JP7355565B2 - Solar cell group manufacturing equipment, wall structure manufacturing equipment, and machine learning programs - Google Patents

Description

The present invention relates to a solar cell group manufacturing apparatus and a wall structure manufacturing apparatus. The present invention also relates to a machine learning program that can be suitably used when manufacturing solar cell groups.

BACKGROUND ART Conventionally, a so-called back-contact type solar cell in which an electrode is provided only on the back surface side and no electrode is provided on the light-receiving surface side has been known (for example, Patent Document 1).
In back-contact solar cells, the electrodes are concentrated on the back surface, which allows the solar cell to have a larger light-receiving surface and capture more light.
In addition, in solar cell modules that use back-contact solar cells, the wiring members that connect each solar cell are also provided on the back side, so indoors, etc., it is difficult to give viewers the impression that the appearance is uniform. can.

Back-contact solar cells are provided with an anti-reflection film on the light-receiving surface side in order to confine the received light within the solar cell, and the appearance of the solar cell is largely determined by the anti-reflection film.

Japanese Patent Application Publication No. 2018-170482

Normally, in the production of mass-produced solar cells, an antireflection film is formed on the light-receiving surface of a large number of solar cells at the same time under the same conditions. However, even if the anti-reflective film is formed under the same conditions in each solar cell, there may be slight differences in the anti-reflective film between the solar cells due to the effects of the film forming position, heating temperature, etc. when manufacturing the anti-reflective film. The thickness may be different, or the refractive index of the antireflection film may be different.

In such a case, if you arrange solar cells randomly and connect them with wiring members to create a module, it will hardly be felt indoors with low illuminance as shown in Fig. 15(a), but as shown in Fig. 15(b). Color unevenness may appear when sunlight shines through. Therefore, in conventional solar cell modules, when viewed under sunlight, there is color unevenness between the solar cells, poor color balance, and problems with the uniformity of color elements.

Moreover, a solar cell module may form one wall surface using a plurality of solar cell modules. In such a case, if the color balance of the solar cells in each solar cell module is different, the color balance of the entire wall surface will be poor when the wall surface is formed, and there is a problem in that the uniformity of color elements is affected.

Therefore, an object of the present invention is to provide an apparatus for manufacturing a solar cell group that gives a good impression of good color balance even when a person visually recognizes the solar cell group under sunlight.

In response to the above-mentioned problem, the present inventor considered the following.
It is thought that the above problem can be solved by simply using solar cells with similar color elements side by side to improve color balance. However, if only solar cells with similar color elements are simply used, there will be a problem that some solar cells will be unusable even though there is no abnormality in performance, resulting in a decrease in yield. Therefore, in order to maintain yield, it is preferable to manufacture a solar cell module using as many solar cells as possible.

By the way, the human brain recognizes images recognized by the eyes as still images, and recognizes movement by connecting still images. As a result, there is an optical illusion, where even if the color is uniform, it may appear mottled, or even if the color distribution is not uniform, it may appear to be uniform.
Therefore, the present inventor wondered if it would be possible to utilize this human optical illusion to use more solar cells.

However, although human optical illusions tend to occur, the principles behind them are often not understood scientifically, and we have no choice but to consider the arrangement based on what has actually been determined by humans. As a result, an optical illusion occurs from among a huge number of combinations of solar cells, and it is difficult to artificially derive combinations that appear to have good color balance.
Therefore, the inventor of the present invention incorporated a machine learning program into the manufacturing equipment and made the machine learn to judge the balance between the color and arrangement of solar cells and the human color, thereby making the human feel that the color balance is good. The idea was to have the manufacturing equipment itself derive the layout.

One aspect of the present invention derived based on the above idea includes a placement operation section for arranging a plurality of solar cells constituting a solar cell group, and a machine learning section, wherein the solar cell group solar cells are arranged in a plane, each of the plurality of solar cells has a light-receiving surface, and an anti-reflection film is formed on part or all of the light-receiving surface. In some cases, the thickness of the anti-reflection film or the refractive index of the anti-reflection film may vary, resulting in variations in color elements. The machine learning unit performs machine learning using the correlation with the determination result of determining the color balance of the solar cell group in the arrangement of the plurality of solar cells as training data, and the machine learning unit manufactures the solar cell group. At this time, a placement model of the solar cells that is predicted to be judged as having a good color balance of the solar cell group when visually recognized by a person is generated based on information on the color elements of each solar cell, and the placement operation is performed. The part is a solar cell group manufacturing apparatus that arranges each solar cell based on the arrangement model.

"Color balance" here includes not only homogeneity, but also the color balance of the designs and patterns when they are made of solar cells.

Although the solar cells of this aspect are formed through the same manufacturing process, the difference in the thickness of the anti-reflection film or the refractive index of the anti-reflection film causes disturbances in color elements between the solar cells. It is something.
According to this aspect, the machine learning section judges the color balance based on human subjectivity, and the machine learning section determines that the color balance of the solar cell group is good based on slight disturbances in color elements. Then, a predicted placement model is generated. Therefore, the arrangement can be set based on human optical illusions, and more solar cells can be used to manufacture solar cell groups. Therefore, the yield can be improved compared to simply arranging items with similar color distributions.
Furthermore, according to this aspect, since the color balance can be determined within the manufacturing apparatus in a way that is close to human sensitivity, it is also possible to automate the placement of solar cells.

In a preferred aspect, the solar cell group is a solar cell module in which the plurality of solar cells are electrically connected by a wiring member, and each solar cell is connected by the wiring member on a side opposite to the light receiving surface. This is what is being done.

A preferred aspect is that the solar cell is a solar cell module including a plurality of solar cells sandwiched between two sealing members.

By the way, under sunlight with high illuminance, variations in brightness are more noticeable than variations in chromaticity.

Therefore, a preferable aspect is that some of the plurality of solar cells have variations in brightness due to differences in the thickness of the anti-reflection film or the refractive index of the anti-reflection film, and the machine learning unit , the machine learning is performed using a correlation between the arrangement of the plurality of solar cells and a judgment result obtained by a person judging the variation in brightness of the group of solar cells in the arrangement of the plurality of solar cells as training data, and the machine When manufacturing the solar cell group, the learning section is configured to determine, based on the information on the brightness of each solar cell, that the variation in brightness of the plurality of solar cells in the solar cell group appears small when visually recognized by a person. The goal is to generate a predicted placement model.

In a preferred aspect, the distribution of color elements of 500 or more solar cells is obtained, and a predetermined number of solar cells are extracted from the 500 or more solar cells so that the distribution of the color elements is substantially maintained. The machine learning unit uses, as training data, the correlation between the arrangement of the predetermined number of solar cells and the determination result of determining the color balance of a solar cell group extracted by a person and in which the predetermined number of solar cells are arranged. , is to do machine learning.

Here, "extracting a predetermined number of solar cells so that the distribution of color elements is substantially maintained" means that when the color distribution of a predetermined number of solar cells as samples is obtained, the tendency of the color distribution is normal. It means that it matches the tendency of the distribution of color elements in the group.

A preferred aspect is that the machine learning section performs learning based on a neural network having four or more layers.

One aspect of the present invention includes a placement operation section for arranging a plurality of solar cells constituting a solar cell group, a machine learning section, and a second machine learning section, wherein the solar cell group The plurality of solar cells have a light-receiving surface, an anti-reflection film is formed on part or all of the light-receiving surface, and the plurality of solar cells have a light-receiving surface. In some cases, variations in color elements occur due to differences in the thickness of the anti-reflection film or the refractive index of the anti-reflection film, and the machine learning unit is configured to determine the color elements and arrangement of the plurality of solar cells in advance. , machine learning is performed using the correlation with the judgment result of human judgment of the color balance of the solar cell group as training data, and the color balance of the solar cell group is machine learned based on information on the color elements and arrangement of each solar cell. It is possible to predict the determination result of determining the color balance of the solar cell group at the time, and the second machine learning unit replaces the arrangement of the plurality of solar cells in the solar cell group, Information on the color elements and arrangement of the solar cells is given to the machine learning unit to determine the color balance of the solar cell group, and the correlation between the arrangement of the plurality of solar cells and the determination result determined by the machine learning unit. The second machine learning section predicts that when manufacturing the solar cell group, the machine learning section will determine that the color balance of the solar cell group is better. The solar cell group manufacturing apparatus generates an arrangement model of the solar cells based on the arrangement model, and the arrangement operation section arranges each solar cell based on the arrangement model.

Although the solar cells of this aspect are formed through the same manufacturing process, the difference in the thickness of the anti-reflection film or the refractive index of the anti-reflection film causes disturbances in color elements between the solar cells. It is something.
The machine learning part of this aspect performs machine learning in advance on the correlation between the color elements and arrangement of solar cells and the judgment results of human judgment of the color balance of a group of solar cells as training data, and makes judgments close to human subjectivity. Have standards.
According to this aspect, the second machine learning unit performs machine learning based on the judgment of color balance by the machine learning unit, which is similar to human subjectivity, and predicts that a person will judge that the color balance of the solar cell group is good. The second machine learning unit generates a placement model. Therefore, the arrangement can be set based on human optical illusions, and more solar cells can be used to manufacture solar cell groups. Therefore, the yield can be improved compared to simply arranging items with similar color distributions.
Furthermore, according to this aspect, since the color balance can be determined within the manufacturing apparatus in a way that is close to human sensitivity, it is also possible to automate the placement of solar cells.
Furthermore, according to this aspect, since the machine learning section provides the teacher data to the second machine learning section, the teacher data can be provided to the second machine learning section without human judgment.

One aspect of the present invention includes a placement operation section for arranging a plurality of solar cell modules constituting a wall structure, and a machine learning section, wherein the wall structure has a plurality of solar cell modules arranged in a plane. The plurality of solar cell modules have a light-receiving surface, and a solar cell is sandwiched between two sealing members, and the plurality of solar cell modules have a light-receiving surface. An antireflection material is interposed between the sealing member and the solar cell, and the sealing member on the light-receiving surface side has translucency, and there is a In some cases, color elements may vary due to differences in the thickness of the anti-reflection material or the refractive index of the anti-reflection material, and the machine learning section is designed to determine the arrangement of the plurality of solar cell modules and the human ability to Machine learning is performed using the correlation with the determination result of determining the color balance of the wall structure in the arrangement of the plurality of solar cell modules as training data, and the machine learning unit performs machine learning when manufacturing the wall structure. , based on the information on the color elements of each solar cell module, generate a placement model of the solar cell module that is predicted to be judged as having good color balance of the wall structure when visually recognized by a person; is a wall structure manufacturing apparatus that arranges each solar cell module based on the arrangement model.

Although the solar cell modules of this aspect are formed through the same manufacturing process, due to the difference in the thickness of the antireflection material or the refractive index of the antireflection material, color elements may appear between the solar cell modules. Disturbances occur.
According to this aspect, the machine learning section judges the color balance based on human subjectivity, and the machine learning section determines that the color balance of the solar cell module is good based on slight disturbances in color elements. Then, a predicted placement model is generated. Therefore, the arrangement can be set based on the illusion of the human eye, and more solar cell modules can be used in the wall structure. Therefore, the yield can be improved compared to simply arranging items with similar color distributions.
Further, according to this aspect, since the color balance can be determined in the manufacturing apparatus close to human sensitivity, it is also possible to automate the arrangement of solar cell modules.

One aspect of the present invention is a machine learning program that provides a placement model to a placement operation unit that places a plurality of solar cells constituting a solar cell group based on a placement model, wherein the solar cell group solar cells are arranged in a plane, each of the plurality of solar cells has a light-receiving surface, an anti-reflection film is formed on a part or all of the light-receiving surface, and the plurality of solar cells have In some cases, variations in color elements occur due to differences in the thickness of the anti-reflection film or the refractive index of the anti-reflection film, and the arrangement of the plurality of solar cells and the difference in the number of solar cells This is a machine learning method that uses the correlation with the judgment result of determining the color balance of the solar cell group in the arrangement as training data, and based on the information on the color elements of each solar cell, when visually recognized by a human. This is a machine learning program that generates a solar cell placement model that is predicted to determine that the color balance of the solar cell group is good.

Although the solar cells of this aspect are formed through the same manufacturing process, the difference in the thickness of the anti-reflection film or the refractive index of the anti-reflection film causes disturbances in color elements between the solar cells. It is something.
According to this aspect, the machine learning section judges the color balance based on human subjectivity, and the machine learning section determines that the color balance of the solar cell group is good based on slight disturbances in color elements. Then, a predicted placement model is generated. Therefore, the arrangement can be set based on the illusion of the human eye.

According to the solar cell group manufacturing apparatus and the wall surface structure manufacturing apparatus of the present invention, even when a person visually recognizes the solar cell group under sunlight, it is easy to feel that the color balance is good.
According to the machine learning program of the present invention, it is possible to generate a solar cell arrangement model that gives a good impression of good color balance even when viewed under sunlight.

FIG. 1 is a block diagram of a module manufacturing apparatus according to a first embodiment of the present invention. 2 is an explanatory diagram schematically showing a solar cell module that can be manufactured by the module manufacturing apparatus of FIG. 1, in which (a) is a perspective view seen from the front side, and (b) is a perspective view seen from the back side. be. It is an exploded perspective view of the solar cell module of FIG.2(b). FIG. 4 is a perspective view of a main part of the solar cell module of FIG. 3; 4 is a cross-sectional view of the solar cell module of FIG. 3. FIG. FIG. 2 is an explanatory diagram of the deep learning unit in FIG. 1, in which (a) is a schematic diagram showing a neuron model, and (b) is a schematic diagram showing a neural network model. FIG. 2 is an explanatory diagram showing the relationship between population solar cells and sample solar cells manufactured in the production department of FIG. 1, in which (a) is a graph of the number of solar cells in the population versus brightness; It is a graph of the number of sample solar cells versus brightness. These are images simulating a solar cell of a solar cell module manufactured in the manufacturing department in Figure 1, where (a) shows the case when light is irradiated at low illuminance, and (b) shows the case when light is irradiated with simulated sunlight. represent. FIG. 2 is a block diagram of a wall manufacturing apparatus according to a second embodiment of the present invention. 10 is a perspective view schematically showing a wall structure that can be manufactured by the wall manufacturing apparatus shown in FIG. 9. FIG. FIG. 10 is a perspective view of the wall structure of FIG. 10 viewed from another direction. 12 is a cross-sectional view of the solar cell module of FIG. 11. FIG. It is a block diagram of the module manufacturing apparatus of 3rd Embodiment of this invention. It is a sectional view of the solar cell module manufactured by the manufacturing device of other embodiments of the present invention. These are images simulating a solar cell of a conventional solar cell module, in which (a) shows the case where light is irradiated with low illuminance, and (b) shows the case where light is irradiated with simulated sunlight.

Hereinafter, the module manufacturing apparatus 1 according to the first embodiment of the present invention will be described in detail.

A module manufacturing apparatus 1 according to a first embodiment of the present invention is for manufacturing a solar cell module 200 (solar cell group) shown in FIG. 2 that includes a plurality of solar cells 201 (solar cells).
The module manufacturing apparatus 1 includes a manufacturing section 2, a control section 3, and a measuring section 5, as shown in FIG.
In the module manufacturing apparatus 1, the control unit 3 is provided with a deep learning unit 20 that operates using a machine learning program, and a placement model of the solar cell 201 is generated based on the results of machine learning performed in advance by the deep learning unit 20. It is. One of the characteristics of the module manufacturing apparatus 1 is that the solar cells 201 are arranged and manufactured according to the generated arrangement model.

As shown in FIG. 1, the manufacturing section 2 includes a cell forming section 10, a housing section 11, a placement operation section 12, and a wiring connection section 15 as main components. In addition, various devices such as a sealing part for sealing the solar cell 201 are provided, but since they are the same as conventional ones, their explanation will be omitted.

The cell forming unit 10 is a part where the solar cells 201 are formed, and includes a plurality of film forming apparatuses such as CVD apparatuses.
The housing section 11 is a housing member that temporarily houses the solar cells 201 measured by the measuring section 5 . The storage unit 11 is provided with a plurality of rooms that can accommodate the solar cells 201.
The placement operation unit 12 is a part that takes out and places a predetermined solar cell 201 from the storage unit 11 based on the placement model formed by the deep learning unit 20 of the control unit 3 . Note that the solar cells 201 formed in the cell forming section 10 may be directly arranged.
The wiring connection part 15 is a part that connects the wiring member 202 between the solar cells 201, 201 arranged in a predetermined arrangement by the arrangement operation part 12.

As shown in FIG. 1, the control unit 3 includes a deep learning unit 20 (machine learning unit, placement determining device), a storage unit 21, a measurement result acquisition unit 22, and an input/output unit 23 as main components. There is.
The control section 3 may be provided in a different building from the manufacturing section 2 and the measuring section 5. In this case, it is preferable that the control section 3 is communicably interconnected with the manufacturing section 2 and the measuring section 5 via a network such as an intranet. Further, the control section 3 may be connected to the manufacturing section 2 and the measuring section 5 via the Internet or the like. By doing so, it is also possible to collectively manage the manufacturing department 2 and the measuring department 5 at multiple locations in different buildings.

The deep learning unit 20 is a machine learning unit that can operate based on a machine learning program.
The deep learning unit 20 has a function of performing machine learning by itself using the color elements and arrangement of each solar cell 201 and the result of human judgment of the color balance of the solar cell module 200 as training data, and uses the results of machine learning as training data. Based on the color elements of each solar cell 201 acquired by the measurement result acquisition unit 22, it is possible to create an arrangement model of the solar cells 201 that is expected to be judged by a person as having good color balance.
The deep learning unit 20 is capable of performing supervised learning in accordance with algorithms such as a neural network, which will be described later.
Here, "supervised learning" means that a large amount of training data, that is, a set of input and result data, is given to the deep learning unit 20 to learn the features in those data sets, and to learn the characteristics from the input and result data. This is a model (error model) that estimates the relationship between the input and the result inductively.

The deep learning unit 20 of this embodiment uses the color elements of each solar cell 201, the arrangement of each solar cell 201 in the solar cell module 200, and the ability of a person to determine the color balance of the solar cell module 200 in that arrangement. The determination results are linked and machine learning is performed on the correlation between the color elements of the solar cells 201 and the arrangement of each solar cell 201 and how a person makes a quality determination. Based on the results of machine learning, the deep learning unit 20 can generate an arrangement model that is expected to be judged as good by humans from the information on the color elements of each solar cell 201. In addition, the deep learning unit 20 of the present embodiment can further predict the determination result of the color balance of the solar cell module 200 when visually recognized by a person from the information on the color elements of each solar cell 201. There is.
Details of the deep learning unit 20 of this embodiment will be described later.

The storage unit 21 includes a storage device such as a memory or a hard disk, and stores data used in machine learning in the deep learning unit 20, past and present manufacturing parameters used in manufacturing the solar cell 201 in the manufacturing unit 2, and a measurement unit. This is a part that stores and accumulates various measurement parameters such as the power generation characteristics and color elements of each solar cell 201 measured in step 5, the arrangement model generated by the deep learning unit 20, and the like.

The measurement result acquisition unit 22 is a unit that acquires measurement results such as power generation characteristics and color elements measured by the measurement unit 5 and transmits them to the storage unit 21 and/or the deep learning unit 20.

The input/output unit 23 is a unit that inputs and outputs input to and from the manufacturing unit 2, and is a unit that outputs the placement model generated by the deep learning unit 20 to the placement operation unit 12 of the manufacturing unit 2.

The measuring section 5 is a section that measures the characteristics of the solar cell 201 formed in the cell forming section 10 of the manufacturing section 2, and includes a power generation characteristic measuring section 30 and a color element measuring section 31.
The power generation characteristic measurement section 30 is a part that measures the power generation characteristics of the solar cell 201.
The color element measurement unit 31 is a part that measures color elements of the solar cell 201.

Next, the solar cell module 200 to be manufactured will be explained.

As shown in FIGS. 2 and 3, the solar cell module 200 has a plurality of solar cells 201 electrically connected to each other by a wiring member 202 between two sealing base materials 205 and 206, and a sealing base. The space between the materials 205 and 206 is filled with sealing materials 207 and 208.
The solar cell module 200 is plate-shaped, and has solar cells 201 arranged in a plane based on the above-described arrangement model, and wiring members 202 are provided only on the back side of the solar cells 201. It is something that

As shown in FIG. 2A, the solar cell module 200 of this embodiment includes a total of 20 or more solar cells 201, and the solar cells 201 are arranged in a grid pattern.
The shortest distance L between vertically and horizontally adjacent solar cells 201, 201 of solar cell module 200 shown in FIG. 4 is preferably 5 mm or less, respectively.
Within this range, the solar cells 201 can be densely spread, and the power generation efficiency per installation area can be improved.

The solar cell module 200 of this embodiment includes a module side identification part 223 on the back surface 221 side, as shown in FIG. 2(b).
The module-side identification unit 223 is a part that assigns a unique ID to each solar cell module 200, and is specifically a one-dimensional code or a two-dimensional code. That is, at least the ID assigned to the solar cell module 200 can be detected by identifying the module-side identification section 223 using a dedicated reading device.

The solar cell 201 is a so-called back contact type solar cell, and as shown in FIG. 5, electrode layers 213 and 216 are provided on the back surface 221 side, and electrode layers 213 and 216 are not provided on the light receiving surface 220 side. .
Specifically, the solar cell 201 includes an antireflection film 211 (antireflection material) on the light receiving surface 220 side of the first conductivity type semiconductor substrate 210. On the other hand, the solar cell 201 has a first conductivity type semiconductor layer 212 and a first conductivity type side electrode layer 213 on the back surface 221 (principal surface opposite to the light receiving surface 220) of the first conductivity type semiconductor substrate 210. They are stacked in order. Further, in the solar cell 201, a second conductive type semiconductor layer 215 is provided on the back surface 221 side of the first conductive type semiconductor substrate 210 and is separate from the first conductive type semiconductor layer 212 and the first conductive type side electrode layer 213. and a second conductivity type side electrode layer 216 are laminated. The color elements of the solar cell 201 are substantially determined by the antireflection film 211 provided on the light receiving surface 220 side.
The first conductive type semiconductor layer 212 has the same conductive type as the first conductive type semiconductor substrate 210 and has the opposite conductive type to the second conductive type semiconductor layer 215. That is, in the solar cell 201, when the first conductive type semiconductor layer 212 and the first conductive type semiconductor substrate 210 are n-type, the second conductive type semiconductor layer 215 is p-type, and the first conductive type semiconductor layer 212 and the first conductive type semiconductor substrate 210 are n-type. When the first conductivity type semiconductor substrate 210 is p-type, the second conductivity type semiconductor layer 215 is n-type.

The antireflection film 211 is a reflective sealing material that confines received light within the solar cell 201 . The anti-reflection film 211 can be made of, for example, silicon nitride.

In this embodiment, there is a slight amount of anti-reflection between the solar cells 201 that forms the solar cell module 200 due to the influence of the film forming position and heating temperature of the anti-reflection film 211 during manufacturing. The thickness of the film 211 and the refractive index of the antireflection film 211 are different.

As shown in FIG. 4, the solar cell 201 has a cell-side identification portion 217 on the back surface 221 side.
The cell-side identification unit 217 is a part that assigns a unique ID, specifically a one-dimensional code or a two-dimensional code. That is, by identifying the cell-side identification unit 217 using a dedicated reading device, it is possible to detect at least the ID assigned to the solar battery cell 201.

The wiring member 202 is a so-called interconnector, and as shown in FIG. 3, it physically and electrically connects adjacent solar cells 201, 201.

The first sealing base material 205 is a sealing member that seals the solar cell 201, and is a light-transmitting insulating substrate or a transparent insulating sheet having light-transmitting properties and insulating properties, and is, for example, made of glass or transparent Materials made of resin can be used.
The second sealing base material 206 is a sealing member that seals the solar cell 201, and is an insulating substrate or an insulating sheet having insulation properties, and may be made of glass or resin, for example.

The sealing materials 207 and 208 are light-transmitting adhesives having light-transmitting properties and adhesive properties, and for example, sheets such as EVA can be used.

Next, a method for manufacturing the solar cell module 200 will be explained.

First, the solar cell 201 is formed in the cell forming section 10 of the manufacturing section 2 (solar cell forming step, cell forming step).
Specifically, as can be read from FIG. 5, a first conductive type semiconductor layer 212 and a first conductive type side electrode layer 213 are laminated in this order on a part of one side of the first conductive type semiconductor substrate 210, and A second conductive type semiconductor layer 215 and a second conductive type side electrode layer 216 are laminated in this order on the other part. Then, an antireflection film 211 is formed on the opposite surface of the first conductive type semiconductor substrate 210.

Next, the power generation characteristics of each solar cell 201 formed in the solar cell formation process are measured by the power generation characteristic measurement section 30, and at the same time, the color elements are measured by the color element measurement section 31 (color measurement step). .
In this embodiment, the power generation characteristic measuring section 30 measures the IV characteristics, resistance, etc., and the color element measuring section 31 measures the brightness of each solar cell 201.

The measurement results of each solar battery cell 201 in the color measurement process are transmitted to the measurement result acquisition unit 22 of the control unit 3 (transmission process), and the deep learning unit 20 receives the measurement results of each solar battery cell 201 in the color measurement process. It is linked to the cell-side identification section 217 of each solar cell 201 and stored in the storage section 21 (linking step).

If necessary, the linking step is completed, and the solar cell 201 with the measurement result and the cell-side identification section 217 linked is stored in the storage section 11 (accommodation step).
At this time, the accommodation form in the accommodation section 11 is not particularly limited. They may be placed in the accommodating section 11 in the order of manufacture, or they may be sorted by color element and placed in a different accommodating section 11 for each color element.

Next, the deep learning unit 20 of the control unit 3 determines the arrangement of the solar cells 201 in the solar cell module 200 based on the measurement results received by the measurement result acquisition unit 22 (arrangement determination step).
In the present embodiment, the deep learning unit 20 uses the color elements of each solar cell 201 measured in the past solar cell module 200 and the arrangement of each solar cell 201, and the balance of the color of the past solar cell module 200. Based on the result of determining the quality of each solar battery cell 201, each solar battery cell is adjusted so that a person determines that the color balance of the solar battery module 200 to be formed is good based on the measurement result of the measurement unit 5 of each solar battery cell 201. 201 is generated, and the arrangement is determined based on the arrangement orientation of the solar cells 201.

The placement operation unit 12 of the manufacturing unit 2 takes out the solar cells 201 from the storage unit 11 and places the solar cells 201 based on the placement model generated by the deep learning unit 20 in the placement determination process (placement process).

During or after the placement process, wiring members 202 are connected between adjacent solar cells 201, 201 to electrically connect each solar cell 201 (wiring connection process).
At this time, as can be read from FIG. 3, the wiring member 202 connects the first conductivity type side electrode layer 213 and the second conductivity type side electrode layer 216 on the back surface 221 side of the adjacent solar cell 201. That is, the wiring member 202 is not connected to the light-receiving surface 220 side of the solar cell 201 .

Thereafter, a frame, connector members, etc. are attached as appropriate using a known method to complete the solar cell module 200.

Next, the deep learning unit 20 of this embodiment will be explained.

The deep learning unit 20 performs learning in accordance with a neural network of four or more layers, and includes an arithmetic unit and an arithmetic unit that implement a neural network incorporating a neuron model as shown in FIG. 6(a) as an approximation algorithm for a value function. It consists of memory, etc.
That is, as shown in FIG. 6(a), the neuron outputs an output y for m inputs x _i (i is a positive integer), and each x _i has an output y corresponding to this input x _i It is multiplied by a weight w _i and outputs an output y expressed by the following equation (1). Note that the input x _i , output y, and weight w _i are all vectors.

Here, b is the bias and f is the activation function.

The neural network of the deep learning unit 20 of this embodiment includes an input layer 300, a middle layer 301, and an output layer 302, as shown in FIG. 6(b). ) are combined to form a deep neural network having a thickness of p layers (p is a positive integer of 4 or more). That is, the intermediate layer 301 includes p-layer intermediate layers D1 to Dp.
The neural network of this embodiment receives S inputs X (X ₁ to X _S : S is a positive integer) from an input layer 300, passes through an intermediate layer 301, and outputs T results Y from an output layer 302. (Y ₁ to Y _T : T is a positive integer) are output.

Specifically, the input X (X ₁ to X _S ) of the input layer 300 is multiplied by the corresponding weight W1 and input to each neuron N1 of the first intermediate layer D1 of the intermediate layer 301. Each neuron N1 of the first hidden layer D1 outputs a feature vector Z1, and the feature vector Z1 is multiplied by a corresponding weight W2 and input to each neuron N2 of the second hidden layer D2 of the hidden layer 301. Ru.

The feature vector Z1 is a feature vector between the weights W1 and W2, and can be regarded as a vector that extracts the feature amount of the input vector.
The feature vector Z1 is applied to each neuron N2 of the second intermediate layer D2 of the intermediate layer 301 by a corresponding weight W2 and input.
Each neuron N2 of the second hidden layer D2 outputs a feature vector Z2, and the feature vector Z2 is multiplied by a corresponding weight W3 and input to each neuron N3 of the third hidden layer D3 of the hidden layer 301. Ru.
The above process is repeated in each intermediate layer of the intermediate layer 301, and each neuron Np of the terminal P-th intermediate layer Dp outputs a feature vector Zp, and the feature vector Zp is output to the output layer 302. As a result, the neural network outputs the result Y (Y ₁ -Y _T ).
The weights W1 to Wp can be learned by the error backpropagation method. The error backpropagation method is a method of adjusting (learning) the weight W of each neuron so as to reduce the difference between the output y when the input x is input and the true output y (teacher). .

Next, the procedure of machine learning in the deep learning section 20 will be explained.

First, 500 or more solar cells 201 (hereinafter also referred to as population solar cells 201a) are manufactured, and the distribution of color elements of each solar cell 201 of population solar cells 201a is calculated.
In this embodiment, as shown in FIG. 7, the brightness distribution of each solar cell 201 of 1000 population solar cells 201a is calculated.

Subsequently, as shown in FIG. 7, a predetermined number (for example, 30) of solar cells 201 are selected from among the population solar cells 201a so that the brightness distribution of the population solar cells 201a is substantially maintained. (hereinafter also referred to as sample solar cells 201b) are extracted as samples, the sample solar cells 201b are arranged randomly, and the solar cell module 200 is assembled.

The arrangement of each photovoltaic cell 201 and the color element of each photovoltaic cell 201 are input to the input layer 300 of the deep learning unit 20, and the determination result of whether the color balance is good or bad is obtained from the output layer 302. In addition, in a separate process, a person is asked to visually check the solar cell module 200 and judge whether it is good or bad.
At this time, in this embodiment, a person uses brightness as a criterion and determines whether the color balance is good or bad based on the brightness balance.

Then, the pass/fail judgment result obtained from the output layer 302 of the deep learning unit 20 is compared with the pass/fail judgment result judged by a person, the weights are adjusted so that the difference in the pass/fail judgment results match, and the arrangement of the solar cell 201 is and machine learning.

Finally, typical physical properties of the solar cell module 200 manufactured by the module manufacturing apparatus 1 of this embodiment will be explained.

The solar cell module 200 has a low overall brightness and appears uniform as shown in FIG. 8(a) in a place with low illuminance such as indoors, and even in a place with high illuminance such as outdoors. 8(b), there is no sense of color unevenness as a whole.

Here, the CIE1976 (L*, a*, b*) color system is generally used as a method for displaying object colors. However, the CIE1976 (L*, a*, b*) color system is not suitable for quantifying differences in appearance due to differences in the illumination environment of an object.
Therefore, as mentioned above, it cannot be used for the purpose of quantifying the appearance of objects, which is particularly noticeable under conditions of high illuminance such as outdoor direct sunlight.

Therefore, the inventor investigated a method for numerically expressing the color of an object under direct sunlight in an outdoor environment. The display of object color in this specification reflects the results of the above study, and specifically uses numerical values measured under the following conditions.
A solar simulator is prepared as a light source, and AM1.5 light with a radiation intensity of 1000 W/m ² is vertically irradiated onto the object. A camera is prepared as a measuring device and placed in a position facing the object as directly as possible without specularly reflected light entering.

The camera settings are as follows, and a JPEG image of the object is taken with the camera. The RGB values of each pixel of the object are read out from the photographed JPEG image. The read RGB values are converted into the CIE1976 (L*, a*, b*) color system using the white point of the 10° visual field of the light source D65.

[Camera settings]
Nikon digital camera D5500 with standard lens NIKKOR 18-55mm 1:3.5-5.6 GI lens attached, aperture 8, film sensitivity ISO 400, shutter speed 1/100 seconds, white balance "Sunny", picture control "Standard", color space sRGB, active D-lighting OFF, high dynamic range OFF.

The chromaticity coordinates in the CIE1976 (L*, a*, b*) color system described below are the coordinates obtained by quantifying the object color using the method described above.

The solar cell module 200 is one of the following (1) to (3) in the CIE1976 (L*, a*, b*) color system calculated from a photographed image of a solar cell under direct sunlight. It is preferable that the following conditions are satisfied.
(1) The difference between the maximum and minimum brightness L* of each solar cell 201 is 3.0 or more, and the difference between the brightness L* of adjacent solar cells 201, 201 is 1.5 or less. be.
(2) The difference between the maximum and minimum values of chromaticity a* of each solar cell 201 is 2.0 or more, and the difference between chromaticity a* of adjacent solar cells 201, 201 is 1.0 or less It is.
(3) The difference between the maximum and minimum values of chromaticity b* of each solar cell 201 is 5.0 or more, and the difference between chromaticity b* of adjacent solar cells 201, 201 is 2.5 or less It is.
The solar cell module 200 is arranged adjacent to at least three solar cells 201 so as to surround one solar cell 201, and the relationship between the one solar cell 201 and the three solar cells 201 is based on CIE1976. In the (L*, a*, b*) color system, it is preferable that any of the following conditions (4) to (6) be satisfied.
(4) The condition of (1) is satisfied, and the difference in brightness L* between one solar cell 201 and three solar cells 201 is 1.8 or less.
(5) The condition of (2) is satisfied, and the difference in chromaticity a* between one solar cell 201 and three solar cells 201 is 1.5 or less.
(6) The condition of (3) is satisfied, and the difference in chromaticity b* between one solar cell 201 and three solar cells 201 is 4.0 or less.
Note that the values of lightness L*, chromaticity a*, and chromaticity b* may be values at one measurement point, or may be average values measured at a plurality of measurement points.

As described above, although the solar cells 201 manufactured in the manufacturing section 2 of the module manufacturing apparatus 1 of this embodiment are formed through the same manufacturing process, the thickness of the antireflection film 211 is Alternatively, the difference in the refractive index of the anti-reflection film 211 may cause disturbance of color elements between the solar cells 201, 201.
According to the module manufacturing apparatus 1 of the present embodiment, the deep learning section 20 makes a judgment of the color balance based on human subjectivity, and the deep learning section 20 detects the slight disturbance of the color elements between the solar cells 201 described above. Based on this, an arrangement model is generated in which it is predicted that a person will judge that the color balance of the solar cell module 200 is good. Therefore, the arrangement of the solar cells 201 can be set to have a uniform appearance based on the illusion of the human eye, and more solar cells 201 can be used for manufacturing the solar cell module 200. Therefore, the yield can be improved compared to simply arranging items with similar color distributions.
Furthermore, according to the module manufacturing apparatus 1 of the present embodiment, since the color balance can be determined within the module manufacturing apparatus 1 in a way that is close to human sensitivity, it is also possible to automate the arrangement of the solar cells 201.

According to the solar cell 201 of this embodiment, the wiring member 202 is provided on the side opposite to the light-receiving surface 220, that is, on the back surface 221 side, so that the wiring member 202 does not get in the way of receiving light such as sunlight. , the power generation efficiency can be improved compared to the case where the wiring member 202 is provided on the light receiving surface 220 side.

According to the module manufacturing apparatus 1 of this embodiment, since the variation in brightness is used as a criterion for color balance, it is possible to achieve better color balance under sunlight such as outdoors, and to further improve the design.

According to the module manufacturing apparatus 1 of the present embodiment, the sample solar cells 201b are obtained from the color distribution of the population solar cells 201a whose tendency has been smoothed so that the distribution is substantially maintained. Can do machine learning well.

In the solar cell module 200 of this embodiment, the absolute value of at least one of the color elements of lightness L*, chromaticity a*, and chromaticity b* of each solar cell 201 is large, and the color element of the entire solar cell 201 is There are variations. That is, if they are simply arranged, the color elements will not be uniform and color unevenness will occur.
According to the solar cell module 200 of the present embodiment, even if there are variations in the color elements of the entire solar cell 201, the brightness L*, chromaticity a*, and chromaticity b* of the adjacent solar cells 201, 201 Since the difference in at least one of the color elements is small, the solar cell module is uniform and homogeneous as a whole, and the color elements are uniform.

The solar cell module 200 of the present embodiment has at least the brightness L*, the chromaticity a*, and the chromaticity b* of one solar cell 201 and the solar cells 201 adjacent to each side of the one solar cell 201. Since the difference in any of the color elements is small, the solar cell module has more uniform color elements.

In the solar cell module 200 of this embodiment, the solar cells 201 are arranged in a grid pattern, and the intervals between adjacent solar cells 201, 201 are narrow, so that the solar cell module 200 has more uniform color elements. becomes.

According to the manufacturing method of the solar cell module 200 of the present embodiment, the arrangement of the solar cell 201 is determined using the measurement results in the color measurement step of measuring the color elements. A solar cell module 200 with good color balance can be manufactured.

According to the manufacturing method of the solar cell module 200 of this embodiment, in the placement determination step, the color elements of each solar cell 201 measured in the past solar cell module 200 and each solar cell of the past solar cell module 200 are used. Based on the quality of the color balance of each solar cell 201, the arrangement of the solar cells 201 is determined based on the measurement results in the color measurement step so that the color balance of each solar cell 201 is good. Therefore, the color balance of each solar cell 201 is better than when the solar cells 201 are arranged randomly.

According to the manufacturing method of the solar cell module 200 of the present embodiment, in the placement step, the solar cells are linked to the measurement results in the linking step based on the placement model determined by the deep learning unit 20 in the placement determination step. The cell 201 is taken out from the storage section 11 and placed. Therefore, the solar cells 201 accommodated in the accommodation section 11 and temporarily stocked can be taken out at a necessary timing.

According to the method for manufacturing the solar cell module 200 of the present embodiment, the color elements and power generation characteristics of the solar cell 201 are simultaneously measured in the color measurement process, so that defective products can be excluded before being placed in the storage section 11. This can shorten manufacturing time.

According to the method for manufacturing the solar cell module 200 of this embodiment, the orientation of the solar cells 201 (which side should face which position, etc.) is determined based on the measurement results. Therefore, more types of solar cells 201 can be used in the solar cell module 200.

According to the manufacturing method of the solar cell module 200 of this embodiment, the color elements of the solar cell 201 are measured at a plurality of measurement points in the color measurement step, and the average value of the color elements is used in the placement determination step. Since the arrangement of the solar cells 201 is determined, the arrangement of the solar cells 201 can be determined more accurately.

Next, a wall manufacturing apparatus 400 according to a second embodiment of the present invention will be described. Note that the same configurations as the module manufacturing apparatus 1 of the first embodiment are given the same reference numerals and the description thereof will be omitted.

The wall manufacturing apparatus 400 of the second embodiment forms a wall structure 500 (solar cell group) in which a plurality of solar cell modules 200 (solar cells) are arranged in a plane, as shown in FIG.
Like the module manufacturing apparatus 1 in the first embodiment, the wall manufacturing apparatus 400 also includes a deep learning section 20, and a placement model of the solar cell module 200 is generated based on the results of machine learning performed in advance by the deep learning section 20. It is something that will be done. One of the characteristics of the module manufacturing apparatus 1 is that the solar cell modules 200 are arranged and manufactured according to the generated arrangement model.
That is, in the module manufacturing apparatus 1 according to the first embodiment, the deep learning unit 20 forms an arrangement model of each solar cell 201 and adjusts the color balance of the solar cell module 200, whereas in the second embodiment In the wall manufacturing apparatus 400 according to the embodiment, the deep learning unit 20 adjusts the arrangement of each solar cell module 200 to adjust the color balance of the wall structure 500.

The wall manufacturing apparatus 400 includes a manufacturing section 402, a control section 403, and a measuring section 5, as shown in FIG.
The manufacturing department 402 includes a module manufacturing device 1 and a placement operation section 412 as main components.

The placement operation unit 412 is a part that arranges the solar cell modules 200 based on the placement model formed by the deep learning unit 20 of the control unit 403.

The control unit 403 includes a deep learning unit 20 , a storage unit 21 , a measurement result acquisition unit 422 , and an input/output unit 23 .
The deep learning unit 20 of this embodiment has a function of performing machine learning by itself using the color elements and arrangement of each solar cell module 200 and the result of human judgment of the color balance of the wall structure 500 as training data. Based on the results, it is possible to create an arrangement model of solar cell modules 200 that is expected to be judged by a person to have good color balance from the color elements of each solar cell module 200 acquired by the measurement result acquisition unit 422. .
Regarding the machine learning procedure in the deep learning unit 20, the solar cell 201 of the first embodiment becomes the solar cell module 200 of the second embodiment, and the solar cell module 200 of the first embodiment becomes the wall surface of the second embodiment. The structure is the same except for the structure 500, so the explanation will be omitted.

The measurement result acquisition unit 422 is a unit that acquires measurement results such as power generation characteristics and color elements measured by the measurement unit 5 and transmits them to the storage unit 21 and/or the deep learning unit 20.

The measuring section 5 of this embodiment is a section that measures the characteristics of the solar cell module 200 formed by the module manufacturing apparatus 1 of the manufacturing section 402, and includes a power generation characteristic measuring section 30 and a color element measuring section 31. The measurement target of the sections 30 and 31 is the solar cell module 200.

Next, the wall structure 500 to be manufactured will be described.

In the wall structure 500, a plurality of solar cell modules 200 are arranged in a plane as shown in FIG. 11, and each solar cell module 200 is electrically connected by a connector member 502 provided on the back surface 221. In the wall structure 500 of this embodiment, a total of 20 or more solar cell modules 200 are arranged in a checkerboard shape.

The shortest distance between adjacent solar cell modules 200, 200 is preferably 5 cm or less, more preferably 2 cm or less, and particularly preferably 5 mm or less.
Within this range, the solar cell modules 200 can be densely laid out and the power generation efficiency per installation area can be improved.

In the solar cell module 200 of this embodiment, as shown in FIG. 12, a second antireflection film 501 is formed on the sealing base material 205 on the light receiving surface 220 side. That is, in the solar cell module 200 of this embodiment, an antireflection film 211 is interposed between the sealing base material 205 on the light receiving surface 220 side and the solar cell 201, and the light receiving surface is further interposed with respect to the solar cell 201. A second antireflection film 501 is formed on the outer surface of the sealing base material 205 on the 220 side.

Next, a method for manufacturing the wall structure 500 will be described.

First, the solar cell module 200 is formed using the module manufacturing apparatus 1 of the manufacturing department 402 (solar cell module forming step).
At this time, a second antireflection film 501 is formed on the outer surface of the sealing base material 205 on the light receiving surface 220 side.

Next, for each solar cell module 200 formed in the solar cell module forming step, the power generation characteristics are measured by the power generation characteristic measurement section 30, and at the same time, the color elements are measured by the color element measurement section 31 (color measurement step). ).

The measurement results of the solar cell module 200 in the color measurement process are transmitted to the measurement result acquisition unit 422 of the control unit 403 (transmission process), and the deep learning unit 20 transmits the measurement results of the solar cell module 200 in the color measurement process to each solar cell. It is linked to the module-side identification unit 223 of the battery module 200 and stored in the storage unit 21 (linking step).

Next, the deep learning unit 20 of the control unit 403 determines the placement of the solar cell modules 200 within the wall structure 500 based on the measurement results received by the measurement result acquisition unit 422 (placement determination step).
Specifically, similar to the first embodiment, the color elements of each solar cell module 200 and the arrangement of each solar cell module 200 measured in the past wall structure 500 by the deep learning unit 20, and the Based on the results of determining the quality of the color balance of the 500 colors, a person can determine that the color balance of the wall structure 500 to be formed is good based on the measurement results in the measurement unit 5 of each solar cell module 200. A placement model of each solar cell module 200 is generated and the placement is determined.

The placement operation unit 412 of the manufacturing department 402 places the solar cell modules 200 based on the placement model generated by the deep learning unit 20 in the placement determination process (placement process).

During or after the placement process, the connector members 502 provided on the back surface 221 are connected between adjacent solar cell modules 200, 200 to electrically connect each solar cell module 200 (connector connection process).

Thereafter, the wall structure 500 is completed by appropriately attaching crosspieces and the like using a known method.

According to the wall manufacturing apparatus 400 of the present embodiment, although the solar cell modules 200 manufactured by the module manufacturing apparatus 1 are formed through the same manufacturing process, the antireflection coatings 211 and 501 are The difference in thickness or refractive index of the antireflection films 211 and 501 causes color element disturbance between the solar cell modules 200 and 200.
According to the wall surface manufacturing apparatus 400 of this embodiment, the deep learning section 20 judges the color balance based on human subjectivity, and the deep learning section 20 detects the slight disturbance of color elements between the solar cell modules 200 described above. Based on this, an arrangement model is generated in which it is predicted that a person will judge that the color balance of the wall structure 500 is good. Therefore, the arrangement of the solar cell modules 200 can be set to have a uniform appearance based on the illusion of the human eye, and more solar cell modules 200 can be used to manufacture the wall structure 500. Therefore, the yield can be improved compared to simply arranging items with similar color distributions.
Further, according to the wall surface manufacturing apparatus 400 of this embodiment, since the color balance can be determined within the wall surface manufacturing apparatus 400 in a way that is close to human sensibilities, it is also possible to automate the arrangement of the solar cell modules 200.

According to the wall manufacturing apparatus 400 of this embodiment, the wall structure 500 is formed by arranging the solar cell modules 200, so that the color balance, particularly the uniformity of color elements, is good over a wide range.

In the wall structure 500 of the present embodiment, the absolute value of at least one of the color elements of lightness L*, chromaticity a*, and chromaticity b* of each solar cell module 200 is large, and the color element of the entire solar cell module 200 has a large absolute value. There is variation. That is, if they are simply arranged, the color elements will not be uniform and color unevenness will occur.
According to the wall structure 500 of the present embodiment, even if there are variations in the color elements of the entire solar cell module 200, at least the lightness L*, chromaticity a*, and chromaticity b* of the adjacent solar cell modules 200, 200 Since the difference in any one of the color elements is small, the wall surface structure has a cohesive and homogeneous overall structure with uniform color elements.

According to the manufacturing method of the wall structure 500 of the present embodiment, the placement of the solar cell module 200 is determined using the measurement results in the color measurement step of measuring color elements, so the colors based on the measurement results of the color elements are determined. A wall structure 500 with good balance can be manufactured.

Next, a manufacturing apparatus 600 according to a third embodiment of the present invention will be described. Note that the same components as those of the manufacturing apparatus 1,400 of the first and second embodiments are given the same reference numerals, and the description thereof will be omitted.

In the manufacturing apparatus 600 of the third embodiment, the structure of the control section 603 is different from the control section 3 of the first embodiment.
That is, as shown in FIG. 13, the control unit 603 of the manufacturing apparatus 600 includes a deep learning unit 20 (machine learning unit), a storage unit 21, a measurement result acquisition unit 22, and an input/output unit 23 as main components. In addition, a second deep learning unit 605 (second machine learning unit) is provided, and the second deep learning unit 605 generates a placement model of the solar battery cells 201.
The second deep learning unit 605 replaces the arrangement of each solar cell 201 in the solar cell module 200 and provides the deep learning unit 20 with information on the color element of each solar cell 201 when the arrangement is replaced. The color balance is determined, and the correlation between the arrangement of the plurality of solar cells 201 and the determination result determined by the deep learning unit 20 is used as training data for machine learning.
The second deep learning unit 605 is capable of generating an arrangement model of the solar cell 201 that is predicted to be determined by the deep learning unit 20 to have better color balance of the solar cell module 200.

The second deep learning unit 605, like the deep learning unit 20, performs learning based on a neural network of four or more layers. The neural network of the second deep learning unit 605, like the deep learning unit 20, includes an input layer 300, a middle layer 301, and an output layer 302, and the above neurons (neurons N1 to Np) are combined as the middle layer 301. , a deep neural network having a thickness of p layers (p is a positive integer of 4 or more).

Next, the procedure of machine learning in the second deep learning unit 605 will be explained.

First, the population solar cell 201a is manufactured, and the distribution of color elements of each solar cell 201 of the population solar cell 201a is calculated.

Next, sample solar cells 201b are extracted from the population solar cells 201a as samples, and the sample solar cells 201b are randomly arranged so that the brightness distribution of the population solar cells 201a is substantially maintained. They are lined up and the solar cell module 200 is assembled.
The arrangement of each photovoltaic cell 201 and the color element of each photovoltaic cell 201 are input to the input layer 300 of the second deep learning unit 605, and the determination result of whether the color balance is good or bad is obtained from the output layer 302. Further, the deep learning section 20 determines whether the solar cell module 200 is good or bad.
Then, the pass/fail judgment result obtained from the output layer 302 of the second deep learning unit 605 is compared with the pass/fail judgment result judged by the deep learning unit 20, and the weights are adjusted so that the difference in the pass/fail judgment results match. Machine learning is performed by changing the arrangement of battery cells 201.

The deep learning unit 20 of this embodiment performs machine learning in advance on the correlation between the color elements and arrangement of the solar cell 201 and the judgment result of the human judgment of the color balance of the solar cell module 200 as training data. It has judgment criteria that are close to the subjective nature of
According to the manufacturing apparatus 600 of the present embodiment, the second deep learning unit 605 performs machine learning based on the color balance judgment by the deep learning unit 20, which is close to human subjectivity, and the person determines the color balance of the solar cell module 200. The second deep learning unit 605 generates a placement model predicted to be determined to be good. Therefore, the arrangement can be set based on the illusion of the human eye, and more solar cells 201 can be used to manufacture the solar cell module 200. Therefore, the yield can be improved compared to simply arranging items with similar color distributions.
According to the manufacturing apparatus 600 of this embodiment, the deep learning unit 20 provides the second deep learning unit 605 with the teaching data, so it is possible to provide the second deep learning unit 605 with the teaching data without human judgment. can.

In the above-described embodiment, a back contact type solar cell 201 in which the electrode layers 213, 216 and the wiring member 202 are provided on the back surface 221 side is used as the solar cell 201, but the present invention is applicable to this. It is not limited. Other types of solar cells in which electrode layers and wiring members are provided on the light-receiving surface 220 side can also be used as the solar cell 201.

In the embodiment described above, each solar cell 201 was electrically connected by the wiring member 202, but the present invention is not limited to this. As shown in FIG. 14, the electrode layers 213 and 216 of the solar cell 201 may directly contact each other and be electrically connected.

In the first and third embodiments described above, the solar battery cell 201 was provided with the cell-side identification portion 217 on the back surface 221 side, but the present invention is not limited to this. A cell-side identification section 217 may be provided on the light-receiving surface 220 side.

In the first embodiment described above, each solar cell 201 is identified by providing the cell-side identification portion 217 on the back surface 221 side, but the present invention is not limited to this. do not have. The solar cell 201 may be identified using information such as the storage position of the solar cell 201 on the production line. In this case, the cell-side identification section 217 does not need to be provided on the solar cell 201.

In the second embodiment described above, the solar cell module 200 was provided with the module side identification part 223 on the back surface 221 side, but the present invention is not limited to this. A module-side identification section 223 may be provided on the light-receiving surface 220 side.

In the second embodiment described above, each solar cell module 200 is identified by providing the module side identification part 223 on the back surface 221 side, but the present invention is not limited to this. do not have. The solar cell module 200 may be identified using information such as the storage position of the solar cell module 200 on the production line. In this case, the module-side identification section 223 does not need to be provided on the solar cell module 200.

In the embodiment described above, the first conductivity type semiconductor substrate 210 of one conductivity type and the second conductivity type semiconductor layer 215 of the opposite conductivity type are directly bonded to form a pn junction as the solar cell 201. The present invention is not limited to this. The solar cell 201 may be a heterojunction solar cell in which an intrinsic semiconductor layer is interposed between the first conductive type semiconductor substrate 210 and the second conductive type semiconductor layer 215. In this case, the apparent color elements between the solar cells 201 may be affected to some extent depending on the thickness of the intrinsic semiconductor layer and the like. In this case, the deep learning unit 20 will generate an arrangement model that is predicted to determine that the color balance is good, also taking into account the influence of the intrinsic semiconductor layer.

In the first embodiment described above, the entire light-receiving surface 220 of the solar cell 201 is coated with the antireflection film 211, but the present invention is not limited thereto. A portion of the light receiving surface 220 of the solar cell 201 may be coated with an antireflection film 211.

In the first embodiment described above, the human judgment of the color balance of the solar cell module 200 in multiple stages is used as the teacher data, but the present invention is not limited to this. It is also possible to use, as the teacher data, only the quality of the color balance of the solar cell module 200 determined by a person.

In the second embodiment described above, the second antireflection film 501 was formed on the surface of the sealing base material 205 on the light receiving surface 220 side, but the present invention is not limited to this. It is not necessary to form the second antireflection film 501 on the surface of the sealing base material 205 on the light-receiving surface side.

In the second embodiment described above, the solar cell module 200 includes solar cells 201 that are provided separately and independently, electrically connected by wiring members 202, and sealed with sealing base materials 205 and 206. However, the present invention is not limited thereto. As the solar cell module 200, a module in which layered solar cells are formed on a sealing support substrate, such as a thin film solar cell, may be used.

In the first and third embodiments described above, when a person or the deep learning unit 20 determines the color balance of the solar cell module 200, the color balance is determined based on brightness, but the present invention It is not limited. When a person or the deep learning unit 20 determines the color balance of the solar cell module 200, the color balance may be determined based on chromaticity. Further, the color balance may be determined based on the designs, patterns, etc. formed between the solar cells 201.
Similarly, in the second embodiment, when a person judges the color balance of the wall structure 500, the color balance is judged based on brightness, but the present invention is not limited to this. When a person determines the color balance of the wall structure 500, the color balance may be determined based on chromaticity. Alternatively, the color balance may be determined based on the designs, patterns, etc. formed between the solar cell modules 200.

In the embodiments described above, the deep learning unit 20, 605 that learns according to the algorithm of a deep neural network with four or more layers as the machine learning unit or second machine learning unit of the present invention has been described. , but is not limited to this. Learning may be performed in accordance with an algorithm of a neural network having three or fewer layers.

In the embodiments described above, each component can be freely replaced or added between the embodiments as long as it is within the technical scope of the present invention.

1,500,600 Module manufacturing equipment 12,412 Placement operation unit 20 Deep learning department (machine learning department)
31 Color element measurement unit 200 Solar cell module 201 Solar cell 201a Population solar cell 201b Sample solar cell 202 Wiring member 205 First sealing base material 206 Second sealing base material 207, 208 Sealing material 211 Anti-reflection Film 220 Light-receiving surface 221 Back surface 400 Wall manufacturing device 500 Wall structure (solar cell group)
501 Second anti-reflection film 502 Connector member 605 Second deep learning section (second machine learning section)

Claims (9)

It has a placement operation section for arranging a plurality of solar cells constituting a solar cell group, and a machine learning section,
The solar cell group is a plurality of solar cells arranged in a plane,
The plurality of solar cells have a light-receiving surface, and an anti-reflection film is formed on part or all of the light-receiving surface,
Among the plurality of solar cells, there are some that cause variations in color elements due to differences in the thickness of the antireflection film or the refractive index of the antireflection film,
The machine learning unit performs machine learning using teacher data as a correlation between the arrangement of the plurality of solar cells and a determination result obtained by a human being determining the color balance of the group of solar cells in the arrangement of the plurality of solar cells. and
When manufacturing the solar cell group, the machine learning unit is predicted to determine that the color balance of the solar cell group is good when visually recognized by a person, based on information on color elements of each solar cell. generating a placement model of the solar cells;
The arrangement operation unit is a solar cell group manufacturing device that arranges each solar cell based on the arrangement model. The solar cell group is a solar cell module in which the plurality of solar cells are electrically connected by wiring members,
The solar cell group manufacturing apparatus according to claim 1, wherein each solar cell is connected by the wiring member on a side opposite to the light-receiving surface. 3. The solar cell group manufacturing apparatus according to claim 1, wherein the solar cell is a solar cell module including a plurality of solar cells sandwiched between two sealing members. Some of the plurality of solar cells have variations in brightness due to differences in the thickness of the anti-reflection film or the refractive index of the anti-reflection film,
The machine learning unit performs machine learning using teacher data as a correlation between the arrangement of the plurality of solar cells and a judgment result obtained by a human judge of the variation in brightness of the group of solar cells in the arrangement of the plurality of solar cells. and
When manufacturing the solar cell group, the machine learning unit is configured to determine, based on information on the brightness of each solar cell, that the variation in brightness of the plurality of solar cells in the solar cell group is small when visually recognized by a person. The solar cell group manufacturing apparatus according to any one of claims 1 to 3, which generates a placement model that is predicted to be visible. Obtaining the distribution of color elements of 500 or more solar cells, extracting a predetermined number of solar cells from the 500 or more solar cells so that the distribution of the color elements is substantially maintained,
The machine learning unit uses as training data the correlation between the arrangement of the predetermined number of solar cells and the determination result of determining the color balance of a solar cell group in which a predetermined number of solar cells are arranged, extracted by a human. The solar cell group manufacturing apparatus according to any one of claims 1 to 4, wherein the solar cell group manufacturing apparatus performs learning. 6. The solar cell group manufacturing apparatus according to claim 1, wherein the machine learning section performs learning based on a neural network having four or more layers. comprising a placement operation unit for arranging a plurality of solar cells constituting a solar cell group, a machine learning unit, and a second machine learning unit,
The solar cell group is a plurality of solar cells arranged in a plane,
The plurality of solar cells have a light-receiving surface, and an anti-reflection film is formed on part or all of the light-receiving surface,
Among the plurality of solar cells, there are some that cause variations in color elements due to differences in the thickness of the antireflection film or the refractive index of the antireflection film,
The machine learning unit performs machine learning in advance on the correlation between the color elements and arrangement of the plurality of solar cells and the determination result of a human determination of the color balance of the solar cell group as training data, It is possible to predict the determination result of determining the color balance of the solar cell group when visually recognized by a person based on the color element and arrangement information of each solar cell,
The second machine learning unit replaces the arrangement of the plurality of solar cells in the solar cell group, and provides information on the color element and arrangement of each solar cell at the time of replacement to the machine learning unit to learn the solar cell group. The color balance of the solar cells is determined, and the correlation between the arrangement of the plurality of solar cells and the determination result determined by the machine learning unit is machine learned as training data,
The second machine learning unit generates a placement model of the solar cells that is predicted to be determined by the machine learning unit to have better color balance in the solar cell group when manufacturing the solar cell group,
The arrangement operation unit is a solar cell group manufacturing device that arranges each solar cell based on the arrangement model. It has a placement operation unit that arranges a plurality of solar cell modules that make up a wall structure, and a machine learning unit,
The wall structure is one in which the plurality of solar cell modules are arranged in a plane,
The plurality of solar cell modules have a light-receiving surface, and a solar cell is sandwiched between two sealing members,
In the plurality of solar cell modules, an antireflection material is interposed between the sealing member on the light-receiving surface side and the solar cell,
The sealing member on the light-receiving surface side has translucency,
Some of the plurality of solar cell modules have variations in color elements due to differences in the thickness of the antireflection material or the refractive index of the antireflection material,
The machine learning unit performs machine learning using teacher data as a correlation between the arrangement of the plurality of solar cell modules and a determination result obtained by a person determining the color balance of the wall structure in the arrangement of the plurality of solar cell modules. It is a thing,
When manufacturing the wall structure, the machine learning unit is configured to determine, based on information on the color elements of each solar cell module, the wall structure that is predicted to be judged to have good color balance when visually recognized by a person. Generate a solar module placement model,
The placement operation unit is a wall structure manufacturing device that places each solar cell module based on the placement model. A machine learning program that provides the placement model to a placement operation unit that arranges a plurality of solar cells constituting a solar cell group based on the placement model,
The solar cell group is a plurality of solar cells arranged in a plane,
The plurality of solar cells have a light-receiving surface, and an anti-reflection film is formed on part or all of the light-receiving surface,
Among the plurality of solar cells, there are some that cause variations in color elements due to differences in the thickness of the antireflection film or the refractive index of the antireflection film,
Machine learning is performed using the correlation between the arrangement of the plurality of solar cells and the judgment result of a human judgment of the color balance of the group of solar cells in the arrangement of the plurality of solar cells as training data, and A machine learning program that generates a placement model of the solar cells that is predicted to determine that the color balance of the solar cell group is good when visually recognized by a person, based on information on color elements of the solar cells.

Images