Display screen dynamic thermal management system based on multi-feature fusion and intelligent prediction

Abstract

The invention discloses a display screen dynamic thermal management system based on multi-feature fusion and intelligent prediction. The dynamic thermal management system for the display screen comprises a multi-dimensional feature perception and fusion module used for fusing multi-dimensional features to obtain a comprehensive state vector; the multi-dimensional features comprise at least one of a display load, ambient light data, historical display screen temperature data and a historical brightness adjustment sequence, and current display screen temperature data; the intelligent thermal prediction and anomaly detection module is used for inputting the comprehensive state vector into a neural network model, so that the neural network model outputs a prediction result; performing anomaly detection according to the prediction result to obtain an anomaly detection result; and the adaptive temperature control strategy execution module is used for executing a corresponding temperature control strategy according to the prediction result and the anomaly detection result. Through the above mode, passive response can be turned to active prediction, single-dimension monitoring can be turned to multi-feature fusion, and rough control can be turned to refined and intelligent management.

Description

Technical Field

[0001] The present application relates to the field of display screen technology, and in particular to a dynamic thermal management system for display screens based on multi-feature fusion and intelligent prediction. Background Art

[0002] Traditional NTC (negative temperature coefficient) thermistor (NTC) temperature control solutions for displays rely on real-time detection of the overall display temperature or the temperature of specific points using an NTC resistor. When the temperature reaches a preset threshold (e.g., 85°C), the system takes measures such as linearly reducing screen brightness to dissipate heat and prevent overheating. This simple and straightforward solution provides some degree of protection.

[0003] However, with the development of display technology, especially the trend towards high resolution, high refresh rate, high brightness and integration of more functions, the limitations of traditional NTC solutions are becoming increasingly prominent:

[0004] Delayed response: Traditional solutions are usually “after-the-fact” responses, that is, intervention is initiated only when the temperature is already high, which may not be able to effectively prevent local instantaneous overheating or heat accumulation in the early stages.

[0005] Lack of predictability: Unable to predict future temperature changes and potential hot spots based on current operating conditions and environmental trends.

[0006] "One-size-fits-all" problem: Brightness adjustment based on a single temperature threshold may be too rough and fail to consider the non-uniformity of heat distribution. This may lead to unnecessary global brightness reduction, affecting the user experience, or inadequate heat dissipation in local overheating areas.

[0007] Single information dimension: Relying only on temperature, a single dimension of information, fails to fully utilize other relevant data generated during the operation of the display, such as load, image content characteristics, etc. Summary of the Invention

[0008] The dynamic thermal management system for display screens based on multi-feature fusion and intelligent prediction provided in this application can shift from passive response to active prediction, from single-dimensional monitoring to multi-feature fusion, and from rough control to refined and intelligent management.

[0009] In the first aspect, the present application provides a dynamic thermal management system for a display screen based on multi-feature fusion and intelligent prediction, and the dynamic thermal management system for the display screen includes: a multi-dimensional feature perception and fusion module, which is used to fuse multi-dimensional features to obtain a comprehensive state vector; wherein the multi-dimensional features include at least one of display load, ambient light data, historical display screen temperature data, historical brightness adjustment sequence, and current display screen temperature data; an intelligent thermal prediction and anomaly detection module, which is used to input the comprehensive state vector into a neural network model so that the neural network model outputs a prediction result; and perform anomaly detection based on the prediction result to obtain anomaly detection result; wherein the prediction result includes at least one of the predicted temperature trend of the display screen, the predicted hot spot position on the display screen and the temperature corresponding to the hot spot position, and the frequency-space thermal behavior analysis result; an adaptive temperature control strategy execution module, which is used to execute the corresponding temperature control strategy based on the prediction result and the anomaly detection result.

[0010] Among them, the neural network model includes a model input layer and a main network. The intelligent thermal prediction and anomaly detection module is also used to input the comprehensive state vector into the model input layer for format conversion to obtain input features, and input the input features into the main network to obtain prediction results.

[0011] Among them, the main network includes: an encoder, a frequency domain cross-scale interaction layer, a decoder and a prediction head; the intelligent thermal prediction and anomaly detection module is also used to input the input features into the encoder for encoding and output features at different scales; and input the features at different scales into the frequency domain cross-scale interaction layer for fusion and output fused features; and input the fused features into the decoder for decoding and output decoded features; and input the decoded features into the prediction head for prediction and output the prediction results.

[0012] The encoder includes a frequency-space domain heat dissipation encoder, which is used to extract heat dissipation features from the space domain and the frequency domain.

[0013] Among them, the frequency domain cross-scale interaction layer uses embedding, multi-head attention, and cross-attention mechanisms to perform cross-scale feature fusion in the frequency domain.

[0014] Among them, the main network is trained using spatial loss and frequency loss.

[0015] Among them, the intelligent thermal prediction and anomaly detection module is also used to establish a dynamic baseline based on the prediction results and historical normal operation data; and perform anomaly detection based on the current display temperature data, prediction results and baseline to obtain anomaly detection results.

[0016] Among them, the intelligent thermal prediction and anomaly detection module is also used to obtain the feature mean based on WMA calculation, and perform anomaly detection in combination with the sensitivity factor and the upper and lower thresholds to obtain anomaly detection results; among them, the upper and lower thresholds can be dynamically adjusted.

[0017] Among them, the adaptive temperature control strategy execution module is also used to obtain the thermal risk level based on the prediction results and the anomaly detection results, and execute the corresponding temperature control strategy according to the thermal risk level.

[0018] Among them, the thermal risk levels include the first risk level, the second risk level, the third risk level and the fourth risk; among them, the first risk level is lower than the second risk level, the second risk level is lower than the third risk level, and the third risk level is lower than the fourth risk.

[0019] The beneficial effects of the present application are as follows: Different from the prior art, the present application provides a dynamic thermal management system for a display screen based on multi-feature fusion and intelligent prediction, which includes: a multi-dimensional feature perception and fusion module for fusing multi-dimensional features to obtain a comprehensive state vector; wherein the multi-dimensional features include at least one of display load, ambient light data, historical display screen temperature data, historical brightness adjustment sequence, and current display screen temperature data; an intelligent thermal prediction and anomaly detection module for inputting the comprehensive state vector into a neural network model so that the neural network model outputs a prediction result; and performing anomaly detection based on the prediction result to obtain anomaly detection results; wherein the prediction result includes at least one of the predicted temperature trend of the display screen, the predicted hot spot position on the display screen and the temperature corresponding to the hot spot position, and the frequency-space thermal behavior analysis result; an adaptive temperature control strategy execution module for executing the corresponding temperature control strategy based on the prediction result and the anomaly detection result. The present application solves the problems of the existing NTC temperature control scheme for display screens, such as delayed response, lack of prediction capability, single control strategy, and insufficient information dimension. Specifically, before the display screen temperature reaches the traditional NTC threshold, it can detect potential thermal risks in advance through prediction and abnormal pattern recognition, and provide early warning and proactive intervention. It can not only sense the global temperature, but also identify and locate local hotspots inside the display screen, and perform differentiated brightness adjustments or other heat dissipation measures based on the severity and location of the hotspots, achieving precise positioning and differentiated regulation. In resource-constrained display screen controllers, it can implement intelligent thermal management with lower hardware overhead and computational complexity, taking into account both lightweight and high efficiency, and enable the thermal management system to adapt to different display content, ambient temperature and usage modes, maintain robustness and efficiency, and improve adaptability to multiple working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without inventive efforts. Among them:

[0021] Figure 1 This is a structural diagram of an embodiment of a dynamic thermal management system for a display screen based on multi-feature fusion and intelligent prediction provided by this application;

[0022] Figure 2 This is a schematic diagram of the workflow of the dynamic thermal management system for display screens based on multi-feature fusion and intelligent prediction in this application. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It will be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some, rather than all, structures related to the present application are shown in the drawings. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0024] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0025] See Figure 1 , Figure 1 This is a schematic diagram of an embodiment of a dynamic thermal management system for display screens based on multi-feature fusion and intelligent prediction, as provided in this application. The dynamic thermal management system for display screens includes a multi-dimensional feature perception and fusion module, an intelligent thermal prediction and anomaly detection module, and an adaptive temperature control strategy execution module.

[0026] The multi-dimensional feature perception and fusion module is used to fuse multi-dimensional features to obtain a comprehensive state vector; wherein the multi-dimensional features include at least one of display load, ambient light data, historical display screen temperature data, historical brightness adjustment sequence, and current display screen temperature data.

[0027] Display load can be obtained by monitoring the current processing load of the display controller or graphics processing unit (GPU / CPU, if applicable), rendering frame rate, and image content complexity (e.g., by analyzing image histograms, edge density, etc.). These can be used as proxy indicators of "congestion" or "activity intensity," that is, as display load.

[0028] Ambient light data can be collected by ambient light sensors because changes in ambient light can affect the user's perception of brightness, which in turn may affect the preset brightness strategy.

[0029] The historical display temperature data and the historical brightness adjustment sequence may be NTC temperature and brightness adjustment sequences over a period of time in the past.

[0030] In some embodiments, the multi-dimensional feature perception and fusion module is used to pre-process the collected multi-dimensional features (such as normalization and time series alignment), and perform preliminary fusion and smoothing through a lightweight algorithm (such as weighted moving average WMA, Weighted Moving Average) to form a comprehensive state vector to obtain a comprehensive state vector.

[0031] The intelligent thermal prediction and anomaly detection module is used to input the comprehensive state vector into the neural network model so that the neural network model outputs a prediction result; and perform anomaly detection based on the prediction result to obtain anomaly detection result; wherein the prediction result includes at least one of the predicted temperature trend of the display screen, the predicted hot spot position on the display screen and the temperature corresponding to the hot spot position, and the frequency-space thermal behavior analysis result.

[0032] In some embodiments, the neural network model can be a lightweight neural network model, such as a small MLP, RNN, or a simplified CNN variant. The input of the neural network model is the fused multi-dimensional feature vector (comprehensive state vector).

[0033] The training data for this neural network model can be obtained through precise finite element simulation or extensive real-world testing of displays under different operating conditions. This data set includes input features, corresponding NTC temperatures, detailed temperature distribution across the display screen (thermal imaging data), hotspot locations, and evolution trends. Particular attention is paid to thermal behavior in different display modes (video, gaming, text), at different brightness settings, and at different ambient temperatures.

[0034] In some embodiments, the neural network model includes a model input layer and a main network. The intelligent thermal prediction and anomaly detection module is also used to input the comprehensive state vector into the model input layer for format conversion to obtain input features, and input the input features into the main network to obtain prediction results.

[0035] The input of the model input layer is the comprehensive state vector collected and preprocessed by the multi-dimensional feature perception and fusion module.

[0036] The input of the model input layer will be the characteristics unique to the display, such as: NTC real-time temperature, display load, ambient light, historical temperature / brightness data series, etc.

[0037] The model input layer is equipped with PPNet (Parameter Preprocessing Network), which is used to convert the original multi-dimensional sensor data and status information into a format suitable for neural network model input.

[0038] In some embodiments, the main network includes: an encoder, a frequency domain cross-scale interaction layer, a decoder, and a prediction head.

[0039] The intelligent thermal prediction and anomaly detection module is also used to input the input features into the encoder for encoding and output features at different scales; input the features at different scales into the frequency domain cross-scale interaction layer for fusion and output fused features; input the fused features into the decoder for decoding and output decoded features; and input the decoded features into the prediction head for prediction and output the prediction results.

[0040] The encoder includes a frequency-spatial domain thermal dissipation encoder, which is used to extract thermal dissipation features from both the spatial and frequency domains. The encoder gradually extracts information from the input features at different scales through downsampling. The FSTE (Frequency-Spatial Domain Thermal Encoder) module simultaneously extracts thermal dissipation features from both the spatial and frequency domains through its internal HL-FSE (High-to-low frequency and spatial domain embedding) structure. FSTE (particularly HL-FSE): Its dual-branch (spatial and frequency domain) design is the core, ensuring that the model simultaneously focuses on local details and global / frequency characteristics. DWConv (depthwise separable convolution), 3D-DCT (three-dimensional discrete cosine transform), 3D-IDCT (three-dimensional inverse discrete cosine transform), and learnable frequency weights are key technologies to achieve this goal.

[0041] The encoder can learn the overall features and time series relationships to perform short-term NTC temperature prediction.

[0042] FSTE's ability to extract both spatial and frequency features (high frequencies correspond to rapidly changing hotspots, while low frequencies correspond to slowly accumulating heat) enables hotspot location and intensity prediction. The spatial branch of HL-FSE focuses on local spatial details, while the frequency branch focuses on global and variable-rate components.

[0043] Indirectly implement "frequency-space thermal behavior analysis": By learning these features, the neural network model can understand how heat is distributed and evolves in different areas of the display (space) and over time (frequency / rate of change), thereby indirectly implementing frequency-space thermal behavior analysis.

[0044] The frequency domain cross-scale interaction layer utilizes embedding, multi-head attention, and cross-attention mechanisms to perform cross-scale feature fusion in the frequency domain. This layer fuses features extracted by the encoder at different scales, helping to bridge the semantic gap and achieve global-to-local thermal dissipation perception. The neural network model comprehensively considers both the influence of large-scale background temperatures and localized heating conditions, resulting in more accurate overall temperature and hotspot predictions.

[0045] In some embodiments, the main network is trained using spatial loss and frequency loss. It includes spatial loss (directly comparing the predicted temperature map and the actual temperature map) and frequency loss (comparing the amplitude and phase after performing a 2D DFT transformation on the temperature map). When training the neural network model offline, a similar loss function also needs to be designed. Spatial loss ensures that the predicted temperature value is accurate, while frequency loss can help the model better learn the gradient changes of heat and high-frequency details (such as sharp hotspot edges), thereby improving the accuracy of hotspot location and intensity prediction.

[0046] The prediction objectives of the neural network model are as follows:

[0047] Short-term NTC temperature prediction: Predict the temperature trend of the NTC monitoring point in the next few seconds or tens of seconds.

[0048] Hotspot Location and Intensity Prediction: Predicts possible local hotspots and their approximate temperatures that may appear on the screen.

[0049] Frequency-spatial thermal behavior analysis: including spatial perception and time / frequency perception.

[0050] Spatial Perception: During training, the neural network model indirectly learns the distribution of heat across different areas of the screen by studying thermal imaging data. In practice, while complex frequency-spatial transformations are not performed directly, the neural network model can infer potential heat concentrations based on input features (such as indicators of high-load areas).

[0051] Time / frequency awareness: By analyzing the time series of input features (such as temperature change rate and load fluctuation frequency), the neural network model can identify patterns of rapid temperature rise (high-frequency thermal events) or continuous accumulation (low-frequency thermal events).

[0052] In some embodiments, the intelligent thermal prediction and anomaly detection module is also used to establish a dynamic baseline based on the prediction results and historical normal operation data; and perform anomaly detection based on the current display screen temperature data, prediction results and baseline to obtain anomaly detection results.

[0053] Among them, the intelligent thermal prediction and anomaly detection module is also used to obtain the feature mean based on WMA calculation, and perform anomaly detection in combination with the sensitivity factor and the upper and lower thresholds to obtain anomaly detection results; among them, the upper and lower thresholds can be dynamically adjusted.

[0054] In some embodiments, the intelligent anomaly detection process is as follows:

[0055] Baseline establishment: Based on the prediction results of the agent model and historical normal operation data, a dynamic "normal thermal behavior" baseline is established.

[0056] Deviation analysis: Deviation between real-time NTC readings, fused features, and predicted values ​​and baseline.

[0057] Abnormality Detection: When the actual NTC temperature, fused features, or their changing trends significantly deviate from the prediction model output or normal baseline (for example, using simplified statistical methods such as dynamic thresholds based on WMA (weighted moving average) and standard deviation multiples, the upper and lower limits (U Threshold) and L Threshold are dynamically set based on the feature mean calculated by WMA and an adjustable sensitivity factor (σTuning) to set the dynamic normal range for each key feature (such as NTC temperature, predicted temperature, predicted hotspot intensity, etc.). However, the standard deviation can be approximated using a simplified method such as the mean absolute deviation (MAD), and efficient calculation is achieved using shift operations), the dynamic upper and lower thresholds are stored in the threshold register, and the current real-time feature value is compared with the corresponding dynamic threshold in the threshold register. If one or more features exceed their dynamic threshold range, the system is determined to be abnormal.

[0058] Well, this application has designed a dynamic threshold formula with a certain degree of creativity, which not only takes the mean and dispersion into consideration, but also incorporates the changing trend of the features and the volatility self-regulation mechanism.

[0059] Quantitative calculation of anomaly judgment: trend perception and fluctuation adaptive dynamic weighted threshold.

[0060] Core idea:

[0061] Baseline: Still centered around the weighted moving average (WMA) of the feature.

[0062] Discrepancy: Uses the WMA of the recent absolute deviations of the feature from its WMA as a dynamic, lightweight approximation of the standard deviation

[0063] Trend perception (mainly affects the upper limit): When the WMA of the feature shows a rapid upward trend, dynamically "tighten" the distance between the upper threshold and the WMA (that is, the effective K value becomes smaller) to make it more sensitive.

[0064] Fluctuation adaptation (affecting both upper and lower limits): When the feature itself has recently fluctuated greatly, the threshold range is appropriately "relaxed" (that is, the effective K value becomes larger) to tolerate normal noise and reduce false positives; conversely, when the feature is stable, the threshold range can be relatively tightened.

[0065] Formula components (Chinese expression):

[0066] The weighted moving average of the feature at the current time t (WMA_t):

[0067] WMA_t (characteristics)

[0068] Estimation of characteristic recent dynamic dispersion

[0069]

[0070] |feature-WMA_t(feature)|: Calculates the absolute deviation of each data point from its own WMA.

[0071] WMA_t(...): Performs weighted moving average on the above absolute deviation series again to obtain the smoothed recent mean absolute deviation.

[0072] c: proportionality constant used to adjust the mean absolute deviation to an order of magnitude similar to the standard deviation (for example, for a normal distribution, the standard deviation is approximately equal to 1.253 times the mean absolute deviation, but here c can be adjusted based on experience and can even be a value similar to the WMA). t (feature) itself, to achieve more complex adaptation).

[0073] Trend strength factor (F_trend, t): Mainly used to adjust the sensitivity of the upper threshold to the upward trend:

[0074] F_trend,

[0075] (WMA_t(feature)-WMA_{t-Δt}(feature)) represents the change in WMA over the past Δt time window (i.e., short-term trend).

[0076] The trend change is normalized using dynamic discreteness, and ε is a small constant to prevent division by zero.

[0077] tanh(...): Hyperbolic tangent function, which maps the normalized trend strength to the interval (-1, 1), providing a bounded trend indicator that is less sensitive to extreme values. Positive values ​​indicate an increase, negative values ​​indicate a decrease.

[0078] λ(lambda): Trend influence weight coefficient (positive value), which controls the extent of trend adjustment on K value.

[0079] Volatility Adaptation Factor (F_volatility,t) - used to adjust the overall threshold width based on recent volatility:

[0080]

[0081] Calculate the normalized dynamic dispersion (an approximation of the coefficient of variation) to indicate the relative magnitude of fluctuations. A reference, expected, or historical average normalized dispersion level.

[0082] Indicates the deviation of the current relative volatility from the reference level. If the current volatility is greater than the reference level, this term is positive, otherwise it is negative.

[0083] μ(mu): Fluctuation impact weight coefficient (positive value), which controls the amplitude of fluctuation adjustment on K value.

[0084] Basic sensitivity coefficient (K_basic):

[0085] This is the benchmark for σTuning or K value, a preset constant.

[0086] Final dynamic threshold formula:

[0087] Dynamic upper threshold (U_Th, t):

[0088]

[0089] max(0, (F_trend, t) - 1)): This application is only interested in the tightening effect of an upward trend on the upper limit. When (F_trend, t) is greater than 1 (indicating an upward trend), this term is positive, causing the denominator to exceed 1, thereby reducing the effective K value K_base / (...), bringing the upper limit closer to the WMA. If there is no significant upward or downward trend ((F_trend, t) is close to 1 or less than 1), the denominator is close to 1, and the trend-adjusting effect is weakened.

[0090] (F_fluctuation, t): If the recent fluctuation is large ((F_fluctuation, t)>1), it will be multiplied by a factor greater than 1 to moderately relax the upper limit; if the recent stability ((F_fluctuation, t)<1), it will be multiplied by a factor less than 1 to moderately tighten the upper limit.

[0091] Dynamic lower threshold (L_Th, t):

[0092]

[0093] The lower bound is usually less critical to overheating, so its trend sensitivity can be designed differently or omitted. For asymmetry, (F_trend, t) is not directly used for tightening (since the primary goal is to prevent upward breakouts). If sensitivity to rapid declines is also required (for example, to prevent the system from overcooling or a certain indicator from falling too low), a factor tailored to the downtrend can be similarly designed.

[0094] (F_fluctuation, t) still acts on the lower limit, which is consistent with the logic of the upper limit: large fluctuations lead to a wide range, while small and stable fluctuations lead to a narrow range.

[0095] Parameter description and adjustment:

[0096] Δt: The size of the time window to look back when calculating trends.

[0097] c,λ,μ,K_base, These are all adjustable parameters that need to be set and optimized through experiments or domain knowledge based on the specific application scenarios and statistical characteristics of the features.

[0098] ε: a small constant, such as 1e-6.

[0099] The advantages are as follows:

[0100] Trend perception: More sensitive to rapidly rising risks (upper limit).

[0101] Fluctuation adaptation: It can dynamically adjust the threshold width according to the recent stability of the feature to achieve a balance between noise and real changes.

[0102] Asymmetry: The adjustment strategies for the upper and lower limits can be different, which is more in line with actual needs (for example, overheat protection mainly focuses on the upper limit).

[0103] Computationally friendly: It mainly relies on WMA and basic arithmetic operations. The tanh function can also be lightweight through table lookup or piecewise linear approximation.

[0104] In some embodiments, the adaptive temperature control strategy execution module is used to execute a corresponding temperature control strategy based on the prediction results and the abnormality detection results.

[0105] Among them, the adaptive temperature control strategy execution module is also used to obtain the thermal risk level based on the prediction results and the anomaly detection results, and execute the corresponding temperature control strategy according to the thermal risk level.

[0106] Among them, the thermal risk levels include the first risk level, the second risk level, the third risk level and the fourth risk; among them, the first risk level is lower than the second risk level, the second risk level is lower than the third risk level, and the third risk level is lower than the fourth risk.

[0107] In some embodiments, the adaptive temperature control strategy execution module operates as follows:

[0108] Temperature status assessment: Combining the real-time NTC temperature, short-term temperature prediction, hotspot prediction, and anomaly detection results of the neural network model, a comprehensive assessment of the current and future thermal status of the display is conducted, and different thermal risk levels are classified.

[0109] The graded responses are as follows:

[0110] Level 0 (Normal / Safe): The NTC temperature is well below the threshold, and no risk is predicted. Temperature Control Strategy: Operates according to user settings or ambient adaptive brightness. Level 0 corresponds to the first risk level described above.

[0111] Level 1 (Warning / Minor Risk): The NTC temperature is trending upward, or the model predicts it may approach the traditional NTC warning line in the future, or a minor thermal anomaly pattern has been detected. Level 1 corresponds to the second risk level mentioned above.

[0112] Measures (temperature control strategy): Perform a small, user-imperceptible global brightness smooth decrease in advance, or perform more precise, small-scale brightness adjustments for predicted local hotspots (if the display hardware supports zoned light control).

[0113] Delay mechanism optimization: The 5-minute delay judgment in the original plan can be retained, but combined with the predicted temperature rise rate, if the rate is extremely high, the delay will be shortened or enter the next level immediately.

[0114] Level 2 (Moderate Risk / Confirmed Anomaly): The NTC temperature has approached or reached the action threshold of traditional NTC solutions (e.g., ≥85°C), or the model predicts it will soon exceed it, or a significant thermal anomaly has been detected. Level 2 corresponds to the third risk level mentioned above.

[0115] Measures (temperature control strategy): Perform a more significant linear brightness reduction than the traditional solution (for example, day mode from 800cd / m 2 Down to 400cd / m 2 Rather than 500cd / m 2 , or night mode from 600cd / m 2 Down to 300cd / m 2Rather than 350cd / m 2 ). Prioritize reducing the brightness of predicted hotspot areas (if supported).

[0116] Level 3 (High Risk / Severe Anomaly): The NTC temperature remains above the threshold and cooling measures are ineffective, or the model predicts a significant temperature increase, or a severe thermal anomaly is detected (possibly indicating a hardware failure or extreme operating conditions). Level 3 corresponds to the fourth risk level mentioned above.

[0117] Measures (temperature control strategy): Implement maximum brightness reduction, issue an overheating warning to the user, and even trigger a protective screen shutdown in extreme cases.

[0118] Recovery mechanism: When the NTC temperature falls back to a safe range (e.g., <80°C; a recovery threshold different from 85°C can be set to avoid frequent oscillations), and the model predicts a low thermal risk for a period of time, brightness is gradually restored. The recovery rate can also be adaptively adjusted based on historical data and current operating conditions.

[0119] The following lightweight implementation considerations can be made in this application:

[0120] Model compression and quantization: Pruning and quantization are performed on the trained neural network model to adapt to the limited computing and storage resources of the display controller.

[0121] Hardware acceleration: For WMA and simplified statistical calculations, try to use hardware displacement, addition and other basic operation units as much as possible, and use displacement operations instead of multiplication and division.

[0122] Look-up table (LUT): For some complex functions with limited input range, you can consider using the look-up table approximation.

[0123] In this embodiment, the dynamic thermal management system for a display screen includes: a multi-dimensional feature perception and fusion module for fusing multi-dimensional features to obtain a comprehensive state vector; wherein the multi-dimensional features include at least one of display load, ambient light data, historical display screen temperature data, historical brightness adjustment sequences, and current display screen temperature data; an intelligent thermal prediction and anomaly detection module for inputting the comprehensive state vector into a neural network model so that the neural network model outputs a prediction result; and performing anomaly detection based on the prediction result to obtain an anomaly detection result; wherein the prediction result includes at least one of a predicted temperature trend for the display screen, a predicted hotspot location on the display screen and the temperature corresponding to the hotspot location, and a frequency-spatial thermal behavior analysis result; and an adaptive temperature control strategy execution module for executing the corresponding temperature control strategy based on the prediction result and the anomaly detection result. This solves the problems of existing NTC temperature control schemes for display screens, such as delayed response, lack of prediction capability, single control strategy, and insufficient information dimensionality. Specifically, before the display screen temperature reaches the traditional NTC threshold, it can detect potential thermal risks in advance through prediction and abnormal pattern recognition, and provide early warning and proactive intervention. It can not only sense the global temperature, but also identify and locate local hotspots inside the display screen, and perform differentiated brightness adjustments or other heat dissipation measures based on the severity and location of the hotspots, achieving precise positioning and differentiated regulation. In resource-constrained display screen controllers, it can implement intelligent thermal management with lower hardware overhead and computational complexity, taking into account both lightweight and high efficiency, and enable the thermal management system to adapt to different display content, ambient temperature and usage modes, maintain robustness and efficiency, and improve adaptability to multiple working conditions.

[0124] In some embodiments, see Figure 2 The workflow of the display screen dynamic thermal management system based on multi-feature fusion and intelligent prediction in this application is as follows:

[0125] Step 1: Data collection.

[0126] Collect NTC temperature, display load, ambient light, historical data, etc.

[0127] Step 2: Feature preprocessing and fusion.

[0128] Normalization and WMA smoothing are used to perform feature preprocessing and fusion on the collected data to form a comprehensive state vector.

[0129] Step 3: Intelligent prediction and anomaly detection.

[0130] For example, the state vector is input to a lightweight neural network model, which outputs short-term NTC temperature predictions, hotspot location and intensity predictions.

[0131] Combine real-time NTC readings, predicted values, and dynamic baselines for anomaly detection.

[0132] Step 4: Thermal risk assessment.

[0133] Combining various outputs, the thermal risk level (0-3) is assessed.

[0134] Step 5: Hierarchical adaptive temperature control decision.

[0135] Select the corresponding brightness adjustment strategy (including adjustment amplitude, range, rate and delay) according to the risk level.

[0136] Step 6: Policy execution.

[0137] Controls display brightness.

[0138] Step 7: Status monitoring and feedback.

[0139] Continuously monitor the NTC temperature and system status and feed back to step 1 to form a closed-loop control.

[0140] In some embodiments, the specific process can be referred to the above embodiments and will not be described in detail here.

[0141] In summary, this application is expected to achieve the following technical effects through the combination and innovation of the above technical means:

[0142] Enhanced predictability and proactivity: The intelligent prediction module enables prediction of thermal issues before they actually occur or worsen, enabling a shift from passive response to proactive prevention, effectively avoiding instantaneous overheating and excessive heat accumulation.

[0143] Improve the accuracy and efficiency of thermal management: Based on multi-dimensional feature input and indirect perception of frequency-spatial thermal behavior, it can more accurately identify potential local hotspots, implement more refined temperature control strategies, and avoid unnecessary global performance sacrifices.

[0144] The graded response mechanism enables temperature control measures to match the actual thermal risk level, improving resource utilization efficiency and user experience.

[0145] Achieve low-overhead intelligence: The use of lightweight models, efficient computing methods (such as WMA and displacement operations), and model compression technology makes it possible to deploy on resource-limited display controllers, achieving the goal of low hardware overhead.

[0146] Greater robustness and adaptability: By learning from a large amount of data under different working conditions, the neural network model has the generalization ability to better adapt to changing display content, environmental factors and user habits, providing continuous and stable thermal management.

[0147] Improved user experience: Smoother and smarter brightness adjustment reduces sudden and large brightness changes caused by overheating, improving visual comfort.

[0148] Extending display life: Through more effective active thermal management, the stress of components working at high temperatures for a long time is reduced, which helps to extend the service life of the display and maintain long-term reliability.

[0149] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules or units is merely a logical functional division. In actual implementation, other division methods may be used, such as combining or integrating multiple units or components into another system, or ignoring or not implementing certain features.

[0150] If the integrated units in the above other embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0151] The above description is only an implementation method of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A dynamic thermal management system for display screens based on multi-feature fusion and intelligent prediction, characterized in that: The display screen dynamic thermal management system includes: a multi-dimensional feature perception and fusion module, configured to fuse multi-dimensional features to obtain a comprehensive state vector; wherein the multi-dimensional features include at least one of display load, ambient light data, historical display screen temperature data, historical brightness adjustment sequences, and current display screen temperature data; an intelligent thermal prediction and anomaly detection module, configured to input the integrated state vector into a neural network model so that the neural network model outputs a prediction result; and perform anomaly detection based on the prediction result to obtain an anomaly detection result; wherein the prediction result includes at least one of a predicted temperature trend of the display screen, a predicted hot spot location on the display screen and the temperature corresponding to the hot spot location, and a frequency-space thermal behavior analysis result; The adaptive temperature control strategy execution module is used to execute the corresponding temperature control strategy according to the prediction result and the abnormality detection result.

2. The display screen dynamic thermal management system according to claim 1, characterized in that: The neural network model includes a model input layer and a main network. The intelligent thermal prediction and anomaly detection module is also used to input the comprehensive state vector into the model input layer for format conversion to obtain input features, and input the input features into the main network to obtain the prediction results.

3. The display screen dynamic thermal management system according to claim 2, characterized in that: The main network includes: an encoder, a frequency domain cross-scale interaction layer, a decoder and a prediction head; the intelligent thermal prediction and anomaly detection module is further used to input the input features into the encoder for encoding and output features at different scales; and input the features at different scales into the frequency domain cross-scale interaction layer for fusion and output fused features; and input the fused features into the decoder for decoding and output decoded features; and input the decoded features into the prediction head for prediction and output the prediction results.

4. The display screen dynamic thermal management system according to claim 3, characterized in that: The encoder includes a frequency-space domain heat dissipation encoder, which is used to extract heat dissipation features from the space domain and the frequency domain.

5. The display screen dynamic thermal management system according to claim 3, characterized in that: The frequency domain cross-scale interaction layer utilizes embedding, multi-head attention, and cross-attention mechanisms to perform cross-scale feature fusion in the frequency domain.

6. The display screen dynamic thermal management system according to claim 3, characterized in that: The main network is trained using spatial loss and frequency loss.

7. The display screen dynamic thermal management system according to claim 1, characterized in that: The intelligent thermal prediction and anomaly detection module is further used to establish a dynamic baseline based on the prediction results and historical normal operation data; and perform anomaly detection based on the current display screen temperature data, the prediction results and the baseline to obtain anomaly detection results.

8. The display screen dynamic thermal management system according to claim 7, characterized in that: The intelligent thermal prediction and anomaly detection module is further used to calculate the feature mean based on the WMA, and perform anomaly detection in combination with the sensitivity factor and the upper and lower thresholds to obtain anomaly detection results; wherein the upper and lower thresholds can be dynamically adjusted.

9. The display screen dynamic thermal management system according to claim 1, characterized in that: The adaptive temperature control strategy execution module is further configured to obtain a thermal risk level according to the prediction result and the abnormality detection result, and execute a corresponding temperature control strategy according to the thermal risk level.

10. The display screen dynamic thermal management system according to claim 9, characterized in that: The thermal risk levels include a first risk level, a second risk level, a third risk level and a fourth risk level; wherein the first risk level is lower than the second risk level, the second risk level is lower than the third risk level, and the third risk level is lower than the fourth risk.

Images