Disclosure of Invention
Aiming at the defects of the existing control algorithm, the invention provides an improved finite control set model prediction control method which is suitable for a battery energy storage converter comprising an LCL type passive filter, can reduce the number of sensors and improve the system performance and the network access current quality.
The technical scheme of the invention is to provide a limited control set model prediction control method of an LCL type energy storage converter, which is characterized in that based on a discretized LCL model, a first relation between estimated values of three state variables at the time k, the switching state of a switching tube in the LCL type energy storage converter at the time k and the estimated values of the three state variables at the time k +1 is deduced;
deducing estimated values of the three state variables at the moment of k +1, switch states at the moment of k +1 and a second relation between the estimated values of the three state variables at the moment of k +2 on the basis of the discretized LCL model;
after the estimated value of the first state variable at the moment k is made to approach the measured value of the first state variable at the moment k through correction, estimating according to the measured value of the first state variable at the moment k to obtain estimated values of three state variables;
obtaining estimated values of the three state variables at the k +1 moment based on the first relation through estimated values of the three state variables at the k moment and the switch state at the k moment which is obtained at the k-1 moment;
adopting a time delay compensation algorithm, obtaining predicted values of three state variables corresponding to the k +2 moment and the combinations respectively based on a second relation according to all different combinations of the switch states existing at the k +1 moment, selecting the predicted value with the minimum error with reference values of the three state variables from the predicted values of the three state variables at the k +2 moment, and taking the switch state corresponding to the predicted value with the minimum error as the optimal switch state at the k +1 moment;
the three state variables are converter side current, capacitor voltage and power grid current of the LCL type energy storage converter; the first state variable is any one of the three state variables that is currently measured.
Optionally, the discretized LCL model is:
x(k+1)=A1x(k)+B1vi(k)+B2vg(k) (2)
in a battery storage converter including an LCL filter, x ═ i1i2uc]Representing three state variables, corresponding to converter side current, power grid current and capacitor voltage; v. ofiRepresenting converter output voltage vgRepresenting the grid voltage, TsRepresents a sampling time; l is1Is a converter side inductance value, L2The inductance on the filter side and the capacitance on the filter side.
Optionally, the first relationship is represented as
The second relationship is represented as
Wherein, the 'Λ' above each parameter represents the estimated value of the corresponding parameter; v. of
gαβ(k) Is the sampled power grid voltage to obtain
v
ioptThe optimal converter output voltage at the moment k is a known value obtained at the moment k-1 and corresponds to a switching state at the moment k; v. of
i(k +1) is the converter output voltage vector, corresponding to all the different combinations of switching states that exist at time k + 1.
Optionally, an error between the measured value of the first state variable and the estimated value of the first state variable at the time k is added to the discretized LCL model through a correction component obtained by correcting the matrix L to obtain a "corrected estimated LCL model", so as to obtain estimated values of the three state variables;
the corrected estimated LCL model is:
wherein the output y corresponds to a first state variable, CcIs an output matrix; the lambda above each parameter represents the estimated value of the corresponding parameter; v. ofioptThe optimal converter output voltage at the moment k is a known value obtained at the moment k-1, and is obtained by adopting a finite control set model to predict and control the obtained optimal converter output voltage.
Optionally, the process of obtaining the correction matrix L includes the following processes:
the form of the "corrected estimated LCL model" is changed to:
the characteristic polynomial is: (z) ═ zl- (a)1-LCc)|;
Obtaining a correction matrix L by finding a desired pole;
for the estimator, the desired pole is p1、p2、p3The desired characteristic equation is:
f*(z)=(z-p1)(z-p2)(z-p3) (5)
therefore, | zI- (A)1-LCc)|=(z-p1)(z-p2)(z-p3);
Let the characteristic equation in the continuous domain corresponding to the desired characteristic equation in the discrete domain be
Thereby deducing:
wherein, ω isorFor undamped oscillation angular frequency, ζorFor optimum damping ratio, αodIs a non-dominant pole.
Optionally, undamped oscillation angular frequency ωorTaking half of resonant frequency of LCL filter, optimal damping ratio zetaor0.707, non-dominant pole αodIs 5 omegaor~10ωor。
Optionally, a delay compensation algorithm is adopted, and the method further comprises the following processes:
based on three state variables at estimated time k
And the sampled grid voltage v
gαβ(k) The optimal converter output voltage vector v found at the time k-1
iopt(k) And calculating the predicted values of the three state variables at the k +1 moment:
solving the grid voltage at the k +1 moment;
according to 7 different converter output voltage vectors v existing at the moment of k +1i(k +1), calculating the predicted values of the three state variables at the time k + 2:
constructing a cost function:
wherein λ is a weight coefficient; epsiloni1α,εi1βRespectively, errors between a reference value and an estimated value of the current at the converter side under the static coordinate system; epsiloni2α,εi2βRespectively, errors between a reference value and an estimated value of the power grid current under a static coordinate system; epsilonucα,εucβRespectively the error between the reference value and the estimated value of the capacitor voltage under the static coordinate system;
and selecting the converter output voltage vector when the cost function obtains the minimum value as the optimal converter output voltage vector according to the constructed cost function, so as to calculate the optimal driving signal corresponding to the k +1 moment at the k moment to control the on and off of the switching tube.
Optionally, the first state variable is a grid current;
acquiring reference values of three state variables, further comprising the following processes:
step 1, collecting and obtaining instantaneous values i of a phase a and a phase b of power grid current and power grid voltage2a,i2b,vga,vgb;
Step 2, calculating instantaneous values i of the grid current and the grid voltage in the phase c2c,vgc;
Step 3, the current i of the power grid is measured2a,i2b,i2cAnd the network voltage vga,vgb,vgcObtaining the power grid current i under a static coordinate system through the claick transformation2α,i2βAnd a network voltage component vgα,vgβ;
Step 4, giving a grid current reference value i under a rotating coordinate system2dref,i2qrefGenerating a grid current reference value i under a static coordinate system through coordinate transformation2αref,i2βref;
Step 5, obtaining the power grid current i under the alpha beta coordinate axis at the moment k
2αβGrid current reference value
Network voltage v
gαβComplex vector form of (d):
vgαβ(k)=vgα(k)+j×vgβ(k) (13)
step 6, according to the grid current reference value i under the static coordinate system at the moment k2αref,i2βrefCalculating a converter side current reference value and a capacitor voltage reference value under a static coordinate system at the moment k;
wherein, ω, L2C is the angular frequency of the power grid, the inductance value of the side of the filter grid and the capacitance value of the filter respectively;
step 7, calculating the reference values of the three state variables at the time k to the time k +2 by adopting a Lagrange extrapolation method to obtain the reference values of the three state variables at the time k + 2;
wherein x represents the converter-side current i1Grid current i2And the capacitor voltage uc。
The invention discloses an improved finite control set model predictive control method for a battery energy storage converter comprising an LCL filter. Aiming at the adverse effects caused by excessive sensors and unconsidered calculation delay, a method combining state variable estimation and delay compensation is provided, the current at the side of a converter, the capacitor voltage and the power grid current are estimated by sampling the power grid current, the error between the sampled power grid current and the estimated power grid current passes through a correction matrix, so that the influence caused by model mismatch and parameter drift is reduced, the estimated state variable passes through a limited control set model prediction control algorithm with a delay compensation link, so that the system performance is improved, and the control of the LCL type energy storage converter is finally realized. The improved algorithm can reduce the number of sensors, reduce the cost and improve the reliability of the system; and the influence of the calculation delay on the system performance is eliminated by combining a delay compensation algorithm, and the network access current quality is improved.
Detailed Description
The invention discloses an improved finite control set model predictive control method for a battery energy storage converter comprising an LCL filter.
State variable estimation
The first aspect of the present invention is to estimate the converter side current, the capacitor voltage and the grid current based on the measured grid current, and the specific estimation method is shown in fig. 1. Firstly, the LCL model is converted into a state space equation, and then the state space equation is discretized to obtain a discretized LCL model.
x=[i1i2uc]Represents three state variables, viRepresenting converter output voltage vgRepresenting the grid voltage, TsRepresents a sampling time; the state space equation:
wherein the content of the first and second substances,
B=[1/
L 1 0 0]
T,B
d=[0 -1/L
2 0]
T,
discretized LCL model:
x(k+1)=A1x(k)+B1vi(k)+B2vg(k) (2)
since the variable sampled in this example is the grid voltage, the output y is the grid current i2Output matrix Cc=[0 1 0]. To reduce estimation errors due to model mismatch and parameter drift, a correction matrix L ═ L is introduced1 l2 l3]. As can be seen from fig. 1, when there is a deviation between the measured grid current and the estimated grid current, the component obtained by correcting the deviation by the matrix is added to the original estimation system. Thus, the corrected estimateThe calculated LCL model can be written as:
wherein v is
ioptThe method is used for obtaining the optimal converter output voltage by adopting the finite control set model predictive control. In other examples, another two state variables i may be used instead
1、u
cOne kind of (1). That is, it is determined which of the three state variables is measured, and the output matrix C corresponding to the state variable is obtained
c(i.e. C)
cThe position of the medium value "1" corresponds to the measured state variable) so that the output is made
Is an estimate corresponding to the measured state variable.
As is known from equation (3), in order to obtain the estimated state variables, the correction matrix L must be first found.
The above equation can be written as follows:
the characteristic polynomial is: (z) ═ zl- (a)1-LCc)|
Engineering desires that the estimate be able to quickly approximate the measured value, which requires that the poles of the estimator reach the desired poles, which can be achieved by adjusting the correction matrix.
Assume that the desired pole is p1、p2、p3Then the estimator expects the characteristic equation to be:
f*(z)=(z-p1)(z-p2)(z-p3) (5)
therefore, | zI- (A)1-LCc)|=(z-p1)(z-p2)(z-p3)。
From the above equation, to find the correction matrix L, only three desired poles need to be known.
Since the points in the discrete domain correspond one-to-one to the points in the continuous domain, it is sufficient to find three points in the continuous domain that correspond to the desired poles in the discrete domain. Assuming that the characteristic equation in the continuous domain corresponding to the desired characteristic equation in the discrete domain is
This can be deduced:
wherein the undamped oscillation angular frequency omegaorThe optimum damping ratio Zeta should be about half of the resonant frequency of LCL filteror0.707, non-dominant pole αodAbout 5 to 10 omegaor。
Delay compensation algorithm
The second aspect of the present invention is to provide a delay compensation algorithm based on the three state variables estimated from the first aspect. The specific content of the algorithm is as follows: the method predicts one more step on the basis of predicting one step by the traditional control algorithm, so that the switching tube driving signal required by the current moment is calculated from the previous moment instead of the current moment, and the influence of calculation delay on control is eliminated. As shown in fig. 2, taking the grid current as an example, at the time k, since the switching tube driving signal at the time is already calculated from the time k-1, the grid current value at the time k +1 can be directly calculated by the formula (2). Then 7 different power grid current values at the k +2 moment are obtained according to 7 different switch states at the k +1 moment
![Figure BDA0002197488270000081](https://patentimages.storage.googleapis.com/c3/b3/1d/5d64aec2b25be5/BDA0002197488270000081.png)
And selecting the value with the minimum error with the grid current reference value, wherein the corresponding switching state is the optimal switching state. By adopting the method, the on-off state at the moment k +1 can be calculated at the moment k, and the influence of calculating the time delay can be eliminated by analogy.
The invention will be described in detail below with reference to a three-phase two-level LCL type tank converter as an example, as shown in fig. 3. U shapedcRepresenting the DC voltage, v, across the batteryiRepresenting the output voltage, i, of the phase-leg of a three-phase two-level circuit1Representing three-phase current, i, flowing through the inductance of the converter side2Representing the current into the network, i.e. the three-phase current, u, flowing through the inductance of the network sidecIs the voltage across the capacitor, vgIs the grid voltage.
The invention relates to an improved finite control set model predictive control method, which comprises the following steps:
step 1, sampling and conditioning electrical physical quantity in a circuit, performing analog-to-digital conversion on the sampled physical quantity, then performing conversion and numerical processing on numerical values obtained by an analog-to-digital conversion module in a controller, and acquiring instantaneous values i of two phases of power grid current and power grid voltage a and b2a,i2b,vga,vgb;
Step 2, calculating instantaneous values i of the grid current and the grid voltage in the phase c2c,vgc;
i2c=-(i2a+i2b) (7)
vgc=-(vga+vgb) (8)
Step 3, the sampled and calculated power grid current i2a,i2b,i2cAnd the network voltage vga,vgb,vgcObtaining the power grid current i under a static coordinate system through the claick transformation2α,i2βAnd a network voltage component vgα,vgβWherein the transformation matrix (constant amplitude transformation) is
Step 4, giving a grid current reference value i under a rotating coordinate system2dref,i2qrefGenerating a grid current reference value i under a static coordinate system through coordinate transformation2αref,i2βrefWhich isThe transformation matrix adopted in is
Where θ is the Phase Locked Loop (PLL) output.
Step 5, in order to simplify the operation, the complex vector is used for calculation, and assuming that each variable is obtained by sampling at the time k, the complex vector is expressed as follows:
i2αβ(k)=i2α(k)+j×i2β(k) (11)
vgαβ(k)=vgα(k)+j×vgβ(k) (13)
wherein the content of the first and second substances,
the power grid current, the power grid current reference value and the power grid voltage under the alpha beta coordinate axis are respectively.
Step 6, obtaining a grid current reference value i under a static coordinate system2αref,i2βrefCalculating a converter side current reference value and a capacitor voltage reference value under the coordinate system through a formula (14);
wherein, ω, L2And C is the angular frequency of the power grid, the inductance value of the side of the filter grid and the capacitance value of the filter.
Step 7, calculating the reference values of the three state variables at the time k to the time k +2 by adopting a Lagrange extrapolation method as shown in a formula (15), and obtaining the reference values of the three state variables at the time k + 2;
wherein x represents the converter-side current i1αβGrid current i2αβAnd the capacitor voltage ucαβ。
Step 8, the power grid current i obtained in the step 5 is processed
2αβAnd the power grid current obtained by the state variable estimation method
Make a difference (C in the example of FIG. 1)
c=[0 1 0]Corresponding to
Namely the estimated value of the grid current at the moment k
The specific value of the corrected LCL model can be obtained), the obtained current error is corrected by a correction matrix L to obtain a correction component, and then the correction component is added into the estimated LCL model to obtain the corrected estimated LCL model (formula 3); after obtaining the correction matrix L (equations 4-6), the converter side current can be estimated
Voltage of capacitor
And the current of the power grid
The specific numerical value of (1).
Step 9, adopting a delay compensation algorithm;
step 9.1, first, based on the three estimated state variables at time k
And the sampled grid voltage v
gαβ(k) The optimal converter output voltage vector v found at the time k-1
iopt(k) Determining the prediction of the time of the three state variables k +1A value;
9.2, solving the grid voltage at the moment k + 1;
step 9.3, outputting voltage vectors v according to 7 different converters existing at the moment of k +1
i(k +1) calculating the predicted values of the three state variables at the time of k +2
Since the converter has 6 switches, the voltage vector v is obtained by 8 combination modes and 2 combination modesiAre identical and therefore have 7 different viBy substituting it into equation (18), 7 different predicted values can be obtained. That is, each state variable has a different predicted value corresponding to a different converter output voltage vector, and thus each state variable has 7 different predicted values.
Step 10, constructing a cost function;
wherein λ is a weight coefficient; epsilon is the error between the reference and estimated values of the state variable
That is to say that the first and second electrodes,
and step 11, according to the constructed cost function, selecting the converter output voltage vector when the cost function obtains the minimum value as the optimal converter output voltage vector, so as to calculate the corresponding optimal driving signal to control the switching tube to be switched on and off (namely, the optimal converter output voltage vector at the moment k +1 is calculated at the moment k).
The grid-connected converter can stably work according to the implementation of the steps. To verify the analysis, the present embodiment uses an experimental platform to perform actual measurement, and the obtained waveforms are shown in fig. 4 to 7, fig. 4 to 6 respectively show the relationship between the estimated values and the measured values of the grid current, the inverter-side current and the capacitor voltage (the waveforms corresponding to symbols 41,51,61 are the measured values, the waveforms corresponding to symbols 42,52,62 are the estimated values, and the waveforms corresponding to symbols 43,53,63 are the errors therebetween), please note that: these three measurements are only used for comparison with the estimates and not in the control. As can be seen from these three figures, the errors between the estimated values and the measured values are relatively small (the waveforms corresponding to the symbols 41 and 42, 51 and 52, 61 and 62 are substantially coincident), so that it can be proved that the method of the present invention for controlling by using the estimated values instead of the measured values has no problem. Fig. 7 shows the grid current obtained by the control method provided by the present invention, and it can be seen from the graph that the sine degree of the grid current is very high, and it can be seen that the grid-connected converter controlled by the method works normally to meet the grid-connected requirement, thereby further explaining the effectiveness of the control method provided by the present invention.
The invention can accurately estimate the non-sampled state variable by using the state variable estimation method (experiments prove that the estimated value can not be displayed by an oscilloscope, so the estimated value is displayed by a control esk interface connected with dSPACE), the estimated variable is adopted to replace the sampled variable (inverter side current, capacitance voltage and power grid current) in the traditional prediction control method, two sensors (at least 4 sensors) for sampling the inverter side current and the capacitance voltage can be saved, only the sensors for sampling the power grid current a and the power grid current b are needed to be reserved, the sensor cost is greatly reduced, the control complexity is reduced, and the reliability of the system is improved (when the sensor fails, the method can be used as a standby method). In addition, the delay compensation algorithm provided by the invention improves the quality of the network access current.
While the present invention has been described in detail with reference to the preferred embodiments, it should be understood that the above description should not be taken as limiting the invention. Various modifications and alterations to this invention will become apparent to those skilled in the art upon reading the foregoing description. Accordingly, the scope of the invention should be determined from the following claims.