Design method of organic photovoltaic cell guided by physical fusion and deep learning

Abstract

The invention discloses a design method of an organic photovoltaic cell guided by physical fusion deep learning, and the method comprises the steps: building an ST-OPV performance prediction model of the physical fusion deep learning, screening out an optimal optical regulation and control structure through the ST-OPV performance prediction model, preparing an ST-OPV device according to the optimal optical regulation and control structure, and testing the PCE and AVT of the ST-OPV device. The invention provides a design method of an organic photovoltaic cell guided by physical fusion and deep learning, which is a design method of a high-performance transparent organic photovoltaic cell guided by physical fusion and deep learning, and comprises the following steps of: fusing prior physical information of a device optical analysis model into a construction and prediction process of a deep learning model; the prediction capability and the physical accuracy of the model in an ST-OPV sparse data scene are significantly improved, and efficient optical design is guided so as to realize collaborative optimization of device performance (PCE and AVT).

Description

Technical Field

[0001] The present invention relates to the field of solar cells, and relates to a design method of an organic photovoltaic cell guided by physical fusion deep learning, and particularly relates to a design method of a high-performance transparent organic photovoltaic cell guided by physical fusion deep learning. Background Art

[0002] Under the background of global energy transformation and the "dual carbon" goal, the innovation of new energy technologies has become the core driving force for reshaping the energy pattern, and solar photovoltaic technology is one of the most promising research directions. Compared with traditional silicon-based solar cells, organic photovoltaic devices (OPVs) have advantages such as solution processability and flexibility, and are regarded as a new generation of green energy technologies with great potential. In recent years, with the continuous emergence of new organic photovoltaic materials, the power conversion efficiency (PCE) of single-junction OPVs has exceeded 20%, showing great potential for industrial application. More importantly, organic photovoltaic materials can precisely regulate their absorption spectra through chemical tailoring, achieving the unique property of semi-transparency in the visible light region, thereby expanding the photovoltaic application scenarios to emerging fields such as building-integrated photovoltaics, automotive skylights, and photovoltaic agriculture, injecting new vitality into the photovoltaic industry.

[0003] Currently, the development of high-performance semi-transparent organic photovoltaic devices (ST-OPV) still faces a core challenge: how to achieve excellent photoelectric conversion efficiency while ensuring a high average visible light transmittance (AVT), break through the mutual restraint relationship between the two key indicators of PCE and AVT, and then improve the light utilization efficiency (LUE) of ST-OPV. This places higher requirements on the light absorption ability of the device in the non-visible light region, especially for near-infrared light that accounts for more than 50% of the solar energy and has not been efficiently utilized. The performance of ST-OPV based on the existing material system still has a large gap compared with the theoretical optimal value, and it is urgent to further improve the light utilization efficiency of the device on the basis of developing new materials. Developing efficient optical regulation means is an important way to improve the performance of ST-OPV. One of the most effective methods is to introduce an optical regulation structure on the transparent electrode of the device. Based on the mechanism of light wave interference and the matching relationship between the optical coefficients and thicknesses of each layer in the thin film optical system, precise regulation of the optical properties of the device can be achieved, and the visible light transmittance and near-infrared light absorption of ST-OPV can be improved synergistically. However, the design of such structures is complex and requires a supporting efficient optical design method for multiple targets and variables. However, the prediction ability of the high-throughput calculation relying only on the optical model for the multi-scale collaborative optimization scenario is still limited. In recent years, the deep learning model that has received much attention in the field of material design can better explore the structure-property relationship between multi-dimensional data with its powerful high-dimensional feature extraction ability, so as to guide the efficient optical design of the device. However, such models highly rely on a large amount of experimental data and have insufficient modeling ability for complex physical laws, and it is necessary to solve the problem of low model accuracy in the small experimental data samples of ST-OPV. Summary of the Invention

[0004] In view of the above defects of the prior art, the technical problem to be solved by the present invention is how to improve the photoelectric conversion efficiency (PCE) while ensuring a high average visible light transmittance (AVT), and break through the mutual restraint relationship between PCE and AVT of ST-OPV devices. In addition, the present invention aims to overcome the problem of low prediction accuracy of the deep learning model in the sparse experimental data scenario. By introducing a physics-guided deep learning method, the accuracy of the ST-OPV performance prediction model is improved, and the optimal optical regulation structure is efficiently screened to achieve the high-performance design of the transparent organic photovoltaic cell. The present invention provides a design method for an organic photovoltaic cell guided by physical fusion deep learning, which is a high-performance transparent organic photovoltaic cell design method guided by physical fusion deep learning. By integrating the prior physical information of the device optical analysis model into the construction and prediction process of the deep learning model, the prediction ability and physical accuracy of the model in the sparse data scenario of ST-OPV are significantly improved, guiding efficient optical design and then realizing the collaborative optimization of the device performance (PCE and AVT).

[0005] To achieve the above object, the present invention provides a design method for a high-performance transparent organic photovoltaic cell guided by physical fusion deep learning, establishes a ST-OPV performance prediction model of physical fusion deep learning, screens out the optimal optical regulation structure through the ST-OPV performance prediction model, and finally prepares a ST-OPV device according to the optimal optical regulation structure, and tests the PCE and AVT of the ST-OPV device.

[0006] Further, establishing a ST-OPV performance prediction model of physical fusion deep learning includes physically-guided data processing and physically-guided pre-training; wherein, the physically-guided data processing includes embedding physical information such as the ST-OPV device structure and optical characteristics into a deep learning model; the physically-guided pre-training includes using an optical model based on the transfer matrix method to generate a large amount of physical simulation data for pre-training, and then using the collected experimental observation data for fine-tuning.

[0007] Further, the physically-guided data processing specifically includes collecting the optical characteristics of the materials used in ST-OPV to obtain a material optical characteristic data set; then, using the material optical characteristic data set for data processing, converting physical information such as the optical characteristics of the materials into the form of neuron nodes and data streams through an attention mechanism algorithm, and embedding them into a deep learning model.

[0008] Further, the optical characteristics of the materials used in ST-OPV include multiple physical characteristics, namely wavelength W, material type M, material refractive index n, and material extinction coefficient k.

[0009] Further, the materials used in ST-OPV include organic active layer materials, metal materials, and dielectric materials.

[0010] Further, the refractive index and extinction coefficient are averaged and sampled at intervals of 10 nm from a wavelength of 300 nm to 1000 nm, and a total of 71 refractive index sampling points and 71 extinction coefficient sampling points are obtained.

[0011] Further, the physically-guided pre-training specifically includes using an optical model to generate a large amount of physical simulation data to obtain a physical simulation data set; then using the physical simulation data set to pre-train a deep learning model to obtain a pre-trained deep learning model; collecting experimental observation data of ST-OPV from an existing database to obtain an experimental observation data set; then using the experimental observation data set to fine-tune and train the pre-trained deep learning model to obtain a physically-guided ST-OPV performance prediction model.

[0012] Furthermore, the optimal optical modulation structure is screened out through the ST-OPV performance prediction model. Specifically, a physics-guided ST-OPV performance prediction model is used to find the optical modulation structure corresponding to the optimal device PCE and AVT through random grid search, including the material selection and the thickness of each layer; the optimal optical modulation screening is based on the product LUE of PCE and AVT, and the optical modulation structure corresponding to the highest LUE is screened out.

[0013] Furthermore, the ST-OPV device structure from bottom to top is an anode, a hole transport layer, an organic active layer, an electron transport layer, a transparent cathode, and an optical modulation layer.

[0014] Furthermore, finally, the ST-OPV device is prepared according to the optimal optical modulation structure, including preparing the anode, the hole transport layer, the organic active layer, the electron transport layer, the transparent cathode, and the optical modulation layer according to the optimal optical modulation structure.

[0015] Technical effects

[0016] A design method of a high-performance transparent organic photovoltaic cell guided by physical fusion deep learning according to the present invention establishes a physical fusion deep learning ST-OPV performance prediction model through physics-guided data processing and physics-guided pre-training, focusing on solving the challenge of the prediction accuracy of ST-OPV performance in sparse data scenarios and breaking through the limit of traditional optical simulation. Based on the accurate prediction of the model, an optical modulation structure that can maximize the synergistic improvement of PCE and AVT is screened. The LUE of the ST-OPV designed by the present invention is greater than 5%, proving the high efficiency of the present application in device design. The design method of the present application can make the design of photovoltaic devices shift from relying on traditional experience and trial-and-error methods to a more scientific and systematic data-driven design process. By introducing a physics-guided deep learning model, designers can identify and optimize key optical structure parameters in a shorter time, thereby realizing the rapid iteration and improvement of the performance of photovoltaic devices, providing theoretical support and optimization directions for the design of high-performance ST-OPV devices.

[0017] The concept, specific structure, and technical effects of the present invention will be further described below in conjunction with the drawings to fully understand the purpose, features, and effects of the present invention. Brief description of the drawings

[0018] Figure 1 It is a schematic diagram of a physical fusion deep learning ST-OPV performance prediction model of a preferred embodiment of the present invention;

[0019] Figure 2 It is a schematic diagram of the ST-OPV performance optimization guided by physical fusion deep learning of a preferred embodiment of the present invention;

[0020] Figure 3 Schematic diagram of the ST-OPV device structure of a preferred embodiment of the present invention;

[0021] Figure 4 Graph for evaluating and analyzing the model prediction accuracy of a preferred embodiment of the present invention;

[0022] Figure 5 Scatter plot of AVT-PCE for optimal optical regulation screening of a preferred embodiment of the present invention;

[0023] Figure 6 Current density-voltage curve of the optimal high-performance ST-OPV designed by physical fusion deep learning of a preferred embodiment of the present invention;

[0024] Figure 7 Transmittance curve of the optimal high-performance ST-OPV designed by physical fusion deep learning of a preferred embodiment of the present invention. Detailed implementation manners

[0025] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0026] In the following description, specific details such as specific internal programs and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present invention. However, those skilled in the art should clearly understand that the present invention can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits and methods are omitted to avoid unnecessary details from interfering with the description of the present invention.

[0027] As Figure 1-2 shown, the present invention provides a design method for a high-performance transparent organic photovoltaic cell guided by physical fusion deep learning, establishes a physical fusion deep learning-based ST-OPV performance prediction model, screens out the optimal optical regulation structure through the ST-OPV performance prediction model, and finally prepares an ST-OPV device according to the optimal optical regulation structure and tests the PCE and AVT of the ST-OPV device. As Figure 3 shown, in the embodiment of the present invention, the device structure includes an anode, a hole transport layer, an organic active layer, an electron transport layer, a transparent cathode and an optical regulation layer.

[0028] Among them, a ST-OPV performance prediction model based on physical fusion deep learning is established, including physics-guided data processing and physics-guided pre-training; among them, physics-guided data processing includes embedding physical information such as ST-OPV device structure and optical characteristics into the deep learning model; physics-guided pre-training includes using an optical model based on the transfer matrix method to generate a large amount of physical simulation data for pre-training, and then using the collected experimental observation data for fine-tuning.

[0029] An embodiment of the present invention provides a design method for a high-performance transparent organic photovoltaic cell guided by physical fusion deep learning, specifically including the following steps:

[0030] Step 1, physics-guided data processing: Collect the optical characteristics of the materials used in ST-OPV to obtain a material optical characteristic data set; then, use the material optical characteristic data set for data processing, and convert physical information such as the optical characteristics of the materials into forms such as neuron nodes and data streams through the attention mechanism algorithm, and embed them into the deep learning model;

[0031] Specifically, the material optical characteristics include multiple groups of physical features, namely wavelength W, material type M, material refractive index n, and material extinction coefficient k. The materials used in ST-OPV include organic active layer materials, metal materials, and dielectric materials. The optical characteristics of the materials used in ST-OPV include wavelength-dependent refractive index and wavelength-dependent extinction coefficient. The refractive index and extinction coefficient are averaged and sampled at intervals of 10 nm from 300 nm to 1000 nm in wavelength, and a total of 71 refractive index sampling points and 71 extinction coefficient sampling points are obtained. In this embodiment, the storage method of the material optical characteristic data set is: "material name" + 71 refractive index sampling points + 71 extinction coefficient sampling points. The sorting method of the above sampling points is sorted from small to large based on the corresponding wavelength, and the storage format is csv.

[0032] The initial inputs are material optical characteristics (including wavelength W, material type M, material refractive index n, material extinction coefficient k) and material thickness d. Wavelength W, material refractive index n, and material extinction coefficient k are all real variables, and are mapped to high-dimensional vectors through a real number embedding network:

[0033] E W =f W (W,n,k)

[0034] Among them, E W is the material optical property embedding, and f W (·) is an independent fully connected neural network for learning the feature representation of material optical properties.

[0035] The material type M is a categorical variable and is transformed using an embedding matrix:

[0036] E M = f M (M)

[0037] where E M is the material type embedding, and f M (·) is an independent neural network for mapping the material type of discrete categories into vectors of a fixed dimension.

[0038] The material thickness d is a real variable and is embedded in a similar way to the wavelength W:

[0039] E d = f d (d)

[0040] where E d is the material optical property embedding, and f d (·) is an independent fully connected neural network for the feature representation of the thickness of each functional layer in the device.

[0041] After the above steps, the deep learning model obtains three embedding vectors: the wavelength embedding E W , the material embedding E M and the thickness embedding E d . Connecting these three embeddings gives the feature vector:

[0042] E = [E W , E M , E d

[0043] Step 2. Physics-guided pre-training: Use the optical model to generate a large amount of physical simulation data to obtain a physical simulation data set; then use the physical simulation data set to pre-train the deep learning model to obtain a pre-trained deep learning model; collect the experimental observation data of ST-OPV from the existing database to obtain an experimental observation data set; then use the experimental observation data set to fine-tune the pre-trained deep learning model to obtain a physics-guided ST-OPV performance prediction model.

[0044] ​Among them, the method of generating a large amount of physical simulation data using an optical model is to define a virtual device structure, which includes an anode, a hole transport layer, an organic active layer, an electron transport layer, a transparent electrode 1, a transparent electrode 2, a dielectric layer 1, and a dielectric layer 2, a total of 8 layers. Among them, the transparent electrode 1 and the transparent electrode 2 form a transparent cathode, and the dielectric layer 1 and the dielectric layer 2 form an optical modulation layer. Randomly replace the specific materials of the above device structure. The thickness of the anode is randomly set between 100 nm and 200 nm, the thickness of the hole transport layer is randomly set between 30 nm and 60 nm, the thickness of the organic active layer is randomly set between 40 nm and 150 nm, the thickness of the transparent electrode 1 and the transparent electrode 2 is randomly set between 5 nm and 30 nm, and the thickness of the dielectric layer 1 and the dielectric layer 2 is randomly set between 0 nm and 400 nm. The optical model will output the corresponding device performance (PCE and AVT) according to the above input parameters. By looping the above steps, any number of physical simulation data can be obtained. Preferably, the number of physical simulation data is greater than 400,000.

[0045] The experimental observation data includes the ST-OPV device structure, material optical properties, and corresponding device performance (PCE and AVT) extracted from published literature. Preferably, the number of experimental observation data is greater than 200.

[0046] The storage method of the physical simulation data set and the experimental observation data set is: "Material 1" + "Material 2" + "Material 3" + "Material 4" + "Material 5" + "Material 6" + "Material 7" + "Material 8" + "Thickness 1" + "Thickness 2" + "Thickness 3" + "Thickness 4" + "Thickness 5" + "Thickness 6" + "Thickness 7" + "Thickness 8" + "AVT" + "PCE", and the storage format is csv.

[0047] Performance prediction includes predicting the corresponding device performance (PCE and AVT) from the given device structure and thickness, and inversely deducing the corresponding device structure from the optimal device performance (PCE and AVT).

[0048] The method used for performance prediction is Transformer encoding, and its initial input is the feature vector E obtained in step one. First, introduce the positional encoding PE:

[0049] E′ = E + PE

[0050] The core algorithm of the Transformer is the multi-head attention mechanism, and its calculation formula is as follows:

[0051]

[0052] Among them, Q, K, and V are the query, key, and value matrices respectively, which are linearly transformed from E′. d kis the dimension of the key matrix.

[0053] The multi-head attention can be expressed as:

[0054] Multihead(E′)=Concat(head1,…,head h )W O

[0055] where head i is obtained by separate attention calculations.

[0056] After the attention output, it passes through a two-layer feed-forward neural network:

[0057] FFN(x)=max(0,xW1+b1)W2+b2

[0058] where W1, W2, b1, and b2 are network parameters.

[0059] After passing through N layers, the Transformer outputs a vector T with feature transformation:

[0060] T=Transformer(E′)

[0061] The vector T obtained from the above transformation is input into an independent neural network to predict the key performance (AVT and PCE) of ST-OPV, and the prediction function is:

[0062] [AVT,PCE]=g(T)

[0063] where g(·) is a multi-layer neural network.

[0064] As Figure 4 shown, after evaluating the prediction accuracy of the model in this embodiment through the k-fold cross-validation method, the coefficient of determination R 2 > 0.85, and the Pearson correlation coefficient and Spearman correlation coefficient > 0.99.

[0065] Step 3. Optimal optical regulation screening: Using the physically-guided ST-OPV performance prediction model, find the optical regulation structure corresponding to the optimal PCE and AVT of the device through random grid search, including its material selection and the thickness of each layer; the optimal optical regulation screening is based on the product LUE of PCE and AVT, and screen the optical regulation structure corresponding to the highest LUE.

[0066] Preferably, the algorithm used in the method for searching for the optimal LUE is the random grid search algorithm. Its input includes the variables to be optimized and the objective function. The variables to be optimized include the material type M and the thickness d. The material type M is the set of possible material types {M1, M2,…, M N}. The thickness d ranges from [d min , d max . In this embodiment, M is {ITO, PEDOT:PSS, PM6:BTP-eC9:L8-BO, PNDIT-F3N, Au, Ag, m1, m2}, where m1 and m2 are the materials of the first optical modulation layer 61 and the second optical modulation layer 62 respectively. The thickness of ITO is 150 nm, the thickness of PEDOT:PSS is 30 nm, the thickness of the organic active layer is 90 nm, the thickness of PNDIT-F3N is 10 nm, the thickness of Au is 1 nm, the thickness of Ag is 10 nm. The thickness range of m1 and m2 is [0, 400], with the unit of nm. The objective functions include AVT and PCE.

[0067] Randomly generate candidate samples among the above variables to be optimized. Randomly select m1 and n2 from the material type set {ITO, PEDOT:PSS, PM6:BTP-eC9:L8-BO, PNDIT-F3N, Au, Ag, M1, M2}, and randomly select the corresponding thicknesses d1 and d2 of m1 and m2 within the thickness range [0, 400] to generate N s groups of random candidate structures:

[0068] {(M1, d1), (M2, d2), …, (M Ns , d NS )}

[0069] where N S is the random sampling quantity.

[0070] Use the prediction method in step two to traverse N S candidate structures and find the (M * , d * ) corresponding to the maximum value of the objective function:

[0071]

[0072] Preferably, N S is taken as 10000 to ensure sufficient search coverage. As Figure 5 shown, the output results are sorted in descending order of the AVT×PCE value to predict 1000 groups of optical modulation structures. In this embodiment, the preferred (M * , d * ) is ({ITO, PEDOT:PSS, PM6:BTP-eC9:L8-BO, PNDIT-F3N, Au, Ag, TeO2}, {150, 30, 90, 10, 1, 10, 141}), and the thickness unit is nm.

[0073] Step 4. Experimental verification: Fabricate the ST-OPV device using the optimal optical modulation structure selected by the physically-guided ST-OPV performance prediction model, and test the PCE and AVT of the device. The experimental verification is to fabricate the ST-OPV device using the optimal optical modulation structure predicted by the deep learning model, test the device performance (PCE and AVT), and then input the experimentally measured values as feedback into the deep learning model.

[0074] The fabrication steps of the ST-OPV device are as follows: successively perform anode fabrication, hole transport layer fabrication, organic active layer fabrication, electron transport layer fabrication, transparent cathode fabrication, and optical modulation layer fabrication. The thickness of each layer is obtained by the random grid search algorithm in Step 3.

[0075] Specifically, for anode fabrication: deposit indium tin oxide (ITO) on the surface of the transparent substrate by magnetron sputtering to obtain an ITO substrate composed of the transparent substrate and the indium tin oxide conductive thin film; then, clean the ITO substrate.

[0076] Specifically, for hole transport layer fabrication: spin-coat a poly(3,4-ethylenedioxythiophene):poly(styrenesulfonate) (PEDOT:PSS) solution on the surface of the ITO substrate treated in an ultraviolet-ozone environment, and perform annealing treatment to obtain a PEDOT:PSS thin film.

[0077] Specifically, for organic active layer fabrication: First, mix PM6, BTP-eC9, and L8-BO according to mass to obtain an organic donor-acceptor blend solution; then, spin-coat the organic donor-acceptor blend solution on the obtained PEDOT:PSS thin film and perform annealing treatment to obtain an organic active layer thin film.

[0078] Specifically, for electron transport layer fabrication: spin-coat a PNDIT-F3N solution on the organic active layer thin film to obtain a PNDIT-F3N thin film.

[0079] Specifically, for transparent cathode fabrication: In a vacuum environment, successively deposit Au and Ag on the surface of the PNDIT-F3N thin film by thermal evaporation to obtain a transparent cathode thin film.

[0080] Specifically, for optical modulation layer fabrication: In a vacuum environment, successively deposit dielectric materials on the transparent cathode thin film. The specific materials are obtained by optical modulation screening in Step 3 of the design method of a high-performance transparent organic photovoltaic cell guided by physical fusion deep learning, thereby obtaining the ST-OPV device.

[0081] Further, the cleaning step in the anode preparation process is specifically as follows: First, ultrasonically clean the surface of the ITO substrate with dishwashing liquid, deionized water, acetone, and isopropyl alcohol respectively; then, place the cleaned ITO substrate in a high-temperature air-circulating drying oven and dry it at 60 - 100 °C; thereafter, place the dried ITO substrate in a plasma cleaner and clean it for 4 - 6 min.

[0082] In the process of preparing the hole transport layer, the aqueous solution concentration of PEDOT:PSS is 1.3 - 1.7 wt%, the rotation speed of spin coating is 2500 - 3500 rpm, and the spin coating time is 28 - 32 s; the annealing treatment temperature is 125 - 135 °C and the annealing time is 8 - 12 min.

[0083] In the process of preparing the organic active layer, the mass ratio of PM6, BTP-eC9, and L8-BO is 0.6:0.4:0.8, and the total concentration is 16 mg mL -1 , and in the process of preparing the organic active layer, the spin coating of the organic donor-acceptor blend solution is carried out in an atmosphere filled with nitrogen; the rotation speed of spin coating is 3000 - 6000 rpm, and the spin coating time is 28 - 32 s; the annealing treatment temperature is 95 - 105 °C and the annealing time is 8 - 12 min; before implementing spin coating, the organic donor-acceptor blend solution is heated at a constant temperature of 40 - 60 °C for 1 - 3 h.

[0084] In the process of preparing the electron transport layer, the concentration of the PNDIT-F3N solution is 0.95 mg mL -1 , the rotation speed of spin coating is 2500 - 3500 rpm, and the spin coating time is 28 - 32 s;

[0085] In the process of preparing the transparent cathode, the deposition rate of Au is 0.01 - 0.02 nm / s, and the deposition rate of Ag is 0.1 - 0.2 nm / s; in the process of preparing the optical modulation layer, the deposition rate of the dielectric material is 0.2 - 0.4 nm / s.

[0086] The present invention provides a physical fusion deep learning-guided ST-OPV preparation method, specifically as follows:

[0087] Step 1: Magnetron sputter ITO on the surface of a transparent substrate to obtain an ITO conductive film, and the transparent substrate and the ITO conductive film constitute an ITO substrate.

[0088] Step 2: Use a cleaning machine to clean the surface of the ITO substrate, and place the cleaned ITO substrate in a high-temperature air-circulating drying oven for drying. Preferably, the cleaning agent includes dishwashing liquid, deionized water, acetone, and isopropyl alcohol. Ultrasonically clean the ITO substrate with dishwashing liquid, deionized water, acetone, and isopropyl alcohol respectively, and dry it at 80 °C in a high-temperature air-circulating drying oven.

[0089] Step 3: Clean the dried ITO substrate with a plasma cleaner. Preferably, the cleaning time is 5 minutes.

[0090] Step 4: Spin-coat the hole transport layer PEDOT:PSS solution on the surface of the ITO substrate to obtain a PEDOT:PSS thin film. The ITO substrate and the PEDOT:PSS thin film constitute a PEDOT:PSS substrate. Preferably, the PEDOT:PSS solution is filtered through a mixed cellulose aqueous filter head with a pore size of 0.22 μm before use to obtain a filtered PEDOT:PSS solution. Spin-coat the filtered PEDOT:PSS solution on the ITO substrate at a speed of 3000 rpm for 30 seconds, and then anneal it on a 130°C annealing table for 10 minutes to obtain a PEDOT:PSS thin film with a thickness of 30 nm.

[0091] Step 5: Configure the organic donor PM6, the organic acceptor BTP-eC9, and L8-BO according to the mass ratio to obtain a PM6:BTP-eC9:L8-BO blend solution. Preferably, dissolve PM6:BTP-eC9:L8-BO in chloroform at a mass ratio of 0.6:0.4:0.8, and add 1,3,5-trichlorobenzene with a mass concentration of 10 mg mL -1 to obtain a PM6:BTP-eC9:L8-BO blend solution with a concentration of 16 mg mL -1 .

[0092] Step 6: Spin-coat the PM6:BTP-eC9:L8-BO blend solution on the surface of the PEDOT:PSS substrate to obtain a PM6:BTP-eC9:L8-BO thin film. The PM6:BTP-eC9:L8-BO thin film and the PEDOT:PSS substrate constitute a PM6:BTP-eC9:L8-BO substrate. Preferably, use a magnetic stirrer to stir the PM6:BTP-eC9:L8-BO blend solution at 50°C for 1 hour before spin-coating to obtain a fully dissolved PM6:BTP-eC9:L8-BO blend solution. Spin-coat the fully dissolved PM6:BTP-eC9:L8-BO blend solution on the PEDOT:PSS substrate at a speed of 5000 rpm in a glove box with an oxygen and moisture level less than 0.01 ppm for 30 seconds, and then anneal it on a 100°C annealing table for 10 minutes to obtain a PM6:BTP-eC9:L8-BO thin film with a thickness of 90 nm.

[0093] Step 7: Configure the electron transport layer PNDIT-F3N according to the concentration ratio to obtain a PNDIT-F3N solution. Preferably, dissolve PNDIT-F3N at 1.0 mg mL -1is dissolved in methanol containing 5 vol% acetic acid co-solvent to obtain a PNDIT-F3N solution.

[0094] Step Eight: Spin-coat the PNDIT-F3N solution on the surface of the PM6:BTP-eC9:L8-BO substrate to obtain a PNDIT-F3N thin film. The PNDIT-F3N thin film and the PM6:BTP-eC9:L8-BO substrate form a PNDIT-F3N substrate. Preferably, spin-coat the PNDIT-F3N solution on the PM6:BTP-eC9:L8-BO substrate at a speed of 3000 rpm for 30 seconds in a glove box with oxygen and moisture levels less than 0.01 ppm to obtain a PNDIT-F3N thin film with a thickness of 10 nm.

[0095] Step Nine: In a vacuum environment, sequentially evaporate and deposit Au and Ag on the surface of the PNDIT-F3N substrate to obtain a composite transparent conductive thin film. The composite transparent conductive thin film and the PNDIT-F3N substrate form an Ag substrate. Preferably, sequentially thermally evaporate and deposit Au with a thickness of 1 nm at a rate of 0.0015 nm s -1 at a rate of, and thermally evaporate and deposit Ag with a thickness of 10 nm at a rate of 0.02 nm s -1 at a rate of.

[0096] Step Ten: In a vacuum environment, evaporate and deposit a dielectric material TeO2 thin film on the surface of the Ag substrate. Preferably, evaporate and deposit a dielectric material TeO2 thin film with a thickness of 141 nm at a rate of 0.03 nm s -1 on the surface of the Ag substrate.

[0097] The physical fusion deep learning-guided high-performance transparent organic photovoltaic cell design method of the present invention embeds the prior physical information of the optical model into an advanced deep learning architecture to construct a physical fusion deep learning ST-OPV performance prediction model, effectively solving the problem of insufficient prediction ability of the model in sparse data scenarios. Guided by the performance prediction model, accurately explore the optimal combination of optical materials and device structures, significantly improve the near-infrared light capture ability of the device while ensuring excellent visible light transmittance, and achieve the optimal balance of high optical transmittance and high photoelectric conversion in ST-OPV. Finally, as Figure 6-7 shown, in this embodiment, the optimal ST-OPV designed by deep learning can achieve a PCE of 13.19%, an AVT of 38.0%, and an LUE of 5.01%.

[0098] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative efforts. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field based on the concept of the present invention through logical analysis, reasoning, or limited experiments on the basis of the prior art should fall within the protection scope determined by the claims.

Claims

1. A design method for organic photovoltaic cells guided by physical fusion deep learning, characterized in that: A ST-OPV performance prediction model integrating physics and deep learning is established, and the optimal optical regulation structure is screened out through the ST-OPV performance prediction model. Finally, ST-OPV devices are prepared according to the optimal optical regulation structure, and the PCE and AVT of the ST-OPV devices are tested.

2. A method for designing an organic photovoltaic cell guided by physical fusion deep learning as claimed in claim 1, characterized in that: A ST-OPV performance prediction model integrating physics and deep learning is established, including physics-guided data processing and physics-guided pre-training; wherein, the physics-guided data processing includes embedding physical information such as ST-OPV device structure and optical properties into the deep learning model; the physics-guided pre-training includes using an optical model based on the transfer matrix method to generate massive physical simulation data for pre-training, and then using the collected experimental observation data for fine-tuning.

3. A design method for an organic photovoltaic cell guided by physical fusion deep learning as claimed in claim 2, characterized in that: Physics-guided data processing specifically includes collecting the optical properties of materials used in ST-OPV to obtain a data set of material optical properties. Then, the data set of material optical properties is used for data processing to convert physical information such as the optical properties of the materials into the form of neuron nodes and data streams through an attention mechanism algorithm, and then embedded in a deep learning model.

4. A method for designing an organic photovoltaic cell guided by physical fusion deep learning as claimed in claim 3, characterized in that: The optical properties of the materials used in the ST-OPV include multiple sets of physical characteristics, namely wavelength W, material type M, material refractive index n, and material extinction coefficient k.

5. The design method of an organic photovoltaic cell guided by physical fusion deep learning as claimed in claim 4, characterized in that: The materials used in ST-OPV include organic active layer materials, metal materials and dielectric materials.

6. A method for designing an organic photovoltaic cell guided by physical fusion deep learning as claimed in claim 5, characterized in that: The refractive index and the extinction coefficient are sampled on average at wavelengths of 300 nm to 1000 nm and intervals of 10 nm, and a total of 71 refractive index sampling points and 71 extinction coefficient sampling points are obtained.

7. The design method of an organic photovoltaic cell guided by physical fusion deep learning as claimed in claim 1, characterized in that: The physical-guided pre-training specifically includes using an optical model to generate massive physical simulation data to obtain a physical simulation data set; then using the physical simulation data set to pre-train a deep learning model to obtain a pre-trained deep learning model; collecting experimental observation data of ST-OPV from an existing database to obtain an experimental observation data set; then using the experimental observation data set to fine-tune the pre-trained deep learning model to obtain a physical-guided ST-OPV performance prediction model.

8. The method for designing an organic photovoltaic cell guided by physical fusion deep learning as claimed in claim 7, characterized in that: The optimal optical regulation structure is screened out through the ST-OPV performance prediction model. Specifically, the physics-guided ST-OPV performance prediction model is used to find the optical regulation structure corresponding to the optimal PCE and AVT of the device through random grid search, including its material selection and the thickness of each layer; the optimal optical regulation screening is based on the product of PCE and AVT, LUE, as the standard, to screen the optical regulation structure corresponding to the highest LUE.

9. A method for designing an organic photovoltaic cell guided by physical fusion deep learning as claimed in claim 8, characterized in that: The ST-OPV device structure comprises, from bottom to top, an anode, a hole transport layer, an organic active layer, an electron transport layer, a transparent cathode and an optical regulation layer.

10. The design method of an organic photovoltaic cell guided by physical fusion deep learning as claimed in claim 9, characterized in that: Finally, an ST-OPV device is prepared according to the optimal optical regulation structure, including preparing an anode, a hole transport layer, an organic active layer, an electron transport layer, a transparent cathode and an optical regulation layer according to the optimal optical regulation structure.

Images