AI Large Model Regulation Method and System for Display Screen Parameters

Abstract

The present invention discloses an AI large model control method and system for display screen parameters. The method includes: according to the operating state of the display screen, collecting multi-source data in real time through a hardware sensing layer, a video parsing layer, and an environment perception layer to obtain multi-modal feature data; constructing a mechanism-data hybrid model based on the multi-modal feature data to obtain an initial output of the hybrid model; according to the initial output of the hybrid model, using a recursive least squares method based on a sliding window to update the hybrid model coefficients in real time and obtain an optimized output of the hybrid model; designing a cascaded anomaly detection mechanism according to the optimized output of the hybrid model to obtain a final control result for the display screen parameters. By using the embodiments of the present invention, it is possible to achieve intelligent and refined management of the display screen, improve the stability, reliability, safety, and credibility of the electronic display screen during use, and provide more powerful scientific basis and technical guidance for the safe operation of electronic display screens in public areas.

Description

Technical Field

[0001] The present invention belongs to the field of AI technology, and particularly relates to an AI large model control method and system for display screen parameters. Background Art

[0002] With the rapid development of intelligent technology, display screens, as important media for information transmission and visual interaction, are widely used in multiple fields such as advertising, public information release, entertainment, and education. Modern display screens not only require high display quality and resolution but also need to have intelligent control functions to adapt to changing environments and usage requirements. Therefore, how to effectively adjust the parameters of display screens to improve their performance and extend their service life has become an urgent technical problem to be solved.

[0003] Existing display screen parameter control methods mainly rely on simple monitoring means, such as the application of temperature and current sensors. However, these methods often only provide single monitoring data and ignore the complex relationship between the operating state of the display screen and environmental factors. In addition, there are bottlenecks in the current technology for processing multi-source data, and it is impossible to achieve real-time and accurate feature fusion and parameter adjustment, resulting in problems such as overheating and overcurrent in display screens under high load and adverse environments, affecting the stability and reliability of the equipment. Summary of the Invention

[0004] The purpose of the present invention is to provide an AI large model control method and system for display screen parameters to solve the deficiencies in the prior art, which can achieve intelligent and refined management of display screens, improve the stability, reliability, safety, and credibility of electronic display screens during use, and provide more powerful scientific basis and technical guidance for the safe operation of electronic display screens in public areas.

[0005] An embodiment of the present application provides an AI large model control method for display screen parameters, and the method includes:

[0006] According to the operating state of the display screen, multi-source data is collected in real time through a hardware sensing layer, a video parsing layer, and an environmental perception layer. Among them, the hardware sensing layer deploys temperature, current, and vibration sensor arrays, the video parsing layer extracts content features in real time through the YOLOv7 model, the environmental perception layer integrates light and crowd density detection technologies, and the multi-source data is subjected to feature fusion using time series alignment technology to obtain multi-modal feature data;

[0007] According to the multi-modal feature data, a mechanism-data hybrid model is constructed. Among them, the mechanism kernel of the hybrid model is based on the display screen thermodynamics equation, the data kernel of the hybrid model uses a fusion network of 3D-CNN and Transformer, and the output weights of the mechanism kernel and the data kernel are dynamically adjusted through an adaptive weighting algorithm to obtain the initial output of the hybrid model;

[0008] Based on the initial output of the hybrid model, the recursive least squares method based on a sliding window is used to update the hybrid model coefficients in real time. Among them, a teacher-student model architecture is established through knowledge distillation technology. Combining incremental learning, a feature replay buffer is designed to save key samples, the hybrid model parameters are dynamically optimized, and the optimized hybrid model output is obtained.

[0009] Based on the optimized hybrid model output, a cascaded anomaly detection mechanism is designed. Among them, through the three-dimensional joint diagnosis of hardware-content-environment, combined with the three-level response strategy of over-temperature protection and over-current protection and the illegal content feature library interception mechanism, hardware protection, content interception, and environmental adaptation decisions are executed in real time to obtain the final display screen parameter regulation result.

[0010] Optionally, according to the operating state of the display screen, multi-source data is collected in real time through the hardware sensing layer, video parsing layer, and environmental sensing layer. Among them, the hardware sensing layer deploys temperature, current, and vibration sensor arrays. The video parsing layer extracts content features in real time through the YOLOv7 model. The environmental sensing layer integrates light and crowd density detection technologies, and uses time series alignment technology to fuse the features of multi-source data to obtain multi-modal feature data, including:

[0011] According to the operating state of the display screen, hardware state data is collected in real time through the temperature sensor, current sensor, and vibration sensor deployed in the hardware sensing layer. The video parsing layer uses the YOLOv7 model to extract 48-dimensional content features in real time. The environmental sensing layer integrates light sensors and crowd density detection technologies to collect environmental data in real time.

[0012] For the hardware state data, content features, and environmental data, time series alignment technology is used, combined with the timestamp alignment algorithm to eliminate time differences; the missing data is supplemented by interpolation filling method to generate a time-synchronized multi-source data set.

[0013] For the time-synchronized multi-source data set, a feature fusion method based on a cross-modal attention mechanism is used to map the hardware state data, content feature data, and environmental data to a unified multi-modal feature space. Through the feature weighting mechanism, the correlation relationship between different modalities is captured to generate the final multi-modal feature data.

[0014] Optionally, according to the multi-modal feature data, a mechanism-data hybrid model is constructed. Among them, the mechanism kernel of the hybrid model is based on the display screen thermodynamic equation, and the data kernel of the hybrid model uses a fusion network of 3D-CNN and Transformer. The output weights of the mechanism kernel and the data kernel are dynamically adjusted through an adaptive weighting algorithm to obtain the initial output of the hybrid model, including:

[0015] Based on the hardware status data in the multi-modal feature data, the mechanism kernel is initialized using the display thermodynamic equation. Through the dynamic weight allocation mechanism, the rationality and accuracy of the initial parameters of the mechanism kernel are ensured, and a preliminary mechanism model is generated;

[0016] For the content feature data and environmental data in the multi-modal feature data, 3D-CNN is used to extract features in the spatial and temporal dimensions. Combining with the Transformer model to capture long-term sequential dependencies, a data kernel is constructed. Through the adaptive learning rate adjustment mechanism, the training process of the data kernel is optimized, and a preliminary data model is generated;

[0017] For the preliminary mechanism model and the preliminary data model, the adaptive weighted algorithm is adopted. According to the temperature change rate and current fluctuation, the output weights of the mechanism kernel and the data kernel are dynamically adjusted. Through the error feedback mechanism, the weight allocation is optimized to obtain the mechanism-data hybrid model and generate the initial output of the hybrid model.

[0018] Optionally, based on the initial output of the hybrid model, the recursive least squares method based on a sliding window is used to update the hybrid model coefficients in real time. Among them, a teacher-student model architecture is established through knowledge distillation technology. Combining with incremental learning, a feature replay buffer is designed to save key samples, and the hybrid model parameters are dynamically optimized, and the optimized output of the hybrid model is obtained, including:

[0019] Based on the initial output of the hybrid model, the recursive least squares method based on a sliding window is used to calculate the model output error in real time. Through the dynamic weight adjustment mechanism, the coefficients of the mechanism kernel and the data kernel are updated to generate a preliminary optimized model;

[0020] For the preliminary optimized model, knowledge distillation technology is adopted. The hybrid model is used as the teacher model to train a lightweight student model. Through the adaptive learning rate adjustment mechanism, the accuracy and efficiency of the student model are ensured, and a preliminary compressed model is generated;

[0021] For the preliminary compressed model, the incremental learning method is adopted. Combining with the feature replay buffer to save key samples, through the sample weight allocation mechanism, the model parameters are dynamically optimized to generate the final optimized hybrid model and obtain the output of the optimized hybrid model.

[0022] Optionally, based on the output of the optimized hybrid model, a cascaded anomaly detection mechanism is designed. Among them, through the three-dimensional joint diagnosis of hardware-content-environment, combining with the three-level response strategy of over-temperature protection and over-current protection and the violation content feature library interception mechanism, hardware protection, content interception and environmental adaptation decisions are executed in real time to obtain the final display parameter regulation result, including:

[0023] According to the output of the optimized hybrid model, an anomaly detection method based on hardware status data is adopted to monitor temperature, current, and vibration data in real time. Through a three-level response strategy, the hardware protection mechanism is triggered to generate a preliminary hardware protection result;

[0024] For the preliminary hardware protection result, a content detection method based on a violation content feature library is adopted to identify violation content in real time. Through image processing technology, the key features of the violation content are captured, and the content interception mechanism is triggered to generate a preliminary content interception result;

[0025] For the preliminary content interception result, an anomaly detection method based on environmental data is adopted to monitor light intensity and crowd density in real time. Through the light intensity adaptive adjustment algorithm, the brightness of the display screen is optimized to generate a preliminary environmental adaptation result;

[0026] For the preliminary hardware protection result, content interception result, and environmental adaptation result, an output method based on visualization technology is adopted to integrate them into the final display screen parameter regulation result. Through the real-time monitoring mechanism, the stability and efficiency of decision execution are ensured to generate the final regulation result.

[0027] Another embodiment of the present application provides an AI large model regulation system for display screen parameters, and the system includes:

[0028] An acquisition module, configured to collect multi-source data in real time according to the operating state of the display screen through a hardware sensing layer, a video parsing layer, and an environmental perception layer. Among them, the hardware sensing layer deploys temperature, current, and vibration sensor arrays, the video parsing layer extracts content features in real time through the YOLOv7 model, the environmental perception layer integrates light and crowd density detection technologies, and uses time series alignment technology to perform feature fusion on the multi-source data to obtain multi-modal feature data;

[0029] A construction module, configured to construct a mechanism-data hybrid model according to the multi-modal feature data. Among them, the mechanism kernel of the hybrid model is based on the display screen thermodynamics equation, the data kernel of the hybrid model uses a fusion network of 3D-CNN and Transformer, and the output weights of the mechanism kernel and the data kernel are dynamically adjusted through an adaptive weighting algorithm to obtain the initial output of the hybrid model;

[0030] An update module, configured to update the hybrid model coefficients in real time according to the initial output of the hybrid model by using the recursive least squares method based on a sliding window. Among them, a teacher-student model architecture is established through knowledge distillation technology, combined with incremental learning, a feature replay buffer is designed to save key samples, the hybrid model parameters are dynamically optimized, and the optimized output of the hybrid model is obtained;

[0031] A regulation module, configured to design a cascaded anomaly detection mechanism according to the output of the optimized hybrid model. Among them, through the three-dimensional joint diagnosis of hardware-content-environment, combined with the three-level response strategy of over-temperature protection and over-current protection and the violation content feature library interception mechanism, the hardware protection, content interception, and environment adaptation decisions are executed in real time to obtain the final display screen parameter regulation result.

[0032] Another embodiment of the present application provides a storage medium in which a computer program is stored. Among them, the computer program is set to execute the method described in any one of the above when running.

[0033] Another embodiment of the present application provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is set to run the computer program to execute the method described in any one of the above.

[0034] Compared with the prior art, an AI large model regulation method for display screen parameters provided by the present invention collects multi-source data in real time through a hardware sensing layer, a video parsing layer, and an environment perception layer according to the operating state of the display screen to obtain multi-modal feature data; constructs a mechanism-data hybrid model according to the multi-modal feature data to obtain the initial output of the hybrid model; updates the hybrid model coefficients in real time by using the recursive least squares method based on a sliding window according to the initial output of the hybrid model, and obtains the optimized output of the hybrid model; designs a cascaded anomaly detection mechanism according to the optimized output of the hybrid model to obtain the final display screen parameter regulation result, so as to realize the intelligent and refined management of the display screen, improve the performance of the display screen, extend the service life, and ensure safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a hardware structure block diagram of a computer terminal for an AI large model regulation method of display screen parameters provided by an embodiment of the present invention;

[0036] Figure 2 It is a flow schematic diagram of an AI large model regulation method of display screen parameters provided by an embodiment of the present invention;

[0037] Figure 3 It is a structure schematic diagram of an AI large model regulation system of display screen parameters provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be construed as a limitation to the present invention.

[0039] An embodiment of the present invention first provides an AI large model control method for display screen parameters. This method can be applied to electronic devices, such as computer terminals, specifically, ordinary computers, etc.

[0040] Hereinafter, taking the operation on a computer terminal as an example, it will be described in detail. Figure 1 The following is a hardware structure block diagram of a computer terminal for an AI large model control method for display screen parameters provided by an embodiment of the present invention. As Figure 1 shown, the computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory may include a non-volatile storage medium and an internal memory.

[0041] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions. When the program instructions are executed, the processor can execute any AI large model control method for display screen parameters.

[0042] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0043] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any AI large model control method for display screen parameters.

[0044] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 1 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0045] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0046] See Figure 2, embodiments of the present invention provide an AI large model control method for display screen parameters, which may include the following steps:

[0047] S201, according to the operating state of the display screen, multi-source data is collected in real time through the hardware sensing layer, video parsing layer, and environmental perception layer. Among them, the hardware sensing layer deploys temperature, current, and vibration sensor arrays, the video parsing layer extracts content features in real time through the YOLOv7 model, the environmental perception layer integrates light and pedestrian flow density detection technologies, and the multi-source data is feature fused using time series alignment technology to obtain multi-modal feature data;

[0048] The core of this method is to collect multi-source data of the display screen in real time through the collaborative work of the hardware sensing layer, video parsing layer, and environmental perception layer. These data sources include hardware state information such as temperature, current, and vibration, content features extracted through the YOLOv7 model, and environmental monitoring information (such as light intensity and pedestrian flow density). By using time series alignment technology, these multi-source data can eliminate problems such as time delay or out-of-sync, thus forming a high-quality multi-modal feature data set. This process ensures that the system can make accurate responses and decisions in a real-time dynamic environment.

[0049] By collecting multi-source data in real time and performing feature fusion, the system can comprehensively understand the operating state of the display screen, thereby optimizing the display effect and extending the device life. The generation of multi-modal feature data enables the subsequent mechanism-data hybrid model to be trained and optimized based on more accurate and comprehensive information, thereby improving the operating efficiency and safety of the display screen. This process not only enhances the intelligent control ability of the display screen but also lays a solid data foundation for subsequent anomaly detection and response mechanisms.

[0050] Specifically, according to the operating state of the display screen, hardware state data can be collected in real time through the temperature sensor, current sensor, and vibration sensor deployed in the hardware sensing layer, 48-dimensional content features can be extracted in real time through the YOLOv7 model in the video parsing layer, and environmental data can be collected in real time through the environmental perception layer integrating light sensors and pedestrian flow density detection technologies;

[0051] The real-time collection of hardware state data, content features, and environmental data provides multi-dimensional information support for the construction of multi-modal feature data. The combination of these data provides rich background information for the training and optimization of subsequent models, enabling the system to respond flexibly in complex environments, thereby improving the safety and adaptability of the display screen.

[0052] In specific implementation, multiple temperature sensors are first installed around the display screen to monitor the operating temperature of the display screen. These sensors can provide real-time feedback of temperature data to ensure that the display screen operates within a safe temperature range, preventing equipment failures caused by overheating. Meanwhile, current sensors will monitor the power consumption and current fluctuations of the display screen, and detect abnormal current states in a timely manner to prevent potential safety hazards caused by excessive power loads. In addition, vibration sensors can monitor the vibration of the display screen in real time to evaluate its stability, ensuring that there is no physical damage under the influence of environmental factors such as strong winds and vibrations.

[0053] The data of these hardware sensors are transmitted to the data processing module in real time via wireless or wired means. In the data processing module, the data collected by the temperature, current, and vibration sensors will be uniformly formatted and associated with timestamps, providing a basis for subsequent data analysis. Throughout the process, the system will also perform data verification regularly to ensure the accuracy of the output data of the sensors and avoid data errors caused by sensor failures. For example, verification is carried out through regular self-checks or the use of backup sensors to ensure the real-time performance and accuracy of each sensor during operation.

[0054] In the video parsing layer, the YOLOv7 model is responsible for processing the content played on the display screen in real time. This model can perform object recognition and image analysis, extracting 48-dimensional features from the video stream, including information such as the category, location, and size of the objects. The collected content feature data provides rich context, helping the system understand the played content and the interaction situation of the audience. At the same time, these content features will be integrated and analyzed together with other data from hardware and the environment, providing comprehensive insights for the system. For example, when the system detects that a promotional advertisement is being played on the display screen, combined with the audience number data from the environmental perception layer, the system can dynamically adjust the display strategy of the advertisement content, thereby enhancing the effectiveness of the advertisement.

[0055] For the hardware status data, content features, and environmental data, the time series alignment technology is adopted, and the time difference is eliminated by combining the timestamp alignment algorithm; the missing data is supplemented by the interpolation filling method to generate a time-synchronized multi-source data set;

[0056] The time series alignment technology enables the inputs from different data sources to be consistent in time, eliminating data differences caused by different collection times. By interpolating and filling the missing data, the integrity and continuity of the multi-source data set are ensured. This process improves the effectiveness of data fusion, provides a solid data foundation for the subsequent establishment of the hybrid model, and thus enhances the stability and reliability of the overall system.

[0057] At this stage, the temporal alignment of data is achieved by recording the timestamps output by each sensor and the video parsing model. Since different data sources may be updated at different frequencies, for example, the temperature sensor is updated once a minute, while the YOLOv7 model provides content features every five seconds, the system needs to synchronize these data in time. During this process, a timestamp alignment algorithm is adopted to achieve the consistency of multiple data sources in the time dimension. Specifically, by comparing the timestamps of each data source, the nearest time point is found, and its data is mapped to the same time node.

[0058] During the processing, the system may encounter a situation where the data of a certain data source is not successfully collected at a certain time point. To solve this problem, an interpolation filling method is adopted, and the existing data is used for calculation to ensure the integrity and continuity of the dataset. For example, if the temperature data is missing at a certain moment and the data at adjacent moments is relatively stable, the system can use the average value or linear interpolation of these time points to fill in the missing data. This not only ensures the comprehensiveness of the data but also avoids the deviation of the analysis results caused by missing data.

[0059] After the temporal alignment and interpolation filling are completed, the finally generated time-synchronized multi-source dataset will contain complete data from each sensor and the video parsing layer. This multi-source dataset lays the foundation for subsequent analysis and model training. Through accurate time alignment, the system can more effectively identify the relationships between hardware states, video content, and environmental changes, and thus provide a basis for intelligent decision-making. For example, when the system detects an increase in temperature while there is a crowded scene in the video content, the system will comprehensively consider these two pieces of information to determine whether to take cooling measures or adjust the display content to guide the behavior of the audience.

[0060] For the time-synchronized multi-source dataset, a feature fusion method based on cross-modal attention mechanism is adopted to map the hardware state data, content feature data, and environmental data into a unified multi-modal feature space. Through the feature weighting mechanism, the correlation relationships between different modalities are captured to generate the final multi-modal feature data.

[0061] By performing feature fusion through the cross-modal attention mechanism, the important information interactions between different modalities can be effectively identified and enhanced. The feature weighting mechanism further improves the flexibility of the system, enabling it to dynamically adapt to data changes under different environmental conditions. This multi-modal feature dataset provides a rich information basis for the subsequent construction of the mechanism-data hybrid model and enhances the adaptability of the model to complex scenarios.

[0062] In this step, the multi-source data set with time synchronization is preprocessed first. To ensure that the data of each modality can be effectively integrated in the feature fusion stage, the system will standardize the hardware state data, content feature data, and environmental data, making them consistent in the numerical range and format. For example, all data are mapped to the range of [0, 1] through the normalization method. This step is crucial because data from different sources may have different scales. Without standardization, the fusion result may be greatly affected by a certain modality, ignoring the importance of other modalities.

[0063] Next, the system adopts a cross-modal attention mechanism to fuse the data of different modalities. Specifically, the system will calculate the weights of the features of each modality and use the self-attention mechanism to capture the important relationships between modalities. For example, assume that at a specific time point, the hardware state data shows that the temperature has risen to the warning line and at the same time the pedestrian flow density has also increased. The system will assign higher weights to these key features through the attention mechanism. When the importance of the hardware state data and environmental data is higher than that of the content features in the generated multi-modal feature data, the system will enhance them accordingly so that the subsequent model training can more accurately reflect the real situation.

[0064] Finally, after feature weighting and fusion, the generated multi-modal feature data will contain rich information from each data source. These data can not only reflect the hardware state of the display screen and the content features being played, but also comprehensively consider the surrounding environmental factors. The ultimate goal of this process is to make the model more efficient and accurate in dealing with complex scenarios by fusing the correlation information between modalities. For example, when the display screen plays specific advertising content under high pedestrian flow density, the system can use the fused multi-modal feature data to dynamically adjust the display method, color, and brightness of the advertising content to achieve the best visual effect and audience response.

[0065] S202. Construct a mechanism-data hybrid model according to the multi-modal feature data, where the mechanism core of the hybrid model is based on the display screen thermodynamics equation, the data core of the hybrid model adopts a fusion network of 3D-CNN and Transformer, and the output weights of the mechanism core and the data core are dynamically adjusted through an adaptive weighting algorithm to obtain the initial output of the hybrid model;

[0066] In the step of constructing a mechanism-data hybrid model, it is first necessary to build a comprehensive model based on the obtained multi-modal feature data. This model consists of two core parts: the mechanism kernel and the data kernel. The mechanism kernel is based on the display screen's thermodynamic equation, which describes the physical properties such as heat transfer and temperature change of the display screen under different working conditions. At the same time, the data kernel uses a fusion network of 3D-CNN and Transformer, focusing on processing time-series data and its feature extraction to capture the changes in content features and environmental states. This process uses an adaptive weighting algorithm to dynamically adjust the output weights of the mechanism kernel and the data kernel, ensuring that the two can work together to improve the accuracy and response ability of the model and generate the initial output of the hybrid model.

[0067] The significance of constructing a mechanism-data hybrid model is that it provides a scientific and reliable data basis for the real-time regulation of the display screen. The physical equation of the mechanism kernel can accurately relate the operating state of the display screen to its physical properties, and the data kernel can efficiently capture environmental changes and content characteristics through deep learning methods. This combination ensures that the model can not only be adjusted based on physical laws but also flexibly adapt to the changes in real-time data, thus realizing the intelligent regulation and optimization of the display screen parameters and ultimately improving the performance and safety of the display screen.

[0068] Specifically, according to the hardware state data in the multi-modal feature data, the mechanism kernel can be initialized using the display screen's thermodynamic equation. Through a dynamic weight allocation mechanism, the rationality and accuracy of the initial parameters of the mechanism kernel are ensured, and a preliminary mechanism model is generated.

[0069] In this step, the system uses the temperature, current, and vibration information extracted from the hardware state data and combines it with the display screen's thermodynamic equation to initialize the mechanism kernel. The thermodynamic equation can describe the heat distribution and heat dissipation of the display screen under different working conditions. Through this physical model, the operating characteristics of the display screen in different environments can be simulated. This process also includes a dynamic weight allocation mechanism that adjusts the parameters in the model by real-time monitoring of data feedback to ensure that the mechanism kernel has high accuracy at the beginning.

[0070] The implementation of this step ensures the scientificity and accuracy of the mechanism kernel, thus laying a good foundation for subsequent model training. Through precise initialization, the model can better reflect the actual physical state of the display screen, providing a reliable reference framework for further data fusion and analysis, and thus improving the prediction ability of the overall model.

[0071] In actual implementation, first, it is necessary to obtain real-time hardware status data from various sensors (temperature, current, vibration). These sensors can be connected to the central control unit through serial ports or wireless networks to send the collected data in real time. Taking the temperature sensor as an example, the sensor will monitor the temperature of each component of the display screen and send the corresponding values to the control unit at certain time intervals. The current and vibration sensors work in the same way to ensure comprehensive coverage of the operating status of the display screen.

[0072] Next, the system will introduce these real-time data into the thermodynamic equation of the display screen for calculation. The thermodynamic model may include multiple variables, such as the thermal conductivity of materials, ambient temperature, power input, etc. By establishing a mathematical model, these variables are related. For example, the relationship between temperature change and current can be described by a simple linear regression equation or calculated using a more complex heat conduction model (such as Fourier's law). The results obtained from real-time calculation will be used to initialize the parameters of the mechanism kernel, enabling the mechanism kernel to reflect the current hardware status.

[0073] Finally, a dynamic weight allocation mechanism is constructed to continuously optimize the parameters of the mechanism kernel. This step can automatically adjust the weights based on the monitored temperature, current operating mode, and historical data. For example, when the sensor data indicates that the temperature of a certain component has risen to the set safety threshold, the system will increase the weight of this parameter to ensure that the mechanism kernel can respond more sensitively to temperature changes. The implementation method of this dynamic adjustment ensures the accuracy and rationality of the mechanism model, enabling it to react promptly in case of abnormal situations.

[0074] For example, in the LED display screen of a large shopping mall, a group of temperature sensors, current sensors, and vibration sensors are deployed to monitor the operating status of the display screen in real time. The temperature sensors are installed in the key components of the display screen, such as the power module and the display panel, to ensure comprehensive monitoring. The current sensors are also installed in the power supply line to be able to provide real-time feedback on the current consumption of the display screen, while the vibration sensors are placed at the base to monitor the physical stability of the display screen. All these sensors transmit data to the monitoring center in real time via the Wi-Fi network. After receiving the sensor data, the monitoring center first sorts out these data. Then, using the display screen thermodynamics equation, such as based on Fourier's law of heat conduction, combined with the current temperature and current data, it calculates the heat distribution of each component of the display screen. Suppose at a certain moment, the temperature sensor detects that the ambient temperature is 30°C and the operating current of the display screen is 5A. The system calculates according to the thermodynamics equation and obtains that the temperature of each component inside the current display screen should be 45°C. Subsequently, these calculated temperature values will be used as the initial parameters of the mechanism core. Finally, the dynamic weight allocation mechanism starts to operate. When it is continuously monitored that the temperature rises to 55°C, the system determines this as an abnormal state. At this time, the weight allocation mechanism will automatically adjust the weight of the mechanism core to improve the reaction sensitivity to temperature parameters, so as to ensure that the model can quickly adapt to the current hardware state. This mechanism combines the data fed back in real time by the sensors, making the output of the mechanism model more accurate, thus ensuring the safe operation of the display screen.

[0075] For the content feature data and environmental data in the multi-modal feature data, 3D-CNN is used to extract spatial and temporal dimension features, combined with the Transformer model to capture long-time sequence dependencies, construct a data core, and through the adaptive learning rate adjustment mechanism, optimize the training process of the data core to generate a preliminary data model;

[0076] In this step, the system uses 3D-CNN to process the content features and environmental data to extract their spatial and temporal features. The advantage of 3D-CNN is that it can use 3D convolution operations to analyze the changes in data in both the time and space dimensions, thereby capturing the features of the content or environment at different time points. At the same time, combined with the Transformer model, it can effectively capture the dependencies in a longer time series, improving the understanding and modeling ability of time series data. Through the adaptive learning rate adjustment mechanism, the system can dynamically optimize the learning rate during the training process to improve the model convergence speed and accuracy, and finally generate a preliminary data model.

[0077] The significance of this step lies in enabling the model to deeply analyze complex time-series data through advanced deep learning techniques. By extracting features from content characteristics and environmental changes, the system can more accurately understand the relationship between the current state and future trends, thus providing support for achieving more intelligent display parameter regulation.

[0078] In this step, the data processing flow begins with the integration and preprocessing of multi-modal feature data. First, the system extracts content features from the video stream and uses the YOLOv7 model for real-time object recognition. This model can parse and extract 48-dimensional feature vectors in real time to describe the characteristics of the objects appearing on the display screen. These features represent the distribution and changes of the content and are crucial for subsequent processing. At the same time, environmental data is collected through light sensors and people flow density detectors to form a time-synchronized data set. The environmental data will be combined with the content feature data for effective fusion in subsequent model construction.

[0079] Subsequently, the generated multi-modal data is input into the 3D-CNN network. In this process, 3D-CNN will adopt three-dimensional convolution operations, considering both spatial and temporal feature extraction. Specifically, 3D-CNN will input multiple frames of data, extract the spatial features of different time frames in the video through the convolutional layer, and establish connections in time to capture dynamic changes. In addition, the system will design pooling layers and fully connected layers to effectively reduce the dimension and obtain the final feature representation. When implementing this operation, a larger stride is adopted to reduce the computational complexity, enabling the model to have a higher real-time response ability.

[0080] To enhance the performance of the data core, the Transformer model is combined to capture long-term temporal dependencies. The self-attention mechanism of the Transformer model can help the system focus on important moments in the time series, ensuring that the model effectively captures the key changes in the video content. The training process of the model is optimized through an adaptive learning rate adjustment mechanism. In the training process, a relatively high learning rate can be used initially to accelerate the convergence of the model. As the training progresses, a dynamic adjustment method is adopted to gradually reduce the learning rate, which can refine the model parameters and ensure the best fit to the training data. Through continuous iterative training, a preliminary data model is finally generated.

[0081] For example, in the application scenario of a dynamic advertising display screen, the system first analyzes the advertising video stream through the YOLOv7 model, identifies key elements in the advertisement, such as products, brand logos, and text information, and extracts a 48-dimensional content feature vector. At the same time, environmental sensors collect data on light intensity and pedestrian flow density in real time. The light sensor is placed above the advertising screen, while the pedestrian flow density sensor is installed at the passage in front of the advertising screen. These environmental data will record the changes in the surrounding environment to provide real-time feedback on the audience. Through the time series alignment technology, the system synchronizes the content features and environmental data so that they match at the same timestamp. Suppose at a certain moment, the content feature data has been extracted, the measured light intensity by the light sensor is 1000 Lux, and the pedestrian flow density is 50 people. Next, all these time-aligned data will be input into the 3D-CNN network for feature extraction in the spatial and temporal dimensions. The 3D-CNN model can process these time series video data to capture the dynamic changes in the video content. To further enhance the performance of the model, the system combines the Transformer model to capture long-term time series dependencies. For example, the Transformer will analyze the temporal frame changes of a certain product in the video through the attention mechanism to ensure that the model focuses on persistent information. The adaptive learning rate adjustment mechanism will monitor the loss function in real time during the model training process. When the loss significantly decreases, it will automatically adjust the learning rate to ensure efficient convergence. After multiple iterative trainings, the system finally generates a preliminary data model that can effectively express the relationship between content features and environmental data.

[0082] For the preliminary mechanism model and the preliminary data model, the adaptive weighted algorithm is adopted to dynamically adjust the output weights of the mechanism kernel and the data kernel according to the temperature change rate and current fluctuation. Through the error feedback mechanism, the weight allocation is optimized to obtain the mechanism-data hybrid model and generate the initial output of the hybrid model.

[0083] This step integrates the outputs of the mechanism model and the data model through the adaptive weighted algorithm. According to the real-time monitored temperature change rate and current fluctuation, the system can dynamically adjust the output weights of the two, so that under specific conditions, the model can pay more attention to the output of one aspect. For example, when the temperature rises significantly, the weighting of the thermodynamics part can be increased to make the model pay more attention to the factors causing overheating. Through the error feedback mechanism, the system will regularly evaluate the accuracy of the model output and optimize the weight allocation accordingly to ensure that the combined effects of the mechanism kernel and the data kernel are fully exerted.

[0084] The implementation of this step increases the flexibility of the model, enabling it to self-adjust according to the real-time operation status, and enhancing the accuracy and reliability of the prediction. Through this dynamic adjustment, the system can better adapt to the changing environment and operating conditions, providing support for the real-time adjustment of the display screen parameters and ensuring the efficient and safe operation of the display screen.

[0085] In this step, first, the output results of the preliminary mechanism model and the preliminary data model are compared and analyzed. The system will continuously monitor the temperature and current changes in the display screen environment and dynamically adjust the output weights of the mechanism kernel and the data kernel according to this data. For example, when the temperature sensor detects that the temperature change rate exceeds the set threshold, the system will increase the weight of the mechanism kernel, making it play a more important role in the model output. Correspondingly, if the current fluctuation is within the normal range, the system will reduce the weight of the data kernel to ensure that the output of the mechanism kernel can sensitively reflect the impact of temperature changes on the display effect.

[0086] Secondly, an error feedback mechanism is implemented to optimize the weight allocation. The system regularly calculates the error between the model output and the real monitoring data and adjusts the weights according to the feedback information. For example, if the output display effect does not meet the expectations, the system will analyze the reasons for the error and adjust the output weights of the relevant kernels. Through several consecutive iterations, the weights will be gradually optimized, enabling the model to gradually approach the real state. Specifically, a mechanism for regularly updating the gradient can be set to adjust the weights in real time, so that each weight adjustment is carried out according to the specific currently monitored state.

[0087] Finally, the initial output of the mechanism-data hybrid model is generated. Combining the information of the mechanism kernel and the data kernel, the system integrates them into a final output result, which will be used to adjust the working state of the display screen in real time. During the implementation process, a visualization interface can be used to display the changes in each weight and the prediction effect of the final model. For example, the impact of temperature changes on the display effect over different time periods can be shown through a heat map, intuitively reflecting the actual applicability and efficiency of the model. Through such a real-time feedback and adjustment process, the finally generated hybrid model can better reflect the operating conditions of the display screen and ensure its stability and performance under various environmental conditions.

[0088] For example, assume that the predicted temperature output by the preliminary mechanism model is 50°C, while the prediction value given by the data model based on the surrounding environment is 48°C. In this case, the system will analyze these two output results, detect that the temperature change rate exceeds the set threshold, which may indicate an overheating condition of the display screen under a certain load. At the same time, through the dynamic weight allocation algorithm, the system will adjust the outputs of the mechanism core and the data core according to the current temperature change rate and current fluctuation. For example, if it is detected that the temperature rapidly rises to 60°C within a short period of time, the weight of the mechanism core will be increased to strengthen its influence in the output and ensure that abnormal temperature fluctuations are taken seriously. And if the current fluctuates greatly but the temperature remains stable, the weight of the data core may be increased to better reflect the impact of the current on the display effect. Subsequently, the system will regularly record the deviation between the model output and the actual value according to the error feedback mechanism. When it is detected that the output deviation exceeds the preset limit, the system will automatically adjust the weights to make the model more accurate. For example, if the output of the mechanism core fails to reflect the actual safety state in a high-temperature situation in a timely manner, the system will increase the weight of the mechanism core to ensure that the subsequent model can better respond to this situation. By continuously cycling through this process, an initial output of the mechanism-data hybrid model is finally obtained, and this output will be used to adjust the display screen parameters in real time to ensure its safe and efficient operation.

[0089] S203, according to the initial output of the hybrid model, use the recursive least squares method based on a sliding window to update the hybrid model coefficients in real time. Among them, establish a teacher-student model architecture through knowledge distillation technology, combine incremental learning, design a feature replay buffer to save key samples, dynamically optimize the hybrid model parameters, and obtain the optimized hybrid model output;

[0090] According to the initial output of the hybrid model, use the recursive least squares method based on a sliding window to update the hybrid model coefficients in real time. This process aims to optimize the parameters of the model so that it can more effectively adapt to the performance requirements of the display screen in a dynamic environment. Specifically, the sliding window technology allows the model to use the latest data samples for adaptive adjustment to ensure the shortest possible response time. Through the recursive least squares method, the update of the model coefficients can be calculated efficiently without retraining the entire model, thus meeting the requirements of real-time performance. In addition, establishing a teacher-student model architecture through knowledge distillation technology can further improve the learning efficiency and prediction accuracy of the model, enabling the lightweight student model to have faster inference ability while retaining the performance of the original model.

[0091] The implementation of this method greatly enhances the adaptive ability of the display screen system, enabling it to better cope with the challenges brought by external environmental changes. By updating the model coefficients in real time, the system can automatically adjust the parameters according to the current state and environmental conditions during the operation of the display screen, thereby improving the display effect and safety.

[0092] Specifically, based on the initial output of the hybrid model, the recursive least squares method based on a sliding window can be adopted to calculate the model output error in real time. Through the dynamic weight adjustment mechanism, the coefficients of the mechanism kernel and the data kernel are updated to generate a preliminary optimized model.

[0093] In this step, the system obtains the output data within a period of time through the sliding window technique and calculates the error of these data using the recursive least squares method. This process allows the model to dynamically adjust its parameters based on the latest information. The sliding window not only provides real-time monitoring of the data but also maintains good adaptability in the data flow. At the same time, the dynamic weight adjustment mechanism enables the model to adjust the weights of the mechanism kernel and the data kernel in real time when facing different types of inputs (such as temperature changes, flow fluctuations, etc.), thereby ensuring the accuracy and stability of the model.

[0094] The implementation of this step can significantly improve the real-time performance of the hybrid model, enabling the model to remain efficient and reliable in a rapidly changing environment. By calculating the output error in real time, the system can react immediately, make timely adjustments, and avoid potential risks caused by lag. This adaptive ability is particularly important in the operation of the display screen, ensuring the best display effect and operation safety under different conditions and improving the overall quality of the user experience.

[0095] In the first step, to implement the recursive least squares method based on a sliding window, it is first necessary to determine the window size N and design a real-time data collection system. When the display screen is working, this system continuously monitors relevant input and output data, including the display content, environmental status (such as temperature, humidity, etc.), and information such as current and vibration read by sensors. Each window covers N latest data points. Whenever new data comes in, the system removes the oldest data, and this process continues to cycle, ensuring that the model can always calculate based on the latest input.

[0096] Next, using the data within the sliding window, the recursive least squares method is applied to calculate the error of the model output. Specifically, the system can set the output of the initial model and the target value (theoretical output), and then calculate the difference between the actual output and the target output within the current window to obtain the error. Using these errors, the recursive least squares method can efficiently update the coefficients of the mechanism kernel and the data kernel to optimize the model. For this purpose, the weighted least squares method can be used to assign higher weights to the latest data points to ensure the sensitivity of the model to recent dynamic changes.

[0097] Finally, implement a dynamic weight adjustment mechanism. This mechanism adjusts the weights of the mechanism kernel and data kernel based on the output error calculated in real time, enabling the model to flexibly respond to various emergencies. For example, when the readings of the temperature sensor suddenly increase, the system will automatically increase the weight of the thermodynamic equation and correspondingly reduce the weights of the current or vibration sensors, improving the model's response ability to temperature changes. This dynamic adjustment ensures that the hybrid model can maintain accurate output and stable performance under different operating conditions.

[0098] For the preliminary optimized model, adopt the knowledge distillation technique. Use the hybrid model as the teacher model to train the lightweight student model. Through the adaptive learning rate adjustment mechanism, ensure the accuracy and efficiency of the student model, and generate a preliminary compressed model.

[0099] In this step, utilize the knowledge distillation technique. Set the optimized hybrid model as the teacher model and use its output to guide the training process of the lightweight student model. The core of knowledge distillation lies in helping the student model learn effective feature representations more quickly through the soft labels or intermediate layer features of the teacher model. The importance of the adaptive learning rate adjustment mechanism in this process is that it can dynamically adjust the learning rate according to the learning progress of the student model, ensuring that the student model learns quickly in the initial stage of training and reducing the learning rate when approaching convergence to achieve a balance between accuracy and efficiency.

[0100] This process greatly improves the learning efficiency and prediction ability of the student model by learning the knowledge of the teacher model, enabling the lightweight model to still maintain high performance under resource-constrained conditions. This lightweight design is very suitable for embedded devices or mobile applications, meeting the requirements of real-time performance and low power consumption, thus broadening the application prospects in intelligent displays and other real-time dynamic systems.

[0101] In the second step, implement the knowledge distillation technique. First, an appropriate student model structure needs to be selected. Smaller convolutional networks such as MobileNet or SqueezeNet can be used. These networks have fewer parameters and can reduce the demand for computing resources while maintaining relatively high accuracy. Before preparing to generate the student model, initialize it using the already optimized hybrid model as the teacher model, which will serve as an important reference for the student model to learn.

[0102] During the training process, the output of the teacher model will be transformed into soft labels through a softmax layer. These labels respectively include the predicted probability distributions of the teacher model. The student model is then trained based on these soft labels and can also incorporate hard labels (ground truth) to provide stronger supervision. By calculating the loss between the teacher model and the student model, the system can efficiently correct the learning direction of the student model. To increase the flexibility of training, an adaptive learning rate adjustment mechanism can be adopted, enabling the learning rate to be dynamically adjusted according to the changes in the loss during training. This helps to converge quickly in the initial stage and slow down the learning rate in the later stage to improve the accuracy of the model.

[0103] In addition, an early stopping mechanism can be introduced. If the performance of the student model no longer improves on a specific validation set, the training will automatically stop. This ensures that the student model does not overfit and also saves computational resources. Finally, after multiple iterative trainings, the student model will be optimized to generate a preliminary compressed model, which has the ability to perform rapid inference in practical applications while maintaining performance similar to that of the teacher model.

[0104] For the preliminary compressed model, an incremental learning method is adopted. By combining a feature replay buffer to save key samples and through a sample weight assignment mechanism, the model parameters are dynamically optimized to generate the final optimized hybrid model and obtain the output of the optimized hybrid model.

[0105] In this final step, the preliminary compressed model is further optimized through an incremental learning strategy. Incremental learning allows the model to retain the knowledge it has learned while continuously receiving new data, avoiding forgetting previous information. The introduction of the feature replay buffer enables key samples to be saved and reused in subsequent learning, enhancing the model's ability to quickly adapt to new environments. At the same time, the sample weight assignment mechanism ensures that the model focuses on the most important samples during training, improving learning efficiency and accuracy.

[0106] The implementation of this method can significantly improve the performance and stability of the model in a dynamic environment, support its rapid adaptation under changing conditions, and enhance its ability to make real-time decisions. By dynamically optimizing the model parameters, the finally generated optimized hybrid model can achieve higher accuracy and lower resource consumption in practical applications, enhancing the flexibility and maintainability of the system and adapting to various possible future usage scenarios.

[0107] In the final step, the implementation of the incremental learning method requires careful design of the feature replay buffer to facilitate the dynamic preservation of key samples. First, the system needs to set a buffer size to ensure that it can store the most recent n key samples. During each model training phase, the system determines whether new data samples are important in the buffer. Important samples should be saved, while unimportant samples can be replaced. This process ensures that the model can always maintain diversity and promptly learn new knowledge crucial for improving its performance.

[0108] During the incremental learning process, the model will be trained by combining the key samples in the buffer. Using these samples, the system can effectively perform parameter tuning, thus avoiding the high cost of retraining the entire model. The sample weight assignment mechanism further enhances the model's attention to important samples by assigning weights to each sample. For example, the better a sample performs in the previous training, the higher its weight. In subsequent training, the model will rely more on these important samples, thereby accelerating convergence and improving accuracy.

[0109] Finally, based on the output of the model on new data, the performance of the model will be monitored and analyzed in real time to continuously update and optimize the model parameters. For example, in a specific application, if the model performs poorly in high-traffic and high-temperature environments, the system will automatically adjust the model parameters and increase the learning frequency of key samples for those environmental conditions. This dynamic optimization strategy ensures that the finally generated optimized hybrid model can handle various complex real-world scenarios, provide reliable outputs, and ensure continuous effectiveness and stability in different application environments.

[0110] S204, design a cascaded anomaly detection mechanism according to the output of the optimized hybrid model. Among them, through the three-dimensional joint diagnosis of hardware-content-environment, combined with the three-level response strategy of over-temperature protection and over-current protection and the violation content feature library interception mechanism, perform hardware protection, content interception, and environmental adaptation decisions in real time to obtain the final display screen parameter regulation result.

[0111] Design a cascaded anomaly detection mechanism according to the output of the optimized hybrid model, aiming to ensure the safety and stability of the display screen during operation through the joint diagnosis of three dimensions: hardware, content, and environment. This mechanism combines the three-level response strategy of over-temperature protection and over-current protection, as well as the violation content feature library interception mechanism, to achieve comprehensive monitoring and real-time response to the display screen status. This mechanism not only focuses on the physical state of the hardware but also takes into account the compliance of the display content and the adaptability of the external environment, providing a solid foundation for the intelligent regulation of the display screen.

[0112] The implementation of this cascaded anomaly detection mechanism is of great significance for improving the safety, reliability, and intelligence level of the display screen. Through real-time monitoring and automatic response, the display screen can quickly take corresponding measures in case of anomalies, preventing damage, misoperation, or improper content display, and ensuring the user experience and the long-term use of the device. At the same time, the three-dimensional joint diagnosis method of this mechanism effectively enhances the system's adaptive ability, enabling it to still operate stably under changing environmental conditions, thus providing strong technical support for application scenarios such as commercial advertisements and public information.

[0113] Specifically, according to the output of the optimized hybrid model, an anomaly detection method based on hardware status data can be adopted to monitor temperature, current, and vibration data in real time. Through a three-level response strategy, the hardware protection mechanism is triggered to generate a preliminary hardware protection result.

[0114] This method is based on the real-time data collected by sensors to analyze the current hardware status and detect whether there is a situation exceeding the safety threshold. If the temperature, current, or vibration data exceeds the preset standard, the system will immediately judge it as an abnormal state and activate the corresponding protection mechanism. By implementing the anomaly detection method based on hardware status data, it is possible to effectively prevent equipment failures caused by overheating, overcurrent, or abnormal vibration. The timely response mechanism of this method can significantly reduce the risk of equipment damage, improve the safety and reliability of the display screen. At the same time, this process can provide an accurate basis for the hardware status for subsequent decision-making and regulation, thereby further improving the intelligence level of the system.

[0115] First, to ensure data accuracy, the system installs precise sensors at key components of the display screen. These sensors collect real-time data on the temperature, current, and vibration of the display screen operation. The temperature sensor can be placed near the power module of the display screen to ensure any heat increase can be monitored in a timely manner; while the current and vibration sensors can be distributed across various parts of the display screen to comprehensively monitor the overall operation status. Secondly, after data collection, all information is sent to a central processing unit (CPU) for analysis. The processing unit sets multiple safety thresholds to facilitate the judgment of whether the device is operating normally. For example, the temperature threshold is set at 70°C. Once the temperature is found to exceed this value, the CPU will activate the alarm mechanism and initiate temporary protection measures. At the same time, if the current flow exceeds the set range, the processing unit will also analyze the vibration data to evaluate whether there are potential risks in the device. Such a system ensures that multiple factors can be considered simultaneously and responses can be made in a timely manner. Finally, after detecting an abnormal situation, the system will handle it through a three-level response strategy. The primary response can be that the system automatically issues an alarm to alert the management staff; the intermediate response may involve adjusting the working parameters of the display screen, such as reducing the brightness or cooling through a fan; and the advanced response will be to directly cut off the power supply to protect the safety of the display screen. Through this mechanism, potential damage to the device can be prevented in a timely manner, improving the working safety and reliability of the display screen.

[0116] Suppose in a large indoor commercial exhibition, the display screen is used to play brand advertisements. To ensure the normal operation of the device, the sensor array installed inside the display screen starts to collect data in real time.

[0117] First, the temperature sensor detects that the operating temperature of the display screen continuously rises to 75°C, exceeding the set safety threshold of 70°C. At the same time, the current sensor monitors that the current data fluctuates, reaching a current of 5.5A in one path, while the upper limit of the normal operating current of this display screen is 5.0A. The vibration sensor also reports vibrations caused by nearby construction, with the value exceeding the normal range. At this time, after receiving this data, the central processing unit (CPU) makes a comprehensive judgment.

[0118] Based on the detected abnormalities, the system first issues an audible and visual alarm to alert the management staff of the risks of overheating and overcurrent in the device. Immediately afterwards, the CPU initiates the intermediate response, reducing the working load by adjusting the brightness of the display screen from 100% to 70%. At the same time, the fan automatically turns on to increase the heat dissipation effect. Finally, when the temperature further rises to 80°C, the system decides to adopt the advanced response strategy and immediately cuts off the power supply to prevent device damage. Through such a response mechanism, the safety protection of the display screen in an abnormal state is ensured.

[0119] For the preliminary hardware protection results, a content detection method based on the illegal content feature library is used to identify illegal content in real time. Through image processing technology, the key features of illegal content are captured, the content blocking mechanism is triggered, and preliminary content blocking results are generated;

[0120] This method uses deep learning algorithms to analyze the images and video content played on the display screen to identify whether there is prohibited information or inappropriate content. Implementing this content detection method not only helps to maintain the legality and compliance of the content played on the display screen, but also effectively reduces the legal liability and social opinion risks caused by the mistaken broadcast of inappropriate content. Through the automated content monitoring and interception process, the system can ensure that high-quality, standard-compliant content is continuously displayed to the audience, protecting the brand image while improving the overall experience of the audience.

[0121] First, a feature library containing various illegal contents needs to be established. These features can be images, texts, and other forms of information. The feature library can contain content such as bad advertisements, images of sensitive topics, or any content that may affect the brand image. By formatting this data, the system can perform real-time detection more efficiently. During real-time monitoring, the camera system continuously captures the images played on the display screen. Subsequently, these image data are processed and analyzed by deep learning models such as YOLOv7. Based on pre-training, these models can automatically identify illegal elements in the displayed content. For example, if an advertisement contains inappropriate characters or images, the YOLOv7 model will quickly identify and mark the corresponding content as illegal. The rapid response capability of this process is crucial to ensure the compliance of the content displayed on the display screen and prevent potential legal risks. When the illegal content is confirmed, the system will automatically trigger the content blocking mechanism, stop playing the relevant content, and may output alarms or notification information to let managers know the system status. At this time, the background system will also record the illegal events and conduct data analysis for future improvements. For example, if similar illegal content is detected multiple times in a row, the system will automatically update the feature library to enhance its detection capabilities, thereby achieving intelligent and automated management of displayed content.

[0122] Imagine that at a commercial exhibition, the display screen plays advertising content, and the system uses the deployed YOLOv7 model to detect the content being played in real time.

[0123] At the beginning of this process, the video parsing layer first collects image data and continuously monitors the advertisements played on the display screen. Assuming that an advertisement contains inappropriate images, the system identifies this content through the YOLOv7 model and marks it as a violation. The system's own illegal content feature library already contains some key features, such as specific image logos or text, which can be compared very quickly.

[0124] When the content is judged to be illegal, the system automatically triggers the content blocking mechanism to stop the advertisement immediately and display a warning message on the display screen to remind the audience that the current advertisement content is unavailable. At the same time, the background system will record this event and update it to the feature library so that similar illegal content can be identified more effectively in the future. Managers will also receive notifications so that they can take further measures in a timely manner, such as reviewing and adjusting the advertisement content. Through this content detection and blocking mechanism, the system effectively maintains the compliance of the content displayed on the display screen.

[0125] For the preliminary content interception results, an anomaly detection method based on environmental data is used to monitor the light intensity and crowd density in real time. The brightness of the display screen is optimized through the adaptive light adjustment algorithm to generate preliminary environmental adaptation results.

[0126] This method can automatically adjust the brightness and content display of the display according to changes in environmental conditions to ensure that users always get the best viewing experience. The implementation of this environmental adaptation method will significantly improve the adaptability of the display under different environmental conditions. By real-time monitoring and intelligently adjusting the brightness, it can effectively avoid content visibility problems caused by changes in lighting and improve user experience. At the same time, the ability to adapt to changes in the external environment enables the display to perform optimally in various scenarios (such as outdoor advertising, exhibitions, etc.), enhancing commercial value.

[0127] Specifically, light sensors and crowd density sensors are first set up around the display screen to obtain the lighting information and audience flow of the surrounding environment in real time. When the intensity of the ambient light changes greatly, the light sensor will capture different light values and transmit these data to the processing unit. The processing unit will analyze the relationship between the current light intensity and the brightness setting of the display screen. When the light intensity decreases, the system can automatically adjust the brightness of the display screen to ensure that the audience can see clearly. For example, in an outdoor environment, when the surrounding daylight intensity decreases, the brightness of the display screen will automatically increase to prevent the information display from being unclear. In an environment with strong light, the system will lower the brightness to protect the viewer's eyes and save energy. In addition, the crowd density sensor also plays an important role. When an increase in the flow of people is detected, the system can consider adjusting the display content to guide more audience attention, or change the content playback strategy according to the flow of people, such as increasing the display frequency of advertisements or information. Through continuous monitoring and adjustment, the entire system ensures that the best visual display effect can always be provided under different environmental conditions to improve the user's viewing experience.

[0128] For example, in an outdoor open-air event, the display screen is used to play event information and advertisements. First, the environmental perception layer deploys light sensors and crowd density sensors. When the light intensity sensor detects direct sunlight and the light intensity reaches 8000 lux, the system automatically analyzes this data and determines that the current environment requires an increase in the brightness of the display screen. At the same time, the crowd density sensor shows a high flow of people in front of the display screen, approximately 100 people per minute, which means that the attractiveness of the display screen needs to be improved for important information display. Based on the data of light intensity and crowd density, the system activates the light adaptive adjustment algorithm to automatically increase the brightness of the display screen from 350 nits to 550 nits to ensure that the information remains clearly visible. In the case of a high flow of people, the system may also increase the playback frequency of the advertisement content to attract the attention of more viewers. Through such dynamic adjustments, the display screen can continuously provide the best visual experience according to environmental changes and viewer needs.

[0129] For the preliminary hardware protection results, content interception results, and environmental adaptation results, an output method based on visualization technology is adopted to integrate them into the final display screen parameter regulation results. Through a real-time monitoring mechanism, the stability and efficiency of decision execution are ensured to generate the final regulation results.

[0130] Integrate the monitoring and regulation results to enable decision-makers to clearly understand the current status on the visualization interface and provide data support for formulating subsequent operations. The integrated output using visualization technology can greatly enhance the transparency and operability of data processing and decision-making. Decision-makers can quickly identify the current status and potential risks of the display screen through an intuitive graphical interface and make rapid feedback and adjustments. This not only improves work efficiency but also provides a strong backing for actual operations to ensure the effective implementation of various response measures.

[0131] In practice, first, a visual dashboard needs to be designed to integrate all monitoring data and processing results, including hardware status, content compliance, and environmental adaptability. This dashboard can display the current operating status in real time through graphics, colors, and charts, enabling managers to quickly understand the overall condition of the device. Secondly, the dashboard will update the display content through real-time data streams. Managers can intuitively view various parameters of the device, such as temperature, brightness, and current, and make timely decisions based on the displayed information. For example, when the real-time data shows that the temperature is too high, the dashboard will mark it with a red alert item to notify the manager to take cooling measures immediately. In this way, decision-makers can quickly respond to any potential anomalies to ensure that the system is always in the best operating state. Finally, the visual system also needs to integrate a feedback mechanism to evaluate the output results after the decision is executed. When managers adjust the display parameters or content, the background logic will automatically record these changes and conduct an effectiveness evaluation to analyze the response effects and optimization space under different circumstances. Through this design, the final generated regulation results can provide data support for subsequent management decisions, promoting the continuous optimization and improvement of the intelligent display system.

[0132] Taking the display management system of a large-scale exhibition as an example, all monitoring and decision-making results will be presented on an integrated visual dashboard. On this dashboard, managers can simultaneously view real-time data from hardware status monitoring, such as the temperature, current, and vibration conditions of the display screen. In addition, the content detection module is updated in real time to display the compliance of the currently playing content. Once a violation is intercepted, a warning will flash on the dashboard. At the same time, the environmental monitoring part will also display the current light intensity and crowd density data, reflecting environmental changes in real time. Based on this comprehensive information, managers can quickly respond. For example, when the ambient light intensity decreases, they can immediately brighten the display screen; when a content violation is detected, they can quickly replace the advertisement. The dashboard is also equipped with an automatic recording function to store the historical data of all decisions for subsequent analysis and optimization.

[0133] Another embodiment of the present invention provides an AI large model regulation system for display screen parameters. Refer to Figure 3 , the system may include:

[0134] An acquisition module 301, configured to collect multi-source data in real time according to the operating status of the display screen through a hardware sensing layer, a video parsing layer, and an environmental perception layer. Among them, the hardware sensing layer deploys temperature, current, and vibration sensor arrays, the video parsing layer extracts content features in real time through the YOLOv7 model, the environmental perception layer integrates light and crowd density detection technologies, and uses time series alignment technology to perform feature fusion on the multi-source data to obtain multi-modal feature data;

[0135] A construction module 302 is configured to construct a mechanism-data hybrid model according to the multimodal feature data. The mechanism core of the hybrid model is based on the display screen thermodynamic equation, and the data core of the hybrid model uses a fusion network of 3D-CNN and Transformer. The output weights of the mechanism core and the data core are dynamically adjusted through an adaptive weighting algorithm to obtain the initial output of the hybrid model.

[0136] An update module 303 is configured to update the hybrid model coefficients in real time according to the initial output of the hybrid model by using the recursive least squares method based on a sliding window. A teacher-student model architecture is established through knowledge distillation technology. Combined with incremental learning, a feature replay buffer is designed to save key samples, and the hybrid model parameters are dynamically optimized to obtain the optimized output of the hybrid model.

[0137] A regulation module 304 is configured to design a cascaded anomaly detection mechanism according to the optimized output of the hybrid model. Through three-dimensional joint diagnosis of hardware-content-environment, combined with a three-level response strategy of over-temperature protection and over-current protection and a violation content feature library interception mechanism, hardware protection, content interception, and environment adaptation decisions are executed in real time to obtain the final display screen parameter regulation result.

[0138] An embodiment of the present invention also provides a storage medium, in which a computer program is stored. The computer program is configured to execute the steps in any one of the above method embodiments when running.

[0139] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for executing the following steps:

[0140] S201, according to the operating state of the display screen, multi-source data is collected in real time through a hardware sensing layer, a video parsing layer, and an environment perception layer. The hardware sensing layer deploys temperature, current, and vibration sensor arrays. The video parsing layer extracts content features in real time through the YOLOv7 model. The environment perception layer integrates light and pedestrian flow density detection technologies, and uses a time series alignment technology to perform feature fusion on the multi-source data to obtain multimodal feature data.

[0141] S202, according to the multimodal feature data, construct a mechanism-data hybrid model. The mechanism core of the hybrid model is based on the display screen thermodynamic equation, and the data core of the hybrid model uses a fusion network of 3D-CNN and Transformer. The output weights of the mechanism core and the data core are dynamically adjusted through an adaptive weighting algorithm to obtain the initial output of the hybrid model.

[0142] S203. Based on the initial output of the hybrid model, the recursive least squares method based on a sliding window is used to update the hybrid model coefficients in real time. Among them, a teacher-student model architecture is established through knowledge distillation technology. Combining incremental learning, a feature replay buffer is designed to save key samples, the hybrid model parameters are dynamically optimized, and the optimized output of the hybrid model is obtained.

[0143] S204. Based on the optimized output of the hybrid model, a cascaded anomaly detection mechanism is designed. Among them, through the three-dimensional joint diagnosis of hardware-content-environment, combined with the three-level response strategy of over-temperature protection and over-current protection and the violation content feature library interception mechanism, hardware protection, content interception, and environment adaptation decisions are executed in real time to obtain the final display screen parameter regulation result.

[0144] An embodiment of the present invention also provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0145] Specifically, the above electronic device may further include a transmission device and an input / output device. Among them, the transmission device is connected to the above processor, and the input / output device is connected to the above processor.

[0146] Specifically, in this embodiment, the above processor may be configured to execute the following steps through a computer program:

[0147] S201. According to the operating state of the display screen, multi-source data is collected in real time through a hardware sensing layer, a video parsing layer, and an environment perception layer. Among them, a temperature, current, and vibration sensor array is deployed in the hardware sensing layer. The video parsing layer extracts content features in real time through the YOLOv7 model. The environment perception layer integrates light and crowd density detection technologies, and a multi-modal feature data is obtained by using a time series alignment technology to fuse the features of multi-source data.

[0148] S202. According to the multi-modal feature data, a mechanism-data hybrid model is constructed. Among them, the mechanism kernel of the hybrid model is based on the display screen thermodynamics equation, and the data kernel of the hybrid model uses a fusion network of 3D-CNN and Transformer. The output weights of the mechanism kernel and the data kernel are dynamically adjusted through an adaptive weighting algorithm to obtain the initial output of the hybrid model.

[0149] S203. Based on the initial output of the hybrid model, the recursive least squares method based on a sliding window is used to update the hybrid model coefficients in real time. Among them, a teacher-student model architecture is established through knowledge distillation technology. Combining incremental learning, a feature replay buffer is designed to save key samples, the hybrid model parameters are dynamically optimized, and the optimized output of the hybrid model is obtained.

[0150] S204. Design a cascaded anomaly detection mechanism according to the output of the optimized hybrid model. Among them, through the joint diagnosis of the three dimensions of hardware-content-environment, combined with the three-level response strategy of over-temperature protection and over-current protection and the interception mechanism of the illegal content feature library, the decisions of hardware protection, content interception, and environmental adaptation are executed in real time to obtain the final display screen parameter regulation result.

[0151] The structure, features, and effects of the present invention have been described in detail based on the embodiments shown in the drawings. The above is only the preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the drawings. Any changes made according to the concept of the present invention, or equivalent embodiments modified into equivalent changes, should still be within the protection scope of the present invention as long as they do not exceed the spirit covered by the specification and the drawings.

Claims

1. An AI large model regulation method for display screen parameters, characterized in that, The method includes: According to the running state of the display screen, multi-source data is collected in real time through the hardware sensing layer, video parsing layer, and environmental perception layer. Among them, the hardware sensing layer deploys temperature, current, and vibration sensor arrays. The video parsing layer extracts content features in real time through the YOLOv7 model. The environmental perception layer integrates light and pedestrian flow density detection technologies, and uses time series alignment technology to fuse the features of multi-source data to obtain multi-modal feature data; According to the multi-modal feature data, a mechanism-data hybrid model is constructed. Among them, the mechanism kernel of the hybrid model is based on the display screen thermodynamics equation, and the data kernel of the hybrid model uses a fusion network of 3D-CNN and Transformer. The output weights of the mechanism kernel and the data kernel are dynamically adjusted through adaptive weighting to obtain the initial output of the hybrid model; According to the initial output of the hybrid model, the recursive least squares method based on a sliding window is used to update the hybrid model coefficients in real time. Among them, a teacher-student model architecture is established through knowledge distillation technology, combined with incremental learning, a feature replay buffer is designed to save key samples, the hybrid model parameters are dynamically optimized, and the optimized output of the hybrid model is obtained; According to the optimized output of the hybrid model, a cascaded anomaly detection mechanism is designed. Among them, through three-dimensional joint diagnosis of hardware-content-environment, combined with over-temperature protection and over-current protection three-level response strategies and a mechanism for intercepting illegal content feature libraries, hardware protection, content interception, and environmental adaptation decisions are executed in real time to obtain the final display screen parameter regulation result.

2. The method according to claim 1, wherein The multi-source data is collected in real time through the hardware sensing layer, video parsing layer, and environmental perception layer according to the running state of the display screen. Among them, the hardware sensing layer deploys temperature, current, and vibration sensor arrays. The video parsing layer extracts content features in real time through the YOLOv7 model. The environmental perception layer integrates light and pedestrian flow density detection technologies, and uses time series alignment technology to fuse the features of multi-source data to obtain multi-modal feature data, including: According to the running state of the display screen, the hardware state data is collected in real time through the temperature sensor, current sensor, and vibration sensor deployed in the hardware sensing layer. The 48-dimensional content features are extracted in real time through the YOLOv7 model in the video parsing layer. The environmental data is collected in real time through the integration of light sensors and pedestrian flow density detection technologies in the environmental perception layer; For the hardware state data, content features, and environmental data, time series alignment technology is used, combined with the timestamp alignment algorithm to eliminate time differences; the missing data is supplemented through interpolation filling methods to generate a time-synchronized multi-source data set; For the time-synchronized multi-source data set, a feature fusion method based on a cross-modal attention mechanism is used to map the hardware state data, content feature data, and environmental data to a unified multi-modal feature space. Through the feature weighting mechanism, the correlation relationships between different modalities are captured to generate the final multi-modal feature data.

3. The method according to claim 2, wherein Construct a mechanism-data hybrid model based on the multimodal feature data. Among them, the mechanism kernel of the hybrid model is based on the display thermodynamic equation, and the data kernel of the hybrid model uses a fusion network of 3D-CNN and Transformer. The output weights of the mechanism kernel and the data kernel are dynamically adjusted through an adaptive weighting algorithm to obtain the initial output of the hybrid model, including: According to the hardware state data in the multimodal feature data, initialize the mechanism kernel using the display thermodynamic equation, and through a dynamic weight allocation mechanism, ensure the rationality and accuracy of the initial parameters of the mechanism kernel, and generate a preliminary mechanism model; For the content feature data and environmental data in the multimodal feature data, use 3D-CNN to extract features in the spatial and temporal dimensions, combine with the Transformer model to capture long-term sequence dependencies, construct a data kernel, and through an adaptive learning rate adjustment mechanism, optimize the training process of the data kernel to generate a preliminary data model; For the preliminary mechanism model and the preliminary data model, use an adaptive weighting algorithm to dynamically adjust the output weights of the mechanism kernel and the data kernel according to the temperature change rate and current fluctuation. Through an error feedback mechanism, optimize the weight allocation to obtain a mechanism-data hybrid model and generate the initial output of the hybrid model.

4. The method according to claim 3, characterized in that, According to the initial output of the hybrid model, use the recursive least squares method based on a sliding window to update the hybrid model coefficients in real time. Among them, establish a teacher-student model architecture through knowledge distillation technology, combine incremental learning, design a feature replay buffer to save key samples, dynamically optimize the hybrid model parameters, and obtain the optimized output of the hybrid model, including: According to the initial output of the hybrid model, use the recursive least squares method based on a sliding window to calculate the model output error in real time, and through a dynamic weight adjustment mechanism, update the coefficients of the mechanism kernel and the data kernel to generate a preliminary optimized model; For the preliminary optimized model, use knowledge distillation technology, use the hybrid model as the teacher model, train a lightweight student model, and through an adaptive learning rate adjustment mechanism, ensure the accuracy and efficiency of the student model to generate a preliminary compressed model; For the preliminary compressed model, use the incremental learning method, combine the feature replay buffer to save key samples, and through the sample weight allocation mechanism, dynamically optimize the model parameters to generate the final optimized hybrid model and obtain the output of the optimized hybrid model.

5. The method according to claim 4, characterized in that, According to the output of the optimized hybrid model, design a cascaded anomaly detection mechanism. Among them, through three-dimensional joint diagnosis of hardware-content-environment, combine a three-level response strategy of over-temperature protection and over-current protection and a violation content feature library interception mechanism, and execute hardware protection, content interception, and environmental adaptation decisions in real time to obtain the final display parameter regulation result, including: According to the output of the optimized hybrid model, use an anomaly detection method based on hardware state data to monitor temperature, current, and vibration data in real time, and through a three-level response strategy, trigger the hardware protection mechanism to generate a preliminary hardware protection result; For the preliminary hardware protection results, a content detection method based on a violation content feature library is adopted to identify violation content in real time. Through image processing technology, the key features of the violation content are captured, triggering a content interception mechanism to generate preliminary content interception results; For the preliminary content interception results, an anomaly detection method based on environmental data is adopted to monitor the light intensity and pedestrian flow density in real time. Through the light intensity adaptive adjustment algorithm, the brightness of the display screen is optimized to generate preliminary environmental adaptation results; For the preliminary hardware protection results, content interception results and environmental adaptation results, an output method based on visualization technology is adopted to integrate them into the final display screen parameter regulation results. Through a real-time monitoring mechanism, the stability and efficiency of decision execution are ensured to generate the final regulation results.

6. An AI large model control system for display screen parameters, characterized in that, The system includes: An acquisition module, which is used to collect multi-source data in real time according to the operating state of the display screen through a hardware sensing layer, a video parsing layer and an environmental perception layer. Among them, the hardware sensing layer deploys temperature, current and vibration sensor arrays, the video parsing layer extracts content features in real time through the YOLOv7 model, the environmental perception layer integrates light and pedestrian flow density detection technologies, and uses time series alignment technology to perform feature fusion on the multi-source data to obtain multi-modal feature data; A construction module, which is used to construct a mechanism-data hybrid model according to the multi-modal feature data. Among them, the mechanism kernel of the hybrid model is based on the display screen thermodynamic equation, and the data kernel of the hybrid model uses a fusion network of 3D-CNN and Transformer. The output weights of the mechanism kernel and the data kernel are dynamically adjusted through an adaptive weighting algorithm to obtain the initial output of the hybrid model; An update module, which is used to update the hybrid model coefficients in real time according to the initial output of the hybrid model by using the recursive least squares method based on a sliding window. Among them, a teacher-student model architecture is established through knowledge distillation technology, combined with incremental learning, a feature replay buffer is designed to save key samples, the hybrid model parameters are dynamically optimized, and the optimized output of the hybrid model is obtained; A regulation module, which is used to design a cascaded anomaly detection mechanism according to the optimized output of the hybrid model. Among them, through a three-dimensional joint diagnosis of hardware-content-environment, combined with a three-level response strategy of over-temperature protection and over-current protection and a violation content feature library interception mechanism, hardware protection, content interception and environmental adaptation decisions are executed in real time to obtain the final display screen parameter regulation results.

7. The system according to claim 6, wherein The acquisition module is specifically used for: According to the operating state of the display screen, the hardware state data is collected in real time through the temperature sensor, current sensor and vibration sensor deployed in the hardware sensing layer, the 48-dimensional content features are extracted in real time through the YOLOv7 model in the video parsing layer, and the environmental data is collected in real time through the light sensor and pedestrian flow density detection technology integrated in the environmental perception layer; For the hardware state data, content features and environmental data, time series alignment technology is adopted, and the time difference is eliminated by combining the timestamp alignment algorithm; the missing data is supplemented by the interpolation filling method to generate a time-synchronized multi-source data set; For a time-synchronized multi-source dataset, a feature fusion method based on a cross-modal attention mechanism is adopted to map the hardware state data, content feature data, and environmental data into a unified multi-modal feature space. Through a feature weighting mechanism, the correlation relationships between different modalities are captured to generate the final multi-modal feature data.

8. The system according to claim 7, wherein The construction module is specifically configured to: According to the hardware state data in the multi-modal feature data, initialize the mechanism kernel using the display thermodynamic equation, and through a dynamic weight allocation mechanism, ensure the rationality and accuracy of the initial parameters of the mechanism kernel to generate a preliminary mechanism model; For the content feature data and environmental data in the multi-modal feature data, extract the spatial and temporal dimension features using 3D-CNN, combine the Transformer model to capture the long-term time-dependent relationships, construct a data kernel, and through an adaptive learning rate adjustment mechanism, optimize the training process of the data kernel to generate a preliminary data model; For the preliminary mechanism model and the preliminary data model, adopt an adaptive weighting algorithm to dynamically adjust the output weights of the mechanism kernel and the data kernel according to the temperature change rate and current fluctuation. Through an error feedback mechanism, optimize the weight allocation to obtain a mechanism-data hybrid model and generate the initial output of the hybrid model.

9. A storage medium, characterized in that, A computer program is stored in the storage medium, wherein the computer program is set to execute the method according to any one of claims 1-5 when running.

10. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is set to run the computer program to execute the method according to any one of claims 1-5.

Images