An Adaptive Cooperative Control Method for Intelligent Mining Tunneling Equipment
By combining multimodal geological sensing with communication status monitoring, and utilizing Bayesian deep learning and hierarchical fine-tuning, the problem of autonomous collaborative control of mining tunneling equipment in complex environments was solved. This achieved deep collaboration between geological self-adaptation and communication robustness, improving the stability and consistency of the system.
Patent Information
- Application Number
- CN202611120567.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-27
- Publication Date
- 2026-11-10
AI Technical Summary
Existing mining tunneling equipment struggles to achieve efficient and stable autonomous collaborative control in the face of complex geological conditions and harsh communication environments. The lack of uncertainty quantification in geological identification results and the disconnect between communication status and control strategies lead to insufficient reliability of control decisions and poor system adaptability.
Deploy a multimodal geological sensing array and a communication status monitoring module, construct a probabilistic geological identification model and a lightweight communication anomaly pattern recognition model, combine a Bayesian deep learning network to achieve probabilistic identification of geological categories and quantification of uncertainty, and optimize geological adaptability and communication robustness through hierarchical fine-tuning and collaborative arbitration processes.
It improves the reliability and smoothness of geological adaptive control, enhances system stability and control consistency in complex downhole environments, and ensures high reliability and smoothness of equipment under different operating conditions.
Smart Images

Figure CN123064474A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of intelligent manufacturing equipment industry and energy-saving equipment retrofit technology, specifically to an adaptive collaborative control method for intelligent mining tunneling equipment. Background Technology
[0002] In the process of intelligent mining construction, autonomous and collaborative control of tunneling equipment is a key link to achieving less-manned or unmanned operations. Due to the complex and variable geological conditions and harsh communication environment underground, existing tunneling equipment control methods generally suffer from poor adaptability and weak coordination when dealing with the dual uncertainties of geology and communication, making it difficult to ensure efficient and stable operation under complex working conditions.
[0003] In existing technologies, research on geological adaptive control often relies on a single geological exploration method or a deterministic neural network model for lithology identification. When the tunneling face is in a geological transition zone, mixed rock layer, or outside the distribution range of the training samples, the model still forces the output of a certain geological category, leading to control decision deviation, which in turn causes cutting parameter mismatch, equipment overload, or efficiency reduction.
[0004] Meanwhile, the tunneling equipment group relies on a central control node for data exchange and command issuance during collaborative operations. The complex underground roadway environment, coupled with factors such as equipment movement, electromagnetic interference, and network topology changes, often leads to increased end-to-end transmission delays, data loss, or out-of-order delivery, resulting in communication quality issues. Traditional control methods treat communication as a transparent channel, failing to consider communication status as an active input for control decisions. When conflicts arise between geological adaptive requirements and communication constraints, the system lacks a coordination mechanism, easily causing control command oscillations or decision failures.
[0005] Furthermore, existing technologies lack coordination in the division of labor between edge computing and central control. Geological perception, communication status, and control decision-making are isolated from each other and cannot form an organic synergy, which restricts the engineering adaptability of intelligent tunneling systems in complex environments. Summary of the Invention
[0006] The purpose of this invention is to provide an adaptive collaborative control method for intelligent mining tunneling equipment to solve the aforementioned background problems.
[0007] The objective of this invention can be achieved through the following technical solution: an adaptive collaborative control method for intelligent mining tunneling equipment, comprising the following steps: Deploy a multimodal geological sensing array and a communication status monitoring module to collect multimodal geological sensing data and record communication status, and encapsulate it into a fusion input vector; A probabilistic geological identification model is constructed and adopted. Based on the fusion input vector, the predicted probability distribution of geological categories is inferred and the output state is evaluated. The basic adjustment value of the current cut-off parameter is determined by combining the output state. Obtain the current communication status. If it is not in a healthy state, perform a second fine-tuning based on the basic adjustment value to form a second adjustment value for the truncation parameter. Calculate the comprehensive difference index by combining the basic adjustment value and the secondary adjustment value, determine whether to trigger the collaborative arbitration process, and generate the final control instruction; Furthermore, the communication status is obtained as follows: The communication status monitoring module continuously collects network communication metrics between edge computing nodes and central control nodes based on a preset fixed sampling period and a sliding time window containing a preset number of sampling periods ending at the current sampling time. These metrics include end-to-end transmission delay sequence, data packet loss rate, data packet out-of-order rate, and data freshness sequence. A three-level state classification is performed based on network communication indicators: if the end-to-end transmission delay sequence and data packet loss rate are both lower than the corresponding preset first delay threshold and first packet loss threshold, the data packet out-of-order rate is 0, and the data freshness sequence has not exceeded the limit, it is judged as a healthy state; if the end-to-end transmission delay sequence exceeds the preset second delay threshold, or if either the data packet loss rate or the data packet out-of-order rate exceeds the preset corresponding second packet loss threshold and out-of-order rate threshold, or if the number of times the data freshness sequence exceeds the limit reaches the fault count standard, it is judged as a fault state; otherwise, it is judged as a sub-healthy state. Furthermore, the probabilistic geological identification model is built on Bayesian deep learning, using a Bayesian neural network as the core architecture. The network weight parameters are set as random variables that follow a prior distribution. The preprocessed fused input vector is used as input, and posterior inference is performed through Monte Carlo Dropout approximation to realize the geological category identification result accompanied by the quantitative output of uncertainty. The output layer uses the Softmax activation function to output the predicted probability distribution of the geological category. Furthermore, the base adjustment value is obtained as follows: The output status index is calculated based on the predicted probability distribution of geological categories, and the output status is evaluated based on the output status index. The output status includes high confidence, medium confidence and low confidence. If the output state is high confidence, the geological category corresponding to the highest predicted probability is directly adopted, and the corresponding optimal cut-off parameter is retrieved and extracted from the preset parameter mapping library as the basic adjustment value. If the output state is medium confidence, the geological categories whose predicted probability reaches the preset confidence benchmark are marked as candidate geological categories. The corresponding optimal cut-off parameters are retrieved from the parameter mapping library and weighted averaged. The weight coefficients are obtained by taking the predicted probability of each candidate geological category and normalizing it within all candidate geological categories. The cut-off parameter change rate is constrained by the progressive parameter adjuster and used as the basic adjustment value. If the output status is low confidence, the preset safety baseline parameter will be used as the basic adjustment value. Furthermore, the calculation method for the output status index is as follows: The current fusion input vector is input into the trained probabilistic geological identification model for inference, and the predicted probability distribution of geological categories is output. The predicted probability distribution is then substituted into the Shannon entropy formula to calculate the entropy value as the output state index. Furthermore, the secondary refinement adopts a hierarchical processing strategy: If the communication status is in a sub-healthy state, a pre-built and trained lightweight communication anomaly pattern recognition model is used based on network communication indicators to identify and output the anomaly pattern of the current sub-healthy state. Based on the anomaly pattern, a communication adaptation correction strategy is determined, and the basic adjustment value of the truncation parameter is processed to form a secondary adjustment value. If the current communication status is in a fault state, a forced locking strategy is executed, the cutting parameters are switched to the preset safe operating parameters as secondary adjustment values, and the local status predictor is started to locally estimate the status changes of the tunneling equipment during the fault state based on the equipment kinematic model and historical operating data. Furthermore, the lightweight communication anomaly pattern recognition model adopts a lightweight gradient booster as the core algorithm, uses standardized network communication indicators as input features, and is trained through a training dataset jointly constructed by a communication simulation platform and field measured data. The identified anomaly patterns are used as outputs. The anomaly patterns include single anomaly patterns or compound anomaly patterns. The determination of compound anomaly patterns is that if multiple single anomaly patterns are identified at the same time, it is determined that a compound anomaly pattern has been identified. Furthermore, the method for determining the communication adaptation correction strategy is as follows: For a single abnormal mode, query the preset mode correction strategy mapping library. The mode correction strategy mapping library stores the correspondence between various single abnormal modes and communication adaptation correction strategies, and obtain the corresponding communication adaptation correction strategy. For compound anomaly patterns, query the communication adaptation correction strategies corresponding to the individual anomaly patterns included in the compound anomaly pattern, and use a pre-trained semantic recognition model to perform conflict analysis on different communication adaptation correction strategies to determine whether there is an implementation conflict. If it is determined that it does not exist, the communication adaptation correction strategy is selected to be executed in parallel or sequentially according to the independent logic of each communication adaptation correction strategy to obtain the communication adaptation correction strategy corresponding to the composite abnormal mode. If a conflict is detected, a conflict resolution process is triggered, and a decision is made based on the predefined single abnormal mode priority level: the communication adaptation correction strategy corresponding to the abnormal mode with the highest priority level is retained and adopted. When multiple communication adaptation correction strategies have the same priority level, the system further adopts the conservative priority principle and selects the communication adaptation correction strategy with smaller adjustment range of truncation parameters, stricter change rate limit, and higher equipment operation safety. Furthermore, the method for determining whether a collaborative arbitration process has been triggered is as follows: Each cutting parameter is normalized based on its rated operating range. The basic adjustment value and the secondary adjustment value after normalization are processed to obtain a comprehensive difference index. If the overall difference index is less than the preset difference standard, the secondary adjustment value of the truncation parameter will be directly issued as the final control command for execution without triggering the collaborative arbitration process; otherwise, it will be triggered. Furthermore, the collaborative arbitration process specifically includes: If the communication status is in a healthy or sub-healthy state, the geological-led principle is adopted, and the basic adjustment value of the cutting parameter is used as the final control command for execution. The direct communication supplementary transmission mechanism between tunneling equipment is activated, and the basic adjustment value is shared by using the direct communication link established between the equipment. If the communication status is in a fault state, the principle of communication adaptation priority is adopted, and the secondary adjustment value of the truncation parameter is used as the final control command to be issued and executed.
[0008] The beneficial effects of this invention are as follows: 1. This invention constructs a multimodal geological perception system that complements active detection, passive response, and visual perception, and combines it with a Bayesian deep learning network to achieve probabilistic identification of geological categories and quantitative output of uncertainty. This solves the problems of insufficient geological information fusion and lack of representation of the reliability of identification results in existing methods. Based on confidence level, the basic adjustment value of the truncation parameter is determined. At medium confidence, weighted fusion and progressive parameter constraints are adopted, and at low confidence, safety parameters are switched to ensure that geological adaptive control has high reliability and smoothness under different working conditions.
[0009] 2. This invention uses the three-level communication status output by the communication status monitoring module as the active input for control decisions, establishing a graded and refined adjustment mechanism for healthy, sub-healthy, and faulty states. It also incorporates a collaborative arbitration process to resolve conflicts between geological adaptation and communication constraints. When there are significant differences between geological adjustments and communication constraints, based on the principle of prioritizing geological performance or communication adaptation, and supplemented by direct inter-device communication, a deep synergy between geological adaptation effectiveness and communication robustness is achieved, significantly improving system stability and control consistency in complex downhole environments. Attached Figure Description
[0010] Figure 1 This is a flowchart illustrating the steps of an adaptive collaborative control method for intelligent mining tunneling equipment according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the specific steps for obtaining the basic adjustment value in an adaptive collaborative control method for intelligent mining tunneling equipment according to an embodiment of the present invention. Detailed Implementation
[0011] Example 1 Please see Figure 1 As shown in the embodiment of the present invention, an adaptive collaborative control method for intelligent mining tunneling equipment aims to solve the problem of insufficient reliability of control decisions due to the lack of uncertainty quantification in geological identification results in existing tunneling equipment control methods. Simultaneously, it addresses the problems of disconnect between communication status and control strategy, poor adaptive capability under complex underground communication conditions, and the lack of a collaborative arbitration mechanism when conflicts arise between geological adaptive adjustment and communication constraints. This achieves deep collaboration between geological adaptive control and communication robustness, specifically including the following steps: S1: Deploy a multimodal geological sensing array and a communication status monitoring module to collect multimodal geological sensing data and record communication status, and encapsulate it into a fusion input vector; Specifically, the multimodal geological sensing array and communication status monitoring module are deployed on the tunneling equipment participating in the collaborative operation. The tunneling equipment includes tunneling machines, transfer machines, and transportation equipment. The tunneling equipment deployed by the multimodal geological sensing array is the tunneling machine, which consists of an active detection layer, a passive response layer, and a visual perception layer. The communication status monitoring module is deployed between the edge computing nodes of each tunneling equipment and the central control node. It should be noted that each tunneling machine is equipped with an independent edge computing node, which is responsible for local data acquisition, preprocessing and some control decisions, and interacts with the central control node through a communication link, aiming to reduce the computing pressure on the central node and improve the real-time response of the local machine. The active detection layer uses an acoustic ground-penetrating radar installed at the front end of the cutting arm of the tunneling machine to emit high-frequency acoustic signals to the rock mass in front of the tunneling machine. Based on the echo delay and energy attenuation characteristics, the spatial distribution of geological stratification interfaces, joint development zones and anomalies within a preset range in front is inverted, and active detection data is obtained. Exemplary anomalies include faults, fracture zones and cavities. The passive response layer integrates a cutting motor current sensor, a cutting arm three-dimensional vibration accelerometer, a hydraulic system pressure sensor, and a cutting head temperature sensor. It collects the dynamic response signal of the cutting medium reacting on the tunneling machine in real time during the cutting process and sorts it to obtain passive response data. The cutting medium refers to the rock mass that the tunneling machine cutting head directly acts on and breaks. The visual perception layer uses a hyperspectral imager and an infrared thermal imager, which are fixedly installed at the front of the tunneling machine to simultaneously capture the texture, mineral composition distribution, and heat accumulation characteristics of the rock cutting surface, and then process the data to obtain visual perception data. At edge computing nodes, active detection data, passive response data, and visual perception data are spatiotemporally aligned according to a unified timestamp to obtain multimodal geological perception data. It should be noted that the active detection layer aims to obtain prior geological information of the undisturbed rock mass in front of the cut, providing advanced data for predictive control. The passive response layer reflects the instantaneous state of the interaction between the current cut process and the rock mass, used to correct control decisions in real time. The visual perception layer provides high-dimensional physical property information of the cut surface, serving as a supplement and verification to the active detection and passive response. The three perception methods complement each other in terms of physical principles, detection depth, and response time, and together constitute a multimodal, all-view geological perception system. The communication status monitoring module continuously collects network communication indicators between edge computing nodes and central control nodes at a preset fixed sampling period, including end-to-end transmission delay sequence, data packet loss rate, data packet out-of-order rate, and data freshness sequence. The end-to-end transmission delay sequence and data freshness sequence are time series sequences within a preset number of sampling periods, with the current sampling time as the endpoint. The data packet loss rate and data packet out-of-order rate are calculated within the sliding time window. Based on the collected network communication indicators, the communication status monitoring module performs a three-level state division and records the communication status. Specifically, if the end-to-end transmission delay sequence and data packet loss rate are both lower than the corresponding preset first delay threshold and first packet loss threshold, the data packet out-of-order rate is 0 and no exceedance is detected on the data freshness sequence, the communication status is determined and recorded as a healthy state. If the end-to-end transmission delay sequence exceeds the preset second delay threshold, or if any network communication indicator in the data packet loss rate and data packet out-of-order rate exceeds the preset corresponding second packet loss threshold or out-of-order rate threshold, or if the number of times the data freshness sequence is detected to exceed the limit reaches the fault count standard, the communication status is determined and recorded as a fault status. If the communication status is neither healthy nor faulty, then the communication status is determined and recorded as sub-healthy. It should be noted that end-to-end transmission delay represents the one-way transmission time interval between two edge computing nodes, data packet loss rate represents the proportion of lost data packets per unit time to the total number of sent data packets, data packet out-of-order rate reflects the degree of disorder in the transmission order of data packets, and data freshness represents the time span from the time of data acquisition to the successful arrival of data at the central control node. The first delay threshold, the first packet loss threshold, the second delay threshold, the second packet loss threshold, the out-of-order rate threshold, and the fault count standard are all stored in the network indicator threshold group. The network indicator threshold group is pre-calibrated according to the control cycle of the tunneling equipment, the communication network topology, and the actual underground environment. At the edge computing node, the multimodal geological sensing data and the communication status output by the current communication status monitoring module are jointly encapsulated into a fusion input vector; S2: Construct and adopt a probabilistic geological identification model, infer the predicted probability distribution of geological categories based on the fused input vector and evaluate the output state, and determine the basic adjustment value of the current cut-off parameter in combination with the output state; like Figure 2 As shown, the specific steps for obtaining the basic adjustment value are as follows; Specifically, the probabilistic geological identification model is built on Bayesian deep learning and uses a Bayesian neural network as its core architecture. The network structure includes an input layer, multiple hidden layers, and an output layer. The number of nodes in the input layer corresponds to the dimension of the fused input vector. The hidden layer adopts a fully connected structure and introduces a Dropout layer to enhance generalization ability. The output layer uses the Softmax activation function to output the probability distribution of geological categories. It should be noted that, unlike deterministic neural networks, the weight parameters of Bayesian neural networks are not fixed values, but are set as random variables that follow a certain prior distribution. By introducing a probability distribution rather than fixed-point estimation into the network weight parameters, the geological category identification results are accompanied by the quantitative output of uncertainty. Training samples are collected to construct a training dataset. Each training sample contains input data and output labels. The input data uses fused input vectors collected during historical tunneling. The output labels are determined based on core data or geological categories identified by geological experts. The input data is preprocessed, including standardization to ensure that data of different dimensions are on the same order of magnitude, outlier removal to avoid sensor noise interfering with model training, and missing value imputation to ensure the continuity of the data sequence and ensure stable distribution of input data. Exemplary geological categories include coal seams, mudstone, sandstone, limestone, and transitional mixed rock formations of various types; The probabilistic geological identification model is trained using a training dataset. Within the Bayesian framework, a normal distribution with zero mean and controllable variance is used as the prior distribution for the network weight parameters. Based on Bayes' theorem, the posterior distribution of the network weight parameters is proportional to the product of the prior distribution and the likelihood function. The likelihood function measures the fit of the probabilistic geological identification model to the training dataset under the current network weight parameters. The cross-entropy loss function is used as the negative logarithmic form of the likelihood function. Minimizing the cross-entropy loss is equivalent to maximizing the likelihood function. By minimizing the difference between the predicted probability and the output label of the training samples, the probabilistic geological identification model learns the mapping relationship from multimodal perception data to geological categories. The Monte Carlo Dropout approximation is used for posterior inference. Dropout is used as a regularization method to randomly discard neurons, enabling the probabilistic geological identification model to learn parameters that are robust to the inactivation of network weight parameters, thus completing the training of the probabilistic geological identification model. Define an identification cycle. At the end of each identification cycle, preprocess the current fused input vector and input it into the trained probabilistic geological identification model for inference. Output the predicted probability distribution of the geological category. Substitute the predicted probability distribution into the Shannon entropy formula to calculate the entropy value as the output state index. Evaluate the output state by comparing the output state index with the preset high uncertainty threshold and low uncertainty threshold: when the output state index is lower than the preset low uncertainty threshold, the output state is determined to be high confidence; when the output state index is between the preset low uncertainty threshold and high uncertainty threshold, the output state is determined to be medium confidence; otherwise, the output state is determined to be low confidence. It should be noted that Bayesian deep learning endows probabilistic geological identification models with the ability to express their cognitive limitations in sparse data regions or beyond the training distribution range. Entropy quantifies the information richness and ambiguity of the prediction results. The lower the entropy value, the higher the model's confidence in the output results. The higher the entropy value, the greater the ambiguity of the model's output or the greater the ambiguity of its reliable identification range. If the output state is high confidence, the geological category with the highest predicted probability in the predicted probability distribution of the geological category output by the probabilistic geological identification model is directly adopted. The preset parameter mapping library is called. The parameter mapping library stores the mapping relationship between each geological category and the optimal cutting parameters. Based on the currently adopted geological category, the corresponding optimal cutting parameters are retrieved and extracted as the basic adjustment value of the current cutting parameters. The cutting parameters include traction speed, cutting head rotation speed and cutting depth. It should be noted that the preset parameter mapping library is based on a comprehensive calibration of historical best practices, equipment manufacturer recommendations, and offline experimental data; If the output state is medium confidence, it is determined that the current tunneling face is in a geological transition zone or mixed rock layer area. The optimal cutting parameters of a single geological category cannot accurately match the actual working conditions. The progressive parameter adjuster is enabled. Based on the probabilistic geological identification model, the predicted probability distribution of geological categories is output. Geological categories whose predicted probabilities reach the preset confidence benchmark are marked as candidate geological categories. The preset parameter mapping library is called to retrieve and extract the corresponding optimal cutting parameters and perform weighted average processing as the basic adjustment value of the current cutting parameters. The weight coefficient is obtained by taking the predicted probability of each candidate geological category and normalizing it within all candidate geological categories. The progressive parameter adjuster constrains the rate of change of the cutting parameters and calculates the amount of change between the current recognition cycle and the basic adjustment value of the previous recognition cycle. If the amount of change of any cutting parameter exceeds the preset physical threshold, the basic adjustment value of the corresponding cutting parameter is limited to the range of the physical threshold to ensure that the cutting parameters change smoothly in the transition region and avoid the impact of sudden changes in cutting parameters on the cutting mechanism. If the output status is low confidence, it is determined that the current working condition exceeds the reliable identification range of the probabilistic geological identification model and cannot make a reliable geological category judgment based on the output of the probabilistic geological identification model. The cutting parameters are automatically switched to the preset safety benchmark parameters as the basic adjustment value. The safety benchmark parameters are a set of conservative cutting parameters to ensure the safe operation of the equipment under unknown or extreme geological conditions. The fused input vector is sent to the operator's terminal to assist the operator in making manual intervention judgments. It should be noted that the three-layer hierarchical processing mechanism constitutes an adaptive control strategy based on model confidence, ensuring that geological identification can output reliable basic adjustment values under various working conditions. The final output basic adjustment values of the cutting parameters include three core parameters: traction speed, cutting head rotation speed, and cutting depth. These reflect the optimal control decision of the system based on the current multimodal geological sensing data and the fusion of probabilistic geological identification model confidence quantification. It should be noted that the basic adjustment value, as an independent geological optimal benchmark, enables the separation of geological decision-making and communication adaptation in computation, while maintaining synergy in results. This is a prerequisite for subsequent secondary fine-tuning and collaborative arbitration. Without this independent benchmark, it is impossible to quantify the degree of distortion of geological decision-making by communication constraints, nor to make an informed arbitration choice in case of conflict. This ensures that even when the geological model is unreliable or communication is completely interrupted, the system still has a clear safety benchmark to follow. Any scheme that omits the basic adjustment value cannot simultaneously achieve high accuracy and high security. S3: Obtain the current communication status. If it is not in a healthy state, perform a second fine-tuning based on the basic adjustment value to form a second adjustment value for the truncation parameter. Specifically, if the current communication status is healthy, it is determined that the communication link has the ability to transmit the data required for geological adaptive decision-making in its entirety. The basic adjustment value of the truncation parameter is kept unchanged and directly used as the secondary adjustment value for output, ensuring that the geological adaptive decision-making is transmitted to the subsequent execution stage with the most complete information. It should be noted that geological adaptive decision-making refers to the control process of dynamically adjusting cutting parameters to ensure that the tunneling equipment maintains optimal cutting efficiency under different geological conditions. If the current communication status is not healthy, a second fine-tuning is performed, and a hierarchical processing strategy is adopted, implementing different adjustment logic for sub-healthy status and fault status respectively; Specifically, if the current communication status is in a sub-healthy state, it is determined that the communication link cannot fully guarantee the real-time performance and integrity of geological adaptive decision-making. A pre-built and trained lightweight communication anomaly pattern recognition model is adopted. Taking the network communication index of the current sliding window as input, it identifies and outputs the anomaly pattern of the current sub-healthy state. Based on the anomaly pattern, it queries the preset pattern correction strategy mapping library to determine the communication adaptation correction strategy for the secondary adjustment value. According to the matched communication adaptation correction strategy, it processes the basic adjustment value of the truncation parameter in combination with the network communication index to form the secondary adjustment value. The lightweight communication anomaly pattern recognition model uses a lightweight gradient booster as its core algorithm, takes standardized network communication indicators as input features, and trains a training dataset jointly constructed by a communication simulation platform and field measured data. It uses a multi-class cross-entropy loss function for training and takes the identified anomaly patterns as output. The anomaly patterns include single anomaly patterns or compound anomaly patterns. The determination of compound anomaly patterns is that if multiple single anomaly patterns are identified at the same time, it is determined that a compound anomaly pattern has been identified. Among them, the pattern correction strategy mapping library stores the correspondence between various single abnormal patterns and communication adaptation correction strategies; In an exemplary single abnormal mode, under sub-health conditions, if the end-to-end transmission delay sequence is continuously higher than the first delay threshold, or the data packet loss rate is not 0 and there is periodic packet loss, it is determined that there is intermittent communication blockage. A first-order low-pass filter is applied to the traction speed of the basic adjustment value. The filter time constant increases linearly according to the mean of the end-to-end transmission delay sequence. At the same time, the cutting depth is reduced by a preset first correction ratio, so that the tunneling equipment adopts a more conservative but more stable operating attitude in the communication uncertainty environment. In an exemplary single abnormal mode, under sub-health conditions, if the variance of the end-to-end transmission delay sequence exceeds a preset fluctuation threshold, it is determined that there is severe jitter in the communication channel. The progressive parameter adjuster further tightens the constraint on the rate of change of the truncation parameter, appropriately narrows the range of the physical threshold, and calculates the change in the truncation parameter of the basic adjustment value between the current identification period and the second adjustment value of the previous identification period, limiting the change to the range of the narrowed physical threshold, ensuring that the truncation parameter will not change abruptly due to data packet disorder or delay when the communication quality is poor. Priorities are predefined for all single anomaly patterns. If a compound anomaly pattern is identified, the corresponding communication adaptation correction strategy is found for the single anomaly patterns included in the compound anomaly pattern. A pre-trained semantic recognition model is used to perform conflict analysis on the communication adaptation correction strategies corresponding to different single anomaly patterns to determine whether there is an implementation conflict. Specifically, the semantic recognition model adopts a lightweight natural language processing model based on the Transformer architecture. It takes the structured description text of each communication adaptation correction strategy as input, extracts the semantic feature vector of each communication adaptation correction strategy through the encoder, and calculates the conflict score between communication adaptation correction strategies in the semantic embedding space. The calculation of the conflict score takes into account the multi-dimensional features of the communication adaptation correction strategy, including the target, direction, intensity, and execution sequence. When two or more communication adaptation correction strategies have opposite directions on the same target, or when the superposition of their intensity exceeds the preset safe range, the semantic recognition model determines that there is an implementation conflict and outputs the conflict type. If the semantic recognition model determines that there is no implementation conflict, the system executes each communication adaptation correction strategy in parallel or sequentially according to its independent logic, processes the basic adjustment value of the truncation parameter, and forms a secondary adjustment value. If an implementation conflict is determined, the system triggers a conflict resolution process and makes a decision based on the predefined single abnormal mode priority level: retain and adopt the communication adaptation correction strategy corresponding to the abnormal mode with the highest priority level; when multiple communication adaptation correction strategies have the same priority level, the system further adopts the conservative priority principle and selects the communication adaptation correction strategy with smaller adjustment range of truncation parameters, stricter change rate limit, and higher equipment operation safety to be executed. It should be noted that the introduction of the communication adaptation correction strategy aims to balance the contradiction between the geological adaptation requirements and communication constraints. In the sub-healthy state of communication, although the system cannot completely rely on remote geological decisions, it still retains the geological adaptation direction of the basic adjustment value. Through parameter smoothing and change rate limitation, the equipment maintains continuous and stable operation when communication fluctuates, and avoids violent oscillation of truncation parameters due to communication quality problems. If the current communication status is in a fault state, it is determined that the communication link can no longer guarantee the reliable transmission of geological sensing data and control commands. A forced locking strategy is executed, and all cutting parameters are switched to preset safe operating parameters as secondary adjustment values to ensure that the equipment can still operate stably according to conservative but safe parameters even if communication is completely interrupted. Each edge computing node starts the local state predictor. The local state predictor estimates the state changes of the tunneling equipment during the fault state based on the equipment kinematic model and historical operating data. Specifically, the local state predictor uses the cutting parameters and fused input vector at the last moment before the fault state as initial conditions, and uses the kinematic equations of the tunneling equipment to recursively predict the position, attitude and cutting load changes at subsequent moments. The prediction results are used for local state monitoring and safety protection of the edge computing node. The local state predictor continues to run until communication is restored. When the communication state monitoring module outputs a healthy or sub-healthy state again, the local state predictor exits the fault handling process and resumes the reception and execution of remote control commands. It should be noted that the forced locking under fault conditions and the local state prediction mechanism constitute a safety redundancy guarantee for communication interruption conditions. The forced locking ensures that the equipment will not enter a dangerous condition due to parameter loss when it loses remote control commands. The local state predictor maintains the continuity of state estimation based on the equipment's own kinematic model, providing a state benchmark for a smooth transition after communication is restored. The three-level communication state hierarchical processing mechanism enables the system to output secondary adjustment values that adapt to the current communication constraints under various communication conditions, realizing the synergistic optimization of geological self-adaptation and communication robustness. S4: Calculate the comprehensive difference index by combining the basic adjustment value and the secondary adjustment value, determine whether the collaborative arbitration process is triggered, and generate the final control instruction; Specifically, the difference between the basic adjustment value and the secondary adjustment value of the cutting parameters is monitored in real time. The quantitative index of the difference is the comprehensive difference index. Each cutting parameter is normalized based on the rated operating range, and the absolute difference between the basic adjustment value and the secondary adjustment value after normalization is calculated. The absolute differences of each cutting parameter are weighted and fused to obtain the comprehensive difference index. The weight coefficients are pre-calibrated according to the degree of influence of each cutting parameter on the safety and efficiency of the cutting process. If the comprehensive difference index is less than the preset difference standard, it is determined that there is no substantial conflict between the geological adaptive requirements and the communication constraints. The collaborative arbitration process is not triggered. The secondary adjustment value of the truncation parameter is directly issued as the final control command for execution. The complete record of this adjustment process is stored in the database, including timestamp, fusion input vector, predicted probability distribution of geological category and corresponding output status, communication status, basic adjustment value, secondary adjustment value, final control command and equipment status feedback after execution. If the comprehensive difference index is greater than or equal to the preset difference standard, it is determined that there is a serious conflict between the geological adaptive requirements and the communication constraints. At this time, the collaborative arbitration process is triggered. The collaborative arbitration process performs branch processing based on the communication status level output by the current communication status monitoring module. Specifically, if the communication status is in a sub-healthy state, but it is determined that the communication link still has the ability to transmit key geological data or some control commands, the geological-led principle shall be given priority. Under the geological-led principle, the basic adjustment value of the cutting parameter shall be used as the final control command for issuance and execution, and the direct communication supplementary transmission mechanism between tunneling equipment shall be activated. By utilizing the direct communication link established between tunneling equipment, the basic adjustment value that may be blocked due to the communication constraints of the central control node shall be shared among the equipment to ensure that each tunneling equipment can still obtain consistent and coordinated final control commands when working together. If the communication status is in a fault state, it is determined that the communication link can no longer guarantee reliable data transmission. At this time, the communication adaptation priority principle is switched to. Under the communication adaptation priority principle, the secondary adjustment value of the truncation parameter is used as the final control command to be issued and executed to ensure stability when communication is completely interrupted. Each edge computing node continuously performs state estimation based on the local state predictor that is started, providing a state benchmark for a smooth transition after communication is restored. It should be noted that when the communication status is healthy, the control commands are identical and will not enter the collaborative arbitration mechanism; The process of generating a final control command is defined as a control cycle. After the final control command is issued and executed, the next control cycle begins. It should be noted that the core of the collaborative arbitration mechanism is to resolve the conflict between geological adaptive optimization and communication constraints. When the difference between the geological adjustment direction and the adjustment direction under communication constraints is small, the secondary adjustment result is directly executed to simplify the control process. When the difference is significant, the communication status determination principle is used. The geological-dominant principle aims to maximize the geological adaptive efficiency, while the communication adaptation priority principle aims to ensure the system stability under extreme working conditions. The direct communication supplementary transmission mechanism between devices serves as a backup for the central communication link, enhancing decision consistency under the geological-dominant principle. The technical solution of this invention is as follows: Deploy a multimodal geological sensing array and a communication status monitoring module to collect multimodal geological sensing data and record the communication status, encapsulate it into a fusion input vector, use the constructed probabilistic geological identification model to infer the predicted probability distribution of geological categories based on the fusion input vector and evaluate the output status, determine the basic adjustment value of the current truncation parameter based on the output status, obtain the current communication status, if it is not in a healthy state, perform a second fine adjustment based on the basic adjustment value to form a second adjustment value of the truncation parameter, calculate a comprehensive difference index by combining the basic adjustment value and the second adjustment value, determine whether to trigger the collaborative arbitration process, and generate the final control command.
[0012] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.
Claims
1. An adaptive collaborative control method for intelligent mining tunneling equipment, characterized in that: Includes the following steps: Deploy a multimodal geological sensing array and a communication status monitoring module to collect multimodal geological sensing data and record communication status, and encapsulate it into a fusion input vector; A probabilistic geological identification model is constructed and adopted. Based on the fusion input vector, the predicted probability distribution of geological categories is inferred and the output state is evaluated. The basic adjustment value of the current cut-off parameter is determined by combining the output state. Obtain the current communication status. If it is not in a healthy state, perform a second fine-tuning based on the basic adjustment value to form a second adjustment value for the truncation parameter. The comprehensive difference index is calculated by combining the basic adjustment value and the secondary adjustment value to determine whether the collaborative arbitration process is triggered and to generate the final control instruction.
2. The adaptive collaborative control method for intelligent mining tunneling equipment according to claim 1, characterized in that: The communication status is obtained as follows: The communication status monitoring module continuously collects network communication metrics between edge computing nodes and central control nodes based on a preset fixed sampling period and a sliding time window containing a preset number of sampling periods ending at the current sampling time. These metrics include end-to-end transmission delay sequence, data packet loss rate, data packet out-of-order rate, and data freshness sequence. A three-level state classification is performed based on network communication indicators: if the end-to-end transmission delay sequence and data packet loss rate are both lower than the corresponding preset first delay threshold and first packet loss threshold, the data packet out-of-order rate is 0, and the data freshness sequence does not exceed the limit, it is judged as a healthy state; if the end-to-end transmission delay sequence exceeds the preset second delay threshold, or if either the data packet loss rate or the data packet out-of-order rate exceeds the preset corresponding second packet loss threshold and out-of-order rate threshold, or if the number of times the data freshness sequence exceeds the limit reaches the fault count standard, it is judged as a fault state; otherwise, it is judged as a sub-healthy state.
3. The adaptive cooperative control method for intelligent mining tunneling equipment according to claim 1, characterized in that: The probabilistic geological identification model is built on Bayesian deep learning and uses a Bayesian neural network as its core architecture. The network weight parameters are set as random variables that follow a prior distribution. The model takes a fused input vector that has been preprocessed as input and performs posterior inference through Monte Carlo Dropout approximation to achieve the output of geological category identification results with quantitative uncertainty. The output layer uses the Softmax activation function to output the predicted probability distribution of geological categories.
4. The adaptive cooperative control method for intelligent mining tunneling equipment according to claim 3, characterized in that: The base adjustment value is obtained as follows: The output status index is calculated based on the predicted probability distribution of geological categories, and the output status is evaluated based on the output status index. The output status includes high confidence, medium confidence and low confidence. If the output state is high confidence, the geological category corresponding to the highest predicted probability is directly adopted, and the corresponding optimal cut-off parameter is retrieved and extracted from the preset parameter mapping library as the basic adjustment value. If the output state is medium confidence, the geological categories whose predicted probability reaches the preset confidence benchmark are marked as candidate geological categories. The corresponding optimal cut-off parameters are retrieved from the parameter mapping library and weighted averaged. The weight coefficients are obtained by taking the predicted probability of each candidate geological category and normalizing it within all candidate geological categories. The cut-off parameter change rate is constrained by the progressive parameter adjuster and used as the basic adjustment value. If the output status is low confidence, the preset safety baseline parameter will be used as the basic adjustment value.
5. The adaptive cooperative control method for intelligent mining tunneling equipment according to claim 4, characterized in that: The output status index is calculated as follows: The current fusion input vector is input into the trained probabilistic geological identification model for inference, and the predicted probability distribution of geological categories is output. The predicted probability distribution is then substituted into the Shannon entropy formula to calculate the entropy value as the output state index.
6. The adaptive collaborative control method for intelligent mining tunneling equipment according to claim 1, characterized in that: The secondary fine-tuning adopts a hierarchical processing strategy: If the communication status is in a sub-healthy state, a pre-built and trained lightweight communication anomaly pattern recognition model is used based on network communication indicators to identify and output the anomaly pattern of the current sub-healthy state. Based on the anomaly pattern, a communication adaptation correction strategy is determined, and the basic adjustment value of the truncation parameter is processed to form a secondary adjustment value. If the current communication status is in a fault state, a forced locking strategy is executed, the cutting parameters are switched to the preset safe operating parameters as secondary adjustment values, and the local state predictor is started to locally estimate the state changes of the tunneling equipment during the fault state based on the equipment kinematic model and historical operating data.
7. The adaptive cooperative control method for intelligent mining tunneling equipment according to claim 6, characterized in that: The lightweight communication anomaly pattern recognition model uses a lightweight gradient booster as its core algorithm. It takes standardized network communication metrics as input features and trains the model using a training dataset jointly constructed by a communication simulation platform and field test data. The model outputs the identified anomaly patterns, which can be either single or compound anomaly patterns. A compound anomaly pattern is determined by identifying multiple single anomaly patterns simultaneously.
8. The adaptive cooperative control method for intelligent mining tunneling equipment according to claim 7, characterized in that: The communication adaptation correction strategy is determined as follows: For a single abnormal mode, query the preset mode correction strategy mapping library. The mode correction strategy mapping library stores the correspondence between various single abnormal modes and communication adaptation correction strategies, and obtain the corresponding communication adaptation correction strategy. For compound anomaly patterns, query the communication adaptation correction strategies corresponding to the individual anomaly patterns included in the compound anomaly pattern, and use a pre-trained semantic recognition model to perform conflict analysis on different communication adaptation correction strategies to determine whether there is an implementation conflict. If it is determined that it does not exist, the communication adaptation correction strategy is selected to be executed in parallel or sequentially according to the independent logic of each communication adaptation correction strategy to obtain the communication adaptation correction strategy corresponding to the composite abnormal mode. If a conflict is detected, a conflict resolution process is triggered, and a decision is made based on the predefined single exception mode priority level: the communication adaptation correction strategy corresponding to the exception mode with the highest priority level is retained and adopted. When multiple communication adaptation correction strategies have the same priority level, the system further adopts the conservative priority principle and selects the communication adaptation correction strategy with smaller adjustment range of truncation parameters, stricter change rate limit, and higher equipment operation safety.
9. The adaptive cooperative control method for intelligent mining tunneling equipment according to claim 1, characterized in that: The method for determining whether a collaborative arbitration process has been triggered is as follows: Each cutting parameter is normalized based on its rated operating range. The basic adjustment value and the secondary adjustment value after normalization are processed to obtain a comprehensive difference index. If the overall difference index is less than the preset difference standard, the secondary adjustment value of the truncation parameter will be directly issued as the final control command for execution without triggering the collaborative arbitration process; otherwise, it will be triggered.
10. The adaptive cooperative control method for intelligent mining tunneling equipment according to claim 9, characterized in that: The collaborative arbitration process specifically includes: If the communication status is in a sub-healthy state, the geological-led principle is adopted, and the basic adjustment value of the cutting parameter is used as the final control command for execution. The direct communication supplementary transmission mechanism between tunneling equipment is activated, and the basic adjustment value is shared by using the direct communication link established between the equipment. If the communication status is in a fault state, the principle of communication adaptation priority is adopted, and the secondary adjustment value of the truncation parameter is used as the final control command to be issued and executed.
