A multi-roller coordinated belt deviation correction control method and system

Abstract

The present invention provides a multi-roller coordinated belt deviation control method and system, comprising: comparing the current belt edge with a set belt edge value to obtain the offset distance #imgabs1# and offset speed #imgabs2# of the pixel-related point #imgabs0# where each deviation-correcting roller intersects the standard edge line; combining this with the deviation-correcting roller angle #imgabs3# to construct a state vector #imgabs4#; #imgabs5# outputting an action through a policy network of a reinforcement learning model; combining the action and the state vector into a state-action pair, which is input into an action reconstruction model to reconstruct the action, and outputting a safe action for subsequent motor control; the state-action pair is also input into the action value network and action cost network of the reinforcement learning model to evaluate the action output by the policy network. The present invention can perform belt deviation correction control.

Description

Technical Field

[0001] The present invention relates to the technical field of belt deviation correction control, and in particular to a multi-roller coordinated belt deviation correction control method and system. Background Art

[0002] Belt conveyor systems are critical material transportation equipment in mines, ports, logistics, and manufacturing industries. However, belt deviation is common during operation due to complex structures, volatile environments, and long-term operation. This deviation can cause equipment wear, material leakage, energy waste, and associated downtime, seriously affecting production efficiency and equipment life. Therefore, precise correction control is crucial to ensuring production safety and economic efficiency. Currently, the treatment of belt deviation in industrial sites mainly relies on the installation of correction devices, such as correction rollers. These devices are controlled manually or automatically to adjust the lateral friction of the belt or the roller angle, gradually guiding the belt back to its normal trajectory. This method has become a common solution for engineering applications due to its high execution efficiency and low modification costs.

[0003] Since belt deviation is a gradual process, its local deviation will gradually spread through the overall transmission effect of the belt, eventually causing global operational abnormalities. Although the current technical practice of using a single correction roller for a specific area is relatively common, it has obvious limitations - especially in long-distance conveying scenarios. The effective range of a single correction roller usually does not exceed 20 meters, and when the conveying distance exceeds 50 meters, multiple correction units must be deployed. However, the parallel deployment of multiple rollers will lead to new technical bottlenecks. The coordination issues such as control timing matching and force coupling between the correction rollers require the establishment of a real-time communication mechanism. Otherwise, the correction torque hedging phenomenon may occur. This uncoordinated action will not only fail to correct the deviation, but will increase the risk of system oscillation, making the complexity of the correction control exponentially increase. This technical dilemma not only reveals the essential difference between single-point control and global optimization, but also confirms the inevitable trend of the correction system to transform to distributed intelligent control.

[0004] This transformation demand still presents many problems in traditional control architectures. For example, although conventional PID control algorithms can achieve basic deviation correction functions, their fixed parameter settings make it difficult to adapt to dynamic changes in belt tension; while mechanical linkage devices based on preset thresholds can achieve simple coordination, they cannot analyze the nonlinear coupling relationship between multiple rollers. More importantly, traditional systems lack the ability to predict the propagation path of deviation, and often miss the optimal control window when a significant deviation is detected. These inherent defects lead to the dual dilemma of system response lag and inaccurate deviation correction accuracy in existing solutions when dealing with long-distance transportation and variable load conditions - it is impossible to effectively intercept the deviation during the propagation stage, and it is difficult to maintain trajectory stability during dynamic changes. This structural lack of control capability is precisely the fundamental reason why most industrial scenarios are difficult to get rid of manual real-time monitoring and compensation operations.

[0005] On the other hand, adopting intelligent control strategies faces challenges such as difficulty in acquiring training data for multi-roller collaboration and safety issues during model deployment. First, building efficient intelligent control methods typically relies on large amounts of high-quality data for training and optimization. However, in real industrial environments, complex operating conditions and a diverse range of equipment make acquiring comprehensive and rich training data extremely difficult. Furthermore, online learning and data collection processes can disrupt normal production operations, further limiting the effective application of traditional data-driven approaches. Furthermore, in actual system deployment, the design of corrective control strategies must ensure the safety of the belt conveyor system to avoid equipment damage, belt tearing, or overall operational instability caused by excessive control or frequent adjustments. Therefore, the control system must not only achieve corrective effects but also strictly adhere to the physical limitations of the equipment and possess robustness to adapt to the ever-changing industrial environment. However, many intelligent control strategies lack sufficient safety and robustness modeling, which can lead to failures and safety hazards in actual operation. Summary of the Invention

[0006] In order to solve the technical problems existing in the above-mentioned prior art, the present invention provides a multi-roller coordinated belt deviation correction control method and system, and the technical solution is as follows:

[0007] On the one hand, a multi-roller coordinated belt deviation correction control method is provided, the method comprising:

[0008] S1. Obtain the current belt edge through visual edge detection technology;

[0009] S2. Compare the current belt edge with the belt edge setting value to obtain the pixel correlation point where each correction roller intersects with the standard edge line. Belt edge deviation values, including offset distance and offset speed , combined with the angle of the correction roller , construct the state vector , i is the number of the correction roller, n is the total number of correction rollers;

[0010] S3, the state vector The policy network of the offline-trained multi-agent deep reinforcement learning model outputs an action, which is then combined with the state vector to form a state-action pair. This pair is then input into the offline-trained action reconstruction model, which reconstructs the action and outputs a safe action for subsequent motor control.

[0011] S4. The state-action pair is also input into the action value network and action cost network of the multi-agent deep reinforcement learning model completed by offline training, the action output by the policy network is evaluated, and the policy network, action value network and action cost network are trained online;

[0012] S5. The output safety action is processed and converted into an electrical signal and input into the servo motor driver. The driver controls the rotation of the servo motors of each correction roller to perform belt correction control.

[0013] On the other hand, a multi-roller coordinated belt deviation correction control system is provided, the system comprising:

[0014] A detection module is used to detect the current belt edge through visual edge detection technology;

[0015] A construction module is used to compare the current belt edge with the belt edge setting value to obtain the pixel correlation point where each correction roller intersects with the standard edge line Belt edge deviation values, including offset distance and offset speed , combined with the angle of the correction roller , construct the state vector , i is the number of the correction roller, n is the total number of correction rollers;

[0016] Output module for the state vector The policy network of the offline-trained multi-agent deep reinforcement learning model outputs an action, which is then combined with the state vector to form a state-action pair. This pair is then input into the offline-trained action reconstruction model, which reconstructs the action and outputs a safe action for subsequent motor control.

[0017] An evaluation module, which is used to input the action value network and action cost network of the multi-agent deep reinforcement learning model trained offline into the state-action pair, evaluate the actions output by the policy network, and train the policy network, action value network, and action cost network online;

[0018] The deviation correction control module is used to process the output safety action and convert it into an electrical signal which is input into the servo motor driver. The driver controls the rotation of the servo motors of each deviation correction roller to perform belt deviation correction control.

[0019] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the above-mentioned multi-roller coordinated belt deviation correction control method.

[0020] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the above-mentioned multi-roller coordinated belt deviation correction control method.

[0021] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0022] 1) To address the problem of requiring coordinated adjustment of multiple correcting rollers in long-distance belt conveyor systems, this paper proposes a solution based on a multi-agent deep reinforcement learning model. Multiple agents independently learn their own correcting roller control strategies, and a linkage mechanism ensures coordinated action among these rollers, effectively improving the overall system efficiency and stability. To capture the temporal and spatial linkages between different correcting rollers, this paper constructs a spatiotemporal graph model. This model effectively represents the relationships between the various correcting rollers in the belt conveyor system, including their positions, states, and interactions with the belt. By introducing a spatiotemporal graph attention mechanism, the interactions between the correcting rollers are dynamically weighted, ensuring effective modeling of information transfer and decision-making during belt operation. This spatiotemporal graph attention mechanism intelligently evaluates the actions of different correcting rollers based on their current state and historical data, enabling each agent to make decisions not only by considering its own state but also by effectively incorporating dynamic information from other agents. This mechanism ensures consistent temporal response while achieving spatial coordination, avoiding system instability caused by information lag or inconsistent decisions between rollers. In the event of an emergency, the system can quickly identify the correcting roller that needs adjustment and, through collaboration between agents, quickly respond and make appropriate adjustments. This not only improves the operating efficiency of long-distance belt conveyor systems, but also significantly enhances their adaptability and safety in complex working environments.

[0023] 2) To address the difficulty in acquiring training data for multi-agent deep reinforcement learning, this paper employs a data augmentation method based on multi-roller collaboration. This method generates collaborative offline data for multi-agent deep reinforcement learning through a diffusion model, addressing the difficulty in obtaining large-scale, high-quality training data in real-world environments. By enhancing the diversity and representativeness of this data, the generalization capability of the reinforcement learning model is significantly improved, enabling the model to better cope with complex operating environments.

[0024] 3) In the practical deployment of multi-roller collaborative control, the action strategy of each agent is crucial to the safety and stability of the system. Traditional methods typically rely solely on simple constraints to limit the agent's range of motion, which cannot effectively cope with the dynamic changes and potential risks in complex environments. This invention further enhances the safety of the agent by introducing a deep learning-driven safety constraint mechanism. During the reinforcement learning training phase, an action reconstruction model is employed to perform dual functions: First, the action reconstruction model extracts the distribution deviation of the agent's action strategy and quantifies the difference between the action strategy and the prior distribution by calculating the KL divergence, thereby detecting whether the action deviates from the normal range. Second, the action reconstruction model also imposes physical constraints on the reconstructed action, ensuring that the generated action conforms to the data distribution characteristics while adhering to predefined physical constraints. When the agent's action approaches or exceeds the predetermined safety boundary, the action reconstruction model promptly identifies the distribution deviation and applies physical constraints to the reconstructed action, preventing the occurrence of out-of-distribution actions that may lead to unsafe consequences. Furthermore, the KL divergence and physical constraint loss are calculated simultaneously during the training process to optimize the model's safety and physical feasibility. This mechanism ensures that all agents execute actions within physical constraints and safety limits, significantly reducing the risk of belt damage caused by incorrect decisions and improving the model's reliability and practical application value. Furthermore, when an action is detected outside the normal distribution range, the system replaces the original action with a sample of the nearest in-distribution action, further ensuring its safety. This innovative approach not only ensures the safe operation of multi-agents in complex environments but also strongly supports its feasibility in real-world scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 This is a flow chart of a multi-roller coordinated belt deviation correction control method provided by an embodiment of the present invention;

[0027] Figure 2 This is an overall block diagram of a multi-roller coordinated belt deviation correction control method provided by an embodiment of the present invention;

[0028] Figure 3 1 is a schematic diagram of a training process of a multi-roller coordinated belt deviation correction control method provided by an embodiment of the present invention;

[0029] Figure 4 This is a structural block diagram of the intelligent agent deep reinforcement learning model provided by an embodiment of the present invention;

[0030] Figure 5 This is a block diagram of the spatiotemporal graph attention mechanism provided by an embodiment of the present invention;

[0031] Figure 6 This is a block diagram of the attention layer structure of each spatiotemporal graph provided by an embodiment of the present invention;

[0032] Figure 7 This is a flow chart of a data augmentation method for a diffusion model based on collaborative reward-driven implementation provided by an embodiment of the present invention;

[0033] Figure 8 This is a structural diagram of an action reconstruction model provided by an embodiment of the present invention;

[0034] Figure 9 This is a block diagram of a multi-roller coordinated belt deviation correction control system provided by an embodiment of the present invention;

[0035] Figure 10 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0037] An embodiment of the present invention provides a multi-roller coordinated belt deviation correction control method, which can be implemented by an electronic device, and the electronic device can be a terminal or a server. Figure 1 The flow chart of this method is shown below. Figure 2 The overall block diagram of the method is shown below. The processing flow may include the following steps:

[0038] S1. Obtain the current belt edge through visual edge detection technology;

[0039] The visual belt edge technology of the embodiment of the present invention can adopt existing technology, which will not be described in detail here.

[0040] S2. Compare the current belt edge with the belt edge setting value to obtain the pixel correlation point where each correction roller intersects with the standard edge line. Belt edge deviation values, including offset distance and offset speed , combined with the angle of the correction roller , construct the state vector , i is the number of the correction roller, n is the total number of correction rollers;

[0041] S3, the state vector The policy network of the offline-trained multi-agent deep reinforcement learning model outputs an action, which is then combined with the state vector to form a state-action pair. This pair is then input into the offline-trained action reconstruction model, which reconstructs the action and outputs a safe action for subsequent motor control.

[0042] S4. The state-action pair is also input into the action value network and action cost network of the multi-agent deep reinforcement learning model completed by offline training, the action output by the policy network is evaluated, and the policy network, action value network and action cost network are trained online;

[0043] S5. The output safety action is processed and converted into an electrical signal and input into the servo motor driver. The driver controls the rotation of the servo motors of each correction roller to perform belt correction control (at the same time, the motor feeds back various state parameters of the motor and adds them to the motor state sequence. The belt then passes through the visual edge detection to feedback the edge of the belt, realizing a closed loop).

[0044] Optionally, the offline training process of the multi-agent deep reinforcement learning model is:

[0045] By manually controlling the deviation correction roller motor, multi-dimensional data of the belt during normal operation is collected and recorded. The real-time status characteristics of each correction roller, including the correction roller angle , Control the motor rotation angle increment corresponding to the deviation correction roller action , establish the standard edge position of the belt based on historical operating data As a benchmark reference (such as Figure 3 The dotted line of the on-site belt multi-roller system shows that at the standard edge position, the pixel correlation point on the image is defined for each correcting roller at the intersection of the standard edge line and the correcting roller. , obtain the actual belt edge position in real time through visual edge detection technology , by calculating the reference point The actual edge of the belt at the same horizontal position Offset distance and offset speed , build contains The state vector of is the total number of correction rollers, and at the reward function design level, defines the immediate reward ,in To offset the speed penalty coefficient, this design fuses the position deviation and dynamic characteristics through nonlinear weighting to finally generate an offline training dataset:

[0046]

[0047] Data contains The complete interaction trajectory of group length T, where Represents multi-roller control action;

[0048] Then, the reinforcement learning model is trained offline, including offline training of the policy network, action value network, and action cost network. The policy network outputs work, the action value network evaluates the long-term impact of the adjustment action on the belt state, and influences the trend of the policy network to generate actions through value output. The action cost network evaluates the safety of the action and suppresses the generation of unsafe actions through cost output.

[0049] To meet the needs of multi-roller coordinated control in belt conveyor systems, a data augmentation method based on a collaborative reward-driven diffusion model is used to effectively expand the exploration range of the multi-agent state-action space. The integrated data is then placed in an experience replay pool for training the reinforcement learning model.

[0050] In addition, based on the action reconstruction model, part of the model outputs the distribution deviation and action cost of the action, which are used to update the action cost network, and part of the model outputs the reconstructed action as the actual output safety action.

[0051] Optionally, the online training process of the multi-agent deep reinforcement learning model is:

[0052] After the state-action pair is input into the action value network and the action cost network, the loss of the action value network is: the long-term cumulative reward value output by the action value network and immediate rewards after the action Calculate the loss. The loss of the action cost network is: the distribution deviation of the action reconstruction model output and the cost value output by the action cost network Calculate the loss and take the weighted sum of the outputs of the two networks as the comprehensive reward , and state-action pairs are organized into The data is fed into the experience replay pool in the form of data, and the policy network is optimized through small-batch incremental training to positively reinforce high-return and low-risk actions and reduce the impact weight of high-risk actions, thereby improving the adaptability and robustness of the system.

[0053] Alternatively, as Figure 4 As shown in the figure, the multi-agent deep reinforcement learning model consists of three networks: action value network, action cost network and policy network. All three networks take a [state, action] trajectory sequence as input, extract the corresponding feature vectors through the spatiotemporal graph attention mechanism, and finally output the corresponding goals through the fully connected layer.

[0054] The policy network , output the mean of the Gaussian distribution and variance , , and through action sampling Generate randomly sampled actions, which will be evaluated as input to the action value network and the action cost network;

[0055] Action value network, which maps the feature vector extracted by the spatiotemporal graph attention mechanism into a long-term cumulative reward value , and based on the Bellman error loss Update the action-value network, where Represents the state Take action Instant rewards when is the discount factor, which takes the value Between, used to balance the relative importance of current rewards and future rewards, Indicates execution of an action Then transfer to the next state, is the target network's next state and all optional actions The estimated value of Indicates the next state Next, all optional actions The maximum value in reflects the highest expected reward that can be obtained by choosing the optimal action in the future state;

[0056] The action cost network learns the potential risks of actions based on historical trajectories and outputs high cost values for dangerous actions. , by combining the distribution deviation of the action reconstruction model output and the action cost network output Make the difference and update the network as the loss value of the network;

[0057] Then the policy network obtains the comprehensive reward through weighted calculation based on the two evaluation results. , used to dynamically adjust the action generation strategy, the overall optimization goal is defined as ,in is an adjustable risk factor, is the entropy regularization coefficient with a value of 0.01-0.1, and the policy entropy It is used to measure the randomness of the strategy. A high entropy value indicates a more random strategy, which can improve the action reward while avoiding safety risks. During the training process, the state action trajectory randomly sampled by the experience replay pool is constructed as spatiotemporal graph data, driving the three networks to alternately update parameters to form a collaborative optimization. Among them, the policy network calculates the policy gradient through Monte Carlo. Implement parameter update.

[0058] Alternatively, as Figure 5As shown in Figure 2, the core of the spatiotemporal graph attention mechanism is to establish an adaptive modeling method that can simultaneously capture spatial correlation and temporal dynamics. It consists of multiple spatiotemporal graph attention layers connected by residuals, and the hidden feature vectors are continuously transmitted to the deep network.

[0059] Each spatiotemporal graph attention layer, such as Figure 6 As shown, the multi-roller system augmented data is first modeled as a graph structure ,node Indicates the The attribute vector of the correction roller, including the angle of the correction roller , association points , the offset distance between the associated point and the standard edge line and offset speed ,In order to effectively extract features, all attribute vectors are normalized to the range of [0, 1].,These features are stored in the form of time series, ,side Indicates the relative distance between each correction roller;

[0060] Then calculate the spatial adjacency matrix, the calculation process is:

[0061] Through the edge E in the graph structure, the geometric relationship between discrete nodes is converted into a matrix form, and the belt axis is used as the one-dimensional coordinate axis to extract Installation position of each guiding roller node , a nonlinear attenuation function based on a Gaussian kernel is used to map the physical distance to an adjacency weight in the interval [0, 1]):

[0062]

[0063] in For the correction roller and guiding roller The adjacency weight between It is the correction roller and guiding roller The installation location, is the standard deviation of the Gaussian kernel, which controls the neighborhood attenuation rate. In order to balance the computational efficiency and accuracy, a distance threshold is introduced. Perform hard truncation on the spatial adjacency matrix when hour, Set to 0, only retain the adjacency relationship within the effective range, and finally generate a highly sparse spatial adjacency matrix ;

[0064] Then the spatial attention mechanism is calculated to dynamically calculate the node influence through the cross-dimensional interaction of the triple weight matrix. The calculation process is:

[0065] Using time compression weights For input Perform temporal feature aggregation and then use channel-time interaction weights Establish cross-dimensional association rules to capture the mutual influence between feature channels and time dimensions, and then use spatial projection weights Generate the raw attention matrix between nodes , the formula is: ,in is a learnable weight vector used to adjust the scaling of the original attention matrix, is the Sigmoid activation function, is the bias term;

[0066] The original attention matrix is processed by row-wise Softmax normalization to output the dynamic attention weight that conforms to the probability distribution characteristics , each element of which Clearly characterize the correction roller in the current space-time state Correction roller The influence strength of is:

[0067]

[0068] Dynamic attention weights are converted to and spatial adjacency matrix Fusion, get the enhanced adjacency matrix : , and then construct the spatial attention hidden features. In the rth spatiotemporal graph attention layer, the hidden features of the spatial attention output are calculated by the hidden features of the previous spatiotemporal graph attention layer, which is expressed as:

[0069]

[0070] The hidden features of the first spatiotemporal graph attention layer are calculated by node attribute features: ; Then calculate the time attention mechanism, the calculation process is:

[0071] The temporal attention mechanism captures dynamic dependencies in the temporal dimension through parameterized modeling. Its core operation is formalized as follows:

[0072]

[0073] in is a learnable weight vector used to adjust the scaling of the attention matrix, The weight matrix used to compress the node dimensions to a scalar, The weight matrix used to construct the association rules between feature channels and nodes, The weight vector used to extract the feature combination across time steps, is a bias term used to adjust the output of the model;

[0074] The resulting original attention matrix After normalization, the probability distribution weight matrix is generated: , the weight matrix dynamically adjusts the input time series features through matrix multiplication, enabling the model to strengthen the correlation strength of key time segments:

[0075]

[0076] Then the spatiotemporal features are fused through the spatiotemporal feature fusion module. The process is as follows:

[0077] In the spatial dimension, the graph convolution operation is designed based on the spectral graph theory, and in the temporal dimension, conventional convolution is used to extract features. The formula for spatial graph convolution is: ,in is the feature vector after spatial graph convolution The representation of time steps, It is The node attribute vector of time steps, are the learnable polynomial coefficients, For the The Chebyshev polynomial of order is recursively defined as , initial term , , represents the scaled Laplacian matrix, where is the graph Laplacian matrix, for The maximum eigenvalue of is the identity matrix, is the order of the polynomial expansion, which controls the neighborhood range of node feature aggregation. is the spatial attention weight matrix, which is obtained by Hadamard product Adjust the adjacency relationship. This formula combines polynomial expansion and spatial attention mechanism. Order polynomial Limit node feature aggregation to arrive In the range of order neighbors, we avoid the computational burden of directly decomposing the large-scale graph Laplacian matrix and introduce the spatial attention matrix Enhance dynamic spatial relationship modeling and adjust adjacency relationships through Hadamard product, so that the feature aggregation process can adapt to dynamically changing spatial associations;

[0078] The temporal convolution part is used to process the temporal features Perform the convolution operation, defined as:

[0079]

[0080] in is the output feature vector after time convolution; represents the temporal convolution kernel parameters, Indicates the The temporal features of the spatiotemporal graph attention layer have the following channels: , the time step is The feature vector at , extracts features at different time steps, sums all channels and time offsets, aggregates feature information from different channels and time steps to capture and enhance dynamic information, is the corresponding bias term, The sigmoid function enables the model to capture the local dynamic pattern of each belt-correcting roller in the time dimension, complementing the spectral domain operation of spatial graph convolution to jointly build spatiotemporal feature extraction capabilities.

[0081] After splicing in the channel dimension, Convolution integrates information across channels:

[0082]

[0083] in represents the concatenation operation in the channel dimension, Represents the fused convolution kernel, which is used for the linear combination of cross-channel information. is the bias term of the fusion layer, is a standard convolution operation;

[0084] Finally, multiple spatiotemporal graph attention layers optimize feature extraction capabilities through residual connections, enhancing the depth and expressiveness of the model.

[0085] Optionally, the multi-roller system augmented data is obtained by augmenting real sampled data through a data augmentation method based on a collaborative reward-driven diffusion model. The augmentation method guides the data generation process through a collaborative reward mechanism, so that the generated data is not only diverse, but also fully reflects the collaborative characteristics of the multi-roller system, and improves the learning efficiency and generalization ability of the multi-roller correction strategy (to solve the problems of scarcity of training data in multi-agent deep reinforcement learning and lack of collaboration of traditional augmentation methods), such as Figure 7 Shown, including:

[0086] First, the real sampled multi-agent trajectory data Perform preprocessing to construct a trajectory sequence in a unified format , including the state vector of each correction roller ,action And the global reward for this trajectory , represents the cumulative reward value obtained for each step of the trajectory, and the trajectory sequence is expressed as:

[0087]

[0088] in, represents the number of agents, is the number of time steps;

[0089] The following collaborative reward mechanism is designed to make data generation collaborative:

[0090] 1) Motion coordination reward, aimed at suppressing the conflict between adjacent correction rollers, motion coordination reward It is defined as the sum of the squares of the differences in the movements of all adjacent correcting rollers:

[0091]

[0092] in, is the weight coefficient of the action coordination constraint, and Respectively represent and The action of the correcting roller;

[0093] 2) Geometric smoothness bonus, which aims to eliminate sudden changes in the curvature of the belt edge. It is defined as the sum of squares of the second-order differences of the edge positions of the guide roller:

[0094]

[0095] in, is the weight coefficient of the geometric smoothness constraint, is the reference point The pixel points of the actual edge of the belt at the same horizontal position value;

[0096] 3) Global collaborative rewards, which aim to balance overall deviation and local deviation. Defined as the horizontal pixel position of the associated point and The weighted sum of the average value of the deviation and the maximum deviation value:

[0097]

[0098] in, is the weight coefficient of global collaborative optimization, is the weight coefficient of local deviation, is the adjustable coefficient;

[0099] During the training of the conditional diffusion model, a noise function is constructed , its specific implementation is as follows: In the initial stage of generating data, the model generates a fuzzy sample by adding noise to the real data. Assuming there is a real trajectory data , which is the starting point of generation. During the noise addition process, the trajectory state is gradually contaminated by noise over time. This process is modeled as follows:

[0100]

[0101] In this formula, It is in The trajectory state generated by the step, It is in The trajectory state generated by the step, Controls the influence of the true trajectory in each update step, and is the added noise, the scaling factor and Ensure that noise and true trajectories are balanced in updates;

[0102] Based on the noise adding function , a dual loss mechanism is used to achieve joint optimization of data reconstruction and collaborative guidance, where the data distribution reconstruction loss It is used to ensure the consistency of the distribution of generated data and real data, which is defined as:

[0103]

[0104] in, is the original trajectory data, In the diffusion process Step noise trajectory data, is a noise sample, is the number of time steps;

[0105] Loss of multi-roller synergy The data used to guide the generation of data is collaborative and is defined as:

[0106]

[0107] in, is the global value function, equal to the compound reward The cumulative value of ;

[0108] Overall objective function Combining these two losses:

[0109]

[0110] in, is the weight coefficient of collaborative loss;

[0111] In the denoising sampling stage, the model aims to remove the noise-contaminated trajectory. The noise is gradually removed in order to restore or generate clear trajectory data. The formula is:

[0112]

[0113] In this formula, is the trajectory state generated after denoising, The weight of the control noise reflects the impact of the noise on the trajectory update. It is before Steps The cumulative product of , the predicted noise Based on the current trajectory status and number of steps Calculated, is the standard deviation, which controls the amplitude of the added noise, is the noise sampled from a standard normal distribution , adjustment coefficient Controls the degree of influence of reward gradient on trajectory updates, Collaborative rewards About the current track status gradient.

[0114] In the actual deployment of multi-roller coordinated control, the action strategy of each agent is crucial to the safety and stability of the system. To cope with the dynamic changes and potential risks in complex environments and ensure the physical feasibility and safety of the agent's actions, an action reconstruction model is designed in this embodiment of the present invention.

[0115] Alternatively, as Figure 8 As shown in the figure, the action reconstruction model includes a dual-branch encoder and a decoder. The core task of the encoder is to extract the latent characteristics of the input data and parse the dynamic constraint parameters, while the decoder reconstructs the action based on the latent variables generated by the encoder and generates the nearest neighbor safe action within the constraint range through physical rules containing constraint parameters.

[0116] In the encoder, the state-action pairs sampled from the offline data are used as input, and after dual-branch processing, the main branch is processed by Extracting the mean of latent features and variance , describing the latent space The probability distribution of For input, is the encoder; then, by sampling from the latent space, a reparameterization technique is used to generate latent variables from a distribution defined by mean and variance. :

[0117]

[0118] The constraint branch of the encoder generates dynamic safety parameters, including margin coefficients, through the multi-layer perceptron MLP , offset impact factor , nonlinear index and tolerance factor ;

[0119] The decoder is responsible for converting the latent variables Perform reconstruction, generate safe actions, and constrain the actions through dynamic safety parameters to ultimately generate physically compliant reconstruction actions. Define the safe physical rules as follows:

[0120] First, the dynamic corner constraint is defined as ,in , is the maximum rotation angle of the motor. This constraint means that when a large deviation of the belt near the correction roller is detected, the system will relax the rotation angle limit of the correction roller to allow a larger correction action. Secondly, the rate constraint is defined as ,in , is the maximum angular velocity of the motor, is the belt speed vector at each roller position. This constraint indicates that when the overall belt speed exceeds the safety threshold When , the upper limit of the action rate decreases exponentially; finally, the coordination constraint is defined as and ,This constraint ensures that the adjustment amplitude ratio of adjacent correcting rollers does not exceed the ratio of the belt offset by a specific coefficient, and achieves the optimal synthesis of the correcting force while avoiding mechanical interference;

[0121] The overall optimization objective loss of the action reconstruction model consists of KL divergence, reconstruction loss and constraint loss;

[0122] Among them, KL divergence directly measures the difference between the decoder output action and the real action to ensure that the model can accurately restore the input data, potential distribution loss, and calculate the distribution deviation between the potential distribution and the prior distribution through KL divergence. The calculation formula is , where the prior distribution Use the mean based on offline data and covariance matrix To define, ,In this way, the generation of the latent space will be closer to the distribution of offline data;

[0123] Reconstruction loss ;

[0124] The constraint loss term quantifies the degree of violation of the safety margin:

[0125]

[0126] The KL divergence, reconstruction loss, and constraint loss are weighted together to form a total distribution constraint loss, and the encoder and decoder parameters are optimized based on this loss. Through this mechanism, the model improves data reconstruction accuracy while maintaining the distribution rationality of the latent space and the physical feasibility of the action.

[0127] The action reconstruction model uses bimodal indicators to determine whether the input action exceeds the normal range and defines ; Detection logic uses dynamic threshold mechanism to make judgments: When Score > Or there is a single physical constraint violation> When , it is determined to be an abnormal action, and is a preset threshold, at which point the security correction mechanism is triggered to generate compliance actions .

[0128] like Figure 9 As shown, an embodiment of the present invention further provides a multi-roller coordinated belt deviation correction control system, the system comprising:

[0129] Detection module 910, used to detect the current belt edge through visual edge detection technology;

[0130] Construction module 920 is used to compare the current belt edge with the belt edge setting value to obtain the pixel correlation point where each correction roller intersects with the standard edge line Belt edge deviation values, including offset distance and offset speed , combined with the angle of the correction roller , construct the state vector , i is the number of the correction roller, n is the total number of correction rollers;

[0131] Output module 930, for the state vector The policy network of the offline-trained multi-agent deep reinforcement learning model outputs an action, which is then combined with the state vector to form a state-action pair. This pair is then input into the offline-trained action reconstruction model, which reconstructs the action and outputs a safe action for subsequent motor control.

[0132] Evaluation module 940, for inputting the state-action pairs into the action value network and action cost network of the multi-agent deep reinforcement learning model trained offline, evaluating the actions output by the policy network, and training the policy network, action value network, and action cost network online;

[0133] The deviation correction control module 950 is used to process the output safety action, convert it into an electrical signal and input it into the servo motor driver. The driver controls the rotation of the servo motors of each deviation correction roller to perform belt deviation correction control.

[0134] The functional structure of a multi-roller cooperative belt deviation correction control system provided by an embodiment of the present invention corresponds to the multi-roller cooperative belt deviation correction control method provided by an embodiment of the present invention, and will not be repeated here.

[0135] Figure 10 It is a structural diagram of an electronic device 1000 provided in an embodiment of the present invention. The electronic device 1000 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 1001 and one or more memories 1002, wherein at least one instruction is stored in the memory 1002, and the at least one instruction is loaded and executed by the processor 1001 to implement the steps of the above-mentioned multi-roller collaborative belt deviation correction control method.

[0136] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory device containing instructions. These instructions are executable by a processor in a terminal to implement the multi-roller coordinated belt deviation control method. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device.

[0137] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0138] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-roller coordinated belt deviation correction control method, characterized in that: The method comprises: S1. Obtain the current belt edge through visual edge detection technology; S2. Compare the current belt edge with the belt edge setting value to obtain the pixel correlation point where each correction roller intersects with the standard edge line. Belt edge deviation values, including offset distance and offset speed , combined with the angle of the correction roller , construct the state vector , i is the number of the correction roller, n is the total number of correction rollers; S3, the state vector The policy network of the offline-trained multi-agent deep reinforcement learning model outputs an action, which is then combined with the state vector to form a state-action pair. This pair is then input into the offline-trained action reconstruction model, which reconstructs the action and outputs a safe action for subsequent motor control. S4. The state-action pair is also input into the action value network and action cost network of the multi-agent deep reinforcement learning model completed by offline training, the action output by the policy network is evaluated, and the policy network, action value network and action cost network are trained online; S5. The output safety action is processed and converted into an electrical signal and input into the servo motor driver. The driver controls the rotation of the servo motors of each correction roller to perform belt correction control. The multi-agent deep reinforcement learning model consists of three networks: an action value network, an action cost network, and a policy network. All three networks take a sequence of [state, action] trajectories as input, extract corresponding feature vectors through a spatiotemporal graph attention mechanism, and finally output corresponding targets through a fully connected layer. The core of the spatiotemporal graph attention mechanism is to establish an adaptive modeling method that can simultaneously capture spatial correlation and temporal dynamics. It consists of multiple spatiotemporal graph attention layers connected by residuals, and the hidden feature vectors are continuously passed to the deep network. Each spatiotemporal graph attention layer first models the multi-roll system augmented data as a graph structure ,node Indicates the The attribute vector of the correction roller, including the angle of the correction roller , association points , the offset distance between the associated point and the standard edge line and offset speed ,In order to effectively extract features, all attribute vectors are normalized to the range of [0, 1].,These features are stored in the form of time series, ,side Indicates the relative distance between each correction roller; Then calculate the spatial adjacency matrix; Then the spatial attention mechanism is calculated to dynamically calculate the node influence through the cross-dimensional interaction of the triple weight matrix; Then calculate the temporal attention mechanism; Then the spatiotemporal features are fused through the spatiotemporal feature fusion module; Finally, multiple spatiotemporal graph attention layers optimize feature extraction capabilities through residual connections, enhancing the depth and expressiveness of the model.

2. The method according to claim 1, characterized in that The offline training process of the multi-agent deep reinforcement learning model is as follows: By manually controlling the deviation correction roller motor, multi-dimensional data of the belt during normal operation is collected and recorded. The real-time status characteristics of each correction roller, including the correction roller angle , Control the motor rotation angle increment corresponding to the deviation correction roller action , establish the standard edge position of the belt based on historical operating data As a reference, at the standard edge position, the intersection of the standard edge line and the correction roller is used to define the pixel association point on the image for each correction roller. , obtain the actual belt edge position in real time through visual edge detection technology = , by calculating the reference point The actual edge of the belt at the same horizontal position Offset distance and offset speed , build contains The state vector of is the total number of correction rollers, and at the reward function design level, defines the immediate reward ,in To offset the speed penalty coefficient, this design fuses the position deviation and dynamic characteristics through nonlinear weighting to finally generate an offline training dataset: ; Data contains The complete interaction trajectory of group length T, where Represents multi-roller control action; Then, the reinforcement learning model is trained offline, including offline training of the policy network, action value network, and action cost network. The policy network outputs work, the action value network evaluates the long-term impact of the adjustment action on the belt state, and influences the trend of the policy network to generate actions through value output. The action cost network evaluates the safety of the action and suppresses the generation of unsafe actions through cost output. To meet the needs of multi-roller coordinated control in belt conveyor systems, a data augmentation method based on a collaborative reward-driven diffusion model is used to effectively expand the exploration range of the multi-agent state-action space. The integrated data is then placed in an experience replay pool for training the reinforcement learning model. In addition, based on the action reconstruction model, part of the model outputs the distribution deviation and action cost of the action, which are used to update the action cost network, and part of the model outputs the reconstructed action as the actual output safety action.

3. The method according to claim 2, characterized in that The online training process of the multi-agent deep reinforcement learning model is as follows: After the state-action pair is input into the action value network and the action cost network, the loss of the action value network is: the long-term cumulative reward value output by the action value network and immediate rewards after the action Calculate the loss. The loss of the action cost network is: the distribution deviation of the action reconstruction model output and the cost value output by the action cost network Calculate the loss and take the weighted sum of the outputs of the two networks as the combined reward , and state-action pairs are organized into The data is fed into the experience replay pool in the form of data, and the policy network is optimized through small-batch incremental training to positively reinforce high-return and low-risk actions and reduce the impact weight of high-risk actions, thereby improving the adaptability and robustness of the system.

4. The method according to claim 3, characterized in that The policy network of the multi-agent deep reinforcement learning model Output Gaussian distribution mean and variance , , and through action sampling Generate randomly sampled actions, which will be evaluated as input to the action value network and the action cost network; Action value network, which maps the feature vector extracted by the spatiotemporal graph attention mechanism into a long-term cumulative reward value , and based on the Bellman error loss Update the action-value network, where Represents the state Take action Instant rewards when is the discount factor, which takes the value Between, used to balance the relative importance of current rewards and future rewards, Indicates execution of an action Then transfer to the next state, is the target network's next state and all optional actions The estimated value of Indicates the next state Next, all optional actions The maximum value in reflects the highest expected reward that can be obtained by choosing the optimal action in the future state; The action cost network learns the potential risks of actions based on historical trajectories and outputs high cost values for dangerous actions. , by combining the distribution deviation of the action reconstruction model output and the action cost network output Make the difference and update the network as the loss value of the network; Then the policy network obtains the comprehensive reward through weighted calculation based on the two evaluation results. , used to dynamically adjust the action generation strategy, the overall optimization goal is defined as ,in is an adjustable risk factor, is the entropy regularization coefficient with a value of 0.01-0.1, and the policy entropy It is used to measure the randomness of the strategy. A high entropy value indicates a more random strategy, which can improve the action reward while avoiding safety risks. During the training process, the state action trajectory randomly sampled by the experience replay pool is constructed as spatiotemporal graph data, driving the three networks to alternately update parameters to form a collaborative optimization. Among them, the policy network calculates the policy gradient through Monte Carlo. Implement parameter update.

5. The method according to claim 4, characterized in that The calculation process of the computation space adjacency matrix is as follows: Through the edge E in the graph structure, the geometric relationship between discrete nodes is converted into a matrix form, and the belt axis is used as the one-dimensional coordinate axis to extract Installation position of each guiding roller node , a nonlinear attenuation function based on a Gaussian kernel is used to map the physical distance to an adjacency weight in the interval [0, 1]): ; in For the correction roller and guiding roller The adjacency weight between It is the correction roller and guiding roller The installation location, is the standard deviation of the Gaussian kernel, which controls the neighborhood attenuation rate. In order to balance the computational efficiency and accuracy, a distance threshold is introduced. Perform hard truncation on the spatial adjacency matrix when hour, Set to 0, only retain the adjacency relationship within the effective range, and finally generate a highly sparse spatial adjacency matrix ; The computational spatial attention mechanism achieves dynamic calculation of node influence through cross-dimensional interaction of triple weight matrices. The calculation process is as follows: Using time compression weights For input Perform temporal feature aggregation and then use channel-time interaction weights Establish cross-dimensional association rules to capture the mutual influence between feature channels and time dimensions, and then use spatial projection weights Generate the raw attention matrix between nodes , the formula is: ,in is a learnable weight vector used to adjust the scaling of the original attention matrix, is the Sigmoid activation function, is the bias term; The original attention matrix is processed by row-wise Softmax normalization to output the dynamic attention weight that conforms to the probability distribution characteristics , each element of which Clearly characterize the correction roller in the current space-time state Correction roller The influence strength of is: ; Dynamic attention weights are converted to and spatial adjacency matrix Fusion, get the enhanced adjacency matrix : , and then construct the spatial attention hidden features. In the rth spatiotemporal graph attention layer, the hidden features of the spatial attention output are calculated by the hidden features of the previous spatiotemporal graph attention layer, which is expressed as: ; The hidden features of the first spatiotemporal graph attention layer are calculated by node attribute features: ; The calculation process of the time attention mechanism is as follows: The temporal attention mechanism captures dynamic dependencies in the temporal dimension through parameterized modeling. Its core operation is formalized as follows: ; in is a learnable weight vector used to adjust the scaling of the attention matrix, The weight matrix used to compress the node dimensions to a scalar, The weight matrix used to construct the association rules between feature channels and nodes, The weight vector used to extract the feature combination across time steps, is a bias term used to adjust the output of the model; The resulting original attention matrix After normalization, the probability distribution weight matrix is generated: , the weight matrix dynamically adjusts the input time series features through matrix multiplication, enabling the model to strengthen the correlation strength of key time segments: ; The spatiotemporal features are fused by the spatiotemporal feature fusion module, and the process is as follows: In the spatial dimension, the graph convolution operation is designed based on the spectral graph theory, and in the temporal dimension, conventional convolution is used to extract features. The formula for spatial graph convolution is: ,in is the feature vector after spatial graph convolution The representation of time steps, It is The node attribute vector of time steps, are the learnable polynomial coefficients, For the The Chebyshev polynomial of order is recursively defined as , initial term , , represents the scaled Laplacian matrix, where is the graph Laplacian matrix, for The maximum eigenvalue of is the identity matrix, is the order of the polynomial expansion, which controls the neighborhood range of node feature aggregation. is the spatial attention weight matrix, which is obtained by Hadamard product Adjust the adjacency relationship. This formula combines polynomial expansion and spatial attention mechanism. Order polynomial Limit node feature aggregation to arrive In the range of order neighbors, we avoid the computational burden of directly decomposing the large-scale graph Laplacian matrix and introduce the spatial attention matrix Enhance dynamic spatial relationship modeling and adjust adjacency relationships through Hadamard product, so that the feature aggregation process can adapt to dynamically changing spatial associations; The temporal convolution part is used to process the temporal features Perform the convolution operation, defined as: ; in is the output feature vector after time convolution; represents the temporal convolution kernel parameters, Indicates the The temporal features of the spatiotemporal graph attention layer have the following channels: , the time step is ( The feature vector at , extracts features at different time steps, sums all channels and time offsets, aggregates feature information from different channels and time steps to capture and enhance dynamic information, is the corresponding bias term, The sigmoid function enables the model to capture the local dynamic pattern of each belt-correcting roller in the time dimension, complementing the spectral domain operation of spatial graph convolution to jointly build spatiotemporal feature extraction capabilities. After splicing in the channel dimension, Convolution integrates information across channels: ; in represents the concatenation operation in the channel dimension, Represents the fused convolution kernel, which is used for the linear combination of cross-channel information. is the bias term of the fusion layer, It is a standard convolution operation.

6. The method according to claim 5, characterized in that The multi-roller system augmented data is obtained by augmenting real sampled data using a collaborative reward-driven diffusion model data augmentation method. The augmentation method guides the data generation process through a collaborative reward mechanism, making the generated data not only diverse but also fully reflecting the collaborative characteristics of the multi-roller system, thereby improving the learning efficiency and generalization ability of the multi-roller correction strategy, including: First, the real sampled multi-agent trajectory data Perform preprocessing to construct a trajectory sequence in a unified format , including the state vector of each correction roller ,action And the global reward for this trajectory , represents the cumulative reward value obtained for each step of the trajectory, and the trajectory sequence is expressed as: ; in, represents the number of agents, is the number of time steps; The following collaborative reward mechanism is designed to make data generation collaborative: 1) Motion coordination reward, aimed at suppressing the conflict between adjacent correction rollers, motion coordination reward It is defined as the sum of the squares of the differences in the movements of all adjacent correcting rollers: ; in, is the weight coefficient of the action coordination constraint, and Respectively represent and The action of the correcting roller; 2) Geometric smoothness bonus, which aims to eliminate sudden changes in the curvature of the belt edge. It is defined as the sum of squares of the second-order differences of the edge positions of the guide roller: ; in, is the weight coefficient of the geometric smoothness constraint, is the reference point The pixel points of the actual edge of the belt at the same horizontal position value; 3) Global collaborative rewards, which aim to balance overall deviation and local deviation. Defined as the horizontal pixel position of the associated point and The weighted sum of the average value of the deviation and the maximum deviation value: ; in, is the weight coefficient of global collaborative optimization, is the weight coefficient of local deviation, is the adjustable coefficient; During the training of the conditional diffusion model, a noise function is constructed , its specific implementation is as follows: In the initial stage of generating data, the model generates a fuzzy sample by adding noise to the real data. Assuming there is a real trajectory data , which is the starting point of generation. During the noise addition process, the trajectory state is gradually contaminated by noise over time. This process is modeled as follows: ; In this formula, It is in The trajectory state generated by the step, It is in The trajectory state generated by the step, Controls the influence of the true trajectory in each update step, and is the added noise, the scaling factor and Ensure that noise and true trajectories are balanced in updates; Based on the noise adding function , a dual loss mechanism is used to achieve joint optimization of data reconstruction and collaborative guidance, where the data distribution reconstruction loss It is used to ensure the consistency of the distribution of generated data and real data, which is defined as: ; in, is the original trajectory data, In the diffusion process Step noise trajectory data, is a noise sample, is the number of time steps; Loss of multi-roller synergy The data used to guide the generation of data is collaborative and is defined as: ; in, is the global value function, equal to the compound reward The cumulative value of ; Overall objective function Combining these two losses: ; in, is the weight coefficient of collaborative loss; In the denoising sampling stage, the model aims to remove the noise-contaminated trajectory. The noise is gradually removed in order to restore or generate clear trajectory data. The formula is: ; In this formula, is the trajectory state generated after denoising, The weight of the control noise reflects the impact of the noise on the trajectory update. It is before Steps The cumulative product of , the predicted noise Based on the current trajectory status and number of steps Calculated, is the standard deviation, which controls the amplitude of the added noise, is the noise sampled from a standard normal distribution , adjustment coefficient Controls the influence of reward gradient on trajectory update, Collaborative rewards About the current track status gradient.

7. The method according to claim 1, characterized in that The action reconstruction model includes a dual-branch encoder and a decoder. The encoder's core task is to extract the latent characteristics of the input data and parse the dynamic constraint parameters, while the decoder reconstructs the action based on the latent variables generated by the encoder and generates safe actions within the constraint range through physical rules containing the constraint parameters. In the encoder, the state-action pairs sampled from the offline data are used as input, and after dual-branch processing, the main branch is processed by Extracting the mean of latent features and variance , describing the latent space The probability distribution of For input, is the encoder; then, by sampling from the latent space, a reparameterization technique is used to generate latent variables from a distribution defined by mean and variance. : ; The constraint branch of the encoder generates dynamic safety parameters, including margin coefficients, through the multi-layer perceptron MLP , offset impact factor , nonlinear index and tolerance factor ; The decoder is responsible for converting the latent variables Perform reconstruction, generate safe actions, and constrain the actions through dynamic safety parameters to ultimately generate physically compliant reconstruction actions. Define the safe physical rules as follows: First, the dynamic corner constraint is defined as ,in , is the maximum rotation angle of the motor; secondly, the rate constraint is defined as ,in , is the maximum angular velocity of the motor, is the belt velocity vector at each roller position; finally, the coordination constraint is defined as and ; The overall optimization target loss of the action reconstruction model consists of KL divergence, reconstruction loss and constraint loss; Among them, KL divergence directly measures the difference between the decoder output action and the real action to ensure that the model can accurately restore the input data, potential distribution loss, and calculate the distribution deviation between the potential distribution and the prior distribution through KL divergence. The calculation formula is , where the prior distribution Use the mean based on offline data and the covariance matrix To define, ; Reconstruction loss ; The constraint loss term quantifies the degree of violation of the safety margin: ; The KL divergence, reconstruction loss, and constraint loss are weighted together to form a total distribution constraint loss, and the encoder and decoder parameters are optimized based on this goal. Through this mechanism, the model improves data reconstruction accuracy while maintaining the distribution rationality of the latent space and the physical feasibility of the action. The action reconstruction model uses bimodal indicators to determine whether the input action exceeds the normal range and defines ; Detection logic uses dynamic threshold mechanism to make judgments: When Score > Or there is a single physical constraint violation> When , it is determined to be an abnormal action, where and is a preset threshold, at which point the security correction mechanism is triggered to generate compliance actions .

8. A multi-roller coordinated belt deviation correction control system, characterized in that: The system comprises: A detection module is used to detect the current belt edge through visual edge detection technology; A construction module is used to compare the current belt edge with the belt edge setting value to obtain the pixel correlation point where each correction roller intersects with the standard edge line Belt edge deviation values, including offset distance and offset speed , combined with the angle of the correction roller , construct the state vector , i is the number of the correction roller, n is the total number of correction rollers; Output module for the state vector The policy network of the offline-trained multi-agent deep reinforcement learning model outputs an action, which is then combined with the state vector to form a state-action pair. This pair is then input into the offline-trained action reconstruction model, which reconstructs the action and outputs a safe action for subsequent motor control. An evaluation module, which is used to input the action value network and action cost network of the multi-agent deep reinforcement learning model trained offline into the state-action pair, evaluate the actions output by the policy network, and train the policy network, action value network, and action cost network online; The deviation correction control module is used to process the output safety action and convert it into an electrical signal which is input into the servo motor driver. The driver controls the rotation of the servo motors of each deviation correction roller to perform belt deviation correction control. The multi-agent deep reinforcement learning model consists of three networks: an action value network, an action cost network, and a policy network. All three networks take a sequence of [state, action] trajectories as input, extract corresponding feature vectors through a spatiotemporal graph attention mechanism, and finally output corresponding targets through a fully connected layer. The core of the spatiotemporal graph attention mechanism is to establish an adaptive modeling method that can simultaneously capture spatial correlation and temporal dynamics. It consists of multiple spatiotemporal graph attention layers connected by residuals, and the hidden feature vectors are continuously passed to the deep network. Each spatiotemporal graph attention layer first models the multi-roll system augmented data as a graph structure ,node Indicates the The attribute vector of the correction roller, including the angle of the correction roller , association points , the offset distance between the associated point and the standard edge line and offset speed ,In order to effectively extract features, all attribute vectors are normalized to the range of [0, 1].,These features are stored in the form of time series, ,side Indicates the relative distance between each correction roller; Then calculate the spatial adjacency matrix; Then the spatial attention mechanism is calculated to dynamically calculate the node influence through the cross-dimensional interaction of the triple weight matrix; Then calculate the temporal attention mechanism; Then the spatiotemporal features are fused through the spatiotemporal feature fusion module; Finally, multiple spatiotemporal graph attention layers optimize feature extraction capabilities through residual connections, enhancing the depth and expressiveness of the model.

9. An electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, characterized in that: The at least one instruction is loaded and executed by the processor to implement the multi-roller coordinated belt deviation correction control method as described in any one of claims 1-7.

10. A computer-readable storage medium, wherein at least one instruction is stored in the storage medium, characterized in that: The at least one instruction is loaded and executed by the processor to implement the multi-roller coordinated belt deviation correction control method as described in any one of claims 1-7.

Images