Method for controlling working mode of Boost converter

Abstract

The invention discloses a method for controlling the working mode of a Boost converter, and the method comprises the following steps: S1, collecting the input voltage Vin, the output voltage Vout and the output current Io of the Boost converter, carrying out the multiplication of the output voltage Vout and the output current Io, obtaining the output power Pout of a load, judging the working state of a Boost circuit, and obtaining a period T and a duty ratio D; and S2, constructing a target function, determining a reward function and defining and updating a Q value table through the obtained period T and the duty ratio D and corresponding state space and action. And S3, using a Q-learning algorithm to train and solve the optimal D and T by taking the minimum current stress and the minimum power error as targets according to different input parameters (Vin, Vout and Pout). And S4, converting the obtained optimal D and T into driving signals, and controlling the on-off of a switching tube, so that the Boost converter is in the optimal working mode. The problem that it is difficult for an existing Boost converter control method to enable the Boost converter to be in the needed working mode is solved.

Description

Technical Field

[0001] The invention belongs to the technical field of electronic power, and in particular relates to a method for controlling the working mode of a Boost converter. Background Art

[0002] Boost converter circuit belongs to the boost topology circuit, which is widely used in daily life, such as switching power supply control, DC output, power management chip, etc., especially in low voltage boosting occasions. Due to its simple structure, Boost circuit and Boost-type topology derived from Boost, such as forward circuit, half-bridge circuit, full-bridge circuit, PFC and other topologies are widely used in various fields that require DC boosting, such as power supply design in the converter valve system. The power converter generally adopts a two-stage structure, namely the front-stage PFC converter + the rear-stage isolated DC / DC converter. The front-stage PFC stage realizes the power factor correction function, and the rear-stage isolated DC / DC converter performs step-up and step-down.

[0003] At present, there are many control methods for Boost converters, and many scholars are studying the control methods of Boost circuits, hoping to obtain better control characteristics. The most basic control methods are peak current control, average current control, hysteresis current control, and single cycle control. Peak current control and hysteresis current control are control methods based on input current ripple. They have fast response speed but are very sensitive to current ripple. In addition, when the duty cycle is greater than 0.5, there is a subharmonic oscillation problem in peak current control. Slope compensation needs to be introduced. The hysteresis width of hysteresis current control has a great influence on system performance and switching frequency, resulting in the switching frequency being not constant within one power frequency cycle. Average current control has the ability to filter high-frequency switching noise because its controller contains a current error amplifier, but it also leads to a slow response speed. Since the working state of the Boost converter is relatively complex and there are many nonlinear parameters, it is difficult to accurately model it with traditional mathematical modeling methods. Therefore, when solving its optimal modulation strategy, it will face problems such as large amount of calculation, high complexity, and difficulty in finding the best solution. Faced with such a high-dimensional optimization solution system with complex modeling and changes, artificial intelligence (AI) technology has shown its advantages in many system optimization decisions with large data volume, complex modeling and uncertainty, and is very suitable for solving the optimization problem of high-dimensional complex systems.

[0004] The patent document with publication number CN116207988A discloses a four-tube Buck-Boost converter ZVS dynamic control method, and obtains the ZVS dynamic control core algorithm based on the relationship between the duty cycle and the power curve under the soft switching condition. The present invention judges the circuit working mode according to the transformation ratio and the output of the compensator on the basis of feedback closed-loop control of the inductor current, input voltage and output voltage, and calculates the duty cycle required for the circuit to enter the steady state according to the working mode of the circuit, and controls the phase shift duty cycle and the duty cycle of the switch tube to make the circuit enter the steady state. On the basis of the soft switching condition, the present invention considers the influence of the circuit resonance in the dead time on the inductor current, and proposes a corresponding compensation strategy; the present invention can judge the maximum power output point of the circuit according to the changes in the input and output voltages, thereby improving the conversion efficiency and dynamic response speed of the converter.

[0005] The patent document with publication number CN114977790A discloses a two-phase interleaved Boost-Sepic type high-gain converter based on coupled inductors and a control method, belonging to the field of power electronics technology. The two-phase interleaved Boost-Sepic structure of the present invention includes an input source Vin, a first independent inductor L1, a second independent inductor L2, a first switch tube S1 and a second switch tube S2, and a Sepic part composed of a first coupled inductor primary winding Lc1, a second coupled inductor primary winding Lc2, a capacitor C1, and a capacitor C3, and a Boost part composed of a diode D1, a diode D4, a capacitor C5, and a capacitor C6; a first voltage doubling unit includes a diode D2, a capacitor C2, and a first coupled inductor secondary winding Lc1', and a second voltage doubling unit includes a diode D5, a capacitor C4, and a second coupled inductor secondary winding Lc2'; a split capacitor structure includes a capacitor C7 and a capacitor C8, which constitute the output part of the converter. The present invention has the advantages of symmetrical topological structure, simple switch control, ultra-high voltage gain, small input current ripple and ultra-low switch tube voltage stress.

[0006] In practical applications, the Boost converter needs to work in different modes according to the actual working conditions. For example, it needs to work in DCM mode under light load or to achieve soft switching, and sometimes it needs to work in CRM or CCM mode. However, some existing control methods optimize voltage stress, while others optimize soft switching. However, while optimizing one indicator, other performance of the circuit cannot be taken into account, and the optimal overall performance cannot be achieved. Summary of the invention

[0007] In view of this, the present invention aims at the deficiencies of the prior art and provides a method for controlling the working mode of a Boost converter, which solves the problem that the existing Boost converter control method is difficult to put the Boost converter in a desired working mode.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for controlling the working mode of a Boost converter, comprising the following steps: S1, collecting the input voltage Vin, output voltage Vout, and output current Io of the Boost converter, multiplying the output voltage Vout and the output current Io, obtaining the output power Pout of the load, judging the working state of the Boost circuit through the data obtained by the above collection, and obtaining the period T and duty cycle D. S2, constructing the corresponding state space and action by using the period T and duty cycle D obtained in step 1, constructing the objective function, determining the reward function, and defining and updating the Q value table. S3, using the Q-learning algorithm, according to different input parameters (Vin, Vout, Pout), training with the minimum current stress and minimum power error as the goal to solve the optimal D and T. S4, converting the optimal D and T obtained in step 3 into a driving signal, controlling the on and off of the switch tube, so that the Boost converter is in the optimal working mode.

[0009] Furthermore, the step S1 also includes: collecting the input voltage Vin, the output voltage Vout, and the output current Io of the Boost converter, obtaining the output power Pout of the load, and obtaining the period T and the duty cycle D.

[0010] Furthermore, the step S2 further includes: constructing the corresponding state space and action, constructing the objective function, determining the reward function and reconstructing the Q value table. The specific process is as follows:

[0011] Set up the state space, action space and state space defined as S, action space defined as A, reward function as r(s, a) and Q value

[0012] S=[D,T]

[0013] A={C D ,C T}

[0014]

[0015]

[0016] Furthermore, the S2 also includes: randomly sampling Vin, Vout, and calculated Pout to construct a corresponding state space defined as S, an action space defined as A, a reward function as r(s, a) and a Q value.

[0017] Furthermore, S2 also includes: in the reinforcement learning method, the action at the previous moment determines the change of the state at the current moment. The process of using the Q-learning algorithm to solve the optimal modulation amount of the Boost converter is essentially a process of reaching the optimal control state s* from the current state s through a certain action strategy.

[0018] Furthermore, S2 also includes: when ΔG>0, it means that the peak current of the current state is greater than the peak current of the previous state, and a negative reward value is given to the action, and the larger the increment, the smaller the reward value. When ΔG<0, it means that the corresponding peak current decreases after executing the action, and a positive reward value is given to the action. When the current stress of the system is less than the defined minimum loss value Gmin, a large reward value is given, indicating that the optimal state has been reached from the initial state.

[0019] Furthermore, S2 also includes: when using as many exploration strategies as possible, trying to explore new operation strategies and retaining the optimal state under each strategy. Gmin of each stage of training is the lowest current stress and minimum power error state obtained in the previous stage of training.

[0020] Furthermore, in S3, the Q-learning algorithm, a classic algorithm in reinforcement learning, is used to train and solve the optimal D and T according to different input parameters (Vin, Vout, Pout) with the minimum current stress and the minimum power error as the goal.

[0021] Furthermore, S4 also includes: obtaining optimal D and T through S3, determining its working state to be CCM mode, CRM mode, or DCM mode, so that its Boost converter works in an optimal state with minimum current stress and minimum power error.

[0022] Since the working state of the Boost converter is relatively complex and there are many nonlinear parameters, it is difficult to accurately model it with traditional mathematical modeling methods. Therefore, when solving its optimal modulation strategy, it will face problems such as large amount of calculation, high complexity, and difficulty in finding the best solution. Faced with such a high-dimensional optimization solution system with complex modeling and changes, artificial intelligence technology has shown its advantages in many system optimization decisions with large data volume, complex modeling, and uncertainty, and is very suitable for solving the optimization problem of high-dimensional complex systems; the patent document with publication number CN116207988A discloses a four-tube Buck-Boost converter ZVS dynamic control method, and obtains the core algorithm of ZVS dynamic control based on the relationship between duty cycle and power curve under soft switching conditions; however, while optimizing one indicator, it fails to take into account other performance of the circuit, and cannot achieve the best comprehensive performance effect. .

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] The present invention provides a method for controlling the working mode of a Boost converter by using an intelligent algorithm. The mathematical relationship between the input voltage, the output voltage, the output power, the period T and the duty cycle D is established according to the required working mode of the Boost converter, and a function is constructed with minimum current stress and minimum transmission error as the objectives. The optimal D and T are solved by using the classical algorithm Q-learning algorithm in reinforcement learning. The required working mode of the Boost converter is obtained by obtaining the optimal D and T, thereby further improving the working characteristics of the Boost converter, enhancing the practical applicability of the Boost converter, and making it have good anti-interference ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a control block diagram of a converter controlled by using a Q-learning algorithm according to Embodiment 1 of the present invention;

[0026] Figure 2 This is a specific process of combining the algorithm of Embodiment 1 of the present invention with the Boost converter;

[0027] Figure 3 is a working waveform diagram of the Boost converter of Example 1 of the present invention working in the CRM mode;

[0028] Figure 4 This is a working waveform diagram of the Boost converter according to Embodiment 1 of the present invention working in the DCM mode. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0030] Embodiment 1:

[0031] This embodiment provides a method for controlling the working mode of a Boost converter. Figure 1-Figure 4 As shown, the following steps are included:

[0032] S1, collect the input voltage Vin, output voltage Vout, and output current Io of the Boost converter, multiply the output voltage Vout and the output current Io to obtain the output power Pout of the load, and judge the working state of the Boost circuit through the above collected data to obtain the period T and duty cycle D.

[0033] S2, constructs the corresponding state space and action, constructs the objective function, determines the reward function, and defines the updated Q value table.

[0034] S3, using the Q-learning algorithm, according to different input parameters (Vin, Vout, Pout) with the minimum current stress and minimum transmission power error as the goal to solve the optimal D and T

[0035] S4, converting the optimal D, T obtained in step 3 into a driving signal to control the on and off of the switch tube so that the Boost converter is in the optimal working mode.

[0036] Furthermore, the specific operation process of step S1 is as follows:

[0037] S11: Collect the input voltage Vin, output voltage Vout, and output current Io of the Boost converter. According to the general principle of Boost, we can get

[0038]

[0039]

[0040] P out =V out I0 (3)

[0041] Step S12 can be obtained from step S11 that the circuit works in the CCM / CRM / DCM state. The value range of the period T of the circuit working in this mode is

[0042]

[0043] Step S13: From step S12, it can be obtained that the circuit works in the state of CCM / CRM / DCM. The duty cycle D of the circuit working in this mode is

[0044]

[0045] Furthermore, the specific operation process of step S2 is as follows:

[0046] Step S21 randomly samples Vin, Vout, and the calculated Pout to construct the corresponding state space defined as S, the action space defined as A, the reward function as r(s, a) and the Q value.

[0047] Step S22 For a boost converter, its environmental state usually consists of input voltage Vin, output voltage Vout and output power Pout. For a certain environmental state, the corresponding peak current Ip is determined by the converter's (D, T). Therefore, the state space S is defined as:

[0048] State space S = [D, T] (6)

[0049] Step S23 In the reinforcement learning method, the action at the previous moment determines the change of the current state. The process of solving the optimal modulation amount of the Boost converter using the Q-learning algorithm is essentially the process of reaching the optimal control state s* from the current state s through a certain action strategy. The change of the system depends on the change of D and T. The transition between states can be achieved by changing the values ​​of D and T. Therefore, the variable space CD and CT are used to quantify the value of state S.

[0050]

[0051] Where δ is the value state change of D and T.

[0052] Therefore, the system action space A is defined as:

[0053] A={C D ,C T} (8)

[0054] Step S24 The purpose of the Q-learning algorithm is to find the (D, T) corresponding to the minimum peak current, so the objective function g(x) can be defined as

[0055] g(x)=min[Ip] (9)

[0056] Where Ip is the current peak value calculated by the corresponding phase shift angle (D, T) in the current state. According to the working principle and operation mode of the Boost converter, the corresponding constraint condition can be defined as

[0057]

[0058] Pout is the transmission power calculated during the training process, p′ out is the expected transmission power, so the transmission power error is defined as η(D,T)

[0059] η(D,T)=|p′ out -p out | (11)

[0060] In order to obtain the minimum power error and the minimum peak current, the objective function G(D, T) is defined as

[0061] as follows

[0062] G(D,T)=βI p (D,T)+λη(D,T) (12)

[0063] Where β and λ are the corresponding penalty coefficients. For β value 1, a too small λ value will lead to a large power error, while a too large λ value will lead to poor current stress performance. Therefore, the trial and error method is adopted.

[0064] Step S25 can be evaluated based on step S24 by the objective function G. The smaller the value of the objective function G, the better the performance. In order to evaluate the quality of the selected action, a reward function r(s, a) is defined:

[0065]

[0066] In the formula, Gref is the reference value corresponding to the objective function G, and the value of Gref is greater than 0. Gmin is the minimum value corresponding to the objective function G, and the definition of Ip value of Gmin is the optimal state. ΔG represents the difference of the objective function of two adjacent states, and its expression is:

[0067] ΔG=G c -G p (14)

[0068] In the formula, Gc represents the loss in the current state, and GP represents the loss in the previous state. According to formula (14), the reward function of the system depends on the increment of the loss. When ΔG>0, it means that the peak current of the current state is greater than the peak current of the previous state. At this time, a negative reward value is given to the action, and the larger the increment, the smaller the reward value. When ΔG<0, it means that the corresponding peak current decreases after executing the action, and a positive reward value is given to the action. When the current stress of the system is less than the defined minimum loss value Gmin, a large reward value is given, indicating that the optimal state has been reached from the initial state.

[0069] Step S26 is based on the update of the Q value table and the selection of actions in step S25. The optimal strategy of Q-learning is determined by stepwise iteration. For strategy π, the Q value can be calculated by the following formula.

[0070]

[0071] In the formula, k represents the number of iterations. In fact, the state transition process of the agent can be regarded as a Markov decision process. Therefore, the update of the Q value table can be obtained by the following formula:

[0072]

[0073] Among them, α represents the learning rate, r represents the discount factor, and Q k (s, a) is the Q value under state s and action a.

[0074] In order to obtain the minimum current stress and minimum transmission power error D, T of the Boost converter, the action A = {CD, CT} is selected based on the ε-greedy strategy. When using as many exploration strategies as possible, try to explore new operation strategies and retain the optimal state under each strategy. Gmin of each stage of training is the lowest current stress and minimum power error state obtained in the previous stage of training.

[0075] Furthermore, the specific operation process of step S3 is as follows:

[0076] Step S31 After N times of learning, select the smallest G min The value is used as the intermediate parameter and the maximum Q value is used for action selection until the learned strategy converges, as shown in the following formula; then the optimal D and T are obtained.

[0077]

[0078] Furthermore, the specific operation process of step S4 is as follows:

[0079] Step S41 obtains the optimal D and T through step 3, and determines its working state to be CCM mode, CRM mode, or DCM mode, so that its Boost converter works in the optimal state of minimum current stress and minimum power error.

[0080] Draw the working waveform of the Boost converter when the working mode is CRM mode, such as Figure 3 As shown, the bottom waveform in the figure is the inductor current waveform. It can be seen from the inductor current waveform that in steady state, the inductor current and the horizontal axis have only one intersection point, and the minimum value is zero, indicating that the Boost converter works in CRM mode.

[0081] Draw the working waveform of the Boost converter when it works in DCM mode, such as Figure 4 As shown, the bottom waveform in the figure is the inductor current waveform. It can be seen from the inductor current waveform that in the steady state, the switch tube is still not turned on after the freewheeling of the inductor current ends, and there is a period of time when the inductor current is zero, indicating that the Boost converter operates in the DCM mode. The control method of the present invention can accurately control the Boost converter in the required operating mode, improve the operating characteristics of the Boost converter, and enhance the practical applicability and anti-interference ability of the Boost converter.

[0082] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for controlling the working mode of a Boost converter, characterized in that: The following steps are involved: S1, collecting the input voltage Vin, output voltage Vout, and output current Io of the Boost converter, multiplying the output voltage Vout and the output current Io to obtain the output power Pout of the load, judging the working state of the Boost circuit through the above collected data, and obtaining the period T and duty cycle D; S2, by using the period T and duty cycle D obtained in step 1 to construct the corresponding state space and action, construct the objective function, determine the reward function, and define the updated Q value table; S3, using the Q-learning algorithm, according to different input parameters (Vin, Vout, Pout), training is performed with the goal of minimum current stress and minimum power error to solve the optimal D and T; S4, converting the optimal D, T obtained in step 3 into a driving signal to control the on and off of the switch tube so that the Boost converter is in the optimal working mode.

2. The method for controlling the working mode of a Boost converter according to claim 1, characterized in that: The step S1 also includes: collecting the input voltage Vin, the output voltage Vout, and the output current Io of the Boost converter, obtaining the output power Pout of the load, and obtaining the period T and the duty cycle D.

3. The method for controlling the working mode of a Boost converter according to claim 1, characterized in that: The step S2 also includes: constructing the corresponding state space and action, constructing the objective function, determining the reward function and reconstructing the Q value table. The specific process is as follows Set up the state space, action space and state space defined as S, action space defined as A, reward function as r(s, a) and Q value S=[D,T] A={C D ,C T } 4. The method for controlling the working mode of a Boost converter according to claim 1, characterized in that: The S2 also includes: randomly sampling Vin, Vout, and calculated Pout to construct a corresponding state space defined as S, an action space defined as A, a reward function as r(s, a) and a Q value.

5. The method for controlling the working mode of a Boost converter according to claim 1, characterized in that: The S2 also includes: in the reinforcement learning method, the action at the previous moment determines the change of the state at the current moment. The process of using the Q-learning algorithm to solve the optimal modulation amount of the Boost converter is essentially a process of reaching the optimal control state s* from the current state s through a certain action strategy.

6. The method for controlling the working mode of a Boost converter according to claim 3, characterized in that: The S2 also includes: when ΔG>0, it means that the peak current of the current state is greater than the peak current of the previous state, and a negative reward value is given to the action, and the larger the increment, the smaller the reward value. When ΔG<0, it means that the corresponding peak current decreases after executing the action, and a positive reward value is given to the action. When the current stress of the system is less than the defined minimum loss value Gmin, a large reward value is given, indicating that the optimal state has been reached from the initial state.

7. The method for controlling the working mode of a Boost converter according to claim 3, characterized in that: The S2 also includes: when using as many exploration strategies as possible, trying to explore new operation strategies and retaining the optimal state under each strategy. The Gmin of each stage of training is the lowest current stress and minimum power error state obtained in the previous stage of training.

8. The method for controlling the working mode of a Boost converter according to claim 1, characterized in that: In S3, the Q-learning algorithm, a classic algorithm in reinforcement learning, is used to train and solve the optimal D and T according to different input parameters (Vin, Vout, Pout) with the minimum current stress and the minimum power error as the goal.

9. The method for controlling the working mode of a Boost converter according to claim 1, characterized in that: The S4 also includes: obtaining the optimal D and T through S3, determining its working state to be CCM mode, CRM mode, or DCM mode, so that its Boost converter works in the optimal state of minimum current stress and minimum power error.

Images