the vector error in the prediction error to drive the posterior update of the calibration parameters, obtaining a smooth calibration term available in the current control cycle. This allows the control system to correct the control model online based on the actual deviation in the previous prediction interval, instead of simply relying on fixed gain or offline mapping tables. The model building module calculates the time-varying coefficient matrix in real time based on the operating condition feature vectors, and uses the time-varying coefficient matrix to build a discrete-time linear time-varying model, enabling the control model to be parametrically adjusted according to changes in operating conditions such as pump station pressure supply, dynamic inclination angle of the working face, and roof pressure. The control group division module and decision generation module... The module divides adjacent hydraulic supports into multiple control groups and establishes a collaborative control relationship between the master control frame and the follower frame within each control group. Based on the state vector difference and center distance deviation between the master control frame and the follower frame, it generates intra-group collaborative input and combines this with the follower frame compensation amount to generate follower frame compensation control input, transforming the control system from independent single-frame control to group-linked control. The collaborative control module performs linked control of the hydraulic supports according to the collaborative control decisions. The state prediction module, within the current control cycle, uses the time-varying coefficient matrix, master control frame control input, intra-group collaborative input, and smoothing calibration term to predict the predicted state vector for the next control cycle, and compares this predicted state vector with the measured state vector at the beginning of the next control cycle. Therefore, this invention can achieve a closed-loop operation mechanism of "prediction error feedback - online calibration - time-varying modeling - group collaborative decision-making - linked control - next cycle state prediction" at the program control level, improving the control system's adaptability to complex working conditions on steeply inclined working faces, reducing disturbance amplification caused by hydraulic coupling and surrounding rock transmission between adjacent hydraulic supports, reducing the risk of chain instability of supports within the group, and thus improving the coordination, stability, and safety of the hydraulic support group's linked control. Attached Figure Description

[0008] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1This is a schematic diagram of the module execution flow of the hydraulic support center distance management linkage control system for steeply inclined working faces provided in Embodiment 1 of the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. The illustrative embodiments and descriptions of this invention are for illustrative purposes only and are not intended to limit the invention. The embodiments described below are some, but not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0010] In the following description, numerous specific details are set forth to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other embodiments, well-known structures, materials, or methods are not specifically described to avoid obscuring the invention. Unless otherwise specified, the materials, instruments, and reagents used in the following embodiments are commercially available. Unless otherwise specified, the techniques used in the embodiments are conventional methods well known to those skilled in the art.

[0011] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0012] Example 1: In actual production on steeply inclined working faces, neither the fixed-gain control nor the multi-dimensional mapping table control technologies for hydraulic supports can achieve online adaptive adjustment of control parameters to changes in working conditions. When the dynamic fluctuation of pump station pressure reaches ±10%, the dynamic inclination angle of the working face changes in real time within ±5°, and the pressure intensity from the roof plate changes periodically within the peak pressure range of 30-50MPa, the pre-tuned control parameters cannot track these changes in real time, resulting in low control accuracy of a single support and a high risk of support instability. Furthermore, the aforementioned hydraulic support control technologies all treat a single support as an independent control object, without considering the strong mechanical coupling relationship between adjacent supports through the hydraulic system and the surrounding rock geology. Once a single support undergoes an aggressive action, the resulting impact force and displacement disturbance will be amplified to adjacent supports through the hydraulic coupling path and the surrounding rock geology transmission path, evolving into synchronous chain instability of three or more supports within the group in a very short time.

[0013] To address the aforementioned technical issues, this embodiment provides a linkage control system for the center distance management of hydraulic supports on steeply inclined working faces. The system's core logic is based on "data acquisition and preprocessing - error feedback from the previous prediction interval - posterior calibration - parameterized modeling of working condition characteristics - grouped collaborative decision-making - simultaneous periodic linkage control and prediction of the next cycle." The overall control route is achieved collaboratively through a data acquisition module, a data preprocessing module, an online calibration module, a model building module, a control group division module, a decision generation module, a collaborative control module, and a state prediction module. At the beginning of the k-th control cycle, the data acquisition module acquires the measured state vector formed after the (k-1)-th control cycle and the global working condition vector of the working face. The data preprocessing module performs preprocessing on the acquired data, including timestamp alignment, outlier removal, and signal filtering, to construct a working condition feature vector. The online calibration module compares the preprocessed measured state vector with the predicted state vector for the k-th control cycle obtained from the (k-1)-th control cycle to obtain the prediction error. This prediction error is then used as a driving signal to posteriorly update the calibration parameters using an online recursive learning algorithm, yielding the result for the k-th control cycle. The system first obtains a smoothing calibration term available for the current control cycle. Then, the model building module calculates the four time-varying coefficient matrices of the discrete-time linear time-varying model (hereinafter referred to as the "linear time-varying model") in real time based on the working condition feature vector. The obtained smoothing calibration term and the time-varying coefficient matrix are used to construct a discrete-time linear time-varying model usable in this control cycle. Next, the control group partitioning module groups the hydraulic supports on the working face into even-numbered groups, establishing a master-slave mapping structure between the master control frame and the follower frame within each group. The decision generation module generates the control input for the master control frame based on the discrete-time linear time-varying model and the smoothing calibration term available for the k-th control cycle. It also generates the group-wide collaborative input and the follower frame compensation control input based on the group-wide state difference and center distance deviation. Finally, the collaborative control module performs linkage control on the supports according to the above collaborative control decisions within the k-th control cycle. Simultaneously, the state prediction module predicts and stores the single-frame state vector for the (k+1)-th control cycle using the time-varying coefficient matrix, single-frame control input, group-wide collaborative input, and the smoothing calibration term available for the k-th control cycle, for comparison with the measured state vector at the beginning of the (k+1)-th control cycle.

[0014] Based on the above overall technical approach, this system is composed of various functional modules according to... Figure 1 The modules shown execute their logic. Specifically, the data acquisition module, data preprocessing module, online calibration module, model building module, control group division module, and decision generation module form a coordinated control decision at the beginning of the k-th control cycle. The coordinated control module and the state prediction module execute linked control and state prediction for the next cycle, respectively, during the k-th control cycle. The execution logic of each functional module is as follows: (a) Data Acquisition Module

[0015] The data acquisition module is used to acquire the measured state vector of each hydraulic support and the global working condition vector of the working face at the beginning of the k-1 control cycle.

[0016] The purpose of the data acquisition module is to fully acquire the raw sensor data required to support error verification of the previous prediction interval, a posteriori calibration update, current cycle operating condition modeling, collaborative control decision-making, and state prediction of the next control cycle at the beginning of each control cycle, and to build a two-layer data acquisition system covering single-frame state variables and global operating condition variables.

[0017] The specific implementation method of the data acquisition module is as follows: 1. Collect the individual status parameters of each hydraulic support. At the start of each control cycle, the following data are independently collected for each support: "column stroke, column circuit pressure, support resistance, roof pressure, roll attitude angle, slippage, and current action amount." The "column stroke, support resistance, roll attitude angle, and slippage" are then used to construct a single-support state vector. This single-support state vector will serve as the core input for subsequent linear time-varying model prediction and online calibration.

[0018] (1) Column travel The column stroke is a core state quantity for measuring the support height of the hydraulic support on the roof plate, measured in millimeters. In this embodiment, the column stroke is measured using a laser displacement sensor with a range of 0–500 mm and a measurement accuracy of no less than ±0.1 mm. The laser displacement sensor is installed on the lower surface of the middle part of the roof beam, with a sampling frequency of 5 Hz. The laser displacement sensor can also be replaced with a magnetostrictive displacement sensor or a wire encoder, or other displacement measuring devices of equivalent accuracy. The safe control range for the column stroke is typically 100 mm to 400 mm, where 100 mm is the lower limit of the stroke corresponding to the minimum support height, and 400 mm is the upper limit of the stroke corresponding to the maximum extension of the column. The specific range should be determined according to the design stroke parameters of the support model.

[0019] (2) Pressure in the column circuit The pressure in the column circuit is directly measured by a pressure transmitter installed at the cylinder inlet of the column, with a range of 0 to 50 MPa, an accuracy of not less than ±0.5%FS, and a sampling frequency of 10 Hz.

[0020] (3) Support resistance Support resistance is the actual supporting force of the columns on the roof slab, calculated by multiplying the column circuit pressure by the effective cross-sectional area of ​​the columns. F i ( k )= P col,i ( k )× S col .in,F i ( k ) represents the support resistance of the i-th hydraulic support, in kilonewtons (kN). P col,i ( k ) represents the column circuit pressure of the i-th hydraulic support; S col Let be the effective cross-sectional area of ​​the column of the i-th hydraulic support, determined by the support design parameters. Taking the ZFY4000 / 18 / 38 hydraulic support as an example... S col The typical value is 0.015m²; the effective cross-sectional area of ​​the column varies for different models of hydraulic supports and should be determined according to the design parameters of the specific support model.

[0021] (4) Top plate pressure The top plate pressure can be replaced by the column circuit pressure. Although the values ​​are similar, their physical meanings are different. The column circuit pressure is the measured hydraulic pressure at the column cylinder port, while the top plate pressure is the top plate load pressure borne by the hydraulic support. The difference between the two values ​​is negligible within the allowable range of engineering accuracy.

[0022] (5) Roll attitude angle The roll attitude angle is a key state quantity measuring the degree of lateral tilt of the support beam along the working face's advancing direction, measured in radians (rad). A positive roll attitude angle indicates the beam is tilted upwards, while a negative roll attitude angle indicates it is tilted downwards. The roll attitude angle is measured using an inertial measurement unit (IMU), with a measurement range of ±180° and an accuracy of no less than ±0.05°. The IMU is installed at the center of gravity of the support beam, with a sampling frequency of 10Hz. The safe range for the roll attitude angle is typically -0.05rad to +0.05rad (corresponding to -2.86° to +2.86°). The IMU can also be replaced by a combination of a dual-axis tilt sensor and a gyroscope, or a multi-point level array can be used for attitude estimation.

[0023] (6) Slip Slippage is the cumulative displacement of the hydraulic support along the inclined direction of the working surface, reflecting the degree of downward sliding of the hydraulic support relative to its initial position, measured in millimeters. Slippage is calculated cumulatively over the stroke of the push cylinder. The stroke of the push cylinder is measured using a magnetostrictive displacement sensor installed in the cylinder body, with a range of 0–800 mm, an accuracy of no less than ±0.2 mm, and a sampling frequency of 5 Hz. The safe control range for slippage is typically -100 mm to +100 mm.

[0024] (7) Current number of actions performed The current action quantity is the valve opening degree or target stroke increment signal from the previous control cycle, which is used to construct the action change rate component of the subsequent operating condition feature vector.

[0025] 2. Collect global working conditions of the working face. At the start of each control cycle, global operating parameters of the working face are collected, including: pump station pressure supply, dynamic inclination angle of the working face, roof pressure cycle index, coal mining machine position, and roof and floor crushing index. These global operating parameters are collected through a sensor network distributed across the working face.

[0026] (1) Pump station pressure supply The pump station pressure is measured by a pressure sensor installed on the pump station outlet manifold, with a range of 0–40 MPa, an accuracy of no less than ±0.3%FS, and a sampling frequency of 10 Hz. The pump station pressure is a key operating condition input for subsequent calculation of the control input coefficients of the linear time-varying model, reflecting the real-time fluid supply capacity of the hydraulic power system.

[0027] (2) Dynamic dip angle of working face The dynamic tilt angle of the working face is measured by an array of tilt sensors distributed along the working face, with a range of 0–90°, an accuracy of no less than ±0.1°, and a sampling frequency of 10Hz. The tilt sensor array consists of multiple tilt sensors spaced apart along the working face; for example, one tilt sensor is installed every 10 hydraulic supports along the working face. The measured values ​​of each tilt sensor are interpolated to obtain an estimate of the local tilt angle at the position of each support.

[0028] (3) Top plate pressure cycle index The roof pressure cycle index is a pressure stage judgment index constructed based on historical roof pressure data (such as roof pressure over the past 30 days). In this embodiment, the historical roof pressure data is discretized into three stages: a value of 0 for the stable period, a value of 1 for the initial stage of pressure, and a value of 2 for the peak stage. The specific judgment method is as follows: A sliding window statistical method is used to identify the periodic roof pressure peak on the roof pressure time series over the past 30 days; if the current roof pressure > a first pressure threshold (e.g., current roof pressure > 90% of the historical roof pressure peak), then the current roof pressure is determined to be in the peak stage of pressure; if the second pressure threshold ≤ the current roof pressure ≤ the first pressure threshold (e.g., 50% of the historical roof pressure peak ≤ the current roof pressure ≤ 90% of the historical roof pressure peak), then the current roof pressure is determined to be in the initial stage of pressure; if the current roof pressure < the second pressure threshold (e.g., current roof pressure < 50% of the historical roof pressure peak), then the current roof pressure is determined to be in the stable period. The first and second pressure thresholds can be calibrated and adjusted based on the quantile statistics of historical roof pressure data from the field. The roof pressure cycle is typically between 5 and 30 days, with peak pressures reaching 30 to 50 MPa.

[0029] (4) Coal mining machine location The coal mining machine position indicates the relative distance between the coal mining machine and the target hydraulic support, in meters. Underground positioning is preferentially achieved using ultra-wideband positioning technology, but can also be replaced by a combination of inertial navigation and odometer, encoder odometer estimation, or roadway base station positioning. The preferred positioning accuracy is no less than ±0.5m.

[0030] (5) Crushing index of top and bottom plates The roof and floor fragmentation index is a normalized objective indicator of roof and floor integrity, ranging from 0 to 1; 0 represents intact roof and floor, and 1 represents extremely fragmented roof and floor. This index is obtained through normalization mapping based on at least one or a combination of roof pressure fluctuation coefficient, root mean square value of equipment vibration, acoustic emission and impact count, and surrounding rock hardness test results. The weights of each component and the normalization mapping relationship can be determined by historical data quantile statistics, field calibration, or regression fitting.

[0031] It should be noted that among the above five global operating conditions, the pump station pressure supply and the dynamic inclination angle of the working face are mandatory parameters, while the roof pressure cycle index, coal mining machine position, and roof and floor crushing index are extended parameters.

[0032] In the two-layer data acquisition system constructed by the data acquisition module, the single-frame state quantity is the direct input for subsequent linear time-varying model prediction calculation and error verification of the previous prediction interval, while the global operating condition quantity is used for subsequent online calculation of the time-varying coefficients of the linear time-varying model. Both together support the calculation process of subsequent functional modules.

[0033] (ii) Data Preprocessing Module

[0034] The data preprocessing module is used to preprocess the data collected by the data acquisition module and construct the working condition feature vector.

[0035] The purpose of the data preprocessing module is to perform quality processing on the raw data collected by the data acquisition module, eliminate the interference of inconsistent sampling frequencies of different sensors, occasional outliers, and high-frequency signal jitter on the accuracy of subsequent calculations, and normalize the processed measured state vector and global operating condition vector into a standard format operating condition feature vector; and use this operating condition feature vector as a common interface for subsequent calculation of the time-varying coefficients of the linear time-varying model and the regression vector of the online recursive learning algorithm.

[0036] When performing preprocessing, the data preprocessing module includes timestamp alignment, outlier removal, and signal filtering; the data preprocessing module constructs a working condition feature vector based on the preprocessed data.

[0037] The data preprocessing module performs preprocessing on the data, including timestamp alignment, outlier removal, and signal filtering. Based on the preprocessed data, the data preprocessing module constructs a feature vector of operating conditions.

[0038] 1. Timestamp alignment Because the sampling frequencies of each sensor differ, the sampling points of each sensor cannot directly correspond to the same moment at the beginning of each control cycle Ts. The timestamp alignment method uses the start time of the control cycle as the reference: for sensors with sampling frequencies higher than 1 / Ts, the arithmetic mean of all sampling points within the control cycle is taken as the representative value for that control cycle; for sensors with sampling frequencies equal to 1 / Ts, the sampling value at that sampling moment is directly taken; for sensors with sampling frequencies lower than 1 / Ts, linear interpolation is used to supplement missing data. Linear interpolation can also be replaced by spline interpolation, which is suitable for sensor combinations with large differences in sampling frequencies.

[0039] 2. Remove outliers Within a sliding window spanning multiple recent control cycles, anomaly detection is performed on each sensor signal. For signals following a normal distribution, the 3σ criterion is used (i.e., a sampled value exceeding the mean ± 3 times the standard deviation is considered an outlier). For signals not following a normal distribution, the interquartile range criterion is used (i.e., a sampled value below the first quartile minus 1.5 times the interquartile range or above the third quartile plus 1.5 times the interquartile range is considered an outlier). Either criterion can be selected based on the signal distribution characteristics, or they can be used in parallel.

[0040] When a sampled value is identified as an outlier, it is replaced with the valid sampled value from the previous control cycle, and the anomaly count of the sensor is incremented by 1. If an outlier occurs in the same sensor for multiple consecutive control cycles (e.g., 3 consecutive control cycles), a sensor anomaly alarm is triggered, and the safety interlock process of the output constraint and safety interlock module is entered.

[0041] In this embodiment, the length of the sliding window is preferably 50 control cycles, and can be adjusted within the range of 30 to 100 control cycles according to the actual situation on site. The larger the window, the more stable the statistics, but the lower the real-time performance.

[0042] 3. Signal filtering For signals with large fluctuations, such as pressure and displacement, a first-order low-pass filter or an m-point moving average is used for signal filtering.

[0043] (1) First-order low-pass filter The first-order low-pass filter equation is: x f ( k )=λ· x f ( k -1)+(1-λ)· x raw ( k Where λ is the low-pass filter coefficient; x f ( k() represents the signal value after first-order low-pass filtering in the k-th control cycle; x f ( k -1) is the signal value after first-order low-pass filtering in the k-1th control cycle; x raw ( k The value is the original signal acquired during the k-th control cycle. The low-pass filter coefficient λ is determined based on the signal jitter amplitude: when the signal jitter amplitude does not exceed 0.1mm, λ = 0.85 can be taken; when the jitter amplitude is greater than 0.1mm, λ = 0.90 can be taken. The selectable range of λ is 0.80 to 0.95. The larger the λ is, the stronger the filtering effect, but the slower the response to changes in the real signal.

[0044] (2) m-point moving average The equation for the moving average at point m is: x avg (k)=[ x raw ( k -m+1)+…+ x raw ( k )] / m. Among them, x avg ( k ) represents the signal value after m-point moving average processing in the k-th control cycle. In the m-point moving average, m=5 is preferred for pressure signals, and m=3 is preferred for displacement signals.

[0045] 4. Constructing the characteristic vector of the working condition After normalizing the preprocessed sensor data, the data are spliced ​​together to obtain a 6-dimensional working condition feature vector containing "pump pressure ratio, normalized value of roof pressure, sine value of working face inclination angle, sign of roll attitude angle, segmentation of coal mining machine position, and rate of change of action".

[0046] (1) Pump pressure ratio Pump pressure ratio = P_ref / P pump ( k );in, P ref Rated pump pressure, P pump ( k The measured pressure of the pump station is 1.0. When the pump station pressure is lower than the rated pump pressure, the pump pressure ratio is greater than 1.0, indicating insufficient hydraulic power. In this case, the pressure should be cut off to 1.0 to maintain the normalization consistency of the characteristic vector.

[0047] (2) Normalized value of top plate pressure Normalized value of top plate pressure = P roof,i (k ) / P roof,max ;in, P roof,i ( k Let be the top plate pressure corresponding to the i-th hydraulic support in the k-th control cycle. P roof,max The normalized upper limit of the roof pressure is determined based on the statistical analysis of historical peak roof pressure at the working face; in this embodiment, it is taken as... P roof,max =50MPa, the upper limit of the peak pressure range of 30~50 MPa for the roof of the emergency inclined working face; this value can be calibrated and adjusted according to the measured roof pressure quantile (such as the 99th quantile) on site.

[0048] (3) Sine value of the working face inclination angle The formula for calculating the sine value of the working face inclination angle is: sine value of working face inclination angle = sinθ(k), where θ(k) is the dynamic inclination angle of the working face in the kth control cycle. When the working face inclination angle is in the range of 35° to 60°, the value range of the inclination angle influence term is [0.57, 0.87].

[0049] (4) Roll attitude angle symbol item The roll attitude angle sign is used to distinguish the direction of the top beam deflection when calculating the collaborative input coefficients. It takes a value of +1 when the roll attitude angle is not less than 0, and a value of -1 when the roll attitude angle is less than 0.

[0050] (5) Coal mining machine position segmentation The segmentation threshold for the coal mining machine's position segmentation can be adjusted based on historical data. For example, the value is 2 when the relative distance between the coal mining machine and the target support is less than 5m, 1 when the distance is between 5m and 10m, and 0 when the distance is greater than 10m.

[0051] (6) Rate of change of motion Rate of change of motion = | u i ( k )- u i ( k -1)| / Ts; where, u i ( k ) represents the control input for the i-th hydraulic support in the k-th control cycle. u i ( k -1) is the control input for the i-th hydraulic support in the (k-1)-th control cycle.

[0052] The functions of the working condition feature vector constructed by the data preprocessing module include: (1) serving as the regression vector for the online recursive learning algorithm in the online calibration module, driving the recursive update of calibration parameters; (2) serving as the input for calculating the time-varying coefficients of the linear time-varying model in the model building module, achieving parameterized adaptation through the mapping relationship between working conditions and coefficient matrices. Normalization eliminates the influence of the difference in the magnitude of physical quantities in each field on the stability of numerical calculation, ensuring the convergence of the online recursive algorithm.

[0053] (III) Online calibration module The online calibration module is used at the beginning of the k-th control cycle to compare the measured state vector after the end of the (k-1)-th control cycle, which is formed by the data preprocessing module, with the predicted state vector of the k-th control cycle obtained by the (k-1)-th control cycle prediction, and calculate the prediction error corresponding to the (k-1)-th prediction interval. Then, using the vector error in the prediction error as the driving signal, the online recursive learning algorithm is used to perform posterior update of the calibration parameters to obtain the smooth calibration term available for the k-th control cycle.

[0054] 1. Calculation of prediction error The prediction error includes a vector error used for posterior updates in online calibration and a scalar error used for state deviation evaluation or safety interlock determination; the scalar error is a weighted synthesis of the prediction errors of each state component.

[0055] The purpose of the prediction error calculation function in the online calibration module is to: at the beginning of the k-th control cycle, use the already formed measured state vector of the k-th control cycle to verify the predicted state vector of the k-th control cycle stored in the previous control cycle, and construct the vector error used to drive the posterior update of the online calibration and the scalar error used for state deviation evaluation or safety interlock determination, forming a dual-channel error feedback path serving two different purposes. This function is executed before the linear time-varying model construction of the model building module, thus providing the driving signal for the posterior update function of the online calibration module.

[0056] The specific implementation method of the prediction error calculation function of the online calibration module is as follows: First, consider the state vector of a single frame. x i ( k The prediction error is calculated for each of the four components.

[0057] The formula for calculating the prediction error of the column stroke is: e h ( k )= h i meas ( k )- h i pred( k (Unit: mm) Support resistance prediction error e F ( k )= F i meas ( k )- F i pred ( k ), unit kN; Roll attitude angle prediction error e φ ( k )= f i meas ( k )- f i pred ( k ( ), unit: rad; Slip prediction error e s ( k )= s i meas ( k )- s i pred ( k (), unit mm.

[0058] Then, construct the prediction error. e x ( k The prediction error is used to drive the matrix operations of the online recursive learning algorithm. It is necessary to normalize the magnitude of each error component to eliminate the influence of differences in the dimensions of different physical quantities on the numerical stability of the matrix operations. The expression for the prediction error is: e x ( k )=[ e h ( k ), e F ( k ) / 10,100· e φ ( k ), e s ( k )] T ;in e F ( kDividing by 10 is because the typical error level of support resistance is about 10 times the error level of column stroke (kN to mm). e φ ( k Multiplying by 100 is because the typical magnitude of attitude angle error is on the order of rads, while the other components are on the order of mm. The scaling described above makes the four components more consistent in magnitude. The vector error retains the directional information of each component error and is used for the component-by-component recursive update of the calibration parameters.

[0059] Next, construct the scalar error. e i ( k Scalar error is used for determining safety interlocks. It requires compressing multidimensional errors into a single comparable value to simplify the interlock determination logic. The scalar error is obtained by weighted summation of the absolute values ​​of the component errors, and its expression is: e i ( k )= w h ·| e h ( k )|+ w F ·| e F ( k )| / 10+ w φ ·| e φ ( k )|×100+ w s ·| e s ( k The sum of all weights is 1, i.e. w h + w F + w φ + w s =1. Each weight is determined by field testing and calibration; the optimal value is... w h =0.3、 w F =0.2、 w φ =0.4、 w s =0.1. Roll attitude angle component weights w φIt is assigned the highest weight of 0.4 because the roll attitude angle is the core state quantity that directly characterizes the risk of support collapse and has the highest priority in the anti-collapse and anti-slip control of steeply inclined working faces.

[0060] Furthermore, this embodiment defines a preset comprehensive error threshold. e th As a unified benchmark for judging scalar error exceeding limits: e th Predictable deviation Δ based on column stroke h th Allowable prediction deviation of support resistance Δ F th Δ f th And slip allowable prediction deviation Δ s th It is obtained by weighting with the same weight as the scalar error, i.e. e th = w h ·Δ h th + w F ·Δ F th / 10+ w φ ·Δ f th ·100+ w s ·Δ s th Preferred value Δ h th =2mm、Δ F th =10kN, Δ f th =0.01rad, Δ s th =1mm; when scalar error e i (k)> e th When the prediction error exceeds the limit, the result is output to the output constraint and safety interlock module. The vector error constructed by the online calibration module retains direction information to serve the matrix recursive update of calibration parameters, and the scalar error is used to determine the safety interlock or state deviation evaluation through a single threshold comparison.

[0061] 2. Post-calibration parameter update The online calibration module uses the vector error obtained during the prediction error calculation process as the driving signal and adopts an online recursive learning algorithm to update the calibration parameters a posteriori, and calculates the smooth calibration term available in the k-th control cycle. After applying amplitude constraints and rate of change constraints to the smooth calibration term simultaneously, it is stored for use in the linear time-varying model prediction calculation of the model building module, the collaborative control decision generation of the decision generation module, and the state prediction of the state prediction module.

[0062] The purpose of the posterior update function in the online calibration module is to use the vector error obtained during the prediction error calculation process. e x ( k The feedback driving signal is used to learn the calibration parameter matrix through an online recursive learning algorithm. i i ( k The posterior update is performed, and then the available smooth calibration term Δfiuse(k) for the k-th control period is calculated. The smooth calibration term is then stored after being subject to dual constraints of amplitude and rate of change constraints.

[0063] The specific implementation method of the a posteriori update function of the online calibration module is as follows: The online recursive learning algorithm described in this embodiment is mainly a recursive least squares algorithm, which is suitable for prediction errors with linear distribution, and has the advantages of low computational complexity, adjustable forgetting factor, and strong feasibility for real-time industrial control.

[0064] Regression vector f i ( k The input feature vector (4×1) is the recursive least squares algorithm's input feature vector. The regression vector... f i ( k ) is represented as: f i ( k )=[ P ref / P pump ( k ), P roof,i ( k ) / P roof,max sin i ( k ),| u i ( k )- u i ( k -1)| / Ts] TThe fields of the regression vector correspond to the pump pressure ratio, the normalized value of the roof pressure, the sine of the working face inclination angle, and the rate of change of motion, respectively, and are taken from the working condition feature vector constructed by the data preprocessing module. F i ( k The corresponding fields in ) realize the unification of the interface for operating condition perception and calibration parameter calculation.

[0065] Furthermore, the calibration parameters are a 4×4 matrix, with each row corresponding to the regression coefficient of one component of the single-frame state vector; the initial values ​​can be set to a zero matrix or obtained from offline calibration. The recursive update formula is as follows: Gain vector K ( k Calculation of ) K ( k )= P ( k -1)· f i ( k ) / (λ+ f i T (k)· P ( k -1)· f i ( k )); covariance matrix P ( k Update: P ( k )=( P ( k -1)- K (k)· f i T ( k )· P ( k -1)) / λ+ pI ; Parameter matrix i i ( k Update: i i ( k )= i i ( k -1)+ e x ( k )· K T (k); Where: λ=0.98 is the forgetting factor. When λ=0.98, the weight of old data decays to 50% of the initial value after about 35 control cycles, ensuring that the calibration parameters maintain sufficient tracking sensitivity to recent changes in operating conditions. The selectable range of λ is 0.95 to 0.99; ρ=1e -6 This is a regularization parameter used to prevent the covariance matrix from being regularized. P ( k If a numerical singularity occurs during the recursion process, the selectable range of the regularization parameter is 1e. -7 up to 1e -5 ; P ( k The initial value of ) is taken as the diagonal matrix diag([1,1,1,1]); K ( k () is a 4×1 gain vector; e x ( k )· K T (k) is a 4×4 matrix, and i i ( k -1) Consistent dimensions ensure dimensional consistency in recursive operations.

[0066] Furthermore, the calibration term is derived from the product of the parameter matrix and the regression vector: Δ f i ( k )= i i ( k )· f i ( k The dimension is 4×1, and each component corresponds to the correction amount for the prediction deviation of the column stroke, the prediction deviation of the support resistance, the prediction deviation of the roll attitude angle, and the prediction deviation of the slip.

[0067] Then, amplitude constraints and rate of change constraints are simultaneously applied to the calibration term, including: The calculated calibration item Δ f i ( k Simultaneously apply amplitude constraints and rate of change constraints to prevent drastic changes in calibration items due to single large errors or continuous errors in the same direction, and ensure that the calibration process itself does not introduce new control instability factors.

[0068] An upper limit is imposed on the absolute value of each component of the calibration term, i.e., |Δ f i ( k )|≤Δ f max The upper limit vector of the amplitude constraint Δ f maxThe recommended values ​​are [3mm, 10kN, 0.03rad, 2mm]. T Each component can be calibrated and adjusted within ±50% of the recommended value according to the bracket model and on-site working conditions; when a component of the calibration item exceeds the amplitude constraint, the component is truncated to the corresponding upper limit absolute value.

[0069] An upper limit is imposed on the change in the calibration term between adjacent control periods, i.e., |Δ f i ( k )-Δ f i ( k -1)|≤ r f The upper limit vector of the rate of change constraint r f The recommended values ​​are [1 mm / cycle, 5 kN / cycle, 0.01 rad / cycle, 0.5 mm / cycle]. T Each component can be adjusted within ±50% of the recommended value; when the change of a component of the calibration item exceeds the corresponding rate of change constraint, the component will be forcibly corrected to the value of the previous period plus or minus the rate of change constraint limit (the plus or minus sign is determined according to the positive or negative direction of the difference).

[0070] Finally, the smoothed calibration term Δ after double constraint processing f i use ( k The data is stored in the controller memory and sequentially output to the linear time-varying model construction of the model building module, the collaborative control decision generation of the decision generation module, and the single-frame state prediction of the k+1 control cycle of the state prediction module; at the same time, the smooth calibration term participates in the rate of change constraint as a historical calibration term updated a posteriori in the next control cycle.

[0071] The smooth calibration mechanism is the core technical means of this system to suppress control overshoot caused by transient disturbances. It is achieved by specifying that the Δ value in the k-th control period is updated based on the posterior error of the previous prediction interval and processed by amplitude and rate of change constraints. f i use ( k This is achieved through smooth calibration. The physical significance of smooth calibration lies in the fact that the calibration terms used in the current control cycle are derived from the measured feedback of the completed prediction interval, rather than from future measured states that have not yet occurred, thus avoiding the timing contradiction of using unknown measured results to back-calculate the current control quantity within the same control cycle.

[0072] If calibration terms without amplitude and rate-of-change constraints are immediately included in the control calculation, it will cause excessive and sudden adjustments in the control output within a single control cycle, leading to peak impact pressures in the hydraulic system exceeding safe limits and damaging hydraulic components. By combining posterior updates, amplitude constraints, and rate-of-change constraints, the smooth calibration terms available in the k-th control cycle can reflect the true error feedback of the (k-1)-th prediction interval without directly amplifying transient disturbances into abrupt changes in control action within a single control cycle. Taking the roll attitude angle component as an example: if the top plate breaks... f i ( k If the instantaneous jitter reaches 0.05 rad (approximately 2.86°), the attitude angle calibration component obtained from the posterior update may exceed the rate of change constraint threshold. However, after correction through the rate of change constraint, the actual Δ value entering the control calculation will be... f i use ( k The hydraulic shock peak is effectively suppressed as the variation is limited to the allowable range.

[0073] Based on smooth calibration, the gain compensation coefficient E i ( k Simultaneous application of amplitude constraints E i ( k (The absolute value of each element does not exceed 0.8) and the rate of change constraint ( E i ( k (The variation of each element in adjacent periods does not exceed 0.1 / period), further ensuring that the intensity of collaborative compensation within the group does not change abruptly. The update cycle of the smoothing calibration item is 1 control cycle by default; it can be adjusted to the calibration item after weighted averaging over 2 to 3 consecutive control cycles according to the frequency of disturbances on site. The larger the smoothing window, the stronger the smoothing effect, but the more delayed the response to the actual changes in operating conditions. It should be reasonably selected in combination with the actual disturbance frequency on site.

[0074] It should be noted that this embodiment also provides the following two alternative online recursive learning algorithms. If the prediction error exhibits strong nonlinear characteristics (e.g., the residual of the linear time-varying model is large when the working conditions are extremely complex), the following two alternatives can be used to replace the recursive least squares algorithm: Alternative 1 is a lightweight incremental gradient boosting tree algorithm: using a 3-layer decision tree structure, with 10 split nodes in each layer, based on the regression vector. f i ( k ) is the input feature, with the smoothing calibration term Δ f i use ( kThe output is [missing information]; it is suitable for scenarios where the prediction error exhibits a strongly nonlinear distribution and the controller has sufficient computing power; the comprehensive error index [missing information]. e total Equivalent to the scalar error calculated by the online calibration module e i ( k ) and preset comprehensive error threshold e th The ratio, i.e. e total = e i ( k ) / e th When the comprehensive error index e total When the error exceeds 3.0 (i.e., the scalar error exceeds three times the preset comprehensive error threshold), the operating condition is considered to have entered the abnormal zone, and the decision tree model update is frozen to prevent significant parameter drift. Scheme 2 uses the extended Kalman filter algorithm: smoothing the calibration term Δ... f i use ( k This is considered an extended state, and is related to the single-frame state vector. x i ( k Joint estimation; the expanded state vector dimension is 8×1, i.e. x ext ( k )=[ x i ( k ) T , Δ f i use ( k ) T ] T ; Extended coefficient matrix A ext ( k The matrix is ​​8×8, where the first 4 rows and 4 columns are the original time-varying coefficient matrix. A i ( k The last four rows and four columns form the identity matrix; the parameter values ​​of the state noise covariance matrix P0 and the observation noise covariance matrix R should be determined based on the sensor accuracy calibration; this is suitable for scenarios that require simultaneous estimation of model state and calibration parameters. Both of the above alternative schemes require the same amplitude constraints and rate of change constraints to double-limit the smoothing calibration term, and the model update should be frozen synchronously when an interlock is triggered in an abnormal state.

[0075] (iv) Model building module

[0076] The model building module calculates the time-varying coefficients of the discrete-time linear time-varying model in real time based on the working condition feature vector constructed by the data preprocessing module, and constrains the range of values ​​of the time-varying coefficients through the amplitude limiting function. Combined with the smooth calibration term available in the k-th control cycle output by the online calibration module, a discrete-time linear time-varying model that can be used for this control cycle is constructed, which serves as the unified calculation basis for the decision generation module to generate the main control frame control command and the state prediction module to predict the single frame state vector in the k+1-th control cycle.

[0077] The purpose of the model building module is to establish a discrete-time linear time-varying model for each hydraulic support. The time-varying coefficients of this model are updated in real time with the characteristic vector of the current control cycle, so that the model has parameterized adaptive capability to multiple working condition changes. The smoothing calibration terms output by the online calibration module are superimposed on the basic prediction equation in the form of additive correction to obtain the calibrated final prediction model.

[0078] The specific implementation of the model building module is as follows: First, a discrete-time linear time-varying model is established. The basic model expression of the discrete-time linear time-varying model described in this embodiment is: x i pred,base ( k +1)= A i ( k )· x i ( k )+ B i ( k )· u i ( k )+ C i ( k )· W i ( k )+ D i ( k )·M; where, x i pred,base ( k +1) represents the value when the cross-cycle smoothing calibration term is not superimposed, i.e., the... i The basic predicted state vector of the hydraulic support in the (k+1)th control cycle; x i ( k () is the state vector of a single frame, a 4×1 column vector, defined as: x i ( k )=[ hi ( k ), F i ( k ), f i ( k ), s i ( k )] T , h i ( k ) represents the column stroke of the i-th hydraulic support in the k-th control cycle, in mm; F i ( k ) represents the support resistance of the i-th hydraulic support in the k-th control cycle, in kN; f i ( k ) represents the roll attitude angle of the i-th hydraulic support in the k-th control cycle, in rad; s i ( k ) represents the slippage of the i-th hydraulic support in the k-th control cycle, in mm; h i ( k ), F i ( k ), f i ( k ), s i ( k ) is the state vector of a single frame. x i ( k The four components in the equation correspond to four physical states in the anti-tipping and anti-slip control of the hydraulic support: the roll attitude angle component corresponds to the anti-tipping requirement, the slippage component corresponds to the anti-slip requirement, and the column stroke and support resistance components together characterize the roof support state; furthermore, u i ( k () is the control input for a single hydraulic support, including valve-controlled type (valve opening percentage) and target quantity type (target stroke increment), in mm; W i ( kThe input is the intra-group collaborative input, with a dimension of 4×1, generated by the control group partitioning module and the decision generation module; M is the action compensation constant vector, with a dimension of 4×1, used to compensate for relatively stable fixed bias errors (such as systematic constant deviations caused by uneven support installation, inherent zero drift of the hydraulic system, etc.); the action compensation constant vector M should be determined after the support is installed and leveled through no-load trial operation or historical stable operating condition average calibration. For example, M can be taken as [0.5mm, 2.0kN, 0.01rad, 0.3mm]. T ; A i ( k ) is the state transition coefficient matrix, used to characterize the natural evolutionary influence of the current state on the next cycle state. The state transition coefficient matrix is ​​a 4×4 matrix. B i ( k ) is the control input coefficient matrix, used to characterize the driving effect of the control input on the state. When there is a single-channel control input, the control input coefficient matrix is ​​a 4×1 matrix, and when there is a multi-channel control input, the control input coefficient matrix is ​​a 4×m matrix, where m is the number of control input channels; C i ( k ) is the collaborative input coefficient matrix, used to characterize the influence of intra-group collaborative input on the state of a single frame. The collaborative input coefficient matrix is ​​a 4×4 matrix. D i ( k The matrix ) represents the slow-varying correction coefficient matrix, used to characterize the additional effects of slow-varying operating conditions such as roof pressure cycle, coal mining machine position, and roof and floor crushing on the model. The slow-varying correction coefficient matrix is ​​a 4×4 matrix. A i ( k ), C i ( k )and D i ( k The preferred matrix is ​​a diagonal matrix, with each diagonal element corresponding to one of the four state components: column stroke, support resistance, roll attitude angle, and slip. Other diagonal elements, except for the core diagonal element, can be calculated using the same mapping relationship based on the corresponding state components. Off-diagonal elements can be set to 0 or calibrated as coupling coefficients based on historical stable operating conditions. B i ( k Each element corresponds to the driving effect of the control input on the column stroke, support resistance, roll attitude angle and slip amount, which can be identified through on-site step response test or historical control data.

[0079] Furthermore, the four coefficient matrices of the time-varying coefficient discrete-time linear time-varying model are all calculated using the operating condition eigenvectors and uniformly expressed using the clipping function (clip). z , z min , z max )=max( z min ,min( z , z max Constrain the range of values ​​for each element to prevent abnormal coefficient calculation results from causing model divergence. The specific calculation method is as follows: (1) Calculation method of state transition coefficient State transition coefficients A i ( k The core diagonal element A) i,11 (k) and top plate pressure P roof,i ( k The relationship is inverse: the greater the pressure on the roof, the more... A i,11 ( k The smaller the value, the lower the inertia weight of the prediction calculation for the current state, and the more conservative the control strategy. The calculation formula is as follows: A i,11 ( k )=clip(α / ( P roof,i ( k )+ε),0.5,1.0), where α=8.0, ε=0.1MPa (ε is used to prevent the denominator from being zero); the clip function will A i,11 ( k The constraints are within the range of [0.5, 1.0]. The lower limit of 0.5 prevents the coefficient from being too small, which would cause the model to be insufficient in response to the current state. The upper limit of 1.0 corresponds to the extreme case when the pressure on the top plate approaches zero. Here, α is the state transition proportionality coefficient, which is uniformly taken as 8.0 in this embodiment as an engineering example value and is determined by the calibration results of historical stable working conditions.

[0080] (2) Calculation method of control input coefficient Control input coefficients B i ( k The core elements of ) B i,11 ( k ) and pump station pressure supply P pump ( k Negative correlation: The lower the pump station's supply pressure, Bi,11 ( k The larger the value, the better to compensate for the lag in hydraulic system operation caused by insufficient pressure supply. The calculation formula is as follows: B i,11 ( k =clip(β· P ref / ( P pump ( k )+ε),1.2,2.0), where β=1.5, P ref =31.5MPa; the clip function will B i,11 ( k The constraint is within the range of [1.2, 2.0], and this range is determined based on the calibration of the hydraulic dynamic characteristics.

[0081] (3) Cooperative input coefficient Cooperative input coefficients C i ( k The core elements of ) C i,11 ( k ( ) and working face inclination angle i ( k and roll attitude angle f i ( k Related: The larger the tilt angle and the more severe the deviation of the attitude angle from zero, the stronger the synergistic effect. The calculation formula is as follows: C i,11 ( k )=clip(1+γ·sin i ( k )·sign( f i ( k ),0.8,1.5), where γ=0.5; the clip function will C i,11 ( k The constraint is within the range of [0.8, 1.5].

[0082] (4) Slow variation correction coefficient Slow variation correction coefficient D i ( k The core elements of ) D i,11 (k) Cycle index is suppressed by the top plate. cycle idx ( k Coal mining machine location shearer pos ( k) and the fracture index of top and bottom plates broken idx ( k The three-level piecewise function g, composed of , is determined as follows: D i,11 ( k )=1+γ d ·g( cycle idx ( k ), shearer pos ( k ), broken idx ( k )), where γ d =0.2; The rule for the value of the three-level piecewise function g is: when cycle idx ( k )=2、 shearer pos ( k <5m and broken idx ( k When ≥0.7, g=2 (corresponding to the worst combined working condition of peak pressure, coal mining machine in close proximity, and extremely broken roof and floor plates); when cycle idx ( k )=0、 shearer pos ( k >10m and broken idx ( k When g < 0.3, g = 0 (corresponding to the optimal working condition of stable roof, coal mining machine far away, and intact roof and floor); in other cases, g = 1. The threshold values ​​for each level are preferred values ​​and can be adjusted through on-site historical data statistics or calibration tests.

[0083] Then, a smoothing calibration term available for the k-th control cycle, output by the online calibration module, is introduced on top of the basic model to obtain the final discrete-time linear time-varying model, expressed as: x i pred ( k +1)= x i pred,base (k+1)+Δ f i use ( k ).in, x i pred ( k +1) is the final predicted state vector of the i-th hydraulic support in the k+1 control cycle after superimposing the smoothing calibration term; Δf i use ( k ) is the smooth calibration term available in the k-th control cycle. The smooth calibration term is obtained by the online recursive learning algorithm in the online calibration module, which performs posterior updates based on the prediction error corresponding to the (k-1)-th prediction interval and processes it with amplitude constraints and rate of change constraints.

[0084] The discrete-time linear time-varying model constructed in the above manner updates its time-varying coefficient matrix in real time with the characteristic vector of the operating condition in each control cycle. This enables the model to have an inherent parameterized adaptive capability to various operating condition disturbances such as pump station pressure fluctuations, working face inclination changes, and roof pressure intensity changes, without the need for offline recalibration. At the same time, the linear structure of the linear time-varying model ensures the interpretability of the model, facilitates the online recursive learning algorithm to recursively update the calibration parameters, and has a computational complexity suitable for online operation in industrial real-time control environments.

[0085] (v) Control Group Division Module and Decision Generation Module

[0086] After the model building module constructs a discrete-time linear time-varying model that can be used in the k-th control cycle, the control group division module divides the hydraulic supports on the working face into multiple groups according to the principle of adjacent grouping. Each group contains an even number of supports, with odd-numbered supports in the group being the main control frame and even-numbered supports being the follower frames. The decision generation module enables the main control frame to generate control commands based on its own discrete-time linear time-varying model and the smoothing calibration terms available in the k-th control cycle. Based on the difference between the state vectors of the main control frame and the follower frames, and the center distance deviation characterized by the difference in slippage between the main control frame and the follower frames, the module generates intra-group cooperative inputs and superimposes them onto the cooperative input terms of the main control frame model. The follower frames generate follower frame compensation control inputs according to the compensation amount, which is jointly determined by the gain compensation coefficient, the main control frame state vector, the working condition correction parameter, and the smoothing calibration terms available in the k-th control cycle of the follower frames. The decision generation module synthesizes the main control frame control commands, intra-group cooperative inputs, and follower frame compensation control inputs into a cooperative control decision.

[0087] The purpose of the control group division module and the decision generation module is to, on the basis of single-frame closed-loop control, use even-number scale adjacent groups, master-slave mapping structure and collaborative compensation control based on the state difference within the group to bring the key states of each frame in the group toward consistency in terms of displacement, attitude, support resistance and center distance, so as to prevent the radical actions of a single frame from evolving into chain instability within the group through hydraulic coupling and surrounding rock transmission.

[0088] The specific implementation methods of the control group division module and the decision generation module are as follows: In this embodiment, the center distance management of adjacent supports is achieved through center distance deviation. The center distance is the projected distance between the preset reference points of the main control frame and the follower frame along the arrangement direction of adjacent supports. To eliminate ambiguity, this embodiment further clarifies that the "center distance management direction" is defined as the arrangement direction of adjacent supports, i.e., the direction of the line connecting the geometric centers of adjacent supports arranged along the working face advancing direction. The center distance deviation is the difference between the real-time center distance and the installation calibration center distance. The real-time center distance can be calculated based on the initial installation center distance between the main control frame and the follower frame, the stroke of the pushing cylinder, the sliding amount, and the support attitude angle. In a simplified implementation, the center distance deviation can be obtained from the projection difference of the sliding amount between the main control frame and the follower frame in the center distance management direction. By using the center distance deviation as a component of the group's collaborative input, the follower frame compensation amount can respond not only to the column stroke, support resistance, and attitude angle differences, but also to the center distance deviation between adjacent supports, thereby promoting the convergence of the center distance of the supports within the group towards the installation calibration center distance.

[0089] Specifically, the center distance deviation can be expressed as: Δ d m,s ( k )= d m,s ( k )- d m,s 0 , where Δ d m,s ( k The deviation between the center distances of the main control frame and the follower frame during the k-th control cycle is denoted as . d m,s ( k () represents the real-time center distance. d m,s 0 To calibrate the center distance during installation. In a simplified implementation, the center distance deviation can be expressed as: Δ d m,s ( k )= s s ( k )- s m ( k ),in, s s ( k sm(k) represents the projection of the sliding amount of the follower frame in the center distance management direction, and sm(k) represents the projection of the sliding amount of the main control frame in the center distance management direction. The real-time center distance is calculated using the following formula: d m,s ( k )= d m,s 0 +( Lpush,s ( k )− L push,m ( k ))−( s s ( k )− s m ( k ))· cosθ ( k ) +Δ att ( k ),in L push,m ( k ) is the stroke of the main control frame push cylinder. L push,s ( k ) represents the stroke of the follower frame pushing cylinder, Δ att ( k The term ) is a geometric correction introduced by the attitude angle difference between the main control frame and the follower frame, which can be calibrated by the geometric projection method.

[0090] First, select the support group configuration and group size. Divide the hydraulic supports of the entire working face into several control groups according to the principle of adjacent grouping. Each group has an even number of supports (selectable as 2, 4, 6, 8, or 10 supports). Different group sizes are suitable for different levels of working condition complexity: a 2-support group is suitable for working faces with stable geology and low pressure intensity (working face inclination angle between 35° and 45°, and relatively intact roof and floor slabs). At this size, the response speed within the group is fast, suitable for relatively simple working conditions; a 4-6 support group is suitable for moderately complex working conditions (working face inclination angle between 45° and 55°, and roof pressure cycle between 10 and 20 days), achieving a good balance between stability and response speed; a 8-10 support group is suitable for working faces with complex geology and high pressure intensity (working face inclination angle not less than 55°, and broken roof and floor slabs). The larger group size improves the overall anti-disturbance capability, allowing more adjacent supports to be included in the coordinated control range. For short working faces (length not exceeding 100m), groups of 2 to 4 frames are preferred; for long working faces (length not less than 200m), groups of 6 to 10 frames are preferred.

[0091] Then, a master-slave mapping is constructed. Within each group, odd-numbered supports are selected as master control frames, and even-numbered supports as follower frames. Taking a group of 4 frames as an example, the 1st and 3rd frames are master control frames, and the 2nd and 4th frames are follower frames, forming a two-to-one master-slave mapping structure where "the 1st master control frame corresponds to the 2nd follower frame, and the 3rd master control frame corresponds to the 4th follower frame." Taking a group of 8 frames as an example, a "4 master, 4 slave" structure is formed according to the odd-even numbering mapping rule. The master control frame is responsible for receiving global operating condition information, independently generating control commands based on its own linear time-varying model and online calibration, and sending group-wide collaborative information to the follower frames. The follower frames adjust their actions based on their own state and the master control frame's collaborative commands. Even-numbered grouping ensures an equal number of master control frames and follower frames within each group, simplifying the logic of collaborative control. The adjacent grouping strategy utilizes the strong force correlation between adjacent supports, maximizing the collaborative control effect.

[0092] Subsequently, the state differences within the group are calculated. In each control cycle, the differences of four state components are calculated for each pair of master and follower frames within the group: Column travel difference Δ h = h main ( k )- h slave ( k (Unit: mm) Support resistance difference Δ F = F main ( k )- F slave ( k ), unit kN; Roll attitude angle difference Δ f = f main ( k )- f slave ( k ( ), unit: rad; Center distance deviation Δ, characterized by slip difference s = s main ( k )- s slave ( k (), unit mm.

[0093] The center distance deviation Δ is characterized by the difference in the projection of the slippage between the main control frame and the follower frame in the center distance management direction. d = s slave ( k )- s main( k (), unit mm.

[0094] It should be noted that the formula Δ s = s main ( k )- s slave ( k ) and Δ d = s slave ( k )- s main ( k The signs are opposite but the physical meanings are different: Δ s The difference in slippage between the main control frame and the follower frame is used as a state difference component to participate in intra-group coordination; Δ d The center distance deviation relative to the follower frame is denoted by Δ, which is the deviation of the follower frame from the downhill direction. d >0, facilitating synergistic gain K d Always take positive values.

[0095] Next, generate collaborative input within the group. W i ( k Collaborative input within a group. W i ( k The vector is 4×1, and each component is obtained by multiplying the corresponding state difference by the cooperative gain coefficient. W i ( k )=[ K h ·Δ h , K F ·ΔF, K φ ·Δφ, K d ·Δ d ] T The synergistic gain coefficient was calibrated by field tests: Preferred... K h =0.2 (displacement difference cooperative gain) K F =0.1 (resistance difference synergistic gain) K φ =0.3 (Attitude difference cooperative gain) K d =0.15 (center distance deviation collaborative gain), the calibration target is to ensure that the state difference within the group converges to Δh≤2mm, Δ F≤5kN, Δφ≤0.01rad (approximately 0.57°), Δ d Within the allowable range of ≤1mm. K φ The maximum value reflects the priority of roll attitude difference in anti-tipping control. When the main control frame column stroke change rate exceeds 1mm / cycle and the working surface tilt angle θ exceeds 30°, the system automatically... K h , K φ , K d Each will be increased by 50% to enhance synergy when operating conditions change rapidly. For groups of 8 to 10 aircraft, the range of differences within the group will increase. K h It can be adjusted up to 0.3. K φ This can be adjusted up to 0.4 to ensure that the state difference still converges effectively with larger group sizes. (Generated intra-group co-input) W i ( k Simultaneously, the co-input terms are superimposed on the fundamental prediction equations of the master control frame linear time-varying model. C i ( k )· W i ( k It participates in the state prediction calculation of the main control frame, so that the prediction calculation of the main control frame itself contains the intra-group coupling effect.

[0096] Finally, calculate the compensation amount of the follower frame. Y slave ( k ). Follower frame compensation amount Y slave ( k () is a 4×1 vector, and the calculation formula is: Y slave ( k )= E i ( k )· x main ( k )+ V i ( k )+Δ f slave use ( k ).in: x main ( k ) is the main control frame state vector; E i ( kThe matrix is ​​a 4×4 gain compensation coefficient matrix, where the absolute value of each element does not exceed 0.8 (amplitude constraint upper limit), and the change in each element between adjacent periods does not exceed 0.1 / period (change rate constraint upper limit). V i ( k The vector of working condition correction parameters (4×1) is used to correct fixed deviations caused by slowly varying working conditions such as the roof pressure cycle. For the peak pressure period, it is recommended to use [value missing]. V i ( k = [1mm, 3kN, 0.02rad, 0.5mm] T During a stable period, it is recommended to take V i ( k =[0,0,0,0] T ;Δ f slave use ( k ) represents the smoothing calibration term available to the servo in the k-th control cycle. The servo will compensate for the amount... Y slave ( k The input is superimposed on its own control reference input to generate the follower frame compensation control input, realizing the group linkage control logic of "master control guidance, follower tracking, and difference compensation".

[0097] The initial value of Ei(k) is taken as a diagonal matrix diag([0.3,0.2,0.4,0.3]), and the diagonal elements are related to the cooperative gain coefficients. K h , K F , K φ , K d The optimal value remains on the same order of magnitude; during operation, it follows the recursive formula. Hey ( k )= Hey ( k -1)+ or · e slave ( k )· x main T ( k Update, in which or The learning rate is preferably 0.01. e slave ( k ) represents the prediction error vector of the servo frame; after updating, the above amplitude and rate of change constraints are applied. Vi(k) is the periodic index based on the top plate pressure. cycle idx ( kValues: cycle idx =0 (stable period) takes [0,0,0,0]T; cycle idx =1 (Initial pressure) Take [0.5mm, 1.5kN, 0.01rad, 0.25mm] T ; cycle idx =2 (peak pressure period) is taken as [1mm, 3kN, 0.02rad, 0.5mm] T .

[0098] Finally, the main control rack control commands and intra-group collaborative inputs will be integrated. W i ( k The inputs of the control group and the servo frame compensation control are combined to form the collaborative control decision of the control group in the kth control cycle, and output to the collaborative control module and the state prediction module.

[0099] (vi) Cooperative Control Module and State Prediction Module

[0100] During the k-th control cycle, the collaborative control module performs linkage control on the hydraulic support of the working face according to the collaborative control decision generated by the decision generation module. At the same time, during the k-th control cycle, the state prediction module uses the time-varying coefficient matrix of the discrete-time linear time-varying model constructed by the model building module, the single-frame control input, the intra-group collaborative input, and the smoothing calibration term available in the k-th control cycle output by the online calibration module to predict the single-frame state vector of the k+1-th control cycle, obtain the predicted state vector of the k+1-th control cycle, and store the predicted state vector of the k+1-th control cycle.

[0101] Specifically, the collaborative control module and the state prediction module both contain the following two parallel sub-processes: (1) Collaborative Control Execution Sub-process: The collaborative control decision (including main control frame control commands, intra-group collaborative inputs, and follower frame compensation control inputs) synthesized by the control group division module and the decision generation module is sent to the electro-hydraulic control valve group of each hydraulic support. The electro-hydraulic control valve group drives the main control frame and follower frame's columns, push cylinders, and other actuators to complete the corresponding actions in the k-th control cycle according to the collaborative control decision. During the execution process, the output constraint and safety interlock module monitors the changes in top plate pressure in adjacent control cycles, sensor outputs, communication status, pump station pressure supply, and scalar prediction errors in real time. When any abnormality is triggered, the output constraint and safety interlock module outputs an interlock signal to the collaborative control module, causing it to switch to a conservative control loop and triggering an audible and visual alarm.

[0102] (2) Next cycle state prediction sub-process: The measured state vector of the k-th control cycle is... x imeas ( k ), Single-frame control input u i ( k Collaborative input within groups W i ( k Substituting the equations into the basic prediction equations of the discrete-time linear time-varying model constructed by the model building module, we obtain the basic prediction state vector. x i pred,base ( k +1); then superimpose the smoothing calibration term Δ available in the kth control cycle output by the online calibration module. f i use ( k This yields the final predicted state vector. x i pred ( k +1); and will x i pred ( k +1) Stored in the controller memory, it is used to compare the measured state vector formed at the beginning of the (k+1)th control cycle with that formed at the end of the kth control cycle, thereby completing the closed-loop timing sequence of "acquisition and preprocessing → error calculation → online calibration → model building → grouping → collaborative decision-making → linkage control and prediction of the state of the next cycle".

[0103] Through the above cycle, the system forms a closed-loop time sequence of "prediction-actual verification-post-calibration-model building-collaborative decision-prediction in the next cycle", achieving simultaneous improvement in the prediction accuracy of a single support frame and the collaborative stability within the group.

[0104] (vii) Output constraint and safety interlock module The system receives sensor health status from the data preprocessing module, comparison results of scalar prediction error and preset comprehensive error threshold from the online calibration module, and information such as pump station pressure and communication status from global operating conditions. This information is used to monitor five types of anomalies during the k-th control cycle and output an interlock signal when any of the following anomalies occur: (1) the change in top plate pressure exceeds 5 MPa in adjacent control cycles; (2) the sensor output exceeds its range or remains at a fixed value for three consecutive control cycles; (3) the communication packet loss rate exceeds 30% or there is no data transmission for two consecutive control cycles; (4) the pump station pressure is below 15 MPa; (5) the scalar prediction error exceeds the preset comprehensive error threshold. The output constraint and safety interlock module includes a relay for switching conservative control loops and an alarm for issuing audible and visual alarms.

[0105] Example 2: A linkage control method for center distance management of hydraulic supports on steeply inclined working faces, corresponding to the system in Example 1, is provided, in which S1 to S8 are executed at the beginning of the kth control cycle; S1: Data Acquisition.

[0106] Collect the measured state vector of each support and the global working condition vector of the working face after the (k-1)th control cycle. S2: Data preprocessing and construction of working condition feature vectors.

[0107] Construct a condition feature vector based on the measured state vector and the global condition vector; S3: Calculate the prediction error.

[0108] The measured state vector is compared with the predicted state vector of the k-th control period obtained from the prediction of the (k-1)-th control period to obtain the prediction error. S4: Drive a posteriori update of calibration parameters with vector error.

[0109] Using the vector error in the prediction error as the driving signal, an online recursive learning algorithm is used to update the calibration parameters posteriorly to obtain the smooth calibration term available in the kth control cycle. S5: Construct a discrete-time linear time-varying model.

[0110] The time-varying coefficient matrix is ​​calculated in real time based on the operating condition feature vector, and a discrete-time linear time-varying model is constructed using the time-varying coefficient matrix. S6: Grouping.

[0111] The hydraulic supports at the working face are divided into multiple control groups according to the principle of adjacent grouping; the odd-numbered supports in each group are the main control supports and the even-numbered supports are the follow-up supports.

[0112] S7: Generate collaborative control decisions.

[0113] Based on the discrete-time linear time-varying model and the smoothing calibration term available in the k-th control cycle, the master control frame control input is generated; based on the state vector difference and center distance deviation between the master control frame and the follower frame, the intra-group cooperative input is generated; the follower frame compensation control input is generated according to the follower frame compensation amount; and a cooperative control decision including the master control frame control input, the intra-group cooperative input, and the follower frame compensation control input is generated.

[0114] S8: Execute linkage control and predict the state of the (k+1)th control cycle.

[0115] During the k-th control cycle, coordinated control is performed based on the collaborative control decision. Simultaneously, the state vector of a single frame in the k+1-th control cycle is predicted using the time-varying coefficient matrix of the k-th control cycle, the main frame control input, the intra-group collaborative input, and the smoothing calibration terms available in the k-th control cycle. This predicts the state vector of the k+1-th control cycle, which is then used to compare the predicted state vector at the beginning of the k+1-th control cycle with the measured state vector formed after the end of the k-th control cycle.

[0116] In the method described in this embodiment, the specific implementation of S1-S8 corresponds to Embodiment 1, and will not be repeated in this embodiment.

[0117] Example 3: Based on the system provided in Example 1 and the method provided in Example 2, this example provides a computer device that executes the method described in Example 2, including a memory, a processor, and a transceiver connected in sequence. The memory stores a computer program, the transceiver sends and receives messages, and the processor reads the computer program and executes the method described in Example 2 or any method that may involve the method described in Example 2. Specifically, the memory may include, but is not limited to, random-access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or last-in-first-out (FILO) memory, etc.; the processor may include, but is not limited to, a microprocessor of the STM32F105 series. Furthermore, the computer device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0118] The working process, working details and technical effects of the aforementioned computer device provided in this embodiment can be found in the method described in Embodiment 2, and will not be repeated here.

[0119] Example 4: This example provides a computer-readable storage medium that stores instructions that include the method described in Example 2 or any other method that may involve the method described in Example 2. Specifically, the computer-readable storage medium stores instructions that, when executed on a computer, perform the method described in Example 2. The computer-readable storage medium refers to a carrier for storing data, and may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0120] The working process, working details and technical effects of the aforementioned computer-readable storage medium provided in this embodiment can be found in the method described in Embodiment 2, and will not be repeated here.

[0121] Example 5: This example provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the method described in Example 2. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0122] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A center distance management and linkage control system for hydraulic supports on steeply inclined working faces, characterized in that, include: The data acquisition module is used to acquire the measured state vector of each support and the global working condition vector of the working face at the beginning of the k-1 control cycle. The data preprocessing module is used to perform timestamp alignment, outlier removal, and signal filtering on the measured state vector and global operating condition vector output by the data acquisition module, and to construct operating condition feature vectors based on the preprocessed data. The online calibration module is used to compare the measured state vector with the predicted state vector of the kth control period obtained from the (k-1)th control period to obtain the prediction error. The vector error in the prediction error is used as the driving signal, and the calibration parameters are updated posteriorly using an online recursive learning algorithm. After amplitude constraint and rate of change constraint processing, a smooth calibration term usable in the kth control period is obtained. The model building module is used to calculate the time-varying coefficient matrix in real time based on the working condition feature vector, constrain the value range of each time-varying coefficient in the time-varying coefficient matrix through the amplitude limiting function, and construct a discrete-time linear time-varying model. The control group division module is used to divide the hydraulic supports of the working face into multiple control groups according to the principle of adjacent grouping. In each group, the odd-numbered supports are the main control frames and the even-numbered supports are the follower frames. The decision generation module is used to generate the master control frame control input based on the discrete-time linear time-varying model and smoothing calibration term, generate the intra-group cooperative input based on the state vector difference and center distance deviation between the master control frame and the follower frame, generate the follower frame compensation control input according to the follower frame compensation amount, and generate a cooperative control decision that includes the master control frame control input, the intra-group cooperative input and the follower frame compensation control input. The collaborative control module is used to perform linkage control on the hydraulic support of the working face according to the collaborative control decision during the kth control cycle. The state prediction module is used to predict the state vector of a single frame in the (k+1)th control cycle by using the time-varying coefficient matrix, single frame control input, intra-group cooperative input, and smoothing calibration term. The predicted state vector of the (k+1)th control cycle is then used to compare the predicted state vector at the beginning of the (k+1)th control cycle with the measured state vector formed at the end of the kth control cycle.

2. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 1, characterized in that, When the data preprocessing module performs timestamp alignment, it uses the start time of the control cycle as the reference. For sensors with a sampling frequency higher than the control cycle frequency, it takes the arithmetic mean of each sampling point within the control cycle. For sensors with a sampling frequency lower than the control cycle frequency, it uses interpolation to supplement missing data. When the data preprocessing module removes outliers, it uses the 3σ criterion or the interquartile range criterion for detection within a sliding window. The sampling value of the outlier is replaced with the effective value of the previous control cycle. When an outlier appears in the same sensor for a set number of consecutive control cycles, a sensor abnormality alarm is triggered. The data preprocessing module uses first-order low-pass filtering or moving average filtering to filter the pressure and displacement signals.

3. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 1, characterized in that, The measured state vector includes column stroke, support resistance, roll attitude angle and slippage; the global working condition vector includes mandatory working condition quantity and extended working condition quantity. The mandatory working condition quantity includes pump station pressure supply and working face dynamic inclination angle. The extended working condition quantity includes at least one of roof pressure cycle index, coal mining machine position and roof and floor crushing index. The working condition feature vector includes pump pressure ratio, normalized value of roof pressure, sine value of working face inclination angle, sign term of roll attitude angle, coal mining machine position segmentation and motion change rate; the time-varying coefficient matrix is ​​associated with the current roof pressure, pump station pressure supply, working face inclination angle, roll attitude angle and slow-changing working condition indicators.

4. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 1, characterized in that, The basic model expression of the discrete-time linear time-varying model constructed by the model building module is as follows: x i pred,base (k+1)= A i ( k )· x i meas ( k )+ B i ( k )· u i ( k )+ C i ( k )· W i ( k )+ D i ( k )·M; in, x i pred,base (k+1) is the basic predicted state vector of the i-th hydraulic support in the k+1 control cycle without the superimposed smoothing calibration term. A i ( k ) is the state transition coefficient matrix, x i meas ( k () represents the measured state vector of the i-th hydraulic support obtained at the start of the k-th control cycle. B i ( k () represents the coefficient matrix of the single-frame control input. u i ( k ) represents the single-frame control input for the i-th hydraulic support in the k-th control cycle. C i ( k ) represents the collaborative input coefficient matrix. W i ( k (This is for collaborative input within the group.) D i ( k ) represents the slow-varying correction coefficient matrix, and M represents the action compensation constant vector; the state prediction module uses a smoothing calibration term to correct the basic model expression, and the corrected prediction equation is: x i pred (k+1)= x i pred,base (k+1)+Δ f i use ( k );in, x i pred (k+1) is the final predicted state vector of the i-th hydraulic support in the k+1 control cycle after the superposition of the smoothing calibration term, Δ f i use (k) is the smoothing calibration term available in the k-th control cycle.

5. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 4, characterized in that, When the model building module calculates the time-varying coefficient matrix: it calculates the state transition coefficient matrix based on the current roof pressure. A i ( k The core element of the state transition coefficient matrix is ​​clipped using the clipping function. A i ( k The core elements of the equation are constrained within the range of [0.5, 1.0]; among them, the state transition coefficient matrix... A i ( k The core element decreases as the current pressure on the top plate increases; The coefficient matrix of the single-frame control input is calculated based on the current pump station pressure supply. B i ( k The core element of the control is the coefficient matrix of the single-frame control input, which is clipped using the clipping function. B i ( k The core elements of the control are constrained within the range of [1.2, 2.0]; among them, the coefficient matrix of the single-frame control input is... B i ( k The core element of the pump increases as the current pump station pressure decreases; Calculate the cooperative input coefficient matrix based on the current working face inclination angle and the current roll attitude angle. C i ( k The core element of the ) is to clip the cooperative input coefficient matrix through the clipping function. C i ( k The core elements are constrained within the range of [0.8, 1.5]. Based on the roof pressure cycle index, coal mining machine position, and roof and floor crushing index, a three-stage piecewise function is determined, and a slow-variable correction coefficient matrix is ​​calculated based on the three-stage piecewise function. D i ( k The core element of ).

6. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 1, characterized in that, The prediction errors obtained by the online calibration module include vector errors and scalar errors. The vector errors are obtained by normalizing the magnitude of the prediction errors of each predicted state vector component and then concatenating them. This is used to drive the matrix operations of the online recursive learning algorithm. The scalar errors are obtained by weighted summation of the column travel prediction error, support resistance prediction error, roll attitude angle prediction error, and slip prediction error according to their respective weights. The sum of each weight is 1, and the weight corresponding to the roll attitude angle prediction error is greater than the weights of the other three components.

7. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 6, characterized in that, The online calibration module is also used to determine the prediction error exceeds the limit and output the prediction error exceeds the limit signal when the scalar error is greater than the preset comprehensive error threshold. The preset comprehensive error threshold is obtained by weighting the allowable prediction deviations of the four components, namely column stroke, support resistance, roll attitude angle and slip, with the same weight as the scalar error.

8. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 1, characterized in that, The online calibration module employs a recursive least squares algorithm for its online recursive learning. The module constructs a regression vector using the pump pressure ratio, normalized top pressure, sine of the working face inclination angle, and the rate of change of control input actions. A forgetting factor and regularization parameters are introduced into the recursive formula for updating the calibration parameters, and the parameter matrix is ​​updated using a matrix-based recursive formula. The calibration term for the k-th control cycle is calculated based on the product of the updated parameter matrix and the regression vector. Amplitude constraints and rate of change constraints are then applied sequentially to the calibration term to obtain a smooth calibration term usable in the k-th control cycle. Specifically, the amplitude constraint truncates each component of the calibration term according to the upper limit vector of the amplitude constraint. The rate of change constraint corrects the component when the absolute value of the difference between any component of the calibration term and the corresponding component of the smooth calibration term usable in the (k-1)-th control cycle exceeds the corresponding rate of change constraint limit. The correction is then the value of the corresponding component of the smooth calibration term usable in the (k-1)-th control cycle plus or minus the corresponding rate of change constraint limit.

9. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 8, characterized in that, When the prediction error exhibits a nonlinear distribution, the online calibration module replaces the recursive least squares algorithm with the extended Kalman filter algorithm, and uses the same amplitude and rate of change constraints as the recursive least squares algorithm to process the smoothing calibration term.

10. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 1, characterized in that, The decision generation module generates a 4×1 vector for intra-group collaborative input. Each component of the intra-group collaborative input is determined by multiplying the column stroke difference, support resistance difference, roll attitude angle difference, and center distance deviation (characterized by slip difference) between the main control frame and the follower frame by their respective collaborative gain coefficients. The center distance deviation is calculated by the difference in slip projection between the main control frame and the follower frame in the center distance management direction. The center distance management direction is either the arrangement direction of adjacent hydraulic supports or the direction of the line connecting the preset reference points of adjacent hydraulic supports. The follower frame compensation amount is calculated using the formula... Y slave ( k )= E i ( k )· x main ( k )+ V i ( k )+Δ f slave use ( k ) Determined; among which, Y slave ( k ) is the compensation vector of the follower in the i-th control group during the k-th control cycle; x main ( k ) represents the measured state vector of the master control frame in the same control group during the kth control cycle; E i ( k ) is the gain compensation coefficient matrix. E i ( k Subject to amplitude and rate of change constraints; V i ( k ) represents the working condition correction parameter vector for the i-th control group in the k-th control cycle, used to correct fixed deviations caused by slowly changing working conditions such as the roof pressure cycle; Δ f slave use ( k ) is the smooth calibration term available for the servo frame in the kth control cycle.

11. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 1, characterized in that, The data acquisition module includes a displacement sensor installed on the lower surface of the top beam to measure the stroke of the column, a pressure transmitter installed at the cylinder inlet of the column to measure the pressure of the column circuit, an inertial measurement unit installed at the center of gravity of the top beam to measure the roll attitude angle, a displacement sensor installed on the cylinder body of the push cylinder to measure the stroke of the push cylinder to calculate the slip, a pressure sensor installed on the pump station outlet main pipe to measure the pump station supply pressure, tilt sensors distributed along the working face to measure the dynamic tilt angle of the working face, and an underground positioning module for obtaining the position of the coal mining machine.

12. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 1, characterized in that, The control group division module supports the configuration of each group of supports in the form of 2, 4, 6, 8 or 10 supports.

13. The center distance management and linkage control system for hydraulic supports on steeply inclined working faces according to claim 1, characterized in that, It also includes an output constraint and safety interlock module, which is used to monitor abnormalities during the k-th control cycle and output an interlock signal to the collaborative control module when any of the following abnormalities occur, causing the collaborative control module to switch to a conservative control loop and trigger an audible and visual alarm: the change in top plate pressure in adjacent control cycles exceeds the set pressure change threshold; the sensor output exceeds the range or is a fixed value for a set number of consecutive control cycles; the communication packet loss rate exceeds the set packet loss threshold or there is no data transmission for a set number of consecutive control cycles; the pump station pressure is lower than the set lower limit of the pressure supply; the prediction error exceeds the set prediction error threshold.

Images