LED lamp heat attenuation suppression method and device based on junction temperature prediction

Abstract

The invention discloses an LED lamp thermal attenuation suppression method and device based on junction temperature prediction, and belongs to the technical field of LED lamps, and the method comprises the steps: reading a lamp task signal of an LED lamp, and collecting the real-time working data and external environment data of the LED lamp; after the universal deep learning model is read and the characteristic data of the LED lamp is utilized to execute model adjustment of the universal deep learning model, real-time working data and external environment data are input into the junction temperature prediction model; establishing a first lamp control target, establishing a second lamp control target, and configuring a joint target function; and carrying out joint optimization on the lamp control parameters and the heat dissipation control parameters of the LED lamp, and carrying out LED lamp control according to a joint optimization result. The technical problems that in the prior art, due to the fact that hysteretic temperature monitoring and optical and thermal management control are separated from each other, LED thermal attenuation cannot be actively restrained, and the service life and reliability of a lamp are low are solved.

Description

Technical Field

[0001] This invention relates to the field of LED lighting technology, and more specifically to a method and apparatus for suppressing thermal decay of LED lighting fixtures based on junction temperature prediction. Background Technology

[0002] LED lighting fixtures have been widely used in general lighting and special fields due to their advantages such as high efficiency, energy saving, and long lifespan. However, the luminous efficacy and lifespan of LED light sources are closely related to the chip junction temperature. Long-term operation at high temperatures will lead to severe degradation, significantly shortening the effective lifespan of the lamp and increasing maintenance costs.

[0003] Some existing intelligent control solutions attempt to indirectly adjust power by monitoring case temperature or ambient temperature, but they cannot accurately reflect the core junction temperature change trend, resulting in insufficient real-time performance and accuracy in prediction and control. Furthermore, these methods typically treat luminaire performance control and heat dissipation control as independent or conflicting objectives, lacking a synergistic optimization mechanism. This may lead to excessive sacrifice of light output quality to ensure lifespan, or accelerated thermal decay to pursue brightness. Therefore, there is an urgent need for a method that can accurately predict junction temperature and, based on this, synergistically optimize task execution and thermal decay suppression. Summary of the Invention

[0004] This application provides a method and apparatus for suppressing the thermal decay of LED lamps based on junction temperature prediction, aiming to solve the technical problems in the prior art that rely on hysteresis temperature monitoring and the disconnect between optical and thermal management control, which leads to the inability to actively suppress LED thermal decay and low lamp life and reliability.

[0005] In view of the above problems, this application provides a method and apparatus for suppressing the thermal decay of LED lamps based on junction temperature prediction.

[0006] The first aspect disclosed in this application provides a method for suppressing thermal decay of LED lamps based on junction temperature prediction. This method includes: reading the lamp task signal of the LED lamp and simultaneously collecting real-time operating data and external environmental data of the LED lamp; after reading a general deep learning model, performing model adjustment of the general deep learning model using the feature data of the LED lamp, inputting the real-time operating data and the external environmental data into a junction temperature prediction model, and outputting a junction temperature change trend, wherein the junction temperature prediction model is a model-adjusted general deep learning model; establishing a first lamp control objective based on the lamp task signal, obtaining a current state evaluation of the LED lamp, performing lamp decay analysis based on the current state evaluation of junction temperature influence, establishing a second lamp control objective, configuring a joint objective function based on the first and second lamp control objectives; performing joint optimization of the lamp control parameters and heat dissipation control parameters of the LED lamp based on the joint objective function, establishing a joint optimization result, and controlling the LED lamp based on the joint optimization result.

[0007] Another aspect of this application discloses an LED lighting fixture thermal attenuation suppression device based on junction temperature prediction. The device includes: a real-time data acquisition unit for reading the LED lighting fixture's task signal and simultaneously acquiring real-time operating data and external environmental data of the LED lighting fixture; a trend output unit for, after reading a general deep learning model, performing model adjustment on the general deep learning model using the LED lighting fixture's feature data, inputting the real-time operating data and the external environmental data into a junction temperature prediction model, and outputting a junction temperature change trend, wherein the junction temperature prediction model is a model-adjusted general deep learning model; a lighting fixture attenuation analysis unit for establishing a first lighting fixture control objective based on the lighting fixture task signal, obtaining a current state evaluation of the LED lighting fixture, performing lighting fixture attenuation analysis based on the current state evaluation of the junction temperature, establishing a second lighting fixture control objective, and configuring a joint objective function based on the first and second lighting fixture control objectives; and a lighting fixture control unit for jointly optimizing the lighting fixture control parameters and heat dissipation control parameters of the LED lighting fixture according to the joint objective function, establishing a joint optimization result, and controlling the LED lighting fixture according to the joint optimization result.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: By employing a forward-looking analysis based on a junction temperature prediction model and using this model to drive the joint optimization of optical and thermal management parameters, the technical problem of relying on hysteresis temperature monitoring and the disconnect between optical and thermal management control in existing technologies has been solved, which leads to the inability to actively suppress LED thermal decay and low lamp life and reliability. This achieves the technical effect of upgrading thermal decay suppression from a passive response to active collaborative optimization, and significantly improving lamp life and reliability.

[0009] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0010] Figure 1 A flowchart illustrating the method for suppressing thermal decay of LED lamps based on junction temperature prediction is provided for embodiments of this application.

[0011] Figure 2 A schematic diagram of the structure of an LED lamp thermal decay suppression device based on junction temperature prediction is provided for the embodiments of this application.

[0012] Explanation of reference numerals in the attached figures: Real-time data acquisition unit 11, trend output unit 12, lamp attenuation analysis unit 13, lamp control unit 14. Detailed Implementation

[0013] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structure, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0014] The overall concept of the technical solution provided in this application is as follows: This application provides a method and apparatus for suppressing the thermal decay of LED lamps based on junction temperature prediction. By collecting real-time data on the lamp's operation and environment, and inputting this data into a customized junction temperature prediction model, the method proactively obtains temperature change trends. Then, combining specific task requirements with junction temperature lifetime analysis, a multi-objective joint optimization function is established. Finally, the optimal combination of control parameters is calculated through collaborative optimization and implemented, thereby achieving a fundamental shift from passive response to active suppression of thermal decay.

[0015] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0016] Example 1, as Figure 1 As shown in the embodiments of this application, a method for suppressing the thermal decay of LED lamps based on junction temperature prediction is provided. The method includes: Step S100: Read the LED lighting task signal and simultaneously collect the real-time working data of the LED lighting and external environment data.

[0017] Specifically, lighting task signals refer to the lighting instructions or target requirements that the LED luminaire needs to execute. These are typically functional goals set by the upper-level control system or the user, rather than the luminaire's own physical parameters. Examples include the required brightness level, color value, and dynamic change curve. Real-time operating data refers to the instantaneous measurements of key internal electrical and thermal parameters of the LED luminaire during operation. This data directly reflects the luminaire's current operating status. Examples include drive current, voltage across terminals, lamp board temperature, and the PN junction voltage of the light source chip. External environmental data refers to the physical conditions of the environment in which the LED luminaire is located. These parameters are not controlled by the luminaire itself but significantly affect its heat dissipation performance and actual operating status. Examples include ambient air temperature, humidity, and airflow velocity over the heat sink surface.

[0018] Specifically, the device's main control unit, such as a microprocessor (MCU) or system-on-a-chip (SoC), reads lighting task signals from the upper-level controller or user input via communication interfaces such as DMX512, DALI, PWM, or a custom UART / SPI / I2C protocol, and parses them into specific control targets, such as a target luminous flux of 1000 lumens. While receiving and parsing the task signals, the main control unit instructs its built-in analog-to-digital converter and external sensor interfaces to synchronously collect real-time operating data and external environmental data at a fixed sampling frequency. This includes acquiring real-time operating data through current sensors and voltage detection circuits, obtaining the temperature near the junction through thermistors mounted on the lamp board, and acquiring external environmental data through temperature and humidity sensors and anemometers. To achieve synchronization, a hardware timer triggers simultaneous sampling of all data channels, or each data point is timestamped with high precision to ensure that all data are aligned in the time dimension, laying the foundation for subsequently building an accurate junction temperature prediction model.

[0019] This step lays a high-quality data foundation for subsequent intelligent control. By acquiring task commands and internal and external status data with high precision and time synchronization, the prediction model error caused by data asynchrony is effectively eliminated. This allows the subsequent junction temperature prediction model to accurately depict the dynamic process of the lamp status changing with internal and external conditions under specific task-driven conditions, thus providing a crucial input guarantee for achieving accurate and forward-looking thermal attenuation suppression control.

[0020] Step S200: After reading the general deep learning model, the model is adjusted using the feature data of the LED lamps. The real-time working data and the external environment data are then input into the junction temperature prediction model, and the junction temperature change trend is output. The junction temperature prediction model is the general deep learning model after model adjustment.

[0021] Specifically, in this application, the general deep learning model refers to a neural network model pre-trained on a large, diverse dataset of LED luminaires, such as an LSTM (Long Short-Term Memory) network, a Transformer, or a CNN-LSTM hybrid model. It has learned the complex nonlinear mapping relationship between operating data, environmental data, and junction temperature changes, possessing general predictive capabilities, but is not tailored to a specific luminaire. It employs a time-series-based deep prediction structure, with inputs including real-time electrical data, thermal data, external environmental data, and inherent structural feature parameters of the LED luminaire. Specifically, time-series parameters are input to a temporal feature extraction network, while the static structural features of the luminaire are input to a multilayer perceptron feature layer. A feature fusion module then fuses the dynamic and static features to form a unified feature representation for junction temperature prediction. The training process standardizes long-term operating data from multiple models and structures of LED luminaires and generates training samples according to a fixed time window. The mean squared error loss function is used as the basic training objective, and smoothing and physical consistency constraints are introduced. The smoothing constraint suppresses high-frequency oscillations in the predicted junction temperature sequence, while the physical consistency constraint limits the predicted temperature rise to within the physically feasible range derived from the luminaire's thermal resistance network. The model is trained using the ADMA optimizer until the verification error converges, at which point the model is frozen, forming a general deep learning model for subsequent transfer and adjustment. Feature data refers to static parameters describing the inherent properties of the LED luminaire, rather than real-time changing data. These features determine the thermal and electrical characteristics of the luminaire. Examples include the model and quantity of LED chips, the specifications of the driver power supply, the material and size of the heat sink, and the number and material of the PCB board layers. In this application, the junction temperature prediction model specifically refers to a deep learning model optimized for the current LED luminaire after "model adjustment," serving as a concrete example of the general model.

[0022] Specifically, after startup, the device reads a pre-trained general deep learning model from memory. This model file is stored in the embedded device's Flash or SD card. It performs similarity analysis between the feature data of the LED lamps and the lamp features used during the training of the general model. Historical data from the lamps is then used as new training samples to adjust the general model, resulting in a junction temperature prediction model. Specifically, a lamp feature vector is constructed based on characteristic parameters such as the lamp's heatsink material, chip package thermal resistance, number of LEDs, and driver specifications. This vector is then compared with the set of training lamp feature vectors used to train the general deep learning model, and a similarity coefficient is obtained using cosine similarity or Euclidean distance inverse mapping. If the similarity coefficient is high, only the final prediction layer of the general model is fine-tuned; if the similarity coefficient is medium, the middle and later network layers of the general model are locally adjusted; if the similarity coefficient is low, the main temporal layers, except for the feature extraction layer, are significantly adjusted. Furthermore, a first additional adjustment constraint is established based on historical data volume information. When the historical data volume is small, the number of adjustable network layers is limited and a smaller learning rate is used; when the historical data volume is large, deeper parameters can be allowed to participate in the adjustment. A second additional adjustment constraint is established based on the freshness of historical data, assigning higher training weights to recently generated data and reducing the weights or removing older data. Finally, by compensating and fusing the similarity coefficient, the first additional adjustment constraint, and the second additional adjustment constraint, a comprehensive adjustment strategy for the network is formed. Fine-tuning training is then performed based on the weighted historical data to obtain a junction temperature prediction model specifically for the target lighting fixture.

[0023] In the real-time control loop, the synchronously collected real-time working data and external environmental data are input into the junction temperature prediction model to output the junction temperature change trend, providing a decision basis for subsequent optimization control.

[0024] This step enables precise and personalized junction temperature prediction. By employing a transfer learning strategy, the massive amount of data and computational costs required to train a model from scratch for each new luminaire are avoided. Furthermore, by adjusting the model for specific luminaires, the shortcomings of general models in predicting insufficient accuracy when faced with individual differences are overcome. This provides a highly reliable and forward-looking junction temperature change trend for subsequent optimization control, fundamentally improving the effectiveness and reliability of the thermal decay suppression strategy.

[0025] Step S300: Establish a first lighting control target based on the lighting task signal, obtain the current state evaluation of the LED lighting, perform lighting attenuation analysis on the junction temperature effect based on the current state evaluation, establish a second lighting control target, and configure a joint objective function based on the first lighting control target and the second lighting control target.

[0026] Specifically, the first luminaire control objective is a performance objective related to the core functions of the luminaire, directly derived from the luminaire's task signals. Examples include achieving a specified illuminance, a specific color temperature, or completing a smooth dimming curve. The second luminaire control objective is a control objective based on attenuation analysis results, with the core focus on maintaining the long-term health of the luminaire. It focuses on maximizing luminaire lifespan and maintaining performance.

[0027] Specifically, based on the received lighting task signals, a first lighting control objective is established with the perfect achievement of the task as its core. Simultaneously, the current state evaluation of the LED lighting fixture is obtained, including real-time data and future trends output by the junction temperature prediction model. Based on this evaluation, a lighting fixture attenuation analysis of the junction temperature effect is performed, thereby establishing a second lighting fixture control objective with the suppression of thermal attenuation as its core. The attenuation analysis of the junction temperature effect is based on the LED chip material characteristics, packaging structure, and solder joint thermal fatigue characteristics to establish an attenuation model. Specifically, the Arrhenius aging model can be used to characterize the thermally accelerated aging trend at the chip level, describing the light decay rate through the exponential relationship between the material activation energy and the actual junction temperature of the chip. Material thermal fatigue at the packaging level can be characterized using the Coffin-Manson model, assessing the degree of packaging fatigue accumulation based on the junction temperature change amplitude of the LED during task execution. Thermal cycling damage at the solder joint level is calculated based on the strain fatigue characteristics of the solder. By assigning weights to these three types of attenuation factors, a comprehensive lighting fixture attenuation index can be obtained. By combining the future junction temperature change trend output by the junction temperature prediction model, the future attenuation rate and junction temperature risk level corresponding to different control schemes are calculated in real time before task execution, and this serves as the basis for generating the second luminaire control target. This analysis process enables the system to predict the impact of future junction temperature on luminaire lifespan before control, achieving proactive avoidance of attenuation risks.

[0028] A joint objective function is configured based on the two objectives mentioned above. This joint objective function can be constructed as a multi-objective weighted combination including task achievement evaluation terms, junction temperature decay penalty terms, and energy consumption evaluation terms. To achieve adaptive weight configuration, the task deviation impact analysis results are subjected to limit deviation point identification to determine the range within which task execution deviation will cause a severe functional degradation, thus establishing the first transition identification point. Similarly, the junction temperature impact analysis results are subjected to limit influence point identification to determine the risk transition behavior when the predicted junction temperature approaches a safe threshold or exceeds the material's critical temperature, thus establishing the second transition identification point. Furthermore, the two transition identification points are mapped to taboo factors, which can be constructed using sigmoid or exponential functions, so that the corresponding weights rapidly increase when the system state approaches a transition point. Simultaneously, energy consumption impact compensation is performed based on task execution energy consumption and heat dissipation control energy consumption, ensuring that the weights reflect the importance of energy-saving requirements in different scenarios. Through the normalization of taboo factors, energy consumption compensation amounts, and original weights, the adaptive dynamic allocation of the three evaluation terms in the joint objective function is achieved, enabling the control strategy to achieve an optimal balance between task completion, junction temperature risk, and energy consumption.

[0029] This step represents a leap in control strategy from a single objective and passive response to a multi-objective and proactive optimization approach. By establishing two control objectives representing short-term task performance and long-term health lifetime respectively, and using a joint objective function for dynamic trade-offs, optimal decisions can be made under different working scenarios.

[0030] Step S400: Perform joint optimization of the LED lamp control parameters and heat dissipation control parameters according to the joint objective function, establish the joint optimization result, and perform LED lamp control according to the joint optimization result.

[0031] Specifically, lighting control parameters refer to the electrical parameters that directly affect the optical output of LED lights, such as drive current, voltage, and the duty cycle of the PWM signal. Thermal control parameters refer to the parameters that control the operation of the cooling system, such as fan speed and the current of the thermoelectric cooler (TEC).

[0032] Specifically, the lighting control parameters and heat dissipation control parameters together constitute a multi-dimensional solution space. An optimization algorithm, using a joint objective function as the fitness evaluation criterion, performs an intelligent search within this solution space to generate multiple sets of parameter candidate schemes. An established junction temperature prediction model is then used to evaluate the future junction temperature trend corresponding to each scheme. Through iterative comparison and updating, the joint optimization result is converged. The device's main controller then uses this result to control the LED lighting fixtures. In this process, the junction temperature prediction model enables the optimization to preview the long-term effects of different control strategies within seconds, avoiding high-risk trials on real lighting fixtures.

[0033] This step optimizes the optical and thermal management of the luminaire as a whole, breaking the traditional disconnect or even conflict between brightness and temperature control. It not only significantly suppresses thermal decay and extends luminaire lifespan but also maximizes energy efficiency while ensuring task completion.

[0034] Furthermore, configuring a joint objective function based on the first and second lighting control objectives includes: constructing a joint objective function, which includes a task achievement evaluation term, a junction temperature decay penalty term, and an energy consumption evaluation term; performing task parsing on the first lighting control objective to obtain the criticality objective of the task; using the criticality objective to perform an achievement deviation impact analysis on the first lighting control objective and establishing a deviation impact analysis result; performing a junction temperature lifetime impact analysis on the lighting fixture based on the second lighting control objective and establishing a junction temperature impact analysis result; and adaptively configuring the evaluation weights of the joint objective function based on the deviation impact analysis result and the junction temperature impact analysis result to configure the joint objective function.

[0035] Specifically, criticality targets refer to the core, non-negotiable performance indicators derived from the primary lighting control targets. For example, for ambulance siren lights, a specific flashing frequency might be a criticality target, while absolute brightness may allow for a certain range of variation.

[0036] Specifically, a joint objective function is constructed, comprising three main elements: task achievement evaluation term, junction temperature decay penalty term, and energy consumption evaluation term, as follows: ; in, J is a dimensionless cost function, representing the total cost or total price, and indicating the quality of the current decision-making scheme. The smaller the J value, the better the overall performance in terms of task achievement evaluation, junction temperature decay penalty, and energy consumption evaluation. , , Weights are used to balance the importance of different objective items. + + =1; For task achievement evaluation, the deviation between the actual output of the LED luminaire and the "luminaire task signal" is quantified, for example: The unit of brightness is candela (cd). This is a junction temperature degradation penalty term, reflecting the negative impact of the LED chip junction temperature on the device. When the predicted junction temperature is higher, closer to, or exceeds the safety threshold, The value will increase dramatically. For example: max The temperature unit is Kelvin (K). This refers to the energy consumption evaluation item, representing the total electrical energy consumed to achieve the current state, including LED driver energy consumption and heat dissipation system energy consumption, measured in joules (J). For , , Normalization is required. Specifically, Divide by the square of the task objective value, which transforms it into (1 L actual / L target ) 2 , or divided by a maximum permissible deviation threshold. represents the square of the relative error, which is dimensionless. Divide by the square of the safety threshold or the maximum allowable temperature rise of the material. This transforms into [max(0, (T)] j Tthreshold) / Tthreshold)] 2 , which represents the square of the relative degree of overheating, is dimensionless. Dividing by the baseline power consumption or rated power consumption, and converting it into a relative energy consumption ratio, is dimensionless.

[0037] Next, task analysis is performed on the first lamp control objective to obtain its criticality objective. Using this criticality objective, an achievement deviation impact analysis is conducted to establish a quantified deviation impact analysis result. For example, a 10% reduction in illumination distance would significantly affect driving safety, with an impact level of severe. In parallel, a junction temperature lifetime impact analysis is performed based on the second lamp control objective, establishing a quantified junction temperature impact analysis result. For example, predicted junction temperature indicates that if the current power is maintained, the lifetime will accelerate from 30,000 hours to 15,000 hours, with an impact level of high risk. Based on the two analysis results, the evaluation weights of the joint objective function are adaptively configured. For example, when the task deviation impact is severe, the weights are significantly increased. When the junction temperature effect is considered high-risk, it significantly increases. Energy consumption weight This is typically used as a balancing term. This process is performed in real time by a rule-based or fuzzy logic system embedded in the controller, thereby configuring the joint objective function.

[0038] This step transforms a simple weighted sum into a function with adaptively configurable weights based on real-time impact analysis. This allows the device to adjust the priority of the control strategy according to two dimensions: task importance and the degree of luminaire hazard. This enhances the device's adaptability and robustness in different application scenarios, achieving the fundamental goal of maximizing the suppression of thermal decay and extending luminaire lifespan while ensuring core functionality.

[0039] Furthermore, the evaluation weights of the joint objective function are adaptively configured based on the offset influence analysis results and the junction temperature influence analysis results, including: compensating for the impact of task execution energy consumption and heat dissipation control energy consumption of the energy consumption evaluation item according to the first lighting control target and the second lighting control target respectively; identifying the extreme offset point of the offset influence analysis results and establishing a first transition identification point; identifying the extreme influence point of the junction temperature influence analysis results and establishing a second transition identification point; constructing taboo factors for mapping using the first transition identification point and the second transition identification point respectively, and then using the energy consumption evaluation item and the taboo factor after influence compensation to complete the adaptive configuration of the evaluation weights.

[0040] Specifically, impact compensation refers to adjusting the weights based on the contribution or importance of task execution and heat dissipation control to the overall energy consumption of the device, rather than directly using the original energy consumption value. The critical offset point is the point identified in the analysis of task achievement deviation. Once the task execution effect deviates from the target by more than this point, its impact on device function or safety will undergo a qualitative change. The first transition identification point specifically refers to the critical point related to the task achievement deviation. For example, if the illuminance of a surgical shadowless lamp is below 300 Lux, it may affect the doctor's operation; therefore, 300 Lux is the first transition identification point. The limit influence point is the critical junction temperature value identified in the analysis of the impact on junction temperature lifetime. Once the junction temperature exceeds this point, the lifespan degradation rate of the lamp will accelerate dramatically. The second transition identification point specifically refers to the critical point related to the impact on junction temperature lifetime, corresponding to the maximum allowable junction temperature of the LED chip or the inflection point temperature of the material properties. The taboo factor is a concept derived from the taboo search optimization algorithm. Here, it is abstracted as a penalty coefficient. When the device state approaches or exceeds the transition identification point, i.e. the critical point, the corresponding taboo factor will increase sharply, thereby prohibiting or preventing the optimization algorithm from selecting those schemes that would lead to a dangerous state in the weight configuration.

[0041] Specifically, based on the first and second lighting control objectives, the energy consumption for task execution and heat dissipation control in the energy consumption evaluation item are compensated for their impact. That is, if the current task is critical, the cost weight of task execution energy consumption is appropriately reduced, allowing for increased energy consumption to complete the task; if the junction temperature risk is high, the cost weight of heat dissipation control energy consumption is also reduced, allowing for increased energy consumption to enhance heat dissipation. Next, the results of the offset impact analysis are used to identify the limit offset point, establishing the first transition identification point; simultaneously, the results of the junction temperature impact analysis are used to identify the limit influence point, establishing the second transition identification point. Taboo factors for mapping are constructed using these two transition identification points, typically employing an sigmoid function or exponential function, such that the closer the device state is to the critical point, the faster the taboo factor increases. Using the energy consumption evaluation item after impact compensation and these two taboo factors, the final adaptive configuration of the evaluation weights is completed. Specifically, the weight of the task achievement item is positively correlated with the taboo factor of task offset, the weight of the junction temperature penalty item is positively correlated with the taboo factor of junction temperature impact, and the weight of the energy consumption item is balanced based on these factors.

[0042] This step, by introducing transition identification points and taboo factors, transforms the adaptive weight configuration from a simple linear adjustment into a nonlinear intelligent response with early warning and constraint characteristics. This allows the device to keenly sense when the risk of task failure or hardware damage will qualitatively change, and to perform transitional adjustments to the weights near the critical point, thereby strongly guiding the optimization algorithm away from high-risk areas. This achieves proactive preventative protection for LED lighting fixtures.

[0043] Furthermore, based on the joint objective function, the joint optimization of the LED lamp's control parameters and heat dissipation control parameters is performed, and a joint optimization result is established. This includes: configuring a solution space based on the LED lamp's characteristic data; constructing an initial solution within the solution space using a ternary combination channel, wherein the ternary combination channel includes a historical solution selection sub-channel, a random solution generation sub-channel, and a balanced solution space compensation sub-channel; performing an adaptation evaluation of the initial solution using the joint objective function, and establishing an adaptation evaluation result; and performing an update iteration of the initial solution within the solution space using the adaptation evaluation result, and completing the joint optimization based on the update iteration result.

[0044] Specifically, the ternary combination channel is a strategy for generating initial candidate solution sets, incorporating three different solution generation methods to balance search diversity, convergence speed, and robustness. The historical solution selection sub-channel selects a subset of high-performance parameter combinations stored from the past operation of the LED luminaire as initial solutions. The random solution generation sub-channel generates a subset of parameter combinations completely randomly within the solution space. The balanced solution space compensation sub-channel aims to cover regions that might be overlooked by historical and random solutions. For example, it intentionally generates solutions at the center, boundaries, or known performance-sensitive areas of the solution space to ensure comprehensive search coverage.

[0045] Specifically, the solution space is configured based on the characteristic data of the LED lighting fixtures, such as the maximum allowable current and heat sink capacity, and the upper and lower limits of each control parameter are defined to delineate the search boundaries. A ternary combined channel strategy is adopted to construct the initial solution set, that is, the historical solution selection sub-channel, the random solution generation sub-channel, and the balanced solution space compensation sub-channel are run simultaneously, and the solutions generated by the three sub-channels are merged into an initial solution population with good diversity. The joint objective function is used to perform an adaptation evaluation on each initial solution, and an adaptation evaluation result containing each solution and its J value is established. The adaptation evaluation result is used to perform the update iteration of the initial solution in the solution space. Optimization algorithms, such as genetic algorithms or particle swarm optimization algorithms, are used to select excellent parent solutions based on the magnitude of the J value, and offspring solutions are generated through crossover, mutation, and other operations. The population is continuously evolved until the convergence condition is met, thereby completing the joint optimization based on the update iteration result and outputting the optimal combination of control parameters. Among them, the historical solution selection sub-channel is constructed by querying the recent optimal combination of control parameters stored in the local non-volatile memory of the lighting fixture, directly utilizing historical successful experience. The random solution generation subchannel uses a hardware pseudo-random number generator to randomly generate entirely new parameter combinations within the solution space boundary in a uniform distribution, for exploring unknown regions. The balanced solution space compensation subchannel employs experimental design methods such as Latin hypercube sampling to strategically generate uniformly distributed points within the solution space, ensuring that the search covers all key regions, especially blank areas that may be overlooked by historical and random solutions.

[0046] This step generates an initial solution using a ternary combined channel method, balancing the relationships between utilizing historical experience, exploring unknown regions, and ensuring search completeness, thus laying a solid foundation for subsequent iterative optimization. It ensures that the device can quickly and reliably find the globally optimal or near-optimal combination of control parameters under various operating conditions, thereby stably and efficiently achieving the core objective of thermal decay suppression, while significantly improving the algorithm's convergence speed and applicability.

[0047] Furthermore, the adaptation evaluation results are used to perform initial solution update iterations within the solution space, and joint optimization is completed based on the update iteration results, including: performing adaptation segmentation on the adaptation evaluation results, establishing fine-tuning strategies and random search strategies; using the fine-tuning strategies and random search strategies to perform update management of the corresponding initial solutions, and creating an update trajectory for each initial solution; generating global constraints based on the update trajectories, and performing iterative compensation management of the update iteration results through the global constraints.

[0048] Specifically, fit partitioning refers to dividing the initial solution population into groups of different quality levels based on fit evaluation results, so that different update strategies can be applied to solutions of different qualities. Fine-tuning strategy refers to the update method used for high-fit, high-quality solutions. The core of this strategy is to perform a fine-grained local search within a small range near the high-fit solution. Random search strategy is an update method used for low-fit, low-quality solutions or to maintain population diversity. The core of this strategy is to allow for relatively large random jumps within the solution space to explore new possible regions and avoid the algorithm getting trapped in local optima prematurely. Update trajectory refers to recording the path of parameter combinations and fitness changes for each solution with the number of iterations during the iteration process. It reflects the movement history of the solution in the solution space.

[0049] Specifically, the fitness evaluation results are segmented according to fitness, for example, the top 20% of solutions with the smallest J values ​​are classified as high-quality solutions, and the bottom 50% as low-quality solutions. Based on this, fine-tuning strategies are established for high-quality solutions, such as using gradient descent for small-step fine-tuning search, and random search strategies are established for low-quality solutions, such as using mutations with a large mutation rate. These two strategies are used to update and manage the corresponding initial solutions, generating a new generation of parameter combinations for each solution and creating an update trajectory for each initial solution, recording its historical position and fitness changes. Then, global constraints are generated based on the update trajectories of all solutions. For example, if statistical analysis reveals that the J value generally deteriorates when the driving current exceeds a certain threshold, this threshold is set as a global constraint. Iterative compensation management is performed using this global constraint. For example, when it is detected that the trend of most solutions in the current population points to a suspected local optimum, the constraint manager triggers a compensation mechanism, forcibly resetting some solutions to points far from that region, or temporarily increasing the weight of random search to ensure the globality of the search.

[0050] This step, through adaptive segmentation and dual-strategy updates, allows the algorithm to intelligently balance the core contradiction between local exploitation and global exploration, thus ensuring both convergence speed and global optimality. By generating global constraints through trajectory updates and performing iterative compensation management, the algorithm significantly improves robustness and search efficiency when facing complex and nonlinear problems, ultimately ensuring the quality and stability of the joint optimization results and providing a guarantee for the precise control of LED lighting fixtures.

[0051] Furthermore, the model adjustment of a general deep learning model is performed using the feature data of LED lamps, including: acquiring the training data of the general deep learning model and extracting the features of the training LED lamps; performing feature similarity analysis on the feature data of the LED lamps and the features of the training LED lamps to establish a similarity coefficient; using the similarity coefficient as an adjustment constraint, selecting historical data of the LED lamps to perform parameter adjustment of the general deep learning model to complete the model adjustment.

[0052] Specifically, feature similarity analysis is a quantitative comparison method used in this application to evaluate the overall similarity between the features of a specific LED luminaire and the features of all luminaires in the training data. Its purpose is to determine whether the current luminaire is a common type or a special type relative to the known empirical range of a general model. The similarity coefficient reflects the average feature matching degree between the current luminaire and the training luminaire group. A high coefficient, such as close to 1, indicates very high similarity; a low coefficient, such as close to 0, indicates significant differences.

[0053] Specifically, based on big data, metadata for training a general deep learning model is obtained, namely the statistical distribution of the training LED lighting features corresponding to the training data. Feature similarity analysis is performed between the current LED lighting feature data and the statistical distribution of the training features. Similarity coefficients are obtained by calculating algorithms such as Euclidean distance or cosine similarity. These similarity coefficients serve as adjustment constraints to guide subsequent fine-tuning. Specifically, a high similarity coefficient indicates that the general model is already familiar with this type of lighting, thus requiring only a small amount of historical data from that specific lighting to make minor parameter adjustments to the last few layers of the model, i.e., the task-specific layers. A low similarity coefficient indicates that the current lighting is more specialized, potentially requiring more historical data and adjustments to more layers of the model, or even a smaller learning rate to prevent catastrophic forgetting. This process is completed on edge computing devices using a transfer learning framework, ultimately generating a customized model optimized for the current lighting.

[0054] This step enables intelligent and refined model adaptation. It allows the device to automatically and dynamically allocate computing resources and select the optimal adjustment strategy based on the current lighting fixtures. This preserves the powerful generalization capabilities of the general model while achieving personalized adaptation with maximum efficiency and minimum cost, thus improving the practicality and feasibility of large-scale deployment of the method.

[0055] Furthermore, the parameter adjustment of a general deep learning model is performed using historical data of LED lighting fixtures, including: obtaining data volume information of the historical data and establishing a first additional adjustment constraint based on the data volume information; obtaining data timeliness information of the historical data and establishing a second additional adjustment constraint based on the data timeliness information; and after compensating the adjustment constraint using the first additional adjustment constraint and the second additional adjustment constraint, performing parameter adjustment of a general deep learning model based on the historical data of LED lighting fixtures.

[0056] Specifically, historical data volume information refers to the number of data points or the total duration of historical operating data for a specific LED lamp that can be used for model adjustments. The first additional adjustment constraint, determined by the data volume information, is a restriction or guiding strategy for the model adjustment process. Its core principle is: with a small data volume, adjustments should be made cautiously. Data timeliness information refers to the time difference between the point in time the historical data was generated and the current moment. It reflects the freshness of the data. For example, is it data generated just an hour ago, or data from a lamp from a year ago? The second additional adjustment constraint, determined by the data timeliness information, is another restriction or guiding strategy for the model adjustment process. Its core principle is: the older the data, the lower its reference value may be.

[0057] Specifically, the system acquires information on the amount of historical data available for fine-tuning and establishes a first additional adjustment constraint based on this information. Specifically, if the data volume is small, the model is constrained to use a smaller learning rate or even freeze more network layers during adjustment to prevent overfitting due to insufficient data; if the data volume is large, a normal or slightly larger learning rate is allowed for sufficient adjustment. Simultaneously, the system acquires information on the timeliness of this historical data to establish a second additional adjustment constraint, assigning different weights to data from different periods—more weight to recent data and less weight to earlier data—or directly filtering out overly old data. Subsequently, these two additional constraints are used to compensate for the core adjustment constraint determined by the similarity coefficient, forming a fine-tuning scheme. Based on the historical data of LED lighting fixtures, the parameters of a general deep learning model are adjusted. This is automatically completed by an embedded software algorithm module that integrates a lightweight machine learning library, such as TensorFlow Lite, with a fine-tuning interface.

[0058] This step elevates model tuning from a relatively static operation to a dynamic optimization process with self-awareness and adaptability. By introducing intelligent constraints based on data volume and data timeliness, the adjustment strategy can be dynamically and finely controlled according to available data, enhancing the robustness and practicality of the method.

[0059] Furthermore, LED lighting control based on the joint optimization results includes: establishing a target response curve using the joint optimization results; performing response verification monitoring of the LED lighting based on the target response curve and establishing a response verification monitoring deviation; and reporting a response anomaly warning based on the response verification monitoring deviation.

[0060] Specifically, response verification monitoring deviation refers to the difference between the actual response value and the expected value of the target response curve during the response verification monitoring process. For example, two seconds after the command is issued, the actual brightness is 10% lower than the target brightness curve. This is a quantified error value.

[0061] Specifically, after obtaining the joint optimization results, a dynamic target response curve is established using these results. This curve is generated by a control algorithm, such as a feedforward-feedback composite controller, and defines how parameters such as brightness and current should smoothly transition to the target value. Based on this target response curve, the response verification and monitoring of LED lamps are carried out. Data from the light sensor and temperature sensor are collected in real time through an ADC and compared with the target curve to establish a quantified response verification and monitoring deviation. An abnormal response warning is reported based on whether this deviation value exceeds a threshold. Specifically, the warning is uploaded to the cloud monitoring platform or local human-machine interface through a communication module, such as CAN bus, Zigbee, or 4G / 5G, prompting maintenance personnel to intervene.

[0062] This step, by transforming the optimization results into a measurable target response curve and monitoring it in real time, not only ensures that the optimization strategy is executed accurately, but also enables timely detection of performance degradation or potential faults in lighting fixtures by analyzing response verification monitoring deviations. This allows for early warning of abnormal responses at an early stage of the problem, transforming reactive maintenance after the fact into proactive predictive maintenance before the fact.

[0063] Furthermore, a response verification and monitoring deviation is established, including: creating a control-response record based on the joint optimization results and the response verification and monitoring deviation; and using the control-response record for subsequent control optimization management of LED lighting fixtures.

[0064] Specifically, a control-response log refers to a structured data record formed by digitizing a complete control process. It includes timestamps, joint optimization results of the inputs (i.e., control commands), actual monitored response verification deviations, and environmental conditions at the time.

[0065] Specifically, after each control loop is executed, a control-response record is created based on the joint optimization results and the resulting response verification and monitoring deviations. These records are stored in the luminaire's local non-volatile memory or uploaded to a cloud-based time-series database. Historical control-response records are used for subsequent control optimization management. For example, if big data analysis or online learning algorithms reveal that the actual junction temperature of a specific luminaire is consistently about 3°C ​​higher than the model's prediction when the ambient temperature exceeds 40°C, the junction temperature prediction model will be automatically calibrated to compensate for the deviation. Alternatively, if a certain control strategy is found to have slightly higher theoretical energy consumption but excellent brightness stability in actual operation, it can be marked as the preferred strategy for high-reliability scenarios. This knowledge learned from the data is fed back to optimize the weight configuration of the joint objective function, correct the boundary of the solution space, or directly participate in the next optimization as a high-quality historical solution, thereby continuously improving the system.

[0066] This step, through the creation and utilization of control-response logs, allows the device to learn from its long-term operational data, continuously calibrate the model to eliminate prediction biases, and optimize strategies to adapt to the luminaire's own performance degradation and unique characteristics.

[0067] In summary, the LED lamp thermal decay suppression method based on junction temperature prediction provided in this application has the following technical effects: 1. By predicting junction temperature, the risk of thermal degradation is proactively mitigated, and optical control and thermal management are deeply integrated for joint optimization. This changes the traditional approach where temperature control is lagging and disconnected from brightness control, achieving a leap from passive response to proactive and collaborative optimization, and laying a systematic foundation for efficiently suppressing thermal degradation.

[0068] 2. By constructing a joint function with multiple objectives and introducing adaptive weight configuration based on task criticality and thermal risk, the device can intelligently adjust the priority of ensuring performance and protecting lamps in different scenarios, thus realizing the context awareness and fine-tuning of the control strategy.

[0069] 3. By generating an initial solution through a ternary channel that integrates historical experience, random exploration, and equilibrium compensation, the relationship between the algorithm's convergence speed and global search capability is balanced, effectively preventing the optimization process from getting trapped in local optima too early, and laying the foundation for finding high-quality solutions quickly.

[0070] Example 2 is based on the same inventive concept as the LED lamp thermal decay suppression method based on junction temperature prediction in the previous examples, such as... Figure 2 As shown in the embodiment of this application, an LED lamp thermal attenuation suppression device based on junction temperature prediction is provided. The device includes: a real-time data acquisition unit 11, used to read the lamp task signal of the LED lamp and simultaneously acquire the real-time working data and external environment data of the LED lamp; a trend output unit 12, used to read the general deep learning model, perform model adjustment of the general deep learning model using the feature data of the LED lamp, input the real-time working data and the external environment data into the junction temperature prediction model, and output the junction temperature change trend, wherein the junction temperature prediction model is the general deep learning model after model adjustment; a lamp attenuation analysis unit 13, used to establish a first lamp control target based on the lamp task signal, obtain the current state evaluation of the LED lamp, perform lamp attenuation analysis of junction temperature influence based on the current state evaluation, establish a second lamp control target, and configure a joint objective function based on the first lamp control target and the second lamp control target; and a lamp control unit 14, used to perform joint optimization of the lamp control parameters and heat dissipation control parameters of the LED lamp based on the joint objective function, establish a joint optimization result, and perform LED lamp control based on the joint optimization result.

[0071] Furthermore, the luminaire attenuation analysis unit 13 is also used to perform the following steps: constructing a joint objective function, the joint objective function including a task achievement evaluation term, a junction temperature attenuation penalty term, and an energy consumption evaluation term; performing task parsing on the first luminaire control objective to obtain the criticality objective of the task; using the criticality objective to perform an achievement deviation impact analysis on the first luminaire control objective and establishing a deviation impact analysis result; performing a junction temperature lifetime impact analysis on the luminaire based on the second luminaire control objective and establishing a junction temperature impact analysis result; and adaptively configuring the evaluation weights of the joint objective function based on the deviation impact analysis result and the junction temperature impact analysis result to configure the joint objective function.

[0072] Furthermore, the lamp attenuation analysis unit 13 is also used to perform the following steps: perform impact compensation on the task execution energy consumption and heat dissipation control energy consumption of the energy consumption evaluation item according to the first lamp control target and the second lamp control target respectively; identify the extreme offset point of the offset impact analysis result and establish a first transition identification point; identify the extreme impact point of the junction temperature impact analysis result and establish a second transition identification point; construct the taboo factor of the mapping using the first transition identification point and the second transition identification point respectively, and complete the adaptive configuration of the evaluation weight using the impact-compensated energy consumption evaluation item and the taboo factor.

[0073] Furthermore, the lighting control unit 14 is also used to perform the following steps: configuring a solution space according to the characteristic data of the LED lighting fixture; constructing an initial solution within the solution space through a ternary combination channel, wherein the ternary combination channel includes a historical solution selection sub-channel, a random solution generation sub-channel, and a balanced solution space compensation sub-channel; performing an adaptation evaluation of the initial solution using the joint objective function, and establishing an adaptation evaluation result; performing an update iteration of the initial solution within the solution space using the adaptation evaluation result, and completing joint optimization based on the update iteration result.

[0074] Furthermore, the lighting control unit 14 is also used to perform the following steps: performing adaptation segmentation on the adaptation evaluation results, establishing a fine-tuning strategy and a random search strategy; using the fine-tuning strategy and the random search strategy to perform update management of the corresponding initial solutions, and creating an update trajectory for each initial solution; generating global constraints based on the update trajectory, and performing iterative compensation management of the update iteration results through the global constraints.

[0075] Furthermore, the trend output unit 12 is also used to perform the following steps: acquire training data of a general deep learning model and extract features of the training LED lamps; perform feature similarity analysis on the feature data of the LED lamps and the features of the training LED lamps to establish a similarity coefficient; use the similarity coefficient as an adjustment constraint, select historical data of the LED lamps to perform parameter adjustment of the general deep learning model, so as to complete the model adjustment.

[0076] Furthermore, the trend output unit 12 is also used to perform the following steps: obtain the data volume information of the historical data, and establish a first additional adjustment constraint based on the data volume information; obtain the data timeliness information of the historical data, and establish a second additional adjustment constraint based on the data timeliness information; after compensating the adjustment constraint with the first additional adjustment constraint and the second additional adjustment constraint, perform parameter adjustment of a general deep learning model based on the historical data of LED lamps.

[0077] Furthermore, the lighting control unit 14 is also used to perform the following steps: establishing a target response curve using the joint optimization results; performing response verification monitoring of the LED lighting fixture based on the target response curve, and establishing a response verification monitoring deviation; and reporting a response anomaly warning based on the response verification monitoring deviation.

[0078] Furthermore, the lighting control unit 14 is also used to perform the following steps: create a control-response record based on the joint optimization results and response verification monitoring deviation; and use the control-response record to perform subsequent control optimization management of the LED lighting fixture.

[0079] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for suppressing thermal decay of LED lamps based on junction temperature prediction, characterized in that, The method includes: Read the LED lighting task signals and simultaneously collect real-time operating data of the LED lighting fixtures and external environmental data; After reading the general deep learning model, the model is adjusted using the feature data of the LED lamps. The real-time working data and the external environment data are then input into the junction temperature prediction model, and the junction temperature change trend is output. The junction temperature prediction model is the general deep learning model after model adjustment. A first lighting control objective is established based on the lighting task signal, the current state evaluation of the LED lighting is obtained, the lighting attenuation analysis of junction temperature is performed based on the current state evaluation, a second lighting control objective is established, and a joint objective function is configured based on the first lighting control objective and the second lighting control objective. The joint objective function is used to jointly optimize the lighting control parameters and heat dissipation control parameters of the LED lighting fixture, establish the joint optimization result, and control the LED lighting fixture based on the joint optimization result.

2. The LED lamp thermal decay suppression method based on junction temperature prediction as described in claim 1, characterized in that, Configure a joint objective function based on the first lighting control objective and the second lighting control objective, including: Construct a joint objective function, which includes a task achievement evaluation term, a junction temperature decay penalty term, and an energy consumption evaluation term; Perform task analysis on the first lighting control target to obtain the criticality target of the task; Using the aforementioned criticality target, an offset impact analysis of the achievement of the first lighting control target is conducted, and the offset impact analysis results are established. Based on the second lamp control target, an analysis of the impact of junction temperature on lamp life is conducted, and the results of the junction temperature impact analysis are established. Based on the results of the offset influence analysis and the junction temperature influence analysis, the evaluation weights of the joint objective function are adaptively configured to configure the joint objective function.

3. The LED lamp thermal decay suppression method based on junction temperature prediction as described in claim 2, characterized in that, Based on the offset influence analysis results and the junction temperature influence analysis results, the evaluation weights of the joint objective function are adaptively configured, including: Based on the first lighting control target and the second lighting control target, the energy consumption of task execution and heat dissipation control are compensated for the impact of energy consumption evaluation items. The extreme offset point is identified based on the offset influence analysis results, and a first transition identification point is established; The results of the junction temperature influence analysis are used to identify the limit influence point and establish a second transition identification point; After constructing the taboo factors of the mapping using the first transition identification point and the second transition identification point respectively, the evaluation weight adaptive configuration is completed using the energy consumption evaluation item after impact compensation and the taboo factors.

4. The LED lamp thermal decay suppression method based on junction temperature prediction as described in claim 3, characterized in that, Based on the joint objective function, the lighting control parameters and heat dissipation control parameters of the LED lamp are jointly optimized, and the joint optimization result is established, including: Configure the solution space based on the characteristic data of LED lighting fixtures; An initial solution is constructed within the solution space through a ternary combination channel, which includes a historical solution selection sub-channel, a random solution generation sub-channel, and an equilibrium solution space compensation sub-channel. The initial solution is evaluated using the joint objective function, and the evaluation results are established. The initial solution is updated and iterated within the solution space using the adaptation evaluation results, and joint optimization is completed based on the update and iteration results.

5. The LED lamp thermal decay suppression method based on junction temperature prediction as described in claim 4, characterized in that, Using the fitness evaluation results, an initial solution is updated iteratively within the solution space, and joint optimization is completed based on the update iterative results, including: The adaptation evaluation results are used for adaptation segmentation, and fine-tuning and random search strategies are established. The fine-tuning strategy and random search strategy are used to perform update management of the corresponding initial solutions, and an update trajectory is created for each initial solution; Global constraints are generated based on the update trajectory, and iterative compensation management of the update iteration results is performed through the global constraints.

6. The method for suppressing thermal decay of LED lamps based on junction temperature prediction as described in claim 1, characterized in that, Model tuning of a general deep learning model is performed using feature data from LED lighting fixtures, including: Obtain training data for a general deep learning model and extract features from the trained LED lighting fixtures; A feature similarity analysis was performed on the feature data of LED lamps and the features of training LED lamps to establish a similarity coefficient; Using the similarity coefficient as an adjustment constraint, historical data of LED lighting fixtures are selected to perform parameter adjustments on a general deep learning model to complete the model adjustment.

7. The LED lamp thermal decay suppression method based on junction temperature prediction as described in claim 6, characterized in that, Historical data from LED lighting fixtures were selected to perform parameter tuning for a general deep learning model, including: Obtain the data volume information of the historical data, and establish a first additional adjustment constraint based on the data volume information; Obtain the data timeliness information of the historical data, and establish a second additional adjustment constraint based on the data timeliness information; After compensating for the adjustment constraints using the first and second additional adjustment constraints, the parameters of the general deep learning model are adjusted based on the historical data of the LED lamps.

8. The method for suppressing thermal decay of LED lamps based on junction temperature prediction as described in claim 1, characterized in that, LED lighting control is performed based on the joint optimization results, including: The target response curve is established using the joint optimization results; Based on the target response curve, the response verification and monitoring of LED lamps are carried out, and a response verification and monitoring deviation is established. The response verification monitoring deviation is used to report response anomaly warnings.

9. The method for suppressing thermal decay of LED lamps based on junction temperature prediction as described in claim 1, characterized in that, Establish response verification and monitoring deviations, including: Based on the joint optimization results and the response verification monitoring deviation, a control-response record is created. The control-response records are used for subsequent control optimization and management of LED lighting fixtures.

10. A thermal decay suppression device for LED lamps based on junction temperature prediction, characterized in that, The apparatus for performing the LED luminaire thermal decay suppression method based on junction temperature prediction as described in any one of claims 1 to 9, the apparatus comprising: The real-time data acquisition unit is used to read the lighting task signals of the LED lights and simultaneously acquire the real-time working data of the LED lights and external environmental data. The trend change output unit is used to read the general deep learning model, perform model adjustment of the general deep learning model using the feature data of the LED lamp, input the real-time working data and the external environment data into the junction temperature prediction model, and output the junction temperature change trend. The junction temperature prediction model is the general deep learning model after model adjustment. The lamp attenuation analysis unit is used to establish a first lamp control target based on the lamp task signal, obtain the current state evaluation of the LED lamp, perform lamp attenuation analysis on the junction temperature effect based on the current state evaluation, establish a second lamp control target, and configure a joint objective function based on the first lamp control target and the second lamp control target. The lighting control unit is used to jointly optimize the lighting control parameters and heat dissipation control parameters of the LED lighting fixture according to the joint objective function, establish the joint optimization result, and control the LED lighting fixture according to the joint optimization result.

Images