Autonomous protection control method for abnormal feedback of LED switching power supply

Abstract

The invention discloses an autonomous protection control method for abnormal feedback of an LED switching power supply, and the method comprises the steps: obtaining parameter data in a switching period based on power supply operation data in a historical database; the method comprises the following steps: constructing a prediction model of the running state of the LED switching power supply, sampling based on a real-time feature window of the LED switching power supply, outputting a state prediction vector and abnormal probability distribution of a future switching period, constructing a real-time state feature matrix, dynamically matching the real-time state of the LED switching power supply with a prediction abnormal mode, and calculating a feature similarity index. And designing a hierarchical protection mechanism, uploading complete event data to a database, updating a power supply abnormity prediction model, and dynamically adjusting a protection trigger threshold value based on a deviation degree between a prediction result and an actual state. The method has the advantages that accurate pre-judgment and grading protection of the abnormal probability are achieved, model self-optimization and threshold value self-adaptive adjustment are driven through real-time data, and the system reliability and the protection response sensitivity are remarkably improved.

Description

Technical Field

[0001] The invention relates to an autonomous control technology, and in particular to an autonomous protection control method for abnormal feedback of an LED switch power supply. Background Art

[0002] LED switching power supplies are widely used in lighting, display and other fields. However, in actual applications, due to various reasons such as power supply design, load changes, environmental factors, etc., LED switching power supplies often face abnormal conditions such as overload, overvoltage, overtemperature, short circuit, etc. These abnormal conditions may not only cause damage to the LED power supply, but also affect the service life of the LED and the stability of the system. Therefore, how to improve the reliability and stability of the LED switching power supply has become a key issue in the design.

[0003] The current autonomous protection control methods on the market mainly include two categories: one is the overvoltage / overcurrent protection circuit based on a fixed threshold, which can only trigger shutdown when the abnormality reaches the preset threshold, and cannot predict potential risks in advance. The threshold setting relies on manual experience and is difficult to adapt to dynamic load changes; the second is the feedback control using a single sensor (such as temperature or current). Although it can achieve basic protection, it does not integrate multi-source feature data, and the diagnostic accuracy of complex abnormalities is low. In addition, the existing algorithms mostly use static models and lack the adaptive optimization capability driven by historical data, resulting in poor model generalization and easy failure when the power supply ages or environmental parameters change. In contrast, this method can accurately identify early abnormal patterns and quantify risk probabilities. Its closed-loop feedback mechanism further realizes the self-adjustment of protection thresholds and continuous iteration of model parameters. Summary of the invention

[0004] In order to improve the existing autonomous protection control method, an autonomous protection control method for abnormal feedback of LED switching power supply is provided. This method effectively enhances the autonomous protection performance of the power supply through real-time monitoring, intelligent prediction and adaptive adjustment, reduces the risk of failure, and improves the safety and stability of the power supply.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is: An autonomous protection control method for abnormal feedback of an LED switching power supply, comprising: Based on the power supply operation data in the historical database, the time domain voltage waveform, current spectrum characteristics, power device temperature change rate and PWM modulation parameters within the switching cycle are obtained; Build a prediction model for the operating status of LED switch power supplies and pre-train on the historical database using a transfer learning algorithm; Based on the real-time feature window sampling of the LED switching power supply, it is input into the prediction model to output the state prediction vector and abnormal probability distribution of the next switching cycle; Based on the prediction results, when the abnormal probability exceeds the dynamic threshold, a real-time state feature matrix is ​​constructed, the real-time state of the LED switch power supply is dynamically matched with the predicted abnormal pattern, and the feature similarity index is calculated; Based on the calculated similarity, if it exceeds the preset threshold for a long time, the hierarchical protection mechanism will be triggered; After the autonomous protection is executed, the complete event data is uploaded to the database, the power anomaly prediction model is updated, and the protection trigger threshold is dynamically adjusted based on the deviation between the prediction result and the actual state.

[0006] Preferably, the acquisition of the time-domain voltage waveform, current spectrum characteristics, power device temperature change rate and PWM modulation parameters within the switching cycle based on the power supply operation data in the historical database specifically includes: Based on synchronous trigger acquisition and digital filtering to process the power supply voltage signal, extract effective data and ripple coefficient, and obtain the time domain voltage waveform; Analyze the current spectrum based on FFT, calculate the harmonic distortion rate, identify the main harmonic components, and obtain the current spectrum characteristics; Synchronize temperature and switch status data, and obtain the temperature change rate of power devices based on differential calculation of temperature rise rate; Measure PWM drive signal parameters and analyze duty cycle characteristics to identify modulation strategy types.

[0007] Preferably, the construction of the prediction model of the power supply operation status and pre-training on the historical database by using the transfer learning algorithm specifically includes: Based on the LSTM+CNN hybrid structure, a source domain pre-training model is constructed. The input layer receives time series data, and the output layer predicts the remaining life. The reconstruction loss and prediction loss are jointly optimized through the loss function to learn the potential representation of the power supply operation status. Calculate the domain difference metric distance between the source domain and the target domain and map them to the common subspace, freeze the first three layers of the pre-trained model, and fine-tune the top-level parameters for adaptive training; The weight of the data in the historical database is updated based on the TrAdaBoost algorithm, the weight of the misclassified target domain samples is increased, and the influence of irrelevant source domain samples is reduced.

[0008] Preferably, the real-time feature window sampling based on the LED switching power supply, inputting it into the prediction model, and outputting the state prediction vector and abnormal probability distribution of the next switching cycle specifically include: Based on the complete switching cycle, set the real-time feature sampling window length; The input features are dynamically normalized, and the normalization parameters come from the statistical features of the previous cycles; The model output includes the predicted value of voltage fluctuation in the next cycle, the predicted current harmonic distribution, and the predicted temperature change rate; The abnormal probability distribution output is the conditional probability value of each fault mode, including four types of abnormalities: overvoltage, overcurrent, overheating, and short circuit.

[0009] Preferably, based on the prediction results, when the abnormal probability exceeds the dynamic threshold, a real-time state feature matrix is ​​constructed, the real-time state of the LED switch power supply is dynamically matched with the predicted abnormal mode, and the feature similarity index is calculated, specifically including: The initial threshold is obtained based on historical normal data statistics, and the threshold is adjusted dynamically. The initial threshold formula is: , The dynamic threshold adjustment formula is: , in, is the mean value of the characteristic data under normal conditions, is the standard deviation of the characteristics under normal conditions, representing the degree of dispersion of the data. is the forgetting factor, which controls the historical initial threshold weight, is the sliding window length, and the average abnormal probability of the past switching cycles is calculated. is the abnormal probability of the past k-th period; based on In principle, 99.7% of normal data falls within If it exceeds this range, it will be considered abnormal. Construct a real-time state feature matrix based on the feature data in the current switching cycle when an abnormality occurs; Acquire a set of characteristic matrix templates of pre-stored typical abnormal patterns based on a historical database; The real-time state feature matrix is ​​time-aligned with each abnormal template, and the influence of timing drift is eliminated based on dynamic time warping. The dynamic time warping distance formula is: , in, is the DTW distance between the real-time state feature matrix and the i-th abnormal template, is the total step length of the alignment path, that is, the length of the sequence after dynamic matching, is the jth matching point on the alignment path, is the distance function between feature vectors, is the feature vector of the real-time state feature matrix at a certain moment in the time window at path point j, is the feature vector of the i-th abnormal template at path point j; Based on the obtained dynamic time warping distance, a feature similarity index is obtained. The smaller the dynamic time warping distance, the higher the similarity index.

[0010] Preferably, if the similarity obtained based on the calculation exceeds a preset threshold for a long time, triggering the hierarchical protection mechanism specifically includes: Based on the obtained similarity, different levels of protection mechanisms are triggered; In the initial protection stage, when the threshold is exceeded in consecutive cycles, the PWM duty cycle is started for dynamic adjustment, and the adjustment range is: , in, is the adjustment amount of PWM duty cycle, that is, the PWM duty cycle that needs to be adjusted. is the proportional gain coefficient, which is used to amplify the current error. is the integral gain coefficient, which is used to amplify the accumulated error. is the expected value of the characteristic state data, is the actual value of the characteristic status data; In the intermediate protection stage, the power reduction mode is triggered and the rated power is reduced every cycle; In the advanced protection stage, the main power circuit is immediately cut off and the sound and light alarm is triggered.

[0011] Preferably, after the autonomous protection is executed, the complete event data is uploaded to the database, the power anomaly prediction model is updated, and the protection trigger threshold is dynamically adjusted based on the deviation between the prediction result and the actual state. Specifically, it includes: Regularly retrain the power anomaly prediction model based on newly uploaded data, and verify and evaluate the model after it is updated; The protection trigger threshold is dynamically adjusted based on the deviation between the predicted result and the actual state. The deviation calculation formula is: , in, is the predicted abnormal probability at the kth time point, is the actual abnormal probability of the kth cycle, is the number of samples in the statistical window; The threshold is adjusted dynamically based on the PID control concept, and the effect of the new threshold is simulated at the edge computing node. After confirmation, it is synchronized to all LED switch power supplies.

[0012] Compared with the prior art, the advantages of the present invention are: The prediction model based on transfer learning can effectively utilize the common features in historical data, quickly adapt to the dynamic characteristics under different working conditions, and overcome the lag of traditional threshold protection for unknown anomalies; through the fusion analysis of multi-source features such as time-domain voltage waveform, current spectrum, and temperature gradient, it realizes the accurate identification of complex anomalies such as power device aging and load mutation; the dynamic threshold adjustment mechanism combined with the hierarchical protection strategy can not only avoid false triggering or missed reporting caused by a single threshold, but also match the optimal protection action according to the severity of the anomaly, reducing unnecessary downtime; the closed-loop feedback mechanism continuously optimizes the model parameters through real-time data reflux, so that the system has the ability to evolve autonomously and maintain the stability of protection sensitivity throughout the life cycle of the power supply. A complete closed loop is formed from prediction, diagnosis to execution, which not only ensures the safety of the system, but also extends the service life of key components and reduces operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic diagram of the method proposed by the present invention; Figure 2 This is a schematic diagram of parameter collection proposed by the present invention; Figure 3 This is a schematic diagram of the model training proposed by the present invention; Figure 4 This is a schematic diagram of the state prediction proposed by the present invention; Figure 5 This is a schematic diagram of the dynamic matching proposed by the present invention; Figure 6 This is a schematic diagram of the hierarchical protection mechanism proposed by the present invention; Figure 7 This is a schematic diagram of the dynamic adjustment of the threshold value proposed by the present invention. DETAILED DESCRIPTION

[0014] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0015] See also Figure 1 As shown, an autonomous protection control method for abnormal feedback of LED switching power supply includes: Step 1: Based on the power supply operation data in the historical database, obtain the time domain voltage waveform, current spectrum characteristics, power device temperature change rate and PWM modulation parameters within the switching cycle; Step 2: Build a prediction model for the operating status of the LED switch power supply and pre-train it on the historical database through the transfer learning algorithm; Step 3: Based on the real-time feature window sampling of the LED switching power supply, it is input into the prediction model to output the state prediction vector and abnormal probability distribution of the next switching cycle; Step 4: Based on the prediction results, when the abnormal probability exceeds the dynamic threshold, a real-time state feature matrix is ​​constructed, the real-time state of the LED switch power supply is dynamically matched with the predicted abnormal pattern, and the feature similarity index is calculated; Step 5: Based on the calculated similarity, if it exceeds the preset threshold for a long time, the hierarchical protection mechanism is triggered; Step 6: After the autonomous protection is executed, the complete event data is uploaded to the database, the power anomaly prediction model is updated, and the protection trigger threshold is dynamically adjusted based on the deviation between the prediction result and the actual state.

[0016] See also Figure 2 As shown, based on the power supply operation data in the historical database, the time domain voltage waveform, current spectrum characteristics, power device temperature change rate and PWM modulation parameters within the switching cycle are obtained, including: Based on synchronous trigger acquisition and digital filtering to process the power supply voltage signal, extract effective data and ripple coefficient, and obtain the time domain voltage waveform; Analyze the current spectrum based on FFT, calculate the harmonic distortion rate, identify the main harmonic components, and obtain the current spectrum characteristics; Synchronize temperature and switch status data, and obtain the temperature change rate of power devices based on differential calculation of temperature rise rate; Measure PWM drive signal parameters and analyze duty cycle characteristics to identify modulation strategy types.

[0017] Specifically, during data analysis, trigger conditions are set to ensure synchronous acquisition of voltage signals and other related signals (such as current, temperature, etc.), and the voltage signal is filtered using a bandpass filter to calculate the ripple factor to reflect the voltage quality. The current signal is spectrally analyzed by fast Fourier transform (FFT), the main harmonic components are identified, and the harmonic distortion rate (THD) is calculated to evaluate the waveform quality of the current. The extraction of spectral features helps to understand the harmonic distribution in the system and its impact on the equipment.

[0018] See also Figure 3 As shown in the figure, a prediction model for power supply operation status is constructed, and pre-training is performed on the historical database through the transfer learning algorithm. Specifically, the following steps are performed: Based on the LSTM+CNN hybrid structure, a source domain pre-training model is constructed. The input layer receives time series data, and the output layer predicts the remaining life. The reconstruction loss and prediction loss are jointly optimized through the loss function to learn the potential representation of the power supply operation status. Calculate the domain difference metric distance between the source domain and the target domain and map them to the common subspace, freeze the first three layers of the pre-trained model, and fine-tune the top-level parameters for adaptive training; The weight of the data in the historical database is updated based on the TrAdaBoost algorithm, the weight of the misclassified target domain samples is increased, and the influence of irrelevant source domain samples is reduced.

[0019] Specifically, during the model training process, the parameters of the first three layers of the pre-trained model are frozen. These layers are usually responsible for extracting low-level features and maintaining their stability. Then, the parameters of the top layer are fine-tuned to achieve adaptive training, so that the model can better adapt to the data characteristics of the target domain. This method can enhance its adaptability to the target domain while retaining the generalization ability of the pre-trained model.

[0020] See also Figure 4 As shown in the figure, based on the real-time feature window sampling of the LED switching power supply, it is input into the prediction model, and the state prediction vector and abnormal probability distribution of the next switching cycle are output, which specifically include: Based on the complete switching cycle, set the real-time feature sampling window length; The input features are dynamically normalized, and the normalization parameters come from the statistical features of the previous cycles; The model output includes the predicted value of voltage fluctuation in the next cycle, the predicted current harmonic distribution, and the predicted temperature change rate; The abnormal probability distribution output is the conditional probability value of each fault mode, including four types of abnormalities: overvoltage, overcurrent, overheating, and short circuit.

[0021] See also Figure 5 As shown, based on the prediction results, when the abnormal probability exceeds the dynamic threshold, a real-time state feature matrix is ​​constructed, and the real-time state of the LED switch power supply is dynamically matched with the predicted abnormal mode. The feature similarity index is calculated, which specifically includes: The initial threshold is obtained based on historical normal data statistics, and the threshold is adjusted dynamically. The initial threshold formula is: , The dynamic threshold adjustment formula is: , in, is the mean value of the characteristic data under normal conditions, is the standard deviation of the characteristics under normal conditions, representing the degree of dispersion of the data. is the forgetting factor, which controls the historical initial threshold weight, is the sliding window length, and the average abnormal probability of the past switching cycles is calculated. is the abnormal probability of the past k-th period; based on In principle, 99.7% of normal data falls within If it exceeds this range, it will be considered abnormal. Construct a real-time state feature matrix based on the feature data in the current switching cycle when an abnormality occurs; Acquire a set of characteristic matrix templates of pre-stored typical abnormal patterns based on a historical database; The real-time state feature matrix is ​​time-aligned with each abnormal template, and the influence of timing drift is eliminated based on dynamic time warping. The dynamic time warping distance formula is: , in, is the DTW distance between the real-time state feature matrix and the i-th abnormal template, is the total step length of the alignment path, that is, the length of the sequence after dynamic matching, is the jth matching point on the alignment path, is the distance function between feature vectors, is the feature vector of the real-time state feature matrix at a certain moment in the time window at path point j, is the feature vector of the i-th abnormal template at path point j; Based on the obtained dynamic time warping distance, a feature similarity index is obtained. The smaller the dynamic time warping distance, the higher the similarity index.

[0022] It is understandable that when constructing the real-time state feature matrix, multiple feature dimensions may be involved. If the feature dimension is too high, the computational complexity will increase, and it may lead to overfitting, affecting the generalization ability of the model. Feature selection methods can be used to screen features and retain the most meaningful features for anomaly detection, or dimensionality reduction techniques (such as PCA and LDA) can be used to reduce feature dimensions, thereby reducing computational complexity and preventing overfitting.

[0023] See also Figure 6 As shown, based on the calculated similarity, if it exceeds the preset threshold for a long time, the hierarchical protection mechanism is triggered, including: Based on the obtained similarity, different levels of protection mechanisms are triggered; In the initial protection stage, when the threshold is exceeded in consecutive cycles, the PWM duty cycle is started for dynamic adjustment, and the adjustment range is: , in, is the adjustment amount of PWM duty cycle, that is, the PWM duty cycle that needs to be adjusted. is the proportional gain coefficient, which is used to amplify the current error. is the integral gain coefficient, which is used to amplify the accumulated error. is the expected value of the characteristic state data, is the actual value of the characteristic status data; In the intermediate protection stage, the power reduction mode is triggered and the rated power is reduced every cycle; In the advanced protection stage, the main power circuit is immediately cut off and the sound and light alarm is triggered.

[0024] See also Figure 7 As shown in the figure, after the autonomous protection is executed, the complete event data is uploaded to the database, the power anomaly prediction model is updated, and the protection trigger threshold is dynamically adjusted based on the deviation between the prediction result and the actual state. Specifically, it includes: Regularly retrain the power anomaly prediction model based on newly uploaded data, and verify and evaluate the model after it is updated; The protection trigger threshold is dynamically adjusted based on the deviation between the predicted result and the actual state. The deviation calculation formula is: , in, is the predicted abnormal probability at the kth time point, is the actual abnormal probability of the kth cycle, is the number of samples in the statistical window; The threshold is adjusted dynamically based on the PID control concept, and the effect of the new threshold is simulated at the edge computing node. After confirmation, it is synchronized to all LED switch power supplies.

[0025] Specifically, the effect of the new threshold is simulated in the edge computing node, and the impact of different thresholds on system behavior is tested in the local simulation environment to ensure that the new threshold can effectively improve the accuracy of anomaly detection and avoid false positives and false negatives. After confirmation, the adjusted new threshold is synchronized to all LED switch power supplies to ensure that each power supply unit operates at the new threshold.

[0026] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0027] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0028] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An autonomous protection control method for abnormal feedback of LED switching power supply, characterized in that: include: Based on the power supply operation data in the historical database, the time domain voltage waveform, current spectrum characteristics, power device temperature change rate and PWM modulation parameters within the switching cycle are obtained; Build a prediction model for the operating status of LED switch power supplies and pre-train on the historical database using a transfer learning algorithm; Based on the real-time feature window sampling of the LED switching power supply, it is input into the prediction model to output the state prediction vector and abnormal probability distribution of the next switching cycle; Based on the prediction results, when the abnormal probability exceeds the dynamic threshold, a real-time state feature matrix is ​​constructed, the real-time state of the LED switch power supply is dynamically matched with the predicted abnormal pattern, and the feature similarity index is calculated; Based on the calculated similarity, if it exceeds the preset threshold for a long time, the hierarchical protection mechanism will be triggered; After the autonomous protection is executed, the complete event data is uploaded to the database, the power anomaly prediction model is updated, and the protection trigger threshold is dynamically adjusted based on the deviation between the prediction result and the actual state.

2. The autonomous protection control method for abnormal feedback of LED switching power supply according to claim 1, characterized in that: The obtaining of the time domain voltage waveform, current spectrum characteristics, power device temperature change rate and PWM modulation parameters within the switching cycle based on the power supply operation data in the historical database specifically includes: Based on synchronous trigger acquisition and digital filtering to process the power supply voltage signal, extract effective data and ripple coefficient, and obtain the time domain voltage waveform; Analyze the current spectrum based on FFT, calculate the harmonic distortion rate, identify the main harmonic components, and obtain the current spectrum characteristics; Synchronize temperature and switch status data, and obtain the temperature change rate of power devices based on differential calculation of temperature rise rate; Measure PWM drive signal parameters and analyze duty cycle characteristics to identify modulation strategy types.

3. The autonomous protection control method for abnormal feedback of LED switching power supply according to claim 1, characterized in that: The construction of the prediction model of the power supply operation status and the pre-training on the historical database by the transfer learning algorithm specifically include: Based on the LSTM+CNN hybrid structure, a source domain pre-training model is constructed. The input layer receives time series data, and the output layer predicts the remaining life. The reconstruction loss and prediction loss are jointly optimized through the loss function to learn the potential representation of the power supply operation status. Calculate the domain difference metric distance between the source domain and the target domain and map them to the common subspace, freeze the first three layers of the pre-trained model, and fine-tune the top-level parameters for adaptive training; The weight of the data in the historical database is updated based on the TrAdaBoost algorithm, the weight of the misclassified target domain samples is increased, and the influence of irrelevant source domain samples is reduced.

4. The autonomous protection control method for abnormal feedback of LED switching power supply according to claim 1, characterized in that: The real-time feature window sampling based on the LED switching power supply is input into the prediction model to output the state prediction vector and abnormal probability distribution of the next switching cycle, which specifically includes: Based on the complete switching cycle, set the real-time feature sampling window length; The input features are dynamically normalized, and the normalization parameters come from the statistical features of the previous cycles; The model output includes the predicted value of voltage fluctuation in the next cycle, the predicted current harmonic distribution, and the predicted temperature change rate; The abnormal probability distribution output is the conditional probability value of each fault mode, including four types of abnormalities: overvoltage, overcurrent, overheating, and short circuit.

5. The autonomous protection control method for abnormal feedback of LED switching power supply according to claim 1, characterized in that: Based on the prediction results, when the abnormal probability exceeds the dynamic threshold, a real-time state feature matrix is ​​constructed, the real-time state of the LED switch power supply is dynamically matched with the predicted abnormal mode, and the feature similarity index is calculated, which specifically includes: The initial threshold is obtained based on historical normal data statistics, and the threshold is adjusted dynamically. The initial threshold formula is: ， The dynamic threshold adjustment formula is: , in, is the mean value of the characteristic data under normal conditions, is the standard deviation of the characteristics under normal conditions, representing the degree of dispersion of the data. is the forgetting factor, which controls the historical initial threshold weight, is the sliding window length, and the average abnormal probability of the past switching cycles is calculated. is the abnormal probability of the past k-th period; based on In principle, 99.7% of normal data falls within If it exceeds this range, it will be considered abnormal. Construct a real-time state feature matrix based on the feature data in the current switching cycle when an abnormality occurs; Acquire a set of characteristic matrix templates of pre-stored typical abnormal patterns based on a historical database; The real-time state feature matrix is ​​time-aligned with each abnormal template, and the influence of timing drift is eliminated based on dynamic time warping. The dynamic time warping distance formula is: , in, is the DTW distance between the real-time state feature matrix and the i-th abnormal template, is the total step length of the alignment path, that is, the length of the sequence after dynamic matching, is the jth matching point on the alignment path, is the distance function between feature vectors, is the feature vector of the real-time state feature matrix at a certain moment in the time window at path point j, is the feature vector of the i-th abnormal template at path point j; Based on the obtained dynamic time warping distance, a feature similarity index is obtained. The smaller the dynamic time warping distance, the higher the similarity index.

6. The autonomous protection control method for abnormal feedback of LED switching power supply according to claim 1, characterized in that: If the similarity obtained based on the calculation exceeds a preset threshold for a long time, the hierarchical protection mechanism is triggered, specifically including: Based on the obtained similarity, different levels of protection mechanisms are triggered; In the initial protection stage, when the threshold is exceeded in consecutive cycles, the PWM duty cycle is started for dynamic adjustment, and the adjustment range is: ， in, is the adjustment amount of PWM duty cycle, that is, the PWM duty cycle that needs to be adjusted. is the proportional gain coefficient, which is used to amplify the current error. is the integral gain coefficient, which is used to amplify the accumulated error. is the expected value of the characteristic state data, is the actual value of the characteristic status data; In the intermediate protection stage, the power reduction mode is triggered and the rated power is reduced every cycle; In the advanced protection stage, the main power circuit is immediately cut off and the sound and light alarm is triggered.

7. The autonomous protection control method for abnormal feedback of LED switching power supply according to claim 1, characterized in that: After the autonomous protection is executed, the complete event data is uploaded to the database, the power anomaly prediction model is updated, and the protection trigger threshold is dynamically adjusted based on the deviation between the prediction result and the actual state. Specifically, it includes: Regularly retrain the power anomaly prediction model based on newly uploaded data, and verify and evaluate the model after it is updated; The protection trigger threshold is dynamically adjusted based on the deviation between the predicted result and the actual state. The deviation calculation formula is: , in, is the predicted abnormal probability at the kth time point, is the actual abnormal probability of the kth cycle, is the number of samples in the statistical window; The threshold is adjusted dynamically based on the PID control concept, and the effect of the new threshold is simulated at the edge computing node. After confirmation, it is synchronized to all LED switch power supplies.

Images