Buck converter control method based on improved dandelion fuzzy neural network PID algorithm

Abstract

The invention belongs to the technical field of switching power supplies, and particularly relates to a Buck converter control method based on an improved dandelion fuzzy neural network PID (Proportion Integration Differentiation) algorithm, boundary reflection is introduced into a dandelion algorithm, and when a solution generated by the algorithm exceeds a parameter feasible region, the solution is reflected along a boundary direction instead of being directly abandoned to form a mirror image solution. And a spiral search strategy is introduced in a landing stage, so that the problem of insufficient precision caused by the fact that a traditional algorithm is prone to stagnation and voltage stabilization during later local development is solved. In the later stage of the algorithm, a spiral search strategy is adopted, spiral progressive search is carried out with the current optimal solution as the center, parameter adjustment is refined in combination with the gradient descent thought, the number of redundant iterations is reduced, and the purpose that output voltage regulation and control are completed faster and more accurately is achieved. PID parameters can be accurately adjusted to a fine adjustment interval in combination with the fuzzy neural network, and the steady-state error of the output voltage is reduced. According to the invention, accurate calculation of PID parameters is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of switching power supplies, relates to a Buck-type DC-DC converter, and particularly relates to a control method for a Buck converter based on an improved dandelion fuzzy neural network PID algorithm. Background Art

[0002] Due to its wide application fields and significant advantages, switching power supplies exhibit higher energy conversion efficiency compared to linear regulated power supplies, and thus have been widely used in portable electronic devices. In order to further improve the dynamic response ability of the converter, researchers in the field of power electronics have been committed to exploring effective control strategies. Among them, digital control methods have become an important way to improve the dynamic performance of the converter because they can more easily implement advanced, complex, and intelligent control algorithms.

[0003] In a DC-DC Buck power supply, voltage control loop compensation is crucial to ensure that the system stability and response speed meet the standards, and it affects output ripple, load regulation rate, anti-interference, etc. Designing loop compensation is very difficult and requires profound theory and experience. Therefore, researching a method using a digital PID algorithm as an alternative control strategy has positive significance for efficiently implementing closed-loop control of the circuit, improving the convergence speed, and reducing the output voltage overshoot value. Summary of the Invention

[0004] The purpose of the present invention is to provide a control method for a Buck converter based on an improved dandelion fuzzy neural network PID algorithm to solve the problems in the background art.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] A control method for a Buck converter based on an improved dandelion fuzzy neural network PID algorithm includes the following steps:

[0007] Step 1: Obtain the parameters of the Buck converter, including input voltage, output voltage, output current, and duty cycle;

[0008] Step 2: Construct a small-signal model with input voltage, output voltage, output current, and duty cycle to obtain the transfer function of the buck system, and randomly generate a population of candidate solutions within a predefined range to obtain the initial value range of the PID parameters;

[0009] Step 3: Within the value range, use a cost function to calculate the integral of the time-weighted absolute error (ITAE) corresponding to each parameter value used to characterize the PID performance;

[0010] Step 4: Based on the improved dandelion algorithm, use a fuzzy neural network to calculate the PID parameters. The calculation process of the improved dandelion algorithm includes an initial stage, an ascending stage, a descending stage, and an update stage. The implementation process is as follows:

[0011] Step 4.1: In the initial stage, by minimizing the integral of each time-weighted absolute error (ITAE), adjust the initial value range of the PID parameters obtained in Step 3;

[0012] Step 4.2: In the ascending stage and the descending stage, optimize the minimum cost function by balancing exploration measurement and exploitation strategies to obtain the initial parameters of the PID;

[0013] Step 4.3: In the update stage, first set the initial parameters obtained in Step 4.2 as the initial state of the PID, and then use the fuzzy neural network to update the PID parameters to obtain the optimal parameter output.

[0014] Advantages of the present invention: A PID optimization control method for a Buck-type DC-DC converter provided by the present invention combines the improved dandelion algorithm with a fuzzy neural network to realize the calculation of PID parameters. In this process, by introducing boundary reflection in the dandelion algorithm, when the solution generated by the algorithm exceeds the parameter feasible region, it is not directly discarded but reflected along the boundary direction to form a mirror solution. And the spiral search strategy introduced in the landing stage. Traditional algorithms are prone to stagnation in the later stage of local development (such as the "premature convergence" problem of the particle swarm optimization algorithm), resulting in insufficient steady-state accuracy of the Buck system. In the later stage of the algorithm, a spiral search strategy is adopted to perform a spiral progressive search centered on the current optimal solution, and the parameter adjustment is refined in combination with the gradient descent idea, reducing the redundant iteration times, achieving the purpose of more quickly and accurately regulating the output voltage. Combining with the fuzzy neural network can accurately adjust the PID parameters to the fine-tuning interval and reduce the steady-state error of the output voltage. Description of the Drawings

[0015] Figure 1 It is a schematic diagram of the control method loop of the Buck converter in the embodiment;

[0016] Figure 2 It is a comparison of the fitness evolution curves of the improved dandelion algorithm, the traditional dandelion algorithm (DO), and the particle swarm optimization algorithm (PSO).

[0017] Figure 3 It is a comparison of the output voltage simulation results of the Buck converter control method provided in the embodiment with the original dandelion algorithm and the improved dandelion algorithm. Detailed Embodiment

[0018] The technical solution of the present invention will be described in detail below in conjunction with the drawings and embodiments.

[0019] The Buck converter control method based on the improved dandelion fuzzy neural network PID algorithm provided by this embodiment includes the following steps:

[0020] Step 1: Obtain the parameters of the Buck converter, including input voltage, output voltage, output current, and duty cycle.

[0021] Step 2: Construct a small-signal model with input voltage, output voltage, output current, and duty cycle to obtain the transfer function of the buck system, and randomly generate a population of candidate solutions within a predefined range to obtain the initial value range of the PID parameters. The implementation process includes:

[0022] The circuit of the Buck converter consists of a switching tube, an inductor, a capacitor, a freewheeling diode, and a load resistor. Based on the inductor volt-second balance principle and the capacitor charge balance principle, the output transfer function of the Buck is established as follows:

[0023]

[0024] In formula (1), V i is the input voltage, the inductance value is L, the capacitance is C, the resistance is R, and s is the Laplace complex variable. According to the transfer function, the initial value ranges corresponding to the proportional, integral, and differential of the PID parameters can be obtained.

[0025] Step 3: Within the value range, use the cost function to calculate the integral of the time-weighted absolute error (ITAE) corresponding to each parameter value characterizing the PID performance. The specific calculation formula is as follows:

[0026]

[0027] In formula (2), t is the time variable, and e(t) is the error value of the output voltage at time t.

[0028] Step 4: Based on the improved dandelion algorithm, use a fuzzy neural network to calculate the PID parameters. In this embodiment, during the operation of the dandelion algorithm, boundary conditions are set. If an element exceeds the set boundary range, its position after reflection is calculated and adjusted to the valid range to ensure that the solutions in the solution space are always within a reasonable interval. The implementation process includes:

[0029] Step 4.1: In the initial stage, adjust the initial value range of the PID parameters obtained in Step 3 by minimizing the integral of each time-weighted absolute error (ITAE).

[0030] Step 4.2: In the rising and falling stages, optimize the minimum cost function by balancing the exploration measurement and exploitation strategies to obtain the initial parameters of the PID. The specific steps include:

[0031] (1) Construct the position calculation formulas for dandelion seeds in the ascending stage and the position calculation formulas for dandelion seeds in the descending stage respectively:

[0032] The ascending stage corresponds to the exploration stage of the dandelion algorithm. During this stage, due to factors such as weather, dandelion seeds will rise to different heights. Therefore, the ascending stage is expressed in two cases: sunny days and rainy days.

[0033] The expression corresponding to the position of dandelion seeds on sunny days is:

[0034] X t+1 = X t + αv x v y lnY(X s - X t )(3)

[0035] In formula (3), X t is the position of the seed at the t-th iteration; X s is the position randomly selected in the search space during iteration; α is the search step size; v x， v y is the lift coefficient of the dandelion seed.

[0036]

[0037] In formula (4), the function lnY follows a lognormal distribution with a mean μ = 0 and a variance σ 2 = 1;

[0038] On rainy days, dandelion seeds will be affected by factors such as air humidity, resulting in their diffusion in a local area. The corresponding formula is:

[0039] X t+1 = X t k(5)

[0040] In formula (5), the k value adjusts the local search area of the dandelion; the calculation formula for the k value is:

[0041]

[0042] In formula (6), T is the number of iterations. As the number of iterations increases, the parameter k value gets closer and closer to 1. Therefore, it can ensure that the population finally converges to the optimal search individual.

[0043] Based on the above analysis, the position calculation formula for dandelion seeds in the ascending stage of this embodiment is obtained as:

[0044]

[0045] In formula (7), rand(n) is a random number that follows a standard normal distribution.

[0046] Descent stage: After the dandelion seeds rise to a certain height in this stage, they start to descend. The dandelion algorithm uses Brownian motion to imitate the movement trajectory of dandelion seeds, and uses the average position information obtained in the ascent stage for regional optimization. The calculation formula for the descent stage is constructed as follows:

[0047]

[0048]

[0049] In equations (8) and (9), β t is a random number, following Brownian motion with a normal distribution; is the average position information of the population at the t-th iteration, N p is the number of the population, X i is the position information of the population when the population number is i.

[0050] (2) After going through the ascent and descent stages, it enters the landing stage of the dandelion algorithm. In the previous two stages, the dandelion seeds randomly select positions to land. As the number of iterations increases, the algorithm gradually approaches the global optimal solution, that is, the approximate position where the seeds are most likely to survive. In this embodiment, by introducing the whale search strategy in this stage, the characteristics of the whale algorithm in surrounding prey, spiral position update, and random search are combined with the reproduction and propagation mechanisms of the dandelion algorithm to enhance the local search ability and update the individual positions; its expression formula is:

[0051] X t+1 = X t + D k · e bl · cos(2πl) (10)

[0052] D k = |X t - X t+1 | (11)

[0053] In equation (10), b is the logarithmic spiral constant, l is a random number in [-1, 1], and it shrinks towards the ground in a spiral manner during the falling stage to enhance the convergence speed and search ability. The continuously updated optimal position matrix parameter X t+1 is the PID parameter obtained in the system at this time, and the initial PID parameter is obtained in this step.

[0054] Step 4.3: In the update stage, first set the initial parameters obtained in step 4.2 as the initial state of PID, and then use the fuzzy neural network to update the PID parameters to obtain the optimal PID parameter output.

[0055] The fuzzy neural network of this embodiment is a neural network formed by integrating fuzzy rules into a BP neural network based on the Mamdani fuzzy inference method. The network includes: an input layer, a fuzzification layer, a fuzzy inference layer, a normalization layer, and an output layer; the input layer is used to input the error e and the error change amount ec and transmit them to the fuzzification layer; the fuzzification layer performs fuzzification processing on the received error e and error change amount ec, selects trigonometric functions as membership functions, and transmits them to the fuzzy inference layer; the fuzzy inference layer is based on one of the preset fuzzy rules: if input 1 (e) is "NB" and input 2 (ec) is "NB", then output 1 (kp), output 2 (ki), and output 3 (ki) are all "NB", and so on. Based on the fuzzified error e and error change amount ec, the Mamdani inference method is used to complete fuzzy operations; the normalization layer is used to normalize the operation results output by the fuzzy inference layer and then transmit them to the output layer; the output layer is used to perform defuzzification processing on the normalized data to obtain the output of the optimal PID control parameters.

[0056] Figure 1 FIG. is a schematic diagram of the control loop of the Buck converter control method based on the improved dandelion fuzzy neural network PID algorithm of this embodiment. The control loop includes: a basic buck circuit model, cost function calculation, and a PID control method optimized by a fuzzy neural network based on improved dandelion optimization to complete closed-loop control.

[0057] Figure 2 shows the comparison results of the fitness evolution curves of the improved dandelion algorithm with the traditional dandelion algorithm (DO) and the particle swarm optimization algorithm (PSO). As Figure 2 can be seen, with the increase of the number of iterations, the improved dandelion algorithm (DWOA) of this embodiment has the prominent advantage of finding the optimal solution faster.

[0058] Figure 3 is a comparison of the output voltage simulation results of the Buck converter control method provided in the embodiment with the original dandelion algorithm and the improved dandelion algorithm. As Figure 3 can be seen, the initial output voltage overshoot value of this embodiment is smaller, and the time to reach the steady state of the output voltage is shorter.

[0059] In summary, by adopting the PID optimization control method of this embodiment, through the combination of the improved dandelion algorithm and the fuzzy neural network, the circuit can be efficiently closed-loop controlled, the output voltage overshoot value of the Buck control model is reduced, and the convergence speed is accelerated, solving the complex and difficult problems of traditional Buck design loop compensation.

Claims

1. Buck converter control method based on improved dandelion fuzzy neural network PID algorithm, characterized in that: The following steps are involved: Step 1: Obtain the parameters of the Buck converter, including input voltage, output voltage, output current and duty cycle; Step 2: construct a small signal model based on input voltage, output voltage, output current and duty cycle to obtain the transfer function of the buck system, randomly generate a group of candidate solutions within a predefined range, and obtain the initial value range of PID parameters; Step 3: Within the range of values, use the cost function to calculate the integral of the time-weighted absolute error (ITAE) corresponding to each parameter value used to characterize the PID performance; Step 4: Based on the improved Dandelion algorithm, the PID parameters are calculated using a fuzzy neural network. The calculation process of the improved Dandelion algorithm includes an initial stage, an ascending stage, a descending stage, and an updating stage. The implementation process includes: Step 4.1, in the initial stage, by minimizing the integral of each time-weighted absolute error (ITAE), adjust the initial value range of the PID parameters obtained in step 3; Step 4.2, in the ascending stage and descending stage, the minimum cost function is optimized by balancing the exploration measurement and development strategy to obtain the initial parameters of the PID; Step 4.3: In the update phase, the initial parameters obtained in step 4.2 are first set as the PID initial state, and then the fuzzy neural network is used to update the PID parameters to obtain the optimal parameter output.

2. The Buck converter control method based on the improved Dandelion fuzzy neural network PID algorithm according to claim 1 is characterized in that: The implementation method of step 2 includes: Based on the inductor volt-second balance principle and the capacitor charge balance principle, the output transfer function of Buck is established as follows: Among them, V i is the input voltage, inductance value L, capacitance C, resistance R, s is the Laplace complex variable; The proportional-integral-differential of the PID parameters is calculated according to the transfer function, so as to obtain the initial value range of the PID parameters.

3. The Buck converter control method based on the improved Dandelion fuzzy neural network PID algorithm according to claim 2 is characterized in that: In step 3, within the range of values, the cost function is used to calculate the integral of the time-weighted absolute error (ITAE) corresponding to each parameter value used to characterize the PID performance. The specific calculation formula is as follows: Where t is the time variable and e(t) is the error value of the output voltage at time t.

4. The Buck converter control method based on the improved Dandelion fuzzy neural network PID algorithm according to claim 3 is characterized in that: The implementation process of step 4.2 includes: Step 4.2.1, respectively construct the position calculation formula of dandelion seeds in the ascending stage and the position calculation formula of dandelion seeds in the descending stage: The formula for calculating the position of dandelion seeds in the rising stage is: In the formula, X t is the seed position at the tth iteration; X s is a randomly selected position in the search space during iteration; α is the search step length; v x， v y is the lift coefficient of dandelion seeds, and the function lnY has a mean μ = 0 and a variance σ 2 =1 lognormal distribution; k value is used to adjust the local search area of ​​dandelion, rand(n) is a random number that obeys the standard normal distribution; The formula for calculating the position of a dandelion seed in the descending phase is: In the formula, β t is a random number, which obeys the Brownian motion of normal distribution; is the population average position information N at the tth iteration p is the population size, X i is the population location information when the population number is i; Step 4.2.2, after the rising and falling stages, the dandelion algorithm enters the landing stage. In this stage, by introducing the whale search strategy, the characteristics of the whale algorithm in surrounding prey, spiral updating position and random search are combined with the reproduction and propagation mechanism of the dandelion algorithm to enhance the local search capability, which is used to update the individual position, thereby obtaining the initial parameters of the PID; the expression formula of the process is: X t+1 =X t +D k ·e bl ·cos(2πl) D k =X t -X t+1 In the formula, b is the logarithmic spiral constant, l is a random number between -1,1, and it shrinks toward the ground in a spiral during the falling phase, enhancing the convergence speed and search capability, and continuously updating the optimal position matrix parameter X t+1 , which is the PID parameter obtained in the system at this time.

Images