Low-power-consumption LED screen control system and method for outdoor display

Abstract

The invention relates to the field of LED screen control, and discloses a low-power-consumption LED screen control system and method for outdoor display, and the method comprises the steps: additionally arranging a radio frequency signal intensity detection unit, dynamically adjusting node connection, triggering self-inspection, comparing adjacent node data to correct errors, and building a polynomial regression model to correct a light intensity detection value. Based on the corrected light intensity detection value, constructing a reward function, calculating a state adjustment value, dynamically adjusting a driving voltage, constructing a temperature compensation model, determining a compensation coefficient, selecting a plurality of random feature subset training models and verification set errors to evaluate model performance, dynamically adjusting a tolerance value according to verification set error changes, and optimizing a model training process; the method comprises the steps of collecting environment light intensity data, predicting a future illumination change trend, dynamically adjusting a piecewise function turning point for optimal estimation, evaluating the influence of a geographical environment on illumination, optimizing and adjusting parameters, dynamically adjusting turning point parameters according to a prediction error, capturing a turning point of illumination intensity change, and analyzing and reducing power consumption.

Description

Technical Field

[0001] The present invention relates to the technical field of low-power LED screen control, and in particular to a low-power LED screen control system and method for outdoor display. Background Art

[0002] With the widespread use of outdoor LED displays, the performance requirements for their control systems are increasing, including data acquisition accuracy, transmission stability, energy consumption control, and environmental adaptability. Traditional LED screen control systems have many technical limitations and are unable to meet the diverse needs of complex outdoor scenes. In terms of data processing, sensor data is easily interfered with by environmental factors. At the transmission level, signal stability is poor and anti-interference capabilities are weak. In terms of energy consumption control, there is a lack of dynamic optimization mechanisms. In terms of environmental adaptation, the response to changes in lighting, traffic flow, etc. is not intelligent enough.

[0003] In the existing technology, the existing sensor modules do not have a compensation mechanism designed for environmental interference, light intensity detection is easily affected by temperature, and there is a lack of a temperature compensation model, which leads to large deviations in the detection data and affects the accuracy of the LED screen brightness adjustment; the existing network topology is mostly fixed and cannot be dynamically adjusted according to the node signal strength, transmission delay and energy consumption; in complex outdoor electromagnetic environments, signal interference leads to unstable transmission, high delay, and even data loss, which cannot guarantee the real-time data transmission needs of the LED screen; there is a lack of intelligent spectrum sensing and prediction technology to predict interference signals in advance; traditional LED screen drive voltages mostly use fixed or simple graded adjustment methods, which cannot be dynamically optimized according to the actual working status; the nonlinear relationship between temperature and power consumption is not considered, and there is a lack of a temperature compensation model, which cannot accurately calculate the power consumption fluctuations caused by temperature changes; the existing LED screen brightness adjustment is mostly based on fixed thresholds or simple ambient light detection, and does not combine historical lighting data, weather and time factors to predict lighting change trends; in complex lighting environments, the brightness cannot be adjusted in time, which not only affects the visual effect but also causes energy waste;

[0004] In view of this, it is necessary to provide a low-power LED screen control system and method for outdoor display. Summary of the Invention

[0005] The purpose of the present invention is to provide a low-power LED screen control system and method for outdoor display. In order to solve the above-mentioned existing technical problems, the present invention is achieved through the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a low-power LED screen control method for outdoor display, which specifically includes the following steps:

[0007] Add a radio frequency signal strength detection unit to dynamically adjust node connections, trigger self-test, compare adjacent node data to correct errors, and establish a polynomial regression model to correct light intensity detection values;

[0008] Based on the corrected light intensity detection value, a reward function is constructed to calculate the state adjustment value, dynamically adjust the driving voltage, build a temperature compensation model, determine the compensation coefficient, train the model by selecting multiple random feature subsets, evaluate the model performance by the validation set error, and dynamically adjust the patience value according to the change in the validation set error to optimize the model training process;

[0009] Collect ambient light intensity data, predict future lighting change trends, dynamically adjust the turning points of the piecewise function for optimal estimation, evaluate the impact of the geographical environment on lighting, optimize the adjustment parameters, dynamically adjust the turning point parameters according to the prediction error, capture the turning points of light intensity changes, and analyze and reduce power consumption.

[0010] Furthermore, the method for correcting errors by comparing adjacent node data is:

[0011] For the module's built-in three-axis accelerometer, prevent installation deviation and monitor the sensor's vibration status in real time;

[0012] When the accelerometer detects that the vibration amplitude exceeds the set vibration amplitude standard, the system automatically triggers the self-test program;

[0013] By comparing the collected data of adjacent nodes, it is determined whether the own data has errors due to vibration;

[0014] The polynomial regression model of light intensity detection value and temperature is established by formula to calculate the corrected light intensity detection value;

[0015] Furthermore, the method for correcting the light intensity detection value is:

[0016] By Formula I c =I m +a1T 2 +b1T+c1 establishes a polynomial regression model of light intensity detection value and temperature to calculate the corrected light intensity detection value I c ;

[0017] Among them, I m is the original measured light intensity detection value, T is the ambient temperature, a1, b1, c1 are the correction coefficients obtained through experimental calibration;

[0018] Furthermore, the method for calculating the state adjustment value is:

[0019] The deep Q network in reinforcement learning is used to dynamically adjust the drive voltage VDD. The adjustment range of the drive voltage VDD (3.3-12V) is divided into a discrete state space, with each 0.1V being a discrete state. The action space is set to increase, decrease, or maintain VDD.

[0020] Construct the reward function R through the formula Get the status adjustment value;

[0021] Among them, Pp is the power consumption at the previous moment, Pc is the power consumption at the current moment, Ep is the prediction error at the previous moment, and Ec is the prediction error of the model at the current moment. is the reward weight for power consumption reduction, is the penalty weight for changes in prediction error;

[0022] Furthermore, the method for determining the compensation coefficient is:

[0023] A temperature compensation model is constructed based on polynomial regression, and the power consumption change caused by temperature change is calculated through the formula;

[0024] Use a constant temperature chamber to collect power consumption data at more temperature points within a preset temperature range of 25-85°C. Use the least squares method to fit the data and determine the compensation coefficient.

[0025] Furthermore, the method for evaluating model performance is:

[0026] In the color gamut coverage calculation, deep learning image segmentation is used to segment the actual display color gamut and the DCI-P3 standard color gamut into binary images. The number of pixels in the actual display color gamut and the number of pixels in the overlapping part are obtained. The number of pixels in the overlapping part is then compared with the number of pixels in the DCI-P3 standard color gamut to obtain the color gamut coverage.

[0027] Based on the isolation forest algorithm, a feature selection mechanism is used to construct an isolation forest with 100 subtrees and a maximum depth of 8 layers;

[0028] By calculating the information gain rate of image data features, features that contribute more than the preset standard to anomaly detection are screened out;

[0029] Furthermore, the method for optimizing the model training process is:

[0030] Divide the original dataset into training set, validation set and test set;

[0031] Select M random feature subsets;

[0032] For each feature subset S, the model prediction value is obtained by training the deep learning model based on the original data set;

[0033] The validation set error is obtained by calculating the average of the absolute values of the differences between the model prediction value and the actual power consumption value;

[0034] Adaptive early stopping is used to dynamically adjust the patience value based on the changing trend of the validation set error during model training;

[0035] Furthermore, the method for optimizing the adjustment parameters is:

[0036] In the prediction stage, the state prediction value at the current moment is calculated based on the state at the previous moment and the system dynamic model;

[0037] In the update phase, the state prediction value is corrected in combination with the current observation data to obtain the corrected state value and suppress the interference of noise on the data;

[0038] Fourier encoding is used to process seasonal characteristics. Fourier transform decomposes the seasonal periodic signal into a combination of sine and cosine waves of different frequencies.

[0039] By extracting the coefficients of the combined components of sine and cosine waves of different frequencies, the seasonal characteristics can be quantitatively represented.

[0040] Furthermore, the method for analyzing and reducing power consumption is:

[0041] Using the dynamic turning point algorithm, define a differentiable piecewise function Among them, μ i To control the amplitude of the function at each turning point, β x To adjust the steepness of the function, θ i is the turning point parameter, representing the key node of light intensity change;

[0042] Optimization is performed through the back propagation algorithm. Back propagation transmits the error signal from the output layer to the input layer according to the prediction error between the predicted value and the true value, calculates the gradient of the adjustment parameter, and then corrects the adjustment parameter value.

[0043] Based on constraints Among them, η is the learning rate of 0.01, which is used to control the step size of parameter update. Represents the illumination matching loss function L for the turning point parameter θ i The gradient of the illumination matching loss function L is the mean absolute error between the predicted illumination intensity and the current actual illumination intensity;

[0044] When it is detected that the overall brightness of the picture is low and the contrast is not high, the backlight brightness is reduced, and the local dimming algorithm is used to individually dim the areas that need to be highlighted.

[0045] In a second aspect, an embodiment of the present invention provides a low-power LED screen control system for outdoor display, which specifically includes the following modules:

[0046] Data acquisition module: responsible for acquiring collected data;

[0047] Data pre-processing module: Add a radio frequency signal strength detection unit, dynamically adjust node connections, and trigger self-test;

[0048] Data analysis module: Compare adjacent node data to correct errors and establish a polynomial regression model to correct light intensity detection values

[0049] Transmission Optimization Module: Builds a temperature compensation model, determines the compensation coefficient, trains the model by selecting multiple random feature subsets, evaluates model performance based on the validation set error, dynamically adjusts the patience value based on the validation set error, and optimizes the model training process.

[0050] Power consumption control module: Based on the corrected light intensity detection value, a reward function is constructed to calculate the state adjustment value and dynamically adjust the drive voltage. The module also collects ambient light intensity data, predicts future lighting trends, dynamically adjusts the turning point of the piecewise function for optimal estimation, evaluates the impact of the geographical environment on lighting, and optimizes the adjustment parameters.

[0051] Adaptive control module: Dynamically adjusts turning point parameters based on prediction errors, captures turning points of light intensity changes, and analyzes and reduces power consumption.

[0052] Beneficial effects of the present invention:

[0053] 1. Using reinforcement learning algorithms with signal strength, data transmission delay, and energy consumption as reward functions, dynamically adjust the connection relationship between nodes to form an adaptive cellular topology, improve signal transmission quality, and optimize data transmission delay and energy consumption; prevent installation offset and monitor vibration status in real time, automatically trigger the self-test program, and use data repair algorithms to make corrections; establish a polynomial regression model and install a high-precision temperature sensor to collect ambient temperature data in real time to correct the light intensity detection value; use multimodal data fusion methods, combined with binocular cameras and millimeter-wave radars, to construct a three-dimensional model of the human body, calculate crowd density, and avoid counting errors caused by human occlusion; set weights for different channels to calculate the difference between frames; calculate the color gamut coverage of the actual display color gamut and the DCI-P3 standard color gamut to improve the accuracy of color processing; use feature selection mechanisms to screen features that contribute most to anomaly detection, reduce interference from irrelevant features, and dynamically adjust the anomaly score threshold according to data distribution to detect abnormal data;

[0054] 2. Utilize a dynamic parameter adjustment mechanism to adjust the weight of the light intensity sliding window according to the rate of change of light intensity, so that the weight distribution adapts to the dynamic changes in light intensity and balances the weights of historical image data and real-time image data; by constructing a reasonable reward function, power consumption is reduced while ensuring model prediction accuracy; combine the adaptive early stopping method to dynamically adjust the patience value according to the validation set error to optimize model training, avoid overtraining while ensuring model performance, reduce computing resource consumption and thus power consumption; adjust the backlight brightness in real time according to the brightness distribution and contrast of the picture, and perform local dimming on areas that need to be highlighted, reducing power consumption while ensuring visual effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0056] Figure 1 This is a flowchart of a low-power LED screen control method for outdoor display provided by Example 1 of the present invention;

[0057] Figure 2 This is a structural diagram of a low-power LED screen control system for outdoor display provided by Example 2 of the present invention. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0059] Example 1

[0060] like Figure 1 As shown, an embodiment of the present invention provides a low-power LED screen control method for outdoor display, which specifically includes the following steps:

[0061] Step 1: Add a radio frequency signal strength detection unit to dynamically adjust node connections. When vibration exceeds the standard, a self-check is triggered. The error is corrected by comparing the data of adjacent nodes. A polynomial regression model is established to correct the light intensity detection value.

[0062] In a specific embodiment, an adaptive dynamic topology adjustment mechanism is used to adjust the topology every 0.5m. 2 Add a radio frequency signal strength detection unit to the deployed integrated sensor module to obtain collected data and monitor the signal strength between nodes in real time;

[0063] The collected data includes: environmental data, equipment status data, human body detection data and image data;

[0064] Using reinforcement learning algorithms, with signal strength, data transmission delay, and energy consumption as reward functions, the connection relationship between nodes is dynamically adjusted to form an adaptive cellular topology;

[0065] For the module's built-in three-axis accelerometer, prevent installation deviation and monitor the sensor's vibration status in real time;

[0066] When the accelerometer detects that the vibration amplitude exceeds the set vibration amplitude standard, the system automatically triggers the self-test program;

[0067] By comparing the collected data of adjacent nodes, it is determined whether the own data has errors due to vibration;

[0068] If there is an error, the data repair algorithm is activated, using the historical data and current data of adjacent nodes to correct the error data through weighted average, with the weight determined based on the node distance and data similarity;

[0069] The dual photodiode array based on light intensity detection is easily affected by temperature, resulting in a decrease in detection accuracy. A polynomial regression model was established, and a high-precision temperature sensor was added to the sensor module to collect environmental data in real time.

[0070] Specifically, through formula I c =I m +a1T 2 +b1T+c1 establishes a polynomial regression model of light intensity detection value and temperature to calculate the corrected light intensity detection value I c ;

[0071] Among them, I m is the original measured light intensity detection value, T is the ambient temperature, a1, b1, and c1 are the correction coefficients obtained through experimental calibration. The light intensity is measured at different temperatures T using the controlled variable method and compared with the known standard value to ensure that factors other than temperature in the experimental environment remain unchanged;

[0072] Using multimodal data fusion methods, a binocular camera with crowd counting is used to obtain environmental images, the human body outline is recognized through the YOLOv7 model, and the distance information of the human body is detected using millimeter-wave radar;

[0073] It should be noted that the YOLOv7 model is a model for real-time target detection. It treats target detection as a regression problem and directly predicts bounding boxes and category probabilities from images through a single neural network.

[0074] The 2D human body contour identified by the camera is integrated with the 3D distance information detected by the millimeter-wave radar to construct a 3D model of the human body.

[0075] Calculate crowd density through a three-dimensional model to avoid counting errors caused by human body occlusion, further reducing density errors;

[0076] In real-time transmission, intelligent spectrum sensing technology is used, and the sensor module monitors the interference signal strength and spectrum occupancy ratio within the frequency band in real time;

[0077] Use the spectrum prediction model of deep learning to predict the changing trend of interference signals in advance;

[0078] If interference is detected, it automatically switches to a sub-band where no interference is detected for data transmission, and adjusts carrier aggregation parameters to optimize transmission rate and latency;

[0079] For example, the light intensity detection sampling rate is set to 1kHz. In the temperature compensation model, a=-0.001,b=0.05,c=0.1 are obtained through experimental calibration. For the light intensity detection value I measured at any time m =500lux, ambient temperature T = 25℃, then the corrected light intensity detection value I is calculated by the polynomial regression model of light intensity detection value and temperature. c =500-0.001×25×25+0.05×25+0.1=501.25lux;

[0080] Based on crowd statistics, the baseline distance of the binocular camera is B = 12cm, and the human body distance d is calculated by the triangulation principle.

[0081] Where f is the focal length of the camera, and p is the parallax of the human feature points in the left and right environment images;

[0082] Combining the human outline identified by the YOLOv7 model and the distance information detected by the millimeter-wave radar, the two-dimensional outline is projected into three-dimensional space, the volume of the space occupied by the human body is calculated, and then the crowd density is calculated based on the volume of the monitored area;

[0083] For example, the monitoring area volume is V = 10m 3 , the total volume of the human body is obtained by fusion calculation as V p =1m 3 , then the crowd density is ρ p =0.1 person / m 3 ;

[0084] The fast Sobel algorithm is used, and the principle of integral image is utilized to convert convolution operations into addition and subtraction operations. Based on the gray-level co-occurrence matrix (GLCM), dimensionality reduction technology is used to reduce computational complexity. The environmental image is mapped from a high-dimensional space to a low-dimensional space, and principal component analysis is performed on the image. The contrast parameters of the gray-level co-occurrence matrix (GLCM) are calculated based on the low-dimensional space, thus reducing the amount of calculation while ensuring accuracy.

[0085] It should be noted that texture variance is a statistic used to describe the texture characteristics of an image, reflecting the roughness or complexity of the texture in the image. Gray-level co-occurrence matrix is a statistical method used to describe the spatial dependency of gray levels in an image. It describes the texture characteristics of an image by calculating the frequency of occurrence of pixel pairs with specific gray values in a given spatial relationship.

[0086] Based on the Lucas-Kanade algorithm used in time dimension analysis to calculate motion vector amplitude, the pyramid Lucas-Kanade algorithm is introduced to improve the detection capability of large displacement motion by constructing an image pyramid, starting from low-resolution images and gradually refining them towards high-resolution images.

[0087] Specifically, in the HSV color space, the weighted histogram chi-square distance is used to capture color changes. According to the different sensitivities of the human eye to different color channels, weights are set for the H, S, and V channels respectively. The weights are preset to w H =0.5, w S =0.3, w V =0.2, then the weighted inter-frame difference is calculated by the formula IFD=w H *IFD H +w S *IFD S +w V *IFD V Calculate the difference between frames;

[0088] Using deep learning-based image segmentation technology, the actual display color gamut and the DCI-P3 standard color gamut are divided separately, and the overlapping area is accurately calculated by pixel counting;

[0089] By fitting the blackbody radiation trajectory and using the Bayesian estimation method, combined with prior knowledge and current measurement data, the color temperature standard deviation is calculated;

[0090] For example, the image size is M×N. After the fast Sobel algorithm is used, the computational time complexity is reduced from O(M×N) to close to O(1) after the integral image is constructed. In the inter-frame difference IFD calculation, the histogram chi-square distance IFD of the two frames of the environment image in the H channel is H =0.2, S channel IFD S =0.1, V channel IFD V =0.15, then the weighted inter-frame difference IFD=0.5×0.2+0.3×0.1+0.2×0.15=0.16;

[0091] In the color gamut coverage calculation, deep learning image segmentation is used to segment the actual display color gamut and the DCI-P3 standard color gamut into binary images. The number of pixels in the actual display color gamut and the number of pixels in the overlapping part are obtained. The number of pixels in the overlapping part is then compared with the number of pixels in the DCI-P3 standard color gamut to obtain the color gamut coverage.

[0092] Based on the isolation forest algorithm, a feature selection mechanism is used to construct an isolation forest with 100 subtrees and a maximum depth of 8 layers;

[0093] By calculating the information gain rate of image data features, features that contribute to anomaly detection beyond the preset standard are screened out, reducing the interference of irrelevant features;

[0094] Dynamically adjust the anomaly score threshold according to the image data distribution, and calculate the adjusted threshold θ based on the adaptive threshold adjustment strategy using the formula θ = θ0 + β × σ;

[0095] Where θ0 = 0.65 is the initial threshold, β is the adjustment coefficient, which is preset to 0.1, and σ is the standard deviation of the data;

[0096] Using the dynamic parameter adjustment mechanism, the light intensity sliding window weight distribution is calculated by the formula Calculate the sliding window weight to adapt to the dynamic changes of light intensity;

[0097] According to the intensity of the light intensity change, the value of α is adjusted by calculating the rate of change of the light intensity;

[0098] If the light intensity change rate is greater than the preset standard range, the value of α is increased to give more weight to the recent image data;

[0099] If the light intensity change rate is less than the preset standard range, reduce the value of α to balance the weights of image history data and real-time image data;

[0100] For example, in the abnormal cleaning, the standard deviation of the data is obtained as σ = 0.2. According to the adaptive threshold adjustment strategy, the adjusted abnormal score threshold is θ = 0.65 + 0.1 × 0.2 = 0.67. In the light intensity sliding window weight allocation, the current time t is obtained. c , based on the historical moment t c -10s, the calculated light intensity change rate is greater than the preset standard range, adjust α=0.3, calculate the weight

[0101] Step 2: Based on the corrected light intensity detection value, a reward function is constructed, the state adjustment value is calculated, the driving voltage is dynamically adjusted, and a temperature compensation model is constructed. The compensation coefficient is determined by least squares fitting. The model is trained by selecting multiple random feature subsets and calculating the validation set error to evaluate the model performance. The patience value is dynamically adjusted according to the change in the validation set error to optimize the model training process.

[0102] In a specific embodiment, a deep Q network in reinforcement learning is used to dynamically adjust the driving voltage VDD; the adjustment range of the driving voltage VDD (3.3-12V) is divided into a discrete state space, with each 0.1V being a discrete state, and the action space is set to increase, decrease, or maintain VDD;

[0103] Construct the reward function R through the formula Get the state adjustment value R;

[0104] Among them, Pp is the power consumption at the previous moment, Pc is the power consumption at the current moment, Ep is the prediction error at the previous moment, and Ec is the prediction error of the model at the current moment. is the reward weight for power consumption reduction, is the penalty weight for the change in prediction error, preset

[0105] Dynamically adjust the drive voltage VDD based on the obtained state adjustment value R. By continuously interacting with the environment, the optimal drive voltage VDD adjustment strategy is learned to reduce power consumption while ensuring model prediction accuracy.

[0106] The temperature compensation model is constructed based on polynomial regression, and the formula ΔP=a2T is used. 2 +b2T+c2 calculates the power consumption change caused by temperature change;

[0107] A constant temperature chamber was used to collect power consumption data at more temperature points within a preset temperature range of 25-85°C. The data was fitted using the least squares method to determine the compensation coefficients a2, b2, and c2. Polynomial regression was used to simulate the nonlinear relationship between temperature and power consumption changes, improving the accuracy of temperature compensation.

[0108] The synthetic minority oversampling technique SMOTE is used to oversample the minority class samples to generate new minority power consumption samples and obtain the original data set;

[0109] Divide the original dataset into training set, validation set and test set;

[0110] Select M random feature subsets;

[0111] For each feature subset S, the model prediction value is obtained by training the deep learning model based on the original data set;

[0112] The validation set error is obtained by calculating the average of the absolute values of the differences between the model prediction value and the actual power consumption value;

[0113] Adaptive early stopping is used to dynamically adjust the patience value based on the changing trend of the validation set error during model training;

[0114] For example, if the validation set error continues to decrease, the patience value is reduced to speed up the model convergence; if the validation set error fluctuates greatly, the patience value is increased to prevent the model from stopping training prematurely.

[0115] Step 3: Collect ambient light intensity data, predict future lighting trends, dynamically adjust the turning point of the piecewise function for optimal estimation, evaluate the impact of the geographical environment on lighting, optimize the adjustment parameters, and dynamically adjust the turning point parameters based on the prediction error and the lighting matching loss function to timely capture the turning point of light intensity changes, analyze the brightness distribution and contrast of the displayed content, and reduce power consumption;

[0116] A dynamic threshold adjustment mechanism is introduced. By deploying ambient light sensors to collect real-time ambient light intensity data, the mechanism combines historical light data with current scene information, including time and weather. Long short-term memory (LSTM) networks are used to predict light change trends within the future forecast period, dynamically adjusting the turning point of the piecewise function.

[0117] The real-time data acquisition layer serves as the system's sensory antennae, responsible for collecting ambient light and weather information;

[0118] To deal with the instantaneous interference noise in the environment, the system integrates the Kalman filter algorithm;

[0119] The Kalman filter is based on the linear system state space model and makes the best estimate of the data collected by the sensor through two steps: prediction and update.

[0120] Specifically, in the prediction phase, the state prediction value at the current moment is calculated based on the state at the previous moment and the system dynamic model;

[0121] In the update phase, the state prediction value is corrected in combination with the current observation data to obtain the corrected state value and suppress the interference of noise on the data;

[0122] Fourier encoding is used to process seasonal characteristics. Fourier transform decomposes the seasonal periodic signal into a combination of sine and cosine waves of different frequencies.

[0123] By extracting the coefficients of the combined components of sine and cosine waves of different frequencies, the seasonal characteristics can be quantitatively represented.

[0124] For the holiday mode, one-hot encoding is used to map each holiday category into a unique binary vector, where the position of the corresponding category is mapped to 1 and the rest are mapped to 0;

[0125] Based on GPS coordinate information, the spherical coordinate system conversion algorithm is used to accurately analyze the sunshine angle, taking into account the influence of the earth's spherical shape and geographical location on sunshine;

[0126] Based on the Building Information Model (BIM), the building shading coefficient is obtained. The shading coefficient includes detailed three-dimensional structural information of the building, simulates the building's shading of light, and evaluates the impact of the geographical environment on light.

[0127] Using the dynamic turning point algorithm, define a differentiable piecewise function Among them, μ i To control the amplitude of the function at each turning point, β x To adjust the steepness of the function, θ i is the turning point parameter, representing the key node of light intensity change;

[0128] Optimization is performed through the back propagation algorithm. Back propagation transmits the error signal from the output layer to the input layer according to the prediction error between the predicted value and the true value, calculates the gradient of the adjustment parameter, and then corrects the adjustment parameter value.

[0129] Based on constraints Among them, η is the learning rate of 0.01, which is used to control the step size of parameter update. Represents the illumination matching loss function L for the turning point parameter θ i The gradient of the illumination matching loss function L is the mean absolute error between the predicted illumination intensity and the current actual illumination intensity. This ensures that the turning point parameters are dynamically adjusted according to the change direction of the prediction error and the loss function, so that the model can capture the turning point of the illumination intensity change in a timely manner.

[0130] For example, in cloudy scenes, the turning point is lowered to ensure visual comfort and energy saving;

[0131] Based on dynamic backlight scanning DSC technology and computer vision content analysis technology, it analyzes the brightness distribution and contrast of the content displayed in the core area in real time;

[0132] When it detects that the overall brightness of the picture is low and the contrast is not high, the backlight brightness is reduced. At the same time, the local dimming algorithm is used to separately dim the areas that need to be highlighted, reducing power consumption while ensuring visual effects;

[0133] For example, when displaying a nighttime starry sky image, the overall backlight brightness is significantly reduced, and the stars are locally brightened;

[0134] The technical solution of the present invention is: using reinforcement learning algorithm with signal strength, data transmission delay and energy consumption as reward functions, dynamically adjusting the connection relationship between nodes, forming an adaptive cellular topology, improving signal transmission quality, optimizing data transmission delay and energy consumption; preventing installation offset and real-time monitoring of vibration status, automatically triggering the self-test program, and using data repair algorithm to make corrections; establishing a polynomial regression model and adding a high-precision temperature sensor, collecting ambient temperature data in real time to correct the light intensity detection value; using multimodal data fusion method, combined with binocular camera and millimeter wave radar, constructing a three-dimensional model of the human body, calculating the crowd density, and avoiding counting errors caused by human body occlusion; setting weights for different channels to calculate the difference between frames; calculating the color gamut coverage of the actual display color gamut and the DCI-P3 standard color gamut, and improving the accuracy of color processing Accuracy; through the feature selection mechanism, features that contribute greatly to anomaly detection are screened, the interference of irrelevant features is reduced, and the anomaly score threshold is dynamically adjusted according to the data distribution to detect abnormal data; using the dynamic parameter adjustment mechanism, the light intensity sliding window weight is adjusted according to the light intensity change rate, so that the weight distribution adapts to the dynamic change of light intensity and balances the weights of image historical data and real-time image data; by constructing a reasonable reward function, power consumption is reduced while ensuring the prediction accuracy of the model; combined with the adaptive early stopping method, the patience value is dynamically adjusted according to the verification set error to optimize model training, while ensuring model performance, overtraining is avoided, and computing resource consumption is reduced to reduce power consumption; the backlight brightness is adjusted in real time according to the brightness distribution and contrast of the picture, and local dimming is performed on the areas that need to be highlighted, reducing power consumption while ensuring visual effects.

[0135] Example 3

[0136] like Figure 2 As shown, an embodiment of the present invention provides a low-power LED screen control system for outdoor display, which specifically includes the following modules:

[0137] Data acquisition module: responsible for acquiring collected data;

[0138] Data pre-processing module: Add a radio frequency signal strength detection unit, dynamically adjust node connections, and trigger self-test;

[0139] Data analysis module: Compare adjacent node data to correct errors and establish a polynomial regression model to correct light intensity detection values

[0140] Transmission Optimization Module: Builds a temperature compensation model, determines the compensation coefficient, trains the model by selecting multiple random feature subsets, evaluates model performance based on the validation set error, dynamically adjusts the patience value based on the validation set error, and optimizes the model training process.

[0141] Power consumption control module: Based on the corrected light intensity detection value, a reward function is constructed to calculate the state adjustment value and dynamically adjust the drive voltage. The module also collects ambient light intensity data, predicts future lighting trends, dynamically adjusts the turning point of the piecewise function for optimal estimation, evaluates the impact of the geographical environment on lighting, and optimizes the adjustment parameters.

[0142] Adaptive control module: Dynamically adjusts turning point parameters based on prediction errors, captures turning points of light intensity changes, and analyzes and reduces power consumption.

[0143] An embodiment of the present invention is described in detail above, but the content described is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention; the above formulas are all dimensionless and numerical calculations, and the formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field based on actual conditions and historical experience, and can be adjusted according to actual conditions; the above description is only a preferred embodiment of the present invention and is not used to limit the present invention. All equal changes and improvements made according to the scope of application of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A low-power LED screen control system and method for outdoor display, characterized in that: The following steps are involved: Add a radio frequency signal strength detection unit to dynamically adjust node connections, trigger self-test, compare adjacent node data to correct errors, and establish a polynomial regression model to correct light intensity detection values; Based on the corrected light intensity detection value, a reward function is constructed to calculate the state adjustment value, dynamically adjust the driving voltage, build a temperature compensation model, determine the compensation coefficient, train the model by selecting multiple random feature subsets, evaluate the model performance by the validation set error, and dynamically adjust the patience value according to the change in the validation set error to optimize the model training process; Collect ambient light intensity data, predict future lighting change trends, dynamically adjust the turning points of the piecewise function for optimal estimation, evaluate the impact of the geographical environment on lighting, optimize the adjustment parameters, dynamically adjust the turning point parameters according to the prediction error, capture the turning points of light intensity changes, and analyze and reduce power consumption.

2. A low-power LED screen control method for outdoor display according to claim 1, characterized in that: The method for comparing adjacent node data and correcting errors is: For the module's built-in three-axis accelerometer, prevent installation deviation and monitor the sensor's vibration status in real time; When the accelerometer detects that the vibration amplitude exceeds the set vibration amplitude standard, the system automatically triggers the self-test program; By comparing the collected data of adjacent nodes, it is determined whether the own data has errors due to vibration; The corrected light intensity detection value is calculated by establishing a polynomial regression model of the light intensity detection value and temperature through a formula.

3. A low-power LED screen control method for outdoor display according to claim 1, characterized in that: The method for correcting the light intensity detection value is: By Formula I c =I m +a1T 2 +b1T+c1 establishes a polynomial regression model of light intensity detection value and temperature to calculate the corrected light intensity detection value I c ; Among them, I m is the original measured light intensity detection value, T is the ambient temperature, and a1, b1, and c1 are the correction coefficients obtained through experimental calibration.

4. A low-power LED screen control method for outdoor display according to claim 1, characterized in that: The method for calculating the state adjustment value is: The deep Q network in reinforcement learning is used to dynamically adjust the drive voltage VDD. The adjustment range of the drive voltage VDD (3.3-12V) is divided into a discrete state space, with each 0.1V being a discrete state. The action space is set to increase, decrease, or maintain VDD. Construct the reward function R through the formula Get the status adjustment value; Among them, Pp is the power consumption at the previous moment, Pc is the power consumption at the current moment, Ep is the prediction error at the previous moment, and Ec is the prediction error of the model at the current moment. is the reward weight for power consumption reduction, is the penalty weight for changes in prediction error.

5. A low-power LED screen control method for outdoor display according to claim 1, characterized in that: The method for determining the compensation coefficient is: A temperature compensation model is constructed based on polynomial regression, and the power consumption change caused by temperature change is calculated through the formula; A constant temperature chamber is used to collect power consumption data at more temperature points in a preset temperature range. The preset temperature range is set to 25-85°C. The data is fitted using the least squares method to determine the compensation coefficient.

6. A low-power LED screen control method for outdoor display according to claim 1, characterized in that: The method for evaluating model performance is: In the color gamut coverage calculation, deep learning image segmentation is used to segment the actual display color gamut and the DCI-P3 standard color gamut into binary images. The number of pixels in the actual display color gamut and the number of pixels in the overlapping part are obtained. The number of pixels in the overlapping part is then compared with the number of pixels in the DCI-P3 standard color gamut to obtain the color gamut coverage. Based on the isolation forest algorithm, a feature selection mechanism is used to construct an isolation forest with 100 subtrees and a maximum depth of 8 layers; By calculating the information gain rate of image data features, features whose contribution to anomaly detection exceeds the preset standard are screened out.

7. A low-power LED screen control method for outdoor display according to claim 1, characterized in that: The method for optimizing the model training process is: Divide the original dataset into training set, validation set and test set; Select M random feature subsets; For each feature subset S, the model prediction value is obtained by training the deep learning model based on the original data set; The validation set error is obtained by calculating the average of the absolute values of the differences between the model prediction value and the actual power consumption value; Adaptive early stopping method is used to dynamically adjust the patience value according to the changing trend of the validation set error during model training.

8. A low-power LED screen control method for outdoor display according to claim 1, characterized in that: The method for optimizing the adjustment parameters is: In the prediction stage, the state prediction value at the current moment is calculated based on the state at the previous moment and the system dynamic model; In the update phase, the state prediction value is corrected in combination with the current observation data to obtain the corrected state value and suppress the interference of noise on the data; Fourier encoding is used to process seasonal characteristics. Fourier transform decomposes the seasonal periodic signal into a combination of sine and cosine waves of different frequencies. By extracting the coefficients of the combined components of sine and cosine waves of different frequencies, the seasonal characteristics can be quantitatively represented.

9. A low-power LED screen control method for outdoor display according to claim 1, characterized in that: The method for analyzing and reducing power consumption is as follows: Using the dynamic turning point algorithm, define a differentiable piecewise function Among them, μ i To control the amplitude of the function at each turning point, β x To adjust the steepness of the function, θ i is the turning point parameter, representing the key node of light intensity change; Optimization is performed through the back propagation algorithm. Back propagation transmits the error signal from the output layer to the input layer according to the prediction error between the predicted value and the true value, calculates the gradient of the adjustment parameter, and then corrects the adjustment parameter value. Based on constraints Among them, η is the learning rate of 0.01, which is used to control the step size of parameter update. Represents the illumination matching loss function L for the turning point parameter θ i The gradient of the illumination matching loss function L is the mean absolute error between the predicted illumination intensity and the current actual illumination intensity; When it is detected that the overall brightness of the picture is low and the contrast is not high, the backlight brightness is reduced, and the local dimming algorithm is used to individually dim the areas that need to be highlighted.

10. A low-power LED screen control system for outdoor display, the system being used to execute the control method according to any one of claims 1 to 9, characterized in that: include: Data acquisition module: responsible for acquiring collected data; Data pre-processing module: Add a radio frequency signal strength detection unit, dynamically adjust node connections, and trigger self-test; Data analysis module: Compare adjacent node data to correct errors and establish a polynomial regression model to correct light intensity detection values; Transmission Optimization Module: Builds a temperature compensation model, determines the compensation coefficient, trains the model by selecting multiple random feature subsets, evaluates model performance based on the validation set error, dynamically adjusts the patience value based on the validation set error, and optimizes the model training process. Power consumption control module: Based on the corrected light intensity detection value, a reward function is constructed to calculate the state adjustment value and dynamically adjust the driving voltage; Collect ambient light intensity data, predict future lighting trends, dynamically adjust the turning points of the piecewise function for optimal estimation, evaluate the impact of the geographical environment on lighting, and optimize adjustment parameters; Adaptive control module: Dynamically adjusts turning point parameters based on prediction errors, captures turning points of light intensity changes, and analyzes and reduces power consumption.

Images