A tailings mining control method and system

Abstract

This invention belongs to the field of tailings mining technology, specifically relating to a tailings mining control method and system. By constructing a multi-dimensional over-limit criterion based on deviation, vibration spectrum, and load torque, a working condition reconfiguration mechanism is actively triggered. During the freeze period, a response matrix is ​​constructed using steady-state data, and parameter corrections are solved to achieve online updates of interpolation function nodes and slopes. Furthermore, based on the updated mapping relationship, a locally optimal parameter set is generated for multiple devices, and interference risk is dynamically calculated according to the predicted trajectory. If conflicts exist, the area is redistributed according to operational efficiency, and conflict-free paths are generated. This effectively suppresses execution deviations caused by parameter mismatch, avoids multi-machine trajectory interference, ensures stable and coordinated operation of the system under complex disturbance conditions, and significantly improves operational continuity and overall control robustness.

Description

Technical Field

[0001] This invention belongs to the field of tailings mining technology, specifically relating to a tailings mining control method and system. Background Technology

[0002] In tailings recovery operations, existing control methods typically generate control commands based on preset static parameter mapping tables or empirical models to drive the coordinated operation of excavating, conveying, and positioning actuators. While such methods can maintain basic operational efficiency under stable operating conditions, they struggle to adaptively adjust control parameters under complex conditions such as sudden changes in tailings physical characteristics, equipment load fluctuations, or abnormal vibrations. This can easily lead to accumulated execution deviations, interference between multiple machine trajectories, and even abnormal equipment shutdowns.

[0003] The core problem with existing technologies is that under dynamic operating conditions, there is a lack of closed-loop linkage mechanisms for parameter self-correction and multi-machine collaborative avoidance, resulting in poor system stability, high job interruption rate, and difficulty in guaranteeing overall efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a tailings mining control method and system that significantly reduces execution deviations and trajectory interference caused by parameter mismatch, enabling the system to maintain continuous and stable operation under complex working conditions, thereby solving the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a tailings mining control method, comprising:

[0006] Collect physical property data of tailings and mark high-variance areas, then extract the corresponding data of the high-variance areas to form an input vector;

[0007] After normalizing the input vector, the initial parameter vector is obtained by interpolation, and then the control command is generated by the proportional-integral-derivative controller.

[0008] The control command is broken down into independent signals to drive the actuator. The actual output is collected synchronously and compared with the command to calculate the deviation. If the deviation exceeds the limit, it is fed back to the regulator input for correction. At the same time, the vibration spectrum and load torque are collected.

[0009] The proportion of the main frequency energy is calculated based on the vibration spectrum, and the load fluctuation coefficient is calculated based on the load torque. If both of these exceed the limits and the deviation exceeds the limit, the parameters are frozen and the working condition is reconfigured.

[0010] In the condition reconfiguration, a response matrix is ​​constructed based on the physical characteristics and steady-state output data of the equipment collected during the freeze period. The parameter correction is solved by the least squares method, and the node values ​​and slope of the interpolation function are updated.

[0011] Using the updated interpolation function mapping relationship, a local optimal parameter set is generated for multiple devices. The interference index is calculated based on the predicted trajectory of the devices. If interference exists, the work area is redistributed according to the work efficiency and a conflict-free path is generated.

[0012] After the path and parameters are distributed, data from the entire lifecycle are collected to build a data lake. Through principal component analysis and clustering, abnormal and efficient patterns are obtained, a parameter mapping strategy is trained, and the update is completed by replacing the interpolation function with a canary release.

[0013] Preferably, the step of collecting tailings physical characteristic data and marking high-variance areas includes:

[0014] A sensor array is deployed in front of the work area to collect data on density, moisture content, particle size distribution index, and shear strength to obtain physical property data.

[0015] Spatial gradient calculation is performed on physical property data, and regions with gradient values ​​greater than a threshold are marked as high-variable regions;

[0016] The coordinates and physical property data of the high-variable area are packaged and sent to the control center.

[0017] Preferably, the control command generated by the proportional-integral-derivative controller includes:

[0018] Receive the coordinates of the high-variable area, extract the physical characteristic data of the corresponding location to form a four-dimensional input vector, perform minimum and maximum scaling on the input vector, and map it to the [0,1] interval;

[0019] Search for neighboring points with a distance less than 1 in the historical task database, and obtain the initial parameter vector by weighted interpolation based on the inverse of the distance;

[0020] The initial parameter vector and the deviation from the ideal state are input into the proportional-integral-derivative controller. The output correction is superimposed and boundary clamping is performed to form the final control command.

[0021] Preferably, the control command is decomposed into independent signal-driven actuators, and the actual output is synchronously collected and compared with the command to calculate the deviation, including:

[0022] The control command vector is decomposed into pressure setpoint, speed command, and frequency command according to physical quantities, and distributed to the proportional relief valve controller, frequency converter, and motion control module through the industrial real-time communication protocol.

[0023] Every 10 milliseconds, the hydraulic cylinder position, motor speed, and travel speed are collected to form the actual output vector;

[0024] The deviation between the actual output vector and the original instruction is calculated after performing a 5-point moving average filter on the actual output vector.

[0025] If the deviation norm exceeds the limit, the deviation is multiplied by the compensation gain and fed back to the input of the proportional-integral-derivative controller, with a maximum of three compensations.

[0026] Preferably, the step of feeding back to the regulator input for correction if the deviation exceeds the limit includes:

[0027] The three-axis accelerometers installed at the excavator arm hinge point, slewing platform, and drive wheel axle were used to collect vibration signals at a frequency of 2000Hz.

[0028] After performing anti-aliasing filtering on the collected vibration signal, the fast Fourier transform spectrum is calculated;

[0029] The proportion of energy in the main frequency band to the total energy is extracted based on the Fast Fourier Transform spectrum. If the proportion is less than 0.5, it is marked as an abnormal vibration.

[0030] The load torque is collected synchronously by the torque sensor. The ratio of the standard deviation to the mean is calculated based on the most recent 200 data points. If the ratio is greater than 0.3, it is marked as an abnormal load.

[0031] If the vibration anomaly marker and the deviation both exceed the limit, and the condition lasts for more than 0.8 seconds, a condition reconfiguration request will be generated and sent to the control center.

[0032] Preferably, the step of solving for the parameter correction using the least squares method and updating the node values ​​and slope of the interpolation function includes:

[0033] Upon receiving the working condition reconfiguration request, a parameter lock command is sent to the actuator and its speed is reduced to 50% until the operation is stable.

[0034] During parameter locking, capture 10 sets of physical characteristic data and corresponding steady-state data of device output within the last 5 seconds;

[0035] Based on the 10 sets of physical characteristic data and the corresponding steady-state data of device output, the linear regression slope between each physical characteristic dimension and the device output component is calculated, and a response matrix composed of the linear regression slope is formed.

[0036] The current deviation and the historical cumulative deviation are weighted and summed as the objective function. Combined with the response matrix composed of the linear regression slope, the parameter correction amount obtained by the least squares method is obtained by inputting the least squares method.

[0037] Locate the nearest interpolation nodes in the historical operation database, adjust the corresponding parameter values ​​of the nearest interpolation nodes based on the parameter correction amount obtained by the least squares method, and update the relevant parameters of the proportional-integral-derivative controller simultaneously.

[0038] Preferably, the step of reallocating the work area and generating a conflict-free path based on work efficiency if interference exists includes:

[0039] For each device, the updated mapping relationship is invoked to generate a parameter sequence for the next 10 seconds;

[0040] Based on the parameter sequence and current position, predict the motion trajectory and calculate the minimum distance between any two devices;

[0041] The ratio of the minimum distance to the safe distance is used as the interference index; if it is less than 1, interference is considered to exist.

[0042] The ratio of the amount of data collected per unit time to the energy consumption of each device is calculated as the operational efficiency. For interfering devices, the original path of the one with higher efficiency is retained, while the path of the one with lower efficiency is shifted, the timing is staggered, or the area is exchanged.

[0043] Preferably, after the path and parameters are issued, full-cycle data is collected to build a data lake, and anomaly and efficiency patterns are obtained through principal component analysis and clustering, including:

[0044] Physical characteristic data, control command data, actual output data, response matrix data, path trajectory data, and operation efficiency data are stored in a distributed database using time and spatial indexes to form a full-cycle data lake;

[0045] Principal component analysis is performed on the high-dimensional data stored in the distributed database to reduce dimensionality, retaining principal components with a cumulative variance contribution rate of 95%, and then clustering algorithms are applied to extract cluster centers and cluster boundaries to form an abnormal pattern library and an efficient pattern library.

[0046] Using physical characteristic data as state input, equipment parameter data as action output, and the weighted sum of energy consumption data, wear data, and delay data as negative reward, a deep deterministic policy gradient network is trained based on an abnormal pattern library and an efficient pattern library to obtain a parameter mapping policy.

[0047] The parameter mapping strategy first performs shadow running, then performs grayscale testing on a single device, and finally performs a full switch. If the preset circuit breaker condition is triggered at any stage, it will automatically roll back to the original interpolation function.

[0048] Preferably, the grayscale release replaces the interpolation function to complete the update, including:

[0049] The parameter mapping strategy receives the same physical characteristic input data in parallel with the original interpolation function, and only the original interpolation function outputs actual control commands to drive the actuator;

[0050] During parallel operation, the deviation between the predicted output data and the actual execution output data of the parameter mapping strategy is recorded. When the prediction accuracy is higher than 95%, the gray release stage begins.

[0051] During the canary release phase, select an edge computing device to switch to control commands generated by parameter mapping strategy and continuously monitor key performance indicators for 24 hours.

[0052] When 24-hour monitoring results show that the operation efficiency has improved by more than 5% and no new abnormal mode has been triggered, the parameter mapping strategy will be extended to 50% of the devices.

[0053] On the other hand, the present invention proposes a tailings recovery control system, including a sensor array, a control center, and an actuator;

[0054] The sensor array collects physical characteristic data of tailings and marks high-variance areas. The data corresponding to the high-variance areas are extracted to form an input vector and sent to the control center.

[0055] The control center normalizes the input vector and interpolates it to obtain the initial parameter vector, which is then used by the proportional-integral-derivative controller to generate control commands, which are then broken down into independent signals to drive the actuators.

[0056] After the actuator moves, the control center collects the actual output and compares it with the command to calculate the deviation. If the deviation exceeds the limit, feedback is provided for correction. At the same time, the vibration spectrum and load torque are collected.

[0057] The control center calculates the main frequency energy ratio and load fluctuation coefficient. If both exceed the limit, the parameters are frozen and the operating condition is reconfigured.

[0058] The control center constructs the response matrix, solves for the correction using the least squares method, and updates the interpolation nodes and slope.

[0059] The control center generates parameter sets for multiple devices, calculates the interference index, and if interference occurs, it redistributes the region according to efficiency and generates conflict-free paths.

[0060] After the control center issues the path and parameters, the data server collects full-cycle data to build a data lake, obtains patterns through principal component analysis and clustering, trains strategies, and completes the update through canary release.

[0061] Technical effects and advantages of the present invention: The tailings mining control method and system proposed in this invention have the following advantages compared with the prior art:

[0062] This invention proactively triggers a working condition reconfiguration mechanism by constructing a multi-dimensional over-limit criterion based on deviation, vibration spectrum, and load torque. During the freeze period, it utilizes steady-state data to construct a response matrix and solve for parameter corrections, enabling online updates of interpolation function nodes and slopes. Based on the updated mapping relationship, it generates locally optimal parameter sets for multiple devices and dynamically calculates interference risks according to predicted trajectories. If conflicts exist, the region is redistributed according to operational efficiency, and conflict-free paths are generated. This effectively suppresses execution deviations caused by parameter mismatch, avoids multi-machine trajectory interference, ensures stable and coordinated operation of the system under complex disturbance conditions, and significantly improves operational continuity and overall control robustness. Attached Figure Description

[0063] Figure 1 This is a flowchart of a tailings mining control method according to the present invention;

[0064] Figure 2 This is a block diagram of a tailings recovery control system according to the present invention. Detailed Implementation

[0065] 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. The specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention. 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.

[0066] This invention provides, for example Figure 1 The tailings mining control method shown includes the following steps:

[0067] Collect tailings physical property data and mark high-variable areas, extract the corresponding data of high-variable areas to form an input vector; specifically, this includes: deploying a sensor array in front of the working face to collect density, moisture content, particle size distribution index, and shear strength to obtain physical property data; performing spatial gradient calculation on the physical property data, marking areas with gradient values ​​greater than a threshold as high-variable areas; and packaging the coordinates of the high-variable areas and the physical property data and sending them to the control center.

[0068] By identifying regions of abrupt changes in physical characteristics and extracting key data in a targeted manner, the representativeness and sensitivity of the control model input are significantly improved, enabling subsequent parameter generation to respond earlier to changes in geological conditions. This avoids control mismatch or equipment malfunctions caused by uncaptured local abrupt changes, and enhances the system's adaptability and control foresight in complex tailings environments.

[0069] The input vector is normalized and interpolated to obtain the initial parameter vector, which is then used by a proportional-integral-derivative (PID) controller to generate control commands. Specifically, this includes: receiving the coordinates of the high-variable zone, extracting the physical characteristic data of the corresponding location to form a four-dimensional input vector, performing minimum-maximum scaling on the input vector, and mapping it to the [0,1] interval; searching for neighboring points less than 1 in the historical operation database, and interpolating the initial parameter vector by weighting the inverse of the distance; inputting the initial parameter vector and the deviation from the ideal state into the PID controller, and superimposing the output correction amount to perform boundary clamping to form the final control command.

[0070] By constructing highly adaptive initial parameters through normalization and neighborhood interpolation, and combining proportional-integral-derivative closed-loop correction and output limiting, control oscillations caused by uneven distribution of physical characteristics or sparse historical data are effectively suppressed, improving the smoothness and stability of command generation, ensuring that the actuator responds and acts continuously under complex working conditions, and avoiding overshoot or control failure.

[0071] The control command is decomposed into independent signals to drive the actuator. The actual output is collected synchronously and compared with the command to calculate the deviation. Specifically, the control command vector is decomposed into pressure setpoint, speed command, and frequency command according to physical quantities, and distributed to the proportional relief valve controller, frequency converter, and motion control module through an industrial real-time communication protocol. The hydraulic cylinder position, motor speed, and travel speed are collected every 10 milliseconds to form the actual output vector. The actual output vector is filtered by a 5-point moving average and the deviation is calculated with the original command. If the deviation norm exceeds the limit, the deviation is multiplied by the compensation gain and fed back to the input of the proportional-integral-derivative controller, with a maximum of three compensations.

[0072] It achieves physical quantity mapping of control commands and millisecond-level closed-loop tracking. Through real-time deviation detection and finite-order gain compensation, it effectively suppresses control drift caused by actuator response lag and external disturbances. While ensuring the dynamic following performance of the system, it avoids oscillations caused by over-correction, thereby improving the overall control accuracy and operational stability.

[0073] If the deviation exceeds the limit, feedback is sent to the regulator input for correction. Simultaneously, vibration spectrum and load torque are collected. Specifically, this includes: calling the three-axis accelerometers installed at the excavator arm hinge point, slewing platform, and drive wheel axle to collect vibration signals at a frequency of 2000Hz; performing anti-aliasing filtering on the collected vibration signals and calculating the Fast Fourier Transform spectrum; extracting the proportion of energy in the main frequency band to the total energy based on the Fast Fourier Transform spectrum; if the proportion is less than 0.5, it is marked as a vibration anomaly; synchronously calling the torque sensor to collect load torque, calculating the ratio of standard deviation to mean based on the most recent 200 data points; if the ratio is greater than 0.3, it is marked as a load anomaly; if the vibration anomaly flag and the deviation both exceed the limit, and the state lasts for more than 0.8 seconds, a working condition reconfiguration request is generated and sent to the control center.

[0074] By using multi-source sensor collaborative monitoring and dynamic threshold judgment, early abnormal perception of equipment operating status and prediction of operating condition degradation can be achieved. Before control deviations cause serious consequences, the system-level reconfiguration mechanism can be proactively triggered, effectively avoiding equipment damage and operation interruption caused by vibration instability or sudden load changes, and significantly improving the system's fault prevention and control capabilities and operational safety under complex disturbances.

[0075] The proportion of the main frequency energy is calculated based on the vibration spectrum, and the load fluctuation coefficient is calculated based on the load torque. If both of these exceed the limits and the deviation exceeds the limit, the parameters are frozen and the working condition is reconfigured.

[0076] In the condition reconfiguration process, a response matrix is ​​constructed based on the physical characteristics and steady-state output data of the equipment collected during the freeze period. The parameter correction amount is solved using the least squares method, and the node values ​​and slopes of the interpolation function are updated. Specifically, this includes: after receiving the condition reconfiguration request, sending a parameter lock command to the actuator and controlling its speed to decrease to 50% until it stabilizes; during the parameter lock period, capturing 10 sets of physical characteristic data and corresponding steady-state output data of the equipment within the last 5 seconds; based on the 10 sets of physical characteristic data and corresponding steady-state output data of the equipment, calculating the linear regression slope between each physical characteristic dimension and the equipment output component, and forming a response matrix composed of the linear regression slopes; weighted summing of the current deviation and the historical cumulative deviation as the objective function, combined with the response matrix composed of the linear regression slopes, and inputting it to solve for the parameter correction amount obtained by the least squares method; locating the neighboring interpolation nodes in the historical operation database, adjusting the corresponding parameter values ​​of the neighboring interpolation nodes according to the parameter correction amount obtained by the least squares method, and synchronously updating the relevant parameters of the proportional-integral-derivative controller.

[0077] In the early stages of system anomalies, the current control parameters are frozen and the system actively enters the reconfiguration mode. Through the steady-state data-driven least squares parameter correction mechanism, the control model is self-calibrated online, effectively eliminating long-term control deviations caused by geological changes or equipment characteristic drift, restoring the system's dynamic matching capability, avoiding performance degradation or cascading failures caused by continuous parameter mismatch, and significantly enhancing the environmental adaptability and long-term operational robustness of the control system.

[0078] Using the updated interpolation function mapping relationship, a locally optimal parameter set is generated for multiple devices. The interference index is calculated based on the predicted trajectory of the devices. If interference exists, the work area is redistributed according to the work efficiency and a conflict-free path is generated. Specifically, this includes: generating a parameter sequence for the next 10 seconds for each device by calling the updated mapping relationship; predicting the motion trajectory based on the parameter sequence and the current position, and calculating the minimum distance between any two devices; using the ratio of the minimum distance to the safety distance as the interference index, and determining that interference exists if it is less than 1; calculating the ratio of the amount of data collected per unit time to the energy consumption of each device as the work efficiency; for interfering device pairs, retaining the original path of the one with higher efficiency, and performing path translation, time-series staggering, or area exchange for the one with lower efficiency.

[0079] Based on dynamic parameter updates, the system enables real-time optimization and proactive conflict resolution of multi-machine collaborative paths, effectively avoiding spatial interference between equipment and internal consumption of operational resources. While ensuring safe spacing, it maximizes overall operational efficiency, significantly improves the collaborative efficiency and space utilization of multi-machine parallel mining, and ensures orderly, conflict-free, and high-output system operation under complex working conditions.

[0080] After the path and parameters are distributed, data from the entire lifecycle are collected to build a data lake. Through principal component analysis and clustering, abnormal and efficient patterns are obtained, a parameter mapping strategy is trained, and the update is completed by replacing the interpolation function with a canary release.

[0081] Specifically, this includes: storing physical characteristic data, control command data, actual output data, response matrix data, path trajectory data, and operation efficiency data into a distributed database using time and spatial indices to form a full-cycle data lake; performing principal component analysis to reduce the dimensionality of the high-dimensional data stored in the distributed database, retaining principal components with a cumulative variance contribution rate of 95%, and then applying clustering algorithms to extract cluster centers and cluster boundaries to form an abnormal pattern library and an efficient pattern library; using physical characteristic data as state input, equipment parameter data as action output, and the weighted sum of energy consumption data, wear data, and delay data as negative reward, training a deep deterministic policy gradient network based on the abnormal pattern library and the efficient pattern library to obtain a parameter mapping policy; first, performing shadow running on the parameter mapping policy, then performing gray-scale testing on a single device, and finally performing a full switch. If a preset circuit breaker condition is triggered at any stage, the system will automatically roll back to the original interpolation function.

[0082] Further steps include: the parameter mapping strategy and the original interpolation function receive the same physical characteristic input data in parallel, with the original interpolation function only outputting actual control commands to drive the actuator; during parallel operation, the deviation between the predicted output data and the actual execution output data of the parameter mapping strategy is recorded, and when the prediction accuracy is higher than 95%, the gray-scale release phase begins; during the gray-scale release phase, one edge computing device is selected to switch to the control commands generated by the parameter mapping strategy, and key performance indicators are continuously monitored for 24 hours; when the 24-hour monitoring results show that the operational efficiency has improved by more than 5% and no new abnormal modes have been triggered, the parameter mapping strategy is expanded and deployed to 50% of the devices.

[0083] By training pattern recognition and reinforcement learning strategies driven by full-cycle data, the control logic can autonomously evolve from experience interpolation to intelligent mapping. Under the premise of ensuring a stable transition of the system, it gradually replaces the traditional parameter generation mechanism, enabling the control strategy to continuously absorb historical high-efficiency operation experience and avoid abnormal operating conditions. This significantly improves the system's long-term self-optimization capability and environmental adaptability. At the same time, through a progressive gray-scale verification mechanism, it ensures that the new strategy can be replaced safely, controllably, and seamlessly, avoiding operational risks caused by sudden changes in the strategy.

[0084] On the other hand, this invention proposes a tailings recovery control system, such as Figure 2 As shown, it includes a sensor array, a control center, and an actuator;

[0085] The sensor array collects physical characteristic data of tailings and marks high-variance areas. The data corresponding to the high-variance areas are extracted to form an input vector and sent to the control center.

[0086] The control center normalizes the input vector and interpolates it to obtain the initial parameter vector, which is then used by the proportional-integral-derivative controller to generate control commands, which are then broken down into independent signals to drive the actuators.

[0087] After the actuator moves, the control center collects the actual output and compares it with the command to calculate the deviation. If the deviation exceeds the limit, feedback is provided for correction. At the same time, the vibration spectrum and load torque are collected.

[0088] The control center calculates the main frequency energy ratio and load fluctuation coefficient. If both exceed the limit, the parameters are frozen and the operating condition is reconfigured.

[0089] The control center constructs the response matrix, solves for the correction using the least squares method, and updates the interpolation nodes and slope.

[0090] The control center generates parameter sets for multiple devices, calculates the interference index, and if interference occurs, it redistributes the region according to efficiency and generates conflict-free paths.

[0091] After the control center issues the path and parameters, the data server collects full-cycle data to build a data lake, obtains patterns through principal component analysis and clustering, trains strategies, and completes the update through canary release.

[0092] In addition, the aforementioned sensor array, control center, actuator, communication network, and data server are also used to implement other steps of the tailings recovery control method described above, as follows:

[0093] Step 1: Collect local physical characteristic data of the tailings dump

[0094] This step constructs a high-density, high-response, and highly robust in-situ sensing network to continuously, synchronously, and multidimensionally collect key physical quantities within a range of 0.5-3 meters in front of the work area, providing real, timely, and complete status data for subsequent equipment parameter adjustments.

[0095] 1.1 Install multi-channel sensor probes at the front end of the mining equipment, with one set arranged laterally at 0.5-meter intervals, to simultaneously acquire the physical state values ​​of each point on the current working section, denoted as... ,in For planar coordinates, For timestamps;

[0096] This step constructs the hardware infrastructure for physical sensing. An array of sensor probes is deployed horizontally at 0.5-meter intervals at the front end of the mining bucket or excavator arm. Each probe group contains four independent sensing units, used to measure density, moisture content, particle size distribution index, and shear strength, respectively. The probes are encapsulated in a wear-resistant alloy shell and contain piezoresistive, capacitive, acoustic attenuation, and shear torque sensing elements, allowing direct contact with tailings material without wear interference. Each probe synchronously samples at a frequency of 10Hz as the equipment advances, forming a spatial-temporal four-dimensional data cube. (Note: The last sentence appears to be a fragment and doesn't translate directly.) A physical quantity in spatial coordinates ,time The original reading is ,in Corresponding to density Moisture content Particle size distribution index Shear strength This deployment method ensures that during a single excavation cycle, the equipment can cover a working surface at least 6 meters wide, generating no fewer than 12 sets of spatially synchronized data points, avoiding the omission of local characteristics due to sparse sampling. The probe array is rigidly connected to the equipment and moves synchronously with the excavation trajectory, ensuring... The coordinate system is always aligned with the device's own coordinate system, simplifying subsequent spatial mapping calculations.

[0097] 1.2, the information obtained in 1.1 Input a sliding window filter to eliminate transient disturbances and output a smooth sequence of physical characteristics. ,in The time window length is set to ensure data stability;

[0098] This step performs time-domain smoothing on the signal to remove high-frequency noise caused by momentary particle jamming, probe tremors, electromagnetic interference, etc. Although the physical characteristics of tailings exhibit abrupt changes in space, their temporal variation (equipment movement speed approximately 0.2-0.5 m / s) is relatively continuous. Therefore, a moving average filter is used for each spatial point. The time series data are processed independently. Let the time window length be... (corresponding to 0.5 seconds), then for each Calculate the arithmetic mean of its most recent 5 sample points, and use it as the smoothed output for the current time step. This formula is essentially a low-pass filter, preserving low-frequency trends while suppressing high-frequency jitter. For example, if the density reading at a certain point is within 5 consecutive frames... Then output This effectively eliminates random fluctuations on the order of ±0.1 g / cm³. Filtered data. This not only more closely approximates the real physical state, but also provides numerical stability for subsequent gradient calculations, avoiding misjudgments caused by noise amplification.

[0099] 1.3, based on the output of 1.2 Constructing local physical property gradient fields It is used to identify abrupt changes within the tailings pile and to help determine whether the equipment needs to be adjusted in advance.

[0100] Following the smoothed physical field described in section 1.2, this step calculates the spatial gradient to quantify the degree of change in physical properties on the working surface. The gradient is a vector field; its magnitude reflects the rate of change, and its direction indicates the steepest change. For each physical quantity... The gradient is approximated on a discrete grid using the central difference method:

[0101] ;

[0102] in Meters represent the probe spacing. After calculating the gradients for each of the four physical quantities, a combined gradient field is constructed:

[0103] ;

[0104] in These are weighting coefficients (e.g., density 0.4, moisture content 0.3, particle size distribution 0.2, shear strength 0.1), reflecting the contribution of each physical quantity to the equipment load. The unit is "characteristic unit / meter". The larger the value, the more uneven the tailings condition near that point. For example, if a certain area... This indicates the presence of soft-hard interlayers, abrupt changes in water content, or boundaries between coarse and fine particles. Equipment must pre-adjust parameters before entering this region; otherwise, it is prone to "getting stuck" or "being unable to penetrate." The construction of a gradient field enables the system to have "predictive" capabilities, allowing adjustments to be initiated 0.5-1.0 meters before the equipment reaches the high-variance zone, avoiding passive responses.

[0105] 1.4, the gradient field obtained in 1.3 With preset safety threshold In comparison, if If the area is marked as a "high-variability zone", it will trigger subsequent parameter pre-adjustment commands to prevent the equipment from entering an unstable operating condition.

[0106] This step performs a threshold decision, transforming the continuous physical field into discrete control events. A preset safety threshold is set. Set according to the maximum adaptability of the equipment, for example, take For each grid point on the working surface ,like If so, mark it as a "high-variable area" and record its spatial range. Simultaneously, calculate the geometric center of this region. With influence radius This forms a circular warning zone. When the front end of the device is close to... Less than At that time, the system automatically sends a "pre-adjustment request" to step two, along with... All The statistical characteristics (mean, variance, extreme values) are determined. This allows control commands to be initiated 1-2 seconds in advance, giving the equipment sufficient time to smoothly transition to new parameters and avoid sudden shocks. For example, if a high-water-content soft mud area is detected 1.2 meters ahead ( The system will reduce the digging speed and increase torque reserve in advance to prevent the bucket from getting stuck.

[0107] This step achieves "pixel-level" perception of physical state with a resolution of up to 0.5 meters, far exceeding traditional manual sampling or remote sensing estimation; by using moving average and gradient calculation, noisy data is transformed into trend features, improving the signal-to-noise ratio; a threshold trigger mechanism is used to discretize the continuous field, reducing the subsequent computational load while retaining key mutation information; high-risk areas are marked in advance, giving the system "pre-judgment-pre-adjustment" capabilities, fundamentally avoiding "head-on collisions" between devices in unknown mutation areas.

[0108] Furthermore, this process offers strong scalability in engineering implementation: the probe spacing can be scaled according to the equipment size (e.g., up to 1.0 meter for large equipment); the types of physical quantities can be added or removed (e.g., adding pH value, conductivity, etc.); and the filter window length can be adjusted. It can adaptively adjust to device speed (the faster the speed, the shorter the window length); gradient weights Customization is available based on equipment type (e.g., tracked heavy-duty equipment focuses more on shear strength, while wheeled equipment focuses more on moisture content).

[0109] Step 2: Establish the mapping relationship between equipment operating parameters and physical characteristics

[0110] This step will use the physical state data output from step one. The gradient marker "high change zone" is converted into control commands that the device can execute. The key task is to construct a dynamic response function that allows the equipment's digging depth, rotation speed, propulsion force, hydraulic pressure, and other action parameters to automatically match changes in the local physical characteristics of the tailings. This enables measures such as reducing the feed rate in areas of sudden density increases, increasing torque output in areas of excessive moisture content, and increasing vibration frequency in areas with coarse particles.

[0111] 2.1 Based on the location of the "high-variable area" marked in 1.4, extract the corresponding area. It is then normalized to the [0,1] interval to form the input vector. This facilitates the standardization of units;

[0112] The spatial markers for the "high-variable region" are the final output from step one. This step first extracts the physical property values ​​of all grid points within the region. Since the "high-variable region" is usually an irregular polygon, to simplify calculations, the system automatically includes all elements within its bounding rectangular region. The data is packaged and extracted to form raw data blocks. Then, for each physical quantity... Normalization is performed to eliminate the effects of dimensional differences and numerical scale. The normalization formula uses minimum-maximum scaling: ;

[0113] in This refers to the global extreme value of the physical quantity recorded in the equipment's historical operation database (e.g., density range of 1.2-2.4 g / cm³, moisture content of 5%-45%, etc.). After normalization, all physical quantities are compressed to the [0,1] interval, forming an input vector with a uniform scale. These correspond to the normalized density, moisture content, particle size distribution index, and shear strength, respectively. This approach not only prevents a single physical quantity from dominating subsequent calculations due to its excessively large value (e.g., density values ​​much greater than the moisture content percentage), but also provides a standardized input space for interpolation mapping. For example, if the average density in a certain "high-variability zone" is 2.1 g / cm³, then... If the average moisture content is 38%, then Normalized It is not only easy to calculate, but also has physical meaning: the closer the value is to 1, the more "extreme" the characteristic is, and the more the equipment needs to adjust the parameters to cope with it.

[0114] 2.2, based on 2.1 The parameter-feature lookup table in the historical job database is used to generate an initial parameter vector through interpolation. ,in For piecewise linear interpolation mapping;

[0115] Continuing from section 2.1, normalized input vector This step retrieves the nearest working condition record from the system's built-in historical job database to construct the initial control commands. The database stores past successful job cases in a four-dimensional hypercube structure, with each case containing normalized physical inputs. With corresponding optimal equipment parameters ,in To control dimensions (such as digging depth setpoint, main pump pressure, rotary motor speed, track propulsion speed, etc.). The system operates in four-dimensional space. Search distance The eight most recent historical points (forming the vertices of the hypercube) were calculated using a four-dimensional linear interpolation formula. :

[0116] ;

[0117] Weight reflect With the present The similarity is calculated based on proximity, with higher weights for closer points, and it is only included in the calculation when the distance in each dimension is less than 1 (i.e., only neighboring points are effective). For example, if the current... And a certain historical point Then its weight This step makes a significant contribution. It ensures that the system can output safe and usable initial commands even without real-time feedback, thus preventing "cold start" from spiraling out of control.

[0118] 2.3, the result of 2.2 Input the proportional-integral-derivative controller to calculate the dynamic correction. ,in It is an ideal physical state;

[0119] Continuing from the initial instructions generated in 2.2 This step introduces a feedback adjustment mechanism to dynamically optimize the instructions. The system presets an ideal physical state. The optimal value under the equipment's design operating conditions is usually taken, such as... (Medium density, low to medium moisture content, medium particle size, medium to high shear strength). Calculate the current state. Deviation vector from the ideal state .like A positive value indicates that the current tailings are "softer / drier than ideal," and the equipment can be operated more "aggressively"; a negative value indicates that the tailings are "harder / wetter than ideal," and a "conservative" approach is needed. Input the PID controller and calculate the correction:

[0120] ;

[0121] in This is a diagonal gain matrix, amplifying the proportional, integral, and differential terms respectively. Proportional term. Provides immediate response (e.g., immediate pressurization upon density surge); integral term Eliminate steady-state errors (such as prolonged low speed due to excessive humidity); differential terms Suppress overshoot (e.g., slow down in advance when physical properties change rapidly). For example, if (Currently denser and wetter than ideal), then A negative value indicates a decrease in speed and an increase in pressure. PID output Yes Refined corrections.

[0122] 2.4, referring to section 2.3 Overlay To form the final control command The data is then sent to the actuator to ensure that the equipment actions match the current tailings status.

[0123] Following the dynamic correction amount calculated in 2.3 This step completes instruction synthesis and output. The interpolation initial value is then used. With PID correction The final control vector is obtained by adding elements one by one. To ensure the physical feasibility of the instructions, the system... Boundary clamping: If a component exceeds the equipment's allowable range (e.g., pressure > 35 MPa or speed < 0.1 m / s), it is forcibly limited to the boundary value. Subsequently, It is broken down into independent control signals and sent to each actuator (hydraulic valve, frequency converter, servo motor, etc.) via CAN bus or industrial Ethernet. For example, if These correspond to the main pump pressure setting, slewing motor speed command, track travel speed, and boom torque limit, respectively. After the command is issued, the system simultaneously starts a timer. If no confirmation is received from the actuator within 0.5 seconds, the command is resent; if it fails three times consecutively, a fault alarm is triggered and the system switches to safe mode. The output of this step... Not only inheriting historical experience ( ), and further integrates real-time deviation adjustment ( This allows the equipment parameters to dynamically converge around the "current physical state," achieving closed-loop control of "sensing as response and response as matching."

[0124] This step unifies the scale of multiple physical quantities through normalization to avoid dimensional interference; four-dimensional interpolation uses historical data to provide reasonable initial values, reducing the adjustment burden; through this step, a dynamic, adjustable, and robust mapping bridge is established between physical characteristics and equipment parameters, laying a precise decision-making foundation for subsequent execution and feedback.

[0125] Step 3: Perform real-time adjustment of equipment parameters

[0126] This step will generate the control commands from step two. This translates into actual physical movements of the equipment. The goal is not only to "issue commands" but also to "ensure execution," meaning that the actuators, such as hydraulic cylinders, motors, and transmission mechanisms, precisely reproduce the required parameters like digging depth, rotation speed, and thrust. During execution, the deviation between the actual output and the command is continuously monitored, forming the first layer of closed-loop feedback. This ensures the equipment's movement trajectory. Strictly follow instructions Furthermore, it can immediately initiate local correction when deviations occur, preventing error accumulation from affecting overall control stability.

[0127] 3.1, receiving from 2.4 Decompose it into independent control signals These correspond to physical quantities such as digging depth, rotational torque, and travel speed, respectively.

[0128] The control vector output from step two This step first involves semantic parsing and physical mapping. The system has a built-in "instruction-execution unit" lookup table to clearly define each component. The corresponding specific implementing mechanism and physical meaning. For example:

[0129] Main hydraulic cylinder pressure setting value (unit: MPa), controls the downward pressure of the excavator arm;

[0130] : Rotary motor speed command (unit: rpm), controls the bucket rotation speed;

[0131] Track drive motor frequency (unit: Hz), controls the equipment's travel speed;

[0132] : Stick torque limit (unit: kN·m) to prevent overload damage;

[0133] Vibrating screen excitation frequency (unit: Hz) adapts to different particle size distributions;

[0134] : Pump power percentage (unit: %), used to adjust the drainage capacity of water-bearing tailings.

[0135] Each Before sending, dimensional restoration and range verification are required. For example, if... In step two, the normalized value of 0.78 needs to be multiplied by the pressure range of 35 MPa to restore it to 27.3 MPa. If it exceeds the range of [10, 35] MPa, it is clamped to the boundary. Subsequently, each signal is distributed to the corresponding controller through an industrial real-time communication protocol (such as EtherCAT or CANopen): the pressure command is sent to the proportional relief valve controller, the speed command is sent to the frequency converter, and the frequency command is sent to the PLC motion control module, etc. The signal transmission adopts dual-channel redundancy verification, and each frame of data is appended with a CRC16 checksum. If the receiving end fails the verification, it requests a retransmission to ensure that the command arrives without loss.

[0136] 3.2 Based on the signals in 3.1, drive the servo valve and frequency converter to adjust the hydraulic cylinder stroke and motor speed, thus controlling the equipment's movement trajectory. Approximation command requirements, i.e. ;

[0137] This step drives the actuator to produce a real physical response. Taking a hydraulic system as an example: the main controller receives... Then, the proportional relief valve is adjusted to change the valve core opening, stabilizing the pump outlet pressure at 27.3 MPa; simultaneously, the displacement sensor monitors the position of the hydraulic cylinder piston rod in real time. If the actual pressure If the valve opening is increased, the valve core opening will increase; conversely, it will decrease, forming a local pressure closed loop. For motor drive systems: the frequency converter receives... Then, the output frequency and voltage are adjusted to drive the rotary motor to rotate; the encoder provides feedback on the actual rotation speed. ,like If the error is less than ±5 rpm, the output frequency is increased until the error is less than ±5 rpm. Each execution unit has a built-in local PID controller with a control cycle of 1 ms, which is much faster than the upper-level instruction update frequency (100 ms), ensuring rapid response. Overall movement trajectory of the equipment. It is a collection of outputs from each execution unit, which the system collects every 10ms via a real-time bus. and with instructions Compare and calculate instantaneous errors For example, if the instruction requires a travel speed of 0.3 m / s, but the encoder feedback is only 0.28 m / s, then... This triggers local speed loop boost compensation.

[0138] 3.3 During the execution of 3.2, the actual output of the synchronous acquisition device is collected. , and calculate the deviation from the instruction in 3.1 Calculate the deviation , which is used to evaluate the execution accuracy;

[0139] Following the real-time feedback of the device actions in 3.2, this step constructs a performance evaluation system for the execution layer. The system reads the actual physical quantities from each sensor (pressure transmitter, encoder, torque meter, laser rangefinder, etc.) every 20 ms to form the actual output vector . To eliminate sensor noise, perform a 5-point moving average filter on each :

[0140] ;

[0141] Subsequently, calculate the deviation vector between the actual value after filtering and the original instruction in 3.1 :

[0142] ;

[0143] This deviation reflects the tracking accuracy of the actuator. The system defines an "execution tolerance band" , such as pressure ±1 MPa, rotational speed ±10 rpm, speed ±0.02 m / s, etc. If all are within the tolerance, it is determined that the execution is qualified; if any component exceeds the limit, it is marked as an "execution deviation". For example, if (the actual pressure is 1.8 MPa lower than the instruction), exceeding the ±1 MPa tolerance, the compensation process is triggered. At the same time, the system records the duration of the deviation . If , it is upgraded to an "execution failure", and manual intervention is required. This step realizes the quantitative evaluation of the execution quality through high-frequency sampling and deviation calculation, providing a data basis for subsequent compensation.

[0144] 3.4, if in 3.3 , then start the compensation loop, and feed back to the input end of the regulator in step two to form a closed-loop correction until the deviation converges.

[0145] This step starts a cross-level compensation mechanism. The system presets a global maximum allowable deviation norm (such as ). If , it is determined that the local execution loop cannot correct itself and upper-layer intervention is required. At this time, the system sends as an "execution error feedback" to the input end of the PID regulator in step two, and superimposes it on the original deviation : ;

[0146] Among them To compensate for the gain (typically 0.3-0.5), and to avoid overcorrection. For example, if the pressure setting is 1.8 MPa too low ( If this is the case, the PID controller will add an extra pressure command in the next calculation to compensate for insufficient execution. The compensation command is recalculated in step two. And generate new Then, after execution through steps 3.1-3.3, a closed loop of "execution deviation, instruction correction, re-execution, and re-evaluation" is formed. The system is set to have a maximum number of compensation attempts. If after three compensations If the limit is still exceeded, the current command will be frozen, switched to a safe mode (such as speed reduction by 50% or pressure lockout), and reported to the control center.

[0147] This process does not rely on ideal assumptions but directly addresses the nonlinearity, time delays, and disturbances in the execution phase, ensuring precise targeting through multi-layered closed loops. For example, when excavating high-density tailings, if the hydraulic system experiences a slow pressure response due to increased oil temperature ( The system will detect the deviation within 0.5 seconds and increase the command pressure by 2.5 MPa in step two to compensate for the execution loss and bring the actual pressure back to the target value.

[0148] Step 4: Monitoring and Identifying the Status of the Mining Process

[0149] This step involves continuously collecting operational status data such as vibration, load, displacement, and temperature during the equipment's operation, and identifying potential abnormal operating conditions through methods such as spectrum analysis, statistical fluctuation detection, and multi-parameter correlation.

[0150] 4.1 Install a triaxial accelerometer and a pressure sensor on the equipment body to collect the vibration spectrum during the compensation execution in 3.4. With load torque ;

[0151] Following the dynamic process of the equipment executing compensation commands in step three, this step deploys a high-precision state sensing network covering the key force-bearing and moving components of the equipment. Industrial-grade triaxial accelerometers (range ±50g, frequency response 0-2kHz) are installed at eight key locations, including the excavator arm hinge point, the slewing platform base, the track drive wheel axle, and the hydraulic cylinder piston rod, to collect equipment vibration signals in real time. Meanwhile, pressure and torque sensors are installed at the main pump outlet, boom cylinder, and slewing motor input shaft to collect hydraulic pressure data. With load torque (Measuring range 0-50kN·m, accuracy ±0.5%). All sensors are sampled at a uniform frequency of 2kHz to ensure the capture of high-frequency impacts and transient fluctuations. The raw vibration signal is then filtered by anti-aliasing (cutoff frequency 800Hz) before being fed into a Fast Fourier Transform (FFT) module to calculate the spectrum.

[0152] ;

[0153] in The spectral resolution is 0.5Hz, covering the 0-1000Hz frequency band. Load torque. The original time-domain waveform is recorded directly at a frequency of 100Hz. All data is then accurately timestamped. and the instructions in step three and actual output Synchronous alignment creates a four-dimensional data stream encompassing "command, output, vibration, and load." For example, when the equipment is excavating high-density tailings at 28 MPa pressure after compensation, the system synchronously records whether the peak value of the X-axis vibration spectrum of the excavator arm appears at the natural frequency of 85 Hz, and whether the load torque suddenly increases to 45 kN·m. This step, through high-density, highly synchronized sensing, provides a multi-dimensional, high-fidelity data foundation for subsequent anomaly identification.

[0154] 4.2, in relation to 4.1 Perform a Fast Fourier Transform to extract the percentage of the main frequency energy. ,like If so, it is determined to be a stable operating condition;

[0155] This step uses energy distribution analysis to determine whether the equipment's vibration state is "orderly." During normal operation, the vibration energy should be concentrated near a few natural mechanical frequencies (such as rotational frequency 5Hz, hydraulic pulsation frequency 20Hz, structural resonance frequency 85Hz, etc.), forming a clear main peak. If the equipment encounters particle obstruction, foundation loosening, or resonance, the energy will diffuse across a wide frequency band, resulting in "spectral ambiguity." The system defines the main frequency band. The three main operating frequencies listed in the equipment design manual are within a ±5Hz range (e.g., [3,7]Hz, [18,22]Hz, [80,90]Hz). Calculate the energy percentage within the main frequency band:

[0156] ;

[0157] in .like This indicates that over 70% of the vibration energy is concentrated in the expected frequency band, and the equipment is operating smoothly; if This leads to energy dispersion and a risk of abnormal vibration. For example, when the pressure on the excavator arm rises to 29 MPa after compensation, if... A sharp drop from 0.75 to 0.42 indicates that the device may be experiencing "flutter," requiring immediate intervention. The system performs a calculation every 0.5 seconds. And record its rate of change. If the value drops by more than 0.3 within 1 second, it is marked as "vibration degradation".

[0158] 4.3, combining the results of 4.2 with those in 4.1 Calculate the load fluctuation coefficient ,like If so, it will be marked as "load abnormality";

[0159] This step introduces time-domain analysis of the load torque to assess whether the equipment is subjected to "stable" stress. Even if the vibration spectrum is ordered, drastic fluctuations in the load torque can still lead to fatigue damage to transmission components. The system analyzes the load torque over the most recent 2 seconds (200 sampling points). Calculate statistical characteristics:

[0160] ;

[0161] Then calculate the load fluctuation coefficient: ;

[0162] This coefficient reflects the relative degree of load fluctuation. If This indicates a stable load; if The fluctuations are slight and acceptable; if If the load fluctuates drastically, it is marked as "abnormal load". For example, when the equipment attempts to advance at a speed of 0.3 m / s after compensation, if the average load is 30 kN·m, but the standard deviation reaches 12 kN·m ( This indicates an "intermittent hardening" issue, requiring either a reduction in speed or an increase in pressure. The system also performs a check. Compared with 4.2 Relationship: If and If so, it is determined to be a "strong interference condition"; if but This could be a "periodic shock" (such as regular large particles).

[0163] 4.4, align the "Abnormal Load" flag in 4.3 with the deviation in 3.4. If both exceed the limit, a "condition reconfiguration" request is sent to step five to initiate a higher-level adjustment.

[0164] Following the "load anomaly" flag in section 4.3 and the execution deviation in step 3.4... This step performs multi-parameter fusion decision-making to determine whether global parameter reconstruction needs to be initiated. The system defines a "composite anomaly" condition: when "load anomaly" ( ) and "Execution deviation exceeds limit" If both conditions are met simultaneously and the duration exceeds 0.8 seconds, it is determined that the current operating condition has exceeded the local adjustment capabilities of steps two and three, requiring intervention in step five. For example, if the equipment is operating in a high-density area and the pressure execution deviation reaches -2.2MPa (over-limit), and the load fluctuation coefficient rises to 0.45 (abnormal), it indicates that the tailings characteristics are more extreme than predicted, and the historical mapping relationship is invalid. At this time, the system generates an "Operating Condition Reconfiguration Request," which includes the following data packets:

[0165] Current physical input (From step 2.1);

[0166] Current instruction Compared with actual output ;

[0167] Vibration spectrum With load waveform Fragment;

[0168] Deviation vector With volatility coefficient ;

[0169] The data packet is sent to step five via a priority queue, triggering the parameter adaptive update process. Simultaneously, this step sends a "suspend compensation" command to step three to prevent blindly increasing pressure under unclear operating conditions, which could damage the equipment. The system is set to a 60-second "reconstruction cooldown period," during which no new requests are sent to prevent frequent triggering.

[0170] This step achieves "local handling of minor anomalies and global reconstruction of major anomalies." For example, when the equipment is operating in a water-bearing soft mud area, if the load fluctuation coefficient... (Abnormal) but execution deviation The system only logs and does not trigger a refactoring; however, if the pressure execution deviation reaches -1.8MPa (exceeding the limit) at the same time, a refactoring will be initiated immediately to update the pressure mapping relationship.

[0171] Step 5: Closed-loop feedback adjustment and adaptive parameter update

[0172] This step is based on the physical characteristics under the current operating conditions. Actual response of the equipment The deviation is used to deduce a parameter mapping function that better fits the current tailings characteristics, and then update the interpolation database and PID gain matrix in step two to avoid the recurrence of similar anomalies.

[0173] 5.1 Receive the "Operating Condition Reconfiguration" request from 4.4 and freeze the current control parameters. And initiate the parameter recalibration process;

[0174] This step begins with freezing the system state and implementing security isolation. Upon receiving the request, the control center immediately sends a "parameter lock command" to all relevant execution units, locking the current state. Key components such as pressure, rotational speed, and velocity are fixed at their current values. Any automatic adjustment or compensation operations in steps two and three are prohibited to prevent equipment malfunction due to command drift during parameter recalibration. Simultaneously, the system initiates "data snapshot capture," capturing the data contained in the request packet. Once all data is stored in a temporary cache, along with metadata such as timestamps, device IDs, and work surface coordinates, a complete "abnormal operating condition file" is formed. Subsequently, the system activates "recalibration mode," shutting down the regular control loop and prioritizing computational resources for this step. To ensure safety, the equipment automatically reduces its speed to 50% of the original command, and the hydraulic pressure is limited to 80% of its maximum value, entering a "conservative operation" state until the new parameters take effect. For example, if the original command was 28MPa / 800rpm / 0.3m / s, this value will be maintained after freezing, but the system internally stops PID adjustment, waiting for the new mapping relationship to be generated.

[0175] 5.2, based on 5.1 before freezing and Construct the local response matrix It characterizes the sensitivity of the device's output to changes in physical properties;

[0176] This step constructs a "physical-equipment" response model to quantify the impact of minute changes in tailings characteristics on equipment output. The system selects the steady-state data segment within the last 5 seconds before freezing (excluding transient shocks) and extracts... and 10 sets of synchronous sampling points For each physical quantity and output of each device Calculate the approximate value of the partial derivative:

[0177] ;

[0178] in This is the mean. The formula is essentially the slope of a linear regression, reflecting... Unit change caused The average change. For example, if density An increase of 0.1 (normalized unit) leads to increased actual pressure. A decrease of 2.5 MPa, then This indicates that the device is highly sensitive to density changes. (The last part, "All," appears to be a fragment and doesn't translate directly.) Combined into 3D response matrix This matrix reveals "which physical quantity most drastically affects which equipment parameter" under the current operating conditions, providing directional guidance for subsequent parameter adjustments. For example, if... (Density-Pressure) is much greater than (Moisture content - pressure) means that subsequent corrections should prioritize adjusting the pressure mapping rather than the velocity mapping.

[0179] 5.3, using 5.2 The parameter correction coefficients are solved by the least squares method. ,in This is due to historical accumulated bias;

[0180] Building upon the response matrix constructed in section 5.2 This step calculates the optimal parameter correction to make the device output approximate the ideal value. The objective is to eliminate the current execution deviation. Meanwhile, the historical deviation accumulated over the past 30 seconds is also taken into account. (Weighted average, with higher weighting for recent deviations). Define the overall correction target. Assuming parameter correction amount With physical input The product of these can compensate for the output deviation, that is: ;

[0181] Solving for the optimal solution using the least squares method : ;

[0182] This formula makes Minimization means that the correction amount best approximates the target deviation under the response constraints. For example, if (Pressure is 2 MPa too low, speed is 50 rpm too high, and velocity is 0.05 m / s too low), and The density shows the greatest impact on pressure. This will primarily involve corrections to the density mapping. The calculated... It is a 4-dimensional vector (corresponding to 4 physical quantities), and its components are... Indicates "when physical quantity When changes occur, how much additional control input should be added to the original mapping? For example, This means that for every 0.1 unit increase in density, the pressure command should be increased by an additional 0.3 MPa. This step uses least squares optimization to ensure that the correction not only addresses the current problem but also takes into account historical trends, avoiding overfitting to transient noise.

[0183] 5.4, ​​the result of 5.3 Apply the mapping relationship from step two to update the interpolation function. The segmented nodes and slopes make subsequent control commands more closely match the current tailings characteristics.

[0184] This step performs a dynamic update of the mapping relationship. The system locates the interpolation database used in step 2.2 and finds the relationship with the current... The 8 nearest historical sites For each neighboring point, assign its corresponding device parameters. Make corrections: ;

[0185] in This indicates element-wise multiplication. For example, if a certain historical point... Original parameters ,current Then the correction amount New parameters After the update, this historical point will output higher pressure and lower speed in subsequent interpolations, making it more suitable for high-density operating conditions. Simultaneously, the system adjusts the PID gain matrix in step 2.3: if... (If density requires increased pressure), then increase the pressure ring. ;like (If the moisture content needs to be reduced, then increase the speed loop) After the update is complete, the system unfreezes the parameters in step 5.1, loads the new mapping relationship into step two, and sends a "restart control" command to step three. The device smoothly transitions to the new parameters within 0.5 seconds and continues operation.

[0186] This step establishes the self-learning capability of the control system. For example, when the system first encounters ultra-high density tailings ( At this stage, insufficient pressure may occur due to missing historical data; however, after this update, neighboring points... The parameters are adjusted upwards, and the system will automatically output higher pressure the next time a similar working condition is encountered, without the need for reconfiguration. Through this step, the closed-loop system is upgraded from static control to dynamic evolution, providing continuously optimized control assurance for complex tailings mining.

[0187] Step Six: Multi-device Collaboration and Dynamic Planning of Work Paths

[0188] This step undertakes the core task of coordinating multiple mining devices to work together in the spatial dimension, avoiding conflicts, optimizing paths, and balancing loads, based on the adaptive update of single-machine parameters (step five).

[0189] 6.1, Based on the updated mapping relationship in 5.4, generate a locally optimal set of operating parameters for each device. ,in Number the equipment;

[0190] This step first involves all areas within the work area. Tai mining equipment ( Each device independently generates its current optimal set of operating parameters. Based on the local physical characteristics collected by its front-end sensor array (From step one), call the updated interpolation function. Generate initial instructions Then, input its dedicated PID controller (the gain matrix has been updated in step five), combined with the ideal state. Calculate the dynamic correction amount Final synthesis equipment The current optimal parameter vector Because the tailings characteristics differ at the location of each piece of equipment (e.g., equipment 1 requires high pressure in a high-density area, while equipment 2 requires low speed in a soft mud area), its... They are all different. The system will Expanded into a parameter set ,in The second represents the sequence of parameters predicted based on the current physics field over the next 10 seconds (assuming the physics field remains unchanged over 10 seconds). For example, device 1's... It may contain sequences such as [28MPa, 780rpm, 0.28m / s] and [29MPa, 770rpm, 0.26m / s], reflecting that it is about to enter a harder region and needs to gradually increase the pressure and decrease the speed. All Real-time location of the device Orientation Angle Normal vector of the working surface They are uploaded together to the central dispatch server to form a device status cloud.

[0191] 6.2, based on the various provisions in 6.1 Calculate the inter-device interference index ,in For a safe distance, if If so, it is determined that there is a risk of interference;

[0192] Following the "Device Status Cloud" uploaded in section 6.1, this step performs space conflict detection. The system checks each pair of devices... Predict its trajectory over the next 10 seconds. Based on velocity sequence in and orientation angle computing devices Displacement vector:

[0193] ;

[0194] After discretization, the position sequence is obtained. Similarly, computing devices of Define the device. and At any moment Spatial distance: ;

[0195] System preset safety distance The value is set according to equipment size and operational safety regulations (e.g., 5.0 meters for large excavators, 3.5 meters for medium-sized ones). The interference index is calculated as follows: ;

[0196] If in any Within seconds, the device is determined. and There is a risk of "operational interference." For example, if the predicted distance between equipment 1 and equipment 2 is only 2.8 meters at the 6th second, ... Rice, then This triggers an interference alarm. The system records all interference pairs. and its occurrence time window And calculate the severity of interference. A larger value indicates more severe interference.

[0197] 6.3, Regarding the equipment that interferes with 6.2 A priority scheduling method is adopted, based on equipment operating efficiency. Reassign job areas and generate conflict-free paths. This step involves conflict resolution and area reallocation. The system first calculates the operating efficiency of each device. :

[0198] ;

[0199] in Calculated based on digging depth, travel speed, and bucket volume; Efficiency is derived from the integration of engine load, hydraulic pressure, and motor current. The higher the priority, the more likely it is to retain its original path. For interference pairs... ,like Then the equipment Maintain the original plan, equipment It needs to be avoided. Avoidance strategies include:

[0200] Path offset: Move the device Translation of the direction of travel Meters, generate new paths ;

[0201] Timing shifting: If the path cannot be shifted (e.g., near a slope), then delay the device. Startup time Second;

[0202] Area switching: If the device Higher efficiency is related to equipment Exchange work areas and recalculate for both parties. .

[0203] New Path Described using B-spline curves to ensure continuous curvature without sharp turns: ;

[0204] in As control points, These are p-order basis functions. The system verifies all functions under the new path. And equipment Loss of work efficiency For example, if device 2 needs to avoid device 1, the system shifts its path 4 meters to the right and recalculates its path. (Due to the different physical characteristics of the new region), and verified that the efficiency dropped from 8.2 to 7.9 (a loss of 3.7%), which is acceptable.

[0205] 6.4, referring to 6.3 Compared with 6.1 The data are merged to form a spatial-parameter joint instruction package, which is then sent to the controllers of each device to achieve regional adaptive collaborative data acquisition.

[0206] Continuing from the conflict-free path generated in section 6.3 With the parameter set in 6.1 This step involves constructing and reliably issuing a spatiotemporal parameter-joint command. The system will... Discretized into 100 waypoints Each waypoint is associated with a corresponding subset of parameters. (e.g., waypoint 50 corresponds to the parameters at the 5th second). This forms a structured instruction packet:

[0207] ;

[0208] The packet is sent to the device via a high-priority wireless link (such as a 5G private network or industrial mesh). The controller. After parsing, the controller initiates a dual-loop control system of "path tracking + parameter synchronization": the outer loop is a path tracking PID controller, which controls the differential speed of the track to keep the equipment moving along the path. Movement; the inner loop executes PID with parameters to ensure... track Every 0.5 seconds, the system checks whether the actual position of the device matches the command parameters. If the deviation exceeds the limit, the local command segment is resent. Simultaneously, the central server dynamically monitors the status of all devices. If a new device is added or the physical field changes abruptly, this step is triggered to replan within 5 seconds. For example, after device 3 completes its avoidance path, the system immediately checks whether it has entered a new "high-variability zone." If so, steps one through five are linked to update its status. This enables dual adaptive behavior of "path-parameters".

[0209] Step 7: Data-driven continuous model optimization and system self-evolution mechanism

[0210] This step, building upon the single-machine adaptation, multi-machine collaboration, and anomaly reconfiguration achieved in the previous six steps, undertakes the long-term, global, and data-driven continuous optimization of the core mapping model, decision-making algorithm, and scheduling strategy of the entire system, ensuring that the control system continuously improves its performance ceiling over time.

[0211] 7.1 Establish a full-cycle data lake, integrating the outputs of steps one through six. Structured and unstructured data are stored and indexed by timestamps and spatial coordinates; this step builds a unified, traceable, and highly available "industrial data lake." The system deploys a distributed time-series database (such as an InfluxDB cluster) and object storage (such as MinIO), storing data in layers according to data type.

[0212] Physical layer data: from step one gradient field High-variable area marking It is stored as a three-dimensional tensor with a spatial grid (0.5m×0.5m) and a timestamp (100ms) as keys;

[0213] Control layer data: from steps two and three Stored by device ID and millisecond-level timestamp, associated with the corresponding physical location;

[0214] Response layer data: Local response matrix from step five Correction coefficient Archive them according to "abnormal event ID" to form "problem-countermeasure" knowledge pairs;

[0215] Collaboration layer data: Path from step six Interference index ,efficiency Aggregated by "job shift + region ID" to support group behavior analysis;

[0216] Logs and images: Equipment alarm logs, vibration spectrum graphs, load waveform graphs, and on-site camera snapshots (unstructured), stored according to event triggers for root cause analysis.

[0217] All data is accompanied by metadata: equipment model, operator ID, weather conditions (temperature, humidity, rainfall), tailings source (processing plant number, mineral type), maintenance records (last oil change, filter replacement), etc. The data lake uses columnar storage and compression algorithms, supporting petabyte-level capacity and millisecond-level queries. For example, an SQL query can be used to quickly locate inefficient operating conditions by querying "all equipment records operating in areas with a moisture content > 35% and an efficiency η < 7.0 in X year X month".

[0218] 7.2. Principal Component Analysis (PCA) and clustering algorithms are applied to the data in 7.1 to identify high-frequency anomaly patterns and efficient operation patterns, forming a pattern library. This step performs unsupervised learning to automatically discover hidden patterns. First, for high-dimensional data (such as...) Principal Component Analysis (PCA) is performed on the vectors (concatenated into a 20-dimensional vector) to reduce the dimensionality to 3-5 principal components, retaining more than 95% of the variance. For example, the first principal component might represent the "density-pressure-load" coupling axis, and the second principal component might represent the "moisture content-velocity-vibration" coupling axis. Subsequently, the DBSCAN clustering algorithm is applied to the dimensionality-reduced space to automatically identify dense regions.

[0219] Abnormal pattern clusters :correspond The data points are clustered into cluster centers that represent typical abnormal operating conditions (such as "high density + low response" or "high water content + high fluctuation").

[0220] High-efficiency mode cluster :correspond The data points and cluster centers are the golden combination of operation parameters.

[0221] For each cluster, extract its core feature vector. A pattern library is formed by combining the statistical boundary (e.g., mean ± 2 standard deviations):

[0222] ;

[0223] in Let covariance matrix be the variance matrix. For example, anomaly patterns. This may represent a typical failure combination of "high density, low moisture content, medium particle size, high shear strength, pressure 25 MPa, rotation speed 600 rpm, velocity 0.2 m / s, pressure deviation -3.0 MPa, and load 80 kN·m". The system automatically generates a pattern library update report every morning at midnight and pushes it to the control center. 7.3, based on 7.2. A reinforcement learning framework is used to train the parameter mapping strategy. To minimize long-term costs To replace the original interpolation function ;

[0224] This step constructs a reinforcement learning (RL) agent to learn the globally optimal control policy. A Markov Decision Process (MDP) is defined:

[0225] state space Normalized physical input Equipment status (oil temperature, cumulative working hours), environmental status (rainfall, slope);

[0226] Action space Equipment parameters Feasible adjustment ranges (e.g., pressure ±2MPa, speed ±0.05m / s);

[0227] reward function The wear index is calculated from the vibration intensity and load fluctuation.

[0228] Discount factor Long-term optimization is encouraged.

[0229] The policy network is trained using the Deep Deterministic Policy Gradient (DDPG) algorithm. The experience replay buffer samples from the data lake: positive samples come from... Negative samples come from Network input Optimal output The training objective is to minimize the long-run cost:

[0230] ;

[0231] After training converges, the optimal policy is obtained. Its performance surpasses that of the original interpolation function. For example, facing (High density, high shear strength) The output is [28MPa, 750rpm, 0.25m / s], while The output is [30MPa, 700rpm, 0.22m / s]. Although the speed is slightly reduced, the actual efficiency η is increased by 15% due to more accurate pressure matching. The system automatically triggers RL model retraining every quarter or after 1000 hours of cumulative work, and conducts A / B testing on the old and new strategies. The winner is deployed to step two.

[0232] 7.4 Establish a model version management and canary release mechanism, and implement the optimizations in 7.3. The scheduling algorithm is incrementally deployed to the production system, while performance metrics are monitored to ensure system stability and continuity.

[0233] The new model trained in section 7.3 This step involves a secure and controlled production deployment. The system uses a model version repository (such as MLflow) to manage all historical models, recording their training data, hyperparameters, and validation metrics. The deployment process is as follows:

[0234] 1. Shadow Mode Testing: New Model Compared with the old model Running in parallel, receiving the same input However, only the old model outputs actual control commands; the system records the predictions of the new model. With actual implementation The deviation is used to calculate the "prediction accuracy".

[0235] 2. Gray-scale release: If the new model's prediction accuracy is >95% and there are no abnormal alarms in shadow mode, select one non-critical device (such as an edge area operation machine) to switch to the new model and continuously monitor it for 24 hours;

[0236] 3. Indicator Comparison: Calculate the efficiency η, energy consumption E, failure rate F, and other KPIs of the grayscale equipment and the control group (using the old model). If the new model improves η by more than 5% and F does not increase, then expand to 50% of the equipment.

[0237] 4. Full Switchover: If all KPIs are met after one week, fully replace the old model and archive the old version. At the same time, update the scheduling algorithm in step six (e.g., incorporate the efficiency η calculation formula into the RL reward function).

[0238] The system is configured with a "circuit breaker mechanism": if the new model causes any device to... If the "compound anomaly" in step four is triggered more than 3 times per hour, the system will automatically roll back to the old version. All operation logs are stored on the blockchain for auditability. For example, after upgrading the pressure mapping from linear interpolation to the RL strategy in version X of year X, the average efficiency of the entire mining area increased by 8.7%, and fuel consumption decreased by 6.2%.

[0239] This step builds the system's self-evolution capability, accumulates data throughout the entire lifecycle to form industry memory, automatically discovers knowledge patterns through unsupervised learning, implements cost-optimized strategies through reinforcement learning, and ensures evolutionary security through canary releases.

[0240] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A tailings mining control method, characterized in that, include: Collect physical property data of tailings and mark high-variance areas, then extract the corresponding data of the high-variance areas to form an input vector; After normalizing the input vector, the initial parameter vector is obtained by interpolation, and then the control command is generated by the proportional-integral-derivative controller. The control command is broken down into independent signals to drive the actuator. The actual output is collected synchronously and compared with the command to calculate the deviation. If the deviation exceeds the limit, it is fed back to the regulator input for correction. At the same time, the vibration spectrum and load torque are collected. The proportion of the main frequency energy is calculated based on the vibration spectrum, and the load fluctuation coefficient is calculated based on the load torque. If both of these exceed the limits and the deviation exceeds the limit, the parameters are frozen and the working condition is reconfigured. In the condition reconfiguration, a response matrix is ​​constructed based on the physical characteristics and steady-state output data of the equipment collected during the freeze period. The parameter correction is solved by the least squares method, and the node values ​​and slope of the interpolation function are updated. Using the updated interpolation function mapping relationship, a local optimal parameter set is generated for multiple devices. The interference index is calculated based on the predicted trajectory of the devices. If interference exists, the work area is redistributed according to the work efficiency and a conflict-free path is generated. After the path and parameters are distributed, data from the entire lifecycle are collected to build a data lake. Through principal component analysis and clustering, abnormal and efficient patterns are obtained, a parameter mapping strategy is trained, and the update is completed by replacing the interpolation function with a canary release.

2. The tailings mining control method according to claim 1, characterized in that: The process of collecting tailings physical property data and marking high-variance areas includes: A sensor array is deployed in front of the work area to collect data on density, moisture content, particle size distribution index, and shear strength to obtain physical property data. Spatial gradient calculation is performed on physical property data, and regions with gradient values ​​greater than a threshold are marked as high-variable regions; The coordinates and physical property data of the high-variable area are packaged and sent to the control center.

3. The tailings mining control method according to claim 1, characterized in that: The control commands generated by the proportional-integral-derivative controller include: Receive the coordinates of the high-variable area, extract the physical characteristic data of the corresponding location to form a four-dimensional input vector, perform minimum and maximum scaling on the input vector, and map it to the [0,1] interval; Search for neighboring points less than 1 in the historical task database, and obtain the initial parameter vector by weighted interpolation based on the inverse of the distance; The initial parameter vector and the deviation from the ideal state are input into the proportional-integral-derivative controller. The output correction is superimposed and boundary clamping is performed to form the final control command.

4. The tailings mining control method according to claim 1, characterized in that: The control command is broken down into independent signal-driven actuators, and the actual output is synchronously collected and compared with the command to calculate the deviation, including: The control command vector is decomposed into pressure setpoint, speed command, and frequency command according to physical quantities, and distributed to the proportional relief valve controller, frequency converter, and motion control module through the industrial real-time communication protocol. Every 10 milliseconds, the hydraulic cylinder position, motor speed, and travel speed are collected to form the actual output vector; The deviation between the actual output vector and the original instruction is calculated after performing a 5-point moving average filter on the actual output vector. If the deviation norm exceeds the limit, the deviation is multiplied by the compensation gain and fed back to the input of the proportional-integral-derivative controller, with a maximum of three compensations.

5. The tailings mining control method according to claim 1, characterized in that: If the deviation exceeds the limit, feedback is sent to the regulator input for correction, including: The three-axis accelerometers installed at the excavator arm hinge point, slewing platform, and drive wheel axle were used to collect vibration signals at a frequency of 2000Hz. After performing anti-aliasing filtering on the collected vibration signal, the fast Fourier transform spectrum is calculated; The proportion of energy in the main frequency band to the total energy is extracted based on the Fast Fourier Transform spectrum. If the proportion is less than 0.5, it is marked as an abnormal vibration. The load torque is collected synchronously by the torque sensor. The ratio of the standard deviation to the mean is calculated based on the most recent 200 data points. If the ratio is greater than 0.3, it is marked as an abnormal load. If the vibration anomaly marker and the deviation both exceed the limit, and the condition lasts for more than 0.8 seconds, a condition reconfiguration request will be generated and sent to the control center.

6. The tailings mining control method according to claim 1, characterized in that: The step of solving for the parameter correction using the least squares method and updating the node values ​​and slope of the interpolation function includes: Upon receiving the working condition reconfiguration request, a parameter lock command is sent to the actuator and its speed is reduced to 50% until the operation is stable. During parameter locking, capture 10 sets of physical characteristic data and corresponding steady-state data of device output within the last 5 seconds; Based on the 10 sets of physical characteristic data and the corresponding steady-state data of device output, the linear regression slope between each physical characteristic dimension and the device output component is calculated, and a response matrix composed of the linear regression slope is formed. The current deviation and the historical cumulative deviation are weighted and summed as the objective function. Combined with the response matrix composed of the linear regression slope, the parameter correction amount obtained by the least squares method is obtained by inputting the least squares method. Locate the nearest interpolation nodes in the historical operation database, adjust the corresponding parameter values ​​of the nearest interpolation nodes based on the parameter correction amount obtained by the least squares method, and update the relevant parameters of the proportional-integral-derivative controller simultaneously.

7. The tailings mining control method according to claim 1, characterized in that: If interference exists, the work area will be reallocated based on work efficiency and conflict-free paths will be generated, including: For each device, the updated mapping relationship is invoked to generate a parameter sequence for the next 10 seconds; Based on the parameter sequence and current position, predict the motion trajectory and calculate the minimum distance between any two devices; The ratio of the minimum distance to the safe distance is used as the interference index; if it is less than 1, interference is considered to exist. The ratio of the amount of data collected per unit time to the energy consumption of each device is calculated as the operational efficiency. For interfering devices, the original path of the one with higher efficiency is retained, while the path of the one with lower efficiency is shifted, the timing is staggered, or the area is exchanged.

8. The tailings mining control method according to claim 1, characterized in that: After the paths and parameters are distributed, full-cycle data is collected to build a data lake. Principal component analysis and clustering are used to identify anomalies and high-efficiency patterns, including: Physical characteristic data, control command data, actual output data, response matrix data, path trajectory data, and operation efficiency data are stored in a distributed database using time and spatial indexes to form a full-cycle data lake; Principal component analysis is performed on the high-dimensional data stored in the distributed database to reduce dimensionality, retaining principal components with a cumulative variance contribution rate of 95%, and then clustering algorithms are applied to extract cluster centers and cluster boundaries to form an abnormal pattern library and an efficient pattern library. Using physical characteristic data as state input, equipment parameter data as action output, and the weighted sum of energy consumption data, wear data, and delay data as negative reward, a deep deterministic policy gradient network is trained based on an abnormal pattern library and an efficient pattern library to obtain a parameter mapping policy. The parameter mapping strategy first performs shadow running, then performs grayscale testing on a single device, and finally performs a full switch. If the preset circuit breaker condition is triggered at any stage, it will automatically roll back to the original interpolation function.

9. The tailings mining control method according to claim 1, characterized in that: The grayscale release replaces the interpolation function to complete the update, including: The parameter mapping strategy receives the same physical characteristic input data in parallel with the original interpolation function, and only the original interpolation function outputs actual control commands to drive the actuator; During parallel operation, the deviation between the predicted output data and the actual execution output data of the parameter mapping strategy is recorded. When the prediction accuracy is higher than 95%, the gray release stage begins. During the canary release phase, select an edge computing device to switch to control commands generated by parameter mapping strategy and continuously monitor key performance indicators for 24 hours. When 24-hour monitoring results show that the operation efficiency has improved by more than 5% and no new abnormal mode has been triggered, the parameter mapping strategy will be extended to 50% of the devices.

10. A tailings recovery control system for implementing the method as described in any one of claims 1-9, characterized in that: Includes sensor array, control center, and actuators; The sensor array is used to collect data on the physical characteristics of tailings and mark high-variable areas. The data corresponding to the high-variable areas are extracted to form an input vector and sent to the control center. The control center is used to normalize the input vector and interpolate it to obtain the initial parameter vector, which is then used by the proportional-integral-derivative controller to generate control commands, which are then broken down into independent signals to drive the actuators. After the actuator moves, the control center collects the actual output and compares it with the command to calculate the deviation. If the deviation exceeds the limit, feedback is provided for correction. At the same time, the vibration spectrum and load torque are collected. The control center calculates the main frequency energy ratio and load fluctuation coefficient. If both exceed the limit, the parameters are frozen and the operating condition is reconfigured. The control center constructs the response matrix, solves for the correction using the least squares method, and updates the interpolation nodes and slope. The control center generates parameter sets for multiple devices, calculates the interference index, and if interference occurs, it redistributes the region according to efficiency and generates conflict-free paths. After the control center issues the path and parameters, the data server collects full-cycle data to build a data lake, obtains patterns through principal component analysis and clustering, trains strategies, and completes the update through canary release.

Images