Slurry and plugging linkage closed-loop control method for mine anchor cable grouting working condition

Abstract

This invention relates to the field of grouting linkage closed-loop control technology, specifically to a grout and plugging linkage closed-loop control method for mine anchor cable grouting conditions. The method includes the following steps: In this invention, by calling a proportional-integral-derivative (PID) controller, the gain coefficient and disturbance variable are calculated by superimposing the control error and the calculus term, directly eliminating the sudden load interference caused by high-pressure stress deformation in soft rock, ensuring smooth and stable closed-loop feedback voltage output. A long short-term memory (LSTM) network is used to merge flow resistance trends and time delay characteristics, setting frequency and opening variables to calculate the intercept update reference point. Multi-objective optimization and quantitative evaluation are performed to balance the system's dynamic response and steady-state control accuracy, accurately mapping and eliminating mechanical response lag. The converted instructions are superimposed on the calculation node to allocate the oil supply ratio, outputting to the plugging hydraulic cylinder and the grouting pump frequency converter to generate linkage drive signals, achieving high-precision mechanical collaborative drive to maintain effective stress within the hole and strictly control lateral displacement.

Description

Technical Field

[0001] This invention relates to the field of grouting linkage closed-loop control technology, and in particular to a grouting and sealing linkage closed-loop control method for grouting conditions of mining anchor cables. Background Technology

[0002] The grouting linkage closed-loop control technology field specifically refers to a closed-loop feedback control mechanism to maintain the stability of the surrounding rock in underground engineering. This field sets execution standards requiring the grouting pressure fluctuation amplitude to converge to a preset control threshold and requires the coordinated adjustment of multiple parameters, including setting a reference value for the output power of the grouting pump, adjusting the radial expansion volume of the plug, and calculating the negative feedback error signal of the pressure inside the hole. The technical framework of this field aims to overcome the defects caused by the nonlinear time-varying permeability of the cracked rock mass medium, thereby ensuring that the filling degree of cement-based grout and chemical grout reaches the engineering set volume ratio standard.

[0003] The grout and plugging linkage closed-loop control method for grouting of mine anchor cables is specifically an adaptive adjustment scheme for the unsteady flow parameters of the gap between the support borehole walls in coal mine roadways. The scheme aims to establish a partial differential mathematical model of the evolution of the grout viscosity coefficient with the rheological state, construct the fluid dynamic constitutive equation, and output target pump pressure control commands and associated plugging displacement commands. The scheme aims to overcome the anchoring force attenuation caused by high-pressure stress deformation in soft rock roadways, control the pressure of the grouting pipeline in the borehole to maintain it within the set effective stress range, and control the lateral displacement of the borehole plugging device within the system's preset tolerance range.

[0004] Traditional methods rely on establishing a partial differential mathematical model of the grout viscosity coefficient evolving with the rheological state and constructing fluid dynamic constitutive equations. These methods then execute adjustments based on a set baseline value for the grouting pump output power and the negative feedback error signal of the borehole pressure. However, in actual engineering operations, rigid constitutive equations struggle to accurately reflect the drastically nonlinear time-varying permeability of fractured rock masses. High-pressure stress deformation in soft rock tunnels causes frequent alternations between laminar and turbulent flow patterns in the borehole wall gaps. Static partial differential models cannot characterize probabilistic transition boundaries in real time, resulting in severe hysteresis in system feedback. Furthermore, the reliance on a single negative feedback error signal to coordinate the adjustment of the grouting pump power and the plugging device is problematic. The lack of a decoupling mechanism for sudden resistance and a mechanical response delay compensation mechanism in radial expansion volume makes it prone to high-frequency control signal oscillations when facing sudden stress release conditions. This can lead to serious consequences, such as the inability of the sealing cylinder to synchronize with the instantaneous pulse pressurization of the grouting pump. During grouting operations in deep, strongly mined, and fractured roadways, the delayed feedback control response causes the lateral displacement of the borehole sealing device to exceed the preset tolerance range, resulting in a large amount of high-pressure grout overflowing and being lost along the gaps in the borehole wall. This results in the chemical grout filling degree being far below the engineering set volume ratio standard, drastically reducing the long-term anchoring torque of the coal mine anchor cables and endangering the overall roadway surrounding rock support safety. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a closed-loop control method for grouting and sealing in mining anchor cable grouting conditions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a closed-loop control method for grouting and sealing in mining anchor cable grouting operations, comprising the following steps:

[0007] Step 1: Based on the real-time absolute pressure value of the grouting pipeline, the time series is divided and the mean of multiple time windows is calculated as the difference term. The fluctuation variance matrix is ​​constructed, the transition probability and observation probability parameters are extracted and the double product value is calculated. The laminar and turbulent discrete variables in the hole wall gap are output to generate flow characterization parameters.

[0008] Step 2: Based on the flow characterization parameters, compare the state difference with the transition threshold, call the proportional-integral-derivative controller to superimpose the control error and the differential term, calculate the product of the gain coefficient and the disturbance variable and offset the unmodeled drag, suppress the high-frequency action chattering value, and obtain the control voltage matrix;

[0009] Step 3: Based on the control voltage matrix, set the nodes of the grouting pump and the sealing cylinder diagram and configure the coupling connection edge. Call the negative half-axis slope mapping term to calculate the inner product of the node state vectors, output the weight matrix and merge the weighted feature variables to construct the feature aggregation matrix;

[0010] Step 4: Based on the feature aggregation matrix, call the Long Short-Term Memory network to merge the flow resistance trend and time delay features, set the frequency and opening variables, calculate the intercept update reference point, quantize the distance to extract the solution set and perform the summation operation of individual fitness evaluation values, and output the overall linkage parameters;

[0011] Step 5: Based on the control voltage matrix and the overall linkage parameters, compare with the preset threshold and filter individual execution instructions, convert the instructions into duty cycle variables and superimpose them on the calculation node, allocate the oil supply ratio and output to the frequency converter of the sealing hydraulic cylinder and the grouting pump to generate linkage drive signals.

[0012] As a further aspect of the present invention, the flow characterization parameters include discrete variables of laminar flow and turbulent flow in the pore wall gap; the control voltage matrix includes proportional variational terms, integral variational terms, differential variational terms, and disturbance compensation control variables; the feature aggregation matrix includes grouting pump control parameters, plugging cylinder control parameters, and weighted feature variables; the overall linkage parameters include grouting frequency variables, reversing valve opening variables, and corresponding numerical multiplication and summation terms; and the linkage drive signal includes node duty cycle variables, oil supply ratio parameters, and control level.

[0013] As a further aspect of the present invention, the specific steps for generating the flow regime characterization parameters are as follows:

[0014] Based on the real-time absolute pressure values ​​of the grouting pipeline, the continuous time series is divided into equal intervals, the extreme value parameter of the sliding window is extracted, the mean is calculated and the pressure value is subtracted to extract the deviation term, the squared deviation variable is aggregated to fill the time feature matrix unit, and the pressure fluctuation variance matrix is ​​generated.

[0015] Based on the pressure fluctuation variance matrix, the dimensions of the transition probability and the observation probability distribution are aligned, the corresponding numerical product terms of the two parameter matrices are calculated, the extreme value paths of the probability product are screened, and the discrete variables of laminar and turbulent flow in the orifice wall gap are output to generate flow regime characterization parameters.

[0016] As a further aspect of the present invention, the specific steps for obtaining the control voltage matrix are as follows:

[0017] Based on the flow regime characterization parameters, feature parameters of adjacent discrete moments are extracted and the time sequence difference between the two moments is calculated. The difference is compared with the transition threshold and the variation term of the overshoot interval is filtered out. The time dimension control error parameter is extracted. The proportional-integral-derivative controller is called and the error and differential variation term are superimposed to generate the transition superposition control parameter.

[0018] Based on the aforementioned transition superposition control parameters, the grouting continuous state characteristic parameters are matched, the gain coefficient and state disturbance variable are extracted and the corresponding product value of the two terms is calculated, the unmodeled abrupt resistance characteristic term of the borehole wall is quantified, the product value is directly subtracted, and the disturbance compensation control variable is generated.

[0019] Based on the disturbance compensation control variables, smooth continuous variables are extracted and replaced with hard switching mapping terms. The frequency value of proportional valve action chattering is quantified. Multiple compensation parameters are merged and high-frequency action chattering values ​​are suppressed. Multi-dimensional values ​​are summarized to construct matrix units and obtain the control voltage matrix.

[0020] As a further aspect of the present invention, the proportional-integral-derivative (PID) controller first calls the PID controller, extracts the built-in proportional gain coefficient, integral gain coefficient, and derivative gain coefficient, performs a multiplication operation between the time-dimensional control error parameter and the proportional gain coefficient to output a proportional variation term, extracts the discrete values ​​of the historical time window of the time-dimensional control error parameter, performs an accumulation integration operation, and multiplies it by the integral gain coefficient to output an integral variation term, obtains the current value of the time-dimensional control error parameter and the adjacent delay value, performs a difference differentiation operation, and multiplies it by the derivative gain coefficient to output a derivative variation term, summarizes the time-dimensional control error parameter, proportional variation term, integral variation term, and derivative variation term, and performs a multi-parameter linear superposition summation operation.

[0021] As a further aspect of the present invention, the specific steps for constructing the feature aggregation matrix are as follows:

[0022] Based on the control voltage matrix, the control parameters of the grouting pump and the sealing cylinder are extracted and mapped to the corresponding topological discrete nodes. Connecting edges are configured according to the rigid boundary. The negative half-axis slope mapping term is called to calculate the inner product of the node state vector. The values ​​of multiple inner products are summarized to generate the node state weight matrix.

[0023] Based on the node state weight matrix, the feature parameters of neighboring nodes are extracted and the weight distribution dimension is aligned. The product of the weight value and the feature parameter is calculated. The weighted feature variables are accumulated along the topological connected edges. Multiple aggregated variables are summarized to construct matrix units and generate a feature aggregation matrix.

[0024] As a further aspect of the present invention, the specific steps for outputting the overall linkage parameter are as follows:

[0025] Based on the feature aggregation matrix, the long short-term memory network is called to extract the flow resistance variation trend and mechanical delay and merge the multi-dimensional features. The grouting frequency and the opening value of the reversing valve are summarized to set continuous spatial decision variables, construct a parameter distribution coordinate system, and generate a delay decision feature matrix.

[0026] Based on the time delay decision feature matrix, the pressure error integral and stress penalty evaluation term are summarized, the multi-objective dimension edge intercept is quantified and the hyperplane reference point coordinates are updated, the positional relationship is mapped to define the solution space constraint boundary, and a reference point distribution coordinate set is generated.

[0027] Based on the coordinate set of the reference point distribution, calculate the orthogonal vertical distance between the individual and the reference point, compare the distance values ​​with the boundary threshold to filter the frontier solution set, extract the weight coefficient matching optimization evaluation value, calculate the corresponding numerical multiplication and summation term, and obtain the overall linkage parameter.

[0028] As a further aspect of the present invention, the Long Short-Term Memory (LSTM) network extracts feature aggregation matrices corresponding to multiple time-series node variables, inputs these multiple time-series node variables into the LTM network's forget gate and input gate, performs weight matrix multiplication and activation function mapping operations, updates the cell state variables at the current time step, and extracts the continuous-time variation derivative as a flow resistance variation trend parameter. The multiple time-series node variables and the cell state variables at the current time step are input into the LTM network's output gate, calculates the hidden layer state variables at the current time step, and extracts the state variable time difference as a mechanical delay parameter. The flow resistance variation trend parameter and the mechanical delay parameter are then subjected to matrix horizontal concatenation operations, merging multiple dimensional values ​​to output multi-dimensional merged features.

[0029] As a further aspect of the present invention, the step of summarizing the pressure error integral and stress penalty evaluation term involves reading the built-in node pressure prediction variables and radial expansion stress parameters in the time delay decision feature matrix, introducing a preset target pressure reference trajectory and calculating the numerical difference between the corresponding node pressure prediction variables and the target pressure reference trajectory, extracting the absolute value of the corresponding numerical difference and performing cumulative integration along a continuous time window to output the pressure error integral evaluation term, retrieving a preset safe stress boundary threshold and comparing the radial expansion stress parameter with the safe stress boundary threshold, quantifying the overshoot difference exceeding the safe stress boundary threshold and multiplying it by a preset system penalty weight coefficient to output the stress penalty evaluation term, and integrating the pressure error integral evaluation term and the stress penalty evaluation term to construct a multi-objective evaluation vector.

[0030] As a further aspect of the present invention, the specific steps for generating the linkage drive signal are as follows:

[0031] Based on the control voltage matrix and the overall linkage parameters, a preset algebraic threshold is extracted and the values ​​of multiple parameters are compared. After filtering out the single execution instructions that exceed the threshold, the instructions are converted into duty cycle variables. Matrix elements are matched and corresponding control nodes are mapped to generate node duty cycle variables.

[0032] Based on the node duty cycle variable, the duty cycle variable is superimposed onto the closed-loop calculation node, the hydraulic station oil supply ratio parameter is extracted and associated allocation is performed, and after the parameter is converted into a control level, the level is directly sent to the frequency converter of the sealing hydraulic cylinder and the grouting pump to generate a linkage drive signal.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0034] In this invention, by calling the proportional-integral-derivative controller to superimpose the control error and the differential term to calculate the product of the gain coefficient and the disturbance variable and to offset the unmodeled resistance, the high-frequency action chattering value is suppressed to obtain the control voltage matrix, directly eliminating the sudden load interference caused by the high-pressure stress deformation of soft rock, and ensuring the smooth and stable closed-loop feedback voltage output.

[0035] In this invention, the flow resistance trend and time delay characteristics are merged by a long short-term memory network to explore the deep correlation of time series in multiple dimensions. The intercept update reference point is calculated by setting frequency and opening variables. Multi-objective optimization and quantitative evaluation are performed to balance the dynamic response and steady-state control accuracy of the system, and to accurately map and eliminate mechanical response lag.

[0036] In this invention, the conversion instruction is superimposed on the duty cycle variable to the calculation node to allocate the oil supply ratio and output it to the frequency converter of the sealing hydraulic cylinder and the grouting pump to generate a linkage drive signal, thereby achieving high-precision mechanical cooperative drive to maintain the effective stress in the hole and strictly control the lateral displacement. Attached Figure Description

[0037] Figure 1This is a schematic diagram of the main steps of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0039] Example 1

[0040] Please see Figure 1 This invention provides a technical solution: a closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations, comprising the following steps:

[0041] Step 1: Based on the real-time absolute pressure value of the grouting pipeline, the time series is divided and the mean of multiple time windows is calculated as the difference term. The fluctuation variance matrix is ​​constructed, the transition probability and observation probability parameters are extracted and the double product value is calculated. The laminar and turbulent discrete variables in the hole wall gap are output to generate flow characterization parameters.

[0042] Step 2: Based on the flow regime characterization parameters, compare the state difference with the transition threshold, call the proportional-integral-derivative controller to superimpose the control error and differential terms, calculate the product of the gain coefficient and the disturbance variable and offset the unmodeled drag, suppress the high-frequency action chattering value, and obtain the control voltage matrix;

[0043] Step 3: Based on the control voltage matrix, set the nodes of the grouting pump and the plugging cylinder diagram and configure the coupling and connection edges. Call the negative half-axis slope mapping term to calculate the inner product of the node state vectors, output the weight matrix and merge the weighted feature variables to construct the feature aggregation matrix;

[0044] Step 4: Based on the feature aggregation matrix, call the long short-term memory network to merge the flow resistance trend and time delay features, set the frequency and opening variables, calculate the intercept to update the reference point, quantize the distance to extract the solution set, and perform the summation operation of individual fitness evaluation values ​​to output the overall linkage parameters.

[0045] Step 5: Based on the control voltage matrix and overall linkage parameters, compare with the preset threshold and filter individual execution instructions, convert the instructions into duty cycle variables and superimpose them on the calculation node, allocate the oil supply ratio and output to the frequency converter of the sealing hydraulic cylinder and the grouting pump to generate linkage drive signals.

[0046] The flow regime characterization parameters include discrete variables of laminar flow and turbulent flow in the pore wall gap. The control voltage matrix includes proportional variational terms, integral variational terms, differential variational terms, and disturbance compensation control variables. The feature aggregation matrix includes grouting pump control parameters, plugging cylinder control parameters, and weighted feature variables. The overall linkage parameters include grouting frequency variables, reversing valve opening variables, and corresponding numerical multiplication and summation terms. The linkage drive signals include node duty cycle variables, oil supply ratio parameters, and control levels.

[0047] The specific steps for generating flow regime characterization parameters are as follows:

[0048] Based on the absolute pressure values ​​of the grouting pipeline, a sliding window statistical algorithm is used to perform equidistant time series segmentation. The initial sampling frequency is set to 50 Hz, and absolute pressure values ​​are continuously collected. A segmentation command is introduced, setting the sliding window size to 100 consecutive sampling points and the time step span to 10 sampling points. A truncation operation is performed along the continuous time series, reading 100 absolute pressure values ​​stored within a single sliding window. An ascending sort comparison operation is performed to select the first minimum absolute pressure value and the last maximum absolute pressure value to define and extract the sliding window value. The window extreme value parameter calculates the sum of 100 absolute pressure values ​​and divides it by a constant of 100 to calculate the mean. It then extracts each absolute pressure value within the window and subtracts the mean to extract the deviation term. For each deviation term, it performs a square operation to obtain the squared deviation variable. It aggregates the 100 squared deviation variables to construct a one-dimensional array sequence. It sets the target two-dimensional spatial structure parameters to include 10 horizontal row coordinates and 10 vertical column coordinates. It then maps and fills the one-dimensional array sequence into the cells of the two-dimensional time feature matrix in row priority order to generate the pressure fluctuation variance matrix.

[0049] Based on the pressure fluctuation variance matrix, the Viterbi algorithm is used to perform extreme path deduction for unsteady flow. Laminar and turbulent flow identifiers are embedded in the model's implicit state set. A frequency statistics algorithm is executed by reading historical pressure measurement logs, quantifying the probability of laminar flow maintaining its original state as 0.8 and the probability of a sudden transition from laminar to turbulent flow as 0.2. A two-dimensional transition probability set is established based on numerical recombination. The scalar elements within the pressure fluctuation variance matrix are extracted as the external observation input sequence for the algorithm. The initial mean parameter of the Gaussian probability distribution model is set to 5 MPa, and the scatter standard deviation parameter is set to 0.5 MPa. The observation input sequence is substituted into the Gaussian probability distribution model for algebraic operations to calculate the laminar flow classification of each observation. The probability values ​​of the corresponding occurrence of the identifier and the turbulence identifier are used to construct the observation probability matrix. Based on the row and column indexing mechanism, the two-dimensional transition probability value set is aligned with the distribution dimension of the observation probability matrix. Multiple loops are initiated to traverse the hierarchy, retrieve the cumulative maximum path probability variable of the previous time step, extract the variable, multiply it by the current state transition probability, and then multiply it by the current observation probability. The corresponding numerical continuous product terms of the two parameter matrices are calculated. The scalar values ​​of the two sets of probability product terms at the same time step are compared and the maximum value term is retained. The maximum term is extracted and associated with the state node index. All state node indices are sequentially connected to filter out the path with the extreme value of the probability multiplication. The extreme value path is finally mapped to the state identifier output orifice wall gap laminar and turbulent discrete variables to generate flow characterization parameters.

[0050] The specific steps for obtaining the control voltage matrix are as follows:

[0051] Based on flow regime parameters, a discrete-time difference algorithm is used to detect temporal abrupt changes. The discrete time step is set to 20 milliseconds. The system reads the discrete variable values ​​of laminar and turbulent flow in the orifice wall gap at the current sampling moment, and also reads the discrete variable values ​​of laminar and turbulent flow in the orifice wall gap at historical lag times. Arithmetic subtraction is performed to calculate the temporal difference before and after the difference. A preset transition threshold step is executed, loaded from the initial configuration file on the console, with the threshold set to 0.65. The system compares the absolute value of the difference with the threshold value, filtering for variations exceeding 0.65, and extracts the time dimension to control error parameters. A positional proportional-integral-derivative (PID) control algorithm is adopted, with the proportional coefficient set to 2.5, the integral coefficient set to 1.2, and the derivative coefficient set to 0.8. The time-dimensional control error parameter is extracted and multiplied by the proportional coefficient 2.5 to obtain the proportional variation term. The sum of the error values ​​of 50 consecutive control cycles is multiplied by the integral coefficient 1.2 to obtain the integral variation term. The deviation between the current error value and the error value of the previous control cycle is calculated and multiplied by the derivative coefficient 0.8 to obtain the derivative variation term. The positional PID controller is called to calculate the main program and superimpose the aforementioned time-dimensional control error parameter and the aforementioned three differential and integral variation terms to generate the transition superposition control parameters.

[0052] Based on the aforementioned transition superposition control parameters, a linear extended state observer algorithm is used to perform nonlinear resistance estimation and compensation operations. The corresponding continuous grouting state feature parameters are extracted by matching the continuous physical address record blocks in the system memory. The baseline value of the observer system bandwidth is set to 40 radians per second. The system's intrinsic gain formula calculation program is executed, setting the gain coefficient of the first observer to 120, the gain coefficient of the second observer to 4800, and the gain coefficient of the last observer to 64000. The second observer gain coefficient 4800 is extracted, and the corresponding state disturbance variable is output from the system's internal state space equation. A double-term numerical multiplication operation is performed to calculate the corresponding product value. The baseline value of the borehole wall abrupt change resistance is set to 15 MPa. The feature input node values ​​are summarized and quantized to obtain the borehole wall unmodeled abrupt change resistance feature term. The aforementioned corresponding product value is directly subtracted from the absolute value of the borehole wall unmodeled abrupt change resistance feature term to generate the disturbance compensation control variable.

[0053] Based on the aforementioned disturbance compensation control variables, a hyperbolic tangent smoothing filter algorithm is used to perform continuous processing of discontinuous switching signals. The smoothing boundary layer thickness constant is set to 0.05, and the nonlinear input variable reference value is set to the instantaneous voltage amplitude of the current sampling node. The nonlinear input variable is divided by the smoothing boundary layer thickness constant 0.05 to obtain intermediate variables. The system's arithmetic logic unit is read to calculate the corresponding positive and negative natural exponent terms of the intermediate variables, and combined arithmetic operations are performed to extract the smoothed continuous variables. The original step transition logic within the system is removed and the hard switching mapping term is replaced. The number of high and low level toggles of the associated control pin is captured by the lower-level hardware timer. The sampling and statistics period is set to 1 second. The number of level reversals within a single second is calculated to be 85. The jitter frequency value of the proportional valve action is quantized. The system initialization configuration script loads and stores the extreme value operation and executes the preset frequency tolerance step. The preset frequency tolerance is set to 50 times. It is determined that 85 times exceeds the upper limit of the preset frequency tolerance of 50 times. The first-order hysteresis digital filter instruction is retrieved to merge multiple compensation parameters and suppress high-frequency jitter values. The target storage structure is set to a two-dimensional matrix containing two horizontal row coordinates and two vertical column coordinates. All the aforementioned corresponding extracted values ​​are summarized in the order of row first and column last and filled into the two-dimensional matrix to construct the matrix unit, thereby obtaining the control voltage matrix.

[0054] The specific steps for constructing the feature aggregation matrix are as follows:

[0055] Based on the control voltage matrix, a graph convolutional neural network algorithm is used to perform topology node mapping and weight calculation operations. This involves extracting the control parameters corresponding to the grouting pump's voltage amplitude of 24 volts and the sealing cylinder's voltage amplitude of 12 volts, and mapping them to corresponding discrete topology nodes. The grouting pump node number is set to a first digit of 0, and the sealing cylinder node number is set to a second digit of 1. Physical mechanical space constraint parameters are read from the system's initial configuration file and written into a constant matrix to execute the preset rigid boundary steps. Based on the constant matrix corresponding to the preset rigid boundary, interconnecting edges between nodes are configured, and the adjacency matrix elements are assigned a value of 1. The physical memory block is read to extract the corresponding 64-dimensional floating-point state vectors of the nodes, and LeakyR is called. The eLU activation function algorithm sets the negative half-axis slope parameter to 0.01. For the state vectors of node 0 and node 1, the corresponding 64-dimensional internal elements are extracted one by one, and multiplication is performed. Then, the accumulation operation is performed to calculate the inner product of the node state vectors. A nonlinear slope mapping operation is performed. The inner product value of the aforementioned state vectors is compared with the scalar 0. When the inner product value is less than the scalar 0, the inner product value is extracted and multiplied by the negative half-axis slope parameter 0.01 to obtain the negative half-axis slope mapping term. When the inner product value is greater than or equal to the scalar 0, the original inner product value is retained as the positive half-axis mapping term. The inner product value generated by the mapping calculation of multiple terms is filled into the two-dimensional grid structure containing two horizontal coordinate rows and two vertical coordinate columns to generate the node state weight matrix.

[0056] Based on the node state weight matrix, a graph neural network message passing algorithm is used to perform a neighboring node feature space aggregation operation. This operation involves traversing the adjacency matrix of the topological graph to read the correspondence between the source and target nodes, extracting the 32-dimensional floating-point feature parameters corresponding to all neighboring nodes of the target node, and calling a tensor dimension expansion instruction to align the weight distribution dimension. This expands the individual scalar weight values ​​within the node state weight matrix along the column direction to construct a one-dimensional tensor containing 32 identical values. The weight values ​​within the expanded one-dimensional tensor are then extracted and input into arithmetic logic along with the aforementioned 32-dimensional floating-point feature parameters. The unit executes an element-wise multiplication instruction to calculate the product of the weight value and the feature parameter, outputs a 32-dimensional single-item weighted feature variable, and performs a network connectivity dimension accumulation operation. It reads the direction of the graph topology connectivity edge association, performs vector addition along the topology connectivity edge to accumulate the input weighted feature variable of all adjacent nodes of the target node, calculates and obtains the single-item target node level aggregation value, establishes an empty two-dimensional data table with 2 rows and 32 columns, summarizes the multiple target node level aggregation variables, and fills the internal matrix unit of the two-dimensional data table in ascending order of row coordinate index to generate the feature aggregation matrix.

[0057] The specific steps for outputting the overall linkage parameters are as follows:

[0058] Based on the feature aggregation matrix, a Long Short-Term Memory (LSTM) network algorithm is used to analyze flow resistance variation trends and mechanical delays. This analysis involves converting the feature aggregation matrix into a 3D tensor input format by calling internal tensor deformation instructions. The time step is set to 20, the input feature dimension to 32, and the number of hidden layer neurons to 128. The 3D tensor is then input into the LSM network forget gate. The corresponding forget gate weight matrix and bias vector, pre-trained by the internal control unit through backpropagation and written into memory, are read. A dot product operation is performed between the input vector and the forget gate weight matrix, and the bias vector is superimposed. The Sigmoid activation function is then called to process the dot product result and calculate the forget gate control variable. This forget gate control variable is extracted to update the cell state at the previous time step. The input tensor is then substituted into the input gate to perform the corresponding dot product operation and activation function mapping operation to generate candidate cell states. The updated cell states and candidate states are then read. The cell state performs bitwise addition to generate the current cell state output flow resistance change trend parameter. The current cell state is substituted into the output gate, and the Tanh activation function is called to map it to the negative one and positive one continuous interval to calculate the hidden layer output vector and extract it as the mechanical delay parameter. Multi-dimensional feature merging and coordinate system construction operations are performed. The tensor splicing instruction is called to perform a horizontal splicing operation along the last dimension to merge multi-dimensional features with the flow resistance change trend parameter and the mechanical delay parameter. The host computer human-computer interaction interface inputs the initial limit value and performs the preset continuous decision boundary steps. The grouting frequency value is set to 50 Hz, and the reversing valve opening value is set to 60%. The 50 Hz and 60% values ​​are combined to set the continuous spatial decision variables. A Cartesian rectangular coordinate system is defined with the grouting frequency variable on the horizontal axis and the reversing valve opening variable on the vertical axis to construct the parameter distribution coordinate system. The merged multi-dimensional features are filled into the coordinate system network nodes according to the coordinate axis mapping logic to generate the delay decision feature matrix.

[0059] Based on the time-delay decision feature matrix, a simplex lattice point hyperplane distribution algorithm is used to perform multi-objective dimension intercept quantization. This multi-objective dimension intercept quantization operation involves reading fixed numerical parameters from the system environment setting file and executing a preset target reference trajectory step. The target pressure reference trajectory value is set to 20 MPa, and the safety stress boundary threshold is set to 35 MPa. The built-in nodal pressure prediction variables and radial expansion stress parameters in the time-delay decision feature matrix are read. The microprocessor's arithmetic logic unit is called to calculate the absolute difference between the nodal pressure prediction variables and the corresponding values ​​of 20 MPa. The absolute difference is extracted and integrated over 50 consecutive time sampling windows to output the pressure error integral evaluation term. The radial expansion stress parameter is compared with the absolute value of 35 MPa, and the overshoot difference exceeding 35 MPa is quantified and multiplied by the corresponding value read from the system constant memory. The constant penalty weight coefficient is 1.5. The stress penalty evaluation term is output. The pressure error integral evaluation term and the stress penalty evaluation term are summarized to construct a multi-objective evaluation vector. The hyperplane reference point coordinates are updated. The origin coordinates of the multi-dimensional target space are set as the system zero point. The minimum and maximum extreme points of the multi-objective evaluation vector at the coordinate axes of the multi-dimensional target space are found to quantize the multi-objective dimension edge intercepts. The multi-dimensional hyperplane is constructed from the extreme points based on the intercept values. The simplex lattice division layer parameter is set to 10. The hyperplane segmentation command is called to divide the multi-dimensional hyperplane along the edge intercepts to generate spatial discrete intersections and update the hyperplane reference point coordinates. The continuous space decision variables are matched to limit the extreme value range and map the positional relationship to delineate the solution space constraint boundary. The set of all distributed coordinate points inside the constraint boundary is extracted to construct a three-dimensional array and generate the reference point distribution coordinate set.

[0060] Based on the reference point distribution coordinate set, an orthogonal projection non-dominated sorting algorithm is used to perform a front set selection operation. This operation involves reading the built-in floating-point fitness coordinates of individuals in the non-dominated sorting iterative population, calculating the orthogonal perpendicular distance from each fitness coordinate to a reference point within the reference point distribution coordinate set using Euclidean distance calculation instructions, and executing a preset boundary threshold step by reading a fixed constant from the system's initial configuration file (set to 0.02 meters). The orthogonal perpendicular distance values ​​are compared to the absolute value of 0.02 meters, and fitness coordinates corresponding to distances less than 0.02 meters are added to the local candidate set. Finally, the non-dominated comparison operation logic is used to eliminate mutually exclusive individuals within the local candidate set. The system selects the frontier solution set of the dominant individuals and performs a summation operation on the optimization evaluation values. It extracts the weight coefficients to match the optimization evaluation values, sets the first weight coefficient to 0.6, sets the second weight coefficient to 0.4, sets the first optimization evaluation value to the correlation constant of the aforementioned pressure error integral evaluation term, sets the second optimization evaluation value to the correlation constant of the aforementioned stress penalty evaluation term, extracts 0.6 and 0.4 to match the first and second optimization evaluation values, calls the microprocessor floating-point multiplication instruction to calculate 0.6 multiplied by the first optimization evaluation value to obtain the first product term, calculates 0.4 multiplied by the second optimization evaluation value to obtain the second product term, calls the floating-point addition instruction to accumulate the first and second product terms to calculate the corresponding value multiplication summation term, and obtains the overall linkage parameter.

[0061] The specific steps for generating the linkage drive signal are as follows:

[0062] Based on the control voltage matrix and overall linkage parameters, a pulse width modulation instruction conversion algorithm is used to perform parameter comparison and screening. This process involves reading pre-written floating-point values ​​from the non-volatile static area of ​​the programmable logic controller (PLC) and executing a preset algebraic threshold step. The algebraic threshold is set to 24 volts. The output voltage values ​​of the control nodes within the control voltage matrix are extracted and compared with the built-in linkage control values ​​of the overall linkage parameters. The microprocessor's arithmetic logic unit then uses a value comparison instruction to compare multiple parameter values ​​with the absolute value of the 24-volt algebraic threshold. Instructions exceeding the 24-volt threshold are selected for individual execution, and a duty cycle instruction conversion operation is performed to extract the built-in voltage amplitude parameters of these individual execution instructions. Divide the number by the system reference maximum voltage constant of 36 volts, call the floating-point division unit to calculate and obtain a continuous floating-point value in the range of 0 to 1, call the floating-point to integer data type conversion function to multiply the continuous floating-point value by the constant 100 to extract the integer part value, convert the instruction into a duty cycle variable, and perform matrix node mapping operation. Read the two-dimensional row and column index coordinates inside the control voltage matrix, set the row coordinate index of the matrix element corresponding to the grouting pump to 0 and the column coordinate index to 0, set the row coordinate index of the matrix element corresponding to the sealing cylinder to 1 and the column coordinate index to 1, extract the duty cycle variable association identifier to match the matrix element, and call the system physical address pointer to map the corresponding control node to generate the node duty cycle variable;

[0063] Based on the node duty cycle variable, a multi-channel digital-to-analog conversion allocation algorithm is used to perform the computing node superposition operation. This operation involves locking the current physical memory address of the closed-loop computing node by reading the addressing instruction from the closed-loop feedback control loop main program buffer, and then calling the microcontroller's internal register increment operation instruction to superimpose the duty cycle variable into the internal temporary storage register of the closed-loop computing node. Simultaneously, the hydraulic station oil supply allocation operation is performed. The hydraulic station oil supply ratio parameter is extracted by reading the hydraulic station fluid control communication protocol stack bus data frame. The system initialization configuration script loads the preset ratio parameters and executes the oil supply ratio setting steps, setting the grouting pump oil supply ratio parameter to 70% and the sealing hydraulic cylinder oil supply ratio parameter to 30%. The microcontroller is then invoked. The floating-point multiplication unit extracts the duty cycle variable from the temporary register, multiplies it by 70% and 30% respectively, performs associated allocation, and performs control level conversion and transmission. It calls the microcontroller's built-in 12-bit precision digital-to-analog converter peripheral module to read the allocated duty cycle variable, sets the conversion reference voltage to 5 volts, and maps the duty cycle variable to a continuous analog voltage range of 0 volts to 5 volts according to a linear ratio to convert the parameter into a control level. It sets the serial peripheral interface communication baud rate to 115200 bits per second, calls the direct memory access controller chip to directly transmit the control level bus data packet to the analog input control terminal of the proportional servo valve of the sealing hydraulic cylinder and the frequency setpoint terminal of the grouting pump frequency converter, generating a linkage drive signal.

[0064] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations, characterized in that, Includes the following steps: Step 1: Based on the real-time absolute pressure value of the grouting pipeline, the time series is divided and the mean of multiple time windows is calculated as the difference term. The fluctuation variance matrix is ​​constructed, the transition probability and observation probability parameters are extracted and the double product value is calculated. The laminar and turbulent discrete variables in the hole wall gap are output to generate flow characterization parameters. Step 2: Based on the flow characterization parameters, compare the state difference with the transition threshold, call the proportional-integral-derivative controller to superimpose the control error and the differential term, calculate the product of the gain coefficient and the disturbance variable and offset the unmodeled drag, suppress the high-frequency action chattering value, and obtain the control voltage matrix; Step 3: Based on the control voltage matrix, set the nodes of the grouting pump and the sealing cylinder diagram and configure the coupling connection edge. Call the negative half-axis slope mapping term to calculate the inner product of the node state vectors, output the weight matrix and merge the weighted feature variables to construct the feature aggregation matrix; Step 4: Based on the feature aggregation matrix, call the Long Short-Term Memory network to merge the flow resistance trend and time delay features, set the frequency and opening variables, calculate the intercept update reference point, quantize the distance to extract the solution set and perform the summation operation of individual fitness evaluation values, and output the overall linkage parameters; Step 5: Based on the control voltage matrix and the overall linkage parameters, compare with the preset threshold and filter individual execution instructions, convert the instructions into duty cycle variables and superimpose them on the calculation node, allocate the oil supply ratio and output to the frequency converter of the sealing hydraulic cylinder and the grouting pump to generate linkage drive signals.

2. The closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations according to claim 1, characterized in that, The flow characterization parameters include discrete variables of laminar flow and turbulent flow in the orifice wall gap; the control voltage matrix includes proportional variational terms, integral variational terms, differential variational terms, and disturbance compensation control variables; the feature aggregation matrix includes grouting pump control parameters, plugging cylinder control parameters, and weighted feature variables; the overall linkage parameters include grouting frequency variables, reversing valve opening variables, and corresponding numerical multiplication and summation terms; and the linkage drive signal includes node duty cycle variables, oil supply ratio parameters, and control level.

3. The closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations according to claim 1, characterized in that, The specific steps for generating the flow regime characterization parameters are as follows: Based on the real-time absolute pressure values ​​of the grouting pipeline, the continuous time series is divided into equal intervals, the extreme value parameters of the sliding window are extracted, the mean is calculated and the pressure values ​​are subtracted to extract the deviation term, the squared deviation variables are aggregated to fill the time feature matrix units, and the pressure fluctuation variance matrix is ​​generated. Based on the pressure fluctuation variance matrix, the dimensions of the transition probability and the observation probability distribution are aligned, the corresponding numerical product terms of the two parameter matrices are calculated, the extreme value paths of the probability product are screened, and the discrete variables of laminar and turbulent flow in the orifice wall gap are output to generate flow characterization parameters.

4. The closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations according to claim 1, characterized in that, The specific steps for obtaining the control voltage matrix are as follows: Based on the flow regime characterization parameters, feature parameters of adjacent discrete moments are extracted and the time-series differences are calculated. The differences are compared with the transition threshold and the variation terms of the overshoot interval are filtered. The time dimension control error parameters are extracted. The proportional-integral-derivative controller is called and the error and differential variation terms are superimposed to generate the transition superposition control parameters. Based on the aforementioned transition superposition control parameters, the grouting continuous state characteristic parameters are matched, the gain coefficient and state disturbance variable are extracted and the corresponding product value of the two terms is calculated, the unmodeled abrupt resistance characteristic term of the borehole wall is quantified, the product value is directly subtracted, and the disturbance compensation control variable is generated. Based on the disturbance compensation control variables, smooth continuous variables are extracted and replaced with hard switching mapping terms. The frequency value of proportional valve action jitter is quantified. Multiple compensation parameters are merged and high-frequency action jitter values ​​are suppressed. Multi-dimensional values ​​are summarized to construct matrix units and obtain the control voltage matrix.

5. The closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations according to claim 1, characterized in that, The proportional-integral-derivative (PID) controller first calls the PID controller, extracts the built-in proportional gain coefficient, integral gain coefficient, and derivative gain coefficient, performs a multiplication operation between the time-dimensional control error parameter and the proportional gain coefficient to output the proportional variation term, extracts the discrete values ​​of the historical time window of the time-dimensional control error parameter, performs an accumulation integration operation, and multiplies it by the integral gain coefficient to output the integral variation term, obtains the current value of the time-dimensional control error parameter and the adjacent delay value, performs a difference differentiation operation, and multiplies it by the derivative gain coefficient to output the derivative variation term, summarizes the time-dimensional control error parameter, proportional variation term, integral variation term, and derivative variation term, and performs a multi-parameter linear superposition summation operation.

6. The closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations according to claim 1, characterized in that, The specific steps for constructing the feature aggregation matrix are as follows: Based on the control voltage matrix, the control parameters of the grouting pump and the sealing cylinder are extracted and mapped to the corresponding topological discrete nodes. Connecting edges are configured according to the rigid boundary. The negative half-axis slope mapping term is called to calculate the inner product of the node state vector. The values ​​of multiple inner products are summarized to generate the node state weight matrix. Based on the node state weight matrix, the feature parameters of neighboring nodes are extracted and the weight distribution dimension is aligned. The product of the weight value and the feature parameter is calculated. The weighted feature variables are accumulated along the topological connected edges. Multiple aggregated variables are summarized to construct matrix units and generate a feature aggregation matrix.

7. The closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations according to claim 1, characterized in that, The specific steps for outputting the overall linkage parameters are as follows: Based on the feature aggregation matrix, the long short-term memory network is called to extract the flow resistance variation trend and mechanical delay and merge the multi-dimensional features. The grouting frequency and the opening value of the reversing valve are summarized to set continuous spatial decision variables, construct a parameter distribution coordinate system, and generate a delay decision feature matrix. Based on the time delay decision feature matrix, the pressure error integral and stress penalty evaluation term are summarized, the multi-objective dimension edge intercept is quantified and the hyperplane reference point coordinates are updated, the positional relationship is mapped to define the solution space constraint boundary, and a reference point distribution coordinate set is generated. Based on the coordinate set of the reference point distribution, calculate the orthogonal vertical distance between the individual and the reference point, compare the distance values ​​with the boundary threshold to filter the frontier solution set, extract the weight coefficient matching optimization evaluation value, calculate the corresponding numerical multiplication and summation term, and obtain the overall linkage parameter.

8. The closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations according to claim 1, characterized in that, The Long Short-Term Memory (LSTM) network extracts feature aggregation matrices corresponding to multiple time-series node variables. These multiple time-series node variables are input into the LTM network's forget gate and input gate, and weight matrix multiplication and activation function mapping operations are performed to update the cell state variables at the current time step. The derivative of continuous time variation is extracted as a flow resistance variation trend parameter. The multiple time-series node variables and the cell state variables at the current time step are input into the LTM network's output gate to calculate the hidden layer state variables at the current time step and extract the time difference of the state variables as a mechanical delay parameter. The flow resistance variation trend parameter and the mechanical delay parameter are extracted and subjected to matrix horizontal concatenation operations to merge multiple dimensional values ​​and output multi-dimensional merged features.

9. The closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations according to claim 7, characterized in that, The summary pressure error integral and stress penalty evaluation term reads the built-in node pressure prediction variables and radial expansion stress parameters in the time delay decision feature matrix, introduces the preset target pressure reference trajectory and calculates the corresponding numerical difference between the node pressure prediction variables and the target pressure reference trajectory, extracts the absolute value of the corresponding numerical difference, performs cumulative integration along a continuous time window and outputs the pressure error integral evaluation term, retrieves the preset safe stress boundary threshold and compares the radial expansion stress parameter with the safe stress boundary threshold, quantifies the overshoot difference exceeding the safe stress boundary threshold and multiplies it by the system's preset penalty weight coefficient to output the stress penalty evaluation term, integrates the pressure error integral evaluation term and the stress penalty evaluation term and combines them to construct a multi-objective evaluation vector.

10. The closed-loop control method for grout and sealing linkage in mining anchor cable grouting operations according to claim 1, characterized in that, The specific steps for generating the linkage drive signal are as follows: Based on the control voltage matrix and the overall linkage parameters, a preset algebraic threshold is extracted and the values ​​of multiple parameters are compared. After filtering out the single execution instructions that exceed the threshold, the instructions are converted into duty cycle variables. Matrix elements are matched and corresponding control nodes are mapped to generate node duty cycle variables. Based on the node duty cycle variable, the duty cycle variable is superimposed onto the closed-loop calculation node, the hydraulic station oil supply ratio parameter is extracted and associated allocation is performed, and after the parameter is converted into a control level, the level is directly sent to the frequency converter of the sealing hydraulic cylinder and the grouting pump to generate a linkage drive signal.

Images