TWI669589B - Maximum power tracking method for solar cell and system thereof suitable for real-time online environment - Google Patents

Abstract

The invention relates to a solar cell maximum power tracking method and system suitable for a real-time online environment, which mainly enables a microcontroller of a maximum power tracking system to connect and receive the voltage, current signal, and ambient illuminance and temperature of the solar cell. Data, and a fuzzy perturbation method built into the microcontroller, so that the PWM driver is connected between the microcontroller and the DC/DC converter, and the DC/DC converter is connected to the solar cell to be output by the microcontroller. The duty cycle PWM signal drives the PWM driver to drive the DC/DC converter for output, performs neural network-like training in the human-machine interface, and transmits the trained weights to the microcontroller via the communication module; In this way, not only can the disturbance amount be adjusted according to the current position, but also provide more accurate tracking control, and the reaction speed is very fast, which makes the maximum power tracking more suitable for the online environment, and can reduce the burden on the microcontroller, so that it can Suitable for different environments throughout the year.

Description

Solar cell maximum power tracking method and system suitable for real-time online environment

The invention relates to a solar cell maximum power tracking method and system suitable for a real-time online environment, in particular to a method for not only adjusting the amount of disturbance according to the current position, but also providing more accurate tracking control, and the reaction speed is very fast. The solar power is suitable for the real-time online environment, and the burden on the microcontroller can be reduced, so that it can be adapted to different environments throughout the year, and the solar energy is applied to the real-time online environment. Battery maximum power tracking method and system innovation designer.

According to the existing solar cell maximum power tracking method, it mainly includes two aspects of hardware technology and software technology. Hardware technology such as Actual Measurement and Power Compensation Method; software technology methods are mostly Disturbance view Perturbation & Observation (P & O), Incremental Conductance Algorithm, and Fuzzy Logic Method. Hardware technology is less expensive because of its higher cost; software technology mostly writes tracking methods in microcontrollers or control chips, and then controls the converter output through pulse width modulation (PWM) technology to achieve maximum power. Tracking purpose.

Among them, as for the common method of tracking the maximum power of the solar cell, please refer to "Ice Battery Maximum Power Tracking Method" published on September 21, 102, the output power of the solar cell is changed by a conversion unit and the DC voltage value is changed. Output to a load, the conversion unit is controlled by a pulse width modulation signal, and the output current and voltage of the solar cell also change when the pulse width ratio of the pulse width modulation signal is changed. The maximum power tracking method includes the following steps: (A) setting The first initial pulse width modulation signal, and the pulse width ratio of the pulse width modulation signals is small and large, and the first, second, and third pulse width ratios are sequentially transmitted to the conversion unit; (B) respectively Measure the output current and the output voltage value of the solar cell at the first, second, and third pulse width ratios; (C) calculate the output power of the solar cell at the first, second, and third pulse width ratios, And respectively, the first, second, and third output powers are respectively; (D) if the first, second, and third output powers are sequentially incremented, a pulse width interval value is obtained, and the second pulse width ratio is made New first pulse width ratio, order The three-pulse width ratio becomes the new second pulse width ratio, so that the third pulse width ratio plus the pulse width interval value becomes the new third pulse width ratio, and step B is repeated with the new first, second, and third pulse width ratios; E) if the first, second, and third output powers are sequentially decreased, then a pulse width interval value is obtained, and The second pulse width ratio becomes the new third pulse width ratio, so that the first pulse width ratio becomes the new second pulse width ratio, so that the first pulse width ratio minus the pulse width interval value becomes the new first pulse width ratio, and the new first Repeating step B with the second and third pulse width ratios; (F) if the second output power is greater than the first output power and the second output power is greater than the third output power, then the new parameter is obtained by the binomial curve formula a second pulse width ratio, and measuring a new second output power; (G) if the difference between the new second output power and the original second output power is greater than a predetermined ratio of the new second output power, then the new second pulse The width ratio is taken into the second-order curve formula to obtain another new second pulse width ratio, and the new second output power is calculated and measured, and step F is repeated; and (I) if the new second output power and the original second output power The difference between the two is less than a predetermined ratio of the new second output power, and the new second output power is the maximum output power.

Please refer to the "Control Method for Voltage Controlled DC/AC Power Converters with Maximum Power Tracking", published on January 11, 102, No. I382646, which includes: detecting a first AC voltage detector The voltage of the AC power system is sent to a bandpass filter, and the center frequency of the bandpass filter is the fundamental wave frequency of the AC power system, so that the bandpass filter obtains the fundamental wave component of the AC power system, wherein The fundamental wave component is a sine wave signal; the sine wave signal generated by the band pass filter is phase-shifted by 90 degrees by a phase shift circuit; the sine wave signal after a phase shift of 90 degrees by a multiplier And multiplying an output signal of a maximum power tracking control circuit to obtain a vertical vector signal; adding an vertical vector signal to the voltage signal detected by the first AC voltage detector by an adder to obtain an output voltage a reference signal; detecting, by a second AC voltage detector, an output voltage of an output filter of the DC/AC power converter and sending it to an input of a subtractor, the subtractor The other input is connected to the adder, and the subtractor subtracts the output voltage reference signal from the output voltage of the DC/AC power converter; and receives the output signal of the subtractor by a waveform control circuit and forms a modulation signal; receiving a modulation signal outputted by the waveform control circuit by a pulse width modulation circuit and sending the signal to a driving circuit, wherein the driving circuit generates a set of driving signals to control the power of the DC/AC power converter Electronic switch set.

Please refer to 2007023665, "Using Impedance Matching Method for Maximum Power Tracking Technology of Solar Photovoltaic System", which is disclosed on December 7, 1994. The steps include: measuring the open circuit voltage (V _oc ) value of the solar photovoltaic system and a a short circuit current (I _se ) value; dividing the open circuit voltage value by the short circuit current value and multiplying the correction constant (K) to obtain an impedance value in the solar photovoltaic system; determining the solar photovoltaic system according to the impedance value An optimal voltage value and an optimal current value; and determining a subsequent maximum power point based on the optimal voltage value and the optimal current value such that the solar photovoltaic system is maintained at the maximum power point.

Please refer to 201525643, “Recommended to estimate the maximum power point of the solar panel by the Fuzzy DR-LMS algorithm”, which is published on July 1, 104, including: obtaining the known illuminance value and the maximum power voltage value, and the illuminance value. For the input value of the Fuzzy DR-LMS filter, the maximum power voltage value is the output value of the filter, and the coefficient of the filter is adjusted by the Fuzzy DR-LMS algorithm using the estimation capability of the adaptive filter; Adjust the filter coefficient to obtain the relationship between the illuminance and the maximum power voltage value, and directly use the filter coefficient to estimate the maximum illuminance Power voltage; to eliminate the need for complex solar panel maximum power point tracking operation, simplify and accelerate the tracking process of solar energy maximum power.

Please refer to the 201421188 “Improving Illumination Citation Rate for Improving the Maximum Power Tracking of Solar Cells” published on June 1, 103, which includes: obtaining a considerable amount of actual illuminance data; intervening in grey prediction The first three stroke illuminance data is used to predict the predicted illuminance data at the next time point; the illuminance data is added to the actual illuminance data by using the interpolation method to add the predicted illuminance data; the maximum power voltage point sampling value is tracked by the disturbance observation method; The sampled value drives the control unit to gradually adjust the voltage so that the solar cell has a smoother power supply between the two samples.

Moreover, most of the current general solar cell maximum power tracking methods use the perturbation observation method for maximum power tracking. Although this method has the advantages of being simple and easy to implement, it still finds that the tracking speed is too slow in actual operation and implementation. And it is easy to add shock at the maximum efficiency point, so there is still room for improvement in the overall implementation.

Therefore, the inventor has in view of this, and has been rich in design development and actual production experience of the relevant industry for many years, and has researched and improved the existing structure and defects to provide a solar cell maximum power tracking method and system suitable for real-time online environment. In order to achieve better practical value for the purpose.

The main object of the present invention is to provide a suitable environment for the online environment. The maximum power tracking method and system of the solar battery can not only adjust the disturbance amount according to the current position, but also provide more accurate tracking control, and the reaction speed is very fast, so that the maximum power tracking is more suitable for the online environment. It can also reduce the burden on the microcontroller, so that it can be adapted to different environments throughout the year, and it is more practical and useful in its overall implementation.

The main purpose and effect of the method for adapting the maximum power of the solar cell in the instant online environment are achieved by the following specific technical means: the main method is that the maximum power tracking includes the fuzzy perturbation method built in the pre-stage microcontroller. And the neural network method [ANN] performed by the human-machine interface connected to the microcontroller at the subsequent stage; the fuzzy perturbation method [FMPPT] mainly uses the fuzzy inference rule to estimate the next disturbance amount: Disturbance observation method: by using the output voltage and current of the solar cell to the microcontroller of the maximum power tracking system, the PWM controller drives the PWM driver with different duty cycles to utilize the PWM driver The drive changes the output of the DC/DC converter and further changes the terminal voltage and output power of the solar cell; at the same time, observes the relevant illumination [ L _ux ] and temperature [ T ], and compares the DC/DC converter output The output voltage and output power of the solar cell before and after the change determine whether the next output is increased or decreased; Fuzzy perturbation method: the fuzzy inference engine determines the amount of the next disturbance. When the working point is far from the maximum power point [P _max ], the disturbance amount is large; otherwise, the disturbance amount is reduced, and the input power variation is [△ P] and voltage change amount [△V], and the output is the duty cycle adjustment amount [△D], and the two input variables are divided into seven fuzzy intervals to establish a fuzzy knowledge base; the form is: R _i : If ΔP is A ₁ and ΔV is B ₁ Then ΔD is C ₁ ; performing a neural network-like method [ANN]: using the fuzzy perturbation method of the pre-stage to collect input/output data by using a neural-like The network is learning and training. The input layer is 5 input variables, which are solar cell output voltage [ V _pv ], solar cell output current [ I _pv ], solar cell power [ P _pv ], illuminance [ L _ux ] and temperature [ T ], giving the initial input matrix x ⁽⁰⁾ = [ V _pv I _pv P _pv L _ux T ] ^T , expecting the output d , and randomly generating the weight matrix w ⁽¹⁾ and w ⁽²⁾ , the bias weight matrix b ⁽¹⁾ and b ^(2), whose value uniformly distributed in [0,1], of which the desired output is d Fuzzy Output voltage (FMPPT (V _out)) of the dynamic method, the second layer 5 containing neurons in the hidden layer, and therefore a total of 25 weights _{w ji (j = i = 1} ~ 5) with 5 weight bias b _{h ( h =1~5)} , the third-layer output layer is the duty cycle change required to reach the maximum power, consisting of one neuron, and contains a total of five weights w _{k ( k =1~5)} With one bias value b ₁ ; perform forward propagation operation, Hidden layer output y ⁽¹⁾ = f ⁽¹⁾ ( net ⁽¹⁾ ), where f ⁽¹⁾ is a hyperbolic transfer function, Output layer output y ⁽²⁾ = f ⁽²⁾ ( net ⁽²⁾ ), where f ⁽²⁾ is a hyperbolic transfer function and y ⁽²⁾ is the neural network output voltage (ANN( V _out )), Error δ ⁽²⁾ = d - y ⁽²⁾ = FMPPT( V _out )-ANN( V _out ); during the training process, in order to enable the training data to cover all possible environmental conditions, the retraining mechanism is performed; The training mechanism is: a. In the post-stage tracking phase, the fuzzy perturbation method is also executed once every other time, and the output voltage generated is compared with the output voltage of the post-class neural network. When the error between the two is greater than 1% , the relevant information is collected; b. the human-machine interface is activated for retraining; c. The trained weight is transmitted to the microcontroller through the communication module for maximum power tracking control.

The present invention is applicable to a preferred embodiment of a solar cell maximum power tracking method in a real-time online environment, wherein further performing a backpropagation operation and correcting a hidden layer and an output layer weight using a minimum mean square error criterion Adjust the effect of the output layer weight w ⁽²⁾ on E , Adjust the effect of the output layer weight w ⁽¹⁾ on E , Adjust the weight of each layer, Where w ⁽¹⁾ ( t +1) is the weight of the ( t +1) time [or iteration number] hidden layer, and w ⁽²⁾ ( t +1) is the ( t +1) time output layer The weight, the α is a constant constant [momentum constant], the η is a learning rate constant, and the values of α and η are between 0 and 1.

The present invention is applicable to a preferred embodiment of a solar cell maximum power tracking method in a real-time online environment wherein the alpha value is between 0.5 and 0.99 and the η value is between 0.01 and 0.5.

The invention is applicable to the maximum power tracking of solar cells in a real-time online environment A preferred embodiment of the method, wherein the human interface is a LabVIEW-Matlab interface for feeding data into Matlab for retraining.

The main purpose and function of the solar cell maximum power tracking system applicable to the instant online environment are achieved by the following specific technical means: the method includes the maximum power tracking method of the solar cell suitable for the on-line environment, and the main system is the largest The power tracking system comprises a microcontroller, a PWM (Pulse Width Modulation) driver, a DC/DC converter, a communication module and a human machine interface; wherein: the microcontroller is configured to receive and receive a solar cell Voltage, current signal, and ambient illuminance and temperature data, and a fuzzy perturbation method [FMPPT] is built in the microcontroller; the PWM driver is connected to the microcontroller to output different signals from the microcontroller The PWM signal of the duty cycle drives the PWM driver; the DC/DC converter is connected to the PWM driver, and the DC/DC converter is connected to the solar cell, and the PWM driver can be used to drive the DC/DC conversion Outputting the communication module, which is connected to the microcontroller; the human machine interface, and the communication module The group is connected, and the human-machine interface is trained in the neural network method, and the trained weight is transmitted to the microcontroller via the communication module.

The present invention is applicable to a preferred embodiment of a solar cell maximum power tracking system in a real-time online environment, wherein the output of the DC/DC converter is connected to either a DC load or a battery.

The present invention is applicable to a preferred embodiment of a solar cell maximum power tracking system in a real-time online environment, wherein the output of the DC/DC converter simultaneously connects the DC load to the battery.

The present invention is applicable to a preferred embodiment of a solar cell maximum power tracking system for a real-time online environment, wherein the DC/DC converter is a SEPIC converter.

The present invention is applicable to a preferred embodiment of a solar cell maximum power tracking system in a real-time online environment, wherein the communication module performs signal conversion between the RS-485 interface and the TCP/IP interface.

The present invention is applicable to a preferred embodiment of a solar cell maximum power tracking system in a real-time online environment, wherein the human-machine interface adopts LabVIEW graphics monitoring software, and the Matlab software is used to provide a neural network code in the human-machine interface.

(1)‧‧‧Maximum power tracking system

(11)‧‧‧Microcontrollers

(12)‧‧‧PWM driver

(13)‧‧‧DC/DC converter

(14) ‧‧‧Communication Module

(15) ‧‧‧Human Machine Interface

(2) ‧‧‧ solar cells

(3) ‧ ‧ DC load

(4) ‧‧‧Battery

First: Schematic diagram of the system architecture of the present invention

Second: the solar cell power [P _PV ] and voltage [V _PV ] curves of the present invention

Third figure: Schematic diagram of the action flow of the disturbance observation method of the present invention

Figure 4: Block diagram of the fuzzy control system of the present invention

Figure 5: Schematic diagram of the input fuzzy attribution function of the present invention [power variation]

Figure 6: Schematic diagram of the input fuzzy attribution function of the present invention [voltage variation]

Figure 7: Schematic diagram of the output fuzzy attribution function of the present invention [change of duty cycle]

Figure 8: Schematic diagram of the maximum power tracking architecture of the neural network of the present invention

Figure 9: Schematic diagram of the maximum power tracking process of the present invention

Figure 10: Maximum power tracking condition curve of the fuzzy perturbation method [FMPPT] of the present invention (illuminance 20000 lux)

Figure 11: Maximum power tracking condition curve of the fuzzy perturbation method [FMPPT] of the present invention (illuminance 40000 lux)

Twelfth figure: The maximum power tracking condition curve of the neural network method [ANN] of the present invention (illuminance 20000 lux)

Thirteenth figure: The maximum power tracking condition curve of the neural network method [ANN] of the present invention (illuminance 40,000 lux)

Figure 14: Comparison of the tracking state of the fuzzy perturbation method [FMPPT] and the neural network-like method [ANN] of the present invention

The fifteenth figure: the fuzzy perturbation method [FMPPT] and the neural network of the present invention Magnification of the area near the roadway [ANN] tracking point

Figure 16: Comparison of the tracking state of the fuzzy perturbation method [FMPPT] and the neural network-like method [ANN] after the retraining of the present invention

Figure 17: Enlarged map of the vicinity of the tracking point after the retraining fuzzy perturbation method [FMPPT] and the neural network method [ANN] of the present invention

For a more complete and clear disclosure of the technical content, the purpose of the invention and the effects thereof achieved by the present invention, the following is a detailed description, and please refer to the drawings and drawings: First, please refer to 1 is a schematic diagram of a system architecture of the present invention. The present invention mainly provides a maximum power tracking system (1) including a microcontroller (11), a PWM (Pulse Width Modulation) driver (12), a DC/DC converter (13), a communication module (14), and a human-machine interface (15); wherein: the microcontroller (11) is configured to receive and receive the voltage, current signal, and environment of the solar cell (2) Illumination and temperature data, and a fuzzy perturbation method [FMPPT] is built in the microcontroller (11); the PWM driver (12) is connected to the microcontroller (11) to be used by the microcontroller ( 11) output PWM signals with different duty cycles to drive the PWM driver (12); The DC/DC converter (13) is connected to the PWM driver (12), and the DC/DC converter (13) is connected to the solar cell (2), and the DC/DC converter (13) The output end is connected to either the DC load (3), the battery (4), or the output of the DC/DC converter (13) is simultaneously connected to the DC load (3) and the battery (4), the DC The /DC converter (13) can be a SEPIC converter, so that not only the output voltage has no opposite polarity problem, but also can perform the boosting operation, and the switching circuit can be operated to boost when the duty cycle ratio D is adjusted. Or stepping down to increase the flexibility of the type and voltage range of the solar cell (2), and the PWM driver (12) can be used to drive the DC/DC converter (13) for output; the communication module (14), It is connected to the microcontroller (11), and the communication module (14) can perform signal conversion between the RS-485 interface and the TCP/IP interface; the human interface (15), and the communication module ( 14) Connection, the human-machine interface (15) adopts the LabVIEW graphical monitoring software, and the Matlab software provides the class in the human-machine interface (15). The neural network code enables the Matlab Script Node function in the LabVIEW graphical monitoring software to be trained in the neural network method in the Matlab environment, and the trained weights are transmitted through the communication module (14). Transfer to the microcontroller (11).

The present invention, in operational use, includes a pre-stage microcontroller (11) The fuzzy perturbation method built in and the neural network method [ANN] performed by the human-machine interface (15) in the latter stage.

The fuzzy perturbation method [FMPPT] mainly uses the fuzzy inference rule to estimate the next disturbance amount: it first performs the disturbance observation method: mainly by returning the output voltage and current of the solar cell (2) to the maximum power. The microcontroller (11) of the tracking system (1) drives the PWM driver (12) by sending a PWM signal of different duty cycle by the microcontroller (11) to use the PWM driver (12) to drive and change the DC /DC converter (13) output, and further change the terminal voltage and output power of the solar cell (2); at the same time, observe the relevant illuminance [ L _ux ] and temperature [ T ], and compare the DC / DC conversion The output voltage and the output power of the solar cell (2) before and after the output of the device (13) determine whether the next output is increased or decreased.

Please refer to the second diagram of the solar cell power [P _PV ] and voltage [V _PV ] graphs of the present invention, and set the left side of the maximum power point [P _max ] to the A area and the right side to the B area; In the A zone, if the power is to be moved to the maximum power point [P _max ], the output voltage of the solar cell (2) must be increased, that is, the duty cycle ratio D is lowered; and in the B zone, the power is required to be the maximum power point. When [P _max ] is moved, the output voltage of the solar cell (2) must be lowered, that is, the duty cycle ratio D is increased, and the amount of increase or decrease is called the disturbance amount. Please refer to the third figure again. The schematic diagram of the operation diagram of the disturbance observation method of the present invention shows that after reading the voltage and current of the solar cell (2), the output power is calculated, if the current output power is greater than the previous output power. , the microcontroller (11) will adjust the duty cycle [D] to make the output power change in the same direction; conversely, if the output power is less than the previous output power, the output power is changed in the next duty cycle [D]. The direction of change.

Then the fuzzy perturbation method is used: the fuzzy inference engine determines the amount of the next disturbance. When the working point is far from the maximum power point [P _max ], the disturbance amount is large; otherwise, the disturbance amount is reduced. Please refer to the fourth block diagram of the fuzzy control system of the present invention as shown in the figure, which is the input power variation [ΔP] and the voltage variation [ΔV], and the output is the duty cycle adjustment amount [△D]. Please refer to the fifth diagram for the fuzzy attribution function of the input and output of the present invention [power variation], and the sixth diagram of the fuzzy attribution function of the input and output of the present invention [voltage variation] and the seventh diagram. The fuzzy assignment function of the input and output of the invention is shown in the diagram of the duty cycle change, where LN is large negative, MN is medium negative, SN is small negative, ZE is zero, LP is large positive, and MP is The positive and the SP are small positive. Since the two input variables are divided into seven fuzzy intervals, the knowledge base will contain 49 [7×7] inference engines, as shown in Table 1:

Its form is as follows: R _i : If ΔP is A ₁ and ΔV is B ₁ Then ΔD is C ₁

For example, the 10th fuzzy rule: R ₁₀ : If ΔP is MN and ΔV is SN Then ΔD is SN

R ₁₀ indicates that if the amount of power change is medium negative [MN], that is, the power is decreased, and the voltage change amount is small negative [SN], it is judged that the operation is in the A zone, and the voltage is required to increase the previous maximum power point [P _max 〕 Move [from A2 to A1], so the duty cycle [D] must be small negative [SN], that is, the small amplitude is small.

Take the 45th fuzzy rule as an example: R ₄₅ : If △P is LP and △V is SN Then △D is MP

R ₄₅ shows that if the power change amount is large positive [LP], that is, the power is greatly increased, and the voltage change amount is small negative [SN], it is judged that the operation is in the B zone, and the voltage needs to be reduced to continue to the maximum power point. Move [from B2 to B1], so the duty cycle change must be medium positive [MP], that is, the medium amplitude is increased.

Perform Neural Network-like Method [ANN]: Please refer to the eighth diagram again. The neural network of the present invention is shown in the schematic diagram of the maximum power tracking architecture. The data training using the neural-like inverse transfer algorithm can be divided into the following step:

Step 1: Give the initial input matrix x ⁽⁰⁾ = [ V _pv I _pv P _pv L _ux T ] ^T , expect the output d , and randomly generate the weight matrix w ⁽¹⁾ and w ⁽²⁾ , the weighted matrix b ⁽¹⁾ and b ⁽²⁾ , whose values are evenly distributed between [0, 1], where the expected output d is the output voltage of the fuzzy perturbation method (FMPPT( V _out )).

Step 2: Perform a forward propagation operation

Hidden layer output y ⁽¹⁾ = f ⁽¹⁾ ( net ⁽¹⁾ ), where f ⁽¹⁾ is a hyperbolic transfer function,

Output layer output y ⁽²⁾ = f ⁽²⁾ ( net ⁽²⁾ ), where f ⁽²⁾ is a hyperbolic transfer function and y ⁽²⁾ is the neural network output voltage (ANN( V _out )), Error δ ⁽²⁾ = d - y ⁽²⁾ = FMPPT( V _out )-ANN( V _out )

Step 3: Perform a backpropagation operation

Correcting the hidden layer and output layer weights using the minimum mean square error criterion, then

Step 3.1: Adjust the impact of the output layer weight w ⁽²⁾ on E

Step 3.2: Adjust the impact of the output layer weight w ⁽¹⁾ on E

Step 3.3: Adjust the weight of each layer

Where w ⁽¹⁾ ( t +1) is the weight of the ( t +1) time [or iteration number] hidden layer, and w ⁽²⁾ ( t +1) is the ( t +1) time output layer The weight, α is a constant constant [momentum constant], η is a learning rate constant, and α and η are usually set by the user based on experience or experiment, and the value is between 0 and 1. In general, the alpha value is between 0.5 and 0.99, and the η value is between 0.01 and 0.5. The magnitude of α affects the learning convergence rate, and the value of η affects the learning effect.

Step 4: Repeat steps 2 and 3 until the set number of iterations is reached or the program converges within the error range.

The following is a detailed description of the data training by the neural network-like method, using the input/output data collected by the fuzzy perturbation method of the pre-stage microcontroller (11), and the human-machine interface (15) for the latter stage. For learning and training through a neural network, please refer to the eighth diagram. The neural network of the present invention is shown in the schematic diagram of the maximum power tracking architecture. The input layer is 5 input variables, respectively, the solar cell output voltage [ V _Pv ], solar cell output current [ I _pv ], solar cell power [ P _pv ], illuminance [ L _ux ] and temperature [ T ], the second layer is a hidden layer containing 5 neurons, so there are a total of 25 rights The value w _{ji ( j = i =1~5)} and the 5 partial weights b _{h ( h =1~5)} , the third layer of the output layer is the duty cycle change required to reach the maximum power, by 1 neuron The total consists of five weights W _{k ( k =1~5)} and one partial weight b ₁ .

In the above process, the fuzzy perturbation method used by the pre-stage micro-controller (11) can overcome the shortcoming of the conventional perturbation observation method at the maximum power point, and can accurately and quickly track the maximum power point [P _max ]. In the latter stage, the human-machine interface (15) uses the input/output data collected by the pre-stage micro-controller (11) to perform learning training through the neural network, and during the training process, the training data is used. It can cover all possible environmental conditions. In the present invention, the retraining mechanism is utilized, which is: 1. In the latter stage of tracking, the microcontroller (11) fuzzy perturbation method is used at regular intervals (such as 15 seconds). Execute once, the output voltage produced by it is compared with the output voltage of the human-machine interface (15) neural network in the latter stage, when the error between the two is greater than 1% [ie ], then collect the relevant data; 2. Start the communication module (14) LabVIEW-Matlab interface, feed the data into Matlab for retraining; 3. Pass the trained weight through the communication module (14) RS485 to TCP is transmitted to the microcontroller (11) for maximum power tracking control (please refer to the ninth diagram of the present invention for the maximum power tracking process diagram).

In this way, when the invention is tested in the experiment, the solar cell (2) has a maximum output power P _max = 25 W, an open circuit voltage V _OC = 21.7 V, a short-circuit current I _SC = 1.31 A, and the battery (4). Charging voltage 14.2~15V, the battery (4) capacity 17Ah, please refer to the tenth figure of the present invention, the fuzzy perturbation method [FMPPT] maximum power tracking condition graph (illuminance 20000 lux), the eleventh figure of the present invention Fuzzy perturbation method [FMPPT] maximum power tracking condition graph [illuminance 40000 lux], twelfth map, neural network method [ANN] of the present invention, maximum power tracking condition graph (illuminance 20000 lux), thirteenth map The neural network method [ANN] of the invention, the maximum power tracking condition graph (illuminance 40000 lux), the fourteenth graph, the fuzzy perturbation method [FMPPT] of the present invention and the neural network method [ANN] tracking state comparison graph and The fifteenth figure shows the enlarged perturbation map of the vicinity of the tracking point of the fuzzy perturbation method [FMPPT] and the neural network method [ANN] of the present invention, and since it compares the average percentage error e%=1.4935% [>1%], it starts Retraining mechanism The weights of the trained layers are as follows: hidden layer weights

Hidden layer bias

Output layer weight

Output layer bias value

Please refer to the sixteenth figure for the retraining fuzzy perturbation method [FMPPT] and the neural network method [ANN] tracking condition comparison graph and the seventeenth graph of the present invention. The method [FMPPT] and the neural network method [ANN] show an enlarged view of the vicinity of the tracking point. After training, the average error has been reduced to e%=0.732%. The above average error is defined as follows:

From the above, the implementation description of the present invention shows that the present invention has the following advantages in comparison with the prior art means:

1. The present invention uses a fuzzy perturbation method [FMPPT] in the pre-stage microcontroller. It can adjust the amount of disturbance according to the current position, so it can avoid the disadvantage of oscillating near the maximum power point and provide more accurate tracking control.

2. The invention uses the neural network method [ANN] in the human-machine interface of the latter stage, so that its input and output only need some algebraic operation, and the reaction speed is very fast, and the maximum power tracking is more suitable for the real-time online environment.

3. The retraining mechanism of the present invention is trained in the Matlab environment by the Matlab Script Node function of the LabVIEW-Matlab interface of the human machine interface, in addition to reducing the burden on the microcontroller, and making the maximum power tracking control system suitable for the whole year. Different environments.

However, the above-described embodiments or drawings are not intended to limit the structure or the use of the present invention, and any suitable variations or modifications of the invention will be apparent to those skilled in the art.

In summary, the embodiments of the present invention can achieve the expected use efficiency, and the specific structure disclosed therein has not been seen in similar products, nor has it been disclosed before the application, and has completely complied with the provisions of the Patent Law. And the request, the application for the invention of a patent in accordance with the law, please forgive the review, and grant the patent, it is really sensible.

Claims (10)

A solar cell maximum power tracking method suitable for a real-time online environment, mainly for the maximum power tracking including a fuzzy disturbance method built in a pre-stage microcontroller and a human-machine interface connected to the microcontroller at a later stage. The neural network method [ANN]; the fuzzy perturbation method [FMPPT] mainly uses the fuzzy inference rule to estimate the next disturbance amount: it first performs the disturbance observation method: by feeding back the output voltage and current of the solar cell to The microcontroller of the maximum power tracking system sends a PWM signal driving PWM driver with different duty cycles by the microcontroller to use the PWM driver to drive the output of the DC/DC converter, and further change the solar cell The terminal voltage and output power; at the same time, observe the relevant illuminance [ L _ux ] and temperature [ T ], and compare the output voltage and output power of the solar cell before and after the DC/DC converter output change to determine the next time. The output is increased or decreased; then the fuzzy perturbation method is used: the fuzzy inference engine determines the amount of the next disturbance, When the working point is far from the maximum power point [P _max ], the disturbance amount is large; otherwise, the disturbance amount, the input power variation amount [ΔP] and the voltage variation amount [ΔV] are reduced, and the output is the duty cycle adjustment. The quantity [△D] divides the two input variables into seven fuzzy intervals to establish a fuzzy knowledge base; its form is: R _i : If ΔP is A ₁ and ΔV is B ₁ Then △D is C ₁ Performing a neural network method [ANN]: using the input/output data collected by the fuzzy perturbation method of the pre-stage to learn training through a neural network, the input layer is 5 input variables, respectively, solar cells Output voltage [ V _pv ], solar cell output current [ I _pv ], solar cell power [ P _pv ], illuminance [ L _ux ] and temperature [ T ], given initial input matrix x ⁽⁰⁾ = [ V _pv I _pv P _Pv L _ux T ] ^T , expects to output d , and randomly generates weight matrix w ⁽¹⁾ and w ⁽²⁾ , bias weight matrix b ⁽¹⁾ and b ⁽²⁾ , whose values are evenly distributed in [0, 1 ] room, where d is the desired output of the fuzzy output voltage perturbation method (FMPPT (V _out)), a second layer 5 containing neurons Hidden layer, so that a total of 25 weights _{w ji (j = i = 1} ~ 5) and five bias weights _{b h (h = 1 ~ 5} ), the third layer is the output layer to achieve the desired maximum power responsibility The amount of periodic variation consists of one neuron, which contains a total of five weights w _{k ( k =1~5)} and one partial weight b ₁ ; performs a forward propagation operation, Hidden layer output y ⁽¹⁾ = f ⁽¹⁾ ( net ⁽¹⁾ ), where f ⁽¹⁾ is a hyperbolic transfer function, output layer net output net ⁽²⁾ = w ⁽²⁾ × y ⁽¹⁾ + b ⁽²⁾ Output layer output y ⁽²⁾ = f ⁽²⁾ ( net ⁽²⁾ ), where f ⁽²⁾ is a hyperbolic transfer function and y ⁽²⁾ is the neural network output voltage (ANN( V _out )), Error δ ⁽²⁾ = d - y ⁽²⁾ = FMPPT( V _out )-ANN( V _out ); during the training process, in order to enable the training data to cover all possible environmental conditions, the retraining mechanism is performed; The training mechanism is: a. In the post-stage tracking phase, the fuzzy perturbation method is also executed once every other time, and the output voltage generated is compared with the output voltage of the post-class neural network. When the error between the two is greater than 1% , the relevant information is collected; b. the human-machine interface is activated for retraining; c. The trained weight is transmitted to the microcontroller through the communication module for maximum power tracking control. For example, the maximum power tracking method of the solar cell applicable to the on-line environment as described in the first application of the patent scope, further performing a backpropagation operation, and correcting the hidden layer and the output layer weight using the minimum mean square error criterion, Adjust the effect of the output layer weight w ⁽²⁾ on E , =- δ ⁽²⁾ y ⁽²⁾ (1- y ⁽²⁾ ) y ⁽¹⁾ , adjust the effect of the output layer weight w ⁽¹⁾ on E , Adjust the weight of each layer, Where w ⁽¹⁾ ( t +1) is the weight of the ( t +1) time [or iteration number] hidden layer, and w ⁽²⁾ ( t +1) is the ( t +1) time output layer The weight, the α is a constant constant [momentum constant], the η is a learning rate constant, and the values of α and η are between 0 and 1. The maximum power tracking method for a solar cell applicable to a real-time online environment as described in claim 2, wherein the alpha value is between 0.5 and 0.99, and the η value is between 0.01 and 0.5. The maximum power tracking method for a solar cell applicable to a real-time online environment as described in claim 1 of the patent application scope, wherein the human-machine interface is a LabVIEW-Matlab interface, and the data is fed into Matlab for retraining. A solar cell maximum power tracking system suitable for use in a real-time online environment, comprising the method of claim 1, wherein the maximum power tracking system includes a microcontroller, PWM (Pulse Widt) h Modulation, pulse width modulation] driver, DC/DC converter, communication module and human-machine interface; wherein: the microcontroller is connected to receive the voltage, current signal, and ambient illumination and temperature of the solar cell Data, and a fuzzy perturbation method (FMPPT) is built in the microcontroller; the PWM driver is connected to the microcontroller, and the PWM driver is driven by the microcontroller to output PWM signals with different duty cycles; the DC a DC/DC converter coupled to the PWM driver and having the DC/DC converter coupled to the solar cell, wherein the PWM driver is used to drive the DC/DC converter for output; the communication module, and the a microcontroller connection; the human interface is connected to the communication module, the human-machine interface is trained in the neural network method, and the trained weight is transmitted to the microcontroller via the communication module . The solar cell maximum power tracking system applicable to the on-line environment as described in claim 5, wherein the output of the DC/DC converter is connected to either a DC load or a battery. A solar cell maximum power tracking system suitable for use in a real-time online environment as described in claim 5, wherein the output of the DC/DC converter simultaneously connects the DC load to the battery. Solar cells suitable for use in a real-time online environment as described in claim 5 A maximum power tracking system in which the DC/DC converter is a SEPIC converter. The solar cell maximum power tracking system is applicable to the instant online environment as described in claim 5, wherein the communication module performs signal conversion between the RS-485 interface and the TCP/IP interface. The solar cell maximum power tracking system applicable to the on-line environment as described in claim 5, wherein the human-machine interface adopts LabVIEW graphics monitoring software, and the Matlab software is used to provide a neural network program in the human-machine interface. code.