Intelligent power supply regulation and control method and system of LED based on intelligent algorithm

Abstract

The invention relates to the technical field of intelligent power supply regulation and control, and discloses an intelligent power supply regulation and control method and system for an LED based on an intelligent algorithm. The method comprises the following steps: collecting illumination state parameters of an LED and carrying out LED power demand analysis to obtain an optimal power prediction value; calculating the switching frequency of the PFC circuit and the PWM duty ratio of the DC-DC circuit according to the optimal power predicted value; adjusting a power factor correction circuit by using the switching frequency, and outputting a direct-current bus voltage; the DC bus voltage is input into the DC-DC circuit, the PWM duty ratio is used for adjusting the LED driving current, and an LED driving signal is generated, it is ensured that the circuit can maintain the zero-voltage switching state under different power requirements, switching loss is effectively reduced, and then the stability and consistency of LED lighting output are ensured.

Description

Technical Field

[0001] This invention relates to the field of intelligent power control, and in particular to an intelligent power control method and system for LED light-emitting diodes based on intelligent algorithms. Background Technology

[0002] Traditional LED light-emitting diodes (LEDs) cannot adaptively adjust to the actual optical conditions when faced with complex optical changes such as the curvature adjustment of the highlighting lens and the dynamic expansion of the astigmatic reflector in LED searchlight systems. The fixed control mode used leads to a mismatch between power demand and power output during the dynamic adjustment of optical components in the LED system. When the lens curvature and reflector position change frequently, it easily causes fluctuations in drive current and unstable light output, seriously affecting the lighting quality and energy efficiency of the LED searchlight system. Summary of the Invention

[0003] This invention provides an intelligent power control method and system for LED light-emitting diodes based on intelligent algorithms. This invention ensures that the circuit can maintain a zero-voltage switching state under different power requirements, effectively reducing switching losses, and thus ensuring the stability and consistency of LED light-emitting diode lighting output.

[0004] In a first aspect, the present invention provides an intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms, the intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms comprising: Collect the lighting status parameters of LED light-emitting diodes and perform LED power demand analysis to obtain the optimal power prediction value; The switching frequency of the PFC circuit and the PWM duty cycle of the DC-DC circuit are calculated based on the optimal power prediction value. The power factor correction circuit with the switching frequency regulation is used to output DC bus voltage; The DC bus voltage is input into the DC-DC circuit and the LED drive current is adjusted using the PWM duty cycle to generate an LED drive signal.

[0005] In conjunction with the first aspect, in a first implementation of the first aspect of the present invention, the step of collecting the lighting state parameters of the LED light-emitting diode and performing LED power demand analysis to obtain the optimal power prediction value includes: The curvature change of the light-lifting lens is detected by a laser displacement sensor array to obtain the lens bending arc, and the shrinkage displacement of the astigmatic reflective strip is detected by a Hall position sensor to obtain the reflective strip shrinkage ratio. The junction temperature change of the LED chip is detected by an NTC thermistor to obtain the chip junction temperature data, and the light output intensity of the LED is detected by a silicon photodiode to obtain the luminous flux data. The illumination status parameters are obtained by combining the lens curvature, the reflective strip shrinkage ratio, the chip junction temperature data, and the luminous flux data. Based on the lighting condition parameters, LED power demand analysis is performed to obtain the optimal power prediction value.

[0006] In conjunction with the first aspect, in a second implementation of the first aspect of the present invention, the step of performing LED power demand analysis based on the lighting state parameters to obtain the optimal power prediction value includes: The lens curvature, reflector shrinkage ratio, chip junction temperature data, and luminous flux data in the lighting state parameters are normalized and vectorized to obtain a standard input vector. The Euclidean distance is calculated based on the standard input vector and the preset radial basis function center vector, and the activation value of each radial basis function node is calculated using the Gaussian kernel function to obtain the activation vector; The activation vector and the weight matrix are multiplied to obtain the initial power prediction value and prediction confidence of the LED power demand. The optimal power prediction value is obtained by updating the radial basis function center vector and the weight matrix based on the initial power prediction value and the prediction confidence.

[0007] In conjunction with the first aspect, in a third implementation of the first aspect of the present invention, the step of updating the radial basis function center vector and the weight matrix based on the initial power prediction value and the prediction confidence to obtain the optimal power prediction value includes: The difference between the initial power prediction value and the actual LED power value is calculated and weighted by combining the prediction confidence value to obtain the weighted prediction error signal. Based on the weighted prediction error signal, the error gradient is calculated for each radial basis function node, and the weight gradient and center vector gradient are calculated using the backpropagation algorithm to obtain the parameter update gradient vector. The gradient vector of the parameter update is combined with the adaptive learning rate and momentum term to perform gradient descent update of the weight matrix, resulting in the updated weight matrix; The optimal position of the radial basis function center vector is recalculated based on the updated weight matrix, and the center vector is adjusted to obtain the optimal power prediction value.

[0008] In conjunction with the first aspect, in a fourth implementation of the first aspect of the present invention, the step of calculating the switching frequency of the PFC circuit and the PWM duty cycle of the DC-DC circuit based on the optimal power prediction value includes: The optimal power prediction value is compared with the rated power of the LED and then nonlinearly mapped to obtain the PFC power regulation coefficient. The switching frequency of the PFC circuit is obtained by multiplying the PFC power regulation coefficient with the reference switching frequency of the LED. The LED target current is calculated based on the optimal power prediction value and the difference is calculated with the nominal current. Then, a nonlinear transformation is performed through the hyperbolic tangent function to obtain the DC-DC current regulation amount. The DC-DC current adjustment is added to the reference duty cycle, and dynamic compensation is performed by combining the lens curvature and the reflector strip shrinkage ratio to obtain the PWM duty cycle of the DC-DC circuit.

[0009] In conjunction with the first aspect, in a fifth implementation of the first aspect of the present invention, the step of using the switching frequency-adjusted power factor correction circuit to output the DC bus voltage includes: The switching frequency of the PFC circuit is input to the digital PWM controller for frequency setting, and the SiC MOSFET switching transistor is driven to switch according to the switching frequency to obtain the PFC switching control signal. Based on the PFC switch control signal, the leakage current of the PFC circuit is detected by the current transformer, and the switching frequency is automatically adjusted based on the leakage current to obtain the adjusted switching frequency. The PFC inductor value is adjusted in real time according to the adjusted switching frequency to obtain adaptive inductor parameters that match the adjusted switching frequency. The DC bus voltage is obtained by performing PI control calculations on the output voltage of the PFC circuit with the adaptive inductor parameter configuration and the target bus voltage.

[0010] In conjunction with the first aspect, in a sixth implementation of the first aspect of the present invention, the step of detecting the leakage inductance current of the PFC circuit through a current transformer based on the PFC switch control signal, and automatically adjusting the switching frequency based on the leakage inductance current to obtain the adjusted switching frequency, includes: Based on the switching time of the PFC switch control signal, the leakage current in the PFC circuit is sampled and detected in real time by a current transformer to obtain the leakage current at the time of switch turn-off. Calculate the minimum current threshold for zero-voltage switching based on the output capacitance value, maximum drain-source voltage, and dead time; The leakage current at the moment the switch is turned off is compared with the minimum current threshold of the zero-voltage switch. When the leakage current is less than the minimum current threshold of the zero-voltage switch, the switching frequency of the PFC circuit is reduced to obtain the adjusted switching frequency.

[0011] In conjunction with the first aspect, in the seventh implementation of the first aspect of the present invention, the step of inputting the DC bus voltage into the DC-DC circuit and using the PWM duty cycle to adjust the LED drive current to generate an LED drive signal includes: The DC bus voltage is input into the DC-DC converter circuit, and a digital PWM generator is used to control the switching of the GaN HEMT transistor according to the PWM duty cycle to obtain the DC-DC switching control signal. The predicted LED current value for the next moment is calculated based on the DC-DC switch control signal, and the deviation between the predicted LED current value and the target LED current calculated based on the optimal power prediction value is calculated to obtain the current regulation deviation. Based on the current regulation deviation and the LED chip junction temperature data, the LED forward voltage is temperature compensated to obtain the compensated forward voltage. The PWM duty cycle fine-tuning amount is then recalculated using the compensated forward voltage to obtain the target duty cycle parameter. The DC-DC circuit configured with the target duty cycle parameter outputs LED current and luminous flux data for fuzzy control, and outputs LED driving signals.

[0012] In conjunction with the first aspect, in the eighth implementation of the first aspect of the present invention, the step of performing fuzzy control on the LED current and luminous flux data output by the DC-DC circuit configured with the target duty cycle parameter, and outputting an LED driving signal, includes: The target duty cycle parameter is used to drive the DC-DC circuit to output LED current, and the luminous flux data is detected by the photoelectric converter. The luminous flux data is subtracted from the target luminous flux determined according to the lens curvature and the shrinkage ratio of the reflector to obtain the luminous flux control deviation and its time change rate. The light flux control deviation and the time change rate are respectively input into the triangular membership function for fuzzification and quantization to obtain the membership degree corresponding to each fuzziness level; Logical operations are performed based on the membership degree to obtain the fuzzy inference result of the duty cycle fine-tuning. The fuzzy inference result of the duty cycle fine-tuning is defuzzified and accumulated with the target duty cycle parameter to output the LED driving signal.

[0013] Secondly, the present invention provides an intelligent power control system for LED light-emitting diodes based on intelligent algorithms, the intelligent power control system for LED light-emitting diodes based on intelligent algorithms comprising: The data acquisition module is used to collect the lighting status parameters of LED light-emitting diodes and perform LED power demand analysis to obtain the optimal power prediction value; The calculation module is used to calculate the switching frequency of the PFC circuit and the PWM duty cycle of the DC-DC circuit based on the optimal power prediction value. The output module is used to adjust the power factor correction circuit using the switching frequency to output the DC bus voltage; The adjustment module is used to input the DC bus voltage into the DC-DC circuit and use the PWM duty cycle to adjust the LED drive current to generate an LED drive signal.

[0014] The technical solution provided by this invention has the following beneficial effects. Attached Figure Description To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating the intelligent power control method for LED light-emitting diodes based on intelligent algorithms provided in this application embodiment; Figure 2 This is a schematic block diagram illustrating the structure of an intelligent power control system for LED light-emitting diodes based on intelligent algorithms, provided in an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the described order. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change based on the actual situation.

[0018] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0019] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0020] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features described herein can be combined with each other.

[0021] Please see Figure 1 , Figure 1 A flowchart illustrating the intelligent power control method for LED light-emitting diodes based on intelligent algorithms provided in this application embodiment is shown below. Figure 1 As shown in the embodiments of this application, the intelligent power control method for LED light-emitting diodes based on intelligent algorithms includes: Step S100: Collect the lighting status parameters of the LED light-emitting diodes and perform LED power demand analysis to obtain the optimal power prediction value; Specifically, a lighting state parameter sensing system is constructed, integrating multiple sensor components to collect and process key physical states during the operation of the LED light-emitting unit in parallel. A laser displacement sensor array is used to perform non-contact displacement scanning on the light-adjusting lens, detecting real-time changes in the curvature of the lens surface. The normalized lens curvature value is calculated by combining the spatial distribution information of the sensor array and the calibration model. Simultaneously, Hall position sensors are deployed along the movement path of the astigmatic reflector strip to record its linear displacement changes during dimming in real time, and these changes are converted into standardized shrinkage ratio data based on the structural travel range. To obtain internal thermal information of the LED, an NTC thermistor is embedded in the LED chip carrier area. Junction temperature change data is obtained through collaborative sampling with a high-precision analog-to-digital converter, and the data reliability is improved by combining temperature compensation calibration curves. The monitoring of light output intensity is accomplished by a photoelectric conversion path composed of a silicon photodiode and a transimpedance amplifier. The detection signal is linearly amplified and digitized within a certain bandwidth range to form stable and reliable luminous flux data. The lens curvature, reflector shrinkage ratio, LED chip junction temperature, and luminous flux are integrated into a unified state vector structure according to time series, serving as multi-dimensional input information to describe the instantaneous physical behavior of the LED lighting system. Based on the lighting state vector, an improved radial basis function neural network model is constructed as a power prediction engine. During the training phase, feature mapping is performed on the nonlinear photoelectric characteristics, and an online learning mechanism is used to update the weight parameters in real time during operation. This allows the system to predict the optimal power demand value of the LED system at the next moment, i.e., the optimal power prediction value, based on the current state.

[0022] Step S200: Calculate the switching frequency of the PFC circuit and the PWM duty cycle of the DC-DC circuit based on the optimal power prediction value; Specifically, the deviation of the current power demand from the system's nominal capacity is characterized by comparing the optimal power prediction value with the rated power value of the LED. This ratio is then used as an input variable in a sigmoid-type nonlinear function model for mapping, extracting a smooth and adjustable PFC power regulation coefficient. This PFC power regulation coefficient is multiplied by the reference switching frequency of the PFC circuit as a frequency modulation factor to obtain the real-time updated target switching frequency of the PFC circuit, enabling the PFC stage to maintain high power factor operation under different power loads. The target LED current is obtained by dividing the same optimal power prediction value by the current LED forward voltage value. The difference between the target LED current and the LED's nominal operating current is then calculated to quantify the current deviation between the current load demand and the standard operating point. This difference is then normalized and nonlinearly scaled using a hyperbolic tangent function to output a stable and continuous DC-DC current regulation. The DC-DC current regulation is added to the preset DC-DC reference duty cycle to construct the initial PWM control value. Based on this, the dynamic parameters of the optical system, namely the curvature of the light-emitting lens and the shrinkage ratio of the astigmatic reflector, are introduced. By calculating the influence of the lens curvature and the shrinkage ratio of the reflector on the changes in energy distribution and light output efficiency in the power transmission path, a feedforward dynamic compensation term is formed and synthesized with the initial duty cycle to obtain the DC-DC circuit PWM duty cycle that can adaptively match the dynamic power demand of the LED.

[0023] Step S300: Use the switching frequency to adjust the power factor correction circuit to output the DC bus voltage; Specifically, the target switching frequency of the PFC circuit is input into the digital PWM controller. The digital PWM controller sets the PWM period and duty cycle in real time, and drives the SiC MOSFET power switch to perform periodic on / off operations at the set frequency. This generates a stable and fast-responding PFC switching control signal for energy modulation of the main power circuit. To ensure that the switching frequency control has the ability to adapt to load changes, a current transformer embedded in the inductor circuit collects the leakage current signal of the PFC inductor. The leakage current is used as the key basis for determining whether the ZVS (Zero Voltage Switching) condition is met. When current lag or voltage non-zeroing is detected during the switching operation, the controller automatically makes a fine adjustment to the switching frequency according to the set threshold, generating an adjusted switching frequency that adapts to the actual operating state. Based on the adjusted switching frequency, the parameters of the PFC boost inductor are adjusted in real time by controlling the variable permeability core inductor structure. This ensures that the inductor always meets the continuous conduction mode (CCM) operating conditions at different frequencies and maintains the integrity of the inductor current waveform, resulting in adaptive inductor parameters that match the actual frequency. Based on the PFC boost circuit with adaptive inductor parameter configuration, a DC voltage signal is output. The output voltage is compared with the preset target DC bus voltage. The difference is introduced into the PI controller for error processing and adjustment integral to generate a voltage correction quantity. This is used to control the conduction behavior and frequency setting of the PFC in a closed loop, so that the DC bus voltage output by the PFC is stably maintained near the target value, the fluctuation range is controlled within ±1%, and the power factor is maintained above 0.997.

[0024] Step S400: Input the DC bus voltage into the DC-DC circuit and use the PWM duty cycle to adjust the LED drive current to generate an LED drive signal.

[0025] Specifically, the stable DC bus voltage output from the PFC circuit is input to the DC-DC converter module. The DC-DC converter module employs a Buck topology and uses a high-speed GaN HEMT as the main power switching device. Simultaneously, a 16-bit high-precision digital PWM generator loads the PWM duty cycle parameters into the control signal generation unit in real time, driving the GaN HEMT switch to perform high-precision on / off operations at a fixed frequency. This generates a DC-DC switching control signal and achieves primary modulation of the output current. Based on the PWM parameters and bus voltage data within the current control cycle, and combined with the load characteristic model, the theoretical value of the LED output current at the next moment is predicted, forming the LED current prediction value. This predicted LED current value is then compared with the LED target current calculated based on the optimal power prediction value, and the deviation is extracted as the current regulation deviation. This deviation characterizes the degree of difference between the current control state and the expected power supply state. To improve system stability and accuracy, the control logic incorporates the junction temperature data of the LED chip. A linear temperature compensation model is used to correct the LED forward voltage drop, resulting in an effective forward voltage after dynamic temperature compensation. This voltage is then used as a reference to recalculate the duty cycle fine-tuning, correcting current mismatch caused by temperature drift and forming a target duty cycle parameter reflecting the current physical state. After being configured in the DC-DC circuit, the target duty cycle outputs a new drive current signal, which, together with synchronously acquired luminous flux data, forms a closed-loop optical output feedback. A fuzzy controller based on luminous flux error and its rate of change is introduced for dynamic adjustment, ensuring that the LED drive signal not only meets power prediction requirements at the electrical level but also maintains a stable and continuous light intensity response at the optical output level.

[0026] In one specific embodiment, the process of performing step S100 may specifically include the following steps: The curvature change of the light-lifting lens is detected by a laser displacement sensor array to obtain the lens bending arc, and the shrinkage displacement of the astigmatic reflective strip is detected by a Hall position sensor to obtain the reflective strip shrinkage ratio. The junction temperature change of the LED chip is detected by an NTC thermistor to obtain the chip junction temperature data, and the light output intensity of the LED is detected by a silicon photodiode to obtain the luminous flux data. The illumination status parameters are obtained by combining the lens curvature, reflector shrinkage ratio, chip junction temperature data, and luminous flux data. LED power demand analysis is performed based on lighting condition parameters to obtain the optimal power prediction value.

[0027] Specifically, a lighting state sensing system is constructed. A laser displacement sensor array is deployed across multiple key curvature control areas above the lighting lens. The array scans the lens surface at a frequency of once per millisecond, and, combined with the calibration model of the optical components, converts the collected height change data into a two-dimensional curvature distribution map. A fitting algorithm then extracts feature values ​​representing the overall deformation state of the lens, i.e., the normalized lens curvature arc value, reflecting the dynamic adjustment state of the lens during stress deformation or structural light adjustment. Simultaneously, a Hall position sensor detects the linear contraction behavior of the astigmatic reflector strip. The Hall position sensor is fixedly installed on the sliding trajectory of the reflector strip, and the relationship between magnetic field strength and displacement is used to detect the linear contraction behavior. The movement range of the reflector strip is obtained through numerical relationships, and a normalized shrinkage ratio is calculated based on the ratio of the original length to the current shrinkage, thus describing the influence of the reflector mechanism on the beam distribution. An NTC thermistor is integrated in the vicinity of the LED chip substrate; its resistance change has a non-linear relationship with the junction temperature. Temperature values ​​are sampled and digitally converted by linking with a high-resolution ADC module, and the actual chip junction temperature is obtained by combining it with a temperature calibration curve, forming thermal response data. The photoelectric conversion path composed of a silicon photodiode and a transimpedance amplifier is used to monitor the LED light output intensity. The collected electrical signal, after amplification, becomes the luminous flux, reflecting the current luminous efficacy output level of the LED. The lens curvature, reflector strip shrinkage ratio, chip junction temperature, and luminous flux are combined to form a unified state vector, logically representing the global physical state of the LED light-emitting system at a certain instant. The state vector is input into a nonlinear neural network model, which is built on an improved radial basis function neural network. The model performs nonlinear mapping and feature enhancement on the input state space through a Gaussian kernel function, and uses historical power behavior data for offline pre-training to form a stable hidden layer response pattern. During system operation, the radial basis function neural network performs forward propagation calculation based on the state vector collected each time, and outputs the optimal power prediction value for the next moment, which represents the target input power required by the LED system to maintain the best luminous efficiency and thermal stability under the combined effects of the current optical structure configuration, chip thermal state and light output level.

[0028] The process involves using a laser displacement sensor array to detect the curvature change of the lens, obtaining the lens bending arc, and using a Hall position sensor to detect the contraction displacement of the astigmatic reflector strip, obtaining the reflector strip contraction ratio. This includes: arranging multiple laser displacement sensors in a circular array around the lens, with each sensor independently detecting the change in the radius of curvature of the lens surface; fusing the multi-point measurement data into an overall lens bending arc parameter using a least squares fitting algorithm, obtaining spatially distributed lens curvature detection data; using a Kalman filter algorithm to perform time-series processing on the lens curvature detection data, establishing a state-space model of the lens bending motion; filtering out sensor noise and mechanical vibration interference through prediction and correction mechanisms to obtain high-precision lens bending arc; and configuring multiple Hall position sensors distributed... At different locations on the astigmatic reflector strip, the displacement of each part of the reflector strip is detected in real time. The overall shrinkage degree of the reflector strip is calculated using a weighted average algorithm, and a digital filter is used to eliminate the influence of electromagnetic interference on the Hall sensor, resulting in stable reflector strip shrinkage ratio data. The lens curvature and reflector strip shrinkage ratio data are input into a data fusion processor for cross-validation. When the trends of the two optical parameters are inconsistent, a sensor fault diagnosis program is initiated. The missing information is reconstructed using redundant sensor data to obtain optical state parameters with enhanced reliability. Based on the optical state parameters, a motion trajectory prediction model of the LED optical system is established to predict the next motion state of the lens and reflector strip in advance, providing forward-looking information for power demand prediction and obtaining multi-dimensional optical state data with predictive capabilities.

[0029] In one specific embodiment, the process of performing LED power demand analysis based on lighting state parameters to obtain the optimal power prediction value can specifically include the following steps: Normalize and vectorize the lens curvature, reflector shrinkage ratio, chip junction temperature data, and luminous flux data in the lighting status parameters to obtain a standard input vector; The Euclidean distance is calculated based on the standard input vector and the preset radial basis function center vector, and the activation value of each radial basis function node is calculated using the Gaussian kernel function to obtain the activation vector; The activation vector and the weight matrix are multiplied to obtain the initial power prediction value and prediction confidence of the LED power demand. The optimal power prediction value is obtained by updating the radial basis function center vector and weight matrix based on the initial power prediction value and prediction confidence.

[0030] Specifically, the lens curvature, reflector strip shrinkage ratio, chip junction temperature, and luminous flux are linearly transformed according to preset maximum and minimum ranges, reducing their values ​​to between zero and one. Four normalized results are used to construct a standard input vector, describing the current state characteristics of the LED light-emitting system in three physical dimensions: structural dimming, thermal stability, and light output intensity. The standard input vector is then used to calculate Euclidean distances with the center vectors of multiple radial basis functions configured in the neural network to characterize the similarity between the current state and typical states in the historical learning sample space. These Euclidean distances are then input into a Gaussian kernel function to generate nonlinear response values, resulting in the activation value of each node. The activation values ​​of all nodes are sequentially combined to form an activation vector, representing the response spectrum of the entire input state in the network's hidden layers. This activation vector is then input to the output layer and subjected to matrix operations with the weight matrix of the current training epoch to generate the initial power prediction value for the current control cycle, along with a corresponding prediction confidence index. The prediction confidence value is used to evaluate the model's learning coverage of the current input state and the reliability of the prediction results. Based on the deviation between the predicted value and the prior statistical reference data, and the level of confidence, it is determined whether to enter the parameter update stage. When the prediction confidence is low or the predicted value deviates significantly from historical experience, an online learning mechanism is triggered to jointly adjust the center position of the currently used radial basis function and the output weight parameters. During the center vector update process, a sample-driven center migration method is used to make the center distribution closer to the data subspace of the current state, thereby improving the local modeling capability in this region. During the weight update process, the model gradually approximates the true power output through error backpropagation and weight fine-tuning, making the model output results more accurate and stable. After the parameter update is completed, the model re-executes forward inference based on the adjusted structure and generates a corrected power prediction result as the optimal power prediction value at the current moment. The optimal power prediction value is then passed as the target control input to the PFC circuit and DC-DC converter module to guide the entire LED power supply system to perform adaptive energy scheduling in a dynamic environment.

[0031] The process involves calculating the Euclidean distance between the standard input vector and the preset radial basis function center vectors, and then using a Gaussian kernel function to calculate the activation value of each radial basis function node to obtain the activation vector. This includes: inputting the standard input vector into an adaptive K-means clustering algorithm for data distribution analysis; dynamically adjusting the number of cluster centers based on changes in data density and variance; automatically adding cluster centers when a new data pattern is detected; and merging adjacent cluster centers when data patterns overlap, resulting in an adaptively optimized set of radial basis function center vectors; calculating the four-dimensional Euclidean distance between each center vector in the radial basis function center vector set and the standard input vector; and adaptively adjusting the Gaussian kernel function based on the data coverage of each center vector. The width parameter is used to obtain the distance-weighted kernel function response value. This value is then sorted by activation intensity using a competitive learning mechanism. The top N nodes with the highest activation intensity are selected as effective activated nodes, while the remaining nodes are set to zero activation, resulting in a sparse activation vector. Based on the distribution characteristics of the sparse activation vector, the network complexity is evaluated using the information entropy criterion. When the information entropy exceeds a preset threshold, network pruning is triggered, deleting long-term inactive radial basis function nodes to obtain an optimized activation vector. The position and width parameters of the radial basis function center vector are continuously updated using an online learning algorithm, enabling the network structure to adapt to changes in the LED system state, resulting in an activation vector output with self-evolutionary capabilities.

[0032] In one specific embodiment, the process of updating the radial basis function center vector and weight matrix based on the initial power prediction value and prediction confidence to obtain the optimal power prediction value can specifically include the following steps: The difference between the initial power prediction value and the actual LED power value is calculated and weighted by combining the prediction confidence value to obtain the weighted prediction error signal. Based on the weighted prediction error signal, the error gradient is calculated for each radial basis function node, and the weight gradient and center vector gradient are calculated using the backpropagation algorithm to obtain the parameter update gradient vector. The parameter update gradient vector is combined with the adaptive learning rate and momentum term to perform gradient descent update of the weight matrix, resulting in the updated weight matrix. The optimal position of the radial basis function center vector is recalculated based on the updated weight matrix, and the center vector is adjusted to obtain the optimal power prediction value.

[0033] Specifically, the difference between the initial power prediction and the actual LED power is calculated to construct a basic signal representing the network output error. Simultaneously, prediction confidence is introduced as a reliable weighting factor. The error signal is weighted according to the concentration of the current prediction value in the radial basis function response distribution, forming a weighted prediction error signal. This weighted prediction error signal is propagated layer by layer within the model, and partial derivatives are calculated with respect to the activation values ​​of each radial basis function node to determine the gradient contribution of each node to the final output under the current error context, thus forming the error gradient information for the node. The error is then propagated to the output weight and center vector structural layers via backpropagation, generating gradient vectors corresponding to the current weight matrix and the center position of the radial basis function, respectively. These gradient vectors together constitute the directional index for parameter updates. To avoid oscillations or slow convergence during model training, a gradient vector and dynamically adjusted learning rate strategy are combined during parameter updates. Specifically, this involves exponentially decaying the learning rate based on the prediction error magnitude, introducing a momentum term to retain information from the previous gradient direction, enhancing the smoothness of weight adjustments, and then applying the weighted gradient to the current parameters to perform gradient descent, resulting in the updated weight matrix. Using the distribution relationship between the updated weight vector and the activation response, the geometric position of each radial basis function center vector in the input space is reassessed, and its position is corrected through an optimization strategy to bring it closer to regions with higher density representing the current input sample set, enhancing the model's expressive power and fitting accuracy in those regions. The dynamic adjustment of the center vector improves the model's local approximation ability and compresses prediction bias in the overall structure, improving response consistency across the entire input space. After the joint correction of the two structural layers—weight update and center position optimization—the neural network performs another forward propagation operation. The output generated by the updated parameter system at this point is the optimal power prediction value for the current control cycle.

[0034] In one specific embodiment, the process of performing step S200 may specifically include the following steps: The optimal power prediction value is compared with the rated power of the LED and nonlinearly mapped to obtain the PFC power regulation coefficient. The switching frequency of the PFC circuit is obtained by multiplying the PFC power regulation coefficient with the reference switching frequency of the LED. The target LED current is calculated based on the optimal power prediction value and the difference is calculated with the nominal current. Then, a nonlinear transformation is performed through the hyperbolic tangent function to obtain the DC-DC current regulation amount. The PWM duty cycle of the DC-DC circuit is obtained by adding the DC-DC current regulation to the reference duty cycle and then dynamically compensating for the lens curvature and reflector shrinkage ratio.

[0035] Specifically, the optimal power prediction value output by the RBF neural network is used as the basic control parameter and compared with the rated power of the LED in the design specifications to obtain the relative quantitative relationship between the current power demand and the rated operating capacity, reflecting the instantaneous power pressure of the LED load under a specific environmental condition. To enhance the flexible adjustment characteristics of the control curve, a nonlinear mapping function is introduced to process the above ratio. A nonlinear function with smooth boundaries and differentiability is selected so that the ratio input tends to have a linear response near the rated point, while saturation compression occurs in the high load or low load range far from the rated point. This suppresses instantaneous disturbances of over-adjustment and obtains a smooth output. The mapping result is the PFC power regulation coefficient. The PFC power regulation coefficient is used as a weighting factor and multiplied with the preset reference switching frequency of the PFC circuit to generate a real-time PFC switching frequency control quantity. This dynamically adjusts the operating rhythm of the front-end power factor correction circuit, increasing its operating frequency to enhance power supply response capability during high power demand, and appropriately reducing the frequency during low power stages to reduce switching losses and improve overall energy efficiency. Simultaneously, to achieve synchronous response of the subsequent DC-DC circuit, the required LED target current is calculated based on the optimal power prediction value. The LED target current is obtained by dividing the predicted power value by the real-time estimated value of the LED forward voltage, and the difference is calculated with the predefined LED nominal current to obtain the current drive correction requirement of the current system. The difference is used as the current deviation input to a nonlinear compression function for processing. A hyperbolic tangent function is used as the nonlinear mapping model to achieve high sensitivity response for small deviations and saturation for large deviations, avoiding overdrive. The nonlinear output is the DC-DC current regulation. The DC-DC current regulation is added to the DC-DC reference duty cycle to obtain the basic PWM duty cycle control value, which is used to adjust the on-time ratio of the GaN HEMT switch, thereby regulating the DC-DC output current. To improve control accuracy and adapt to the dynamic influence of optical structure on light flux distribution and energy transfer path, two structural state parameters, lens curvature and reflector shrinkage ratio, are introduced during duty cycle generation. By establishing the response mapping relationship between these two parameters and optical load disturbance, a feedforward compensation model is constructed. The influence of lens curvature on beam focusing degree and reflector deformation on scattering efficiency is quantified as a dynamic adjustment increment, which is superimposed on the basic duty cycle as a dynamic compensation factor for optical structure, generating a target PWM duty cycle with dynamic structural response capability.

[0036] In one specific embodiment, the process of performing step S300 may specifically include the following steps: The switching frequency of the PFC circuit is input to the digital PWM controller for frequency setting, and the SiC MOSFET switching transistor is driven to switch according to the switching frequency to obtain the PFC switching control signal. The leakage current of the PFC circuit is detected by a current transformer based on the PFC switch control signal, and the switching frequency is automatically adjusted based on the leakage current to obtain the adjusted switching frequency. The PFC inductor value is adjusted in real time according to the adjusted switching frequency to obtain adaptive inductor parameters that match the adjusted switching frequency. The DC bus voltage is obtained by performing PI control calculations on the output voltage of the PFC circuit with adaptive inductor parameter configuration and the target bus voltage.

[0037] Specifically, the target PFC switching frequency value is input into a high-resolution digital PWM controller. The PWM controller adjusts the internal clock division coefficient and trigger logic in real time according to the target PFC switching frequency setting to generate a drive pulse sequence consistent with the target frequency. It directly controls the gate level of the SiC MOSFET switch through a logic gate array to realize the switching control of the switch at the specified frequency. The output switching pulse signal is the PFC switching control signal. The PFC switching control signal serves as the main driving source in the entire power factor correction topology, controlling the period and time ratio of energy storage and release of the Boost inductor. As the PFC control loop operates, the leakage inductance current becomes a crucial parameter for determining whether the zero-voltage switching condition is met and for detecting the dynamic load response of the system. The leakage inductance current waveform is acquired in real time by a current transformer placed in the Boost inductor circuit. After conditioning, amplification, and analog-to-digital conversion, the leakage inductance current signal enters the digital controller and is compared with the ZVS critical current threshold set at the current frequency. If the leakage inductance current fails to reach the required level at the moment of switch closure, it indicates that the current switching frequency is too high or the inductor has not released enough energy. To prevent turn-on losses and voltage spikes, the frequency adaptive adjustment mechanism is activated. Based on the ZVS judgment logic, the current frequency is slightly corrected, the set value is lowered or raised, and the signal is reloaded into the PWM controller to generate a new adjusted switching frequency that meets the soft-switching condition. Meanwhile, to maintain the continuity of the inductor current waveform and adapt to the energy storage efficiency at different frequencies, the adjusted switching frequency is introduced as an input variable into the inductor adjustment function. This function controls the activation combination of adjustable permeability cores or parallel inductor arrays, thereby dynamically adjusting the equivalent inductance value of the PFC boost inductor to match the current frequency. This ensures the boost inductor still has sufficient energy accumulation capability at high frequencies and avoids response lag caused by excessively large inductance values ​​at low frequencies, resulting in adaptive inductance parameters adapted to the adjusted frequency. Based on this adaptive inductor configuration, the PFC boost circuit continuously calculates the difference between the real-time output voltage and the preset target bus voltage during the process of stabilizing the DC voltage output. This difference is input as an error signal to the PI controller. The PI controller responds quickly to instantaneous errors through the proportional gain term and simultaneously accumulates and corrects long-term deviations through the integral term. The output adjustment is then applied to the switching frequency setting path or the PWM duty cycle fine-tuning path, forming a closed-loop regulation mechanism that ensures the DC bus voltage remains stably around 400V under dynamic load changes.

[0038] The process involves inputting the switching frequency of the PFC circuit into a digital PWM controller for frequency setting, and driving the SiC MOSFET switches to switch according to the switching frequency to obtain the PFC switching control signal. This includes: configuring the PFC circuit as a multi-phase interleaved parallel structure, using three independent power processing units operating in parallel. Each unit contains an independent SiC MOSFET switch and inductor. Current ripple cancellation is achieved through switching timing control with a 120-degree phase difference, resulting in a multi-phase PFC topology; precise phase control of the switching timing of the three power processing units based on the switching frequency, with the first unit switching directly according to the switching frequency, the second unit delayed by 120 degrees, and the third unit delayed by 240 degrees, achieving microsecond-level phase synchronization accuracy through a digital signal processor, resulting in a phase-interleaved multi-channel switching control signal; and driving the SiC MOSFET switches of each unit according to the multi-channel switching control signal. MOSFET switches coordinate switching actions, using phase interleaving to achieve continuous input current and triple the current ripple frequency. Simultaneously, a load current sharing control algorithm balances the current distribution across units, resulting in a low-ripple composite input current. The composite input current is monitored in real-time, and the current imbalance between units is calculated. When the imbalance exceeds 5%, a current sharing correction algorithm is activated, redistributing the current by fine-tuning the duty cycle of each unit, resulting in a current-corrected switching control signal. Multi-phase interleaving control reduces the flux change rate and core loss of individual inductors, decreasing inductor size and increasing power density. Furthermore, current phase difference reduces common-mode noise and electromagnetic interference, resulting in a high-efficiency, low-noise PFC switching control signal.

[0039] In one specific embodiment, the process of performing the steps of detecting the leakage inductance current of the PFC circuit through a current transformer based on the PFC switch control signal, and automatically adjusting the switching frequency based on the leakage inductance current to obtain the adjusted switching frequency can specifically include the following steps: Based on the switching timing of the PFC switch control signal, the leakage current in the PFC circuit is sampled and detected in real time by a current transformer to obtain the leakage current at the moment of switch turn-off. Calculate the minimum current threshold for zero-voltage switching based on the output capacitance value, maximum drain-source voltage, and dead time; The leakage current at the moment the switch is turned off is compared with the minimum current threshold of the zero-voltage switch. When the leakage current is less than the minimum current threshold of the zero-voltage switch, the switching frequency of the PFC circuit is reduced to obtain the adjusted switching frequency.

[0040] Specifically, a high-speed response current transformer is installed in the PFC boost circuit. The secondary output is connected to the analog-to-digital conversion channel through a signal conditioning module and synchronized with the PWM switching logic in the digital controller. By marking the falling edge of the PWM as the trigger reference point for the turn-off event, the leakage inductance current is sampled at a fixed point. The instantaneous current value in the inductor branch is captured at the moment the switch is turned off, and the instantaneous current value is cached as the effective turn-off leakage inductance current within the switching cycle, reflecting the magnitude of the residual electromagnetic energy in the circuit at the moment the switch is turned off. Based on the physical parameters and circuit configuration of the power devices in the PFC circuit, a soft-switching decision model is established. The output capacitor directly affects the gate voltage release rate; the maximum drain-source voltage reflects the switching transistor's voltage withstand capability and the voltage change amplitude during device turn-off; and the dead time defines the time window available for achieving zero voltage between the main switch's turn-on and the start of the next conduction cycle. These three factors are input into the decision model, and the minimum leakage inductance current threshold required to just complete capacitor discharge and reduce the switch's drain-source voltage to near zero under the current dead time and capacitor energy storage conditions is calculated using an equivalent energy release model. This minimum current threshold is used as a reference standard in the soft-switching control strategy, serving as a boundary condition for judging the current leakage inductance current. When the leakage inductance current sampled by the controller at the turn-off moment is lower than the minimum threshold, it indicates that the current energy release of the system is insufficient to fully discharge the gate capacitor voltage. In other words, the zero-voltage turn-on condition cannot be met under the current frequency and inductance conditions. Without adjustment, the turn-on action in the next cycle will occur when a relatively high voltage still exists between the switch's drain and source, leading to turn-on losses, increased thermal stress, or voltage spikes. Therefore, the frequency reduction mechanism is immediately activated. Based on the margin between the current frequency value and the minimum supported frequency, and combined with the error magnitude, the frequency reduction range to be executed in the current cycle is calculated, and the adjusted switching frequency setting value is regenerated.

[0041] In one specific embodiment, the process of performing step S400 may specifically include the following steps: The DC bus voltage is input into the DC-DC converter circuit, and a digital PWM generator is used to control the switching of the GaNHEMT switch according to the PWM duty cycle to obtain the DC-DC switching control signal. The predicted LED current value for the next moment is calculated based on the DC-DC switch control signal, and the deviation between the predicted LED current value and the target LED current calculated based on the optimal power prediction value is calculated to obtain the current regulation deviation. The LED forward voltage is temperature compensated based on the current regulation deviation and the junction temperature data of the LED chip to obtain the compensated forward voltage. The PWM duty cycle fine-tuning is then recalculated using the compensated forward voltage to obtain the target duty cycle parameter. The DC-DC circuit with the target duty cycle parameter configured outputs LED current and luminous flux data for fuzzy control, and outputs LED driving signals.

[0042] Specifically, a stable DC bus voltage is input to the DC-DC converter circuit, which adopts a Buck topology and uses a GaN HEMT as the main power switching device to achieve high-frequency, high-efficiency energy conversion. At the control level, a high-resolution digital PWM generator generates PWM control pulses based on the duty cycle set in the current control cycle. The PWM control pulses directly drive the gate of the GaN HEMT device, causing it to complete power on / off modulation according to the set on / off sequence, thereby controlling the average amplitude of the output inductor current and forming a DC-DC switching control signal. Based on known quantities such as the PWM duty cycle, bus voltage, and output inductor parameters in the previous control cycle, a current dynamic prediction model is constructed to estimate the LED output current at the next moment, forming a predicted LED current value, representing the predicted current trend under the current control strategy. Simultaneously, the theoretically achievable LED target current is derived by dividing the current optimal power prediction value by the LED forward voltage drop. The difference between the predicted current value and the target current value is calculated to obtain the current regulation deviation. To improve current control accuracy and suppress output offset caused by temperature changes, the current junction temperature data of the LED chip is introduced into the control path. The forward voltage drop of the LED is dynamically corrected through the modeled temperature-voltage compensation relationship to obtain the compensated forward voltage value. The compensated voltage is then reused in the power-current calculation path to correct the duty cycle adjustment reference, compensate for the drive error caused by chip temperature rise or ambient temperature change, and recalculate the fine adjustment amount of PWM duty cycle based on this. A target duty cycle parameter that reflects the synergistic effect of temperature state, current deviation and target power is constructed. The DC-DC circuit re-outputs the updated LED drive current according to the new duty cycle setting, and introduces luminous flux data as feedback into the fuzzy control system. The fuzzy controller constructs a fuzzy rule set and performs fuzzy inference based on the error between the current output luminous flux of the LED and the set luminous flux, as well as the error change trend, to form an optical-level feedback correction control quantity. This feedback correction control quantity is then superimposed on the duty cycle adjustment path in a fine-tuning manner, thereby correcting and compensating the LED current adjustment result at the light output layer. This forms a four-dimensional collaborative control mechanism of voltage, current, temperature, and luminous flux, which can output LED drive signals under complex operating scenarios such as dynamic dimming, power fluctuations, temperature changes, or load disturbances.

[0043] The process involves inputting the DC bus voltage into a DC-DC converter circuit and using a digital PWM generator to control the switching of the GaN HEMT transistor according to the PWM duty cycle, thereby obtaining a DC-DC switching control signal. This includes: inputting the DC bus voltage into a non-isolated DC-DC converter containing a coupled inductor and a voltage multiplier; receiving the DC bus voltage through the primary winding of the coupled inductor and controlling the on / off state of the main switch transistor using the PWM duty cycle to obtain the primary winding magnetizing current of the coupled inductor; inducing a voltage in the secondary winding based on the primary winding magnetizing current of the coupled inductor, and establishing a soft-switching condition through the resonance between the leakage inductance of the coupled inductor and the voltage multiplier capacitor; when the main switch transistor is turned off, utilizing the energy stored in the leakage inductance to provide a zero-voltage switching environment for the switch transistor, thus obtaining a secondary induced voltage containing resonant characteristics; and inputting the secondary induced voltage into a converter... A voltage multiplier circuit composed of multi-stage capacitors and diodes is used for voltage boosting. By adjusting the number of voltage multiplier stages and the turns ratio of the coupled inductor in conjunction with the PWM duty cycle, three independent design degrees of freedom are formed to obtain a boost output voltage that meets the requirements of LED driving. Based on the amplitude and ripple characteristics of the boost output voltage, the leakage inductance parameters of the coupled inductor and the capacitance value of the voltage multiplier are optimized to ensure that all switching devices and rectifier diodes operate in a soft-switching state, resulting in a low-switching-loss DC-DC switching control signal. By recovering the energy stored in the leakage inductance of the coupled inductor and transferring it to the load, while using the voltage multiplier to reduce the voltage stress on the semiconductor devices and using low-rated GaN HEMT devices to reduce conduction losses, a high-efficiency LED driving output is obtained.

[0044] In one specific embodiment, the process of performing fuzzy control on the LED current and luminous flux data output by the DC-DC circuit configured with the target duty cycle parameter, and outputting the LED driving signal, can specifically include the following steps: The target duty cycle parameter is used to drive the DC-DC circuit to output LED current, and the luminous flux data is detected by the photoelectric converter. The luminous flux data is subtracted from the target luminous flux determined according to the lens curvature and the shrinkage ratio of the reflector to obtain the luminous flux control deviation and its time change rate. The luminous flux control deviation and the rate of change over time are respectively input into the triangular membership function for fuzzification and quantization to obtain the membership degree corresponding to each fuzziness level; Logical operations are performed based on membership degrees to obtain fuzzy inference results for duty cycle fine-tuning. The fuzzy inference result of the duty cycle fine-tuning is defuzzified and accumulated with the target duty cycle parameter to output the LED driving signal.

[0045] Specifically, the target duty cycle parameter is loaded into a high-resolution digital PWM generator, and its control output pulse width modulation signal drives a GaN HEMT switch. The target duty cycle is used as a control quantity to adjust the conduction time of the DC-DC circuit, thereby adjusting the output current amplitude and stabilizing the current output to the LED light-emitting module to achieve current-type drive control of the light source. During LED output, the luminous flux value emitted by the LED is detected in real time by a photoelectric converter set at the illumination optical axis or calibration position. The luminous flux data is input to the controller in digital form after being amplified by transimpedance and converted from analog to digital. The difference is calculated with the target luminous flux calculated based on the optical structure state parameters. The target luminous flux is determined by the lens curvature and reflector shrinkage ratio obtained in real time. These two structural state parameters characterize the current convergence and divergence state and light output characteristics of the LED beam shaping device. The expected light output value that should be achieved under the current geometric configuration is derived through an optical model, so that the target luminous flux has structural adaptability. The difference between the measured luminous flux and the target luminous flux is used as the luminous flux control deviation signal. The time rate of change of the deviation is calculated by the difference between two consecutive sampling periods, forming a dynamic control variable describing the trend of error change. The luminous flux control deviation and the time rate of change are respectively input into a triangular membership function. The triangular membership function presets multiple fuzzy levels (such as negative large, negative medium, negative small, zero, positive small, positive medium, positive large), and constructs a smooth transition structure between adjacent fuzzy intervals through trigonometric functions. This numerically maps the luminous flux deviation and the rate of change to each membership interval, and then calculates their membership degree at different fuzzy levels. The magnitude of the membership degree reflects the strength of the variable's current state in each level. The fuzzy inference process is executed according to the control rule base. The rule base pre-sets output response schemes under different membership combinations. For example, when the luminous flux deviation is negative and the rate of change is negative, it means that the system's light output is severely insufficient and still decreasing. The controller will infer that the duty cycle adjustment should be increased. When the deviation is zero but the rate of change is positive and small, it indicates that the system is close to the target light intensity but has a slight overshoot tendency. The controller will determine that the duty cycle adjustment should be appropriately reduced. The entire logical inference process uses "if-then" rule statements as the basic structure. It cross-matches and synthesizes the fuzzy level and magnitude of the output duty cycle fine-tuning amount based on the fuzzy state of the input variables. After completing the fuzzy inference, the fuzzy inference result is quantified by the defuzzification module, mapping the fuzzy level to a specific value. A weighted average is then performed based on the membership weights of each level to form a clear duty cycle fine-tuning output value. The duty cycle fine-tuning amount is accumulated with the original target duty cycle parameter to generate a new composite duty cycle control signal, which is immediately reloaded into the PWM controller to adjust the conduction time of the GaN switch in the next cycle, thereby realizing the fine-tuning closed-loop control of the LED output current.

[0046] Please see Figure 2 , Figure 2 A schematic block diagram of the structure of the intelligent power control system 200 for LED light-emitting diodes based on intelligent algorithms provided in this application embodiment is shown below. Figure 2 As shown, the intelligent power control system 200 for LED light-emitting diodes based on intelligent algorithms includes: The acquisition module 210 is used to acquire the lighting status parameters of LED light-emitting diodes and perform LED power demand analysis to obtain the optimal power prediction value; The calculation module 220 is used to calculate the switching frequency of the PFC circuit and the PWM duty cycle of the DC-DC circuit based on the optimal power prediction value. Output module 230 is used to adjust the power factor correction circuit using the switching frequency and output DC bus voltage; The adjustment module 240 is used to input the DC bus voltage into the DC-DC circuit and use the PWM duty cycle to adjust the LED drive current and generate an LED drive signal.

[0047] Through the synergistic cooperation of the aforementioned components, a four-dimensional state parameter system is constructed, incorporating lens curvature, reflector strip shrinkage ratio, chip junction temperature, and luminous flux. This system, combined with a radial basis function neural network, predicts LED power demand. Compared to traditional single-parameter feedback control, this system provides a comprehensive understanding of the LED's operating status. Based on the intelligently predicted power demand, the system simultaneously calculates the PFC circuit switching frequency and the DC-DC circuit PWM duty cycle. Coordinated parameter calculations ensure power transmission matching, improving dynamic response characteristics and control accuracy. Real-time leakage current detection and zero-voltage switching condition determination automatically adjust the PFC switching frequency and adaptively adjust inductor parameters, ensuring the circuit maintains a zero-voltage switching state under varying power demands, effectively reducing switching losses and improving reliability. Finally, a fuzzy control algorithm is used to optimize the LED drive signal. Through fuzzification of luminous flux deviation and inference rule matching, it accurately compensates for light output fluctuations caused by dynamic changes in the optical system, ensuring the stability and consistency of LED lighting output.

[0048] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0049] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, 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. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0050] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for intelligent power supply regulation of LED light-emitting diodes based on intelligent algorithms, characterized in that, include: Collect the lighting status parameters of LED light-emitting diodes and perform LED power demand analysis to obtain the optimal power prediction value; The switching frequency of the PFC circuit and the PWM duty cycle of the DC-DC circuit are calculated based on the optimal power prediction value. The power factor correction circuit with the switching frequency regulation is used to output DC bus voltage; The DC bus voltage is input into the DC-DC circuit and the LED drive current is adjusted using the PWM duty cycle to generate an LED drive signal.

2. The intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms according to claim 1, characterized in that, The process of collecting the lighting status parameters of LEDs and performing LED power demand analysis to obtain the optimal power prediction value includes: The curvature change of the light-lifting lens is detected by a laser displacement sensor array to obtain the lens bending arc, and the shrinkage displacement of the astigmatic reflective strip is detected by a Hall position sensor to obtain the reflective strip shrinkage ratio. The junction temperature change of the LED chip is detected by an NTC thermistor to obtain the chip junction temperature data, and the light output intensity of the LED is detected by a silicon photodiode to obtain the luminous flux data. The illumination status parameters are obtained by combining the lens curvature, the reflective strip shrinkage ratio, the chip junction temperature data, and the luminous flux data. Based on the lighting condition parameters, LED power demand analysis is performed to obtain the optimal power prediction value.

3. The intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms according to claim 2, characterized in that, The step of performing LED power demand analysis based on the lighting state parameters to obtain the optimal power prediction value includes: The lens curvature, reflector shrinkage ratio, chip junction temperature data, and luminous flux data in the lighting state parameters are normalized and vectorized to obtain a standard input vector. The Euclidean distance is calculated based on the standard input vector and the preset radial basis function center vector, and the activation value of each radial basis function node is calculated using the Gaussian kernel function to obtain the activation vector; The activation vector and the weight matrix are multiplied to obtain the initial power prediction value and prediction confidence of the LED power demand. The optimal power prediction value is obtained by updating the radial basis function center vector and the weight matrix based on the initial power prediction value and the prediction confidence.

4. The intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms according to claim 3, characterized in that, The step of updating the radial basis function center vector and the weight matrix based on the initial power prediction value and the prediction confidence to obtain the optimal power prediction value includes: The difference between the initial power prediction value and the actual LED power value is calculated and weighted by combining the prediction confidence value to obtain the weighted prediction error signal. Based on the weighted prediction error signal, the error gradient is calculated for each radial basis function node, and the weight gradient and center vector gradient are calculated using the backpropagation algorithm to obtain the parameter update gradient vector. The gradient vector of the parameter update is combined with the adaptive learning rate and momentum term to perform gradient descent update of the weight matrix, resulting in the updated weight matrix; The optimal position of the radial basis function center vector is recalculated based on the updated weight matrix, and the center vector is adjusted to obtain the optimal power prediction value.

5. The intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms according to claim 1, characterized in that, The calculation of the switching frequency of the PFC circuit and the PWM duty cycle of the DC-DC circuit based on the optimal power prediction value includes: The optimal power prediction value is compared with the rated power of the LED and then nonlinearly mapped to obtain the PFC power regulation coefficient. The switching frequency of the PFC circuit is obtained by multiplying the PFC power regulation coefficient with the reference switching frequency of the LED. The LED target current is calculated based on the optimal power prediction value and the difference is calculated with the nominal current. Then, a nonlinear transformation is performed through the hyperbolic tangent function to obtain the DC-DC current regulation amount. The DC-DC current adjustment is added to the reference duty cycle, and dynamic compensation is performed by combining the lens curvature and the reflector strip shrinkage ratio to obtain the PWM duty cycle of the DC-DC circuit.

6. The intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms according to claim 1, characterized in that, The power factor correction circuit using the switching frequency adjustment outputs a DC bus voltage, including: The switching frequency of the PFC circuit is input to the digital PWM controller for frequency setting, and the SiC MOSFET switching transistor is driven to switch according to the switching frequency to obtain the PFC switching control signal. Based on the PFC switch control signal, the leakage current of the PFC circuit is detected by the current transformer, and the switching frequency is automatically adjusted based on the leakage current to obtain the adjusted switching frequency. The PFC inductor value is adjusted in real time according to the adjusted switching frequency to obtain adaptive inductor parameters that match the adjusted switching frequency. The DC bus voltage is obtained by performing PI control calculations on the output voltage of the PFC circuit with the adaptive inductor parameter configuration and the target bus voltage.

7. The intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms according to claim 6, characterized in that, The step of detecting the leakage inductance current of the PFC circuit through a current transformer based on the PFC switch control signal, and automatically adjusting the switching frequency based on the leakage inductance current to obtain the adjusted switching frequency, includes: Based on the switching time of the PFC switch control signal, the leakage current in the PFC circuit is sampled and detected in real time by a current transformer to obtain the leakage current at the time of switch turn-off. Calculate the minimum current threshold for zero-voltage switching based on the output capacitance value, maximum drain-source voltage, and dead time; The leakage current at the moment the switch is turned off is compared with the minimum current threshold of the zero-voltage switch. When the leakage current is less than the minimum current threshold of the zero-voltage switch, the switching frequency of the PFC circuit is reduced to obtain the adjusted switching frequency.

8. The intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms according to claim 1, characterized in that, The step of inputting the DC bus voltage into the DC-DC circuit and using the PWM duty cycle to adjust the LED drive current to generate an LED drive signal includes: The DC bus voltage is input into the DC-DC converter circuit, and a digital PWM generator is used to control the switching of the GaN HEMT transistor according to the PWM duty cycle to obtain the DC-DC switching control signal. The predicted LED current value for the next moment is calculated based on the DC-DC switch control signal, and the deviation between the predicted LED current value and the target LED current calculated based on the optimal power prediction value is calculated to obtain the current regulation deviation. Based on the current regulation deviation and the LED chip junction temperature data, the LED forward voltage is temperature compensated to obtain the compensated forward voltage. The PWM duty cycle fine-tuning amount is then recalculated using the compensated forward voltage to obtain the target duty cycle parameter. The DC-DC circuit configured with the target duty cycle parameter outputs LED current and luminous flux data for fuzzy control, and outputs LED driving signals.

9. The intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms according to claim 8, characterized in that, The step of performing fuzzy control on the LED current and luminous flux data output by the DC-DC circuit configured with the target duty cycle parameter, and outputting an LED driving signal, includes: The target duty cycle parameter is used to drive the DC-DC circuit to output LED current, and the luminous flux data is detected by the photoelectric converter. The luminous flux data is subtracted from the target luminous flux determined according to the lens curvature and the shrinkage ratio of the reflector to obtain the luminous flux control deviation and its time change rate. The light flux control deviation and the time change rate are respectively input into the triangular membership function for fuzzification and quantization to obtain the membership degree corresponding to each fuzziness level; Logical operations are performed based on the membership degree to obtain the fuzzy inference result of the duty cycle fine-tuning. The fuzzy inference result of the duty cycle fine-tuning is defuzzified and accumulated with the target duty cycle parameter to output the LED driving signal.

10. An intelligent power control system for LED light-emitting diodes based on intelligent algorithms, characterized in that, The intelligent power supply regulation method for LED light-emitting diodes based on intelligent algorithms as described in any one of claims 1-9 includes: The data acquisition module is used to collect the lighting status parameters of LED light-emitting diodes and perform LED power demand analysis to obtain the optimal power prediction value; The calculation module is used to calculate the switching frequency of the PFC circuit and the PWM duty cycle of the DC-DC circuit based on the optimal power prediction value. The output module is used to adjust the power factor correction circuit using the switching frequency to output the DC bus voltage; The adjustment module is used to input the DC bus voltage into the DC-DC circuit and use the PWM duty cycle to adjust the LED drive current to generate an LED drive signal.

Images