LED light output intelligent control method based on high-frequency driving

Abstract

The invention provides an LED light output intelligent control method based on high-frequency driving, and relates to the technical field of intelligent control, and the method comprises the steps: applying a high-frequency detection pulse sequence with the same amplitude and increasing frequency to each lamp bead in an LED array, collecting the light response time constant and peak light intensity data of the lamp bead under the excitation of the detection pulse sequence, and obtaining the light response time constant and peak light intensity data of the lamp bead; and a frequency-light response feature vector matrix is constructed, and grouping management and distributed prediction control of the LED array are realized. According to the invention, the light response characteristic difference between the lamp beads is fully considered, and the accuracy and uniformity of light output control are improved; meanwhile, by establishing a distributed predictive control model, the optimal driving frequency and duty ratio parameter combination of each group can be predicted in real time, the problem of light response nonlinearity under the high-frequency driving condition is effectively solved, and the stability and efficiency of dynamic light output are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and more specifically, to an intelligent control method for LED light output based on high-frequency driving. Background Technology

[0002] With the rapid development of modern industrial automation and intelligent control technologies, lighting systems based on intelligent control methods have gradually become a research hotspot. Against this backdrop, LEDs (Light Emitting Diodes), as a highly efficient and energy-saving light source, are being widely used in home lighting, industrial lighting, displays, and intelligent optical systems. Traditional LED driving control technologies mainly focus on fixed-frequency driving and centralized control methods, achieving light output control through simple pulse width modulation (PWM) or constant current driving. However, as the scale of LED arrays expands and their applications in high-end scenarios become increasingly complex, existing control methods are struggling to meet the dynamic, multi-objective, and high-precision light output requirements, especially under high-frequency driving conditions, where the limitations of traditional control methods become increasingly apparent.

[0003] Existing control technologies for LED arrays face several challenges: First, differences in manufacturing processes, aging processes, and environmental conditions among LED chips lead to significant variations in their photoresponse characteristics. Centralized control methods cannot adequately account for these differences, impacting the uniformity and stability of overall light output. Second, under high-frequency driving conditions, the photoresponse behavior of LED chips exhibits complex nonlinear characteristics. Existing technologies lack effective modeling and predictive control mechanisms, making it difficult to achieve precise control over future light output states. Furthermore, centralized control architectures suffer from response delays and insufficient control precision when handling complex control tasks for large-scale LED arrays, especially in dynamic scenarios requiring rapid adjustments to light output. Traditional methods struggle to meet real-time and stability requirements. These technological bottlenecks severely limit the application potential of LED arrays in industrial control, adjustable lighting, and complex display systems. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an intelligent control method for LED light output based on high-frequency driving. This method can, to some extent, solve the problem that in multi-LED array systems, due to individual differences and nonlinear aging characteristics, the traditional uniform driving method leads to inconsistent light decay among the LED beads in the array, resulting in uneven overall light spots and color patches.

[0005] According to one aspect of the present invention, a method for intelligent control of LED light output based on high-frequency driving is provided, comprising: A high-frequency detection pulse sequence with the same amplitude and increasing frequency is applied to each LED in the LED array. The photoresponse time constant and peak light intensity data of each LED under the excitation of the detection pulse sequence are collected, and the frequency-photoresponse feature vector matrix of each LED is constructed.

[0006] Based on the frequency-light response feature vector matrix, the LED array is divided into several control groups with similar response characteristics, and each group is assigned an independent high-frequency driving channel and a set of control parameters.

[0007] A distributed predictive control model for an LED array is established, using the current light output state of each group of LEDs and the coupling effect of adjacent groups as input variables to predict the optimal combination of driving frequency and duty cycle parameters required by each group in multiple future control cycles.

[0008] Based on the optimal combination of driving frequency and duty cycle parameters, the high-frequency driving signals of each group are synchronously adjusted.

[0009] Furthermore, dividing the LED array into several control groups with similar response characteristics includes the following steps: The weighted cosine similarity metric between LED beads is calculated using an adaptive clustering algorithm based on similarity propagation, and the beads are iteratively grouped. The quality of the groups is evaluated and optimized using a quality evaluation function until the best balance between cohesion and separation is achieved, thus dividing the LED array into several control groups with highly similar response characteristics.

[0010] Furthermore, the quality evaluation function is constructed using the improved Calinski-Harabasz index, as shown in the following equation: ; in, The value of the group quality evaluation function. This represents the total number of groups in the LED array. Let be the number of LED beads contained in the i-th group. Let be the arithmetic mean of the optical response time constants of all LED beads in the i-th group. This represents the overall average of the optical response time constants of all LED beads globally. Let be the variance of the optical response time constant within the i-th group. Let be the mean of the optical response time constant within the i-th group. Let be the light response time constant of the m-th LED in the i-th group.

[0011] Furthermore, the distributed predictive control model divides the LED array into multiple intelligent agent nodes. Each intelligent agent performs rolling time-domain state prediction based on a local dynamic model and a coupling relationship model, and constructs an objective function and iterative calculation of constraints to obtain the optimal control parameter sequence for the next three control cycles of a single intelligent agent node.

[0012] Furthermore, the rolling time-domain state prediction uses its current state and neighbor information as a starting point, and uses local state transition equations and coupling influence vectors to gradually predict the state evolution of multiple future control cycles. Each prediction step superimposes the coupling influence of the neighbors into the local dynamic model to obtain a corrected state prediction value, which is then used as the starting state for the next prediction step.

[0013] Furthermore, the objective function includes a tracking error penalty term, a control input change penalty term, and a coordination consistency penalty term with neighboring agents at each prediction time. The constraints include upper and lower bound limits on the control input, current and temperature safety constraints on the LED devices, and system stability constraints.

[0014] Furthermore, the objective function is shown in the following equation: ; in, To optimize the objective function value of agent i in the three-period finite-time domain, Index for prediction time Let be the predicted value of light output intensity at time t. Let the target set value of light output intensity be given at time t. Nominal light output intensity Let be the change in driving frequency at time t. Let be the change in PWM duty cycle at time t. The nominal drive frequency, The nominal PWM duty cycle, Indexing neighboring intelligent agents, Let i be the set of the neighbors of agent i. For agent i, the control input at time t. Let j be the control intention of the neighboring intelligent agent j at time t. For nominal control input, To track the weighting coefficients of the error penalty term, To control the weighting coefficient of the penalty term for input changes, The weighting coefficient for the consistency penalty term.

[0015] According to another aspect of the present invention, an intelligent control system for LED light output based on high-frequency driving is provided, comprising: The frequency feature detection module is used to apply a high-frequency detection pulse sequence with the same amplitude and increasing frequency to each LED in the LED array, collect the photoresponse time constant and peak light intensity data of each LED under the excitation of the detection pulse sequence, and construct the frequency-photoresponse feature vector matrix of each LED. The intelligent grouping module is used to divide the LED array into several control groups with similar response characteristics based on the frequency-light response feature vector matrix, and to assign an independent high-frequency driving channel and control parameter set to each group. The predictive control module is used to establish a distributed predictive control model for the LED array. It uses the current light output state of each group of LED beads and the coupling effect of adjacent groups as input variables to predict the optimal combination of driving frequency and duty cycle parameters required by each group in the future multiple control cycles. The synchronous drive module is used to synchronously adjust the high-frequency drive signals of each group according to the optimal combination of drive frequency and duty cycle parameters.

[0016] According to another aspect of the present invention, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the intelligent control method for LED light output based on high-frequency driving as described above.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the intelligent control method for LED light output based on high-frequency drive as described above.

[0018] Compared with existing technologies, the intelligent control method for LED light output based on high-frequency driving provided by this invention applies a high-frequency probe pulse sequence with the same amplitude and increasing frequency to each LED in the LED array. It collects the photoresponse time constant and peak light intensity data of the LEDs under the excitation of the probe pulse sequence and constructs a frequency-photoresponse feature vector matrix to achieve grouped management and distributed predictive control of the LED array. This fully considers the differences in photoresponse characteristics among LEDs, improving the accuracy and uniformity of light output control. Simultaneously, by establishing a distributed predictive control model, it can predict the optimal driving frequency and duty cycle parameter combination for each group in real time, effectively solving the nonlinearity problem of photoresponse under high-frequency driving conditions and improving the stability and efficiency of dynamic light output. This method helps to realize intelligent control of multi-channel, large-scale LED arrays, meeting the dynamic light output requirements in complex scenarios, thereby significantly improving the limitations of existing LED control technologies. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart of an intelligent control method for LED light output based on high-frequency driving according to an embodiment of the present invention.

[0020] Figure 2 This is a comparison diagram of the uniformity of LED array light intensity distribution in the intelligent control method for LED light output based on high-frequency driving according to an embodiment of the present invention. Detailed Implementation

[0021] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0022] Figure 1 This is a flowchart of an intelligent control method for LED light output based on high-frequency driving, according to an embodiment of the present invention. Figure 1 As shown, the intelligent control method for LED light output based on high-frequency driving includes: S1: Apply a high-frequency detection pulse sequence with the same amplitude and increasing frequency to each LED in the LED array. The frequency range is 80kHz-150kHz with a step size of 5kHz. Collect the photoresponse time constant and peak light intensity data of each LED under the excitation of the detection pulse sequence, and construct the frequency-photoresponse feature vector matrix of each LED.

[0023] During the initialization phase of the LED array intelligent control system, individual characteristic identification and documentation for each LED bead are required. First, the system applies a standardized high-frequency detection pulse sequence sequentially to each LED bead in the LED array using a high-precision current source. The current amplitude of the detection pulse sequence is uniformly set to 50mA to ensure that each LED bead undergoes characteristic testing under the same current excitation conditions, avoiding measurement errors introduced by inconsistent driving conditions.

[0024] The frequency parameters of the probe pulses are executed according to a preset frequency sweep scheme, with a starting frequency of 80kHz and an ending frequency of 150kHz, and a frequency step size precisely controlled at 5kHz. This means that the entire frequency sweep process includes 15 different frequency points: 80kHz, 85kHz, 90kHz, 95kHz, 100kHz, 105kHz, 110kHz, 115kHz, 120kHz, 125kHz, 130kHz, 135kHz, 140kHz, 145kHz, and 150kHz. At each frequency point, the system sends a square wave pulse signal with a duration of 100μs to the target LED, with the rise and fall time of the pulse controlled within 10ns to ensure the accuracy of the high-frequency characteristic test.

[0025] During each probe pulse application, the system acquires the light output response signal of the LED in real time using a high-speed photodetector. The light response time constant is determined by analyzing the time required for the light output signal to rise from 10% peak to 90% peak; this parameter reflects the dynamic response speed of the LED under a specific frequency drive. Simultaneously, the system records the peak light intensity value reached by the light output signal, which represents the maximum light output capability of the LED under the corresponding frequency excitation.

[0026] By scanning all 15 frequency points, each LED chip generates 15 data pairs. Each data pair contains the frequency value, the corresponding photoresponse time constant, and the peak luminous intensity value. The system organizes this data into a 15×3-dimensional numerical matrix, where the first column represents the frequency parameter, the second column represents the photoresponse time constant, and the third column represents the peak luminous intensity data. This matrix is ​​the frequency-photoresponse eigenvector matrix of the LED chip, which fully describes the individual photoelectric characteristics of the chip under high-frequency driving conditions.

[0027] For an array system containing N LED beads, N independent frequency-light response eigenvector matrices will eventually be obtained.

[0028] S2: Based on the frequency-light response feature vector matrix, an adaptive clustering algorithm is used to divide the LED array into several control groups with similar response characteristics. The difference in the light response time constant of the LED beads in each group is controlled within 15%. An independent high-frequency driving channel and control parameter set are assigned to each group.

[0029] An adaptive clustering algorithm based on similarity propagation is used to intelligently group LED arrays. First, the similarity metric between the feature vector of each LED and the feature vectors of all other LEDs is calculated, constructing a similarity matrix. Then, the attraction and affiliation parameters of each LED are initialized to zero. Next, an iterative update process begins. When LED i sends attraction information to LED k, the system calculates the suitability of LED k as the cluster center for LED i. Simultaneously, LED k feeds back its affiliation information to LED i, indicating its willingness to accept LED i as a member. Through multiple rounds of iterative propagation, the similarity matrix is ​​further refined. LED beads with similar photoresponse characteristics gradually aggregate to form a stable group structure. If the changes in attraction and membership parameters tend to stabilize after multiple iterations, the system determines that the clustering process has converged and identifies the final cluster center. Then, a discreteness evaluation and adaptive adjustment method based on the balance optimization of group cohesion and inter-group segregation is used to classify LED beads with similar photoelectric response characteristics in the LED array into the same control group. Finally, each determined control group is assigned an independent high-frequency drive channel and a dedicated parameter configuration set, including control elements such as group identification code, drive frequency range, current adjustment step size, and response compensation parameters.

[0030] The similarity matrix is ​​constructed using a weighted cosine similarity algorithm to calculate the similarity metric of the feature vectors between LED beads. First, the light response time constant sequence and peak light intensity sequence are extracted from the frequency-light response feature vector matrix of each LED bead as two sub-vectors. Then, a weight coefficient of 0.6 is assigned to the light response time constant sub-vector, and a weight coefficient of 0.4 is assigned to the peak light intensity sub-vector to reflect the importance of the time constant in the control grouping. Next, when calculating the similarity between LED bead i and LED bead j, the system calculates the weighted dot product of the light response time constant vectors, the weighted dot product of the peak light intensity vectors, and the weighted magnitude of each vector. When two vectors are identical, the similarity value is 1; when two vectors are orthogonal, the similarity value is 0. The smaller the angle between the light response feature vectors of two LED beads, the closer the similarity metric value is to 1, indicating that the characteristics are more similar. Then, the calculated similarity metric value is filled into the corresponding position of the similarity matrix, where the row index i and column index j of the matrix represent the numbers of the two LED beads being compared. When i equals j, the diagonal elements are set to 1, indicating that the LED bead is completely similar to itself. When i is not equal to j, the system calculates the similarity metric value between all pairs of LED beads one by one and fills the off-diagonal positions of the matrix, finally constructing a symmetric N×N-dimensional similarity matrix.

[0031] The dispersion assessment and adaptive adjustment method calculates the variance of the photoresponse time constant of all LEDs in each group as the group cohesion index, and calculates the minimum distance between the center of the group and all other group centers as the group separation index. Then, it constructs the ratio of cohesion to separation as the group quality evaluation function. When the quality evaluation function value of a group deviates significantly from the global average quality evaluation function value, the system starts an adaptive optimization program. First, it calculates the Mahalanobis distance from each LED in the group to the group center, identifies outliers in the distance distribution as candidate reassignment objects, and then calculates the attraction strength between these candidate LEDs and adjacent group centers. If the attraction strength of a candidate LED from other groups exceeds that of the current group, the system migrates it to the target group with the strongest attraction. After the migration operation is completed, the system recalculates the quality evaluation function values ​​of the relevant groups. If the overall quality evaluation function tends to improve, the system continues to execute the next round of optimization iteration. If the quality evaluation function converges to a stable state after multiple consecutive iterations, the system determines that the group optimization is complete and locks the final group configuration.

[0032] The quality evaluation function employs a modified Calinski-Harabasz index to construct a grouped quality evaluation function. First, the sum of the squared Euclidean distances from the photoresponse time constants of all LEDs within each group to the group center is calculated as the intra-group dispersion. Then, the intra-group dispersion values ​​of all groups are summed to obtain the overall intra-group dispersion. Next, the squared Euclidean distance from each group center to the global center is calculated and multiplied by the number of LEDs within that group to obtain the inter-group dispersion. Finally, the inter-group dispersion values ​​of all groups are summed to obtain the overall inter-group dispersion. When the overall intra-group dispersion characterizes the tightness within all groups, a smaller value indicates better cohesion. When the overall inter-group dispersion... When characterizing the degree of separation between groups, a larger value indicates better separation. A quality evaluation function is then constructed as the ratio of the overall inter-group dispersion divided by the number of groups minus one, to the overall intra-group dispersion divided by the total number of LED beads minus the number of groups. The numerator represents the average inter-group variance, reflecting the contribution to separation, while the denominator represents the average intra-group variance, reflecting the cost of cohesion. If the local contribution of a group causes a decrease in the quality evaluation function value, the system marks that group as a target group requiring optimization. By maximizing the quality evaluation function, both cohesion and separation of groups are simultaneously optimized. When the quality evaluation function reaches a local optimum, it indicates that the current group configuration has achieved an optimal balance between cohesion and separation. The quality evaluation function is shown in the following equation: ; in, The value of the group quality evaluation function. This represents the total number of groups in the LED array. Let be the number of LED beads contained in the i-th group. Let be the arithmetic mean of the optical response time constants of all LED beads in the i-th group. This represents the overall average of the optical response time constants of all LED beads globally. Let be the variance of the optical response time constant within the i-th group. Let be the mean of the optical response time constant within the i-th group. Let be the light response time constant of the m-th LED in the i-th group.

[0033] Furthermore, the global average quality evaluation function value is calculated by weighting the quality evaluation function values ​​of each group constructed above according to the number of LED beads in the group. The quality evaluation function value of each group is multiplied by the number of LED beads in the group as the quality contribution weight of the group. Then, the quality contribution weights of all groups are summed and divided by the total number of LED beads to obtain the global average quality evaluation function value.

[0034] It should be noted that a significant deviation of the quality evaluation function value of a group from the global average quality evaluation function value is determined by calculating the absolute values ​​of the differences between the quality evaluation function values ​​of all groups and the global average quality evaluation function value to form a deviation sequence. Then, the standard deviation of the deviation sequence is calculated as a measure of the degree of deviation. Next, the absolute value of the difference between the quality evaluation function value of each group and the global average quality evaluation function value is divided by the standard deviation of the deviation sequence to obtain the standardized deviation index of that group. When the standardized deviation index of a group exceeds the sum of the median and the interquartile range of the deviation sequence, the system determines that the group has a significant deviation.

[0035] Finally, based on the total number of groups, the main controller initializes a corresponding number of high-frequency signal generation modules. Each signal generation module is equipped with an independent clock source, frequency synthesizer, and amplitude modulation circuit to ensure that the drive signals between groups are independent of each other. Then, a dedicated parameter configuration set is established for each group, including the reference drive frequency parameter determined based on the reciprocal of the average light response time constant of the LED beads in the group, the current adjustment range parameter set with upper and lower limits based on the statistical distribution of the peak light intensity of the LED beads in the group, the PWM duty cycle range parameter determined based on the linearity of the light response characteristics of the group, and the response compensation coefficient parameter calculated based on the variance of the characteristic vector of the LED beads in the group. When the response characteristics of the LED beads in a group are highly consistent, its response compensation coefficient is set to a smaller value to reduce unnecessary adjustment amplitude. When there are certain characteristic differences in a group, its response compensation coefficient is increased accordingly to provide sufficient compensation capability. Then, the system establishes a one-to-one mapping relationship between the drive channel and the group and stores it in the hardware address allocation table. At the same time, a dedicated feedback acquisition channel is allocated to each group for real-time monitoring of the light output status of the LED beads in the group, ultimately forming a multi-channel parallel control architecture.

[0036] S3: Establish a distributed predictive control model for the LED array, using the current light output state of each group of LED beads and the coupling effect of adjacent groups as input variables, and adopt a rolling time-domain optimization strategy to predict the optimal combination of driving frequency and duty cycle parameters required by each group in the next three control cycles.

[0037] The system establishes a distributed predictive control model based on multi-agent collaboration. First, each control group after the LED array is divided is regarded as an independent agent node. A local state space model is established for each agent, where the state vector contains four state variables: real-time light output intensity of LED beads in the group, light response time constant, junction temperature and drive current. The control input vector contains two control variables: drive frequency and PWM duty cycle. Then, a first-order Markov process is used to describe the state transition law of each agent. The state transition matrix and control input influence matrix are extracted from historical operation data through the least squares parameter identification method. Next, a coupling relationship model between agents is established. The thermal coupling coefficient matrix is ​​established by measuring the distance between groups and thermal resistance parameters, and the optical coupling coefficient matrix is ​​established through optical simulation and experimental calibration. Then, a global coordination layer is constructed to be responsible for information interaction and goal coordination among agents. When an agent needs to predict its future state, it requests the predictive control sequence information from neighboring agents. The state prediction value is calculated by comprehensively considering its own dynamic characteristics and the influence of its neighbors. Finally, a distributed predictive control model architecture that maintains local autonomy and achieves global coordination optimization is formed.

[0038] Preferably, a first-order Markov process is used to establish the discrete-time state-space equations for each agent. It is assumed that the agent's state vector at the next time step depends only on the current state vector and the control input vector. The state transition equation is established as follows: the state at the next time step equals the state transition matrix multiplied by the current state, plus the control input influence matrix multiplied by the current control input, plus a white noise term. The state transition matrix describes the system's free evolution, and the control input influence matrix describes the contribution of the control action to the state change. Then, the system collects state observation data and corresponding control input data from continuous historical time steps to construct the dataset required for parameter identification. The state transition equation is then converted into a linear regression form, where the output data matrix contains all historical data. The state observations at historical moments are used to construct the regression matrix, which consists of the historical state vector and the control input vector arranged in rows. The parameter matrix to be identified contains all the unknown parameters in the state transition matrix and the control input influence matrix. Then, the least squares criterion is applied to construct the objective function of parameter estimation, which is to minimize the sum of squared errors between the predicted state and the actual observed state. By solving the normal equations, the parameter estimates are obtained by multiplying the transpose of the regression matrix by the inverse of the regression matrix, and then by multiplying the transpose of the regression matrix by the output data matrix. When historical data is sufficient and the control input signal has sufficient spectral richness, the parameter identification results converge to the true parameter values. The obtained state transition matrix satisfies the system stability condition, and the numerical range of the control input influence matrix conforms to the physical constraints. As shown in the following equation: ; ; in, This is the optimized state transition matrix. This represents the total number of historical data points. Let be the state vector at time t+1. Let be the state vector at time t. Let be the transpose of the state vector at time t. Index for the current time, Let be the variance of the historical state vector. The mean of the historical state vector. Here is the numerical stability constant. These are the regularization coefficients of the state transition matrix. It is the identity matrix. The optimized control input influence matrix, Let be the control input vector at time t. Let t be the transpose of the control input vector at time t. Let be the change in the control input at time t. The mean of the historical control input vector. For numerical stability protection parameters, The variance of the historical control input vector. To control the input influence matrix regularization coefficients.

[0039] On the other hand, a thermal coupling sub-model is constructed, treating the LED array as a multi-node thermal network. Each agent corresponds to a thermal node. The thermal coupling strength between agent i and agent j is calculated using a thermal resistance network. The thermal resistance value equals the distance between the geometric centers of the two agents divided by the effective thermal conductivity of the heat-conducting medium, and then divided by the heat transfer area. When the power consumption of the LED beads in agent i increases, causing the junction temperature to rise, the excess heat diffuses to the adjacent agent j through the heat conduction path, causing a change in the ambient temperature of agent j and thus affecting the photoelectric conversion efficiency of its LED beads. The system establishes a thermal coupling coefficient matrix by measuring the temperature response relationship between the agents under different operating conditions, and then constructs an optical coupling sub-model, considering adjacent... The light field interaction between LED beads alters the surrounding light field distribution when the light output intensity of agent i changes. This affects the effective light output of neighboring agent j through light scattering and reflection. The optical coupling strength is related to the distance between the two agents, the emission angle of the LED beads, the characteristics of the reflective surface, and the properties of the optical medium. The system uses the Monte Carlo ray tracing method to simulate and calculate the contribution of each agent to the light output of its neighbors, establishing an optical coupling coefficient matrix to describe the quantitative relationship of optical interaction between agents. Then, a comprehensive coupling model is constructed to unify thermal coupling and optical coupling effects into a coupling influence vector. Each element of this vector represents the degree of influence of the corresponding neighboring agent on a specific state variable of the current agent. The obtained coupling influence vector is shown in the following equation: ; in, Let i be the coupling influence vector of agent i. This represents the total number of agents in the array. Number and index neighboring intelligent agents. Let be the thermal coupling coefficient between agents i and j. The heat power generated by agent j For the junction temperature of agent j, The junction temperature of agent i. Let i be the thermal resistance between agents i and j. Let be the distance between the geometric centers of agents i and j. The characteristic length of thermal diffusion. Let be the optical coupling coefficient between agents i and j. The light flux output of agent j. Let the angle between the line connecting agents i and j and the normal vector be... Let be the difference in photoelectric conversion efficiency between agents i and j. This is the light attenuation coefficient.

[0040] Furthermore, each agent performs rolling time-domain prediction based on the established local state space model and coupling relationship model. First, agent i obtains its own real-time state vector at the current time k as the prediction starting point. Then, it sends state information requests to all its neighboring agents j and receives the current state and planned control sequence of the neighbors. Next, agent i assumes that its own control input sequence in the next 3 control cycles is the variable to be optimized. It uses the local state transition equation to perform the first step of prediction, that is, calculates the state prediction value of the next time step based on the current state and the control input of the first control cycle. At the same time, it calculates the coupling influence term from the neighbors based on the coupling coefficient matrix and the state information of the neighboring agents. The coupling influence is superimposed as an additional input into the local state transition equation to obtain the corrected state prediction value. Then, the corrected state prediction value is used as the new starting state to repeat the above process for the second and third steps of prediction. If the coupling influence of a certain neighbor is small, the historical state information of that neighbor can be used for approximate calculation to reduce communication overhead. Finally, each agent obtains the predicted trajectory of its own state variables in the next 3 control cycles.

[0041] Furthermore, each agent constructs a local optimization problem based on the obtained state prediction trajectory to solve for the optimal control parameter sequence. First, agent i establishes a finite-time domain optimization objective function for three control cycles. This objective function includes a tracking error penalty term, a control input change penalty term, and a coordination and consistency penalty term with neighboring agents at each prediction time. The tracking error penalty term calculates the weighted quadratic error between the predicted light output intensity value and the target setpoint at each prediction time and accumulates it over the three prediction cycles. The control input change penalty term calculates the sum of squares of the changes in drive frequency and PWM duty cycle during adjacent control cycles to suppress drastic fluctuations in the control signal. The coordination and consistency penalty term promotes coordination among agents by comparing the differences between its own predicted control sequence and the control intentions announced by neighboring agents. Then, agent i uses a sequential quadratic programming algorithm to solve the constrained nonlinear optimization problem, with the optimization variables being the drive frequency and PWM duty cycle for the next three control cycles. The M duty cycle sequence has constraints including upper and lower bounds for control inputs, current and temperature safety constraints for LED devices, and system stability constraints. When the predicted junction temperature at a certain prediction time approaches the safety upper limit, the optimization algorithm will automatically reduce the power setting at the corresponding time to meet the temperature constraint. If neighboring agents transmit updated control plan information through the communication network, the current agent will recalculate the coordination and consistency term and adjust the optimization result. The optimization solution process uses gradient descent combined with the Lagrange multiplier method to handle equality and inequality constraints. Iterative calculation is performed until the objective function converges to a local optimum. When there are multiple local optima in the optimization problem, the system adopts a multi-starting point strategy to select the globally optimal or near-globally optimal solution. Finally, agent i obtains the optimal control parameter sequence for the next three control cycles, but only executes the control parameters for the first control cycle. The remaining control parameters are used as control intentions to be announced to neighboring agents for the next round of coordination optimization.

[0042] The objective function is shown in the following equation: ; in, ; ; ; The constraints include: ; ; ; ; ; ; in, To optimize the objective function value of agent i in the three-period finite-time domain, Index for prediction time Let be the predicted value of light output intensity at time t. Let the target set value of light output intensity be given at time t. Nominal light output intensity Let be the change in driving frequency at time t. Let be the change in PWM duty cycle at time t. The nominal drive frequency, The nominal PWM duty cycle, Indexing neighboring intelligent agents, Let i be the set of the neighbors of agent i. For agent i, the control input at time t. Let j be the control intention of the neighboring intelligent agent j at time t. For nominal control input, To track the weighting coefficients of the error penalty term, To control the weighting coefficient of the penalty term for input changes, The weighting coefficient for the consistency penalty term. These are the dominant eigenvalues ​​of the state transition matrix A. is the kurtosis parameter of the sigmoid function. To control the smoothness gain constant, The nominal operating temperature, Let be the direct coupling coefficient between agents i and j. and To drive the lower and upper bounds of the frequency, and These are the lower and upper bounds of the PWM duty cycle. This is the coefficient representing the influence of frequency on light output. The duty cycle is the coefficient that affects the light output. This refers to the upper limit of the safe temperature for LED devices. The power-temperature conversion factor, Let be the power at time t. The thermal coupling coefficient is... The thermal power of neighboring intelligent agent j. Let L2 be the coefficient of the coupling influence vector. This is the threshold for the maximum coupling effect.

[0043] In a preferred embodiment, the implementation process of step S3 is described in detail using the brightness uniformity control of an LED display screen in a shopping mall as an example. Assume that a large shopping mall has installed an LED display screen for playing advertising content. The display screen is composed of multiple LED modules. After preliminary cluster analysis, all LED modules are divided into several control groups. Each group contains several LED modules with similar photoelectric characteristics. The system establishes an intelligent agent node for each group. Intelligent agent A is responsible for controlling the LED modules in the upper left area of ​​the display screen, intelligent agent B controls the upper middle area, and intelligent agent C controls the upper right area. The state vector of each intelligent agent contains variables such as the average brightness, average color temperature, thermal state, and current state of the LED modules under its jurisdiction. The control input is the driving frequency and PWM duty cycle. Then, the system identifies the parameters of each intelligent agent through historical operating data and finds that the state transition matrix of intelligent agent A shows that the brightness change has strong inertial characteristics, the color temperature change is relatively slow, and the control input influence matrix shows that the PWM duty cycle has a significant impact on the brightness, while the driving frequency has a relatively small impact.

[0044] Next, the coupling relationship between the agents was established. Measurements revealed a significant thermal coupling effect between agent A and its neighboring agent B. When the LED module power of agent A increases, it affects the operating temperature of agent B through heat conduction, thereby affecting its light output efficiency. At the same time, there is also an optical coupling effect between the two agents. Changes in the brightness of agent A will affect the effective light output of agent B through optical crosstalk. In the rolling time-domain prediction stage, assuming that the brightness of the upper right area of ​​the display screen needs to be increased to improve the display effect, agent C first obtains its current brightness state, temperature state, and drive current state. Then, it requests state information and future control plans from neighboring agent B and other neighbors. It receives a plan from agent B to moderately increase the PWM duty cycle in the next control cycle, and other neighboring agents plan to maintain the current control parameters or make fine adjustments.

[0045] Based on this information, agent C predicts that if it significantly increases the PWM duty cycle in the next cycle, considering the thermal and optical coupling effects of the enhanced control from agent B, it predicts the changes in its temperature and brightness states. Similarly, it performs multi-step predictions to obtain the complete state evolution trajectory for the next few control cycles. In the optimization decision-making stage, agent C constructs an objective function that includes a brightness tracking error term to achieve the desired display brightness, a control smoothness term to limit the range of control parameter changes to avoid display flickering, and a neighbor coordination term to ensure that the brightness changes of neighboring agents are coordinated to maintain the continuity of the display screen. The optimal control parameter sequence for the next few control cycles is obtained through an optimization algorithm. However, the system only executes the control settings for the first cycle and notifies the neighboring agents of the subsequent control plan for coordination optimization in the next round. If a deviation between the actual brightness effect and the prediction is detected during execution, agent C will update the model parameters and perform rolling predictive control again in the next control cycle. Through this distributed predictive control mechanism, the entire LED display screen can maintain excellent display effects and brightness uniformity under various display content and environmental conditions.

[0046] S4: Based on the optimal combination of driving frequency and duty cycle parameters, synchronously adjust the high-frequency driving signals of each group.

[0047] The system, based on the optimal drive frequency and PWM duty cycle parameter combination output by the distributed predictive control model, synchronously sends parameter commands to the corresponding hardware drive channel controllers of each group via a high-speed data transmission bus. Upon receiving the parameter update command, each controller immediately reconfigures its internal frequency synthesis circuit and pulse width modulation circuit. The frequency synthesis circuit adjusts the reference clock to the specified drive frequency using a numerically controlled oscillator and phase-locked loop technology. The pulse width modulation circuit adjusts the ratio of the high-level duration to the signal period in real time according to the duty cycle parameter. The system adopts a unified synchronous triggering mechanism to ensure that the parameter updates of all groups take effect at the same time, avoiding instantaneous output inconsistencies caused by update time differences. When the main controller issues a global synchronization pulse, each group driver simultaneously switches to the new parameter configuration, realizing precise synchronous adjustment of the drive signals of each group of the LED array, thereby ensuring that the entire array operates in the optimal working state calculated by the optimization algorithm.

[0048] It is worth noting that, for example Figure 2The diagram shows a comparison between the proposed solution and a traditional control solution in terms of the uniformity of light intensity distribution in an LED array. It demonstrates the comparative effects of the traditional LED array control method and the distributed predictive control method of the present invention in terms of light intensity distribution uniformity. The horizontal axis represents the spatial distribution of the LED array from 0 to 100 mm, and the vertical axis represents the percentage of light intensity at each position. The red curve indicates that the light intensity distribution of the traditional control method exhibits significant fluctuations, with light intensity values ​​varying greatly between 73% and 94% at different positions, with a maximum deviation of 21%. In contrast, the green curve represents that the light intensity distribution of the proposed method remains highly stable, with light intensity values ​​evenly distributed between 94% and 97% at each position, and a deviation controlled within 3%. The comparison clearly shows that the distributed predictive control algorithm of the present invention effectively eliminates the problem of uneven light intensity in the traditional method, significantly improving the overall light intensity uniformity from 74.2% to 95.8%, fully demonstrating the technical advantages and practical application value of the present invention in LED array light field control.

[0049] In summary, the intelligent control method for LED light output based on high-frequency driving, as described in this invention, is explained. It applies a high-frequency probe pulse sequence with the same amplitude and increasing frequency to each LED in the LED array, collects the photoresponse time constant and peak light intensity data of the LED under the excitation of the probe pulse sequence, and constructs a frequency-photoresponse feature vector matrix to achieve grouped management and distributed predictive control of the LED array. This fully considers the differences in photoresponse characteristics among the LEDs, improving the accuracy and uniformity of light output control. Simultaneously, by establishing a distributed predictive control model, it can predict the optimal driving frequency and duty cycle parameter combination for each group in real time, effectively solving the nonlinearity problem of photoresponse under high-frequency driving conditions and improving the stability and efficiency of dynamic light output. This method helps to realize intelligent control of multi-channel, large-scale LED arrays, meeting the dynamic light output requirements in complex scenarios, thereby significantly improving the limitations of existing LED control technologies.

[0050] Here, those skilled in the art will understand that the specific operations of each step in the above-described intelligent control system for LED light output based on high-frequency drive have been referenced above. Figure 1 and Figure 2 The method for intelligent control of LED light output based on high-frequency driving has been described in detail, and therefore, its repeated description will be omitted.

[0051] In summary, the intelligent LED lighting output control system based on high-frequency driving, as described in this invention, is explained. It applies a high-frequency probe pulse sequence with the same amplitude and increasing frequency to each LED in the LED array, collects the light response time constant and peak light intensity data of the LEDs under the excitation of the probe pulse sequence, and constructs a frequency-light response feature vector matrix to achieve grouped management and distributed predictive control of the LED array. This fully considers the differences in light response characteristics among the LEDs, improving the accuracy and uniformity of light output control. Simultaneously, by establishing a distributed predictive control model, it can predict the optimal driving frequency and duty cycle parameter combination for each group in real time, effectively solving the nonlinearity problem of light response under high-frequency driving conditions and improving the stability and efficiency of dynamic light output. This method helps to realize intelligent control of multi-channel, large-scale LED arrays, meeting the dynamic light output requirements in complex scenarios, thereby significantly improving the limitations of existing LED control technologies.

Claims

1. A method for intelligent control of LED light output based on high-frequency driving, characterized in that, Includes the following steps: A high-frequency detection pulse sequence with the same amplitude and increasing frequency is applied to each LED in the LED array. The photoresponse time constant and peak light intensity data of each LED under the excitation of the detection pulse sequence are collected, and the frequency-photoresponse feature vector matrix of each LED is constructed. Based on the frequency-light response feature vector matrix, the LED array is divided into several control groups with similar response characteristics, and each group is assigned an independent high-frequency driving channel and control parameter set. A distributed predictive control model for an LED array is established, using the current light output state of each group of LED beads and the coupling effect of adjacent groups as input variables to predict the optimal combination of driving frequency and duty cycle parameters required by each group in multiple future control cycles. Based on the optimal combination of driving frequency and duty cycle parameters, the high-frequency driving signals of each group are synchronously adjusted.

2. The intelligent control method for LED light output based on high-frequency driving according to claim 1, characterized in that, The process of dividing the LED array into several control groups with similar response characteristics includes the following steps: The weighted cosine similarity metric between LED beads is calculated using an adaptive clustering algorithm based on similarity propagation, and the beads are iteratively grouped. The quality of the groups is evaluated and optimized using a quality evaluation function until the best balance between cohesion and separation is achieved, thus dividing the LED array into several control groups with highly similar response characteristics.

3. The intelligent control method for LED light output based on high-frequency driving according to claim 2, characterized in that, The quality evaluation function is constructed using the improved Calinski-Harabasz index, as shown in the following equation: ； in, The value of the group quality evaluation function. This represents the total number of groups in the LED array. Let be the number of LED beads contained in the i-th group. Let be the arithmetic mean of the optical response time constants of all LED beads in the i-th group. This represents the overall average of the optical response time constants of all LED beads globally. Let be the variance of the optical response time constant within the i-th group. Let be the mean of the optical response time constant within the i-th group. Let be the light response time constant of the m-th LED in the i-th group.

4. The intelligent control method for LED light output based on high-frequency driving according to claim 3, characterized in that, The distributed predictive control model divides the LED array into multiple intelligent agent nodes. Each intelligent agent performs rolling time-domain state prediction based on a local dynamic model and a coupling relationship model, and constructs an objective function and iterative calculation of constraints to obtain the optimal control parameter sequence for the next three control cycles of a single intelligent agent node.

5. The intelligent control method for LED light output based on high-frequency driving according to claim 4, characterized in that, The rolling time-domain state prediction starts by acquiring its current state and neighbor information, and uses local state transition equations and coupling influence vectors to gradually predict the state evolution of multiple future control cycles. Each prediction step adds the coupling influence of the neighbors to the local dynamic model to obtain the corrected state prediction value, which is then used as the starting state for the next prediction.

6. The intelligent control method for LED light output based on high-frequency driving according to claim 5, characterized in that, The objective function includes a tracking error penalty term, a control input change penalty term, and a coordination consistency penalty term with neighboring agents at each prediction time. The constraints include upper and lower bound limits on the control input, current and temperature safety constraints on the LED devices, and system stability constraints.

7. The intelligent control method for LED light output based on high-frequency driving according to claim 6, characterized in that, The objective function is shown in the following equation: ； in, To optimize the objective function value of agent i in the three-period finite-time domain, For the index of the predicted time, Let be the predicted value of light output intensity at time t. Let the target set value of light output intensity be given at time t. The nominal light output intensity, Let be the change in driving frequency at time t. Let be the change in PWM duty cycle at time t. The nominal drive frequency, The nominal PWM duty cycle, Indexing neighboring intelligent agents, Let i be the set of the neighbors of agent i. For agent i, the control input at time t. Let j be the control intention of the neighboring intelligent agent j at time t. For nominal control input, To track the weighting coefficients of the error penalty term, To control the weighting coefficient of the penalty term for input changes, The weighting coefficient for the consistency penalty term.

8. A smart control system for LED lighting output based on high-frequency driving, characterized in that, include: The frequency feature detection module is used to apply a high-frequency detection pulse sequence with the same amplitude and increasing frequency to each LED in the LED array, collect the photoresponse time constant and peak light intensity data of each LED under the excitation of the detection pulse sequence, and construct the frequency-photoresponse feature vector matrix of each LED. The intelligent grouping module is used to divide the LED array into several control groups with similar response characteristics based on the frequency-light response feature vector matrix, and to assign an independent high-frequency driving channel and control parameter set to each group. The predictive control module is used to establish a distributed predictive control model for the LED array. It uses the current light output state of each group of LED beads and the coupling effect of adjacent groups as input variables to predict the optimal combination of driving frequency and duty cycle parameters required by each group in the future multiple control cycles. The synchronous drive module is used to synchronously adjust the high-frequency drive signals of each group according to the optimal combination of drive frequency and duty cycle parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent control method for LED light output based on high-frequency driving as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent control method for LED light output based on high-frequency drive as described in any one of claims 1 to 7.

Images