Steel support axial force dynamic servo regulation and control method and system based on deep learning

Abstract

The invention relates to the technical field of steel support axial force dynamic servo regulation and control, and discloses a steel support axial force dynamic servo regulation and control method and system based on deep learning, and the method comprises the steps: collecting node data, filtering data noise, and obtaining purification deformation features. Analyzing a deformation propagation path according to the purified deformation characteristics, calculating a node influence coefficient, constructing an initial deformation matrix, outputting an optimized deformation matrix through a graph neural network in combination with soil characteristic influence parameters of each node, screening high-risk nodes, and fusing environment variables and mechanical difference data of the nodes to obtain an optimal deformation matrix; and generating an adjustment deformation matrix, predicting a node regulation score through linear regression, and generating a priority list. And finally, node stress is analyzed according to the list, an adaptive adjustment instruction set is generated to obtain a preliminary regulation and control draft, and a regulation and control scheme meeting requirements is finally formed through simulation verification and unbalance probability optimization. According to the method, the defect that influence factors in multiple aspects cannot be comprehensively processed and an accurate regulation and control strategy cannot be formulated in the prior art is overcome.

Description

Technical Field

[0001] This invention relates to the field of dynamic servo control technology for axial force of steel supports, and in particular to a method and system for dynamic servo control of axial force of steel supports based on deep learning. Background Technology

[0002] In modern construction engineering, the safety and stability of steel support systems are crucial for ensuring the safety of the construction process and the structure. Research in this field directly impacts the success or failure of engineering projects, especially in complex geological environments and under dynamic construction conditions, where steel support control technology becomes a core element in ensuring project quality.

[0003] How to achieve precise management of steel support systems in a changing environment is a key issue that the industry urgently needs to address.

[0004] Under current technology, the deformation of nodes in steel support systems is monitored and analyzed only statically or through simple dynamic means, without considering soil properties, environmental changes, and the mutual influence between support structures. Furthermore, the propagation of deformation characteristics among different parts of the steel support system is not uniformly distributed, and the mechanical effects on support nodes at different locations vary significantly. Static or simple dynamic monitoring cannot accurately capture the propagation relationship of deformation characteristics between different parts. Therefore, during dynamic modeling, it is impossible to accurately simulate and predict the propagation path and degree of impact of deformation in the steel support network, often resulting in delayed or inaccurate control measures.

[0005] Therefore, existing technologies have the drawback of being unable to comprehensively handle multiple influencing factors and formulate precise control strategies. Summary of the Invention

[0006] This invention provides a method and system for dynamic servo control of axial force in steel supports based on deep learning, which can solve the problem that existing technologies cannot comprehensively handle multiple influencing factors and formulate precise control strategies.

[0007] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a deep learning-based method for dynamic servo control of axial force in steel supports, comprising: Obtain the geometric position of each node, record the displacement change data and stress change data of the node, integrate them to obtain the initial deformation characteristics, filter the noise of the initial deformation characteristics to obtain the purified deformation characteristics; Based on the deformation characteristics of the purification process, the deformation propagation direction is analyzed to obtain the deformation propagation path. The influence coefficients between nodes on the deformation propagation path are calculated, and the influence coefficients are used as element values ​​of the matrix to construct an initial deformation matrix. The soil property influence parameters of each node are obtained and used as input parameters along with the initial deformation matrix. The graph neural network is used to output an optimized deformation matrix. High-risk nodes are selected based on the optimized deformation matrix to generate a list of high-risk nodes. The environmental variable data of each node in the high-risk node list is obtained and purified to obtain purified environmental variables. The optimized deformation matrix is ​​adjusted according to the purified environmental variables to obtain the adjusted deformation matrix. The mechanical impact data of the nodes is acquired and fused with the adjustment deformation matrix to form a mechanical impact matrix. At the same time, combined with the purification environmental variables, the node regulation score is predicted by linear regression, and a regulation priority list is generated based on the node regulation score. Based on the control priority list, the stress situation of each node is analyzed, node judgment results are generated, and an applicable set of adjustment instructions is generated based on the node judgment results to obtain a preliminary control draft. The preliminary control draft is simulated, and based on the simulation results, the preliminary control draft is optimized according to the imbalance probability to obtain the final control scheme.

[0008] In one optional implementation, the step of analyzing the deformation propagation direction based on the purification deformation characteristics to obtain the deformation propagation path, calculating the influence coefficients between nodes on the deformation propagation path, and using the influence coefficients as element values ​​of a matrix to construct an initial deformation matrix includes: Based on the geometric position of the nodes in the purification deformation features, the nodes are divided into meshes to generate a node connection network diagram; For the aforementioned node connection network diagram, stress transmission at the nodes is simulated using finite element analysis to determine the stress transmission path; Based on the stress transfer path, the deformation propagation direction is analyzed to obtain the deformation propagation path. Calculate the influence coefficients between nodes along the deformation propagation path, and use these influence coefficients as element values ​​of the matrix to construct an initial deformation matrix.

[0009] In one optional implementation, the step of obtaining the soil property influence parameters of each node and using the initial deformation matrix as input parameters, using a graph neural network to output an optimized deformation matrix, filtering high-risk nodes based on the optimized deformation matrix, and generating a high-risk node list includes: Obtain the soil property influence parameters for each node; The initial deformation matrix and the soil property influence parameters are used as input parameters, and a graph neural network is used for training and optimization to output an optimized deformation matrix. From the optimized deformation matrix, nodes that meet the preset high-risk conditions are used to generate a list of high-risk nodes.

[0010] In one optional implementation, the step of acquiring and purifying the environmental variable data of each node in the high-risk node list to obtain purified environmental variables, and adjusting the optimized deformation matrix based on the purified environmental variables to obtain an adjusted deformation matrix, includes: The environmental variable data of each node in the high-risk node list is obtained from the preset environmental database, and outliers are removed from the environmental variable data to obtain purified environmental variables. Analyze the correlation between the purification environmental variables and the optimized deformation matrix to determine the set of correction coefficients for the node propagation weights of the environmental variables; The optimized deformation matrix is ​​mapped and adjusted according to the set of correction coefficients to obtain the adjusted deformation matrix.

[0011] In one optional implementation, the mechanical impact data of the acquired nodes is fused with the adjusted deformation matrix to form a mechanical impact matrix. Simultaneously, combined with the purification environmental variables, node regulation scores are predicted via linear regression, and a regulation priority list is generated based on the node regulation scores, including: Mechanical influence data for each node is obtained from a pre-set mechanical database. The adjustment deformation matrix and the mechanical influence data are then fused to obtain a mechanical influence matrix. Based on the mechanical influence matrix and the purification environment variables, the node regulation score is predicted by linear regression. If the node's control score exceeds a preset score threshold, it is marked as a high-priority node. For the high-priority nodes, the control scores are sorted to obtain a control priority list.

[0012] In one optional implementation, the step of analyzing the force situation of each node according to the control priority list, generating node judgment results, and generating an applicable set of adjustment instructions based on the node judgment results to obtain a preliminary control draft includes: Based on the control priority list, analyze the force situation of each node to obtain the node judgment result; Based on a preset rule base, the node judgment results are matched to generate corresponding instructions, resulting in an instruction set; By associating the instruction set with real-time monitored node status data, the instruction set is adjusted based on the applicability of the instructions to obtain an adjusted instruction set, and a preliminary control draft is determined.

[0013] In one optional implementation, the step of simulating the preliminary control draft and optimizing the preliminary control draft based on the simulation results and the probability of imbalance to obtain the final control scheme includes: The preliminary draft regulation was simulated to obtain the first simulation results; Based on the first simulation results, the imbalance probability of each region is calculated, and regions with imbalance probabilities higher than a preset imbalance threshold are classified as high-risk regions. Based on the high-risk areas, the preliminary control plan was partially revised to obtain a revised control plan; The simulation is performed according to the modified control scheme to obtain a second simulation result. If the imbalance probability of the second simulation result is lower than the preset imbalance threshold, the final control scheme is output.

[0014] Secondly, the present invention provides a deep learning-based dynamic servo control system for axial force of steel supports, comprising: The deformation feature acquisition module is used to acquire the geometric position of each node, record the displacement change data and stress change data of the node, integrate them to obtain the initial deformation feature, filter noise from the initial deformation feature to obtain the purified deformation feature; The initial deformation matrix construction module is used to analyze the deformation propagation direction based on the purification deformation characteristics, obtain the deformation propagation path, calculate the influence coefficient between nodes on the deformation propagation path, and use the influence coefficient as the element value of the matrix to construct the initial deformation matrix. The risk identification module is used to obtain the soil property influence parameters of each node and use the initial deformation matrix as input parameters. It uses a graph neural network to output an optimized deformation matrix, and filters high-risk nodes based on the optimized deformation matrix to generate a list of high-risk nodes. The deformation matrix acquisition module is used to acquire environmental variable data of each node in the high-risk node list and perform purification processing to obtain purified environmental variables. The optimized deformation matrix is ​​then adjusted based on the purified environmental variables to obtain the adjusted deformation matrix. The priority determination module is used to acquire the mechanical impact data of the nodes and fuse it with the adjustment deformation matrix to form a mechanical impact matrix. At the same time, it combines the purification environmental variables to predict the node control score through linear regression and generates a control priority list based on the node control score. The preliminary control draft generation module is used to analyze the force situation of each node according to the control priority list, generate node judgment results, generate an applicable set of adjustment instructions based on the node judgment results, and obtain a preliminary control draft. The final control scheme generation module is used to simulate the preliminary control draft, and optimize the preliminary control draft based on the simulation results and the imbalance probability to obtain the final control scheme.

[0015] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the deep learning-based intelligent image processing method for digital photo frames described in any one of the above.

[0016] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the deep learning-based intelligent image processing method for digital photo frames described above.

[0017] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention acquires the geometric position, displacement, and stress change data of each node, integrates them to form initial deformation features, and filters out noise to obtain purified deformation features. Based on these features, it analyzes the deformation propagation path and calculates the influence coefficients between nodes to construct the initial deformation matrix. This effectively removes irrelevant information such as complex environmental interference and sensor errors from the deformation features, avoiding these interferences from affecting subsequent stages. At the same time, it analyzes the deformation propagation path based on the actual force transmission logic, and through numerical realization, it restores the influence between nodes, clarifies the order and direction of deformation transmission from the source node to the surrounding nodes, and accurately captures the linkage relationship between nodes.

[0018] (2) This invention combines soil characteristic parameters of each node, outputs an optimized deformation matrix through a graph neural network, and screens high-risk nodes. Subsequently, environmental variable data of high-risk nodes are purified, and mechanical influence data of the nodes are integrated to form a mechanical influence matrix. Combined with the purified environmental variables, linear regression is used to predict control scores and generate a node priority list. This achieves comprehensive processing of multiple influencing factors of nodes, efficient integration and utilization of data from various aspects, and, by leveraging the associative learning capability of graph neural networks and the quantitative evaluation advantages of linear regression, transforms multi-source data into measurable control scores, thereby generating a clear and explicit control priority list. This makes the focus and order of control work clear and explicit, enabling targeted operations to be carried out on high-priority nodes.

[0019] (3) This invention obtains a preliminary control draft by analyzing the force on nodes based on a list and generating a set of adaptive adjustment instructions. After simulation verification and imbalance probability optimization, a control scheme that meets the requirements is finally formed. This method can obtain a precise control scheme and continuously correct the shortcomings in the scheme, avoid potential risks in a timely manner. At the same time, through multiple rounds of simulation and verification, it ensures that the final control scheme not only meets engineering safety standards but also has strong feasibility, effectively guaranteeing the safety and accuracy of the entire control process, and providing efficient and reliable support for dynamic servo control of axial force of steel supports. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the process of the deep learning-based dynamic servo control method for axial force of steel supports provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a deep learning-based dynamic servo control system for axial force of steel supports provided in the second embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1 The first embodiment of the present invention provides a schematic diagram of a deep learning-based dynamic servo control method for axial force of steel supports, including steps S101 to S107, as follows: S101, Obtain the geometric position of each node, and record the displacement change data and stress change data of the nodes. Integrate them to obtain the initial deformation characteristics, filter the noise of the initial deformation characteristics, and obtain the purified deformation characteristics. S102, Based on the deformation characteristics of the purification process, perform deformation propagation direction analysis to obtain the deformation propagation path, calculate the influence coefficient between nodes on the deformation propagation path, and use the influence coefficient as the element value of the matrix to construct an initial deformation matrix; S103, obtain the soil property influence parameters of each node and use the initial deformation matrix as input parameters, use the graph neural network to output the optimized deformation matrix, filter high-risk nodes according to the optimized deformation matrix, and generate a list of high-risk nodes; S104, Obtain the environmental variable data of each node in the high-risk node list and perform purification processing to obtain purified environmental variables. Adjust the optimized deformation matrix according to the purified environmental variables to obtain the adjusted deformation matrix. S105, acquire the mechanical impact data of the nodes and fuse it with the adjustment deformation matrix to form a mechanical impact matrix. At the same time, combine the purification environmental variables and predict the node regulation score through linear regression, and generate a regulation priority list based on the node regulation score. S106, Based on the control priority list, analyze the force situation of each node, generate node judgment results, generate an applicable set of adjustment instructions based on the node judgment results, and obtain a preliminary control draft; S107, Simulate the preliminary control draft, and optimize the preliminary control draft based on the simulation results and the imbalance probability to obtain the final control scheme; In step S101, the geometric positions of each node are acquired, and the displacement and stress change data of the nodes are recorded. These data are then integrated to obtain initial deformation features. Noise is filtered from the initial deformation features to obtain purified deformation features. It is worth noting that the geometric positions of each node in the dynamic servo control of the steel support axial force are collected in real time through a pre-set sensor network, and the displacement and stress changes of the nodes are recorded. The initial deformation characteristics are then integrated. The pre-set sensor network includes positioning sensors, displacement sensors, and stress sensors, which are used to capture the displacement and stress change data of each node in the dynamic servo control of the steel support axial force in real time. These devices are deployed at key nodes, such as support joints or areas of concentrated stress. The collected data includes displacement and stress values, typically in millimeters and megapascals. The initial deformation characteristics are obtained by integrating the collected data.

[0023] It is worth noting that complex geological environments may lead to noise from vibrations or electromagnetic interference in the data. If the noise value in the initial deformation characteristics exceeds a preset noise threshold, it is judged as abnormal data and is removed. This achieves noise filtering of the initial deformation characteristics, resulting in purified deformation characteristics. The preset noise thresholds are specifically a displacement data fluctuation threshold of 1 mm and a stress data fluctuation threshold of 10 MPa. The 1 mm displacement data fluctuation threshold and the 10 MPa stress data fluctuation threshold are determined based on the engineering accuracy requirements of steel support system deformation monitoring and the interference characteristics of complex geological environments. According to historical monitoring data, when the displacement data fluctuation threshold is set to 1 mm and the stress data fluctuation threshold is set to 10 MPa, more than 95% of irrelevant noise such as complex geological vibrations and sensor errors can be filtered out, while retaining the valid deformation and stress data. For example, by collecting data from each node in real time through a preset sensor network, if the displacement of a node suddenly increases from 2.5 mm to 4 mm, or the stress data suddenly increases from 150 MPa to 170 MPa, exceeding the noise threshold, all data from that node are removed, ultimately obtaining purified deformation characteristics.

[0024] In step S102, based on the deformation characteristics of the purification process, deformation propagation direction analysis is performed to obtain the deformation propagation path. The influence coefficients between nodes along the deformation propagation path are calculated, and these influence coefficients are used as element values ​​of a matrix to construct an initial deformation matrix, including: Based on the geometric position of the nodes in the purification deformation features, the nodes are divided into meshes to generate a node connection network diagram; For the aforementioned node connection network diagram, stress transmission at the nodes is simulated using finite element analysis to determine the stress transmission path; Based on the stress transfer path, the deformation propagation direction is analyzed to obtain the deformation propagation path. Calculate the influence coefficients between nodes along the deformation propagation path, and use these influence coefficients as element values ​​of the matrix to construct an initial deformation matrix.

[0025] It is worth noting that, based on the geometric positions of the nodes in the deformation characteristics of the purification system, preset meshing tools such as Ansys, HyperWorks, and MeshWorks are used to mesh the nodes and generate a node connection network diagram. Meshing tools can partition the nodes of the steel support system according to their geometric positions, forming a virtual mesh structure diagram to describe the spatial relationships between the nodes. For example, in a steel support system of an underground engineering project, assuming there are 100 key nodes, the meshing tool divides these nodes into 10x10 mesh regions, with each mesh cell representing the node connection relationships within a certain range.

[0026] It is worth noting that by defining mechanical parameters such as mass, stiffness, and damping for all node-connecting components in the node connection network diagram, and calculating the displacement and stress of each node under load using a finite element analysis model, the direction of force transmission can be determined by combining the positive and negative signs of the stress (positive for tension and negative for compression). If the component is under compression and the node displacement points to the adjacent node, it indicates that the internal force is transmitted from that node to the adjacent node along the component axis. By performing force analysis on each node one by one, the magnitude and direction of the transmission between adjacent nodes are connected in series. Through the visualization of stress cloud diagrams and displacement vector diagrams, the distribution law of stress in the entire network is clearly presented, and the stress transmission path is determined. For example, by establishing a finite element model using ANSYS, the stress of node P1 is calculated to be -135MPa (negative sign indicates compression), and its horizontal displacement of 2.8mm points to the adjacent support node P2; the stress of support node P2 is -110MPa, and its displacement of 2.1mm points to the middle node P3; the stress of middle node P3 is -95MPa, and its displacement of 1.5mm points to the opposite node P4. After analyzing the stress on each node one by one, the stress transmission relationship between adjacent nodes is connected in series, and the stress transmission path is determined to be P1→P2→P3→P4.

[0027] It is worth noting that the displacement data of each node obtained from the finite element analysis were extracted. By comparing the magnitude and direction of the displacement of adjacent nodes on the stress transfer path, the main direction of deformation propagation from the node with the larger displacement value to the node with the smaller displacement value was determined as the deformation propagation direction. Then, nodes with continuous displacement changes and consistent directions were connected in series according to the transmission sequence to form a deformation propagation path corresponding to the stress transfer path. For example, on the stress transfer path P1→P2→P3→P4, the displacements of nodes P1, P2, P3, and P4 are 2.8 mm, 2.1 mm, 1.5 mm, and 0.9 mm, respectively. The displacement directions of all nodes are consistent, pointing towards the inside of the foundation pit. Comparing the displacements of adjacent nodes, it can be seen that the deformation gradually spreads from P1 with the larger displacement value to P2, P3, and P4 with the smaller displacement values, and the displacement changes of each node are continuous without abrupt changes. Based on this, it was determined that the deformation propagation direction is consistent with the stress transfer direction, ultimately forming the deformation propagation path: P1→P2→P3→P4.

[0028] It is worth noting that the influence coefficient is specifically the ratio of the displacement change of node j to the displacement change of node i, reflecting the degree of influence of the deformation of node i on node j. The specific implementation of calculating the influence coefficient between nodes along the deformation propagation path involves extracting the displacement data of N nodes along the deformation propagation path, calculating the influence coefficient of adjacent related nodes, and setting the coefficient of non-related nodes to 0. Finally, the influence coefficients between each node are filled into the corresponding positions in a matrix with node numbers as the matrix, constructing an N×N initial deformation matrix. For example, the displacement changes of each node are extracted: node P1 has a displacement of 4mm, P2 has a displacement of 3mm, and P3 has a displacement of 2mm. Nodes P1 and P2 are adjacent nodes, and nodes P2 and P3 are adjacent nodes. The calculated influence coefficient is K. 12 =3 / 4=0.75, K 23 =2 / 3≈0.67, the influence coefficients of non-adjacent nodes are all set to 0. The initial deformation matrix is... .

[0029] In step S103, soil property influence parameters for each node are obtained and used as input parameters along with the initial deformation matrix. A graph neural network is used to output an optimized deformation matrix. Based on the optimized deformation matrix, high-risk nodes are screened to generate a list of high-risk nodes, including: Obtain the soil property influence parameters for each node; The initial deformation matrix and the soil property influence parameters are used as input parameters, and a graph neural network is used for training and optimization to output an optimized deformation matrix. From the optimized deformation matrix, nodes that meet the preset high-risk conditions are used to generate a list of high-risk nodes.

[0030] It is worth noting that the soil property parameters affecting each node can be obtained from a preset soil parameter database, which is constructed based on field geological survey data and includes soil permeability and shear strength.

[0031] It is worth noting that the graph neural network model structure is based on the graph convolutional network (GCN) architecture. The input consists of two parts: an initial deformation matrix and soil characteristic parameters of each node. The input data is first standardized by Z-score to eliminate dimensional differences and ensure numerical stability. The core of the model consists of three graph convolutional layers. Each graph convolution operation uses the mean aggregation function to aggregate the feature information of the target node and its first-order neighbor nodes, thereby capturing the local topological relationship of deformation propagation. After each convolution, the ReLU activation function is applied to introduce a nonlinear transformation to enhance the model's expressive power. The number of hidden layer nodes is set to 64. The output layer is a fully connected layer that maps the learned node embedding vectors back to an N×N dimensional matrix and normalizes it to the [0,1] interval using the Sigmoid function to generate an optimized deformation matrix, whose element values ​​represent the optimized influence strength between nodes.

[0032] The model's training process relies on finite element simulation data and field monitoring datasets from historical engineering projects, covering deformation matrices and soil parameter labels under diverse working conditions to ensure data diversity and representativeness. During training, mean squared error (MSE) is used as the loss function to measure the deviation between the predicted matrix and the true labels (influence coefficient matrices corrected by expert experience). The optimizer is Adam, with an initial learning rate of 0.001 and a batch size of 32 to balance convergence speed and stability. To prevent overfitting, a 5-fold cross-validation method is used, with each fold training continuing until the validation set loss function converges.

[0033] It is worth noting that the preset high-risk condition is that the node influence intensity is less than 0.2. All elements in the i-th row of the optimized deformation matrix represent the node influence intensity of node i on all other nodes, reflecting the influence intensity of node i in the entire deformation transmission network. The node influence intensity value of node i is the average influence intensity of all nodes adjacent to node i in the i-th row. If the influence intensity of a node in the optimized deformation matrix is ​​lower than the preset high-risk condition, then the node is marked as a node with high risk, and a list of high-risk nodes is generated.

[0034] The high-risk condition is determined based on the node strength design standards and engineering risk control requirements. In historical monitoring data, 95% of the samples showed that the node would pose a bearing risk when the influence intensity was below 0.2. For example, if the corresponding row element of a node is [0, 0.15, 0.05], and the strength is (0 + 0.15 + 0.05) / 2 = 0.10, which meets the preset high-risk condition, then the node is marked as a high-risk node.

[0035] In step S104, environmental variable data for each node in the high-risk node list is obtained and purified to obtain purified environmental variables. The optimized deformation matrix is ​​then adjusted based on these purified environmental variables to obtain an adjusted deformation matrix, including: The environmental variable data of each node in the high-risk node list is obtained from the preset environmental database, and outliers are removed from the environmental variable data to obtain purified environmental variables. Analyze the correlation between the purification environmental variables and the optimized deformation matrix to determine the set of correction coefficients for the node propagation weights of the environmental variables; The optimized deformation matrix is ​​mapped and adjusted according to the set of correction coefficients to obtain the adjusted deformation matrix.

[0036] It's worth noting that the pre-set environmental database is constructed based on on-site geological survey data and long-term environmental monitoring data, containing historical time-series data on environmental variables such as groundwater level, water pressure changes, temperature fluctuations, and humidity for each node. Environmental variable data for each node in the high-risk node list is retrieved from the pre-set environmental database, specifically including water pressure and temperature values ​​at different times. Pre-set data cleaning tools, such as Python's pandas library, are used to remove outliers that significantly deviate from the normal range, resulting in purified environmental variables. For example, if the temperature at a certain moment is 100 degrees Celsius, or the water pressure reaches 2.0 MPa, which clearly does not conform to actual environmental conditions, then that outlier environmental data is removed.

[0037] It is worth noting that the influence intensity value of each high-risk node in the deformation matrix was extracted, and Pearson correlation analysis was used to quantify the correlation between the influence intensity value of each high-risk node and the environmental variable data in each node. The correlation coefficient was calculated, and a modified coefficient model was constructed. Where K is the correction coefficient, k is the correlation coefficient, and y is the environmental variable data of the node. Let σ be the mean of the environmental variable data for each node, and σ be the standard deviation of the environmental variable data for each node. The correlation coefficients between the node's influence intensity value and the water pressure and temperature values ​​are calculated separately to determine the set of correction coefficients for the water pressure and temperature values ​​on the node's propagation weight. For example, the Pearson correlation coefficient between the influence intensity value and water pressure value of high-risk node 1 is 0.8, and the Pearson correlation coefficient between the influence intensity value and temperature value of high-risk node 1 is 0.9. Therefore, the correction coefficient for the influence intensity value of water pressure on the node is calculated to be 2.1, and the correction coefficient for the influence intensity value of temperature on the node is 2.4.

[0038] It is worth noting that the high-risk nodes of the optimized deformation matrix are mapped and adjusted according to the set of correction coefficients. The specific adjustment formula is w'=w×K1×K2, where w' is the adjusted influence strength value of node i on node j, w is the original influence strength value of node i on node j, K1 is the correction coefficient for the influence strength value of water pressure on the node, and K2 is the correction coefficient for the influence strength value of temperature on the node. After adjustment, the mean influence strength of the nodes is recalculated. If the mean influence strength of the adjusted nodes is lower than 0.2 or higher than 1, the node is removed, and the influence strength value of the adjusted nodes is used as the element value of the matrix again to obtain the adjusted deformation matrix. For example, the influence strength value of high-risk node 1 on high-risk node 2 is 0.1, the correction coefficient for the influence strength value of water pressure on the node is 2.1, and the correction coefficient for the influence strength value of temperature on the node is 2.4. According to the formula, the calculated influence strength value of the adjusted node is 0.504, which no longer meets the preset high-risk condition. Therefore, the influence strength value of the adjusted node replaces the original influence strength value.

[0039] In step S105, the mechanical impact data of the nodes is acquired and fused with the adjustment deformation matrix to form a mechanical impact matrix. Simultaneously, combined with the purification environmental variables, the node regulation score is predicted using linear regression, and a regulation priority list is generated based on the node regulation score, including: Mechanical influence data for each node is obtained from a pre-set mechanical database. The adjustment deformation matrix and the mechanical influence data are then fused to obtain a mechanical influence matrix. Based on the mechanical influence matrix and the purification environment variables, the node regulation score is predicted by linear regression. If the node's control score exceeds a preset score threshold, it is marked as a high-priority node. For the high-priority nodes, the control scores are sorted to obtain a control priority list.

[0040] It is worth noting that the preset mechanical database is constructed based on field geological survey data and historical finite element simulation analysis results, containing data with differences in elastic modulus, Poisson's ratio, compressive strength, and shear modulus of materials at each node. Mechanical influence data for each node material is obtained from the preset mechanical database, specifically including differences in elastic modulus and Poisson's ratio. The influence intensity of each node in the adjusted deformation matrix is ​​calculated, and the elastic modulus and Poisson's ratio data of each node material are fused to obtain an M×3 mechanical influence matrix, where M is the total number of nodes, the first column is the influence intensity of each node, the second column is the elastic modulus of each node material, and the third column is the Poisson's ratio data. For example, the mechanical properties of each node material are extracted from the preset mechanical database: node P1: elastic modulus 206 GPa, Poisson's ratio 0.30; node P2: 205 GPa, 0.31. The influence intensity of each node in the adjusted deformation matrix is ​​obtained: node 1 ≈ 0.418, node 2 ≈ 0.432, resulting in a 2×3 mechanical influence matrix. .

[0041] It is worth noting that, based on the mechanical influence matrix and the purification environmental variables, it is also necessary to obtain the environmental variable data of nodes not in the high-risk node list from the preset environmental database. The node regulation score is predicted by linear regression. Specifically, the mechanical influence matrix and the purification environmental variables are concatenated into a 5-dimensional feature vector. After Z-score standardization (using the standard used when solving the multiple linear regression parameters with historical node data), a multiple linear regression model is constructed in the form y=β0+β1x1+β2x2+β3x3+β4x4+β5x5. Where x1 is the standardized influence intensity, x2 is the standardized material elastic modulus, x3 is the standardized Poisson's ratio, x4 is the standardized water pressure value, and x5 is the standardized temperature value. The linear parameters β0, β1, β2, β3, β4, and β5 are obtained by using historical node data, including the standardized mechanical influence matrix and environmental variables, as features, and the actual control effect score as the label. The optimal regression parameters are solved using the least squares method. Then, the standardized features of the node to be predicted are substituted into the model, and the predicted values ​​are mapped to the [0,1] interval using the Sigmoid function to obtain the node control score. Based on the mechanical influence matrix and the purification environmental variables, the node control score is predicted through linear regression. For example, with standardized features of [0.55, 0.68, 0.52, 0.61, 0.49], the predicted value is 0.76 obtained by multiple linear regression y = 0.12 + 0.35x1 = 2.08x2 + 0.12x3 + 0.22x4 + 0.10x5. The predicted value is mapped to 0.68 by the Sigmoid function, and 0.68 is the node regulation score.

[0042] It is worth noting that if a node's control score exceeds a preset threshold of 0.6, it is marked as a high-priority node. The threshold of 0.6 is based on safety control requirements. According to historical control data, when the control score threshold is set to 0.6, over 95% of nodes with abnormal mechanical properties or large environmental fluctuations require close monitoring and thus are considered high-priority nodes. For these high-priority nodes, they are sorted from largest to smallest control score to obtain a control priority list. For example, given five nodes with control scores of 0.82 for node P1, 0.75 for node P2, 0.58 for node P3, and 0.69 for node P4, the scores of P1 (0.82), P2 (0.75), and P4 (0.69) exceed the preset threshold of 0.6 and are thus marked as high-priority nodes. These nodes are then sorted from largest to smallest control score to obtain the final control priority list: Node P1, Node P2, Node P4.

[0043] In step S106, based on the control priority list, the force situation of each node is analyzed, a node judgment result is generated, and an applicable set of adjustment instructions is generated based on the node judgment result to obtain a preliminary control draft, including: Based on the control priority list, analyze the force situation of each node to obtain the node judgment result; Based on a preset rule base, the node judgment results are matched to generate corresponding instructions, resulting in an instruction set; By associating the instruction set with real-time monitored node status data, the instruction set is adjusted based on the applicability of the instructions to obtain an adjusted instruction set, and a preliminary control draft is determined.

[0044] It is worth noting that the stress analysis of each node in the control priority list involves using existing stress analysis software to integrate historical data such as cumulative deformation of the nodes, and incorporating current environmental variables such as water pressure and temperature changes to perform stress analysis on the nodes. Software examples include Tongji Qimingxing BSC5.0, Yingjianke YJK-JKZH, and Midas Civil. The final node assessment results typically include insufficient or excessive node support, and insufficient node stability. For instance, a stress analysis of a high-priority node, combining historical deformation data and current environmental variables, might conclude that the node's support is insufficient and requires increased support.

[0045] It's worth noting that the pre-set rule base is based on steel support system design specifications, underground engineering construction safety standards, and historical control case experience. Specifically, it includes node judgment result types and corresponding processing instructions. For example, if the node judgment result is insufficient support, the corresponding processing instruction is to add support in two stages, each time by 0.5 times the difference in support force, and retest after 2 hours. If the node judgment result is excessive support, the corresponding processing instruction is to reduce support in two stages, each time by 0.5 times the difference in support force, and retest after 2 hours, and so on. The obtained node judgment results are matched with the pre-set rule base to generate corresponding instructions, resulting in an instruction set. For example, if the node judgment result is insufficient support, the matched instruction is to add support in two stages, each time by 0.5 times the difference in support force, and retest after 2 hours.

[0046] It is worth noting that the real-time monitoring of node status data is achieved by monitoring the actual stress borne by the nodes through preset node stress sensors, thereby determining the node's load status. If the actual stress borne by the node is only 30% or less of the material's design limits, such as the material's yield strength or allowable stress range, the node is determined to be in a low-load state; if the actual stress is between 30% and 70% of the design limits, the node is determined to be in a medium-load state; if the actual stress exceeds 70% of the design limits and approaches or reaches the material's yield strength, the node is determined to be in a high-load state. The instruction set is then adjusted based on the node status. If the node is in a low-load state, no instruction adjustment is needed; if the node is in a medium-load state, the corresponding instructions are added or reduced in four rounds, with each adjustment amount being 0.25 times the difference; if the node is in a high-load state, multiple rounds of adjustments are made, with each adjustment amount controlled within 5% of the difference to avoid significant adjustments under high loads, thus obtaining an adjustment instruction set. Based on the adjustment instruction set, a preliminary control draft including adjustment time, required equipment, and personnel arrangements is generated. For example, if a high-load node needs to increase its support, the corresponding instruction will be adjusted to multiple rounds of adjustment, with each adjustment amounting to 5% of the support difference.

[0047] In step S107, the preliminary regulation draft is simulated, and based on the simulation results, the preliminary regulation draft is optimized according to the imbalance probability to obtain the final regulation scheme, including: The preliminary draft regulation was simulated to obtain the first simulation results; Based on the first simulation results, the imbalance probability of each region is calculated, and regions with imbalance probabilities higher than a preset imbalance threshold are classified as high-risk regions. Based on the high-risk areas, the preliminary control plan was partially revised to obtain a revised control plan; The simulation is performed according to the modified control scheme to obtain a second simulation result. If the imbalance probability of the second simulation result is lower than the preset imbalance threshold, the final control scheme is output.

[0048] It is worth noting that the simulation was conducted using a preliminary control draft, and the simulation results of soil pressure data for each region were obtained using pre-set earth pressure sensors. This yielded the first simulation results. Soil pressure data typically includes the distribution of pressure values ​​within a region, reflecting the stress state of the underground structure. For example, the pressure value in a certain region reached 500 kPa per square meter.

[0049] It is worth noting that the simulation results of soil pressure in each region are used to calculate the probability of imbalance in each region. The specific calculation formula is as follows: Where α is the calculated imbalance probability, and Y is the mean soil pressure in the region. The average soil pressure in a given area is based on historical data. A preset imbalance threshold of 20% is used. If simulation results show that the probability of imbalance in a certain area reaches this threshold, then that area is classified as a high-risk area. The 20% imbalance probability threshold is determined based on safety and control requirements and verified through extensive historical monitoring and simulation samples. It is set based on the fact that in 95% of the historical data samples, the soil pressure distribution is stable and the stress at nodes is balanced when the imbalance probability is below 20%. For example, if the average soil pressure in a certain area is 150 kPa and the historical average soil pressure is 210 kPa, the calculated imbalance probability for that area is 28.6%, exceeding the preset 20% imbalance threshold. Therefore, this area is classified as a high-risk area.

[0050] It is worth noting that, based on high-risk areas, the preliminary control draft is partially revised, with corresponding adjustments made to the instructions and operations related to these areas. If the average soil pressure is greater than the historical average, the soil pressure in that area can be reduced by adding a precipitation-induced pressure reduction instruction or a support structure pressure relief instruction. Conversely, if the average soil pressure is less than the historical average, the soil pressure in that area can be increased by adding a preloading instruction or a support structure compression instruction, thus obtaining a revised control plan. For example, if the average soil pressure in a high-risk area is 260 kPa, which is greater than the historical average of 210 kPa, targeted adjustments are needed. This involves adding a support structure pressure relief instruction, performing three stepped pressure reliefs on the steel supports in that area, each time reducing pressure by 2.5 kN, with a 30-minute interval between each step to remeasure the nodal stress and soil pressure. This prevents excessively rapid pressure relief from causing soil deformation exceeding limits and reduces the soil pressure in that area.

[0051] It is worth noting that, based on the modified control scheme, a second simulation result is obtained, which includes soil pressure data for each region after the adjustment command. The imbalance probability for each region is recalculated. If the imbalance probability for each region is lower than the preset imbalance threshold, the final control scheme is output. If there are still regions that have not reached the balance standard, step S107 is repeated until the imbalance probability for each region is lower than the preset imbalance threshold. For example, after the simulation based on the modified control scheme, in the second simulation result, the average soil pressure in a certain region drops to 225 kPa. The recalculated imbalance probability is 12.7%, which is lower than the preset imbalance threshold of 20%. At this point, the imbalance probability for all regions is lower than the preset imbalance threshold, so the modified control scheme is output as the final control scheme.

[0052] In summary, this invention discloses a deep learning-based dynamic servo control method for axial force of steel supports. It acquires data on the geometric position, displacement, and stress changes of each node, integrates these data to form initial deformation features, and filters out noise to obtain purified deformation features. Based on these features, it analyzes the deformation propagation path and calculates the influence coefficients between nodes to construct an initial deformation matrix. Then, combining the soil property influence parameters of each node, it outputs an optimized deformation matrix and screens high-risk nodes through a graph neural network. Subsequently, it purifies the environmental variable data of high-risk nodes to adjust and optimize the deformation matrix, obtaining an adjusted deformation matrix. It then integrates the node mechanical influence data to form a mechanical influence matrix, and combines the purified environmental variables with linear regression to predict control scores and generate a priority list. Finally, based on the list, it analyzes the node stress, generates an adaptive adjustment instruction set to obtain a preliminary control draft, and through simulation verification and imbalance probability optimization, ultimately forms a control scheme that meets the requirements. This method not only effectively eliminates irrelevant information such as complex environmental interference and sensor errors from deformation characteristics, avoiding these interferences from affecting subsequent stages, but also, based on actual analysis of deformation propagation paths, numerically reconstructs the influence between nodes, and comprehensively processes multiple influencing factors of nodes, efficiently integrating and utilizing data from various aspects to obtain a precise control scheme. Furthermore, it continuously corrects the shortcomings of the scheme, and through multiple rounds of simulation and verification, ensures that the final generated control scheme not only meets engineering safety standards but also has strong feasibility, effectively guaranteeing the safety and accuracy of the entire control process. This solves the shortcomings of existing technologies that cannot comprehensively process multiple influencing factors and formulate precise control strategies.

[0053] Reference Figure 2 The second embodiment of the present invention provides a deep learning-based dynamic servo control system for axial force of steel supports, comprising: The deformation feature acquisition module is used to acquire the geometric position of each node, record the displacement change data and stress change data of the node, integrate them to obtain the initial deformation feature, filter noise from the initial deformation feature to obtain the purified deformation feature; The initial deformation matrix construction module is used to analyze the deformation propagation direction based on the purification deformation characteristics, obtain the deformation propagation path, calculate the influence coefficient between nodes on the deformation propagation path, and use the influence coefficient as the element value of the matrix to construct the initial deformation matrix. The risk identification module is used to obtain the soil property influence parameters of each node and use the initial deformation matrix as input parameters. It uses a graph neural network to output an optimized deformation matrix, and filters high-risk nodes based on the optimized deformation matrix to generate a list of high-risk nodes. The deformation matrix acquisition module is used to acquire environmental variable data of each node in the high-risk node list and perform purification processing to obtain purified environmental variables. The optimized deformation matrix is ​​then adjusted based on the purified environmental variables to obtain the adjusted deformation matrix. The priority determination module is used to acquire the mechanical impact data of the nodes and fuse it with the adjustment deformation matrix to form a mechanical impact matrix. At the same time, it combines the purification environmental variables to predict the node control score through linear regression and generates a control priority list based on the node control score. The preliminary control draft generation module is used to analyze the force situation of each node according to the control priority list, generate node judgment results, generate an applicable set of adjustment instructions based on the node judgment results, and obtain a preliminary control draft. The final control scheme generation module is used to simulate the preliminary control draft, and optimize the preliminary control draft based on the simulation results and the imbalance probability to obtain the final control scheme.

[0054] It should be noted that the deep learning-based steel support axial force dynamic servo control system provided in this embodiment of the invention is used to execute all the process steps of the deep learning-based steel support axial force dynamic servo control method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0055] 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 descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that 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 for those skilled in the art.

[0056] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps described in the method embodiments above.

[0057] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0058] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0059] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0060] The memory can be used to store the computer programs and modules. The processor implements various functions of the electronic device by running or executing the computer programs and modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0061] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0062] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0063] 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 descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that 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 for those skilled in the art.

Claims

1. A deep learning-based dynamic servo control method for axial force of steel supports, characterized in that, include: Obtain the geometric position of each node, record the displacement change data and stress change data of the node, integrate them to obtain the initial deformation characteristics, filter the noise of the initial deformation characteristics to obtain the purified deformation characteristics; Based on the deformation characteristics of the purification process, the deformation propagation direction is analyzed to obtain the deformation propagation path. The influence coefficients between nodes on the deformation propagation path are calculated, and the influence coefficients are used as element values ​​of the matrix to construct an initial deformation matrix. The soil property influence parameters of each node are obtained and used as input parameters along with the initial deformation matrix. The graph neural network is used to output an optimized deformation matrix. High-risk nodes are selected based on the optimized deformation matrix to generate a list of high-risk nodes. The environmental variable data of each node in the high-risk node list is obtained and purified to obtain purified environmental variables. The optimized deformation matrix is ​​adjusted according to the purified environmental variables to obtain the adjusted deformation matrix. The mechanical impact data of the nodes is acquired and fused with the adjustment deformation matrix to form a mechanical impact matrix. At the same time, combined with the purification environmental variables, the node regulation score is predicted by linear regression, and a regulation priority list is generated based on the node regulation score. Based on the control priority list, the stress situation of each node is analyzed, node judgment results are generated, and an applicable set of adjustment instructions is generated based on the node judgment results to obtain a preliminary control draft. The preliminary control draft is simulated, and based on the simulation results, the preliminary control draft is optimized according to the imbalance probability to obtain the final control scheme.

2. The deep learning-based dynamic servo control method for axial force of steel supports according to claim 1, characterized in that, The process involves analyzing the deformation propagation direction based on the deformation characteristics to obtain the deformation propagation path, calculating the influence coefficients between nodes along the deformation propagation path, and using these influence coefficients as element values ​​of a matrix to construct an initial deformation matrix, including: Based on the geometric position of the nodes in the purification deformation features, the nodes are divided into meshes to generate a node connection network diagram; For the aforementioned node connection network diagram, stress transmission at the nodes is simulated using finite element analysis to determine the stress transmission path; Based on the stress transfer path, the deformation propagation direction is analyzed to obtain the deformation propagation path. Calculate the influence coefficients between nodes along the deformation propagation path, and use these influence coefficients as element values ​​of the matrix to construct an initial deformation matrix.

3. The deep learning-based dynamic servo control method for axial force of steel supports according to claim 1, characterized in that, The process involves obtaining soil property influence parameters for each node and using the initial deformation matrix as input parameters. A graph neural network is then used to output an optimized deformation matrix. Based on the optimized deformation matrix, high-risk nodes are selected to generate a list of high-risk nodes, including: Obtain the soil property influence parameters for each node; The initial deformation matrix and the soil property influence parameters are used as input parameters, and a graph neural network is used for training and optimization to output an optimized deformation matrix. From the optimized deformation matrix, nodes that meet the preset high-risk conditions are used to generate a list of high-risk nodes.

4. The deep learning-based dynamic servo control method for axial force of steel supports according to claim 1, characterized in that, The process involves acquiring environmental variable data for each node in the high-risk node list and performing purification processing to obtain purified environmental variables. The optimized deformation matrix is ​​then adjusted based on these purified environmental variables to obtain an adjusted deformation matrix, including: The environmental variable data of each node in the high-risk node list is obtained from the preset environmental database, and outliers are removed from the environmental variable data to obtain purified environmental variables. Analyze the correlation between the purification environmental variables and the optimized deformation matrix to determine the set of correction coefficients for the node propagation weights of the environmental variables; The optimized deformation matrix is ​​mapped and adjusted according to the set of correction coefficients to obtain the adjusted deformation matrix.

5. The deep learning-based dynamic servo control method for axial force of steel supports according to claim 1, characterized in that, The mechanical impact data of the acquired nodes is fused with the adjusted deformation matrix to form a mechanical impact matrix. Simultaneously, combined with the purification environmental variables, node regulation scores are predicted using linear regression. Based on these node regulation scores, a regulation priority list is generated, including: Mechanical influence data for each node is obtained from a pre-set mechanical database. The adjustment deformation matrix and the mechanical influence data are then fused to obtain a mechanical influence matrix. Based on the mechanical influence matrix and the purification environment variables, the node regulation score is predicted by linear regression. If the node's control score exceeds a preset score threshold, it is marked as a high-priority node. For the high-priority nodes, the control scores are sorted to obtain a control priority list.

6. The deep learning-based dynamic servo control method for axial force of steel supports according to claim 1, characterized in that, The process involves analyzing the stress state of each node based on the control priority list, generating node judgment results, and generating an applicable set of adjustment instructions based on the node judgment results to obtain a preliminary control draft, including: Based on the control priority list, analyze the force situation of each node to obtain the node judgment result; Based on a preset rule base, the node judgment results are matched to generate corresponding instructions, resulting in an instruction set; By associating the instruction set with real-time monitored node status data, the instruction set is adjusted based on the applicability of the instructions to obtain an adjusted instruction set, and a preliminary control draft is determined.

7. The deep learning-based dynamic servo control method for axial force of steel supports according to claim 1, characterized in that, The process of simulating the preliminary control draft and optimizing it based on the simulation results, taking into account the probability of imbalance, to obtain the final control scheme includes: The preliminary draft regulation was simulated to obtain the first simulation results; Based on the first simulation results, the imbalance probability of each region is calculated, and regions with imbalance probabilities higher than a preset imbalance threshold are classified as high-risk regions. Based on the high-risk areas, the preliminary control plan was partially revised to obtain a revised control plan; The simulation is performed according to the modified control scheme to obtain a second simulation result. If the imbalance probability of the second simulation result is lower than the preset imbalance threshold, the final control scheme is output.

8. A deep learning-based dynamic servo control system for axial force of steel supports, characterized in that, include: The deformation feature acquisition module is used to acquire the geometric position of each node, record the displacement change data and stress change data of the node, integrate them to obtain the initial deformation feature, filter noise from the initial deformation feature to obtain the purified deformation feature; The initial deformation matrix construction module is used to analyze the deformation propagation direction based on the purification deformation characteristics, obtain the deformation propagation path, calculate the influence coefficient between nodes on the deformation propagation path, and use the influence coefficient as the element value of the matrix to construct the initial deformation matrix. The risk identification module is used to obtain the soil property influence parameters of each node and use the initial deformation matrix as input parameters. It uses a graph neural network to output an optimized deformation matrix, and filters high-risk nodes based on the optimized deformation matrix to generate a list of high-risk nodes. The deformation matrix acquisition module is used to acquire environmental variable data of each node in the high-risk node list and perform purification processing to obtain purified environmental variables. The optimized deformation matrix is ​​then adjusted based on the purified environmental variables to obtain the adjusted deformation matrix. The priority determination module is used to acquire the mechanical impact data of the nodes and fuse it with the adjustment deformation matrix to form a mechanical impact matrix. At the same time, it combines the purification environmental variables to predict the node control score through linear regression and generates a control priority list based on the node control score. The preliminary control draft generation module is used to analyze the force situation of each node according to the control priority list, generate node judgment results, generate an applicable set of adjustment instructions based on the node judgment results, and obtain a preliminary control draft. The final control scheme generation module is used to simulate the preliminary control draft, and optimize the preliminary control draft based on the simulation results and the imbalance probability to obtain the final control scheme.

Images