TWI708470B - A power converter - Google Patents

Abstract

A power converter has a power conversion circuit and a control unit, the control unit is used to perform a PWM operation on the power conversion circuit to convert a DC input voltage into a DC output voltage, wherein the control unit is A duty cycle of the PWM operation is generated by executing a PID algorithm, and is characterized by: a first setting value of a proportional term parameter of the PID algorithm, a second setting value of an integral term parameter, and a derivative term parameter The third setting value of is obtained by an external device using a circuit model corresponding to the power conversion circuit to execute a particle swarm algorithm, and the particle swarm algorithm is based on (the proportional term parameter, the integral term parameter, the The differential term parameter) is a combination of particle parameters, and in the case of optimizing the overstepping amount of the step response of the output voltage of the circuit model and the weighted value of the settling time as the goal, the value of the particle parameter combination is updated a predetermined Times to obtain a final particle parameter combination to generate the first set value, the second set value, and the third set value.

Description

A power converter

The present invention relates to a power converter, and in particular, to a power converter with a Proportion Integration Differentiation (PID) controller.

In recent years, switching power supplies have been widely used in industrial and consumer electronic products. The main purpose of switching power supplies is to provide stable output voltage to back-end electronic equipment. The voltage control mode is usually used to stabilize the output voltage. The method will be The error of the output voltage is compared with the reference voltage and used as the controller input. After the controller calculates, it outputs a Pulse Width Modulation (PWM) signal to the switch drive circuit to control the duty cycle of the main switch and regulate the output Voltage.

The performance of the switching power supply depends on the algorithm of the controller and the design of its control parameters. Therefore, many linear and non-linear control methods have been derived. However, the limitations of hardware make it difficult to implement some more complex control methods.

In contrast, Proportion Integration Differentiation (PID) controllers are easy to implement and robust, so they are widely used in various fields. However, if the parameters of the proportional integral derivative controller are set incorrectly, it may cause the overall system to be unstable, so the setting of the proportional integral derivative parameter value is extremely important.

Because the proportional integral derivative controller has the characteristics of simple structure and stable control, but there are problems of difficult adjustment and poor accuracy when adjusting parameters. The method of determining parameters used to rely on engineers to find the most suitable method through trial and error. Parameter value, this method is quite time-consuming and labor-intensive, and not economical.

Many experts have proposed the adjustment method of the parameter value of the proportional integral derivative controller. It can be roughly divided into the empirical adjustment method and the optimized parameter adjustment method. The empirical adjustment method is usually the Ziegler-Nichols method. The method is to calculate the proportional-integral-derivative controller parameters by applying the adjustment formula of the system oscillation period and oscillation gain. Although simple, it has poor accuracy and tends to be too large. There are many artificial intelligence algorithms to choose from, such as genetic algorithm (GA), particle swarm optimization (Particle Swarm OptimizationPSO), etc., to improve the controller. Performance.

Some literature proposes a frequency domain adjustment method, which builds a Bode diagram through a system mathematical model, and realizes a proportional integral derivative controller after compensation design. The parameter design of this method is more accurate, but the compensation is more troublesome and accurate mathematics must be established first. Model, otherwise the result will have a considerable gap with the actual circuit; there is also a literature that proposes ant colony optimization (ACO) to optimize parameters, which is based on the pheromone path left by ants in the biological world. Parameter combination, but the parameters represented by each path point have resolution problems, and it is impossible to determine whether there is a better path in the case of higher resolution, so it is not suitable for occasions that require high precision; another document proposes Genetic Algorithm (GA), which uses selection, crossover and mutation behavior in biological genes to evolve parameters to optimize them. This method must encode the optimization target, so it has high computational complexity It is not easy to realize the shortcomings, and premature convergence is easy to occur in the process, resulting in a situation of falling into a local optimal solution; more literature proposes the Cuckoo Search (CS), which uses the biological behavior of cuckoo bird parasitism and reproduction Carry out an optimized search, and search for regional and global solutions through the Levy flight pattern; another literature proposes a particle swarm optimization (PSO), which mimics the predation behavior of group animals Optimizing the target parameters, the calculation process is relatively simple, easy to implement, and can escape the local optimal solution to the global solution to search, so that the probability of finding the optimal solution is improved.

However, the above-mentioned documents are inadequate in the improvement of conversion efficiency and over-excessive amount. Therefore, a novel power converter is urgently needed in this field.

Proportional integral derivative (PID) controllers have been widely used in switching power supply control, and the parameters of the controller have a great impact on the performance of the controller, although there are simple and widely used controller parameter adjustments However, effective parameter adjustment methods have always been an important issue when applied in the industry.

This case presents the technology of particle swarm optimization (PSO) to adjust the parameters of the proportional integral derivative controller. The particle swarm algorithm is used to obtain the best ratio of the voltage control mode synchronous rectification flyback converter Examples of integral derivative controller parameters.

One purpose of the present invention is to disclose a power converter, which uses a synchronous rectification digital control flyback converter architecture, which has lower power consumption than conventional diode rectifier converters and can also improve the overall conversion effectiveness.

Another object of the present invention is to disclose a power converter, which adjusts the parameters of the proportional integral derivative controller by executing a particle swarm algorithm, and adds random variables to the particle movement process so that it has a chance to escape the best solution of the area (Local Optimal solution), as well as individual particles and particle groups have memory functions, in order to achieve simple calculations, easy implementation, low cost and other purposes.

Another object of the present invention is to disclose a power converter. In the simulation results, the settling time part is 65.11% less than the frequency domain compensation adjustment method of the conventional technology; the maximum overshoot part (0.15V) is also better than the conventional technology Frequency domain compensation adjustment method (2.27V) and the conventional Ziegler-Nichols adjustment method (36.58V).

Another purpose of the present invention is to disclose a power converter. In the actual test results, the settling time of the present invention is 71.83% less than the frequency domain compensation adjustment method of the prior art, and 61.8% less than the Ziegler-Nichols adjustment method of the prior art. ; The maximum overshoot part (1.78V) is better than the conventional frequency domain compensation adjustment method (5.74V) and the conventional technology Ziegler-Nichols adjustment method (14.39V).

To achieve the foregoing objective, a power converter is proposed, which has: a power conversion circuit and a control unit for performing a PWM operation on the power conversion circuit to convert a DC input voltage to a DC Output voltage, wherein the control unit generates a duty cycle of the PWM operation by executing a PID algorithm, and is characterized in that: the first setting value of a proportional term parameter of the PID algorithm and an integral term parameter The second setting value and the third setting value of a differential term parameter are obtained by an external device using a circuit model corresponding to the power conversion circuit to execute a particle swarm algorithm, and the particle swarm algorithm is based on (the ratio Term parameter, the integral term parameter, the derivative term parameter) is a combination of particle parameters, and when the goal is to optimize the step response of the output voltage of the circuit model and the weighted value of the settling time, The value of the particle parameter combination is updated a predetermined number of times to obtain a final particle parameter combination, thereby generating the first set value, the second set value, and the third set value.

In one embodiment, the particle swarm algorithm includes the following steps: randomly initializing the speed and position of a plurality of particles; and updating the position and the speed of each particle for the predetermined number of times according to the weighted value.

In an embodiment, the weighted value is expressed as:

Where F is the weighted value, M _p is the overrun amount, M _pmax is a maximum overrun amount, T _s is the settling time, T _smax is the highest settling time, α is a first weighting coefficient, and β is A second weight coefficient.

In an embodiment, the first weighting coefficient α and the second weighting coefficient β are both 10.

In one embodiment, the number of particles and the predetermined number are both 50.

In one embodiment, the calculation of the circuit model is implemented in Simulink.

In one embodiment, the particle swarm algorithm is implemented in Matlab.

In order to enable your reviewer to further understand the structure, features and purpose of the present invention, drawings and detailed descriptions of preferred specific embodiments are attached as follows.

100: power converter

110: power conversion circuit

120: control unit

200: External device

Step a: Initialize the speed and position of a plurality of particles randomly

Step b: Update the position and the velocity of each particle for the predetermined number of times according to the weighted value

FIG. 1a shows a block diagram of an embodiment of the power converter of the present invention.

FIG. 1b shows a block diagram of an embodiment of the particle swarm algorithm of the present invention.

Fig. 2 shows a block diagram of an embodiment of the proportional integral derivative control system.

Figure 3a illustrates the operation of a flyback converter circuit is a schematic diagram of _{the p-conducting} period of the switch Q.

Figure 3b illustrates the operation of the flyback converter circuit schematic of the switch during the OFF _p Q.

Figure 3c shows the voltage and current waveforms of the flyback converter in continuous current conduction mode.

Figure 4a shows a schematic diagram of the synchronous rectification flyback converter.

Figure 4b shows a waveform diagram of the synchronous rectification flyback converter operating in continuous conduction mode.

Figure 5 illustrates the concept of particles moving in space in the particle swarm algorithm.

Figure 6a shows a schematic diagram of the system control block.

Figure 6b shows the Bode diagram of the system before compensation.

Figure 6c shows the Bode diagram of the system after compensation.

Figure 6d shows a system step response diagram simulated by the frequency domain compensation adjustment method of the prior art.

Figure 6e shows the step response diagram of the system simulated by the Ziegler-Nichols adjustment method of the prior art.

Figure 6f shows the step response diagram of the system simulated by the present invention.

Figure 7a shows the measured performance of the conventional frequency domain compensation adjustment method in stable time.

Figure 7b shows the performance of the conventional Ziegler-Nichols adjustment method measured in stable time.

Figure 7c shows the measured performance of the present invention during stable time.

Figure 7d shows the maximum overshoot measurement measured by the frequency domain compensation adjustment method of the prior art.

Figure 7e illustrates the maximum overshoot measurement measured by the Ziegler-Nichols adjustment method of the conventional technology.

Figure 7f shows the measured maximum overrun measurement of the present invention.

FIG. 8 shows a comparison diagram of the measured efficiency of the present invention at three different input voltages (90V, 110V, 130V).

Please also refer to FIGS. 1a to 1b, in which FIG. 1a shows a block diagram of an embodiment of the power converter of the present invention, and FIG. 1b shows a block diagram of an embodiment of the particle swarm algorithm of the present invention.

As shown in FIG. 1a, the power converter 100 has a power conversion circuit 110 and a control unit 120.

The control unit 120 is used to perform a PWM operation on the power conversion circuit 110 to convert a DC input voltage into a DC output voltage, wherein the control unit 120 generates one of the PWM operations by executing a PID algorithm Responsibility cycle.

The first setting value of a proportional term parameter of the PID algorithm, the second setting value of an integral term parameter, and the third setting value of a derivative term parameter are used by an external device 200 using a circuit corresponding to the power conversion circuit The model is obtained by executing a particle swarm algorithm, and the particle swarm algorithm uses (the proportional term parameter, the integral term parameter, the derivative term parameter) as the particle parameter combination, and the output of the power conversion circuit 110 The weighted value of the overshoot of the step response of the voltage and the settling time When optimization is the goal, the value of the particle parameter combination is updated a predetermined number of times to obtain a final particle parameter combination, thereby generating the first set value, the second set value, and the third set value.

As shown in Figure 1b, the particle swarm algorithm includes the following steps: randomly initialize the speed and position of a plurality of particles; step a; and update the position and the speed of each particle according to the weighted value to reach the The predetermined number of times; step b.

The weighted value is expressed as:

Where F is the weighted value, M _p is the overrun amount, M _pmax is a maximum overrun amount, T _s is the settling time, T _smax is the highest settling time, α is a first weighting coefficient, and β is A second weight coefficient.

The first weighting coefficient α and the second weighting coefficient β are, for example, but not limited to 10; the number of particles and the predetermined number are, for example, but not limited to 50; the calculation of the circuit model is, for example, but not limited to, Simulink Realization; The particle swarm algorithm is, for example, but not limited to, realized by Matlab.

The principle of the present invention will be described below:

Please refer to FIG. 2, which shows a block diagram of an embodiment of a proportional integral derivative control system.

In the control system, the most commonly used control method is the proportional integral derivative control method. As shown in the figure, the proportional integral derivative controller is composed of a proportional unit (P), an integral unit (I) and a derivative unit (D). Determine its characteristics by adjusting the gain of the above-mentioned units.

The proportional integral derivative controller is a linear control method, and the control error e(t) can be obtained according to the given input r(t) value and the actual output value y (t) , as shown in equation (1).

e ( t ) = r ( t )- y ( t ) (1)

Proportional, integral and differential operations are performed on the error e(t) , and the results of the three operations are added together to obtain the control output u(t) . In the continuous time domain, the calculation formula is as shown in equation (2).

Among them, k _p is the proportional coefficient, T _i is the integral time constant, and T _d is the derivative time constant. The functions of each parameter are as follows:

1. Proportional term: proportionally reflect the error signal e(t) of the control system. Once an error occurs, the controller immediately produces a control effect to reduce the error.

2. Integral term: It is mainly used to eliminate steady-state errors to improve the accuracy of the system. The strength of the integral action depends on the integral time T _i . The larger the T _i , the weaker the integral action, and vice versa.

3. Differential term: It reflects the rate of change of the error signal, adjusts the differential output of the error, and can introduce an effective early correction signal into the system before the error signal becomes too large, thereby accelerating the speed of the system.

In the digital control system, sampling control is performed. The control quantity can only be calculated based on the error value at the sampling time. Therefore, the integral and differential terms in equation (2) cannot be used directly. Discretization is required. First, a series of sampling The time point kT represents the continuous time t of the equation (2), and the integration is replaced by the accumulation, and the differential is replaced by the increment, and the approximate conversion can be performed, as shown in the equation (3).

For the convenience of calculation, e(kT ) in equation (3) is simplified to e(k) , that is, T is omitted, as shown in equation (4).

，k ^d =k ^p T ^d ，u(k)為第k次取樣時控制器的輸出值，e(k)為第k次取樣時輸入控制系統的誤差值，e(k-1)為第(k-1)次取樣時輸入控制系統的誤差值， T為取樣週期。 Among them, k _p is the proportional coefficient, k _i is the integral coefficient, k _d is the differential coefficient, and

, K ^d = k ^p T ^d , u(k) is the output value of the controller at the kth sampling, e(k) is the error value input to the control system at the kth sampling, and e(k-1) is the (k-1) Input the error value of the control system during sampling, T is the sampling period.

Since each output is related to all past states, e(k) should be accumulated during calculation. This is not only complicated to calculate, but also takes up a lot of memory space. In order to improve this phenomenon, some literature proposes an algorithm for incremental proportional integral derivative control. This algorithm means that the output of the controller adopts the increment △u(k) of the control quantity.

The algorithm of incremental proportional integral differential control is derived from equation (4), as shown in equation (5).

Equation (5) can be obtained by subtracting equation (4) from equation (4).

△ u ( k ) = k _p [ e ( k )- e ( k -1)]+ k _i e ( k )+ k _d [ e ( k )-2 e ( k -1)+ e ( k -2 ))=( k _p + k _i + k _d ) e ( k )-( k _p +2 k _d ) e ( k -1)+ k _d e ( k -2) (6)

Since the general digital control system adopts a constant sampling period T , once k _p , k _i , and k _{d are} determined, the control increment can be obtained by equation (6) as long as the errors of the previous and next three measurements are used. Although incremental control only makes some improvements in the algorithm, it also adds the following advantages:

1. Since the processor only needs to process the output increment, the impact is small when there is an error, and it can be processed by logical judgment when necessary.

2. There is no need to accumulate in the formula. The value of the control increment △u(k) is only related to the last three sampling values, so it is easier to get a better control effect through weighting.

Please also refer to FIG. 3a. 3C, wherein Figure 3a which illustrates a flyback converter in the circuit operation period of the switch Q _p turned schematic, FIG. 3b which illustrates the flyback converter during a switch Q _p OFF of the operation of the circuit For a schematic diagram, Figure 3c shows the voltage and current waveforms of the flyback converter in continuous current conduction mode.

As shown in Figure 3a, when the power switch Q _{p is} turned on, there is a current passing through the magnetizing inductance L _m on the primary side of the transformer. At this time, the energy is stored in the magnetizing inductance L _m of the transformer, because the primary and secondary sides of the transformer have opposite polarities. The power diode D is reverse biased, and the energy required by the load will be supplied by the output capacitor C.

As shown in Figure 3b, when the power switch Q _p is turned off, the polarity of the voltage across the primary winding of the transformer is reversed. At this time, the power diode D is turned on, and the exciting current is mapped to the secondary side, which is originally stored in the magnetizing inductance of the transformer The energy of L _m is transferred to the output capacitor C and the load terminal through the power diode D.

As shown in Figure 3c, in the continuous conduction mode, when the power switch Q _{p is} turned on, from the power supply side, the input current i _p flows through the energy storage inductor of the primary winding, and the energy is stored in the energy storage inductor. There is a voltage drop, the input current i _p rises linearly, and the power diode D is reverse biased, so the secondary side is treated as an open circuit. At this time, the load energy is completely supplied by the output capacitor C , and the voltage of the output capacitor C will be reduce. Among them, △ _ip is the rate of change of the input current on the primary side of the transformer, △i _s is the rate of change of current on the secondary side, and V _D is the forward bias of the power diode.

When the power switch Q _{p is} turned on, the voltage of the energy storage inductor L _{m on} the primary side of the flyback converter is as shown in equation (7).

The rate of change of input current Δi _{Lm is} shown in equation (8).

來決定；當功率開關Q _p 截止時，電感上之電流必須連續，使得功率二極體D順向偏壓，且感應電流出現在二次側。跨在變壓器二次側繞組上之壓降為V _s =V _o +V _D 。功率開關Q _p 截止時二次側儲能電感L _S 之電壓如方程式(9)所示。 The slope of the input current change when the power switch Q _p is turned on is available

To decide; when the power switch Q _{p is} off, the current on the inductor must be continuous, so that the power diode D is forward biased, and the induced current appears on the secondary side. The voltage drop across the secondary winding of the transformer is V _s = V _o + V _D. When the power switch Q _p is turned off, the voltage of the secondary-side energy storage inductor L _S is shown in equation (9).

During the switch-off period, the input current i _in drops to zero, and Δi _{s is} shown in equation (10).

From the inductance volt-second balance, it can be known that the net change in current in one cycle is zero, as shown in equation (11).

Substituting t _on = DT _s and t _off = (1- D ) T _s , the equation (12) can be obtained.

。 Among them, n is the transformer turns ratio,

.

Please refer to FIGS. 4a to 4b together, in which FIG. 4a is a schematic diagram showing the structure of the synchronous rectification flyback converter, and FIG. 4b is a waveform diagram of the synchronous rectification flyback converter operating in continuous conduction mode.

As shown in Figure 4a, the output side rectification of the flyback converter usually uses a diode, while the synchronous rectification flyback converter uses a MOSFET with extremely low on-resistance to replace the rectifier diode to reduce the conduction loss of the diode. And improve the overall converter efficiency. Because MOSFET is a voltage-controlled component, when using MOSFET as a rectifier, the gate voltage must be synchronized with the phase of the original rectifier diode voltage to perform the rectification function. However, it is necessary to consider the gate charge Q _g of the MOSFET used to determine the applicable gate resistance. If the gate resistance value is too small, the synchronous rectification MOSFET may turn on first before the main switch Q _{p is} completely disconnected. A short circuit occurs, causing excessive current to pass. If the above situation occurs, the loss will increase and the benefit of using synchronous rectification will be lost.

The loss of synchronous rectification can be divided into two parts: switching gate loss P _g and conduction loss P _c , as shown in equations (13) and (14) respectively.

P _g = Q _g . V _g . f _s (13)

P _c = R _{DS ( on )} . I _rms . D (14)

Among them, Q _g is the C _gs charge of the MOSFET, V _g is the gate voltage, f _s is the switching frequency, R _{DS ( on )} is the on-resistance of the MOSFET, I is the drain current, and D is the duty cycle.

和

部分，二次側電流i _s 流過同步整流MOSFET本體二極體，且當一次側開關為導通狀態時同步整流MOSFET本體二極體將有反向恢復Q _PR 之功率損失產生，二次側同步整流之總消耗功率為

+

。 As shown in Figure 4b, the delay time

with

Part, the secondary side current i _s flows through the body diode of the synchronous rectification MOSFET, and when the primary side switch is turned on, the body diode of the synchronous rectification MOSFET will have reverse recovery Q _PR power loss, and the secondary side is synchronous The total power consumption of rectification is

+

.

)如方程式(15)所示。 Conduction loss (

) Is shown in equation (15).

Among them, R _{DS ( on )} is the on-resistance of the synchronous rectification MOSFET, I _O is the output current, △i _s is the secondary-side peak-to-peak filtering current, and V _D and I _D are the opposite of the body diode of the synchronous rectification MOSFET. To voltage and current, D is the duty cycle of primary side switch.

)係二極體因反向恢復Q _RR 而產生，如方程式(16)所示。 Power loss (

) Is a diode produced by the reverse recovery Q _RR , as shown in equation (16).

The body diode reverse voltage V _D of the synchronous rectification MOSFET is V _o +( V _i / n ).

The circuit design specifications of the synchronous rectification flyback converter used in the present invention are shown in Table 1.

The present invention uses the mathematical software MATLAB and the simulation software Simulink developed by the American Mathwork Company to implement the algorithm part, but is not limited to this. Among them, the system first uses Simulink to simulate the action of the hardware to establish a simulation circuit model, and then implements the particle swarm algorithm through MATLAB to determine the part of the model's proportional integral derivative controller parameter optimization.

Part of the particle swarm algorithm:

Particle Swarm Optimization (PSO) from Eberhart blog The optimization algorithm proposed by Dr. Shi and Dr. Kennedy in 1995, which originated from the study of bird predation behavior, uses the concept of biological instinct to share information in society, so that the group can get better results. This biological characteristic makes particles Find the best solution faster and more efficiently. The algorithm is based on swarm intelligence and has characteristics of biological evolution such as adaptive evaluation.

The movement of particles in the particle swarm algorithm refers to the current best solution of the individual and the group. It is similar to the calculation method of Genetic Algorithms (GA). The biggest difference is that the particle swarm algorithm adds random variables to the particle movement process. , So that the particles have a chance to escape the local optimal solution, and both the individual particle and the particle group have the memory function, the mathematical operation is relatively simple and easy to implement, and the operation cost is relatively low.

Fitness values in the solution space of each particle has a corresponding, and each particle know so far the best fitness value and the optimal position, it referred to the above-described characteristics of the individual particles of the optimum value (Particle best value, pBest ), each particle not only has its own best experience, but also knows the best fitness value and best position of all particles. This feature is called the Globle best value ( gBest ) of the particle. After each generation of renewal, the particle will use the experience of the individual particle and the experience of the particle group as a reference value to update the speed and position of the individual particle.

The particle swarm algorithm will randomly scatter particles in the regional solution at the beginning. If there is a particle close to the best solution in the region, the particles in the region will search near the best solution in the region. However, the regional optimal solution is limited to a local region, which does not mean that it is also the location of the global optimal solution. At this time, it is necessary to perturb the random random number so that the particles in the area have the opportunity to jump out of the whole domain and search for the best solution in the whole domain. The concept of particles moving in space in the particle swarm algorithm is shown in Figure 5. The particle swarm algorithm Based on equations (17) and (18) to find the best solution.

v _ij ( t +1) = w . v _ij ( t )+ C ₁ . rand ₁ . [ pBest _ij ( t )- x _ij ( t )]+ C ₂ . rand ₂ . [ gBest _ij ( t )- x _ij ( t )] (17)

x _ij ( t +1) = x _ij ( t )+ v _ij ( t +1) (18)

Where x _ij is the position of the particle, i is the number of particles, and j is the dimension of the particles; v _ij is the speed of the particles, i is the number of particles, and j is the dimension of the particles; the individual learning factor C ₁ is the learning of the particle itself Parameter, between 1 and 4, usually set to 2; group learning factor C ₂ is the mutual learning parameter of particles, between 1 and 4, usually set to 2; rand ₁ is between 0 and 1 random nonce; rand ₂ are random nonce 0 to 1; pBest optimum position of the representative individual particle; gBest optimum position represents groups; inertia weight value w represents a particle velocity of between 0 and 1 between.

Circuit simulation part:

The present invention uses Simulink to simulate various output dynamic behaviors of the flyback converter. Simulink is a simulation tool constructed in the Matlab environment to analyze and simulate the dynamic characteristics of the system. Simulink adopts the window method and forms a complete simulation system through the connection of graphical function blocks. The purpose is to complete the simulation analysis with a simple design process. The most important thing is that the parameter values simulated by Simulink can be returned to Matlab in the form of a matrix to achieve the purpose of combining circuit simulation and algorithm.

When the simulation starts, the output terminal of the circuit will transmit the voltage value back and compare it with the reference potential, and the error potential will be sent to the output pulse width modulation (Pulse Width Modulation, PWM) module after passing through the proportional integral derivative controller. The PWM waveform is switched to the power switch.

In addition, the To Workspace module can be used in the simulation to send the output voltage of each circuit to the MATLAB program in a matrix. The information in the matrix includes time and real-time circuit voltage values, etc. The algorithm performs a circuit simulation for each calculation. After the simulation results are generated, they can be sent back to the program in a matrix form, which is like building a communication bridge between the program and the circuit simulation to ensure that the values calculated by the algorithm can be used as the parameters of the circuit simulation.

Before using any optimization algorithm, it is necessary to correctly describe the optimization problem to be solved. The present invention aims to optimize the step response performance of the output voltage, which is related to the main performance parameters of the step response as follows:

1. Maximum percent overshoot: refers to the ratio of the difference between the maximum (first) peak value of the system unit step response and the expected value of the system to the steady-state output response value.

2. Peak time (Peak time): refers to the time required for the maximum (first) peak of the system unit step response.

3. Rise time: The time required for the unit step response of the system to change from 0 to 100% of the steady state value, or from 10% to 90%.

4. Settling time: The time required for the system unit step response to enter the range of ±5% or ±2% of the steady-state value and no longer leave this range. The present invention uses ±2% of the steady-state value as The result of settling time.

The three parameters of the proportional integral derivative controller will all affect the step response of the system. The proportional term K _p can make the response speed faster and slightly improve the steady-state error; the integral term K _i can improve the steady-state error, but over The large response peak will increase, and if it is too small, the response speed of the system will be slow; the derivative term K _d can improve the part of the transient response. After adding the differential controller, for the step response, the response of the system is at the initial instant There will be a large peak, and as time increases, the system response will decrease to zero.

In summary, the influence of the proportional integral derivative controller on the step response is shown in Table 2.

However, the present invention uses the excess amount and the settling time as the scoring items of the algorithm. The smaller the sum of them, the better the step response (minimization). Since the units of the two scoring items are different, the total will be directly added. Cause a weight problem. The fitness score equation is normalized (that is, the weighted value of the present invention) as shown in equation (19).

Among them, M _p is the maximum overrun, M _pmax is the highest maximum overrun in the other two comparative adjustment methods, T _s is the settling time, T _smax is the highest settling time in the other two adjustment methods, and α is the maximum overrun The weight coefficient of the quantity is set to 10, and the weight coefficient of β is the settling time is set to 10.

The environment settings of this experiment are as follows: The input source is Chroma's AC programmable power supply, the model is 61602, the power stage is a synchronous rectification flyback converter, and the load end uses Chroma's 63108A electronic load and operates. Regulate to 19V in constant voltage mode, The control level adopts the digital signal processor TMS320F28335 introduced by Texas Instrument Company, and the measurement of the measured waveform adopts the DPO4054 oscilloscope introduced by Tektronix Company.

The following compares the step response performance obtained by the particle swarm calculation adjustment method used in the present invention and the conventional Ziegler-Nichols parameter adjustment method and frequency domain compensation adjustment method.

The frequency domain compensation method uses Bode plots to show the phase margin (PM), gain margin (GM), bandwidth (BandWidth, BW) and crossover frequency (Crossover Frequency, f ) of the original system. _c ) Performance indicators in the frequency domain, and reuse loop control according to the conditions to be achieved to make the system performance more stable and less affected by external factors. Generally speaking, the system loop design conditions are as follows:

The crossover frequency f _{c of the} gain must be large enough to make the output of the power supply quickly return to stability in a transient state, but if f _{c is} too large, it may cause high-frequency noise interference. Therefore, f _c is generally set to be less than f _S /4 ~ f _S /20 in application, where f _{S is the} main switching frequency.

Generally speaking, the larger the phase margin, the better the stability of the system, which is less affected by component parameter changes. Ideally, the phase margin should be above 45° and the gain margin should be between 6-20dB.

Simulation results:

Please refer to Figures 6a to 6f together. Figure 6a shows the system control block diagram, Figure 6b shows the system Bode diagram before compensation, Figure 6c shows the system Bode diagram after compensation, and Figure 6d shows the system Bode diagram. The step response diagram of the system simulated by the frequency domain compensation adjustment method of the conventional technology is shown in Fig. 6e, which shows the step response diagram of the system simulated by the Ziegler-Nichols adjustment method of the conventional technology, and Fig. 6f shows the system simulated by the present invention. Step response graph.

As shown in Figure 6a, the principle is to obtain the feedback signal of the output voltage multiplied by the sampling attenuation frequency, compare it with the reference voltage command, and then generate an appropriate voltage command through the compensator C. H represents the attenuation ratio of the voltage divider resistor.

Substituting the circuit design specifications of the synchronous rectification flyback converter in Table 1 into the switching duty cycle vs. output voltage transfer function G _vd ( s ) of the hardware architecture to obtain the system Bode diagram and parameters of the Bode diagram before compensation in Figure 6b as shown in Table 3.

It can be seen from Table 3 that before compensation, the bandwidth, phase margin, and gain margin of G _vd ( s ) do not meet the system stability conditions. Therefore, the frequency domain compensation must be supplemented with a proportional integral derivative controller to meet the stated stability conditions. The parameters are shown in Table 4.

Compensate the transfer function G _vd ( s ) to obtain the compensated system Bode diagram in Figure 6c. The parameters of the compensated system Bode diagram are shown in Table 5. It can be seen that the gain margin, phase margin and The bandwidth meets the design considerations.

Substituting the parameter combinations in Table 4 into the Simulink simulation can obtain the system step response diagram of the frequency domain compensation adjustment method in Figure 6d.

Then use the Ziegler-Nichols adjustment method to adjust the parameters of the proportional integral derivative. First, adjust the K _p parameter in the circuit simulation until the system oscillates. Record the K _p value that makes the system oscillate as K _c , and the system oscillation period is T _c , The proportional integral derivative parameter combination of Ziegler-Nichols adjustment method is shown in Table 6, and the system step response diagram of Ziegler-Nichols adjustment method is shown in Figure 6e.

In the particle swarm algorithm used in the present invention, the number of generations is set to 50, the number of particles is set to 50, the learning factors C ₁ and C _{2 are} each set to 2, and the weight coefficient w is set to 0.9. Substituting the above parameters and optimizing the proportional, integral and differential parameters of the analog circuit, MATLAB records the results of each iteration of the objective function, and uses the best-performing set of proportional, integral and differential parameters as the maximum of the particle swarm algorithm To optimize the result, the combination of proportional, integral and differential parameters of the present invention is shown in Table 7, and the step response diagram of the system of the present invention is shown in Fig. 6f.

Table 7

The scoring standard items of the above three adjustment methods are settling time, the maximum amount of excess and the score substituted into the objective function are shown in Table 8.

It can be seen from the simulation performance of the three adjustment methods in Table 8:

In terms of settling time, the Ziegler-Nichols adjustment method is the fastest, and the frequency domain compensation adjustment method has the slowest performance; the Ziegler-Nichols adjustment method has a larger maximum excess amount, and the present invention has the smallest excess amount among the three; The smaller the score after the objective function, the better the performance, so it can be known that the present invention is the best in terms of simulation performance.

The proportional integral derivative parameter values of the above three adjustment methods are shown in Table 9.

results of testing:

Please also refer to Figures 7a to 7f, where Figure 7a shows the performance of the conventional frequency domain compensation adjustment method measured in the stable time, and Figure 7b shows the conventional technology Ziegler-Nichols adjustment method measured in the stable time Fig. 7c shows the measured performance of the present invention at a stable time, Fig. 7d shows the measured maximum excess measurement of the frequency domain compensation adjustment method of the conventional technique, and Fig. 7e shows one of the conventional techniques The maximum overrun measurement measured by the Ziegler-Nichols adjustment method. FIG. 7f illustrates the maximum overrun measurement measured by the present invention.

The actual measurement specifications of the above three adjustment methods are the same. The circuit specifications input is 110 V _AC , output voltage 19 V _DC , output current 3.5A, switching frequency is 100kHz, and it operates in continuous conduction mode. The scoring standard items of the above three adjustment methods are settling time, maximum overrun and the score substituted into the objective function as shown in Table 10.

From the measured performance of the three adjustment methods in Table 10, it can be seen that the present invention has the smallest maximum overrun among the three and the smallest score after being substituted into the objective function, so the measured performance is the best.

Next, perform an efficiency test for the hardware architecture of the present invention. The output load is increased from light load (1A) to heavy load (3.5A). The measurement data includes input voltage V _in and input power P _in , output current I _out and output voltage V _out , output power P _out and efficiency η are shown in Table 11.

Please refer to FIG. 8, which shows a comparison diagram of the measured efficiency of the present invention at three different input voltages (90V, 110V, 130V).

As shown in the figure, the hardware architecture of the present invention has an efficiency of 87% in the input voltage specification. Above, and the efficiency reaches more than 90% when the input voltage is 110V.

In summary, the simulation and actual performance evaluation of the Ziegler-Nichols adjustment method and the frequency domain compensation adjustment method of the present invention and the prior art show that the present invention performs well in both simulation and actual measurement.

With the design disclosed above, the present invention has the following advantages:

1. The present invention discloses a power converter, which uses a synchronous rectification digital control flyback converter architecture, which consumes less power than the conventional diode rectifier converter and can also improve the overall conversion efficiency.

2. The present invention discloses a power converter, which adjusts the parameters of the proportional-integral-derivative controller by executing a particle swarm algorithm, and adds random variables to the particle movement process to have a chance to escape from the local optimal solution (Local optimal solution) , And particle individuals and particle groups have memory functions to achieve simple calculations, easy implementation, low cost and other purposes.

3. The present invention discloses a power converter. In the simulation results, the settling time part is reduced by 65.11% compared with the frequency domain compensation adjustment method of the conventional technology; the maximum overshoot part (0.15V) is also better than the frequency domain compensation of the conventional technology The adjustment method (2.27V) and the Ziegler-Nichols adjustment method (36.58V) of the conventional technology.

4. The present invention discloses a power converter. In the actual test results, the settling time of the present invention is 71.83% less than the conventional frequency domain compensation adjustment method and 61.8% less than the conventional Ziegler-Nichols adjustment method; The excess part (1.78V) is better than the frequency domain compensation adjustment method (5.74V) of the conventional technology and the Ziegler-Nichols adjustment method (14.39V) of the conventional technology.

The disclosure of the present invention is a preferred embodiment, and any partial changes or modifications that are derived from the technical idea of the present invention and can be easily inferred by those familiar with the art will not depart from the scope of the patent right of the present invention.

In summary, no matter the purpose, means, and effects of the present invention, it is showing its technical characteristics that are very different from the conventional ones, and its first invention is suitable for practical use, and it is also in compliance with the patent requirements of the invention. I sincerely ask your examiner to observe it carefully, and Pray that the patent will be granted as soon as possible to benefit the society.

100: power converter

110: power conversion circuit

120: control unit

200: External device

Claims (7)

A power converter has a power conversion circuit and a control unit, the control unit is used to perform a PWM operation on the power conversion circuit to convert a DC input voltage into a DC output voltage, wherein the control unit is A duty cycle of the PWM operation is generated by executing a PID algorithm, and is characterized by: The first setting value of a proportional term parameter of the PID algorithm, the second setting value of an integral term parameter, and the third setting value of a derivative term parameter are used by an external device using a circuit model corresponding to the power conversion circuit It is obtained by executing a particle swarm algorithm, and the particle swarm algorithm uses (the proportional term parameter, the integral term parameter, and the derivative term parameter) as the particle parameter combination, and is used to make the output voltage of the circuit model step In the case of optimizing the weighted value of the first-order response and the settling time as the goal, the value of the particle parameter combination is updated a predetermined number of times to obtain a final particle parameter combination, thereby generating the first set value and the second set value. The set value and the third set value. In the power converter described in item 1 of the scope of patent application, the particle swarm algorithm includes the following steps: Initialize the speed and position of a plurality of particles randomly; and Updating the position and the velocity of each particle for the predetermined number of times according to the weighted value. For the power converter described in item 2 of the scope of patent application, the weighted value is expressed as:

Where F is the weighted value, M _p is the overrun amount, M _pmax is a maximum overrun amount, T _s is the settling time, T _smax is a maximum settling time,

Is a first weight coefficient and

Is a second weight coefficient. In the power converter described in item 3 of the scope of patent application, the first weighting coefficient α and the second weighting coefficient β are both 10. For the power converter described in item 2 of the scope of patent application, the number of particles and the predetermined number are both 50. In the power converter described in the first item of the scope of patent application, the calculation of the circuit model is realized by Simulink. The power converter described in the first item of the scope of patent application, wherein the particle swarm algorithm is implemented by Matlab.

Images