Construction and scheduling method of LED large screen splicing visual management and control platform

Abstract

This invention pertains to the field of display control technology, specifically the construction and scheduling method of an LED large-screen splicing visualization management and control platform. It addresses the problems of poor splicing synchronization, insufficient color consistency, and inefficient multi-source signal scheduling in traditional LED large-screen management and control. The invention collects hardware, environmental, and multi-source signal parameters, inputs them into an initial policy gradient model to generate a first combination of scheduling control parameters, constructs a Markov jump process control model and solves the Kolmogorov forward equation to generate system state change data, generates a second combination of scheduling control parameters based on model predictive control, solves the system steady-state probability distribution, and calculates core performance indicators such as color consistency deviation, signal switching delay, and load balancing, and obtains a target depth deterministic policy gradient model through a particle swarm optimization algorithm to generate the optimal target scheduling control parameter combination. This invention achieves intelligent management and control of LED large-screen splicing systems, enhancing display quality.

Description

Technical Field

[0001] This invention pertains to the field of display control technology, specifically to the construction and scheduling method of an LED large screen splicing visualization management and control platform. Background Technology

[0002] With the rapid iteration of LED display technology, ultra-large-scale splicing screens have become a key infrastructure for command and dispatch, exhibition and display and smart city core scenarios, and are developing towards ultra-high definition and splicing of hundreds of screens, which puts forward requirements for display accuracy and response speed. Current mainstream LED screen management solutions still rely mainly on manual screen-by-screen debugging and automated control with fixed rules. Clock drift and load differences among multiple units can easily lead to tearing and stuttering in high-speed dynamic images. Furthermore, differences in the light decay characteristics and temperature response of LED chips in different splicing units can cause obvious color block unevenness in the image after long-term operation. It is also impossible to dynamically allocate computing and display resources according to the priority of signal content and the real-time load of the system, which can easily lead to problems such as signal switching delay and frame loss.

[0003] Therefore, a method for constructing and scheduling a visual management and control platform for LED large screen splicing is needed to solve the above problems. Summary of the Invention

[0004] To address the technical problems raised in the background, this invention provides a method for constructing and scheduling an LED large screen splicing visualization management and control platform.

[0005] The objective of this invention can be achieved through the following technical solutions: This invention provides a method for constructing and scheduling a visual management and control platform for LED large screen splicing, comprising the following steps: Step 1: Obtain the hardware parameter set, environmental parameter set, and multi-source signal parameter set of the LED large screen splicing system, and input the parameter set into the preset initial strategy gradient model for scheduling control analysis to generate the first scheduling control parameter combination; Step 2: Create a Markov jump process control model for the LED screen splicing system based on the first combination of scheduling control parameters, and generate corresponding system state change data through the Kolmogorov forward equation; Step 3: Based on the preset model predictive control model, perform control parameter combination analysis on the state change data of the LED large screen splicing system to obtain the second scheduling control parameter combination; Step 4: Real-time control of the LED screen splicing system is achieved through the combination of the second scheduling control parameters, and the steady-state probability distribution of the system is solved to obtain complete steady-state probability distribution data. Based on the steady-state probability distribution data, the control performance of the LED screen splicing system is evaluated to obtain multiple system performance evaluation indicators. Step 5: Optimize the model parameters of the initial strategy gradient model based on the system performance evaluation vector to obtain the target depth deterministic strategy gradient model. Then, perform fault handling analysis and parameter combination prediction on the LED screen splicing system through the target model to obtain the target scheduling control parameter combination. In this application, based on step one, the hardware parameter set, environmental parameter set, and multi-source signal parameter set of the LED large screen splicing system are obtained, and the parameter set is input into a preset initial strategy gradient model for scheduling control analysis to generate a first scheduling control parameter combination. The specific steps are as follows: The main controller of the LED screen establishes communication with the control chips of each splicing unit via the SNMPv3 protocol. It reads factory-fixed parameters and real-time configurable parameters unit by unit to construct a hardware parameter set, including physical, communication, and electrical parameters. A distributed sensor network is then deployed around the LED screen to collect environmental parameters in real time. This network includes light and temperature / humidity sensors, and the environmental parameters include ambient light intensity, ambient temperature, and air humidity. These parameters are transmitted to the edge computing nodes of the management platform via an RS485 bus. After noise removal through sliding window filtering, a standardized environmental parameter set is generated. Finally, all incoming signals are analyzed in real time to obtain a multi-source signal parameter set, which includes technical and business parameters. The three types of parameters collected are normalized by min-max to map all parameters to the [0,1] interval. For discrete parameters, one-hot encoding is used to generate a standardized input vector. The standardized input vector is input into the pre-trained initial policy gradient model, which consists of an actor network and a critic network. The actor network outputs initial scheduling parameters based on the standardized input vector. The critic network calculates the Q value of the action. If the Q value is greater than a preset Q threshold, the parameter is retained; otherwise, it is regenerated until the first set of scheduling parameters that meets the threshold requirement is output. The first set of scheduling parameters includes global control parameters, single-screen control parameters, and signal scheduling parameters.

[0006] In this application, based on step two, a Markov jump process control model for the LED screen is created according to the first combination of scheduling control parameters, and the corresponding system state change data is generated through the Kolmogorov forward equation. The specific steps are as follows: Based on the fault evolution law of the LED large screen splicing system and the operational constraints of the first scheduling control parameters, the discrete state of the system is divided into four mutually exclusive and complete operating states, including normal operation, minor fault, moderate fault, and severe fault. Based on the historical operational fault data of this type of LED large screen, several complete state transition records are extracted, and the initial transition rate is calculated using the maximum likelihood estimation method. The total dwell time of the system in each operating state and the total number of transitions from one state to another are statistically analyzed. The total number of transitions is divided by the total dwell time to obtain the base transition rate. Based on the parameters of the first scheduling control parameters, brightness correction factors, temperature correction factors, and refresh rate correction factors are constructed respectively. The base transition rate is corrected by the brightness correction factors, temperature correction factors, and refresh rate correction factors to obtain the actual transition rate matrix matched by the current control parameters. The Kolmogorov forward equation is established through the actual transition rate matrix, and its calculation logic is as follows: Where cs is the instantaneous rate of change of the actual transfer rate matrix, and t is the time point. The first derivative of the instantaneous rate of change with respect to time, QS, is the state transition vector. The state transition vector is obtained by numerically solving the Kolmogorov forward equation using the fourth-order Runge-Kutta method. The state transition vector is used to generate dwell time by inverse transformation. When the system finishes the state dwell time, a uniform random number in the interval [0,1] is generated. If the random number is less than the actual transition rate, the system transitions to the target state. The above process is repeated to generate a complete state change sequence, including state identifier, start time, end time, dwell time, system control parameters at the transition time, and transition reason.

[0007] In this application, based on step three, the control parameter combination analysis is performed on the state change data of the LED large screen splicing system based on the preset model predictive control model to obtain the second scheduling control parameter combination. The specific steps are as follows: The actual display brightness, actual output gain of the red / green / blue three channels, single-screen image synchronization offset, single-screen internal operating temperature, single-screen control board CPU load, single-screen graphics processor load, single-screen power module output voltage, driver chip operating current, and average pixel failure rate are integrated into single-screen local state variables. The maximum splicing synchronization error, average color deviation between all splicing units, and total power consumption of the entire large-screen system are integrated into global system state variables. The global system state variables and single-screen local state variables are combined into system state variables. Based on the system state variables... and control input vector The state equations are established, and their calculation logic is as follows: Where f is a nonlinear state transition function, This is the process noise vector. The system state variables for the next time step are then used in conjunction with the system state variables and the noise vector. The observation equation is established, and its calculation logic is as follows: ,in For nonlinear observation functions, The observed output vector is used; a nonlinear state space is constructed based on the state equation and the observation equation. The initial value of the state mean is taken as the steady-state operating value of the system corresponding to the first scheduling control parameter. The covariance matrix is ​​initialized as a diagonal matrix, with the diagonal elements representing the initial uncertainties of the corresponding state variables. Based on the current state mean and covariance matrix, a set of sampling points of the state probability distribution are generated in the nonlinear state space and marked as Sigma points. Each Sigma point is substituted into the nonlinear state equation to calculate the corresponding predicted Sigma point. Based on the weight of each Sigma point, a weighted average is performed on all predicted Sigma points to obtain the predicted state mean at the next time step. The deviation from the predicted state mean is used to calculate the predicted covariance matrix for the next time step. The predicted Sigma points are propagated through a nonlinear state space to obtain the predicted observation Sigma points. The predicted observation Sigma points are weighted and averaged to obtain the predicted observation mean. The predicted observation covariance matrix and the state-observation cross-covariance matrix are calculated. The Kalman gain is calculated based on the predicted observation covariance matrix and the cross-covariance matrix. The deviation of the Kalman gain from the actual measured value and the predicted observation value is used to correct the predicted state mean and the predicted covariance matrix to obtain the optimal state estimate and the optimal covariance matrix for the current time step. The current optimal state estimate and optimal covariance matrix are input into the finite-time rolling optimization model predictive controller. The model predictive controller establishes a quadratic programming problem, which is solved by an embedded real-time quadratic programming solver to obtain the optimal control sequence. The first control quantity of the optimal control sequence is extracted and converted into control instructions and register values ​​that can be recognized by the LED screen hardware to obtain the second scheduling control parameter combination, including global control parameters, single-screen control parameters and signal scheduling parameters.

[0008] In this application, based on step four, the LED large screen splicing system is controlled in real time through the second scheduling control parameter combination, and the steady-state probability distribution of the system is solved to obtain complete steady-state probability distribution data. The control performance of the LED large screen splicing system is then evaluated based on the steady-state probability distribution data to obtain multiple system performance evaluation indicators. The specific steps are as follows: Initially, the system is assumed to have a probability of 1 in the normal state and a probability of 0 in other states. The current state probability vector is multiplied by the one-step transition probability matrix to obtain the state probability vector at the next time step. This multiplication process is repeated until the difference between the two state probability vectors obtained is less than the preset convergence threshold, thus obtaining a stationary distribution of state probabilities. The actual transfer rate matrix is ​​decomposed into two parts: absorption state and transient state. Severe fault is the absorption state, and normal operation, mild fault and moderate fault are the transient states. The probability of the system starting from each transient state and absorbing the severe fault state is obtained to obtain the absorption probability. Then, the average time of the system starting from each transient state and absorbing the severe fault state is obtained. The absorption probability and the average time are integrated into steady-state probability distribution data. The RGB color values ​​of the splicing units are converted to CIE LAB color values. The color difference between any two adjacent splicing units is calculated. The maximum value of the color difference between all adjacent units is taken as the global color consistency deviation at that moment. The global color consistency deviation at all moments is classified according to the system state. The average color consistency deviation is calculated for the system under normal operation, minor fault, and moderate fault states, respectively, to obtain the sub-state average color deviation for each state. The sub-state average color deviation for each state is multiplied by the steady-state probability corresponding to that state. The results of all states are added together to obtain the steady-state average color consistency deviation of the system. The timestamp of the switching command issued by the control platform, the timestamp of the signal switching matrix completing the signal routing switch, and the timestamp of the large screen reaching a stable state are integrated into a signal switching event. All signal switching events are classified according to the state of the system when the event occurs. The average signal switching delay is calculated for the system under normal operation, minor fault, and moderate fault states, respectively. The delay time of each switching event is the screen stabilization timestamp minus the command issuance timestamp. The sub-state average signal switching delay for each state is multiplied by the steady-state probability corresponding to that state. The results of all states are added together to obtain the steady-state average signal switching delay of the system. The CPU utilization and GPU utilization of each splicing unit are collected, and the maximum value of the two is taken as the comprehensive resource utilization of the splicing unit at that moment. The comprehensive resource utilization of each splicing unit at all moments is classified according to the system state. For each state, the standard deviation of the resource utilization of each splicing unit at all moments under that state is calculated. The arithmetic mean of the standard deviations is taken to obtain the sub-state average load balance of that state. The sub-state average load balance of each state is multiplied by the steady-state probability corresponding to that state. The results of all states are added together to obtain the steady-state average load balance of the system. The steady-state average color consistency deviation, steady-state average signal switching delay, and steady-state average load balancing degree are min-max normalized and integrated to obtain the system performance evaluation vector.

[0009] In this application, based on step five, the initial strategy gradient model is optimized according to the system performance evaluation vector to obtain a target depth deterministic strategy gradient model. Then, the target model is used to perform fault handling analysis and parameter combination prediction on the LED large screen splicing system to obtain the target scheduling control parameter combination. The specific steps are as follows: The system performance evaluation vector is input into the trained anomaly detection model. Each decision tree in the model analyzes the input vector and outputs a preliminary anomaly score and anomaly type. The output results of all decision trees are combined through weighted voting to obtain the target device anomaly evaluation result. Based on the abnormal assessment results, the weights of the five performance indicators in the optimization function are dynamically adjusted, and the steady-state average color consistency deviation, steady-state average signal switching delay and steady-state average load balance are weighted and summed to obtain the overall system performance score. All network weight parameters of the initial policy gradient model are encoded into high-dimensional vectors, which are used as particles in the particle swarm optimization algorithm. Within a reasonable parameter range, a particle swarm is randomly generated. The comprehensive performance score corresponding to each particle is calculated, representing the overall performance of the system under the model parameters represented by that particle. The individual optimal position of each particle is updated, which is the parameter position corresponding to the lowest comprehensive performance score historically obtained by that particle. Simultaneously, the global optimal position of the entire particle swarm is updated, which is the parameter position corresponding to the lowest comprehensive performance score historically obtained by all particles. Then, based on the individual optimal and global optimal positions, the velocity and position of each particle are updated. This process is repeated until the maximum number of iterations is reached, and the target model parameter set is output. The target model parameter set is input into the initial policy gradient model for model parameter optimization to obtain the target depth deterministic policy gradient model. The target depth deterministic policy gradient model takes the collected real-time state data as input and outputs the optimal scheduling control parameter combination in real time through the actor network and sends it to the local controller of each splicing unit.

[0010] Compared with the prior art, the beneficial effects of the present invention are: A Markov jump process control model with dynamically bound control parameters is constructed. The system state evolution law is numerically solved using the Kolmogorov forward equation, overcoming the limitation of traditional static models in describing random faults and performance degradation. A performance evaluation system based on steady-state probability distribution deeply integrates indicators such as color consistency deviation, signal switching delay, and load balancing with the system's long-term operating state, achieving a comprehensive quantitative evaluation from short-term instantaneous performance to long-term steady-state characteristics, providing a scientific basis for control strategy optimization. Model predictive control solves constrained multi-objective optimization problems in the finite time domain, achieving coordinated optimization of stitching synchronization, color uniformity, and system power consumption, effectively solving the control conflict problem caused by multivariable coupling. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. The following drawings are not drawn to scale according to the actual size, but are intended to illustrate the main idea of ​​the present invention.

[0012] Figure 1 This is a diagram illustrating the method steps of the present invention. Detailed Implementation

[0013] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are also within the scope of protection of the present invention.

[0014] Please refer to Figure 1 As shown, this invention provides a method for constructing and scheduling an LED large screen splicing visualization management and control platform, including the following steps: Step 1: Obtain the hardware parameter set, environmental parameter set, and multi-source signal parameter set of the LED large screen splicing system, and input the parameter set into the preset initial strategy gradient model for scheduling control analysis to generate the first scheduling control parameter combination; Step 2: Create a Markov jump process control model for the LED screen splicing system based on the first combination of scheduling control parameters, and generate corresponding system state change data through the Kolmogorov forward equation; Step 3: Based on the preset model predictive control model, perform control parameter combination analysis on the state change data of the LED large screen splicing system to obtain the second scheduling control parameter combination; Step 4: Real-time control of the LED screen splicing system is achieved through the combination of the second scheduling control parameters, and the steady-state probability distribution of the system is solved to obtain complete steady-state probability distribution data. Based on the steady-state probability distribution data, the control performance of the LED screen splicing system is evaluated to obtain multiple system performance evaluation indicators. Step 5: Optimize the model parameters of the initial strategy gradient model based on the system performance evaluation vector to obtain the target depth deterministic strategy gradient model. Then, perform fault handling analysis and parameter combination prediction on the LED screen splicing system through the target model to obtain the target scheduling control parameter combination. In this application, based on step one, the hardware parameter set, environmental parameter set, and multi-source signal parameter set of the LED large screen splicing system are obtained, and the parameter set is input into a preset initial strategy gradient model for scheduling control analysis to generate a first scheduling control parameter combination. The specific steps are as follows: The main controller of the LED screen establishes communication with the control chips of each splicing unit via the SNMPv3 protocol. It reads factory-fixed parameters and real-time configurable parameters unit by unit to construct a hardware parameter set, including physical, communication, and electrical parameters. A distributed sensor network is deployed around the LED screen to collect environmental parameters in real time. This network includes light and temperature / humidity sensors. Environmental parameters include ambient light intensity, ambient temperature, and air humidity. These parameters are transmitted to the edge computing nodes of the management platform via an RS485 bus. After noise removal through sliding window filtering, a standardized environmental parameter set is generated. All access signals are analyzed in real time to obtain a multi-source signal parameter set, including technical and business parameters. Business parameters include signal source number, access port number, user-preset content priority and content type. User-preset content priority is 1-5 levels, with level 1 being the highest. Content types include command and dispatch, video surveillance, publicity display, and data visualization. Technical parameters include signal resolution, frame rate, bit rate, color space, and bit depth. The three types of parameters collected are normalized by min-max to map all parameters to the [0,1] interval. For discrete parameters, one-hot encoding is used to generate a standardized input vector. The standardized input vector is input into a pre-trained initial policy gradient model, which consists of an actor network and a critic network. The actor network outputs initial scheduling parameters based on the standardized input vector, and the critic network calculates the Q-value of the action. If the Q-value is greater than a preset Q-threshold, the parameter is retained; otherwise, it is regenerated until a first set of scheduling parameters that meets the threshold requirement is output. The first set of scheduling parameters includes global control parameters, single-screen control parameters, and signal scheduling parameters. It should be noted that the global control parameters are the global synchronization offset, the overall brightness coefficient, and the global color correction matrix; the single-screen control parameters are the reference brightness, R / G / B three-channel gain, and gamma correction value for each splicing unit; and the signal scheduling parameters are the image splicing coordinates, frame rate conversion parameters, and bit rate control parameters. The first set of scheduling control parameters is combined and written into the real-time cache of the management platform.

[0015] In this application, based on step two, a Markov jump process control model for the LED screen is created according to the first combination of scheduling control parameters, and the corresponding system state change data is generated through the Kolmogorov forward equation. The specific steps are as follows: Based on the fault evolution law of the LED large screen splicing system and the operational constraints of the first scheduling control parameters, the discrete state of the system is divided into four mutually exclusive and complete operating states, including normal operation, minor fault, moderate fault, and severe fault. Based on the historical operational fault data of this type of LED large screen, several complete state transition records are extracted, and the initial transition rate is calculated using the maximum likelihood estimation method. The total dwell time of the system in each operating state and the total number of transitions from one state to another are statistically analyzed. The total number of transitions is divided by the total dwell time to obtain the base transition rate. Based on the parameters of the first scheduling control parameters, brightness correction factors, temperature correction factors, and refresh rate correction factors are constructed respectively. The base transition rate is corrected by the brightness correction factors, temperature correction factors, and refresh rate correction factors to obtain the actual transition rate matrix matched by the current control parameters. The Kolmogorov forward equation is established through the actual transition rate matrix, and its calculation logic is as follows: Where cs is the instantaneous rate of change of the actual transfer rate matrix, and t is the time point. The first derivative of the instantaneous rate of change with respect to time, QS, is the state transition vector. The state transition vector is obtained by numerically solving the Kolmogorov forward equation using the fourth-order Runge-Kutta method. The state transition vector is used to generate dwell time using the inverse transformation method. After the system finishes the state dwell time, a uniform random number in the interval [0,1] is generated. If the random number is less than the actual transition rate, the system transitions to the target state. The above process is repeated to generate a complete state change sequence, including state identifier, start time, end time, dwell time, system control parameters at the transition time, and transition reason. It should be noted that transition reasons include LED light decay, driver circuit failure, power supply failure, and excessively high ambient temperature.

[0016] In this application, based on step three, the control parameter combination analysis is performed on the state change data of the LED large screen splicing system based on the preset model predictive control model to obtain the second scheduling control parameter combination. The specific steps are as follows: The actual display brightness, actual output gain of the red / green / blue three channels, single-screen image synchronization offset, single-screen internal operating temperature, single-screen control board CPU load, single-screen graphics processor load, single-screen power module output voltage, driver chip operating current, and average pixel failure rate are integrated into single-screen local state variables. The maximum splicing synchronization error, average color deviation between all splicing units, and total power consumption of the entire large-screen system are integrated into global system state variables. The global system state variables and single-screen local state variables are combined into system state variables. Based on the system state variables... and control input vector The state equations are established, and their calculation logic is as follows: Where f is a nonlinear state transition function, This is the process noise vector. The system state variables for the next time step are then used in conjunction with the system state variables and the noise vector. The observation equation is established, and its calculation logic is as follows: ,in It is a nonlinear observation function that describes the mapping relationship between state variables and measured values. The observed output vector is used; a nonlinear state space is constructed based on the state equation and the observation equation. The initial value of the state mean is taken as the steady-state operating value of the system corresponding to the first scheduling control parameter. The covariance matrix is ​​initialized as a diagonal matrix, with the diagonal elements representing the initial uncertainties of the corresponding state variables. Based on the current state mean and covariance matrix, a set of sampling points of the state probability distribution are generated in the nonlinear state space and marked as Sigma points. Each Sigma point is substituted into the nonlinear state equation to calculate the corresponding predicted Sigma point. Based on the weight of each Sigma point, a weighted average is performed on all predicted Sigma points to obtain the predicted state mean at the next time step. The deviation from the predicted state mean is used to calculate the predicted covariance matrix for the next time step. The predicted Sigma points are propagated through a nonlinear state space to obtain the predicted observation Sigma points. The predicted observation Sigma points are weighted and averaged to obtain the predicted observation mean. The predicted observation covariance matrix and the state-observation cross-covariance matrix are calculated. The Kalman gain is calculated based on the predicted observation covariance matrix and the cross-covariance matrix. The deviation of the Kalman gain from the actual measured value and the predicted observation value is used to correct the predicted state mean and the predicted covariance matrix to obtain the optimal state estimate and the optimal covariance matrix for the current time step. The current optimal state estimate and optimal covariance matrix are input into the finite-time domain rolling optimization model predictive controller. The model predictive controller establishes a quadratic programming problem, which is solved by an embedded real-time quadratic programming solver to obtain the optimal control sequence. The first control variable of the optimal control sequence is extracted and converted into control instructions and register values ​​that the LED screen hardware can recognize, resulting in the second combination of scheduling control parameters, including global control parameters, single-screen control parameters, and signal scheduling parameters. It should be noted that the quadratic programming problem is transformed by the objective function. The first part of the objective function is the state tracking error term, which aims to make the system state track the reference state as closely as possible. The reference state is determined by the first scheduling control parameters. This part assigns different weights to different state variables, with the highest weights for splicing synchronization error and color consistency deviation, as they directly affect the display effect; the next highest weight is for total system power consumption, used to control energy consumption; the remaining state variables have lower weights to ensure the overall stability of the system. The second part of the objective function is the control variable increment penalty term, which aims to limit the range of change in the control variable and avoid system oscillations and hardware damage caused by sudden changes in the control variable. This section assigns different weights to different control variables, with the increment of synchronization offset having the highest weight because abrupt changes in synchronization offset can cause screen tearing; the increments of RGB channel gain and brightness have the next highest weights; the increments of other control variables have lower weights. The third part of the objective function is a relaxation factor term, used to handle hard constraint conflicts. When the optimization problem has constraint conflicts that lead to no solution, a relaxation factor is introduced to relax some non-core constraints, ensuring that the optimization problem always has a solution. The penalty coefficient of the relaxation factor is set to a very large value to ensure that constraints are relaxed only when necessary.

[0017] In this application, based on step four, the LED large screen splicing system is controlled in real time through the second scheduling control parameter combination, and the steady-state probability distribution of the system is solved to obtain complete steady-state probability distribution data. The control performance of the LED large screen splicing system is then evaluated based on the steady-state probability distribution data to obtain multiple system performance evaluation indicators. The specific steps are as follows: Initially, the system is assumed to have a probability of 1 in the normal state and a probability of 0 in other states. The current state probability vector is multiplied by the one-step transition probability matrix to obtain the state probability vector at the next time step. This multiplication process is repeated until the difference between the two state probability vectors obtained is less than the preset convergence threshold, thus obtaining a stationary distribution of state probabilities. The actual transfer rate matrix is ​​decomposed into two parts: absorption state and transient state. Severe fault is the absorption state, and normal operation, mild fault and moderate fault are the transient states. The probability of the system starting from each transient state and absorbing the severe fault state is obtained to obtain the absorption probability. Then, the average time of the system starting from each transient state and absorbing the severe fault state is obtained. The absorption probability and the average time are integrated into steady-state probability distribution data. The RGB color values ​​of the splicing units are converted to CIE LAB color values. The color difference between any two adjacent splicing units is calculated. The maximum value of the color difference between all adjacent units is taken as the global color consistency deviation at that moment. The global color consistency deviation at all moments is classified according to the system state. The average color consistency deviation is calculated for the system under normal operation, minor fault, and moderate fault states, respectively. The sub-state average color deviation for each state is obtained. The sub-state average color deviation for each state is multiplied by the steady-state probability corresponding to that state. The results of all states are added together to obtain the steady-state average color consistency deviation of the system. The timestamps of the switching command issued by the control platform, the signal switching matrix completing the signal routing switch, and the large screen reaching a stable state are integrated into a signal switching event. All signal switching events are classified according to the state of the system when the event occurs. The average signal switching delay is calculated for the system in normal operation, minor fault, and moderate fault states. The delay time of each switching event is the screen stabilization timestamp minus the command issuance timestamp. The average signal switching delay of each state is multiplied by the steady-state probability corresponding to that state. The results of all states are added together to obtain the steady-state average signal switching delay of the system. The CPU utilization and GPU utilization of each splicing unit are collected, and the maximum value of the two is taken as the comprehensive resource utilization of the splicing unit at that moment. The comprehensive resource utilization of each splicing unit at all moments is classified according to the system state. For each state, the standard deviation of the resource utilization of each splicing unit at all moments under that state is calculated. The arithmetic mean of the standard deviations is taken to obtain the sub-state average load balance of that state. The sub-state average load balance of each state is multiplied by the steady-state probability corresponding to that state. The results of all states are added together to obtain the steady-state average load balance of the system. The steady-state average color consistency deviation, steady-state average signal switching delay, and steady-state average load balancing degree are min-max normalized and integrated to obtain the system performance evaluation vector.

[0018] In this application, based on step five, the initial strategy gradient model is optimized according to the system performance evaluation vector to obtain a target depth deterministic strategy gradient model. Then, the target model is used to perform fault handling analysis and parameter combination prediction on the LED large screen splicing system to obtain the target scheduling control parameter combination. The specific steps are as follows: The system performance evaluation vector is input into the trained anomaly detection model. Each decision tree in the model analyzes the input vector and outputs a preliminary anomaly score and anomaly type. The outputs of all decision trees are combined through weighted voting to obtain the target device anomaly assessment result. It should be noted that the model is pre-trained before system deployment. The training dataset contains historical operating data for this model of LED screen, covering performance evaluation vector samples under normal operating conditions and various abnormal operating conditions. The model consists of 100 classification and regression trees. Through gradient boosting ensemble learning, each new decision tree fits the residuals of all previous decision trees, gradually improving the model's prediction accuracy. Based on the abnormal assessment results, the weights of the five performance indicators in the optimization function are dynamically adjusted, and the steady-state average color consistency deviation, steady-state average signal switching delay and steady-state average load balance are weighted and summed to obtain the overall system performance score. All network weight parameters of the initial policy gradient model are encoded into high-dimensional vectors, which are used as particles in the particle swarm optimization algorithm. Within a reasonable parameter range, a particle swarm is randomly generated. The comprehensive performance score corresponding to each particle is calculated, representing the overall performance of the system under the model parameters represented by that particle. The individual optimal position of each particle is updated, which is the parameter position corresponding to the lowest comprehensive performance score historically obtained by that particle. Simultaneously, the global optimal position of the entire particle swarm is updated, which is the parameter position corresponding to the lowest comprehensive performance score historically obtained by all particles. Then, based on the individual optimal and global optimal positions, the velocity and position of each particle are updated. This process is repeated until the maximum number of iterations is reached, and the target model parameter set is output. The target model parameter set is input into the initial policy gradient model for model parameter optimization to obtain the target depth deterministic policy gradient model. The target depth deterministic policy gradient model takes the collected real-time state data as input and outputs the optimal scheduling control parameter combination in real time through the actor network and sends it to the local controller of each splicing unit.

[0019] All the above formulas use dimensionless numerical calculations, which can be achieved through standardization, normalization, etc., and will not be elaborated here. The relevant expressions are obtained by collecting a large amount of full life cycle operating data of LED large screen splicing systems of different models and scales, covering various hardware configurations, environmental conditions, signal types and business scenarios, and are obtained through simulation fitting, which closely reflects the actual engineering application state. The preset parameters in the formulas can be set and calibrated by those skilled in the art according to the LED large screen model, the number of splicing units, the application scenario and business requirements.

[0020] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented as a computer program product, which contains one or more computer-executable instructions. When the instructions are loaded and run on a computing device, they fully or partially implement the processes and functions described in this invention. The device includes an LED splicing unit controller, an edge computing node, a large-screen main controller, or a cloud management platform. The instructions can be stored on a computer-readable storage medium or transmitted between devices via wired Ethernet, 5G / 4G wireless, etc. The readable storage medium includes industrial-grade storage media such as solid-state drives, USB flash drives, memory cards, disks, and optical discs.

[0021] The execution order of each process in this invention is based on functional and logical relationships, and is not limited by the number of the sequence. Those skilled in the art will understand that the units and algorithm steps described in this invention can be implemented using electronic hardware or a combination of hardware and software, and appropriate implementation methods can be selected for specific application scenarios, all of which are within the scope of protection of this invention.

[0022] The methods and systems disclosed in this invention can also be implemented in other non-illustrative ways. The unit division is only a logical functional division; other division methods can be used in practice. Multiple units or components can be combined or integrated, and some features can be omitted. The coupling and connection between units can be indirect, and can be electrical, network communication, or other forms.

[0023] The units described as separate components may or may not be physically separated. The components shown as units may be physical units or distributed across multiple network units. Some or all units can be selected to achieve the purpose of the solution according to actual control requirements.

[0024] If the functions of this invention are implemented as software functional units and sold or used as a product, they can be stored in a computer-readable storage medium. This software product includes several instructions for driving an LED splicing unit controller, main controller, edge computing gateway, or cloud server to execute all or part of the steps of this invention. The storage medium includes media capable of storing program code, such as solid-state drives, flash drives, portable hard drives, read-only memory, and random access memory.

[0025] The above description is merely a specific embodiment of the present invention, but the scope of protection is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this invention should be included within the scope of protection. Therefore, the scope of protection of this invention is determined by the claims.

Claims

1. A method for constructing and scheduling a visual management and control platform for LED large screen splicing, characterized in that, Includes the following steps: Step 1: Obtain the hardware parameter set, environmental parameter set, and multi-source signal parameter set of the LED large screen splicing system, and input the parameter set into the preset initial strategy gradient model for scheduling control analysis to generate the first scheduling control parameter combination; Step 2: Create a Markov jump process control model for the LED screen splicing system based on the first combination of scheduling control parameters, and generate corresponding system state change data through the Kolmogorov forward equation; Step 3: Based on the preset model predictive control model, perform control parameter combination analysis on the state change data of the LED large screen splicing system to obtain the second scheduling control parameter combination; Step 4: Real-time control of the LED screen splicing system is achieved through the combination of the second scheduling control parameters, and the steady-state probability distribution of the system is solved to obtain complete steady-state probability distribution data. Based on the steady-state probability distribution data, the control performance of the LED screen splicing system is evaluated to obtain multiple system performance evaluation indicators. Step 5: Optimize the model parameters of the initial strategy gradient model based on the system performance evaluation vector to obtain the target depth deterministic strategy gradient model. Then, perform fault handling analysis and parameter combination prediction on the LED screen splicing system through the target model to obtain the target scheduling control parameter combination.

2. The method for constructing and scheduling an LED large screen splicing visualization management and control platform according to claim 1, characterized in that, In step one, the hardware parameter set, environmental parameter set, and multi-source signal parameter set of the LED large screen splicing system are obtained, and the parameter set is input into a preset initial strategy gradient model for scheduling control analysis to generate the first scheduling control parameter combination. The specific steps are as follows: The main controller of the LED screen establishes communication with the control chips of each splicing unit through the SNMPv3 protocol. It reads the factory-fixed parameters and real-time configurable parameters of each unit to build a set of hardware parameters, including physical parameters, communication parameters and electrical parameters. Then, by deploying a distributed sensor network around the LED screen, the environmental parameters of the LED screen are collected in real time. The distributed sensor network includes light sensors and temperature and humidity sensors. The environmental parameters include ambient light intensity parameters, ambient temperature and air humidity parameters. Environmental parameters are transmitted to the edge computing node of the management and control platform via RS485 bus. After noise is removed by sliding window filtering, a standardized set of environmental parameters is generated. By analyzing all access signals in real time, a set of multi-source signal parameters is obtained, which includes technical parameters and business parameters. The three types of parameters collected are normalized by min-max to map all parameters to the [0,1] interval. For discrete parameters, one-hot encoding is used to generate a standardized input vector. The standardized input vector is input into the pre-trained initial policy gradient model, which consists of an actor network and a critic network. The actor network outputs initial scheduling parameters based on the standardized input vector. The critic network calculates the Q value of the action. If the Q value is greater than a preset Q threshold, the parameter is retained; otherwise, it is regenerated until the first set of scheduling parameters that meets the threshold requirement is output. The first set of scheduling parameters includes global control parameters, single-screen control parameters, and signal scheduling parameters.

3. The method for constructing and scheduling an LED large screen splicing visualization control platform according to claim 1, characterized in that, In step two, a Markov jump process control model for the LED screen is created based on the first combination of scheduling control parameters, and the corresponding system state change data is generated through the Kolmogorov forward equation. The specific steps are as follows: Based on the fault evolution law of the LED large screen splicing system and the operational constraints of the first scheduling control parameter, the discrete state of the system is divided into four mutually exclusive and complete operating states: normal operation, minor fault, moderate fault, and severe fault. Based on the historical operational fault data of this type of LED large screen, several complete state transition records are extracted. The initial transition rate is calculated using the maximum likelihood estimation method. The total dwell time of the system in each operating state and the total number of transitions from one state to another are statistically analyzed. The total number of transitions is divided by the total dwell time to obtain the basic transition rate. Based on the parameters of the first scheduling control parameter, brightness correction factors, temperature correction factors, and refresh rate correction factors are constructed respectively. The base transfer rate is corrected by the degree correction factor, temperature correction factor, and refresh rate correction factor to obtain the actual transfer rate matrix matching the current control parameters. The Kolmogorov forward equation is established through the actual transfer rate matrix, and the Kolmogorov forward equation is numerically solved by the fourth-order Runge-Kutta method to obtain the state transition vector. The dwell time is generated from the state transition vector by the inverse transformation method. When the system finishes the state dwell time, a uniform random number in the interval [0,1] is generated. If the random number is less than the actual transfer rate, the system transitions to the target state. The above process is repeated to generate a complete state change sequence, including the state identifier, start time, end time, dwell time, system control parameters at the time of transition, and reason for transition.

4. The method for constructing and scheduling an LED large screen splicing visualization control platform according to claim 1, characterized in that, In step three, the optimal state estimate and the optimal covariance matrix are obtained. The specific steps are as follows: The actual display brightness, actual output gain of the red / green / blue three channels, single-screen image synchronization offset, single-screen internal operating temperature, single-screen control board CPU load, single-screen graphics processor load, single-screen power module output voltage, driver chip operating current, and average pixel failure rate are integrated into single-screen local state variables. The maximum splicing synchronization error, the average color deviation between all splicing units, and the total power consumption of the entire large-screen system are integrated into global system state variables. The global system state variables and single-screen local state variables are combined into system state variables. A state equation is established based on the system state variables and control input vectors. Then, an observation equation is established through the system state variables and noise vectors. A nonlinear state space is constructed based on the state equations and observation equations. The initial value of the state mean is taken as the steady-state operating value of the system corresponding to the first scheduling control parameter. The covariance matrix is ​​initialized as a diagonal matrix, with the diagonal elements representing the initial uncertainties of the corresponding state variables. Based on the current state mean and covariance matrix, a set of sampling points of the state probability distribution are generated in the nonlinear state space and marked as Sigma points. Each Sigma point is substituted into the nonlinear state equation to calculate the corresponding predicted Sigma point. Based on the weight of each Sigma point, a weighted average is performed on all predicted Sigma points to obtain the predicted state mean at the next time step. The deviation from the predicted state mean is used to calculate the predicted covariance matrix for the next time step. The predicted Sigma points are propagated through a nonlinear state space to obtain the predicted observation Sigma points. The predicted observation Sigma points are then weighted and averaged to obtain the predicted observation mean. The predicted observation covariance matrix and the state-observation cross-covariance matrix are calculated. The Kalman gain is calculated based on the predicted observation covariance matrix and the cross-covariance matrix. The deviation of the Kalman gain from the actual measured value and the predicted observation value is used to correct the predicted state mean and the predicted covariance matrix, thus obtaining the optimal state estimate and the optimal covariance matrix for the current time step.

5. The method for constructing and scheduling an LED large screen splicing visualization control platform according to claim 4, characterized in that, In step three, the second combination of scheduling control parameters, including global control parameters, single-screen control parameters, and signal scheduling parameters, is obtained through the model predictive controller. The specific steps are as follows: The current optimal state estimate and optimal covariance matrix are input into the finite-time rolling optimization model predictive controller. The model predictive controller establishes a quadratic programming problem, which is solved by an embedded real-time quadratic programming solver to obtain the optimal control sequence. The first control quantity of the optimal control sequence is extracted and converted into control instructions and register values ​​that can be recognized by the LED screen hardware to obtain the second scheduling control parameter combination, including global control parameters, single-screen control parameters and signal scheduling parameters.

6. The method for constructing and scheduling an LED large screen splicing visualization management and control platform according to claim 1, characterized in that, In step four, the LED screen splicing system is managed in real time through the second scheduling control parameter combination, and the steady-state probability distribution of the system is solved to obtain complete steady-state probability distribution data. The specific steps are as follows: Initially, the system is assumed to have a probability of 1 in the normal state and a probability of 0 in other states. The current state probability vector is multiplied by the one-step transition probability matrix to obtain the state probability vector at the next time step. This multiplication process is repeated until the difference between the two state probability vectors obtained is less than the preset convergence threshold, thus obtaining a stationary distribution of state probabilities. The actual transfer rate matrix is ​​decomposed into two parts: absorption state and transient state. Severe fault is the absorption state, and normal operation, mild fault and moderate fault are the transient states. The probability of the system starting from each transient state and absorbing the severe fault state is obtained to obtain the absorption probability. Then, the average time of the system starting from each transient state and absorbing the severe fault state is obtained. The absorption probability and the average time are integrated into steady-state probability distribution data.

7. The method for constructing and scheduling an LED large screen splicing visualization control platform according to claim 6, characterized in that, In step four, the control performance of the LED large screen splicing system is evaluated based on the steady-state probability distribution data to obtain multiple system performance evaluation indicators. The specific steps are as follows: The RGB color values ​​of the splicing units are converted to CIE LAB color values. The color difference between any two adjacent splicing units is calculated. The maximum value of the color difference between all adjacent units is taken as the global color consistency deviation at that moment. The global color consistency deviation at all moments is classified according to the system state. The average color consistency deviation is calculated for the system under normal operation, minor fault, and moderate fault states, respectively. The sub-state average color deviation for each state is obtained. The sub-state average color deviation for each state is multiplied by the steady-state probability corresponding to that state. The results of all states are added together to obtain the steady-state average color consistency deviation of the system. The timestamps of the switching command issued by the control platform, the signal switching matrix completing the signal routing switch, and the large screen reaching a stable state are integrated into a signal switching event. All signal switching events are classified according to the state of the system when the event occurs. The average signal switching delay is calculated for the system in normal operation, minor fault, and moderate fault states. The delay time of each switching event is the screen stabilization timestamp minus the command issuance timestamp. The average signal switching delay of each state is multiplied by the steady-state probability corresponding to that state. The results of all states are added together to obtain the steady-state average signal switching delay of the system. The CPU utilization and GPU utilization of each splicing unit are collected, and the maximum value of the two is taken as the comprehensive resource utilization of the splicing unit at that moment. The comprehensive resource utilization of each splicing unit at all moments is classified according to the system state. For each state, the standard deviation of the resource utilization of each splicing unit at all moments under that state is calculated. The arithmetic mean of the standard deviations is taken to obtain the sub-state average load balance of that state. The sub-state average load balance of each state is multiplied by the steady-state probability corresponding to that state. The results of all states are added together to obtain the steady-state average load balance of the system. The steady-state average color consistency deviation, steady-state average signal switching delay, and steady-state average load balancing degree are min-max normalized and integrated to obtain the system performance evaluation vector.

8. The method for constructing and scheduling an LED large screen splicing visualization control platform according to claim 7, characterized in that, In step five, the initial strategy gradient model is optimized based on the system performance evaluation vector to obtain a target depth deterministic strategy gradient model. Then, the target model is used to perform fault handling analysis and parameter combination prediction on the LED screen splicing system to obtain the target scheduling control parameter combination. The specific steps are as follows: The system performance evaluation vector is input into the trained anomaly detection model. Each decision tree in the model analyzes the input vector and outputs a preliminary anomaly score and anomaly type. The output results of all decision trees are combined through weighted voting to obtain the target device anomaly evaluation result. Based on the abnormal assessment results, the weights of the five performance indicators in the optimization function are dynamically adjusted, and the steady-state average color consistency deviation, steady-state average signal switching delay and steady-state average load balance are weighted and summed to obtain the overall system performance score. All network weight parameters of the initial policy gradient model are encoded into high-dimensional vectors, which are used as particles in the particle swarm optimization algorithm. Within a reasonable parameter range, a particle swarm is randomly generated. The comprehensive performance score corresponding to each particle is calculated, representing the overall performance of the system under the model parameters represented by that particle. The individual optimal position of each particle is updated, which is the parameter position corresponding to the lowest comprehensive performance score historically obtained by that particle. Simultaneously, the global optimal position of the entire particle swarm is updated, which is the parameter position corresponding to the lowest comprehensive performance score historically obtained by all particles. Then, based on the individual optimal and global optimal positions, the velocity and position of each particle are updated. This process is repeated until the maximum number of iterations is reached, and the target model parameter set is output. The target model parameter set is input into the initial policy gradient model for model parameter optimization to obtain the target depth deterministic policy gradient model. The target depth deterministic policy gradient model takes the collected real-time state data as input and outputs the optimal scheduling control parameter combination in real time through the actor network and sends it to the local controller of each splicing unit.

Images