LED light source brightness uniformity control method, device and equipment

Abstract

The present invention relates to a method, device and equipment for controlling the brightness uniformity of an LED light source. The method: collects the driving current, PWM duty cycle and light intensity signal of each LED unit in an LED array to establish an LED characteristic data set; extracts working state features to obtain LED working state data, and inputs the data into an LED brightness prediction model to predict the light intensity value to obtain a predicted light intensity value; performs spectrum analysis and resampling calculation on the predicted light intensity value and the measured light intensity value to obtain a state determination result of the LED unit; divides the LED array into multiple driving groups, and controls each driving group to equalize current through energy regulation of an energy storage capacitor to obtain LED driving current parameters; divides the LED array into compensation areas to obtain multiple compensation areas, and calculates the brightness uniformity compensation parameters of each compensation area through a BSA algorithm. The implementation of the present invention improves the overall display quality and operational reliability of the LED array.

Description

Technical Field

[0001] The present invention relates to the technical field of LED light source control, and in particular to a method, device and equipment for controlling brightness uniformity of an LED light source. Background Art

[0002] LED light sources are widely used in display, lighting and other fields due to their high efficiency, energy saving and long life. However, due to differences in manufacturing processes and device characteristics, different LED units in the same LED array often exhibit different light intensity outputs under the same driving conditions, resulting in uneven display or lighting effects and affecting overall performance.

[0003] Especially in large-scale LED array applications, factors such as batch differences, temperature drift, and aging effects of LED units will cause the problem of brightness non-uniformity to become more prominent. Traditional brightness uniformity methods mainly rely on simple table lookup compensation or linear correction, which is difficult to adapt to the dynamic changes of LED units under different working conditions, and the compensation effect is limited by fixed compensation strategies. In addition, fault detection and compensation of LED units usually require multiple sensors and complex hardware circuits, which increases system cost and complexity. Existing uniformity control methods often treat fault diagnosis, drive control, and brightness compensation as independent links, lack a systematic collaborative optimization mechanism, and it is difficult to achieve efficient brightness uniformity control while ensuring system reliability. Summary of the invention

[0004] The main purpose of the present invention is to provide a method, device and equipment for controlling the brightness uniformity of an LED light source. The present invention improves the overall display quality and operational reliability of an LED array.

[0005] To achieve the above object, the present invention provides a method for controlling the brightness uniformity of an LED light source, comprising the following steps:

[0006] The driving current, PWM duty cycle and light intensity signal of each LED unit in the LED array are collected, and the initial light intensity distribution data of the LED array is obtained through pixel-level light intensity detection to establish an LED characteristic data set;

[0007] Extract working state features based on the LED characteristic data set to obtain LED working state data, and input the LED working state data into an LED brightness prediction model to predict light intensity values ​​to obtain predicted light intensity values;

[0008] Performing spectrum analysis and resampling calculation on the predicted light intensity value and the measured light intensity value to obtain a state determination result of the LED unit;

[0009] Based on the state determination result, the LED array is divided into a plurality of drive groups, and current sharing control is performed on each drive group through energy regulation of the energy storage capacitor to obtain LED drive current parameters;

[0010] According to the LED driving current parameter and the state determination result, the LED array is divided into compensation areas to obtain a plurality of compensation areas, and the brightness uniformity compensation parameter of each compensation area is calculated by a BSA algorithm.

[0011] The present invention also provides a device for controlling brightness uniformity of an LED light source, comprising:

[0012] Establish a module for collecting the driving current, PWM duty cycle and light intensity signal of each LED unit in the LED array, obtain the initial light intensity distribution data of the LED array through pixel-level light intensity detection, and establish an LED characteristic data set;

[0013] A prediction module, used for extracting working state features based on the LED characteristic data set to obtain LED working state data, and inputting the LED working state data into an LED brightness prediction model to predict the light intensity value to obtain a predicted light intensity value;

[0014] A calculation module, used for performing spectrum analysis and resampling calculation on the predicted light intensity value and the measured light intensity value to obtain a state determination result of the LED unit;

[0015] A control module, for dividing the LED array into a plurality of drive groups based on the state determination result, and performing current sharing control on each drive group by energy regulation of an energy storage capacitor to obtain LED drive current parameters;

[0016] The compensation module is used to divide the LED array into compensation areas according to the LED driving current parameter and the state determination result to obtain multiple compensation areas, and calculate the brightness uniformity compensation parameter of each compensation area through the BSA algorithm.

[0017] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0018] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0019] In summary, the technical solution provided by the present invention establishes a complete LED characteristic data set including spatial position, temperature influence and batch characteristics through pixel-level light intensity detection combined with multi-dimensional feature fusion technology, providing a reliable data basis for brightness control; adopts a deep learning model combined with domain adversarial training and self-attention mechanism to significantly improve the accuracy of light intensity prediction and the ability to adapt to cross-working conditions; innovatively introduces spectrum analysis and resampling technology to achieve accurate fault diagnosis of LED units, and based on this, adaptive division of drive groups is performed; a double-layer energy storage structure and a dual current equalization control unit are designed, and the drive current ripple is effectively reduced through phase shift control and energy compensation strategies; the BSA algorithm is used to optimize the compensation area, and accurate brightness compensation is achieved in combination with sequence optimization and cyclic adjustment strategies. The present invention realizes the system coordination of fault diagnosis, drive control and brightness compensation, and significantly improves the overall display quality and operational reliability of the LED array while reducing hardware complexity and implementation costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the steps of a method for controlling brightness uniformity of an LED light source in one embodiment of the present invention;

[0021] Figure 2 This is a structural block diagram of a device for controlling brightness uniformity of an LED light source in one embodiment of the present invention;

[0022] Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0023] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0025] Reference Figure 1 This embodiment provides a method for controlling the brightness uniformity of an LED light source, comprising the following steps:

[0026] S1, collect the driving current, PWM duty cycle and light intensity signal of each LED unit in the LED array, obtain the initial light intensity distribution data of the LED array through pixel-level light intensity detection, and establish an LED characteristic data set;

[0027] A high-resolution light intensity sensor array is set for each LED unit in the LED array to achieve accurate light intensity sampling. The relationship between the light intensity sensor array and the LED array is not fixed. Through the spatial resolution matching operation, the position of each LED unit is mapped to the corresponding sensor unit one by one to obtain the mapping relationship data between the LED unit and the sensor unit. Multiple groups of different driving currents and PWM duty cycle combinations are applied to the LED units. Different driving conditions cause the LED units to emit light of different intensities. Through sampling by the light intensity sensor array, the luminous characteristic curve of the LED unit under different driving conditions is obtained, reflecting the brightness change of the LED under various operating conditions. The luminous characteristic curve is input into the spatial mapping processing unit, and the corresponding relationship data between the position and light intensity output of each LED unit is obtained through interpolation calculation. The interpolation calculation process can fill the gap in light intensity data under different driving states and refine the relationship between the LED unit and its corresponding light intensity output. The light intensity of the LED array is sampled under different ambient temperatures, the temperature influence factor of the LED unit is calculated, and the LED unit is sampled and measured under multiple temperature conditions using the factor. By performing standard deviation analysis on these sampled data, the batch characteristic parameters of each LED unit are obtained, reflecting the stability and consistency of the LED unit under different temperatures. The mapping relationship data, corresponding relationship data and batch characteristic parameters are fused to obtain the LED unit feature matrix. The feature matrix contains the relationship between the light intensity output of the LED unit and its driving current and PWM duty cycle, as well as the temperature impact data of the LED unit under different working conditions. The data in the feature matrix are analyzed for correlation, and the dynamic characteristic parameters of the LED unit are obtained through feature extraction methods to reflect the performance changes of the LED unit in real time, especially the response to temperature fluctuations or current changes. Based on the time series management method, the dynamic characteristic parameters are stored and archived to form a complete LED characteristic data set.

[0028] S2, extracting working state features based on the LED characteristic data set to obtain LED working state data, and inputting the LED working state data into the LED brightness prediction model to predict the light intensity value to obtain the predicted light intensity value;

[0029] Specifically, the LED characteristic data set is divided into feature sequences of fixed length according to the time series, and multiple continuous working state data are generated through sliding window processing, which represents the working state changes of the LED unit in different time periods and can capture the dynamic behavior of the LED array. The LED working state data is input into the LED brightness prediction model to predict the light intensity value. The working state data is processed by the feature extractor. The feature extractor adopts a deep convolutional network structure, including four convolutional layers, each of which is followed by a ReLU activation function and a batch normalization layer. This structure can effectively capture the spatial characteristics of the input data, and extract the domain invariant features of the LED unit through the sliding window mechanism of the convolutional layer. The domain invariant features represent the stability characteristics of the LED unit under different working states and environmental conditions. The domain invariant features are input into the domain discriminator of the LED brightness prediction model. The domain discriminator includes three fully connected layers, each of which uses the LeakyReLU activation function and the Dropout layer to enhance the robustness of the model and avoid overfitting. Through the processing of this discriminator, adversarial calculation can be performed on the domain invariant features to obtain a cross-condition feature vector, which reflects the response changes of the LED unit under different working environments, temperatures or current conditions. The cross-condition feature vector has strong generalization ability and adapts to changing working environments. The cross-condition feature vector is input into the brightness predictor in the LED brightness prediction model, especially its encoding module. The encoding module consists of six residual blocks, each of which contains two layers of convolutional networks and a short-circuit connection to form an effective feature encoding structure. These residual blocks convert the cross-condition feature vector into a more abstract and deep LED unit feature by feature encoding. The design of the residual block can effectively avoid the gradient vanishing problem and keep the information flow through the short-circuit connection, making the feature encoding process more stable and efficient. Self-attention calculation is performed on the encoded deep features. The self-attention mechanism can focus on the important information in the input data and suppress the unimportant information. The temporal or spatial position information of the data is captured by position encoding, and then the information weights of different positions are calculated in parallel using the multi-head attention mechanism. These features are weighted to form an attention feature map, which can effectively highlight the key characteristics of the LED unit and reduce the interference of irrelevant data. The attention feature map is input into the decoding module, which consists of four deconvolution layers, each of which is followed by bilinear interpolation upsampling and residual connection. This decoding process restores the low-resolution feature map to a higher resolution through upsampling technology, making the final light intensity prediction result more accurate. The design of the deconvolution layer can restore the spatial resolution of the data and optimize the detailed features in the data, making the prediction result closer to the actual light intensity distribution. Multi-scale fusion of light intensity features. By adopting three feature pyramid networks of different scales and combining jump connections, information from different scales is fused to obtain a more comprehensive and accurate light intensity prediction.Multi-scale fusion technology can process features of different resolutions, making the light intensity prediction value of LED units more diversified and adaptable to lighting scenes of different scales. After multi-scale fusion, the multi-scale light intensity prediction value is input into the output layer of the LED brightness prediction model, and after weighted averaging and normalization, the predicted light intensity value is finally obtained.

[0030] The input domain-invariant features are flattened and converted into a one-dimensional feature sequence, so that the features can be effectively transmitted in the subsequent fully connected layers, while simplifying the dimension of the features. The one-dimensional feature sequence is input to the first fully connected layer of the domain discriminator. The first fully connected layer contains 1024 neurons, which can learn preliminary high-level feature representations from the input feature sequence. The LeakyReLU activation function is used to ensure that the negative value part still has certain gradient information during the training process, which helps to accelerate model convergence and reduce the gradient vanishing problem. At the same time, in order to avoid overfitting, the first layer features are subjected to a Dropout operation with a probability of 0.5 after calculation. The Dropout operation enhances the generalization ability of the model and reduces the risk of overfitting by randomly discarding the activation values ​​of some neurons. The output features of the first layer are passed to the second fully connected layer. The second layer contains 512 neurons, which extracts more abstract representations from the features output by the first layer. In this layer, the same LeakyReLU activation function is used for nonlinear transformation to enhance the feature representation ability, and the Dropout operation with a probability of 0.3 is used to reduce overfitting. The features output by the second layer are input to the third fully connected layer of the domain discriminator. This layer contains 256 neurons, extracts high-dimensional features from the features output by the second layer, and calculates them through the LeakyReLU activation function. In order to maintain the stability and robustness of the model, the output features of the third layer perform a Dropout operation with a probability of 0.2 to reduce the risk of overfitting of the model during training. After this layer of processing, the output features of the third layer are obtained. The output features of the third layer are subjected to a maximum pooling operation. The spatial dimension of the features is reduced by downsampling while retaining the most significant feature information. In order to ensure that the pooling operation can effectively reduce the dimension of the feature map and retain important spatial information, the pooling kernel size is set to 2, that is, each 2x2 area will output a maximum value. The pooled feature vector is input to the gradient reversal layer, which reverses the gradient direction during the back propagation process to achieve adversarial training. By performing a gradient reversal operation in the reversal layer, the model pays more attention to learning cross-condition features during training, rather than simply relying on the performance under a specific working state. This process helps the model improve its generalization ability under various working conditions and adapt to the light intensity prediction task under different environmental conditions. Perform L2 regularization on adversarial features. L2 regularization can limit the complexity of the model, prevent the model from generating too large weight values ​​during training, and ensure the stability and robustness of the model. Obtain regularized features through normalized calculation. Fuse the regularized features with the third-layer output features. Combine feature information from two different sources by splicing or weighting, enhance the model's ability to express cross-condition features, and obtain a cross-condition feature vector.

[0031] S3, performing spectrum analysis and resampling calculation on the predicted light intensity value and the measured light intensity value to obtain a state determination result of the LED unit;

[0032] It should be noted that by comparing the predicted light intensity value with the measured light intensity value, the difference between them is calculated to obtain the light intensity deviation data of each LED unit, reflecting the deviation between the actual working process of the LED array and the predicted value. The light intensity deviation data is sampled in the time domain, and the data is intercepted by setting a fixed time window to obtain a time-dependent light intensity deviation sequence, which effectively represents the light intensity change of the LED unit over a period of time and reflects the dynamic characteristics of the LED working state. The light intensity deviation sequence is spectrally decomposed to identify the periodic characteristics and inherent frequencies in the data, thereby revealing the frequency characteristics of the LED unit under different working states. Through spectral decomposition, the inherent frequency characteristics of the LED unit are extracted, which helps to judge the working state of the LED and identify potential faults. In the process of spectral decomposition, the resonant frequency or other periodic change patterns of the LED unit under specific conditions are revealed, reflecting the health status of the LED unit. The inherent frequency features are resampled according to the spatial position of the LED unit to ensure that the features of the LED units at different positions can be compared in the same space-time framework, and the spatial correlation features are extracted to effectively reveal the relationship between different LED units, helping to determine which units' light intensity changes are caused by common factors (such as temperature, driving current fluctuations, etc.) and which are caused by single unit failures. The spatial correlation features are matched with the preset fault feature data. The preset fault feature data is obtained through historical data accumulation, simulation experiments or fault diagnosis models, and contains typical features of LED units under various types of faults or abnormal working conditions. By matching and calculating the spatial correlation features with these fault features, feature similarity data is obtained, reflecting the similarity between the features of the current LED unit and the preset fault mode. In this way, it is accurately determined whether the LED unit is abnormal and its fault type is classified. The feature similarity data is normalized and converted into a standardized quantized value to obtain the state quantization value of the LED unit. Based on the state quantization value, each LED unit is classified and marked. According to the size of the quantization value, the LED unit is divided into different working state categories. For example, an LED unit with a higher state quantization value represents a normal working state, while a unit with a lower quantization value or abnormal fluctuation indicates a fault or performance degradation. The working state data is compared and analyzed with the preset state threshold to determine whether it meets the preset normal working standard. Through classification and threshold comparison, the state determination result of each LED unit is obtained.

[0033] S4, based on the state determination result, the LED array is divided into a plurality of driving groups, and current equalization control is performed on each driving group through energy regulation of the energy storage capacitor to obtain LED driving current parameters;

[0034] Specifically, the LED array is divided into a plurality of drive groups based on the state determination result. Two drive groups are constructed based on the working state of the LED unit, wherein the first drive group includes LED units in a normal working state, and the second drive group includes LED units in an abnormal working state. Preliminary drive grouping data is obtained by classifying the state of the LED unit. In order to ensure that energy balance can be achieved between the drive groups, different energy storage capacitors are respectively set in the first drive group and the second drive group. Among them, the first energy storage capacitor is used to realize energy exchange between different drive groups, so that when necessary, excess energy can be transferred from one drive group to another drive group to maintain the balance of the overall system. At the same time, the second energy storage capacitor is specifically used to adjust the energy within the group to ensure that the LED units in each group can work under stable current conditions. Through the double-layer energy storage structure, the current is accurately controlled at both the global and local levels to obtain the drive group energy storage parameters. The drive group energy storage parameters are input into the first current sharing control unit to realize current sharing control between channels. The first current sharing control unit includes a plurality of voltage sampling ports and a current detection port, and obtains the current state of each channel by real-time sampling of the voltage and current of different channels. Based on these data, the current balancing of each channel is accurately adjusted through the distributed control strategy to obtain the current balancing data between channels, which reflects the current distribution between different channels. Through this process, the problem of too large or too small current in a single channel is effectively avoided, ensuring that the current distribution between each channel is more uniform. The current balancing data between channels is subjected to the first phase shift control. The phase shift control adjusts the phase of different driving signals so that each LED unit can work in an interleaved manner when driving, avoiding the sudden change or excessive fluctuation of current caused by turning on multiple LED units at the same time. Through the design of staggered conduction timing, synchronous operation between multiple channels can be achieved, while avoiding excessive current shock. After this control step, the first group of driving signal sequences is obtained, which can ensure the orderly switching between each driving channel in time and achieve better current balancing. At the same time, the second current balancing control unit realizes the current balancing within the group through the energy compensation branch, ensuring that the LED units in each driving group work under a stable current. Through compensation control, the second current balancing control unit can make corresponding adjustments according to the imbalance of the current within the group, optimize the driving current of each LED unit, and prevent some LED units from being damaged or degraded due to excessive or too small current. Through this process, the current balancing data within the group is obtained. The second phase shift control is performed on the current balancing data within the group. The second phase shift control dynamically adjusts the phase so that each drive signal works in coordination in time, ensuring that no excessive fluctuations or unstable working conditions are caused during the current compensation process. Through this adjustment, the second set of drive signal sequences is obtained.The first drive signal sequence and the second drive signal sequence are controlled in coordination, and the current distribution is optimized through a double closed-loop feedback mechanism to ensure that each drive group and each LED unit can maintain the optimal working current under different working conditions. Based on the multi-channel drive control parameters, the first energy storage capacitor and the second energy storage capacitor are accurately allocated energy to ensure that the system always maintains energy balance and stability during the entire working process, and the LED drive current parameters are obtained.

[0035] S5, dividing the LED array into compensation areas according to the LED driving current parameters and the state determination result to obtain a plurality of compensation areas, and calculating the brightness uniformity compensation parameters of each compensation area through the BSA algorithm.

[0036] The compensation area of ​​the LED array is divided according to the LED driving current parameter and the state determination result. An optimization sequence containing the spatial position information and the light intensity characteristic information of the LED unit is constructed. The first optimization sequence contains the spatial position information of the LED unit, while the second optimization sequence contains the light intensity characteristic information of the LED unit. The two sequences are processed in a sequence correspondence manner to obtain the optimization data of each LED unit. The LED unit optimization data is partitioned and clustered. The LED array is divided into multiple initial compensation areas by analyzing the brightness similarity of the LED units. By comparing the brightness characteristics of the LED units in different areas, the LED units with higher similarity are gathered in the same area to obtain the initial division result of the compensation area. The initially divided compensation area data is input into the grouping optimization unit of the BSA algorithm, and the division of the compensation area is optimized through the area merging operation of the algorithm to obtain the optimal compensation area data. The brightness uniformity of the LED array is iteratively calculated based on the optimal compensation area data. The first compensation coefficient is obtained by optimizing the compensation parameters. The first compensation coefficient represents the current adjustment amplitude required to achieve brightness uniformity in a given compensation area. The first compensation coefficient is adjusted cyclically. During the adjustment process, the compensation coefficient is optimized according to the light intensity deviation of different LED units by updating the compensation amount to obtain the second compensation coefficient. The first compensation coefficient and the second compensation coefficient are combined and optimized. The combined compensation coefficient is generated by weighting the two coefficients. The combined compensation coefficient can take into account the spatial distribution and light intensity characteristics of the LED units at the same time, ensuring that the brightness of each area can be balanced to the maximum extent in actual operation. In order to ensure the consistency and stability of the compensation effect within the area, the combined compensation coefficient is subjected to boundary processing so that the compensation coefficient will not fluctuate violently at the junction of different compensation areas, avoiding the problem of uneven brightness or over-compensation. Through constrained optimization operations, a set of optimized compensation parameter matrices are obtained. The compensation parameter matrix is ​​mapped to the LED units in each compensation area to obtain the specific brightness uniformity compensation parameters of each LED unit.

[0037] In one example, the driving current, PWM duty cycle and light intensity signal of each LED unit in the LED array are collected, and the initial light intensity distribution data of the LED array is obtained through pixel-level light intensity detection to establish an LED characteristic data set, including:

[0038] A high-resolution light intensity sensor array is set for the LED array, and the mapping relationship data between the LED unit and the sensor unit is obtained through spatial resolution matching operation;

[0039] Apply multiple sets of driving current and PWM duty cycle combinations to the LED unit, and obtain the luminous characteristic curve of the LED unit by measuring the light intensity sensor array;

[0040] The luminous characteristic curve is input into the space mapping processing unit, and the corresponding relationship data between the position of the LED unit and its light intensity output is obtained through interpolation calculation;

[0041] The light intensity of the LED array is sampled at different ambient temperatures, and the temperature influence factor of the LED unit is obtained through temperature compensation calculation. Based on the temperature influence factor, the LED unit is sampled and measured multiple times, and the batch characteristic parameters of the LED unit are obtained through standard deviation analysis.

[0042] The mapping relationship data, corresponding relationship data and batch characteristic parameters are fused to obtain the feature matrix of the LED unit, and the feature matrix and temperature influencing factors are correlated and analyzed. The dynamic characteristic parameters of the LED unit are obtained by feature extraction, and the dynamic characteristic parameters are stored in time series to obtain the LED characteristic data set.

[0043] In this example, a high-resolution light intensity sensor array is set on the LED array to collect the light intensity information emitted by each LED unit in the array. Through the spatial resolution matching operation of the sensor array, a mapping relationship between each LED unit and the sensor unit is established. Through the alignment of spatial resolution, each LED unit corresponds to a certain area in the sensor array, ensuring that the light intensity value of each LED unit can be accurately captured during measurement. Assuming that there are N LED units in the LED array and M sensor units in the sensor array, the spatial resolution matching operation is expressed as:

[0044] ;

[0045] in, Indicates LED unit With sensor unit The spatial distance between is the horizontal coordinate of the LED unit and the sensor unit, As their ordinates. By calculating the spatial distance, the corresponding relationship between the sensor and the LED unit is determined. Multiple sets of driving current and PWM duty cycle combinations are applied to the LED unit to obtain the luminous characteristic curve of the LED unit. Changes in driving current and PWM duty cycle will cause changes in the light intensity output of the LED unit. By measuring the light intensity values ​​under different driving conditions, the luminous characteristic curve of the LED unit is obtained. The luminous characteristic curve is expressed as the relationship between light intensity, current and duty cycle, and is described using the following formula:

[0046] ;

[0047] in, Indicates the LED unit at time The light intensity at that time, is the driving current of the LED unit, is the PWM duty cycle. Through multiple sets of current and PWM duty cycle combinations, the light intensity output of each LED unit under different conditions is measured, and the luminous characteristic curve is plotted. After the luminous characteristic curve is input into the space mapping processing unit, the corresponding relationship data between the position of the LED unit and its light intensity output is obtained through interpolation calculation. The interpolation calculation estimates the light intensity value at other positions or conditions through the known luminous characteristic data points. For example, using the Lagrange interpolation method, the formula is:

[0048] ;

[0049] in, is the predicted value of light intensity obtained by interpolation, is the location of the known data point, is the light intensity value corresponding to the known data point, is the total number of known data points. Interpolation calculation can effectively predict the light intensity data of unmeasured positions and complete the calculation of the corresponding relationship between position and light intensity output. The light intensity of the LED array is sampled at different ambient temperatures. The luminous efficiency and light intensity of the LED will change with the change of temperature, especially under high temperature conditions, the light efficiency of the LED will decrease. The temperature influence factor is obtained by temperature compensation calculation. The calculation of this factor is expressed by a linear model:

[0050] ;

[0051] in, is the ambient temperature, and is the temperature compensation coefficient obtained through experiments. The temperature influence factor is used to adjust the light intensity output of the LED unit at different temperatures to eliminate the impact of temperature changes. Based on the temperature influence factor, multiple sampling measurements are performed on each LED unit, and the batch characteristic parameters of the LED unit are obtained through standard deviation analysis. Standard deviation analysis is used to evaluate the light intensity fluctuations caused by manufacturing differences in the production process of the LED unit. Assume that the light intensity data generated by the LED unit in multiple measurements is , the standard deviation is calculated by the following formula:

[0052] ;

[0053] in, is the average value of the light intensity data, is the standard deviation, is the number of measurement data. Through standard deviation analysis, the batch characteristic parameters of each LED unit are obtained to describe the consistency of the LED unit in the production process. The mapping relationship data, the corresponding relationship data and the batch characteristic parameters are data fused. The measurement data from different sources are integrated to form a more accurate and comprehensive description of the characteristics of the LED unit. The dynamic characteristic parameters of the LED unit are extracted by correlation analysis of the characteristic matrix and the temperature influencing factor. The dynamic characteristic parameters include the law of light intensity change of the LED unit under different environmental conditions and its response to temperature changes. These dynamic characteristic parameters are stored in time series, and finally a complete LED characteristic data set is constructed to reflect the working characteristics, temperature influence and performance changes of each LED unit under different operating conditions.

[0054] In one example, working state feature extraction is performed based on an LED characteristic data set to obtain LED working state data, and the LED working state data is input into an LED brightness prediction model to predict the light intensity value to obtain a predicted light intensity value, including:

[0055] The LED characteristic data set is divided into feature sequences of fixed length according to the time series, and the LED working status data is obtained through sliding window processing;

[0056] The LED working status data is input into the feature extractor of the LED brightness prediction model for processing. The feature extractor contains 4 convolutional layers. Each convolutional layer uses the ReLU activation function and the batch normalization layer to obtain the domain-invariant features of the LED unit through feature mapping.

[0057] The domain invariant features are input into the domain discriminator of the LED brightness prediction model. The domain discriminator contains three fully connected layers. Each layer uses the LeakyReLU activation function and the Dropout layer. The cross-operating condition feature vector is obtained through adversarial calculation.

[0058] The cross-operating condition feature vector is input into the encoding module of the brightness predictor in the LED brightness prediction model. The encoding module contains 6 residual blocks, each of which contains two layers of convolutional networks and a short-circuit connection. The deep features of the LED unit are obtained through feature encoding.

[0059] Perform self-attention calculation on deep features. Self-attention calculation includes position encoding and multi-head attention mechanism, and obtains attention feature mapping through feature weighting;

[0060] The attention feature map is input into the decoding module of the brightness predictor in the LED brightness prediction model. The decoding module contains 4 deconvolution layers. Each layer uses bilinear interpolation upsampling and residual connection to obtain the light intensity features of the LED unit through feature decoding.

[0061] The light intensity features are fused at multiple scales, and the multi-scale light intensity prediction values ​​of the LED units are obtained through feature pyramid networks and jump connections of three different scales. The multi-scale light intensity prediction values ​​are input into the output layer of the LED brightness prediction model for weighted averaging and normalization to obtain the predicted light intensity values.

[0062] In this example, the LED characteristic data set is divided into fixed-length feature sequences according to the time series to capture the dynamic changes of the LED working state. Assuming that the original LED characteristic data set contains feature data of N time steps, a sliding window method is used to divide the data. Set the length of the time window to L, then the feature sequence of the kth time window is expressed as:

[0063] ;

[0064] in, is the LED in time step The characteristic data of is the corresponding time window feature sequence. Through sliding window processing, a series of feature sequences of length L are obtained, each of which contains the dynamic characteristics of the LED unit in a fixed time period. The LED working status data is input into the feature extractor of the LED brightness prediction model for processing. The feature extractor consists of 4 convolutional layers, each of which uses the ReLU activation function and a batch normalization layer. The function of the convolutional layer is to extract local features from the input data and introduce nonlinear transformations through the ReLU activation function, while the batch normalization layer accelerates the training process and prevents the gradient from disappearing. Assume that The input of the layer convolution operation is , the output is , the calculation formula of the convolutional layer is:

[0065] ;

[0066] in, represents the convolution operation, is the convolution kernel, is the bias term. After being processed by 4 convolutional layers, the features of the LED unit are mapped to a high-dimensional space to form domain-invariant features. These features are similar under different working conditions and can effectively help the model predict light intensity under different working conditions. The domain-invariant features are input into the domain discriminator in the LED brightness prediction model. The domain discriminator contains 3 fully connected layers, each of which uses the LeakyReLU activation function and the Dropout layer. The LeakyReLU activation function avoids the "dead neuron" problem of the ReLU activation function during training, while the Dropout layer helps prevent overfitting. In the domain discriminator, through adversarial training, the model learns cross-condition feature vectors, that is, how to maintain the feature consistency of the LED unit under different working environments and states. The calculation formula for the fully connected operation of the layer is:

[0067] ;

[0068] in, is the output of the previous layer, is the weight matrix of the current layer, is the bias term. The cross-condition feature vector is input into the encoding module of the brightness predictor. The encoding module contains 6 residual blocks, each of which includes two layers of convolutional networks and a short-circuit connection. The residual block is designed to allow information to propagate across layers, thereby solving the gradient vanishing problem that occurs in deep networks. The output of each residual block is directly added to the input through a short-circuit connection, and the calculation formula is:

[0069] ;

[0070] in, is the input, and They are the convolution kernel, Represents the convolution operation. In this way, the model can extract deep features from multiple residual blocks and gradually build a deep feature representation of the LED unit. For the obtained deep features, self-attention calculation is used to enhance the expression ability of the model. The self-attention mechanism can capture the dependencies between the various parts of the input features by introducing position encoding and multi-head attention mechanism. During the calculation process, the feature vector of each position is weighted and summed with the feature vectors of other positions to obtain a weighted feature map. Assume that the position encoding is , the input features are , the calculation formula of the multi-head attention mechanism is:

[0071] ;

[0072] in, is the query vector, is the key vector, is a vector of values, is the dimension of the key vector. Through position encoding and multi-head attention mechanism, the model is able to capture the complex dependencies between LED units in multiple dimensions. The attention feature map is input to the decoding module of the brightness predictor. The decoding module contains 4 deconvolution layers, each of which uses bilinear interpolation upsampling and residual connection. The role of the deconvolution layer is to gradually restore the low-dimensional feature map to a higher dimension to generate the final light intensity prediction. Assume that The layer deconvolution operation is:

[0073] ;

[0074] Among them, DeConv is the deconvolution operation, is the deconvolution kernel, is the bias term. The model performs multi-scale fusion of light intensity features. By using three different scale feature pyramid networks and jump connections, features from different levels are fused to obtain the multi-scale light intensity prediction value of the LED unit. When the features of each scale are fused, a weighted average is performed to ensure that the features of different scales can be reasonably combined. Assume that the multi-scale light intensity prediction value is , and its calculation formula is:

[0075] ;

[0076] in, For the The light intensity characteristics of the scale, is the weighting coefficient. After multi-scale fusion, the light intensity prediction value is input into the output layer for weighted averaging and normalization to obtain the final LED unit light intensity prediction value.

[0077] In one example, the domain invariant features are input into the domain discriminator of the LED brightness prediction model. The domain discriminator contains three fully connected layers, each of which uses the LeakyReLU activation function and the Dropout layer. The cross-condition feature vector is obtained through adversarial calculation, including:

[0078] The domain-invariant features are flattened and a one-dimensional feature sequence is obtained by feature recombination.

[0079] The one-dimensional feature sequence is input into the first fully connected layer of the domain discriminator. The first fully connected layer contains 1024 neurons. The first layer features are calculated by the LeakyReLU activation function, and the first layer features are subjected to a Dropout operation with a probability of 0.5 to obtain the first layer output features.

[0080] The first layer output features are input into the second fully connected layer of the domain discriminator. The second fully connected layer contains 512 neurons. The second layer features are calculated by the LeakyReLU activation function, and the second layer features are subjected to a Dropout operation with a probability of 0.3 to obtain the second layer output features.

[0081] The output features of the second layer are input into the third fully connected layer of the domain discriminator. The third fully connected layer contains 256 neurons. The third layer features are calculated by the LeakyReLU activation function, and the third layer features are subjected to a Dropout operation with a probability of 0.2 to obtain the third layer output features.

[0082] Perform a maximum pooling operation on the output features of the third layer, and set the pooling kernel size of the maximum pooling operation to 2 to obtain a pooling feature vector. The pooling feature vector is then input into the gradient reversal layer to obtain adversarial features through the gradient reversal operation.

[0083] The adversarial features are subjected to L2 regularization, and the regularized features are obtained through normalization calculation. The regularized features are fused with the third-layer output features to obtain the cross-working condition feature vector.

[0084] In this example, the domain invariant features are flattened into a one-dimensional feature sequence. The original feature tensor is set to ,in is the height, is the width, is the number of channels. After flattening, the feature tensor becomes a one-dimensional sequence:

[0085] ;

[0086] All spatial dimensions and channel dimensions are compressed into a long vector for easy input to the subsequent fully connected layer. The one-dimensional feature sequence is input to the first fully connected layer of the domain discriminator. Assume that the one-dimensional feature sequence is , the first fully connected layer has 1024 neurons and uses the LeakyReLU activation function. The LeakyReLU activation function is defined as:

[0087] ;

[0088] in, is a small constant, usually 0.01. After this layer of calculation, the feature representation of the first layer is obtained:

[0089] ;

[0090] in, is the weight matrix of the first layer, is the bias term. To prevent overfitting, a Dropout operation is performed on the output features of this layer. Dropout is a regularization technique that reduces the complexity of the model by randomly discarding some neurons in the neural network. The formula for the Dropout operation is:

[0091] ;

[0092] in, It means randomly discarding some neurons with a probability of 50% to obtain the output features of the first layer. Input to the second fully connected layer. The number of neurons in the second layer is 512. The LeakyReLU activation function is used for calculation to obtain the characteristics of the second layer:

[0093] ;

[0094] in, is the weight matrix of the second layer, is the bias term. Similarly, the Dropout operation is performed on the output features of this layer, and the dropout rate is set to 0.3 to obtain the output features of the second layer:

[0095] ;

[0096] The second layer output features Input to the third fully connected layer. The number of neurons in the third layer is 256. The LeakyReLU activation function is used for calculation to obtain the characteristics of the third layer:

[0097] ;

[0098] in, is the weight matrix of the third layer, is the bias term. Similarly, the output features of this layer undergo a Dropout operation of 0.2:

[0099] ;

[0100] After processing by three layers of fully connected layers, the output features of the third layer are obtained In order to reduce redundant information and enhance the ability to extract features, the output of the third layer is subjected to a maximum pooling operation. The size of the feature map is reduced by taking the maximum value of each region. The size of the pooling kernel is set to 2, and the formula for the pooling operation is:

[0101] ;

[0102] Through maximum pooling, we get the pooled feature vector The pooled feature vector is input to the gradient reversal layer. The function of the gradient reversal layer is to reverse the sign of the gradient, which is usually used in adversarial training to make the model better adaptable between different fields. In this layer, the calculation formula of the reversal operation is:

[0103] ;

[0104] After gradient reversal, we get the adversarial feature In order to enhance the generalization ability of features, L2 regularization is performed on adversarial features. L2 regularization avoids overfitting by penalizing excessive weights. Its calculation formula is:

[0105] ;

[0106] in, is the L2 norm of the adversarial feature. The regularized features help the model maintain better generalization ability when facing new data. The regularized adversarial features are combined with the third layer output features Perform feature fusion. Combine feature information from different paths to improve the model's expressiveness. Feature fusion is achieved through addition or concatenation operations:

[0107] ;

[0108] In this way, the cross-operating condition characteristic vector is obtained , including the common characteristics of LED units under different working conditions.

[0109] In one example, spectrum analysis and resampling calculation are performed on the predicted light intensity value and the measured light intensity value to obtain a state determination result of the LED unit, including:

[0110] The difference between the predicted light intensity value and the measured light intensity value is calculated to obtain the light intensity deviation data of the LED unit, and the light intensity deviation data is sampled in the time domain to obtain the light intensity deviation sequence through a fixed time window.

[0111] The light intensity deviation sequence is spectrally decomposed to obtain the inherent frequency characteristics of the LED unit, and the inherent frequency characteristics are resampled according to the spatial position of the LED unit to obtain the spatial correlation characteristics;

[0112] The spatial correlation feature is matched with the preset fault feature data to obtain feature similarity data, and the feature similarity data is normalized to obtain the state quantization value of the LED unit;

[0113] The LED units are classified and marked based on the state quantification value to obtain the working state data of the LED units, and the working state data are compared and analyzed with the preset state threshold to obtain the state determination result of the LED units.

[0114] In this example, the difference between the predicted light intensity value and the measured light intensity value is calculated. Set the predicted light intensity value to And the measured light intensity is ,in Indicates the timestamp, and They are the predicted and actual measured values ​​at the corresponding time points. It is expressed as:

[0115] ;

[0116] The obtained deviation data reflects the difference between the actual working state of the LED unit and the expected state. The deviation data is sampled in the time domain, and a certain length of light intensity deviation sequence is intercepted by setting a fixed time window. Assume that the set time window length is , the deviation data is sampled and processed to obtain a light intensity deviation sequence containing a time series:

[0117] ;

[0118] This sequence describes the evolution of the light intensity deviation of the LED unit within a fixed time window. The light intensity deviation sequence spectrum is decomposed to extract the inherent frequency characteristics of the LED unit. Spectral analysis reveals the periodicity and frequency components in the signal, which helps to determine the working state of the LED unit. The light intensity deviation sequence is converted into the frequency domain using fast Fourier transform to obtain its spectrum representation:

[0119] ;

[0120] in, Indicates frequency, Represents the amplitude of the light intensity deviation at the corresponding frequency. The result of spectrum decomposition can help identify the main frequency components of light intensity fluctuations, that is, the inherent frequency characteristics. These inherent frequency characteristics reflect the dynamic characteristics of LED units. The inherent frequency characteristics are spatially resampled. Since the LED array is a spatially distributed system, the position of the LED unit has a direct impact on its light intensity characteristics, and the spectrum characteristics are spatially resampled. Assume that the spatial position of each unit in the LED array is , by resampling the spectrum characteristics of each LED unit, the spatial correlation characteristics are obtained. The resampling operation is implemented by the interpolation method. Assuming that the original spatial position characteristics are , the resampled features are ,in Represents the new sampling point, and the spatial correlation characteristics after resampling are:

[0121] ;

[0122] The spatial resampling operation can integrate the frequency characteristics of each LED unit to better understand the relative working status of each LED unit. Match the spatial correlation features with the preset fault feature data. The fault feature data is obtained by pre-training the features of the LED unit under different working conditions, and contains the features of different types of faults that may occur in the LED unit. For example, faults such as overheating, short circuit, and aging will show specific frequency characteristics and light intensity fluctuation patterns during the operation of the LED unit. By matching and calculating the spatial correlation features with these preset fault features, feature similarity data is obtained. Feature similarity is measured by calculating cosine similarity or Euclidean distance. Assume that the preset fault feature data is , the spatial correlation feature is , then the similarity calculation formula is:

[0123] ;

[0124] In this way, the similarity data obtained reflects the matching degree between the current state of the LED unit and the preset fault mode. The similarity data is normalized to eliminate the dimensional differences between different fault characteristics, making the data more unified and convenient for subsequent analysis. Assume that the similarity data is , the normalized similarity is:

[0125] ;

[0126] After normalization, the state quantification value of the LED unit is obtained. This quantification value is a digital representation of the working state of the LED unit, which can effectively reflect the health and performance of the LED unit. Based on the state quantification value, the LED unit is classified and marked, and its working state is compared with the preset state threshold. Assume that the preset working state threshold is , the state quantization value of the LED unit is , the status is determined by the following method:

[0127] ;

[0128] In this way, the state determination result of the LED unit is obtained, and the classification mark is used to determine whether it is in a normal working state or has a fault. For example, when the state quantization value is high, the LED unit is in a normal working state, and when the quantization value is low, it means that the LED unit has a fault or performance degradation.

[0129] In one example, based on the state determination result, the LED array is divided into a plurality of drive groups, and current equalization control is performed on each drive group through energy regulation of the energy storage capacitor to obtain LED drive current parameters, including:

[0130] Based on the state determination result, a first driving group and a second driving group are constructed, wherein the first driving group includes LED units in a normal working state, and the second driving group includes LED units in an abnormal working state, and initial driving group data is obtained by state classification;

[0131] A first energy storage capacitor and a second energy storage capacitor are set for the first drive group and the second drive group, wherein the first energy storage capacitor is used for energy exchange between channels, and the second energy storage capacitor is used for energy regulation within the group, and the drive group energy storage parameters are obtained through the double-layer energy storage structure;

[0132] The drive group energy storage parameter is input into the first current balancing control unit to perform current balancing control between channels. The first current balancing control unit includes multiple voltage sampling ports and current detection ports, and obtains current balancing data between channels through distributed control;

[0133] Performing a first phase shift control on the current balancing data between the channels, and obtaining a first set of driving signal sequences by staggered conduction timing;

[0134] The energy storage parameters of the driving group are input into the second current balancing control unit to perform current balancing control within the group. The second current balancing control unit includes multiple energy compensation branches, and current balancing data within the group is obtained through compensation control.

[0135] Performing a second phase shift control on the current balancing data within the group, and obtaining a second group of driving signal sequences through dynamic phase adjustment;

[0136] The first group of drive signal sequences and the second group of drive signal sequences are cooperatively controlled, and multi-channel drive control parameters are obtained through double closed-loop feedback. Energy is distributed between the first energy storage capacitor and the second energy storage capacitor based on the multi-channel drive control parameters to obtain LED drive current parameters.

[0137] In this example, the LED array is grouped according to the working state of the LED unit (normal or abnormal). Based on the state determination results of the LED unit, it is determined which units are in a normal working state and which units are in an abnormal working state. Greater than the preset threshold , then the LED unit is in a normal working state, otherwise, it is in an abnormal working state. Based on this classification result, the LED units are divided into a first drive group and a second drive group, wherein the first drive group contains LED units in a normal working state, and the second drive group contains LED units in an abnormal working state. Preliminary drive grouping data is obtained through the classification method. For these two drive groups, two different energy storage capacitors are set respectively, the first energy storage capacitor and the second energy storage capacitor. Among them, the first energy storage capacitor is used to exchange energy between different channels to ensure current balance between channels; the second energy storage capacitor is mainly used for energy regulation within the group to ensure the stability of the current of the LED units in each group. Assume that the capacitance of the first energy storage capacitor is , the capacitance of the second energy storage capacitor is By setting up a double-layer energy storage structure, the energy is effectively distributed and managed, and the energy storage parameters of the drive group are obtained. The energy storage parameters of the drive group are input into the first current sharing control unit, and the current sharing control between channels is performed. The first current sharing control unit includes multiple voltage sampling ports and current detection ports. Through these ports, the operating voltage and current of the LED unit are monitored in real time, and adjustments are made to ensure the current balance between each channel. The current of each channel is set to The first current balancing control unit adjusts the current of each channel through a distributed control algorithm to achieve current balancing between channels. The current balancing data between channels is obtained by feedback control of the current detection signal. The following conditions are met:

[0138] ;

[0139] in, is the number of channels, It is The current of each channel, is the average value of all channel currents. The first phase shift control is performed on the current balancing data between channels. In order to avoid electromagnetic interference caused by current fluctuations, the phase of multiple drive signals is adjusted by staggered conduction timing. Assume that the phase of the first group of drive signals is By adjusting the phase of the current waveform, the conduction timings of the signals are staggered, thereby reducing the interference of the system. Through the first phase shift control, the first set of driving signal sequences is obtained. , the sequence meets the predetermined timing requirements. After completing the current sharing control between channels, the current sharing control within the group is performed. The energy storage parameters of the drive group are input into the second current sharing control unit, which contains multiple energy compensation branches for adjusting the current distribution within the group. Through compensation control, the current balance of each LED unit in the group can be achieved to avoid the situation where some LED units are overloaded or underloaded. Set the current balance data within the group to , through the compensation control strategy, the current in the group is close to a desired equilibrium value The current sharing control data within the group is adjusted in real time through the compensation branch:

[0140] ;

[0141] The second phase shift control is performed on the current balancing data within the group. The second phase shift control makes the regulation signal of the current within the group smoother through dynamic phase adjustment to avoid instantaneous current fluctuations. Through this type of adjustment, the second group of drive signal sequences are obtained. , these signals work together with the first set of drive signal sequences to achieve optimal current distribution for the LED array. The first set of drive signal sequences and the second set of drive signal sequences are controlled collaboratively. By combining the two sets of drive signal sequences, a dual closed-loop feedback control strategy is adopted to coordinate and adjust the drive current of multiple channels. In this process, the energy distribution of the first energy storage capacitor and the second energy storage capacitor will also be dynamically adjusted to achieve current balance for the entire LED array. The operating state of the LED array is optimized through the calculated control parameters to ensure that each LED unit is within the ideal operating current range. Based on the multi-channel drive control parameters, the energy storage capacitor is energy distributed to obtain the drive current parameters of the LED. These drive current parameters provide a reasonable operating current for each LED unit to ensure the brightness uniformity and stability of the LED array under different working conditions. According to the final control algorithm, the drive current parameters Obtained by the following formula:

[0142]

[0143] in, It is a function based on feedback control, which integrates the results of channel current, intra-group current and balance control.

[0144] In one example, the LED array is divided into compensation areas according to the LED driving current parameters and the state determination result to obtain multiple compensation areas, and the brightness uniformity compensation parameters of each compensation area are calculated by the BSA algorithm, including:

[0145] Constructing a first optimization sequence and a second optimization sequence according to the LED driving current parameters and the state determination result, wherein the first optimization sequence includes the spatial position information of the LED unit, and the second optimization sequence includes the light intensity characteristic information of the LED unit, and obtaining the LED unit optimization data through sequence corresponding processing;

[0146] Perform partition clustering calculation on the LED unit optimization data, obtain the initial division result of the compensation area through brightness similarity analysis, and input the initial division result of the compensation area into the grouping optimization unit of the BSA algorithm, and obtain the optimal compensation area data through regional merging operation;

[0147] Based on the optimal compensation area data, iterative calculation is performed on the brightness uniformity of the LED array, a first compensation coefficient is obtained by optimizing the compensation parameters, and the first compensation coefficient is cyclically adjusted, and a second compensation coefficient is obtained by updating the compensation amount;

[0148] The first compensation coefficient and the second compensation coefficient are combined and optimized, and the combined compensation coefficient is obtained by weight distribution, and the combined compensation coefficient is processed according to the compensation area, and the compensation parameter matrix is ​​obtained by constrained optimization operation;

[0149] The compensation parameter matrix is ​​mapped to the LED units in each compensation area to obtain the brightness uniformity compensation parameters.

[0150] In this example, the first optimization sequence and the second optimization sequence are constructed based on the driving current parameters and state determination results of the LED array. The first optimization sequence mainly contains the spatial position information of the LED unit, because the spatial position of the LED directly affects the light intensity distribution, and the LED units located at different positions in the array will have different light intensity outputs. Set the position coordinates of the LED unit to ,in Indicates LED units, Respectively represent their coordinates in three-dimensional space. These spatial position information is used to optimize the layout of the LED array so that the spatial characteristics of different LED units can be considered in the process of brightness uniformity. The second optimization sequence contains the light intensity characteristic information of the LED unit, which reflects the actual brightness output characteristics of the LED unit. The light intensity characteristics of the LED unit are set to in Indicates The light intensity output value of each LED unit. The light intensity output value is closely related to the driving current of the LED unit. When constructing the second optimization sequence, the driving current parameter and light intensity characteristics Consider together. There is usually a functional relationship between the light intensity characteristics and the operating current of the LED unit:

[0151] ;

[0152] in, Represents the mapping relationship between light intensity and driving current. Through this relationship, the light intensity characteristics of each LED unit are obtained to construct the second optimization sequence. The two optimization sequences are combined through sequence correspondence processing to obtain the optimization data of each LED unit. By matching the spatial position information and the light intensity characteristic information in sequence, the optimization data of each LED unit is obtained. . Perform partition clustering calculation on the optimization data to analyze the brightness similarity between LED units and divide multiple compensation areas. Through cluster analysis, based on the brightness difference between LED units, LED units are divided into multiple areas, and the LED units in these areas have similar brightness output. Set the brightness similarity measure between LED units to ,in Measuring LED units and The specific calculation method is measured by Euclidean distance or cosine similarity:

[0153] ;

[0154] By using this metric, the LED units are divided into multiple clusters with similar brightness. The initial division of the compensation area is set as ,in Representative The compensation area contains a group of LED units with similar brightness characteristics. The initial compensation area division result is passed as input data to the group optimization unit of the BSA algorithm. In the BSA algorithm, the optimal compensation area data is obtained through the area merging operation. By continuously optimizing the area division, the light intensity difference between the LED units in the area is reduced, so that the brightness difference in each compensation area is minimized. The compensation area data output by the BSA algorithm Contains the optimized area division information. Based on the optimal compensation area data, perform iterative calculation of brightness uniformity. Calculate the first compensation coefficient of each compensation area through the compensation parameter optimization algorithm. Assume that the first compensation coefficient is , represents the compensation amount that needs to be adjusted in a certain compensation area to compensate for the brightness difference caused by different performance of LED units and environmental factors. The goal of calculating the first compensation coefficient is to optimize the overall brightness distribution of the LED array by minimizing the light intensity deviation in the area:

[0155] ;

[0156] in, It is the compensation area The target light intensity is determined by the average light intensity of all LED units in the area. The first compensation coefficient is adjusted cyclically, and the second compensation coefficient is obtained by updating the compensation amount. The second compensation coefficient further optimizes the uniformity of light intensity in the compensation area. The first compensation coefficient and the second compensation coefficient are combined and optimized to obtain the combined compensation coefficient through weight allocation. The combined compensation coefficient is calculated using the following weighted average formula:

[0157] ;

[0158] in, and is the weight coefficient, which indicates the importance of the first compensation coefficient and the second compensation coefficient in the final compensation. Through weight allocation, the influence of the first compensation coefficient and the second compensation coefficient is comprehensively considered to obtain an optimal compensation strategy. Perform boundary processing to ensure that the compensation parameters will not have too large discontinuities at the boundaries of the compensation area. Obtain the final compensation parameter matrix through constrained optimization operations. , which contains the compensation parameters of each LED unit in its compensation area. The compensation parameter matrix is ​​mapped to the LED units in each compensation area to obtain the brightness uniformity compensation parameters of the LED units, thus achieving brightness uniformity control of the entire LED array.

[0159] Reference Figure 2 This embodiment provides a device for controlling brightness uniformity of an LED light source, comprising:

[0160] Establish module 1, which is used to collect the driving current, PWM duty cycle and light intensity signal of each LED unit in the LED array, obtain the initial light intensity distribution data of the LED array through pixel-level light intensity detection, and establish an LED characteristic data set;

[0161] Prediction module 2 is used to extract working state features based on the LED characteristic data set to obtain LED working state data, and input the LED working state data into the LED brightness prediction model to predict the light intensity value to obtain the predicted light intensity value;

[0162] The calculation module 3 is used to perform spectrum analysis and resampling calculation on the predicted light intensity value and the measured light intensity value to obtain the state determination result of the LED unit;

[0163] The control module 4 is used to divide the LED array into a plurality of driving groups based on the state determination result, and perform current sharing control on each driving group through energy regulation of the energy storage capacitor to obtain LED driving current parameters;

[0164] The compensation module 5 is used to divide the LED array into compensation areas according to the LED driving current parameters and the state determination result to obtain a plurality of compensation areas, and calculate the brightness uniformity compensation parameters of each compensation area through the BSA algorithm.

[0165] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0166] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0167] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0168] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0169] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0170] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0171] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for controlling brightness uniformity of an LED light source, characterized in that: The following steps are involved: The driving current, PWM duty cycle and light intensity signal of each LED unit in the LED array are collected, and the initial light intensity distribution data of the LED array is obtained through pixel-level light intensity detection to establish an LED characteristic data set; Extract working state features based on the LED characteristic data set to obtain LED working state data, and input the LED working state data into an LED brightness prediction model to predict light intensity values ​​to obtain predicted light intensity values; Performing spectrum analysis and resampling calculation on the predicted light intensity value and the measured light intensity value to obtain a state determination result of the LED unit; Based on the state determination result, the LED array is divided into a plurality of drive groups, and current sharing control is performed on each drive group through energy regulation of the energy storage capacitor to obtain LED drive current parameters; According to the LED driving current parameter and the state determination result, the LED array is divided into compensation areas to obtain a plurality of compensation areas, and the brightness uniformity compensation parameter of each compensation area is calculated by the BSA algorithm; specifically comprising: constructing a first optimization sequence and a second optimization sequence according to the LED driving current parameter and the state determination result, wherein the first optimization sequence includes the spatial position information of the LED unit, and the second optimization sequence includes the light intensity characteristic information of the LED unit, and obtaining the LED unit optimization data through sequence correspondence processing; The LED unit optimization data is subjected to partition clustering calculation, and the initial division result of the compensation area is obtained through brightness similarity analysis, and the initial division result of the compensation area is input into the grouping optimization unit of the BSA algorithm, and the optimal compensation area data is obtained through regional merging operation; based on the optimal compensation area data, the LED array is subjected to brightness uniformity iterative calculation, and the first compensation coefficient is obtained through compensation parameter optimization, and the first compensation coefficient is cyclically adjusted, and the second compensation coefficient is obtained through compensation amount update calculation; the first compensation coefficient and the second compensation coefficient are combined and optimized, and the combined compensation coefficient is obtained through weight allocation, and the combined compensation coefficient is subjected to boundary processing according to the compensation area, and the compensation parameter matrix is ​​obtained through constraint optimization operation; the compensation parameter matrix is ​​mapped to the LED units in each compensation area to obtain brightness uniformity compensation parameters.

2. The LED light source brightness uniformity control method according to claim 1, characterized in that: The driving current, PWM duty cycle and light intensity signal of each LED unit in the LED array are collected, the initial light intensity distribution data of the LED array is obtained through pixel-level light intensity detection, and the LED characteristic data set is established, including: A high-resolution light intensity sensor array is provided for the LED array, and mapping relationship data between LED units and sensor units is obtained through spatial resolution matching operation; Applying multiple sets of driving current and PWM duty cycle combinations to the LED unit, and obtaining a light emission characteristic curve of the LED unit by measuring with a light intensity sensor array; The luminous characteristic curve is input into a space mapping processing unit, and the corresponding relationship data between the position of the LED unit and its light intensity output is obtained by interpolation calculation; The light intensity of the LED array is sampled at different ambient temperatures, the temperature influence factor of the LED unit is obtained by temperature compensation calculation, and the LED unit is sampled and measured multiple times based on the temperature influence factor, and the batch characteristic parameters of the LED unit are obtained by standard deviation analysis; The mapping relationship data, the corresponding relationship data and the batch characteristic parameters are fused to obtain a characteristic matrix of the LED unit, and a correlation analysis is performed on the characteristic matrix and the temperature influencing factor. The dynamic characteristic parameters of the LED unit are obtained by feature extraction, and the dynamic characteristic parameters are stored in a time series to obtain an LED characteristic data set.

3. The LED light source brightness uniformity control method according to claim 2, characterized in that: The extracting working state features based on the LED characteristic data set to obtain LED working state data, and inputting the LED working state data into an LED brightness prediction model to predict the light intensity value to obtain a predicted light intensity value, comprises: The LED characteristic data set is divided into feature sequences of fixed length according to the time series, and the LED working state data is obtained by sliding window processing; Inputting the LED working state data into a feature extractor of an LED brightness prediction model for processing, the feature extractor comprising four convolutional layers, each of which uses a ReLU activation function and a batch normalization layer, and obtaining domain-invariant features of LED units through feature mapping; Inputting the domain invariant features into the domain discriminator of the LED brightness prediction model, the domain discriminator comprises three fully connected layers, each layer uses a LeakyReLU activation function and a Dropout layer, and obtains a cross-operating condition feature vector through adversarial calculation; Inputting the cross-operating condition feature vector into the encoding module of the brightness predictor in the LED brightness prediction model, the encoding module comprises 6 residual blocks, each residual block comprises two layers of convolutional networks and a short circuit connection, and obtaining the deep features of the LED unit through feature encoding; Performing self-attention calculation on the deep features, wherein the self-attention calculation includes position encoding and a multi-head attention mechanism, and obtaining an attention feature map by feature weighting; Input the attention feature map into a decoding module of a brightness predictor in the LED brightness prediction model, wherein the decoding module comprises four deconvolution layers, each layer uses bilinear interpolation upsampling and residual connection, and obtains the light intensity feature of the LED unit through feature decoding; The light intensity features are fused at multiple scales, and the multi-scale light intensity prediction values ​​of the LED units are obtained through feature pyramid networks and jump connections of three different scales. The multi-scale light intensity prediction values ​​are input into the output layer of the LED brightness prediction model for weighted averaging and normalization to obtain the predicted light intensity values.

4. The LED light source brightness uniformity control method according to claim 3, characterized in that: The domain invariant features are input into the domain discriminator of the LED brightness prediction model, and the domain discriminator includes 3 fully connected layers, each layer uses a LeakyReLU activation function and a Dropout layer, and a cross-operating condition feature vector is obtained through adversarial calculation, including: Performing dimension flattening processing on the domain invariant features, and obtaining a one-dimensional feature sequence by feature recombination; Input the one-dimensional feature sequence into the first fully connected layer of the domain discriminator, the first fully connected layer includes 1024 neurons, calculate the first layer features through the LeakyReLU activation function, and perform a Dropout operation with a probability of 0.5 on the first layer features to obtain the first layer output features; Input the output features of the first layer into the second fully connected layer of the domain discriminator, the second fully connected layer contains 512 neurons, calculate the second layer features through the LeakyReLU activation function, and perform a Dropout operation with a probability of 0.3 on the second layer features to obtain the second layer output features; Input the output features of the second layer into the third fully connected layer of the domain discriminator, the third fully connected layer includes 256 neurons, calculate the third layer features through the LeakyReLU activation function, and perform a Dropout operation with a probability of 0.2 on the third layer features to obtain the third layer output features; Performing a maximum pooling operation on the output features of the third layer, wherein the pooling kernel size of the maximum pooling operation is set to 2, to obtain a pooling feature vector, and inputting the pooling feature vector into a gradient reversal layer, to obtain an adversarial feature through a gradient reversal operation; The adversarial feature is subjected to L2 regularization processing, a regularized feature is obtained through normalization calculation, and the regularized feature is fused with the third-layer output feature to obtain a cross-operating condition feature vector.

5. The LED light source brightness uniformity control method according to claim 4, characterized in that: The performing spectrum analysis and resampling calculation on the predicted light intensity value and the measured light intensity value to obtain the state determination result of the LED unit includes: Calculating the difference between the predicted light intensity value and the measured light intensity value to obtain light intensity deviation data of the LED unit, and performing time domain sampling on the light intensity deviation data to obtain a light intensity deviation sequence through a fixed time window interception; Performing spectral decomposition on the light intensity deviation sequence to obtain the inherent frequency characteristics of the LED unit, and resampling the inherent frequency characteristics according to the spatial position of the LED unit to obtain the spatial correlation characteristics; Performing a matching operation on the spatial correlation feature and preset fault feature data to obtain feature similarity data, and performing normalization processing on the feature similarity data to obtain a state quantization value of the LED unit; The LED units are classified and marked based on the state quantification values ​​to obtain working state data of the LED units, and the working state data are compared and analyzed with a preset state threshold to obtain a state determination result of the LED units.

6. The LED light source brightness uniformity control method according to claim 5, characterized in that: Based on the state determination result, the LED array is divided into a plurality of drive groups, and current sharing control is performed on each drive group through energy regulation of the energy storage capacitor to obtain LED drive current parameters, including: Based on the state determination result, a first driving group and a second driving group are constructed, wherein the first driving group includes LED units in a normal working state, and the second driving group includes LED units in an abnormal working state, and initial driving group data is obtained by state classification; A first energy storage capacitor and a second energy storage capacitor are provided for the first drive group and the second drive group, wherein the first energy storage capacitor is used for energy exchange between channels, and the second energy storage capacitor is used for energy regulation within the group, and energy storage parameters of the drive group are obtained through a double-layer energy storage structure; Inputting the energy storage parameters of the drive group into a first current balancing control unit to perform current balancing control between channels, wherein the first current balancing control unit comprises a plurality of voltage sampling ports and a current detection port, and obtains current balancing data between channels through distributed control; Performing a first phase shift control on the current balancing data between the channels to obtain a first set of driving signal sequences by staggered conduction timing; Inputting the energy storage parameter of the driving group into a second current balancing control unit to perform intra-group current balancing control, wherein the second current balancing control unit comprises a plurality of energy compensation branches, and obtains intra-group current balancing data through compensation control; Performing a second phase shift control on the current balancing data within the group, and obtaining a second group of driving signal sequences through dynamic phase adjustment; The first group of drive signal sequences and the second group of drive signal sequences are cooperatively controlled, multi-channel drive control parameters are obtained through double closed-loop feedback, and energy is distributed between the first energy storage capacitor and the second energy storage capacitor based on the multi-channel drive control parameters to obtain LED drive current parameters.

7. A device for controlling brightness uniformity of an LED light source, characterized in that: For implementing the steps of the method according to any one of claims 1 to 6, the LED light source brightness uniformity control device comprises: Establish a module for collecting the driving current, PWM duty cycle and light intensity signal of each LED unit in the LED array, obtain the initial light intensity distribution data of the LED array through pixel-level light intensity detection, and establish an LED characteristic data set; A prediction module, used for extracting working state features based on the LED characteristic data set to obtain LED working state data, and inputting the LED working state data into an LED brightness prediction model to predict the light intensity value to obtain a predicted light intensity value; A calculation module, used for performing spectrum analysis and resampling calculation on the predicted light intensity value and the measured light intensity value to obtain a state determination result of the LED unit; A control module, for dividing the LED array into a plurality of drive groups based on the state determination result, and performing current sharing control on each drive group by energy regulation of an energy storage capacitor to obtain LED drive current parameters; The compensation module is used to divide the LED array into compensation areas according to the LED driving current parameter and the state determination result to obtain multiple compensation areas, and calculate the brightness uniformity compensation parameter of each compensation area through the BSA algorithm.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images