A method for predicting the lifetime of an LED

Abstract

This invention relates to a method for predicting LED lifetime, comprising the following steps: dividing the dataset into a training set and a test set; combining ACO (Ant Colony Algorithm) and GSO (Growth Spirit Algorithm) to form an optimization algorithm based on ACO and GSO (Ant Colony Algorithm + GSO); using the ACO+GSO hybrid intelligent optimization algorithm to optimize the initial parameters of WNN (Wavelet Neural Network) to avoid detection accuracy problems caused by improper selection of initial parameters; using WNN to learn from the training dataset and form an optimized model; wherein, parameters such as temperature, current, initial luminous flux, and initial color coordinates are used as inputs to WNN, and LED lifetime is used as the output of WNN; the formed optimized model is used to detect the LED lifetime in the test set, solving the problems of long prediction time and poor accuracy, improving the computational accuracy and convergence speed of traditional intelligent algorithms, avoiding problems such as local optimization, reducing manual intervention, and improving the reliability of LED lifetime detection.

Description

Technical Field

[0001] This invention belongs to the field of LED testing technology, and specifically relates to a method for predicting LED lifespan. Background Technology

[0002] In recent years, semiconductor lighting has been increasingly widely used in industrial fields and daily life. Light-emitting diodes (LEDs) are electroluminescent semiconductor light-emitting devices, belonging to the category of novel solid-state cold light sources, possessing advantages such as low voltage driving, high energy efficiency, long lifespan, and low cost. However, with the development of LED technology, a significant limiting factor in LED research and development is the gap between its actual lifespan and the theoretical value. Therefore, conducting corresponding lifespan tests and estimating the lifespan of LEDs based on various indicators to improve their reliability and extend their service life has become a necessary strategy. Accurately predicting LED lifespan has thus become an important research topic.

[0003] Currently, data-driven methods for predicting LED lifetime mainly include statistical regression methods and classical machine learning algorithms. Due to the limitations of model selection, traditional statistical methods cannot infer LED lifetime under conditions of multidimensional influencing factors; while classical machine learning algorithms, although possessing good data processing capabilities, still have certain limitations in terms of running time and prediction accuracy, and cannot achieve reliable LED lifetime prediction. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting LED lifetime. This method evaluates LED reliability through multi-dimensional index parameters and effectively and accurately predicts LED lifetime by combining intelligent optimization algorithms and wavelet neural network methods. The method, based on hybrid ant colony optimization (ACO) and glowworm swarm optimization (GSO) optimized wavelet neural network (WNN) (ACO+GSO-WNN), searches and optimizes parameters. This method avoids the shortcomings of both ACO and GSO, achieving complementary advantages, while reducing manual intervention in WNN algorithm parameter tuning. It boasts high prediction accuracy, strong robustness, and promising application prospects.

[0005] To achieve the objectives of the invention described above, the technical solution adopted by the present invention is as follows:

[0006] A method for predicting LED lifetime, comprising hybrid ant colony (ACO), firefly (GSO), and wavelet neural network (WNN), characterized in that the method includes the following steps:

[0007] S1. Obtain LED data and divide the dataset into training and test sets;

[0008] S2, Set ant colony rules N ACO Maximum number of iterations T ACO The volatile coefficient ρ∈[0,1] and the pheromone concentration Q are determined; the ranges of the kernel function parameter σ and the regularization parameter γ are set, and a set of parameter sequences (σ,γ) are randomly generated as the initial position vector of the ant.

[0009] S3. Calculate the fitness value of each ant at its current location; calculate the pheromone concentration at the ant's current location j. The lower the fitness, the higher the pheromone concentration. Fitness is defined by the mean squared error.

[0010] S4. Determine the position with the minimum ant fitness value based on the pheromone concentration of each ant; iteratively update the pheromone concentration to satisfy the fitness condition f ≤ f d The ant position vector (σ,γ) is placed into set X. ACO This continues until the iteration ends.

[0011] S5. Set the firefly population size to N. GSO Maximum number of iterations T GSO The optimized X obtained by the ant colony algorithm ACO Initialize firefly N GSO The location of each firefly is determined by selecting one location from the remaining fireflies, and each individual firefly carries the same concentration of luciferin. o and perception radius r o ;

[0012] S6. Update the firefly's luciferin;

[0013] S7, Find the neighbors of firefly i;

[0014] S8. Determine the direction of firefly i's movement; when the luciferin value of neighboring firefly j is greater than that of firefly i, and the distance between the two fireflies is within the perception range r. s Within a certain time, firefly i will have a probability p ij (t) Select neighboring firefly j and move in the direction of neighboring firefly j;

[0015] S9. Update the position of firefly i, then update the decision domain, and iteratively determine the extreme value p of each individual firefly. best and optimal position;

[0016] S10, calculate the individual extreme values ​​p of each firefly. best fitness value g of the population's optimal position bestIn comparison, if the best position is found, then the best position of that firefly is taken as the best position of the population, and the individual extreme value p of that firefly is determined. best As the population extreme value g best Check if the iteration termination condition is met. If it is met, end the optimization and find the optimal solution (σ,γ); otherwise, return to S3.

[0017] S11. Initialize the weights W of the WNN using the optimal solution (σ,γ) obtained in S10. k And threshold B, the specific parameters are set as follows:

[0018] X = (x1, x2, ..., x n ) T Input data for the input node;

[0019] W k =(w k1 ,w k2 ,...,w kn ) represents the connection weight between the k-th hidden node in the hidden layer and the input layer;

[0020] W o =(w1,w2,...,w k ) represents the connection weights between the output layer and the hidden nodes;

[0021] H o =(h1,h2,...,h k ) represents the output value of each hidden node in the hidden layer;

[0022] B = (b1, b2, ..., b k ) represents the threshold value for each hidden node in the hidden layer;

[0023] b o The threshold value for the output layer nodes;

[0024] f1 is the activation function of the hidden layer.

[0025] The wavelet mother function is expressed as:

[0026]

[0027] The hidden layer output is:

[0028]

[0029] In the formula, b k ,a k These are the translation factor and scaling factor of the wavelet basis functions, respectively;

[0030] The output layer is:

[0031] O=f2(W oH o -b0) (12)

[0032] S12. Based on the output layer, establish an optimized WNN model to predict the LED lifespan and obtain the LED lamp lifespan prediction results.

[0033] Furthermore, the specific formula for calculating the fitness value of each individual ant's current location in step S3 is as follows:

[0034]

[0035] In the formula, y i The actual values ​​for the test set;

[0036] y' i These are the predicted values ​​for the test set;

[0037] N is the number of test sets;

[0038] The specific formula for calculating the pheromone concentration at the current position j of the ant is as follows:

[0039]

[0040] In the formula, The pheromone left by the k-th ant at position j in this cycle;

[0041] t represents the current time.

[0042] s k Let be the fitness value of the k-th ant at this position.

[0043] Furthermore, the specific formula for iteratively updating the pheromone concentration in step S4 is as follows:

[0044] τ j (t+m)=(1-ρ)τ j (t)+Δτ j (t) (4)

[0045] In the formula, ρ (0 < ρ < 1) is the pheromone evaporation coefficient.

[0046] m is the time required for each pheromone iteration;

[0047] j represents the location in the parameter optimization space;

[0048] During the iteration process, the probability formula for the k-th ant randomly choosing position j from the solution space I at time t is:

[0049]

[0050] Furthermore, the specific formula for updating the luciferin of the firefly in step S6 is as follows:

[0051] l i (t)=(1-ρ)l i (t-1)+γJ(x i (t)) (5)

[0052] In the formula, J(x) i (t) represents the position x of each firefly i in iteration t. i The objective function value corresponding to (t);

[0053] l i (t) represents the current luciferin value of the firefly;

[0054] γ represents the luciferin turnover rate;

[0055] ρ is the volatile factor of fluorescein.

[0056] Furthermore, the specific formula for finding the neighbors of firefly i in step S7 is as follows:

[0057]

[0058] In the formula, N i (t) is the set of neighbors of the i-th firefly in the t-th generation;

[0059] ||x|| is the norm of x;

[0060] x j (t) represents the position of the j-th firefly in generation t;

[0061] l j (t) The luciferin value of the j-th firefly in the t-th generation;

[0062] Dynamic decision domain range This determines the number of neighbors;

[0063] Its upper limit of perception range is

[0064] Furthermore, in step S8, the probability p ij (t) Its formula is:

[0065]

[0066] Furthermore, the specific formula for updating the position of firefly i in S9 is as follows:

[0067]

[0068] In the formula, S is the movement step size;

[0069] The specific formula for updating the decision domain is as follows:

[0070]

[0071] In the formula, |N i (t) represents the number of fireflies in the neighborhood of firefly i.

[0072] Compared with existing technologies:

[0073] 1. Based on the working environment conditions of LEDs, the factors affecting LED lifespan are comprehensively considered from multiple dimensions of index parameters to accurately describe the reliability characteristics of LEDs.

[0074] 2. A novel ACO+GSO hybrid intelligent optimization algorithm was designed to improve the computational accuracy and convergence speed of traditional intelligent algorithms and avoid problems such as local optimization.

[0075] 3. WNN is quite sensitive to initial parameters, often requiring experienced technicians to conduct repeated experiments and parameter tuning based on the characteristics of the business scenario. This invention uses the ACO+GSO hybrid intelligent optimization algorithm to automatically optimize and adjust the initial parameters of WNN, reducing manual intervention.

[0076] 4. The optimized WNN can effectively predict LED lifespan and improve the detection performance of LED reliability. Attached Figure Description

[0077] Figure 1 This is a flowchart of the LED lifetime prediction method in an embodiment of the present invention;

[0078] Figure 2 This is a flowchart of the WNN algorithm based on ACO+GSO in an embodiment of the present invention;

[0079] Figure 3 This is the network topology diagram of the WNN in this embodiment of the invention. Detailed Implementation

[0080] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0081] Combination Figure 1 , Figure 2 , Figure 3 As shown,

[0082] Step 1: Using the LED manufacturer's LM-80-08 test report as the data source, calculate the lifespan of each LED using the TM-21-11 method, and divide the dataset into training and test sets. (Collect data sources and divide the dataset into training and test sets)

[0083] Step 2: Set ant colony rules N ACO Maximum number of iterations T ACO The volatile coefficient ρ∈[0,1] and the pheromone concentration Q are set; the range of kernel function parameter σ and regularization parameter γ are set, and a set of parameter sequences (σ,γ) are randomly generated as the initial position vector of the ant.

[0084] Step 3: Calculate the fitness value of each ant at its current position using formula (1), and calculate the pheromone concentration at position j of the ant at the current moment using formula (2). The smaller the fitness, the larger the pheromone concentration. Fitness is defined by mean squared error:

[0085]

[0086] Among them, y i and y' i These represent the actual value and the predicted value of the test set, respectively, and N is the number of test sets.

[0087] The pheromone concentration at position j of the ant at time t:

[0088]

[0089] in, s represents the pheromone left by the k-th ant at position j in this cycle. k This represents the fitness value of the k-th ant at this position.

[0090] Step 4: Determine the position with the minimum ant fitness value based on the pheromone concentration of each ant. Iteratively update the pheromone concentration using formula (4) to find the position that satisfies the fitness condition f ≤ f d (f d The ant position vector (σ, γ) with values ​​determined by verification experience is placed into set X. ACO This continues until the iteration ends. During the iteration process, the probability formula for the k-th ant randomly choosing position j from the solution space I at time t is:

[0091]

[0092] The formula for updating the pheromone concentration at spatial location j in the parameter optimization is as follows:

[0093] τ j (t+m)=(1-ρ)τ j (t)+Δτ j (t) (4)

[0094] Where ρ (0 < ρ < 1) represents the pheromone evaporation coefficient, and m is the time required for each pheromone iteration.

[0095] Step 5: Set the firefly population size to N. GSO Maximum number of iterations T GSO The optimized X obtained from the ant colony algorithm ACO Initialize firefly N GSO The location of each firefly is determined by selecting one location from the remaining fireflies, and each individual firefly carries the same concentration of luciferin. o and perception radius r o .

[0096] Step 6: Update the luciferin of the fireflies using formula (5).

[0097] l i (t)=(1-ρ)l i (t-1)+γJ(x i (t)) (5)

[0098] Wherein, J(x) i (t) represents the position x of each firefly i in iteration t. i The objective function value corresponding to (t), l i (t) represents the current luciferin value of the firefly, γ is the luciferin renewal rate, and ρ is the luciferin volatile factor.

[0099] Step 7: Find firefly i's neighbors:

[0100]

[0101] Where, N i (t) represents the set of neighbors of the i-th firefly in generation t, ||x|| represents the norm of x, and x j (t) represents the position of the j-th firefly in generation t. j (t) The luciferin value of the j-th firefly in generation t; the range of the dynamic decision domain. The number of neighbors is determined, and its upper bound is the perception range.

[0102] Step 8: Determine the direction of firefly i's movement. When the luciferin value of neighboring firefly j is greater than that of firefly i, and the distance between the two fireflies is within the perception range r... s Within a certain time, firefly i will have a certain probability p ij (t) Select neighbor firefly j and move in the direction of neighbor firefly j.

[0103]

[0104] Step 9: Update the position of firefly i using formula (8), update the decision domain using formula (9), and determine the extreme value p of each individual firefly through iteration.best And optimal position:

[0105]

[0106]

[0107] Where S is the movement step size, |N i (t) represents the number of fireflies in the neighborhood of firefly i.

[0108] Step 10: Set the individual extreme values ​​p of each firefly best fitness value g of the population's optimal position best If the comparison shows that the firefly's optimal position is better, then the optimal position of that firefly is taken as the optimal position of the population, and the individual extreme value p of that firefly is determined. best As the population extreme value g best Set the maximum number of iterations, check if the iteration termination condition is met. If it is met, end the optimization and find the optimal solution (σ, γ). Otherwise, return to step 3.

[0109] Step 11: Initialize the weights W of the WNN using the optimal solution (σ,γ) obtained in step (10). k And threshold B, the specific parameters are set as follows:

[0110] X = (x1, x2, ..., x n ) T Input data for the input node;

[0111] W k =(w k1 ,w k2 ,...,w kn ) represents the connection weight between the k-th hidden node in the hidden layer and the input layer;

[0112] W o =(w1,w2,...,w k ) represents the connection weights between the output layer and the hidden nodes;

[0113] H o =(h1,h2,...,h k ) represents the output value of each hidden node in the hidden layer;

[0114] B = (b1, b2, ..., b k ) represents the threshold value for each hidden node in the hidden layer;

[0115] b o This is the threshold value for the output layer nodes.

[0116] f1 is the activation function of the hidden layer, which is the wavelet mother function, expressed as:

[0117]

[0118] The hidden layer output is:

[0119]

[0120] Among them, b k ,a k These are the translation factor and scaling factor of the wavelet basis function, respectively.

[0121] The output layer is:

[0122] O=f2(W o H o -b0) (12)

[0123] Step 12: Use formula (12) to establish an optimized WNN model, predict the LED lifespan, and obtain the LED reliability test results.

[0124] Obviously, the above embodiments of the present invention are merely illustrative examples to clearly illustrate the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting the lifetime of an LED based on ant colony optimization (ACO), glowworm swarm optimization (GSO) and wavelet neural network (WNN), characterized in that, The method includes the following steps: S1. Obtain LED data, divide the dataset into training set and test set, and use temperature, current, initial luminous flux and initial color coordinate parameters as input to WNN; S2, Set the ant colony size N ACO Maximum number of iterations T ACO The volatile coefficient ρ∈[0,1] and the pheromone concentration Q are determined; the ranges of the kernel function parameter σ and the regularization parameter γ are set, and a set of parameter sequences (σ,γ) are randomly generated as the initial position vector of the ant. S3. Calculate the fitness value of each ant at its current location; calculate the pheromone concentration at the ant's current location j. The lower the fitness, the higher the pheromone concentration. Fitness is defined by the mean squared error. S4. Determine the position with the minimum ant fitness value based on the pheromone concentration of each ant; iteratively update the pheromone concentration to satisfy the fitness condition f ≤ f d The ant's position vector (σ, γ) is placed into set X. ACO Continue until the iteration ends; S5. Set the firefly population size to N. GSO Maximum number of iterations T GSO The optimized X obtained by the ant colony algorithm ACO Initialize firefly N GSO The location of each firefly is determined by selecting one location from the remaining fireflies, and each individual firefly carries the same concentration of luciferin. o and perception radius r o ; S6. Update the firefly's luciferin; S7, Find the neighbors of firefly i; S8. Determine the direction of firefly i's movement; when the luciferin value of neighboring firefly j is greater than that of firefly i, and the distance between the two fireflies is within the perception range r. s Within a certain time, firefly i will have a probability p ij (t) Select neighboring firefly j and move in the direction of neighboring firefly j; S9. Update the position of firefly i, then update the decision domain, and iteratively determine the extreme value p of each individual firefly. best and optimal position; S10, calculate the individual extreme values ​​p of each firefly. best fitness value g of the population's optimal position best In comparison, if the best position is found, then the best position of that firefly is taken as the best position of the population, and the individual extreme value p of that firefly is determined. best As the population extreme value g best Check if the iteration termination condition is met. If it is met, end the optimization and find the optimal solution (σ,γ); otherwise, return to S3. S11, using the optimal solution (σ, γ) obtained in S10 to initialize the weights W of the WNN k and threshold B, the specific parameters are set as: X = (x1, x2,..., x n ) T Data input for input node; W k =(w k1 ,w k2 ,...,w kn ) represents the connection weight between the k-th hidden node in the hidden layer and the input layer; W o =(w1,w2,...,w k ) represents the connection weights between the output layer and the hidden nodes; H o =(h1,h2,...,h k ) represents the output value of each hidden node in the hidden layer; B = (b1, b2, ..., b k ) represents the threshold value for each hidden node in the hidden layer; b o The threshold value for the output layer nodes; f1 is the activation function of the hidden layer. The wavelet mother function is expressed as: The hidden layer output is: In the formula, b k ,a k These are the translation factor and scaling factor of the wavelet basis functions, respectively; The output layer is: S12. Establish an optimized WNN model to predict the lifespan of LEDs and obtain the LED lamp lifespan prediction results. The specific formula for calculating the fitness value of each ant's current location in step S3 is as follows: where y i is the actual value of LED lifetime for the test set; y ′ i Test set LED lifetime prediction values; N is the number of test sets; The specific formula for calculating the pheromone concentration at the current position j of the ant is as follows: wherein Pjkis the pheromone left by the kth ant at position j in this cycle; t represents the current time. s k is the fitness value of the kth ant at this position.

2. The method of claim 1, wherein, The specific formula for finding the neighbors of firefly i in step S7 is as follows: wherein N i (t) is the set of neighbors of the i-th firefly at the t-th iteration; || || represents the norm; x j (t) is the position of the j-th firefly at the t-th iteration; l j (t) represents the luciferin value of the j-th firefly in the t-th iteration; Dynamic decision domain range The number of neighbors is determined; Its upper limit of perception range is , .

3. The method for predicting LED lifetime according to claim 2, characterized in that, The probability p in the step S8 ij (t) whose formula is: 。

Images