A self-correcting photovoltaic grid-connected multi-stage current conversion control method, device and equipment

Abstract

The present invention discloses a self-correcting photovoltaic grid-connected multi-stage current conversion control method, device and equipment, and relates to the field of new energy current conversion control technology. First, the difference between the actual value of the controlled target and the corresponding reference value is determined, and the controlled target includes the voltage at both ends of the photovoltaic component, the DC link voltage of the inverter and the grid-connected current. Then, each difference is input into the current conversion controller based on the artificial neural network to obtain a control signal, and then the control signal is used to control the front-stage boost circuit and the rear-stage inverter to obtain the updated controlled target parameter value, and then the control signal output by the current conversion controller is input into the parameter estimator based on the artificial neural network, and the photovoltaic grid-connected system is fitted by the parameter estimator, and the feedback is fed back to the current conversion controller to update the weight parameters of the current conversion controller, and the weight parameters of the current conversion controller are adaptively adjusted through multiple rounds of iterations to achieve rapid tracking of the controlled target. The present invention simplifies the control process and improves the control accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy current conversion control, and in particular to a self-correcting photovoltaic grid-connected multi-stage current conversion control method, device and equipment. Background Art

[0002] At present, with the continuous growth of global energy demand and the increasing attention to environmental protection, photovoltaic power generation systems have become one of the key means to solve energy problems with their clean, sustainable and renewable advantages. Generally speaking, photovoltaic grid-connected systems are composed of photovoltaic modules, boost circuits, inverters, filters and power grids, and the system structure is relatively fixed. However, its line impedance, filter inductance and other parameters are variable, making it difficult to accurately model them. It is a typical nonlinear system with fixed structure and variable parameters. Since the DC voltage generated by the photovoltaic power generation system is usually unstable and low, it cannot directly meet the grid connection requirements. It usually needs to be increased to a stable and grid-connected voltage level with the help of a boost circuit, and then the boosted DC power is converted into AC power through an inverter to achieve efficient grid connection of the photovoltaic power generation system. As the core equipment for photovoltaic grid connection, the control effect of the converter (including boost circuit and inverter) is crucial for the reliable and stable grid connection of the photovoltaic power generation system.

[0003] For the photovoltaic grid-connected system, the controlled targets need to achieve the following effects: 1. Control the voltage across the photovoltaic module to the voltage corresponding to its maximum power point to ensure the maximum power generation of the photovoltaic module; 2. Control the DC link voltage of the inverter to the reference voltage to ensure the power balance and stable operation of the photovoltaic grid-connected system; 3. Control the grid-connected current to maintain the reference current to ensure that the power generated by the photovoltaic module is stably connected to the grid. Due to the large number of controlled targets, if you want to achieve the ideal photovoltaic grid-connected control effect, the three control signals generated by the converter controller must ensure that the three controlled targets are achieved at the same time.

[0004] In the prior art, power converter controllers are divided into linear controllers and nonlinear controllers. For the former, proportional-integral (PI) controllers and proportional resonance (PR) controllers are the most commonly used, which are simple to implement and highly universal, but the controller parameters are constant and the dynamic performance is poor. For nonlinear controllers, the performance of most nonlinear controllers is highly dependent on the accuracy of the built system model. Once the built model has deviations in the description of the system, the nonlinear controller may become unstable. In addition, since designers cannot accurately perceive parameters such as grid impedance and filter impedance, it is difficult to accurately model the system in practice.

[0005] In an ideal grid-connected control strategy, the grid-connected current must be consistent with the reference current. The reference current is determined by the amplitude and phase. Usually, the amplitude is derived from the required power, while the phase is synchronized with the phase of the grid voltage. In terms of obtaining the grid voltage phase, the phase-locked loop is most commonly used.

[0006] In summary, most of the current photovoltaic grid-connected control methods are implemented using a phase-locked loop and three traditional controllers (mainly PI controllers and PR controllers). The front stage uses a controller to control the boost circuit, that is, to control the voltage across the photovoltaic module, and the back stage uses two controllers to control the voltage outer loop and the current inner loop respectively. In the traditional photovoltaic grid-connected control architecture, the back stage control depends on the time scale (the response speed of the current inner loop must be faster than the response speed of the voltage outer loop), and all three controllers require manual parameter adjustment. The photovoltaic grid-connected control process is complicated and the control accuracy is poor. Summary of the invention

[0007] Based on this, it is necessary to provide a self-correcting photovoltaic grid-connected multi-stage current conversion control method, device and equipment to address the above technical problems.

[0008] The present invention adopts the following technical solutions:

[0009] The present invention provides a self-correcting photovoltaic grid-connected multi-stage current conversion control method, which first determines the difference between the actual value of the controlled target and the corresponding reference value, and the controlled target includes the voltage at both ends of the photovoltaic component, the inverter DC link voltage and the grid-connected current. Then, each difference is input into a current conversion controller constructed by an artificial neural network to obtain a control signal including a modulation signal of a front-stage boost circuit, a grid-connected reference current amplitude and a modulation signal of a rear-stage inverter. Then, the front-stage boost circuit is controlled by the modulation signal of the front-stage boost circuit, and the rear-stage inverter is controlled by the modulation signal of the rear-stage inverter to obtain an updated The voltage across both ends of the PV module, the DC link voltage of the inverter and the grid-connected current are calculated, and then the control signal output by the converter controller is input into the parameter estimator constructed by the artificial neural network, and the control update process of the control signal is fitted by the parameter estimator; the weight parameters in the parameter estimator are fed back to the converter controller to update the weight parameters of the converter controller, and the weight parameters of the converter controller are adaptively adjusted through multiple rounds of iterations, so that the voltage across both ends of the PV module, the DC link voltage of the inverter and the grid-connected current can quickly track the corresponding reference values ​​respectively, and the self-correcting photovoltaic grid-connected multi-level converter control is completed.

[0010] The present invention provides a self-correcting photovoltaic grid-connected multi-stage current conversion control device, comprising:

[0011] A difference determination module is used to determine a photovoltaic module voltage difference between a reference voltage at both ends of a photovoltaic module of a photovoltaic grid-connected system and a voltage at both ends of the photovoltaic module, an inverter DC link voltage difference between an inverter DC link reference voltage of the photovoltaic grid-connected system and an inverter DC link voltage, and a grid current difference between a grid-connected reference current and a grid-connected current of the photovoltaic grid-connected system;

[0012] A control signal determination module is used to input the PV module voltage difference, the inverter DC link voltage difference and the grid-connected current difference into a current converter controller constructed by an artificial neural network to obtain a control signal including a modulation signal of a front-stage boost circuit, a grid-connected reference current amplitude and a modulation signal of a rear-stage inverter;

[0013] A control module is used to control the front-stage boost circuit through the modulation signal of the front-stage boost circuit, and control the rear-stage inverter through the modulation signal of the rear-stage inverter, so as to obtain the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current;

[0014] A parameter estimation module is used to input the control signal output by the current converter controller into a parameter estimator constructed by an artificial neural network, and to fit the control update process of the control signal based on the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current;

[0015] The iterative tracking module is used to feed back the weight parameters in the parameter estimator to the converter controller to update the weight parameters of the converter controller; through multiple rounds of iterative adaptive adjustment of the weight parameters of the converter controller, the voltage at both ends of the photovoltaic module, the inverter DC link voltage and the grid-connected current are made to track the reference voltage at both ends of the photovoltaic module, the inverter DC link reference voltage and the grid-connected current respectively, thereby completing the self-correcting photovoltaic grid-connected multi-level converter control.

[0016] The present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the self-correcting photovoltaic grid-connected multi-stage current conversion control method is implemented.

[0017] The present invention provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned self-correcting photovoltaic grid-connected multi-stage current conversion control method when executing the program.

[0018] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects:

[0019] The present invention determines the control signal of photovoltaic grid-connected multi-level current conversion control according to the difference between the controlled target (the voltage at both ends of the photovoltaic module of the photovoltaic grid-connected system, the DC link voltage of the inverter and the grid-connected current) and the corresponding reference value through the current conversion controller based on artificial neural network, and quickly fits the control update process of the control signal with the parameter estimator constructed by artificial neural network, so as to dynamically and real-time adjust the weight parameters in the current conversion controller by using the weight parameters of the parameter estimator, so that the photovoltaic grid-connected system can be accurately modeled, thereby obtaining an accurate control signal. The centralized control and self-correction control of the front-stage boost circuit and the rear-stage inverter are realized, the control process is simplified, and the control accuracy is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0021] Figure 1 A schematic flow chart of a self-correcting photovoltaic grid-connected multi-stage current conversion control method provided by the present invention;

[0022] Figure 2 A schematic diagram of a photovoltaic grid-connected system provided by the present invention;

[0023] Figure 3 A schematic diagram of a parameter estimator network architecture provided by the present invention;

[0024] Figure 4 A schematic diagram of a network architecture of a current converter controller provided by the present invention;

[0025] Figure 5 A schematic diagram of a photovoltaic grid-connected multi-stage current conversion control based on full artificial neural network self-correction control provided by the present invention;

[0026] Figure 6 A schematic diagram of a self-correcting photovoltaic grid-connected multi-stage current conversion control device provided by the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] The technical solutions provided by various embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0029] Figure 1 The present invention is a self-correcting photovoltaic grid-connected multi-stage current conversion control method flow diagram, which specifically includes the following steps:

[0030] S101: Determine a photovoltaic module voltage difference between a reference voltage across photovoltaic modules of a photovoltaic grid-connected system and a voltage across photovoltaic modules, an inverter DC link voltage difference between an inverter DC link reference voltage of a photovoltaic grid-connected system and an inverter DC link voltage, and a grid current difference between a grid-connected reference current of a photovoltaic grid-connected system and a grid current.

[0031] S102: Inputting the PV module voltage difference, the inverter DC link voltage difference and the grid-connected current difference into a current converter controller constructed by an artificial neural network to obtain a control signal including a front-stage boost circuit modulation signal, a grid-connected reference current amplitude and a rear-stage inverter modulation signal.

[0032] S103: Control the front-stage boost circuit by modulating the signal of the front-stage boost circuit, and control the rear-stage inverter by modulating the signal of the rear-stage inverter, so as to obtain updated voltages across the photovoltaic module, inverter DC link voltage, and grid-connected current.

[0033] S104: Inputting the control signal output by the converter controller into a parameter estimator constructed by an artificial neural network, and fitting the control update process of the control signal based on the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current.

[0034] S105: Feedback the weight parameters in the parameter estimator to the converter controller to update the weight parameters of the converter controller; adaptively adjust the weight parameters of the converter controller through multiple rounds of iterations, so that the voltage at both ends of the photovoltaic module, the inverter DC link voltage and the grid-connected current respectively track the reference voltage at both ends of the photovoltaic module, the inverter DC link reference voltage and the grid-connected current, thereby completing the self-correcting photovoltaic grid-connected multi-level converter control.

[0035] For the convenience of explanation, the following description is only based on the server as the execution subject. The server mentioned in the present invention can be a server set up on the business platform, or a device such as a desktop computer, a notebook computer, etc. that can execute the solution of the present invention.

[0036] Figure 2 FIG. 1 is a schematic diagram of a photovoltaic grid-connected system in the present invention. Figure 2 It can be seen that the photovoltaic grid-connected system includes photovoltaic cells, boost circuits, bus capacitors, inverters and filters, and is connected to the grid side.

[0037] The present invention provides a photovoltaic grid-connected multi-stage current conversion control method based on full artificial neural network self-correction control. In one or more embodiments of the present invention, pulse width modulation (PWM) and the main circuit of the photovoltaic grid-connected system are regarded as a whole, that is, a controlled system, wherein the controlled targets are the voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current.

[0038] The present invention constructs a parameter estimator based on an artificial neural network and optimizes its weight parameters online, so that the parameter estimator can accurately fit the controlled system (the input and output of the parameter estimator are consistent with the input and output of the controlled system), and feeds back the weight parameters of the parameter estimator to the constructed converter controller based on the artificial neural network. Driven by three controlled target error signals, the converter controller optimizes the weight parameters of the converter controller using a gradient descent algorithm and the weight parameters fed back by the parameter estimator, so that it generates a reasonable control signal, wherein the grid-connected reference current amplitude output by the converter controller is multiplied by the output of the phase-locked device based on the artificial neural network to obtain the grid-connected reference current at the next moment, and the modulation signal of the front-stage boost circuit and the modulation signal of the rear-stage inverter output by the converter controller respectively control the front-stage boost circuit and the rear-stage inverter by means of PWM, so as to realize photovoltaic grid-connected multi-stage converter control.

[0039] Specifically, when the current converter controller is controlling, it is usually necessary to quickly track the voltage across the photovoltaic module (i.e., photovoltaic cell), the inverter DC link voltage, and the grid-connected current. Based on this, in one or more embodiments of the present invention, the server of the business platform can first obtain the reference voltage across the photovoltaic module and the voltage across the photovoltaic module output by the maximum power point tracking (MPPT) control and determine the photovoltaic module voltage difference between the two, as well as obtain the inverter DC link reference voltage and the inverter DC link voltage and determine the inverter DC link voltage difference between the two, and simultaneously obtain the grid-connected reference current and the grid-connected current and determine the grid-connected current difference between the two.

[0040] Among them, for the grid current difference, in one or more embodiments of the present invention, the server can determine it through a phase lock based on an artificial neural network. Specifically, the server can first obtain the grid voltage and input it into the trained phase lock based on the artificial neural network to obtain the phase angle sine value of the grid voltage, and determine the grid reference current according to the product of the phase angle sine value and the grid reference current amplitude. Then, the server can determine the grid current difference according to the difference between the grid reference current and the grid current. The mapping relationship between the grid voltage and the phase angle sine value of the grid voltage can be accurately obtained through the phase lock based on the artificial neural network, thereby obtaining an accurate grid reference current.

[0041] Further, in one or more embodiments of the present invention, the phase locker based on artificial neural network may include: an input layer, a hidden layer and an output layer. When the photovoltaic power generation system is connected to a single-phase power grid, the input variable and the output variable of the phase locker are the single-phase power grid voltage and the phase angle sine value, respectively. The number of neurons in the double hidden layer can be selected as 10, then the input layer of the phase locker may include 1 neuron, the hidden layer of the phase locker is a double hidden layer and each hidden layer includes 10 neurons, and the output layer of the phase locker includes 1 neuron. When the photovoltaic power generation system is connected to a three-phase power grid, the input variable and the output variable of the phase locker are the three-phase power grid voltage and the phase angle sine value, respectively. The number of neurons in the double hidden layer is also selected as 10, then the input layer of the phase locker may include 3 neurons, the hidden layer of the phase locker is a double hidden layer and each hidden layer includes 10 neurons, and the output layer of the phase locker includes 1 neuron.

[0042] After obtaining the difference parameter of the controlled target, the server can further track the reference value of the controlled target based on the above-mentioned converter controller and parameter estimator. Of course, both based on artificial neural networks can first initialize the network structure, that is, select the input and output node numbers of the parameter estimator and the converter controller according to the input variable dimension and the output variable dimension. In addition, the number of hidden layers and the number of neurons of the converter controller can be designed based on experience, and the parameter estimator has no hidden layer.

[0043] Specifically, in one or more embodiments of the present invention, the above-mentioned converter controller based on artificial neural network may include: an input layer, a hidden layer and an output layer. When the photovoltaic power generation system is connected to a single-phase power grid, the input variables of the converter controller are the difference between the reference voltage at both ends of the photovoltaic module output by the MPPT and the voltage at both ends of the photovoltaic module, the difference between the DC link reference voltage of the inverter and the DC link voltage of the inverter, and the difference between the single-phase grid-connected reference current and the single-phase grid-connected current (a single input value), and the output variables are the modulation signal of the front-stage boost circuit, the grid-connected reference current amplitude, and the modulation signal of the rear-stage single-phase inverter (a single output value). Then the input layer of the converter controller may include 3 neurons, the hidden layer of the converter controller is a double hidden layer and each hidden layer includes 20 neurons, and the output layer of the converter controller includes 3 neurons.

[0044] When the photovoltaic power generation system is connected to the three-phase grid, the input variables of the current converter controller are the difference between the reference voltage at both ends of the photovoltaic module output by the MPPT and the voltage at both ends of the photovoltaic module, the difference between the DC link reference voltage of the inverter and the DC link voltage of the inverter, and the difference between the three-phase grid-connected reference current and the three-phase grid-connected current (three input values), and the output variables are the modulation signal of the front-stage boost circuit, the grid-connected reference current amplitude, and the modulation signal of the rear-stage three-phase inverter (three output values). The input layer of the current converter controller may include 5 neurons, the hidden layer of the current converter controller is a double hidden layer and each hidden layer includes 20 neurons, and the output layer of the current converter controller includes 5 neurons.

[0045] The parameter estimator based on artificial neural network may include: an input layer and an output layer. When the photovoltaic power generation system is connected to a single-phase grid, the input variables of the parameter estimator are the modulation signal of the front-stage boost circuit, the grid-connected reference current amplitude, and the modulation signal of the rear-stage single-phase inverter (a single input value); the output variables are the voltage fitting value across the photovoltaic module, the inverter DC link voltage fitting value, and the single-phase grid-connected current fitting value (a single output value). The input layer of the parameter estimator may include 3 neurons, and the output layer of the parameter estimator may include 3 neurons.

[0046] When the photovoltaic power generation system is connected to the three-phase grid, the input variables of the parameter estimator are the modulation signal of the front-stage boost circuit, the grid-connected reference current amplitude, and the modulation signal of the rear-stage three-phase inverter (three input values); the output variables are the voltage fitting value across the photovoltaic module, the inverter DC link voltage fitting value, and the three-phase grid-connected current fitting value (three output values). The input layer of the parameter estimator may include 5 neurons, and the output layer of the parameter estimator may include 5 neurons.

[0047] After initializing the network structures of the phase-locked device, the inverter and the parameter estimator, their weight parameters can be further initialized. The weight parameters of the phase-locked device and the inverter controller include input weights, hidden layer weights and output weights. All the weight parameters are initialized using the grid search method, that is, [-d, d] is used as the initial search range, and the search range is gradually increased until it reaches [-D, D], and the search range with the smallest training root mean square error (RMSE) is selected for initialization. The weight parameters of the parameter estimator do not need to be initialized.

[0048] In one or more embodiments of the present invention, the server can use [-0.01, 0.01] as the initial search range, gradually increase the search range by ten times until it reaches [-10, 10], and select the search range with the smallest training RMSE for initialization, that is, the weight parameter initialization ranges of the phase-locked device and the power converter controller are selected as [-0.1, 0.1] and [-1, 1] respectively.

[0049] The phase-locked device and the current converter controller can use the gradient descent algorithm to update the weight parameters. In the gradient descent algorithm, only the learning rate needs to be set. The learning rate can be selected by the grid search method, that is, the program is run in the search range of [M1, M2] with a step size of m, and the learning rate with the smallest training RMSE is selected as the optimal learning rate. The parameter estimator can directly obtain the weight parameters by matrix point division without setting the algorithm parameters.

[0050] In one or more embodiments of the present invention, the server can run the program in a search range of [1e-6, 0.001] with a step size of 1e-6, and select the learning rate with the smallest training RMSE as the optimal learning rate, that is, the weight parameter learning rates of the phase-locked unit and the power converter are selected as 0.0001 and 1e-5 respectively.

[0051] The input layers of the phase-locked device and the current converter controller do not have an activation function. The hidden layers of both can select commonly used nonlinear activation functions (tanh, sigmoid, etc.), and the output layers of both can select activation functions according to the data range of the output physical variables. The parameter estimator does not use an activation function.

[0052] In one or more embodiments of the present invention, the output range of the phase lock is [-1, 1]. Except for the input layer, which has no activation function, other neurons use tanh function as the activation function. The numerical range of the modulation signal of the front-stage boost circuit and the modulation signal of the rear-stage inverter output by the converter controller is [-1, 1], and the numerical range of the output grid-connected reference current amplitude is [0, n] (in a single-phase power grid, n is times the total power of the photovoltaic modules divided by the effective value of the grid voltage; in a three-phase grid, n is times the total power of the photovoltaic module divided by 3 times the effective value of the grid voltage). In the present invention, the input layer of the converter controller has no activation function, the second output neuron activation function is set to n·sigmoid function, and the other neuron activation functions are all tanh functions. The parameter estimator does not use an activation function.

[0053] In the control of the entire photovoltaic grid-connected system, the weight parameter optimization methods of the phase-locked device, the current converter controller and the parameter estimator are different. The weight parameters of the phase-locked device need to be trained offline, that is, under the drive of the grid voltage, the weight parameters of the phase-locked device are continuously optimized using the gradient descent algorithm so that its output fits the grid voltage phase, thereby having a mapping relationship between the grid voltage and the grid voltage phase. There are relatively mature technologies for this, and the present invention will not go into details.

[0054] The weight parameters of the converter controller and the parameter estimator are updated online during the grid connection process. Specifically, the parameter estimator does not set a hidden layer due to the simple fitting task, such as Figure 3 As shown, Figure 3 is a schematic diagram of a network architecture of a parameter estimator in the present invention, Figure 3 The voltage across the photovoltaic module is shown in u pv , inverter DC link voltage u dc and grid current i g Schematic diagram of the fitting control output.

[0055] Set up k At the moment, the voltage difference between the two ends of the photovoltaic module, the voltage difference of the photovoltaic module, the DC link voltage difference of the inverter and the grid-connected current difference are input into the current converter controller based on the artificial neural network, and the control signal including the modulation signal of the front-stage boost circuit, the grid-connected reference current amplitude and the modulation signal of the rear-stage inverter is obtained by the following formula: k )= [ u r1 ( k ), i m ( k ), u r2 ( k )], , , ,e( k )=[ e 1( k ), e 2( k ), e 3( k )], e 1( k )= u pv_ref ( k )- u pv ( k ), e 2( k )= u dc_ref ( k )- u dc ( k ), e 3( k )= i g_ref ( k )- i g (k ).

[0056] In the formula, r( k )for k The control signal at the moment, u r1 ( k )for k The previous stage boost circuit modulates the signal at the moment, i m ( k )for k The grid-connected reference current amplitude at the moment, u r2 ( k )for k The modulation signal of the subsequent inverter at the moment, for k The output of the first hidden layer of the inverter controller at time , for k The input weight of the current converter controller at time t, for k The hidden layer weights of the current controller at time t, for k The output weight of the current converter controller at time for k Moment No. i Column elements, for k The output of the second hidden layer of the inverter controller at time , for k The difference parameter of the input converter controller at the moment, e 1( k )for k The voltage difference of the photovoltaic modules at the moment, e 2( k )for k The inverter DC link voltage difference at time e 3( k )for k The grid-connected current difference at the moment, u pv_ref ( k )for k The reference voltage across the photovoltaic module output by the maximum power point tracking control at the moment, u pv ( k )for k The voltage across the photovoltaic module at the moment, u dc_ref ( k )for k The inverter DC link reference voltage at the moment, udc ( k )for k The inverter DC link voltage at time i g_ref ( k )for k The grid-connected reference current at the moment, i g ( k )for k The grid current at the moment.

[0057] Furthermore, the control signal output by the converter controller is input into a parameter estimator based on an artificial neural network, and based on the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current, the weight parameter of the parameter estimator can be determined by the following formula: .

[0058] In the formula, for k The weight parameters of the moment parameter estimator, for k The voltage across the photovoltaic module at the moment, for k The inverter DC link voltage at the moment, for k Grid-connected current at all times.

[0059] Modulate the signal in the front-stage boost circuit u r1 ( k ), grid-connected reference current amplitude i m ( k ), the modulation signal of the subsequent inverter u r2 ( k ), the fitting output of the parameter estimator is: , " ” represents matrix dot product.

[0060] In the formula, for k The voltage fitting value of the two ends of the photovoltaic module at the moment, for k The inverter DC link voltage fitting value at the moment, for k The grid-connected current fitting value at the moment. The weight parameter of the parameter estimator This can achieve fast and accurate fitting of the photovoltaic grid-connected system.

[0061] Furthermore, in one or more embodiments of the present invention, the current controller may use a gradient descent method to update the weight parameters. The current controller includes an input layer, two hidden layers and an output layer. Each layer of neurons is connected in turn through the input weight W in , the hidden layer weight W and the output weight W out Connection, such as Figure 4 As shown, Figure 4 A schematic diagram of a network architecture of a current converter controller in the present invention.

[0062] The loss function of the current converter controller is: ,y( k )= [ u pv ( k ), u dc ( k ), i g ( k )],t( k )= [ u pv_ref ( k ), u dc_ref ( k ), i g_ref ( k )].

[0063] In the formula, for k The loss of the time-varying current controller, for k The reference output of the photovoltaic grid-connected system at the moment, for k The actual output of the photovoltaic grid-connected system at each moment.

[0064] Since the activation functions of the output layer of the converter controller are different, it is inconvenient to obtain the weight gradient uniformly when optimizing the weight parameters of the converter controller using the gradient descent algorithm. Therefore, the weight gradient can be obtained one by one according to the output, as follows:

[0065] The output weight gradient of the current converter controller is: , .

[0066] In the formula, Z(1, m )( k )for Line 1 of m Column elements, is the derivative of the loss function with respect to the output weight when the converter controller only considers the jth output, is the derivative of the loss function of the inverter controller with respect to the output weight.

[0067] The hidden layer weight gradient of the current converter is: , .

[0068] In the formula, is the derivative of the loss function with respect to the hidden layer weights when the converter controller only considers the jth output, is the derivative of the loss function of the variable current controller with respect to the hidden layer weights.

[0069] The input weight gradient of the current converter controller is: , .

[0070] In the formula, is the derivative of the loss function with respect to the input weight when the converter controller only considers the jth output, is the derivative of the loss function of the variable current controller with respect to the input weight.

[0071] The weight parameters in the above parameter estimator are fed back to the input weight gradient, hidden layer weight gradient and output weight gradient of the current converter, and the input weight, hidden layer weight and output weight of the current converter are updated according to the following formula: , , .

[0072] In the formula, is the learning rate, for k The output weight of the current converter at time +1, for k The hidden layer weight of the inverter controller at time +1, is the input weight of the current converter k The input weight of the current controller at time +1.

[0073] By optimizing the weight parameters of the current converter controller in real time, a reasonable control signal is outputted to make the voltage at both ends of the photovoltaic module, the inverter DC link voltage and the grid-connected current track their corresponding reference values ​​respectively, thus realizing photovoltaic grid-connected multi-level current conversion control.

[0074] Figure 5 This is a schematic diagram of a photovoltaic grid-connected multi-level current conversion control based on full artificial neural network self-correction control in the present invention. After the phase-locked device offline training and all initialization processes are completed, the current conversion control process is officially entered: First, the grid voltage is collected ug The obtained phase angle sine value c is input into the phase lock device and compared with the grid-connected reference current amplitude. i m Perform multiplication to obtain the grid-connected reference current c i m Secondly, the difference between the reference voltage across the PV module output by the MPPT and the voltage across the PV module is e 1. The difference between the inverter DC link reference voltage and the inverter DC link voltage e 2 and the difference between the grid-connected reference current and the grid-connected current e 3 is input to the current converter controller, thereby outputting the modulation signal of the previous stage boost circuit u r1 , grid-connected reference current amplitude i m And the modulation signal of the subsequent inverter u r2 .

[0075] Again, the previous stage boost circuit modulates the signal u r1 Modulation signal with the subsequent inverter u r2 The front-stage boost circuit and the rear-stage inverter are controlled separately to obtain the updated voltage across the photovoltaic module. u pv , inverter DC link voltage u dc Grid-connected current i g Finally, the updated voltage across the photovoltaic module is u pv , inverter DC link voltage u dc and grid current i g And the three outputs of the inverter controller u r1 , i m and u r2 The parameters are transmitted to the parameter estimator, which is used to fit the controlled system, that is, to obtain the weight parameter Z, and then the weight parameter Z is fed back to the converter controller, so as to adjust the weight parameter of the converter controller online to generate a suitable control signal r. The above process is repeated in sequence, and the error e is gradually reduced by continuously adjusting the weight parameter of the converter controller, so that the voltage across the photovoltaic module is u pv , inverter DC link voltage u dc and grid current i gQuickly track their reference values ​​respectively u pv_ref , u dc_ref and i g_ref , and realize stable photovoltaic grid connection.

[0076] based on Figure 1 The self-correcting photovoltaic grid-connected multi-stage current conversion control method shown in the figure first determines the difference between the actual value of the controlled target and the corresponding reference value. The controlled target includes the voltage at both ends of the photovoltaic module, the inverter DC link voltage and the grid-connected current. Then, each difference is input into the current conversion controller based on the artificial neural network to obtain a control signal including a modulation signal of the front-stage boost circuit, a grid-connected reference current amplitude and a modulation signal of the rear-stage inverter. Then, the front-stage boost circuit is controlled by the modulation signal of the front-stage boost circuit, and the rear-stage inverter is controlled by the modulation signal of the rear-stage inverter to obtain an updated photovoltaic module. The voltage at both ends of the photovoltaic module, the DC link voltage of the inverter and the grid-connected current are calculated, and the control signal output by the converter controller is input into a parameter estimator based on an artificial neural network, and the control update process of the control signal is fitted by the parameter estimator; the weight parameters in the parameter estimator are fed back to the converter controller to update the weight parameters of the converter controller, and the weight parameters of the converter controller are adaptively adjusted through multiple rounds of iterations, so that the voltage at both ends of the photovoltaic module, the DC link voltage of the inverter and the grid-connected current can quickly track the corresponding reference values ​​respectively, thereby completing the self-correcting photovoltaic grid-connected multi-level converter control.

[0077] The present invention determines the control signal of the photovoltaic grid-connected multi-stage current conversion control according to the difference between the controlled target and the reference value through the current conversion controller based on the artificial neural network, and quickly fits the control process of the photovoltaic grid-connected system through the parameter estimator based on the artificial neural network, so as to dynamically and real-time adjust the weight parameters in the current conversion controller by using the weight parameters of the parameter estimator, so that the photovoltaic grid-connected system is accurately modeled, thereby obtaining an accurate control signal, realizing the centralized control and self-correction control of the front-stage boost circuit and the rear-stage inverter, simplifying the control process, and improving the control accuracy.

[0078] When applying the self-correcting photovoltaic grid-connected multi-stage current conversion control method provided by the present invention, it is not necessary to Figure 1 The steps are executed in the order shown. The specific execution order of the steps can be determined according to needs, and the present invention does not limit this.

[0079] The above is a self-correcting photovoltaic grid-connected multi-stage current conversion control method provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding self-correcting photovoltaic grid-connected multi-stage current conversion control device, such as Figure 6 shown.

[0080] Figure 6 A schematic diagram of a self-correcting photovoltaic grid-connected multi-stage current conversion control device provided by the present invention includes:

[0081] The difference determination module 201 is used to determine the difference between the reference voltage at both ends of the photovoltaic module of the photovoltaic grid-connected system and the voltage at both ends of the photovoltaic module, the difference between the inverter DC link voltage of the inverter DC link of the photovoltaic grid-connected system and the inverter DC link voltage, and the grid current difference between the grid reference current and the grid current of the photovoltaic grid-connected system;

[0082] The control signal determination module 202 is used to input the PV module voltage difference, the inverter DC link voltage difference and the grid-connected current difference into the current conversion controller constructed by the artificial neural network to obtain a control signal including a modulation signal of the front-stage boost circuit, a grid-connected reference current amplitude and a modulation signal of the rear-stage inverter;

[0083] The control module 203 is used to control the front-stage boost circuit through the front-stage boost circuit modulation signal, and control the rear-stage inverter through the rear-stage inverter modulation signal to obtain the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current;

[0084] The parameter estimation module 204 is used to input the control signal output by the current converter controller into the parameter estimator constructed by the artificial neural network, and fit the control update process of the control signal based on the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current;

[0085] The iterative tracking module 205 is used to feed back the weight parameters in the parameter estimator to the converter controller to update the weight parameters of the converter controller; through multiple rounds of iterative adaptive adjustment of the weight parameters of the converter controller, the voltage at both ends of the photovoltaic module, the inverter DC link voltage and the grid-connected current are made to track the reference voltage at both ends of the photovoltaic module, the inverter DC link reference voltage and the grid-connected current respectively, thereby completing the self-correcting photovoltaic grid-connected multi-level converter control.

[0086] For the specific definition of the self-correcting photovoltaic grid-connected multi-level current conversion control device, please refer to the definition of the self-correcting photovoltaic grid-connected multi-level current conversion control method above, which will not be repeated here. Each module in the above-mentioned self-correcting photovoltaic grid-connected multi-level current conversion control device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0087] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1A self-correcting photovoltaic grid-connected multi-stage current conversion control method is provided.

[0088] The present invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 A self-correcting photovoltaic grid-connected multi-stage current conversion control method is provided.

[0089] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0090] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A self-correcting photovoltaic grid-connected multi-stage current conversion control method, characterized in that: include: Determine a photovoltaic module voltage difference between a reference voltage across photovoltaic modules of the photovoltaic grid-connected system and a voltage across photovoltaic modules, an inverter DC link voltage difference between an inverter DC link reference voltage of the photovoltaic grid-connected system and an inverter DC link voltage, and a grid current difference between a grid-connected reference current and a grid-connected current of the photovoltaic grid-connected system; Inputting the PV module voltage difference, the inverter DC link voltage difference and the grid-connected current difference into a current converter controller constructed by an artificial neural network, and obtaining a control signal including a modulation signal of a front-stage boost circuit, a grid-connected reference current amplitude and a modulation signal of a rear-stage inverter; The front-stage boost circuit is controlled by a modulation signal of the front-stage boost circuit, and the rear-stage inverter is controlled by a modulation signal of the rear-stage inverter to obtain updated voltages at both ends of the photovoltaic module, inverter DC link voltage, and grid-connected current; The control signal output by the converter controller is input into a parameter estimator constructed by an artificial neural network, and the control update process of the control signal is fitted based on the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current; The weight parameters in the parameter estimator are fed back to the converter controller to update the weight parameters of the converter controller; the weight parameters of the converter controller are adaptively adjusted through multiple rounds of iterations, so that the voltage at both ends of the photovoltaic module, the inverter DC link voltage and the grid-connected current respectively track the reference voltage at both ends of the photovoltaic module, the inverter DC link reference voltage and the grid-connected reference current, thereby completing the self-correcting photovoltaic grid-connected multi-level converter control; The photovoltaic module voltage difference, the inverter DC link voltage difference and the grid-connected current difference are input into the current converter controller constructed by the artificial neural network to obtain a control signal including a modulation signal of the front-stage boost circuit, a grid-connected reference current amplitude and a modulation signal of the rear-stage inverter, specifically including: The PV module voltage difference, inverter DC link voltage difference and grid-connected current difference are input into the current converter controller constructed by artificial neural network, and the control signal including the modulation signal of the front-stage boost circuit, the grid-connected reference current amplitude and the modulation signal of the rear-stage inverter is obtained by the following formula: r(k)=[u r1 (k),i m (k),u r2 (k)], G2(k)=tanh(G1(k)W(k)),G1(k)=tanh(e(k)W in (k)), e(k)=[e1(k),e2(k),e3(k)],e1(k)=u pv_ref (k)-u pv (k),e2(k)=u dc_ref (k)-u dc (k), e3(k)=i g_ref (k)-i g (k); Among them, r(k) is the control signal at time k, u r1 (k) is the modulation signal of the previous boost circuit at time k, i m (k) is the grid-connected reference current amplitude at time k, u r2 (k) is the modulation signal of the subsequent inverter at time k, G1(k) is the output of the first hidden layer of the converter controller at time k, W in (k) is the input weight of the current controller at time k, W(k) is the hidden layer weight of the current controller at time k, and W out (k) is the output weight of the current converter at time k, W out (:,i)(k)(i=1,2,3) is W at time k out (k) is the i-th column element of the inverter controller at time k, G2(k) is the second hidden layer output of the inverter controller at time k, e(k) is the difference parameter of the input inverter controller at time k, e1(k) is the PV module voltage difference at time k, e2(k) is the inverter DC link voltage difference at time k, e3(k) is the grid-connected current difference at time k, u pv_ref (k) is the reference voltage across the photovoltaic module output by the maximum power point tracking control at time k, u pv (k) is the voltage across the photovoltaic module at time k, u dc_ref (k) is the inverter DC link reference voltage at time k, u dc (k) is the inverter DC link voltage at time k, i g_ref (k) is the grid-connected reference current at time k, i g (k) is the grid-connected current at time k; The control signal output by the current converter controller is input into a parameter estimator constructed by an artificial neural network, and the control update process of the control signal is fitted based on the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current, specifically including: The control signal output by the converter controller is input into the parameter estimator constructed by the artificial neural network, and the weight parameter of the parameter estimator is determined by the following formula based on the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current: Z(k)=[u pv (k) / u r1 (k),u dc (k) / i m (k),i g (k) / u r2 (k)]; Fitting the control update process of the control signal through the weight parameters of the parameter estimator; Among them, Z(k) is the weight parameter of the parameter estimator at time k, u pv (k) is the voltage across the photovoltaic module at time k, u dc (k) is the inverter DC link voltage at time k, i g (k) is the grid-connected current at time k; The variable current controller updates weight parameters using a gradient descent method; Feeding back the weight parameters in the parameter estimator to the current converter controller to update the weight parameters of the current converter controller specifically includes: The loss function of the current converter is: y(k)=[u pv (k),u dc (k),i g (k)],t(k)=[u pv_ref (k),u dc_ref (k),i g_ref (k)]; The output weight gradient of the current converter controller is: The hidden layer weight gradient of the current converter is: The input weight gradient of the current converter is: The weight parameters in the parameter estimator are fed back to the input weight gradient, hidden layer weight gradient and output weight gradient of the converter controller, and the input weight, hidden layer weight and output weight of the converter controller are updated according to the following formula: Among them, E c (k) is the loss of the inverter controller at time k, t(k) is the reference output of the photovoltaic grid-connected system at time k, y(k) is the actual output of the photovoltaic grid-connected system at time k, Z(1,m)(k) is the 1st row and mth column element of Z(k), is the derivative of the loss function with respect to the output weight when the converter controller only considers the jth output, is the derivative of the loss function of the inverter controller with respect to the output weight, is the derivative of the loss function with respect to the hidden layer weights when the converter controller only considers the jth output, is the derivative of the loss function of the inverter controller with respect to the hidden layer weights, is the derivative of the loss function with respect to the input weight when the converter controller only considers the jth output, is the derivative of the loss function of the inverter controller with respect to the input weight, η c is the learning rate, W out (k+1) is the output weight of the current controller at time k+1, W(k+1) is the hidden layer weight of the current controller at time k+1, and W in (k+1) is the input weight of the current controller at time k+1, and ".*" represents matrix dot multiplication.

2. The self-correcting photovoltaic grid-connected multi-stage current conversion control method according to claim 1, characterized in that: Determine the grid-connected current difference between the grid-connected reference current and the grid-connected current of the photovoltaic grid-connected system, specifically including: Obtaining the grid voltage and inputting it into a trained phase-locked device constructed by an artificial neural network to obtain the phase angle sine value of the grid voltage, and determining the grid reference current according to the product of the phase angle sine value and the grid reference current amplitude; The grid-connected current difference is determined according to the difference between the grid-connected reference current and the grid-connected current.

3. The self-correcting photovoltaic grid-connected multi-stage current conversion control method according to claim 1, characterized in that: The current controller constructed by artificial neural network includes: an input layer, a hidden layer and an output layer; When the photovoltaic power generation system is connected to a single-phase power grid, the input layer of the current converter controller includes 3 neurons, and the output layer of the current converter controller includes 3 neurons; When the photovoltaic power generation system is connected to a three-phase power grid, the input layer of the current converter controller includes 5 neurons, and the output layer of the current converter controller includes 5 neurons.

4. The self-correcting photovoltaic grid-connected multi-stage current conversion control method according to claim 1, characterized in that: The parameter estimator constructed by the artificial neural network includes: an input layer and an output layer; When the photovoltaic power generation system is connected to a single-phase power grid, the input layer of the parameter estimator includes 3 neurons, and the output layer of the parameter estimator includes 3 neurons; When the photovoltaic power generation system is connected to a three-phase power grid, the input layer of the parameter estimator includes 5 neurons, and the output layer of the parameter estimator includes 5 neurons.

5. The self-correcting photovoltaic grid-connected multi-stage current conversion control method according to claim 2, characterized in that: The phase-locked device constructed by the artificial neural network comprises: an input layer, a hidden layer and an output layer; When the photovoltaic power generation system is connected to a single-phase power grid, the input layer of the phase lock includes 1 neuron, and the output layer of the phase lock includes 1 neuron; When the photovoltaic power generation system is connected to a three-phase power grid, the input layer of the phase lock includes 3 neurons, and the output layer of the phase lock includes 1 neuron.

6. A self-correcting photovoltaic grid-connected multi-stage current conversion control device, characterized in that: include: A difference determination module is used to determine a photovoltaic module voltage difference between a reference voltage at both ends of a photovoltaic module of a photovoltaic grid-connected system and a voltage at both ends of the photovoltaic module, an inverter DC link voltage difference between an inverter DC link reference voltage of the photovoltaic grid-connected system and an inverter DC link voltage, and a grid current difference between a grid-connected reference current and a grid-connected current of the photovoltaic grid-connected system; The control signal determination module is used to input the PV module voltage difference, the inverter DC link voltage difference and the grid-connected current difference into the current converter controller constructed by the artificial neural network, and obtain the control signal including the modulation signal of the front-stage boost circuit, the grid-connected reference current amplitude and the modulation signal of the rear-stage inverter through the following formula: r(k)=[u r1 (k),i m (k),u r2 (k)], G2(k)=tanh(G1(k)W(k)),G1(k)=tanh(e(k)W in (k)), e(k)=[e1(k),e2(k),e3(k)],e1(k)=u pv_ref (k)-u pv (k),e2(k)=u dc_ref (k)-u dc (k), e3(k)=i g_ref (k)-i g (k); A control module is used to control the front-stage boost circuit through the modulation signal of the front-stage boost circuit, and control the rear-stage inverter through the modulation signal of the rear-stage inverter, so as to obtain the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current; The parameter estimation module is used to input the control signal output by the converter controller into the parameter estimator constructed by the artificial neural network, and determine the weight parameter of the parameter estimator based on the updated voltage across the photovoltaic module, the inverter DC link voltage and the grid-connected current through the following formula: Z(k) = [u pv (k) / u r1 (k),u dc (k) / i m (k),i g (k) / u r2 (k)]; fitting the control update process of the control signal through the weight parameters of the parameter estimator; The iterative tracking module is used to feed back the weight parameters in the parameter estimator to the input weight gradient, hidden layer weight gradient and output weight gradient of the converter controller, and update the input weight, hidden layer weight and output weight of the converter controller according to the following formula: Through multiple rounds of iterative adaptive adjustment of the weight parameters of the current conversion controller, the voltage at both ends of the photovoltaic module, the DC link voltage of the inverter and the grid-connected current are made to track the reference voltage at both ends of the photovoltaic module, the DC link reference voltage of the inverter and the grid-connected reference current respectively, thus completing the self-correcting photovoltaic grid-connected multi-level current conversion control; The variable current controller updates weight parameters using a gradient descent method; The loss function of the current converter is: y(k)=[u pv (k),u dc (k),i g (k)],t(k)=[u pv_ref (k),u dc_ref (k),i g_ref (k)]; The output weight gradient of the current converter controller is: The hidden layer weight gradient of the current converter is: The input weight gradient of the current converter is: Among them, r(k) is the control signal at time k, u r1 (k) is the modulation signal of the previous boost circuit at time k, i m (k) is the grid-connected reference current amplitude at time k, u r2 (k) is the modulation signal of the subsequent inverter at time k, G1(k) is the output of the first hidden layer of the converter controller at time k, W in (k) is the input weight of the current controller at time k, W(k) is the hidden layer weight of the current controller at time k, and W out (k) is the output weight of the current converter at time k, W out (:,i)(k)(i=1,2,3) is W at time k out (k) is the i-th column element of the inverter controller at time k, G2(k) is the second hidden layer output of the inverter controller at time k, e(k) is the difference parameter of the input inverter controller at time k, e1(k) is the PV module voltage difference at time k, e2(k) is the inverter DC link voltage difference at time k, e3(k) is the grid-connected current difference at time k, u pv_ref (k) is the reference voltage across the photovoltaic module output by the maximum power point tracking control at time k, u pv (k) is the voltage across the photovoltaic module at time k, u dc_ref (k) is the inverter DC link reference voltage at time k, u dc (k) is the inverter DC link voltage at time k, i g_ref (k) is the grid-connected reference current at time k, i g (k) is the grid-connected current at time k; Z(k) is the weight parameter of the parameter estimator at time k, u pv (k) is the voltage across the photovoltaic module at time k, u dc (k) is the inverter DC link voltage at time k, i g (k) is the grid-connected current at time k; E c (k) is the loss of the inverter controller at time k, t(k) is the reference output of the photovoltaic grid-connected system at time k, y(k) is the actual output of the photovoltaic grid-connected system at time k, Z(1,m)(k) is the 1st row and mth column element of Z(k), is the derivative of the loss function with respect to the output weight when the converter controller only considers the jth output, is the derivative of the loss function of the inverter controller with respect to the output weight, is the derivative of the loss function with respect to the hidden layer weights when the converter controller only considers the jth output, is the derivative of the loss function of the inverter controller with respect to the hidden layer weights, is the derivative of the loss function with respect to the input weight when the converter controller only considers the jth output, is the derivative of the loss function of the inverter controller with respect to the input weight, η c is the learning rate, W out (k+1) is the output weight of the current controller at time k+1, W(k+1) is the hidden layer weight of the current controller at time k+1, and W in (k+1) is the input weight of the current controller at time k+1, and ".*" represents matrix dot multiplication.

7. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

Images