A photovoltaic maximum power point tracking method predicted by a multi-step quantum Lie group model

Abstract

The present invention proposes a photovoltaic maximum power point tracking method based on a multi-step quantum Lie group model prediction method. The main steps of the multi-step quantum Lie group model prediction method are: (1) measuring the meteorological conditions and electrical quantities of the photovoltaic array; (2) calculating the maximum power point of the photovoltaic module using a quantum Lie group network; (3) estimating the photovoltaic maximum power point reference value using the photovoltaic array maximum power point calculation model; (4) predicting the photovoltaic system state quantities at multiple future moments using a multi-step prediction model; (5) calculating the optimal control sequence using a two-stage optimization function; and (6) outputting the optimal control quantity. The proposed photovoltaic maximum power point tracking method can balance dynamic performance and steady-state accuracy, achieve optimal control function, optimize the photovoltaic maximum power point tracking process, and improve the stability of the photovoltaic maximum power point tracking system.

Description

Technical Field

[0001] The present invention belongs to the fields of quantum computing, artificial intelligence, machine learning, new energy and photovoltaic power generation control, and relates to a method for multi-step quantum Lie group model prediction, which is suitable for photovoltaic maximum power point tracking. Background Art

[0002] Existing photovoltaic maximum power point tracking (PVMPT) methods struggle to balance dynamic performance and steady-state accuracy. Most PVMPT methods focus on searching for the maximum power point but neglect research into optimizing the tracking process. A few PVMPT methods focus on optimizing the tracking process, but their dynamic performance is affected by the maximum power point reference value.

[0003] Furthermore, while existing model predictive control systems have the potential to optimize dynamic processes, their reference values ​​are mostly derived from perturbation-observation or conductance increment methods, which limits dynamic performance and prevents accurate multi-step predictions. Deep neural networks offer excellent fitting properties, enabling rapid and accurate estimation of the maximum power point reference value for photovoltaic modules and can also be used directly to estimate the reference value for the maximum power point of photovoltaic arrays. However, their operation is complex and inefficient.

[0004] Therefore, a multi-step quantum Lie group model prediction method is proposed. This multi-step prediction model is combined with a quantum Lie group network to solve the multi-step accurate prediction of photovoltaic maximum power point tracking control systems, improving dynamic performance and steady-state accuracy. The method of calculating the photovoltaic maximum power point using a quantum Lie group network can simplify operations and improve efficiency. Summary of the Invention

[0005] A photovoltaic maximum power point tracking method based on a multi-step quantum Lie group model prediction method uses the implicit transcendental equation of photovoltaic modules to construct a dynamic conductance model, achieving high fitting accuracy and real-time updating of the calculated dynamic conductance value. The stable operation of the photovoltaic system is ensured by adding bias feedback correction. The multi-step prediction model is combined with a quantum Lie group network to reduce the calculation complexity of the maximum power point reference value of the prediction model and improve the accuracy of the calculated reference value. The global and continuous tracking capabilities of the photovoltaic maximum power point are enhanced, enabling rapid maximum power point search. A two-stage optimization method is used to achieve rapid tracking of the maximum power point reference value and optimization of the maximum power point power. The steps in the use process are as follows:

[0006] Step (1): Using the dynamic conductivity model as the discrete prediction model for photovoltaic modules, the photovoltaic module current I PV (k+1) is:

[0007] I PV (k+1)=G PV (k)(V PV (k+1)-VPV (k))+I PV (k) (1)

[0008] Among them, V PV (k) and I PV (k) is the voltage and current of the photovoltaic module at step k; V PV (k+1) and I PV (k+1) is the PV module voltage and current at step k+1; G PV (k) is the dynamic conductivity of the photovoltaic module at step k;

[0009] Step (2): The dynamic conductivity model is used to predict the next component current. The calculated value of the dynamic conductivity model is updated as the predicted value changes. On the basis of retaining the form of the dynamic conductivity model, the dynamic conductivity calculation formula of the photovoltaic module is derived according to the implicit transcendental equation of the photovoltaic module, thereby incorporating the electrical characteristics of the photovoltaic module into the dynamic conductivity model. The dynamic conductivity G of the photovoltaic module in the kth step is PV (k) is:

[0010]

[0011] Among them, R sh is the shunt resistor; R s is the series resistance; I s is the diode saturation current; V d and V T It is the intermediate variable between DC voltage and battery temperature voltage; V d (k)=V PV (k)+I PV (k)·R s ; V T =n D K b T / q;n D is the diode quality factor; K b is the Boltzmann constant; T is the battery temperature; q is the electron charge; exp() is the exponential function with e as the base;

[0012] Step (3): Formula (1) and Formula (2) together constitute the explicit discrete model of the photovoltaic module. The photovoltaic array discrete model is established by using the explicit discrete model and the series-parallel topology of the photovoltaic array. The photovoltaic array current I of the k+1th step of the constructed photovoltaic array discrete model is array (k+1) is:

[0013] I array (k+1)=G array (k)(V array (k+1)-V array (k))+I array(k) (3)

[0014] Among them, I array (k) is the photovoltaic array current at step k; V array (k) is the photovoltaic array voltage at step k; G array (k) is the dynamic conductance of the photovoltaic array at step k; I array (k+1) is the photovoltaic array current at step k+1; V array (k+1) is the PV array voltage at step k+1;

[0015] Step (4): Construct a discrete model of the Boost converter. The specific model is:

[0016]

[0017] Among them, I L (k) is the inductor current at step k; I L (k+1) is the inductor current in the k+1th step; S is the switching signal of the switching element; S=1 means the switch is on; S=0 means the switch is closed; T s is the time of one cycle; L is the inductor in the Boost converter; C1 is the capacitor in the Boost converter; V DC is the equivalent DC source voltage;

[0018] Step (5): The photovoltaic system discrete prediction model constructed by combining the photovoltaic array discrete model and the Boost converter discrete model is:

[0019]

[0020] Among them, G array (k) is the dynamic conductance of the photovoltaic array at step k;

[0021] Step (6): The matrix form of the k-th step of the discrete single-step prediction model of the photovoltaic system is:

[0022]

[0023] in,

[0024] x(k)=[I L (k) I array (k) V array (k)] T ;y(k)=[I array (k) V array (k)] T ; [] T is the transpose of the matrix; x(k) is the x at step k; x(k+1) is the x at step k+1; y(k) is the y at step k; S(k) is the state of the switching signal S of the switching element in step k;

[0025] Step (7): The multi-step prediction model further transformed from the discrete single-step prediction model of the photovoltaic system is:

[0026]

[0027] Wherein, m is the total number of prediction steps; i and k are the number of steps; x(k+1), x(k+m-1) and x(k+m) are the x of the k+1th step, k+m-1th step and k+mth step respectively; x(i+1), x(i+m-1) and x(i+m) are the x of the i+1th step, i+m-1th step and i+mth step respectively; y(k+1), y(k+m-1) and y(k+m) are the y of the k+1th step, k+m-1th step and k+mth step respectively; A(1), A(m-1) and A(m) are the parameters A in the prediction model matrix equations of the 1st step, m-1th step and mth step respectively; b(1), b(2) and b(m) are the parameters b in the prediction model matrix equations of the 1st step, 2nd step and mth step respectively;

[0028] Step (8): To ensure the robustness of the system, a deviation feedback correction link is added to prevent large deviations in the predicted value. The model for correcting the predicted value through an increment is:

[0029]

[0030] Wherein, ΔI(k) is the deviation between the measured current and the predicted current at step k; ΔV(k) is the deviation between the measured voltage and the predicted voltage at step k; is the PV array current correction value at step k+1; is the k+1th step PV array voltage correction value;

[0031] Step (9): using a quantum Lie group network to calculate the maximum power point reference value of the photovoltaic module, which is further used to estimate the maximum power point reference value of the entire photovoltaic array;

[0032] Take n groups of simulations under different light intensities and temperatures for the photovoltaic system to obtain the voltage and current corresponding to the maximum power point of the photovoltaic system; use the light intensity and temperature of the photovoltaic system as the input data q of the quantum neural network in (h)=(E(h),W(h)), the voltage and current of the corresponding photovoltaic maximum power point are used as output data q out (h)=(I MPP (h),V MPP (h)); the quantum neural network computing model of group h is:

[0033] (I MPP(h),V MPP (h))=QL(E(h),W(h)) (9)

[0034] Among them, I MPP (h) is the current at the maximum power point of the hth group; V MPP (h) is the voltage at the maximum power point of the h-th group; E(h) is the light intensity of the photovoltaic of the h-th group; W(h) is the temperature of the photovoltaic of the h-th group; QL() is the quantum Lie group network;

[0035] Step (10): The quantum Lie group network is composed of a quantum tensor neural network and a Lie group network. The quantum tensor neural network is a type of quantum neural network with a multi-layer hierarchical structure of a classical deep neural network. It can provide an efficient approximate expression of quantum states and simplify the description of quantum circuits. The quantum neural network with a tensor structure can distinguish tasks and generate tasks. The quantum tensor neural network is a quantum circuit with a fixed tree structure. The Lie group network is introduced to update parameters. The Lie group network is a method of updating parameters based on gradient descent in meta-learning. It is divided into two layers of inner loop and outer loop. The outer loop corresponds to a gradient update process of machine learning, and the inner loop is the process of learning new types of tasks based on a small amount of new types of data. The quantum tensor neural network is composed of quantum bits, quantum gates and quantum measurements.

[0036] Step (11): Input n groups of data q in (n) Convert quantum bit encoding into n sets of quantum data Encode quantum data into n quantum bit product states, and Apply a single-qubit rotation gate to realize the input quantum state |x q Preparation of >, input quantum state |x q >For:

[0037]

[0038] Among them, |x q > is the quantum state of a composite system consisting of n quantum bits; is the Kronecker product; cos() is the cosine function; sin() is the sine function; x in (1), x in (2) and x in (n) are the input data of the neural network of the discriminant model The first, second and nth data; and are the first, second and nth quantum bit states respectively; π is the circumference of a circle;

[0039] Step (12): Input quantum state |x q>Apply unitary transformation to obtain the output quantum state |x qout >, perform quantum measurement on some output quantum states to obtain the output result x out (h), output the result x out (h) The measurement is classical bit data; the output data q is obtained using the Lie group network out (h) and the output result x out (h) Compare and update the training parameters of the unitary transformation according to the error until the error meets the accuracy requirement;

[0040] Step (13): The training parameters of the unitary transformation are updated using the inner loop of the Lie group network. q >divided into training data τ tra and validation data τ qry , the training parameter θ′ after the inner loop update of the Lie group network is:

[0041]

[0042] Among them, θ and Φ are the training parameters of the unitary transformation; θ′ is the parameter γ after the inner loop update of the Lie group network; α is the learning rate during the inner loop update process; For the training data τ tra The loss function f [Φ,θ] is the output data value after the Lie group network parameters are updated; is the differential with respect to the parameter θ; is the transpose of the differential of parameter θ;

[0043] Step (14): Use the verification data τ qry Verify the performance of the training parameters θ′ and generate the loss function By continuously reducing the output data q out (h) and the output result x out (h) deviation, and stop training until the accuracy requirement is met; after the photovoltaic maximum power point reference value is calculated by the quantum Lie group network, it is tracked as the control target of the photovoltaic maximum power point tracking system;

[0044] Step (15): Set up a two-stage optimization method, which is divided into the first stage of quadratic optimization function optimization tracking process and the second stage of power optimization function optimization power stability according to different control objectives; when the photovoltaic maximum power point tracking system is in the process of tracking the maximum power point reference value, it is the first stage optimization, and the quadratic optimization function model is used; the quadratic optimization function value g1 is:

[0045]

[0046] Among them, g1 is the value of the quadratic optimization function; is the sum of n data;I,i is the weight coefficient of current; λ V,i is the voltage weight coefficient; is the maximum power point current reference value; is the maximum power point voltage reference value; is the PV array current correction value at step k+i; is the k+i-th step PV array voltage correction value;

[0047] Step (16): The control quantity is the switch signal corresponding to the minimum quadratic optimization function value. The control quantity K1(k+i) of the k+i step is:

[0048] K1(k+i)=f(min{g1}) (13)

[0049] Where K1(k+i) is the control quantity of the k+i step; g1 is the value of the quadratic optimization function; min{} is the minimum value; f(min{g1}) is the output value of the minimum quadratic optimization function value;

[0050] Step (17): After tracking the reference value, the control target is converted to power optimization, and the second stage of power optimization begins. The value of the power optimization function g2 is:

[0051]

[0052] Where g2 is the value of the power optimization function;

[0053] Step (18): The control quantity of the second stage is the switch signal corresponding to the maximum power optimization function value. The control quantity K2(k+i) of the k+i step is:

[0054] K2(k+i)=f(max{g2}) (15)

[0055] Where K2(k+i) is the control quantity of the k+i step; g2 is the value of the quadratic optimization function; max{} is the minimum value; f(max{g2}) is the output value of the maximum power optimization function value;

[0056] Step (19): The switching condition between the quadratic optimization function and the power optimization function is whether the photovoltaic array operating point reaches the maximum power point reference value. The model for judging by the deviation between the photovoltaic array operating point voltage and current and the reference voltage and current is:

[0057]

[0058] Wherein, σ is the deviation discriminant value of the switching condition between the quadratic optimization function and the power optimization function;

[0059] Step (20): Through the multi-step quantum Lie group model prediction method, the photovoltaic maximum power point tracking controller outputs the optimal pulse width modulation signal to change the state of the switching element, so that the output power of the photovoltaic array gradually approaches the maximum power point, and finally achieves the function of outputting the maximum power of the photovoltaic array.

[0060] The present invention has the following advantages and effects compared to the prior art:

[0061] (1) The existing prediction process of photovoltaic maximum power point reference value using deep neural network is complex and computationally inefficient. However, the use of quantum Lie group network to predict photovoltaic maximum power point reference value can improve computational efficiency and reduce computational complexity.

[0062] (2) Existing training parameter update methods using deep neural networks cannot adapt quickly to new data and require multiple iterations to ensure accuracy. However, using Lie group networks to update the parameters of quantum tensor neural networks can enhance the learning speed and accuracy of quantum Lie group networks for photovoltaic voltage and photovoltaic current characteristics under different light intensities and temperatures, achieving accurate and rapid prediction of the photovoltaic maximum power point. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a framework diagram of the photovoltaic maximum power point tracking system of the method of the present invention.

[0064] Figure 2 This is a flow chart of the multi-step quantum Lie group model prediction method of the present invention. DETAILED DESCRIPTION

[0065] The present invention proposes a photovoltaic maximum power point tracking method predicted by a multi-step quantum Lie group model, which is described in detail with reference to the accompanying drawings as follows:

[0066] Figure 1 This is a framework diagram of the photovoltaic maximum power point tracking system of the method of the present invention.

[0067] First, a photovoltaic maximum power point tracking (MPPT) control system is constructed, consisting of a photovoltaic array, a maximum power point tracking (MPPT) controller, and a boost converter. The photovoltaic array consists of several photovoltaic modules connected in series and parallel. The boost converter is composed of an inductor L, capacitors C1 and C2, a diode D, and a switching element Q. The MPPT controller comprises six components: meteorological condition measurement, a multi-step prediction model, feedback correction, a quantum Lie group network, calculation of the PV array's maximum power point, and a two-stage optimization process. The output voltage and output current at both ends of the photovoltaic array are measured as input data for the MPPT controller. The MPPT reference values ​​of the photovoltaic modules are calculated using the quantum Lie group network based on the measured light intensity and temperature. The MPPT reference values ​​of the photovoltaic modules are then used to calculate the MPPT reference values ​​of the photovoltaic array. The multi-step prediction model is used to predict the future state of the photovoltaic system at multiple moments. Feedback correction of the predicted values ​​is used to reduce the deviation between the predicted and actual values. Finally, the optimal control variable is obtained through two-stage optimization tracking. This is output to the switching element Q to control the voltage of the boost converter, thereby regulating the maximum output power of the photovoltaic array.

[0068] Figure 2 This is a flow chart of the multi-step quantum Lie group model prediction method of the present invention.

[0069] First, the voltage, current, light intensity, and temperature of the photovoltaic array are measured. Then, a quantum Lie group network is used to calculate the maximum power point reference value. A multi-step quantum Lie group model prediction model is then used to calculate the future state of the photovoltaic maximum power point tracking control system. Feedback correction is used to calculate a correction to the predicted value and adjust the deviation. Finally, a determination is made as to whether the current output power of the photovoltaic array has reached the maximum power point reference value. If not, a quadratic optimization function is used to calculate the optimal control variable and output it until the output power of the photovoltaic array reaches the maximum power point reference value. If so, the power optimization function is used to optimize the maximum power output of the photovoltaic array.

[0070] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A photovoltaic maximum power point tracking method predicted by a multi-step quantum Lie group model, characterized in that: The dynamic conductance model is constructed using implicit transcendental equations of photovoltaic modules, achieving high fitting accuracy and real-time updating of the calculated dynamic conductance value. The stable operation of the photovoltaic system is ensured by adding bias feedback correction. The combination of a multi-step prediction model and a quantum Lie group network reduces the computational complexity of the maximum power point reference value of the prediction model and improves the accuracy of the calculated reference value. Improve the global and continuous tracking capabilities of the photovoltaic maximum power point, and achieve rapid maximum power point search; use a two-stage optimization method to achieve rapid tracking of the maximum power point reference value and optimize the maximum power point power; the steps in the use process are: Step (1): Using the dynamic conductivity model as the discrete prediction model for photovoltaic modules, the photovoltaic module current I PV (k+1) is: I PV (k+1)=G PV (k)(V PV (k+1)-V PV (k))+I PV (k) (1) Among them, V PV (k) and I PV (k) is the voltage and current of the photovoltaic module at step k; V PV (k+1) and I PV (k+1) is the PV module voltage and current at step k+1; G PV (k) is the dynamic conductivity of the photovoltaic module at step k; Step (2): The dynamic conductivity model is used to predict the next component current. The calculated value of the dynamic conductivity model is updated as the predicted value changes. On the basis of retaining the form of the dynamic conductivity model, the dynamic conductivity calculation formula of the photovoltaic module is derived according to the implicit transcendental equation of the photovoltaic module, thereby incorporating the electrical characteristics of the photovoltaic module into the dynamic conductivity model. The dynamic conductivity G of the photovoltaic module in the kth step is PV (k) is: Among them, R sh is the shunt resistor; R s is the series resistance; I s is the diode saturation current; V d and V T It is the intermediate variable between DC voltage and battery temperature voltage; V d (k)=V PV (k)+I PV (k)·R s ; V T =n D K b T / q;n D is the diode quality factor; K b is the Boltzmann constant; T is the battery temperature; q is the electron charge; exp() is the exponential function with e as the base; Step (3): Formula (1) and Formula (2) together constitute the explicit discrete model of the photovoltaic module. The photovoltaic array discrete model is established by using the explicit discrete model and the series-parallel topology of the photovoltaic array. The photovoltaic array current I of the k+1th step of the constructed photovoltaic array discrete model is array (k+1) is: I array (k+1)=G array (k)(V array (k+1)-V array (k))+I array (k) (3) Among them, I array (k) is the photovoltaic array current at step k; V array (k) is the photovoltaic array voltage at step k; G array (k) is the dynamic conductance of the photovoltaic array at step k; I array (k+1) is the photovoltaic array current at step k+1; V array (k+1) is the PV array voltage at step k+1; Step (4): Construct a discrete model of the Boost converter. The specific model is: Among them, I L (k) is the inductor current at step k; I L (k+1) is the inductor current in the k+1th step; S is the switching signal of the switching element; S=1 means the switch is on; S=0 means the switch is closed; T s is the time of one cycle; L is the inductor in the Boost converter; C1 is the capacitor in the Boost converter; V DC is the equivalent DC source voltage; Step (5): The photovoltaic system discrete prediction model constructed by combining the photovoltaic array discrete model and the Boost converter discrete model is: Among them, G array (k) is the dynamic conductance of the photovoltaic array at step k; Step (6): The matrix form of the k-th step of the discrete single-step prediction model of the photovoltaic system is: in, x(k)=[I L (k) I array (k) V array (k)] T ;y(k)=[I array (k) V array (k)] T ; [] T is the transpose of the matrix; x(k) is the x at step k; x(k+1) is the x at step k+1; y(k) is the y at step k; S(k) is the state of the switching signal S of the switching element in step k; Step (7): The multi-step prediction model further transformed from the discrete single-step prediction model of the photovoltaic system is: Wherein, m is the total number of prediction steps; i and k are the number of steps; x(k+1), x(k+m-1) and x(k+m) are the x of the k+1th step, k+m-1th step and k+mth step respectively; x(i+1), x(i+m-1) and x(i+m) are the x of the i+1th step, i+m-1th step and i+mth step respectively; y(k+1), y(k+m-1) and y(k+m) are the y of the k+1th step, k+m-1th step and k+mth step respectively; A(1), A(m-1) and A(m) are the parameters A in the prediction model matrix equations of the 1st step, m-1th step and mth step respectively; b(1), b(2) and b(m) are the parameters b in the prediction model matrix equations of the 1st step, 2nd step and mth step respectively; Step (8): To ensure the robustness of the system, a deviation feedback correction link is added to prevent large deviations in the predicted value. The model for correcting the predicted value through an increment is: Wherein, ΔI(k) is the deviation between the measured current and the predicted current at step k; ΔV(k) is the deviation between the measured voltage and the predicted voltage at step k; is the PV array current correction value at step k+1; is the k+1th step PV array voltage correction value; Step (9): using a quantum Lie group network to calculate the maximum power point reference value of the photovoltaic module, which is further used to estimate the maximum power point reference value of the entire photovoltaic array; Take n groups of simulations under different light intensities and temperatures for the photovoltaic system to obtain the voltage and current corresponding to the maximum power point of the photovoltaic system; use the light intensity and temperature of the photovoltaic system as the input data q of the quantum neural network in (h)=(E(h), W(h)), the voltage and current of the corresponding photovoltaic maximum power point are used as output data g out (h)=(I MPP (h), V MPP (h)); the quantum neural network computing model of group h is: (I MPP (h),V MPP (h))=QL(E(h),W(h)) (9) Among them, I MPP (h) is the current at the maximum power point of the hth group; V MPP (h) is the voltage at the maximum power point of the h-th group; E(h) is the light intensity of the photovoltaic of the h-th group; W(h) is the temperature of the photovoltaic of the h-th group; QL() is the quantum Lie group network; Step (10): The quantum Lie group network is composed of a quantum tensor neural network and a Lie group network. The quantum tensor neural network is a type of quantum neural network with a multi-layer hierarchical structure of a classical deep neural network. It can provide an efficient approximate expression of quantum states and simplify the description of quantum circuits. The quantum neural network with a tensor structure can distinguish tasks and generate tasks. The quantum tensor neural network is a quantum circuit with a fixed tree structure. The Lie group network is introduced to update parameters. The Lie group network is a method of updating parameters based on gradient descent in meta-learning. It is divided into two layers of inner loop and outer loop. The outer loop corresponds to a gradient update process of machine learning, and the inner loop is the process of learning new types of tasks based on a small amount of new types of data. The quantum tensor neural network is composed of quantum bits, quantum gates and quantum measurements. Step (11): Input n groups of data g in (n) Convert quantum bit encoding into n sets of quantum data Encode quantum data into n quantum bit product states, and Apply a single-qubit rotation gate to realize the input quantum state |x q Preparation of >, input quantum state |x q >For: Among them, |x q > is the quantum state of a composite system consisting of n quantum bits; is the Kronecker product; cos() is the cosine function; sin() is the sine function; x in (1), x in (2) and x in (n) are the input data of the neural network of the discriminant model The first, second and nth data; and are the first, second and nth quantum bit states respectively; π is the circumference of a circle; Step (12): Input quantum state |x q >Apply unitary transformation to obtain the output quantum state |x qout >, perform quantum measurement on some output quantum states to obtain the output result x out (h), output the result x out (h) The measurement is classical bit data; the output data q is obtained using the Lie group network out (h) and the output result x out (h) Compare and update the training parameters of the unitary transformation according to the error until the error meets the accuracy requirement; Step (13): The training parameters of the unitary transformation are updated using the inner loop of the Lie group network. q >divided into training data τ tra and validation data τ qry , the training parameter θ′ after the inner loop update of the Lie group network is: Among them, θ and Φ are the training parameters of the unitary transformation; θ′ is the parameter θ after the inner loop update of the Lie group network; α is the learning rate during the inner loop update process; For the training data τ tra The loss function f [Φ,θ] is the output data value after the Lie group network parameters are updated; is the differential with respect to the parameter θ; is the transpose of the differential of parameter θ; Step (14): Use the verification data τ qry Verify the performance of the training parameters θ′ and generate the loss function By continuously reducing the output data q out (h) and the output result x out (h) deviation, and stop training until the accuracy requirement is met; after the photovoltaic maximum power point reference value is calculated by the quantum Lie group network, it is tracked as the control target of the photovoltaic maximum power point tracking system; Step (15): Set up a two-stage optimization method, which is divided into the first stage of quadratic optimization function optimization tracking process and the second stage of power optimization function optimization power stability according to different control objectives; when the photovoltaic maximum power point tracking system is in the process of tracking the maximum power point reference value, it is the first stage optimization, and the quadratic optimization function model is used; the quadratic optimization function value g1 is: Among them, g1 is the value of the quadratic optimization function; is the sum of n data; I,i is the weight coefficient of current; λ V,i is the voltage weight coefficient; is the maximum power point current reference value; is the maximum power point voltage reference value; is the PV array current correction value at step k+i; is the k+i-th step PV array voltage correction value; Step (16): The control quantity is the switch signal corresponding to the minimum quadratic optimization function value. The control quantity K1(k+i) of the k+i step is: K1(k+i)=f(min{g1}) (13) Where K1(k+i) is the control quantity of the k+i step; g1 is the value of the quadratic optimization function; min{} is the minimum value; f(min{g1}) is the output value of the minimum quadratic optimization function value; Step (17): After tracking the reference value, the control target is converted to power optimization, and the second stage of power optimization begins. The value of the power optimization function g2 is: Where g2 is the value of the power optimization function; Step (18): The control quantity of the second stage is the switch signal corresponding to the maximum power optimization function value. The control quantity K2(k+i) of the k+i step is: K2(k+i)=f(max{g2}) (15) Where K2(k+i) is the control quantity of the k+i step; g2 is the value of the quadratic optimization function; max{} is the minimum value; f(max{g2}) is the output value of the maximum power optimization function value; Step (19): The switching condition between the quadratic optimization function and the power optimization function is whether the photovoltaic array operating point reaches the maximum power point reference value. The model for judging by the deviation between the photovoltaic array operating point voltage and current and the reference voltage and current is: Wherein, σ is the deviation discriminant value of the switching condition between the quadratic optimization function and the power optimization function; Step (20): Through the multi-step quantum Lie group model prediction method, the photovoltaic maximum power point tracking controller outputs the optimal pulse width modulation signal to change the state of the switching element, so that the output power of the photovoltaic array gradually approaches the maximum power point, and finally achieves the function of outputting the maximum power of the photovoltaic array.

Images