KR100886891B1 - Control system of solar cell generation using genetic algorithm and neuro fuzzy controller - Google Patents

Abstract

A control system of a solar cell generation is provided to control MIPPT(Maximum Power Point Tracking) through a neuro fuzzy controller in regardless of the change of MPP(Maximum Power Point). A control system of a solar cell generation includes a solar cell array(10), an electric power conversion unit(20), and MPPT controller(30). The solar cell array outputs DC signal corresponding to sunlight, the electric power conversion unit converts DC signal into AC or DC voltage. The MPPT controller includes an MPP estimation unit(30a) and a neuro-fuzzy controller(30b). The MPP estimation unit estimates MPP by measuring an output voltage and a current of the solar cell array. The neuro-fuzzy controller estimates and controls MPP automatically according to MPP change, the electric power conversion unit operates according to a control signal of the neuro-fuzzy controller.

Description

CONTROL SYSTEM OF SOLAR CELL GENERATION USING GENETIC ALGORITHM AND NEURO FUZZY CONTROLLER}

The present invention relates to a system for performing MPPT control according to the incident conditions of solar light, and in particular to the environmental conditions in which the solar module is located by using a neurofuge controller controlled by a genetic algorithm for tracking the maximum power of the solar power generation system. The present invention relates to a photovoltaic control system that controls maximum power for change.

In general, photovoltaic power generation (Photovolatics) is a power generation system that converts light directly into electrical energy using a solar cell without the aid of a generator, and is a power generation system configured according to actual demand load using a solar cell.

It is composed of power converters such as solar cells, storage batteries, and inverters. When sunlight is irradiated on a solar cell in which a P-type semiconductor and an N-type semiconductor are bonded together, holes are generated in the solar cell by energy of the sunlight. ) And electrons are generated.

At this time, holes are gathered toward the P-type semiconductor and electrons are gathered toward the N-type semiconductor, and when a full position occurs, current flows. In response, the inverter converts the generated DC power into AC of commercial frequency and voltage to connect to the power system. At the same time, electrical monitoring and protection of the DC and AC side of the system.

 Here, the output characteristics of the solar cell change the maximum power point by a variable depending on the amount of solar radiation, operating voltage and temperature, so in order to maximize the efficiency of photovoltaic power generation, maximum power point tracking (Maximum Power Point Tracking) Called MPPT).

At this time, the MPPT is a control method for producing the maximum power in accordance with the change in the amount of insolation in the inverter, the magnitude of the DC voltage is changed when the amount of insolation changes, the efficiency of the inverter is lowered due to the increase or decrease of the voltage, to improve the various The MPPT algorithm of the method is used.

1 is a graph showing the maximum output point by the voltage and current of the solar cell. As shown in FIG. 1, the maximum power point of the solar cell, which varies with the amount of solar radiation and the surface temperature, is estimated.

Here, the solar cell output when the solar radiation and the surface temperature are constant moves along the current-voltage curve (IV Curve), and the operating point that generates the maximum output is called the maximum power point (hereinafter referred to as MPP). MPPT control is required to control the operating point.

Here, in Korean Patent Laid-Open Publication No. 2007-0043746, the perturbation / observation method and the conductance increasing method are most typically used as the MPPT control method. Because of this, it is necessary to constantly monitor the increase in the amount of power, and the amount of calculation is also a heavy burden on the control.

In addition, it does not suggest a stable method in terms of convergence speed and accuracy in approaching MPP, and only suggests an equation based on experience.

As another method, a method of grasping the electrical characteristics of the photovoltaic module in advance and inputting the variation range of the MPP with respect to the change of external conditions in advance to converge the optimal point within the range is proposed. Although it can be raised somewhat, the convergence method has the same problem because it does not deviate from the above method.

The conventional technique of perturbation and observation (PO) control is to increase or decrease the output voltage of the solar cell array periodically and to find the maximum power operating point by comparing the previous output power with the current output power.

In this case, the PO control technique is simple, but when the amount of insolation changes suddenly, the output voltage of the solar cell array is not equal to the voltage at the maximum power operating point.

On the other hand, IncCond (Incremental Conduction) control technique according to the prior art is a method of finding the maximum power operating point by comparing the conductance and incremental conductance of the solar cell output, which is the maximum power operation even when the solar radiation suddenly changes when the maximum power point is reached Although there is an advantage to be estimated as a point, there is a disadvantage that the efficiency is not high for low solar radiation.

In addition, the Neuro-Fuzzy controller uses a fuzzy theory based on the fuzzy set idea to express linguistic ambiguity, and a neural network that mimics the neural structure of the human brain. It is an intelligent controller fused with each other.

Here, fuzzy theory is a theory that tries to mathematically approach the process of the brain judging and deciding various problems in an ambiguous and unclear situation, from the binary logic that each object belongs to or does not belong to each set. It is based on a fuzzy set mathematically expressed by expressing the degree of belonging as a membership function.

In other words, the existing set of crisps consists of a set of elements having a value of 1 when the element x belongs to the set A and a value of 0 when not belonging to the set A, that is, a definite set.

On the other hand, in the fuzzy set, the degree to which the element x belongs to the set A is represented by a value between 0 and 1, and a value between 0 and 1, that is, [0, 1] is called a membership function.

Neural Network Theory is an algorithm modeled using the human brain structure and is an intelligent algorithm that mimics the process of information transfer in the human nervous system.

The neural network can learn (training) by repeatedly changing the weight of each input signal with respect to the input information for the repeatedly input information, the neural network computer performing the learning can store the input information, like the human brain Because of its nondeterministic nature, it can recognize slightly wrong or similar inputs.

It is composed of a three-layer structure of an input side, a hidden layer, and an output layer. The hidden layer and the output layer include a plurality of nodes, and the input layer receives the input data and delivers it to the hidden layer.

In addition, the nodes in the hidden layer and the output layer have an activation function. The active function is typically a sigmoid function, and the neural network uses an error backpropagation algorithm, and the node of the input layer and the hidden layer, the node of the hidden layer, and the output layer. Learn the weights of the connections between nodes.

Genetic algorithms, on the other hand, are one of the optimization techniques based on the biological principle of “deficit survival,” such as natural selection process, in which natural organisms that are well-adapted to the environment can leave more offspring. It is a search algorithm based on the genetic mechanism of the natural world, which is a natural evolutionary process that develops in a good direction through the change of genes.

Here, the genetic algorithm expresses the solution to be searched by chromosome to form an individual, and various chromosomes are formed to form a group, and the chromosomes in the group are optimized by repeatedly performing selection, crossing, mutation and evaluation. Go to

Therefore, the MPPT control system of photovoltaic power generation requires the application of an intelligent control algorithm that combines the inference function of the fuzzy system and the learning function of the neural network using a neurofuge system.

The present invention has been made to solve the above problems, and even if the MPP changes due to a change in external conditions, the neuron is automatically controlled by a genetic algorithm capable of estimating the MPP and controlling the MPP using a neurofuge controller. An object of the present invention is to provide a photovoltaic power generation control system using a fuzzy controller.

Another object of the present invention is to use a neurofuge controller, which is an intelligent algorithm with learning ability, using a neurofuge controller controlled by a genetic algorithm that can effectively perform MPPT control even if the characteristics of the solar generator are nonlinear. It is an object to provide a solar power generation control system.

In order to achieve the above object, the present invention provides a MPPT control system for photovoltaic power generation, comprising: a solar array configured to output a DC signal according to sunlight to receive sunlight and convert it into electrical energy; A power converter which receives the DC signal and converts the signal into an AC or DC voltage; An MPP estimator for estimating the MPP by measuring the output voltage and the current of the solar array; and the power conversion so that an error between the two voltages is minimized by receiving the output voltage of the power converter and the voltage estimated by the MPP estimator; By outputting a control signal for driving the unit, it includes an MPPT control unit composed of a neuro-fuge controller for automatically estimating and controlling the MPP in accordance with the change of the MPP.

The neurofuge controller may include a first layer configured to receive data from an input variable; A second layer that calculates a goodness of fit for each rule using a membership function from the input value; A third layer normalizing the goodness of fit using the membership function for each input variable; A fourth layer for calculating an inference value for each input variable; And a fifth layer for obtaining an output of the final inference value.

In addition, the neurofuge controller receives two input variables (x ₁ , x ₂ ) as inputs, and uses five membership functions for each input variable and has one output variable.

The input variable x ₁ is defined as a difference between a voltage of the MPP and a signal measuring the output of the power converter, and the control signal of the power converter driving circuit uses a value inferred by the neuropurge controller. do.

The neurofuge controller calculates a process in which an input value is calculated while passing through each layer, but the first layer (Layer 1) is composed of two nodes and receives an input to the second layer (Layer 2). The second layer (Layer 2) is composed of 10 nodes, and transfers the goodness of fit of the input value to the third layer (Layer 3) using a membership function defined by experience and learning in each node, The third layer (Layer 3) is composed of 10 nodes to transfer the normalized goodness of fit for each input variable to the fourth layer (Layer 4), the fourth layer (Layer 4) is composed of two nodes The sum multiplied by the connection weight is added to the fifth layer (Layer 5), and the fifth layer (Layer 5) finally outputs an operation value.

The number of membership functions may be optimized using a genetic algorithm.

In addition, the neurofuge controller is trained to have optimal performance using experimental data, but the learning is characterized by using an error backpropagation algorithm used in neural networks.

The error back propagation algorithm is based on a gradient descent method, and the learning is performed to minimize the error based on an error between a desired output and a system output.

Further, in order to improve the learning performance, when the amount of change in the connection weight increases, the amount of change is increased, and when the amount of change in the connection weight decreases, the momentum for reducing the amount of change is added.

Preferably, the momentum is determined by a difference between a current connection weight and a previous connection weight, wherein the momentum is multiplied by a momentum coefficient to be added when adjusting the connection weight.

As described above, the present invention having the configuration as described above can automatically estimate the MPP even if the MPP is changed due to changes in solar radiation or temperature, and can control the MPPT using a neurofuge controller. In addition, the neurofuge controller is an intelligent algorithm with learning ability and can effectively perform MPPT control even if the characteristics of the solar generator change nonlinearly.

1 is a reference diagram showing the voltage-current characteristics of a solar cell.

2 is a block diagram of MPPT control of a solar generator using a neurofuge controller.

3 is a reference diagram schematically illustrating the internal structure of the neurofuge controller.

4 is a reference diagram schematically illustrating the membership function used in the neurofuge controller.

Figure 5 is a reference diagram illustrating the chromosome composition in the genetic algorithm.

Figure 6 is a reference diagram illustrating the flow of the genetic algorithm.

<Brief description of components for main parts of the drawing>

10: solar array 20: power converter

30: MPPT control unit 30a: MPP estimation unit

30b: Neuropurge Controller 40: Load

The present invention is to solve the problem by measuring the output voltage and current of the solar array to obtain the MPP, by controlling the output voltage of the inverter using a neuropurge controller to minimize the error between the output of the MPP and the inverter.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

2 is a schematic block diagram of the MPPT control system for photovoltaic power generation. The MPPT control system for photovoltaic power generation includes a photovoltaic array 10, a power converter 20, an MPPT controller 30, and a load 40.

The power converter 20 is a device for converting a DC signal, which is the output of the solar array, into an AC or DC voltage, and may use a converter or an inverter and may include a gate driving circuit.

The MPPT controller 30 may include an MPP estimator 30a and a neurofuge controller 30b.

The MPPT control unit 30 provided in the present invention may be composed of an MPP estimator 30a and a fuzzy neural network. The MPP estimator 30a estimates the MPP by measuring the voltage and the current which are the output of the solar array, and estimates and outputs the voltage (V _MPP ) of the MPP using a linear reoriented coordinate method (LRCM). The neuropurge controller 30b receives the output voltage of the inverter and v _MPP obtained from the MPP estimator 30a and outputs a control signal for driving the gate of the power converter so that the error of the input value is minimized.

That is, the MPP estimator 30a may obtain a mathematical method using a linear reoriented coordinate method (LRCM) by sensing the voltage and current of the solar array, and the measuring unit capable of measuring the current and voltage may be A microprocessor. It is possible to use more than two channels of / D conversion.

Here, LRCM is well represented in "Development of a Traceable and MPPT Controller of Photovoltaic Power Generation System" (P. 21, P54-63), which presents a dynamic model for grid linkage.

The neuro fuzzy controller 30b needs a measuring unit capable of measuring the output voltage of the power change unit 20, an A / D converter capable of A / D conversion, and a PWM for output of the power converter. If you do, you can use a microprocessor with PWM capability. In addition, neurofuzzy uses a lot of calculations, so it is good to use a processor capable of high-speed calculation.

In addition, the MPP estimator 30a and the neurofuge controller 30b may be implemented as one processor, and two processors may be used to reduce the computational burden. If two processors are used, communication between the two processors is required.

Lastly, the processor may use Mega 128, for example, among the AVRs. Any processor capable of controlling the neurofuge controller 30b may be programmed.

3 is a diagram illustrating the internal network structure of the neurofuge controller 30b in detail, and includes a first layer that receives data from an input variable and a second suitability for each rule using a membership function from the input value. A layer, a third layer for normalizing the goodness of fit obtained by using the membership function for each input variable, a fourth layer for obtaining the inference value for each input variable, and a fifth layer for obtaining the output as the final inference value.

The neurofuge controller 30b receives two input variables as inputs, uses five membership functions for each input variable, and has one output variable. Input variables are defined as x ₁ , x ₂ . x ₁ is the difference between the voltage of the _MPP (v _MPP ) defined by Equation 2 and the signal (v _in ) measuring the output of the power converter, x ₂ is the amount of change of x ₁ . The membership function for each rule is indicated by subscripts A ₁₁ , A ₁₂ , A ₁₃ , A ₁₄ , A ₁₅ , A ₂₁ , A ₂₂ , A ₂₃ , A ₂₄ , A ₂₅ . u ₁₁ , u ₁₂ , u ₁₃ , u ₁₄ , u ₁₅ , u ₂₁ , u ₂₂ , u ₂₃ , u ₂₄ , u _25, and define the output variable Defined as. In other words Is a value inferred from the neuropurge controller and used as a control signal of the driving circuit of the power converter.

The neurofuzzy controller 30b may be represented by ten fuzzy rules, and Equation 3 is a mathematical expression of the fuzzy rules of the neurofuzzy controller.

Where R _j is the j th rule, x _k is the k th input variable, A _kc is the c th membership function for the k th input variable, y _j is the output for the j th rule, and w _kc is the connection weight to be.

That is, Equation 3 is a general formula for the ten fuzzy rules of Equation 5 below, and the measured value is one of the power converter output v _in , and the error with v _MPP is x ₁ , and the rate of change of the error is x ₂ . do.

Equations 5 to 7 illustrate a process in which an input value is calculated while passing through each layer in the neurofuge controller.

Here, in the first layer, the input layer is composed of two nodes, and the input values are transferred to the second layer. The second layer consists of 10 nodes, and each node obtains u _kc , which is a goodness of fit for the input value, using the membership function defined through experience or learning, and delivers it to the third layer.

In addition, the third layer is composed of ten nodes, and after obtaining the normalized goodness of fit for each input variable using Equation 5, the third layer is transferred to the fourth layer. The fourth layer is composed of two nodes, and the output value of each node of the third layer receives values multiplied by the connection weight between the fourth layer and the fourth layer, and then adds these values to the fifth layer. The fifth layer is the output layer and adds the values input from the fourth layer to output the final output.

Here, u _kc is a goodness-of-fit for the c-th membership function of the k-th history, g is the number of membership functions used in the k-th input variable, and in the present invention, the number of membership functions is optimized using a genetic algorithm.

here, Is the normalized goodness-of-fit and w _kc is the connection weight.

Where f _k is the inference value for the kth input, and n is the number of input variables in the present invention.

The neurofuge controller 30b should learn using experimental data. wkc is initially set randomly as the connection weight, and then the neurofuge controller is set to have optimal performance through the learning process.

In the present invention, an error propagation (BP) algorithm based on a gradient descent method commonly used in neural networks is used as a learning algorithm, which is based on an error between the desired output and the output of the system. Learning is done so that (E _v ) is minimal.

Where E _v is the error for each input data, y _v is the target value and Represents the actual output of the system.

In this case, given N input / output data pairs, the final output error of the neurofuge controller is expressed by Equation (9).

The adjustment (learning) of the connection weight may be performed by adding a change of an adjustment value obtained from an error to an existing value by using Equation (10).

Variation in connection weights to minimize error E _v Can be obtained using the gradient descent method as shown in Equation (11). Where η is the learning rate.

In this case, momentum is added to Equation 11 to shorten the learning time and improve learning performance. Momentum means inertia, which increases the amount of change if the amount of change in the connection weight is large, and reduces the amount of change according to the amount of change. Momentum is determined by the difference between the current connection weight and the previous connection weight as shown in Equation 12.

The momentum is multiplied by the momentum coefficient α to add to the adjustment of the connection weight. Adjustment amount for learning of final neurofuge controller Is determined by Equation 13.

FIG. 4 is a diagram illustrating an example of the membership functions that can be considered in the second layer of the neurofuge controller 30b, and a triangle type membership function having an overlapping degree of 0.5 is used for each input variable.

The neurofuge controller 30b is learned by Equation 13, but the control performance is determined by determining the learning rate, the momentum coefficient and the number of membership functions for each input variable and the vertex of the membership function at that time.

Therefore, we need to effectively determine the learning rate, momentum coefficient, number of membership functions for each input variable, and the vertices of the membership function at that time. In the present invention, the above values are determined using a genetic algorithm which is an optimization algorithm.

5 is a diagram showing the shape of the chromosome used in the genetic algorithm. In this task, we used a genetic algorithm based on real coding. The chromosomes were constructed to have values for the learning rate, momentum coefficients, and the number of membership functions to use for the two input variables.

6 is a diagram illustrating a flowchart of a genetic algorithm. Genetic algorithms are algorithms that evolve chromosomes to optimal solutions through initial population generation, evaluation, selection, crossover, and mutation operations, including early chromosome generation.

Initial population generation randomly generates chromosomes (individuals) of the type shown in FIG. 5 to form one population. In the fitness evaluation, the performance of the neurofuge controller shown in FIG. 4 is evaluated by using the chromosome value as a part of evaluating the generated individuals.

At this time, the vertex of the membership function is determined using the c-Means Clustering algorithm. The performance of each individual is obtained from the evaluation, which means the goodness of fit for each individual.

Improving the Pattern Recognition Performance of Back Error Propagation Algorithm by using Genetic Algorithm and Fuzzy c-Means, Korean Society for Military Operations Analysis Vol 21, No. 1, P83 ~ 93) is described in detail, so detailed description thereof will be omitted.

Here, when the goodness of fit is calculated, the reproduction part generates new individuals through selection, crossover, and mutation operation from the initially generated population. In the present invention, a roulette wheel operator is used as a selection operator, an arithmetic operator is used as a crossing operator, and a uniform mutation operator is used as a mutation operator.

In the present invention, the number of individuals in the group was used 100 and the number of repeat households was used 300 generations. As described above, the neurofuge controller is used to implement the MPPT control system of the solar generator.

In the above described exemplary embodiments of the present invention by way of example, the scope of the present invention is not limited only to this specific embodiment, and those skilled in the art within the scope of the claims of the present invention Changes may be made as appropriate.

Claims (10)

In MPPT control system of solar power generation, A solar array for outputting a DC signal according to sunlight to receive sunlight and convert it into electrical energy; A power converter which receives the DC signal and converts the signal into an AC or DC voltage; An MPP estimator for estimating the MPP by measuring the output voltage and the current of the solar array; and the power conversion so that an error between the two voltages is minimized by receiving the output voltage of the power converter and the voltage estimated by the MPP estimator; MPPT control unit composed of neurofuge controller which automatically estimates and controls MPP according to the change of MPP by outputting control signal for driving the unit Including, The neurofuge controller is composed of five layers to receive two input variables as inputs, calculates one output variable by using a membership function for each input variable, and the first layer receives data from the input variables and receives a second input variable. The second layer calculates the goodness of fit for each fuzzy rule using the membership function from the data input from the first layer, and optimizes the genetic algorithm at each node, and implements the experience and error backpropagation algorithm. The fitness for the data is transmitted to the third layer by using the membership function defined as learning, and the third layer normalizes the fitness to the fourth layer by using the membership function for each input variable. The fourth layer calculates an inference value by summing values multiplied by a connection weight for each input variable. And a fifth layer, wherein the fifth layer calculates an output of a final inference value and outputs an operation value. The solar power generation control system using a neurofuge controller controlled by a genetic algorithm. delete delete The method according to claim 1, The input variable is defined as the difference between the voltage of the MPP and the signal measuring the output of the power converter, and the control signal of the power converter driving circuit is controlled by a genetic algorithm, characterized in that using the value inferred from the neurofuge controller Photovoltaic power generation control system using a neurofuge controller. delete delete delete delete The method according to claim 1, The genetic algorithm evolves the optimal chromosome through the generation of the initial group including the initial chromosome generation, fitness evaluation, selection, crossover, and mutation operation, and evaluates the performance of the neurofuge controller using the chromosome value. The peak of the membership function is determined by a clustering algorithm, the solar power generation control system using a neurofuge controller controlled by a genetic algorithm. delete