CN111625998A - Method for optimizing structure of laminated solar cell - Google Patents
Method for optimizing structure of laminated solar cell Download PDFInfo
- Publication number
- CN111625998A CN111625998A CN202010473536.0A CN202010473536A CN111625998A CN 111625998 A CN111625998 A CN 111625998A CN 202010473536 A CN202010473536 A CN 202010473536A CN 111625998 A CN111625998 A CN 111625998A
- Authority
- CN
- China
- Prior art keywords
- layer
- solar cell
- iteration
- series
- layers
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 51
- 238000013528 artificial neural network Methods 0.000 claims abstract description 34
- 238000005457 optimization Methods 0.000 claims abstract description 32
- 210000004027 cell Anatomy 0.000 claims description 95
- 230000006870 function Effects 0.000 claims description 33
- 238000010606 normalization Methods 0.000 claims description 15
- 210000002569 neuron Anatomy 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 12
- 230000003044 adaptive effect Effects 0.000 claims description 12
- 238000013461 design Methods 0.000 abstract description 29
- 238000004088 simulation Methods 0.000 description 22
- 238000012549 training Methods 0.000 description 14
- 239000000463 material Substances 0.000 description 6
- 230000009286 beneficial effect Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 5
- 238000012938 design process Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 230000035772 mutation Effects 0.000 description 2
- 230000002028 premature Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000000547 structure data Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
- H01L31/00—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof
- H01L31/04—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof adapted as photovoltaic [PV] conversion devices
- H01L31/042—PV modules or arrays of single PV cells
- H01L31/043—Mechanically stacked PV cells
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
- H01L31/00—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof
- H01L31/04—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof adapted as photovoltaic [PV] conversion devices
- H01L31/042—PV modules or arrays of single PV cells
- H01L31/05—Electrical interconnection means between PV cells inside the PV module, e.g. series connection of PV cells
- H01L31/0504—Electrical interconnection means between PV cells inside the PV module, e.g. series connection of PV cells specially adapted for series or parallel connection of solar cells in a module
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/50—Photovoltaic [PV] energy
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computer Hardware Design (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Electromagnetism (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Condensed Matter Physics & Semiconductors (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Power Engineering (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Geometry (AREA)
- Sustainable Development (AREA)
- Photovoltaic Devices (AREA)
Abstract
The invention belongs to the field of cell design, and particularly relates to a method for optimizing a laminated solar cell structure, which comprises the following steps: taking the structural information of the stacked solar cell to be optimized as population information of a differential evolution algorithm and taking a cell performance index as an optimization target of the differential evolution algorithm, and initializing the structural information; and controlling a differential evolution algorithm to carry out iterative evolution on the initial structure information for multiple times by adaptively adjusting the scaling factor and the cross probability required by each iteration, wherein each iterative evolution is to carry out joint adjustment on each layer of structure in the laminated solar cell to obtain a new population, and adopting a pre-constructed cell performance prediction neural network to predict the cell performance index according to the new population to finally obtain the optimal structure information. The self-adaptive differential evolution algorithm can jointly adjust each layer structure, avoids the problem of local optimization, combines the differential evolution algorithm with the battery performance prediction neural network, can efficiently and time-saving self-adaptive reverse optimization design the battery structure, and improves the optimization efficiency.
Description
Technical Field
The invention belongs to the field of cell design, and particularly relates to a method for optimizing a laminated solar cell structure.
Background
The types of solar cells are various, and among them, the solar cell adopting the stacked structure is a common solar cell type at present, and the photoelectric conversion efficiency is relatively high. There is always a great deal of interest in how to efficiently design its laminate structure.
The efficiency of a tandem solar cell depends on its structure, for example, the thickness and material used for each layer, the stacking manner between layers, and the shape of each layer all affect the efficiency of the tandem solar cell, and when the material for each layer is determined, the efficiency directly depends on the structure of the tandem solar cell. Therefore, in the design scheme of the traditional laminated solar cell structure, the design is mainly carried out through a numerical simulation tool and a manual adjustment two-cycle process. For a designed laminated solar cell, it is first necessary to determine the structure (including the thickness of each layer, the stacking manner between layers, the shape of each layer, and the like) and the constituent materials of the laminated solar cell. After the materials are determined, the thickness of each layer, the stacking mode among the layers, the shape of each layer and the like are initialized to be reasonable at will, and then a mathematical model required by numerical simulation is established by using a computer language according to the requirements of a simulation tool (such as FDTD software of Lumerical corporation). And then, gradually adjusting the structure of the laminated solar cell according to the logic of adjusting the thickness composition, the stacking mode and the shape of each layer, and also gradually adjusting the structure according to other adjusting logics. For example, the thickness composition is adjusted first, one layer may be selected as the layer to be adjusted, and the thicknesses of the other layers are fixed. And simulating the current structure by using a numerical simulation tool so as to obtain the current value and the efficiency value of the laminated solar cell. Further judging whether the performance of the current laminated solar cell is optimal or not, and if not, guiding how to adjust the current structure to be optimized through the prior knowledge of experts; if the structure is optimal, the next group of structures to be adjusted is selected according to the adjustment logic to be adjusted. When the stacked solar cell structures have all been adjusted to be optimal, the design process is complete.
However, the drawbacks of the conventional stacked solar cell structure design are mainly reflected in the following aspects: (1) each time a different simulation model has to be created after adjusting for one layer of the stacked solar cell structure, this requires the intervention of a professional and also consumes unnecessary time in the creation and fine tuning of the theoretical model. (2) Each time the structure of the stacked solar cell is adjusted, not only the simulation model needs to be re-established, but also a large amount of time needs to be consumed again to perform simulation calculation and model solution to obtain the current and the efficiency of the stacked solar cell. (3) The evaluation and adjustment of the simulation result of the laminated solar cell require manual intervention, and the structure data of the cell is not single dimension, so that the joint adjustment of the structure of the cell is difficult to perform manually. After one variable is adjusted once for optimization, the parameter is fixed to adjust other parameters, so that the performance of the laminated solar cell obtained by adjustment is easy to fall into local optimization, and a global optimal structure is difficult to find.
Disclosure of Invention
The invention provides a method for optimizing a laminated solar cell structure, which is used for solving the technical problem of low optimization efficiency caused by long optimization period and easy local optimization of the structure in the existing method for optimizing the laminated solar cell structure.
The technical scheme for solving the technical problems is as follows: a method for optimizing a stacked solar cell structure, comprising:
taking the structural information of the stacked solar cell to be optimized as population information of a differential evolution algorithm and taking a cell performance index as an optimization target of the differential evolution algorithm, and initializing the structural information;
and controlling the differential evolution algorithm to carry out iterative evolution on initial structure information for multiple times by adaptively adjusting scaling factors and cross probabilities required by each iteration, wherein each iterative evolution is to carry out joint adjustment on each layer of structure in the laminated solar cell to obtain a new population, and predicting the cell performance index by adopting a pre-constructed cell performance prediction neural network according to the new population to finally obtain an optimal cell performance index and corresponding structure information thereof so as to complete the optimization of the laminated solar cell structure.
The invention has the beneficial effects that: the method utilizes the characteristic that the deep neural network can approach the nonlinear function with any precision, the performance attribute is accurately predicted from the structural data through the battery performance prediction neural network, and the battery performance prediction neural network can be used for replacing a professional numerical simulation tool to simulate the performance of the laminated solar battery, so that a large amount of simulation time and calculation resources can be saved. Secondly, the self-adaptive differential evolution algorithm is used for replacing manual adjustment, adjustment time can be saved, meanwhile, combined adjustment can be carried out to find the optimal structure of the laminated solar cell, the defect that the performance of the laminated solar cell is easy to fall into local optimization due to the structure designed in the prior art is overcome, and the performance of the laminated solar cell can reach global optimization. The differential evolution algorithm is combined with a battery performance prediction neural network, the structure of the laminated solar battery can be efficiently and time-saving in a self-adaptive reverse optimization design mode, the performance of the laminated solar battery can reach the expected performance, the design efficiency is improved, and the performance of the final laminated solar battery is superior to that of the laminated solar battery obtained by a traditional design method. In addition, the structure design method of the laminated solar cell provided by the method has strong universality, the laminated solar cell with different layers is used for different selected materials, and the optimal structure can be quickly found without the intervention of professionals in the design process.
On the basis of the technical scheme, the invention can be further improved as follows.
Further, the adaptive adjustment amplitude of the scaling factor and the cross probability in the early stage of evolution is larger than the adaptive adjustment amplitude in the later stage of evolution.
The invention has the further beneficial effects that: the self-adaptive design of the self-adaptive scaling factor and the cross probability enables the self-adaptive differential evolution algorithm to have a wider search range in the early search period and have stronger refined search capability in the later search period.
Further, the scaling factor is F and is represented as
The cross probability is α and is denoted as
Wherein i is the current iteration frequency, e is a natural base number,
to represent
Rounded down.
The invention has the further beneficial effects that: the convergence rate of the algorithm can be increased, the algorithm is prevented from being premature, and in addition, the minimum limit of the scaling factor F is 0.1, so that the adaptive differential evolution algorithm still has a variation function in the later iteration stage. The method has the advantages that the cross probability is higher in the initial stage of iterative optimization, so that the algorithm is easy to jump out of local optimization, the cross probability is lower in the later stage of iterative optimization, the algorithm is not easy to cross other variables, and the local search capability of the algorithm is enhanced.
Further, the iteration termination criterion of the differential evolution algorithm is as follows:
judging whether the values of the target functions of the differential evolution algorithm are better than the values of all the target functions obtained by previous iteration, if so, continuing the iteration; otherwise, selecting to continue or terminate the iteration based on the current iteration times; the target function is obtained by taking boundary conditions and limiting conditions of a battery structure as Lagrange constraints and combining the Lagrange constraints with battery performance indexes obtained by the battery performance prediction neural network prediction.
The invention has the further beneficial effects that: boundary conditions and limiting conditions of the battery structure are used as Lagrange constraints and used for calculating the target function, and algorithm accuracy can be improved.
Further, the objective function is f and is represented by
Wherein n is the type of the battery performance index output by the battery performance prediction neural network g (, the T is the total battery structure information, and w is the total battery structure informationi，gi(T) are respectively a weighting coefficient of the ith battery performance and a value of the ith battery performance predicted by the battery performance prediction neural network, m is the total number of layers of the laminated solar battery to be optimized,
tjand the structure information of the j layer corresponding to the T is Lagrange constraint factor, and the value is 1.
Further, the selecting to continue or terminate the iteration based on the current iteration number specifically includes: if the current iteration times are more than five hundred times and no more optimal objective function value is found in the last three hundred iterations, the optimization is finished, and the iteration is terminated.
Further, the neural network includes: the system comprises a serial input module consisting of a batch normalization layer and a plurality of full connection layers which are sequentially connected in series, a parallel hidden module consisting of a plurality of full connection layer serial branches which are connected in parallel, and a serial output module consisting of a connection layer and a plurality of full connection layers which are sequentially connected in series;
wherein each full-link layer is activated by an activation function, and a plurality of all full-link layers are activated by L1And/or L2Regularization constraints; and the output of the series input module is respectively input to each full-connection layer series branch, and the output of all the full-connection layer series branches is converged to the connection layer.
The invention has the further beneficial effects that: the introduction of the batch normalization layer aims to accelerate the training speed of the network, larger training rate parameters can be used, and more batch training data can be used at one time during training. In thatThe parallel hidden module uses a parallel structure, so that each branch in parallel connection is not connected with each other, and parameters of the network to be trained are reduced. Multiple L's are suitably employed throughout the network1And/or L2And regularizing the constraints so as to prevent over-fitting or under-fitting and improve the network prediction accuracy.
Furthermore, the number of the full-connection layer serial branches is three, and each branch comprises five full-connection layers, the number of the neurons in serial connection is 12, 10, 8 and 5 respectively; the total connection layer in the series connection output module has five layers, the number of neurons in each layer is 20,12,8, 5 and 3 in sequence, and the number of neurons in the connection layer is 18; the total connection layer in the series input module has three layers, and the number of neurons in each layer is 5, 10 and 12 in sequence.
Furthermore, the first full-connection layer connected in series in the series input module adopts L1And L2Regularized constraint operation, L1And L2The normalization coefficients are 0.001 and 0.002 respectively;
l is respectively adopted in the last two full-connection layers in the series output module2The constraint operation is regularized, and the L2 normalization coefficients are all 0.005.
The present invention also provides a machine-readable storage medium having stored thereon machine-executable instructions which, when invoked and executed by a processor, cause the processor to carry out a method of stacked solar cell structure optimization as described above.
Drawings
Fig. 1 is a flowchart of a method for optimizing a stacked solar cell structure according to an embodiment of the present invention;
fig. 2 is a flowchart of another method for optimizing a stacked solar cell structure according to an embodiment of the present invention;
FIG. 3 is a diagram of a performance prediction neural network according to an embodiment of the present invention;
fig. 4 is a flowchart of a structural design of a stacked solar cell according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Example one
A method for optimizing a stacked solar cell structure, as shown in fig. 1, includes:
taking the structural information of the stacked solar cell to be optimized as population information of a differential evolution algorithm and taking a cell performance index as an optimization target of the differential evolution algorithm, and initializing the structural information;
and controlling the differential evolution algorithm to carry out iterative evolution on initial structure information for multiple times by adaptively adjusting scaling factors and cross probabilities required by each iteration, wherein each iterative evolution is to carry out joint adjustment on each layer of structure in the laminated solar cell to obtain a new population, and predicting the cell performance index by adopting a pre-constructed cell performance prediction neural network according to the new population to finally obtain an optimal cell performance index and corresponding structure information thereof so as to complete the optimization of the laminated solar cell structure.
The self-adaptive differential evolution algorithm takes an efficiency prediction network as a proxy function, takes the structure of the laminated solar cell as an independent variable, weights the predicted obtained current according to the influence of the current on the positive and negative of the efficiency and takes the weighted current as a target value to be optimized, continuously and jointly adjusts the structure of the laminated solar cell (among layers) through optimizing the target value, and continuously and self-adaptively adjusts a scaling factor and a cross probability in an iteration process. And until the self-adaptive differential evolution algorithm converges, at the moment, the structure of the laminated solar cell is optimal.
Aiming at the defects of too long simulation time, the need of intervention of professionals and low manual adjustment efficiency in the traditional design method, the method for designing the structure of the laminated solar cell by combining the battery performance prediction network designed based on the deep neural network with the self-adaptive differential evolution algorithm is provided. The characteristic that the deep neural network can approach the nonlinear function with any precision is utilized, the performance attribute is accurately predicted from the structural data through the battery performance prediction neural network, and the battery performance prediction neural network can be used for replacing a professional numerical simulation tool to simulate the performance of the laminated solar battery, so that a large amount of simulation time and calculation resources can be saved. Taking the data of a 12500-group five-layer laminated solar cell as an example, the battery performance prediction neural network only needs 0.22 second, and the traditional FDTD simulation needs 630983 seconds, so that the former can save about 250 ten thousand times of time. And the latter needs to complete calculation simulation on a high-performance server or a computer cluster, and the method can complete the design process on a common household computer.
Secondly, the self-adaptive differential evolution algorithm is used for replacing manual adjustment, adjustment time can be saved, meanwhile, combined adjustment can be carried out to find the optimal structure of the laminated solar cell, the defect that the performance of the laminated solar cell is easy to fall into local optimization due to the structure designed in the prior art is overcome, and the performance of the laminated solar cell can reach global optimization. The differential evolution algorithm is combined with a battery performance prediction neural network, the structure of the laminated solar battery can be efficiently and time-saving in a self-adaptive reverse optimization design mode, the performance of the laminated solar battery can reach the expected performance, the design efficiency is improved, and the performance of the final laminated solar battery is superior to that of the laminated solar battery obtained by a traditional design method. Taking the design of a five-layer laminated solar cell structure as an example, the laminated solar optimal cell efficiency obtained by adopting the design based on the deep neural network and the self-adaptive differential evolution algorithm is 11.8% higher than the optimal cell performance obtained by combining the traditional manual adjustment and the FDTD simulation design. Meanwhile, the total thickness of the optimal laminated solar cell structure optimally found by the self-adaptive differential evolution algorithm is thinner than that designed by the traditional method, and the practical application of the subsequent cell is facilitated. Similarly, taking data of a five-layer laminated solar cell as an example, the total thickness of the structure obtained by the traditional design method is 385nm, and the total thickness of the structure designed by the method is 345nm, which is 40nm thinner than the former.
In addition, the structure design method of the laminated solar cell provided by the method has strong universality, the laminated solar cell with different layers is used for different selected materials, and the optimal structure can be quickly found without the intervention of professionals in the design process.
Preferably, the adaptive adjustment amplitude of the scaling factor and the cross probability in the early stage of evolution is larger than the adaptive adjustment amplitude in the late stage of evolution. The design of the self-adaptive scaling factor and the cross probability is mainly to enable the self-adaptive differential evolution algorithm to have a wider search range in the early search period and have stronger refined search capability in the later search period.
Preferably, an adaptive scaling factor is added to the mutation operation in the adaptive differential evolution algorithm:
this can speed up the convergence speed of the algorithm while preventing the algorithm from becoming premature. Wherein F is the scaling factor of each variation, i is the current iteration number, e is the natural base number,
to represent
Rounded down. Take iteration 0 as an example, F equals 0.6, assuming x1As the individual to be mutated, x2、x3For randomly selecting individuals from the population, the value of the individuals after mutation is V ═ x1+0.6(x2-x3). Obviously, x1Larger variance values are easily obtained, making it easier for the function to jump out of local optima. Taking the 1000 th iteration as an example, F is 0.134, and the variation value obtained by the individual to be varied is smaller, so that the capability of the algorithm in local search is enhanced. Meanwhile, in order that the adaptive differential evolution algorithm still has a variation function in the later iteration stage, the minimum limit of the scaling factor F in the embodiment is 0.1.
Meanwhile, self-adaptive adjustment is adopted for the cross probability:
where α is the crossover probability, i is the current iteration number, α is limited to a size of 0.3 to 0.7, and α is 0.7 for the 0 th iteration, assuming that the variable to be crossed is xi，j(g) Alternative crossover individuals are ui，j(g) The variable after crossing is ui，j(g) Wherein i is the ith individual, j is the jth variable in the ith individual, and g is the number of iterations, then
Apparently, for the same individual, the algorithm has a larger cross probability at the initial stage of iterative optimization, so that the algorithm is easy to jump out of the local optimum, and taking the 1000 th iteration as an example, α is 0.3, at this time, the cross probability of the individual is relatively small, the individual is not easy to cross with other variables, and at this time, the local search capability of the algorithm is enhanced.
Preferably, the iteration termination criterion of the differential evolution algorithm is as follows:
judging whether the value of the target function of the differential evolution algorithm is better than the values of all target functions obtained by previous iteration, if so, continuing the iteration; otherwise, selecting to continue or terminate the iteration based on the current iteration times; the target function is obtained by combining boundary conditions and limiting conditions of the battery structure as Lagrange constraints with battery performance indexes obtained by predicting the battery performance through a neural network.
Preferably, the objective function is f and is expressed as
Wherein n is the type of the battery performance output by the battery performance prediction neural network g (, the T is the total battery structure information, and w is the total battery structure informationi，gi(T) are respectively a weighting coefficient of the ith battery performance and a value of the ith battery performance predicted by the battery performance prediction neural network, m is the total number of layers of the laminated solar battery to be optimized,
tjand the structure information of the j layer corresponding to the T is Lagrange constraint factor, and the value is 1.
Preferably, the selecting to continue or terminate the iteration based on the current iteration number specifically includes: adaptively, if the current iteration time is more than five hundred times and no more optimal objective function value is found in the last three hundred iterations, the optimization is finished, and the iteration is terminated.
As shown in fig. 2, the adaptive differential evolution algorithm initializes the maximum iteration number, the population number, the boundary condition, adds the limiting condition, performs variation and cross selection on the population, and then, when calculating the target value according to the objective function, uses the thickness boundary condition and the thickness limiting condition as the lagrangian constraint, combines the lagrangian constraint with the objective function, and takes the lagrangian constraint factor as a constant 1. When the algorithm fails to find a better value for the last 300 iterations, the iteration is stopped.
Preferably, the neural network includes: the system comprises a serial input module consisting of a batch normalization layer and a plurality of full connection layers which are sequentially connected in series, a parallel hidden module consisting of a plurality of full connection layer serial branches which are connected in parallel, and a serial output module consisting of a connection layer and a plurality of full connection layers which are sequentially connected in series; wherein each full connection layer is activated by an activation function, and a plurality of all full connection layers adopt L1And/or L2Regularization constraints; and the output of the series input module is respectively input to each full-connection layer series branch, and the output of all full-connection layer series branches is converged to the connection layer.
The neural network can receive cell structure information of the laminated solar cell (for example, each layer thickness is used as input data, and corresponding cell performance information (for example, current value) is predicted in a very short time). in all modules of the neural network, basic units are a batch normalization layer, a full connection layer, a Relu activation layer and a regularization layer, as shown in FIG. 3, the neural network is divided into three parts, namely a series input SIN, a parallel hidden PN and a series output SON, wherein the input is each layer thickness of the laminated solar cell, and the output is the current property of the laminated solar cell.
Preferably, among the SINs, the batch normalization layer BN is first used, then the three fully-connected layers FC are connected, and L is used simultaneously after the first fully-connected layer1And L2Regularize the constraint operation and activate all fully-connected layers using the Relu function (f (x) ═ max (0, x)); the intermediate parallel hidden part PN is also composed of a full connection layer FC and an activation function Relu (excluding a batch normalization layer); among the output series layers SON, the three output layers of the PN part are first connected in parallel as a whole layer, and then passed through a series of full connection layers FC (five layers) and activated using Relu. Using two L's at the last two full link layers of the SON simultaneously2And regularizing the constraint layer. The parallel structure is used in the SON, so that each branch in parallel connection is not connected with each other, and parameters of the network needing to be trained are reduced.
The series input part SIN starts with a batch normalization layer BN after accepting the structure of the stacked solar cells as input data, in order to speed up the training of the network, a larger training rate parameter can be used, and more batches of training data can be used at a time during training. Connect full connectivity layer FC with hidden neuron number 5 after batch normalization layer and add L1、L2Normalizing the layer, outputting through a Relu activation function, then passing through a full junction FC with the hidden neuron number of 10 and a Relu activation layer again, and finally passing through the full junction FC with the hidden neuron number of 12 and the Relu activation layer; the parallel hidden part PN is formed by connecting three identical series sub-networks in parallel, each sub-network comprises five full connection layers FC and Relu activation layers, the number of hidden layer neurons contained in the basic units is 12, 10, 8 and 5, and the Relu function is uniformly used by the activation layer function; the serial output part SON firstly connects three sub-networks at the end of the parallel hidden part in parallel to form a whole layer connecting layer (conditioner layer), and then the number of hidden neurons of the three sub-networks is respectively equal to that of the full connecting layer and the active layer of the parallel hidden part PN sub-network through the combined structure of the full connecting layer and the active layerFor 18,20,12,8, the activation layer also uses Relu function, the last two layers of the series output part SON are full connection layer FC and Relu activation layer combined with L2 regularization layer, and the number of hidden layer neurons of these two layers is 5 and 3 respectively.
To connect L in series input section SIN1And L2The normalization coefficients were set to 0.001,0.002 and the last two L2 normalization coefficients were set to 0.005. In addition, on the premise of designing the same hidden neural unit, 3171 parameters need to be trained by adopting a serial design mode, but the 1164 parameters only need to be trained by adopting a parallel design mode, so that the quantity of the parameters can be reduced by 63.3%. Initialization of other parameters employs randomized initialization in the present invention.
For the training of the network, 80% of the simulation data set which has been obtained can be used as a training set, the remaining 20% can be used as a test set, and all data are subjected to packet training and testing in turn by adopting a 5-fold cross validation mode.
For example, as shown in fig. 4, the development concept of the laminated solar cell structure design mainly includes: the method comprises the steps of simulation data set preprocessing, efficiency prediction network building and initialization, efficiency prediction network training and adaptive differential evolution algorithm optimization of a solar cell structure.
Firstly, in the acquisition of a data set, through an FDTD numerical simulation tool, the structural data of the laminated solar cells with the same number of layers and different thickness combinations of the cells are simulated in batch, and then the current data corresponding to each group of structures can be obtained.
Furthermore, in the simulation data set preprocessing module, the structural data in the simulation data set is amplified, so that further feature extraction and observation calculation can be conveniently carried out on the data subsequently. And meanwhile, dividing the data into 5 parts (the number of the last equal parts which cannot be divided is reduced or increased according to the situation) so as to facilitate the subsequent 5-fold cross validation training method.
For the data set preprocessing section, the obtained structural data can be normalized to 10nm units and the current data to mA/cm2 units.
In a second aspect, based on the efficiency prediction network of the deep neural network, the corresponding current properties can be accurately predicted from the structure of the stacked solar cell after training. The neural network can be used for replacing a professional numerical simulation tool to simulate the current of the laminated solar cell.
In a third aspect, an adaptive differential algorithm, in combination with the efficiency prediction network, can adaptively and reversely optimize and design the structure of the stacked solar cell, so that the efficiency of the stacked solar cell reaches the expected efficiency.
Through effective normalized data and reasonable division, a prediction network is effectively trained, the structure of the laminated solar cell is optimally designed by using a self-adaptive differential evolution algorithm, the design efficiency is improved, and the final efficiency of the laminated solar cell is superior to that of the laminated solar cell obtained by the traditional design method.
Example two
A machine-readable storage medium having stored thereon machine-executable instructions which, when invoked and executed by a processor, cause the processor to carry out a method of stacked solar cell structure optimization as described in embodiment one above.
The related technical solution is the same as the first embodiment, and is not described herein again.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A method for optimizing a stacked solar cell structure, comprising:
taking the structural information of the stacked solar cell to be optimized as population information of a differential evolution algorithm and taking a cell performance index as an optimization target of the differential evolution algorithm, and initializing the structural information;
and controlling the differential evolution algorithm to carry out iterative evolution on initial structure information for multiple times by adaptively adjusting scaling factors and cross probabilities required by each iteration, wherein each iterative evolution is to carry out joint adjustment on each layer of structure in the laminated solar cell to obtain a new population, and predicting the cell performance index by adopting a pre-constructed cell performance prediction neural network according to the new population to finally obtain an optimal cell performance index and corresponding structure information thereof so as to complete the optimization of the laminated solar cell structure.
2. The method of claim 1, wherein the scaling factor and the crossover probability are larger in magnitude for adaptive adjustment in early stages of evolution than in late stages of evolution.
3. The method of claim 2, wherein the scaling factor is F and is expressed as
The cross probability is α and is denoted as
Wherein i is the current iteration frequency, e is a natural base number,
to represent
Rounded down.
4. The method of claim 1, wherein the iteration termination criterion of the differential evolution algorithm is:
judging whether the values of the target functions of the differential evolution algorithm are better than the values of all the target functions obtained by previous iteration, if so, continuing the iteration; otherwise, selecting to continue or terminate the iteration based on the current iteration times; the target function is obtained by taking boundary conditions and limiting conditions of a battery structure as Lagrange constraints and combining the Lagrange constraints with battery performance indexes obtained by the battery performance prediction neural network prediction.
5. The method of claim 4, wherein the objective function is f and is expressed as
Wherein n is the type of the battery performance index output by the battery performance prediction neural network g (, the T is the total battery structure information, and w is the total battery structure informationi，gi(T) are respectively a weighting coefficient of the ith battery performance and a value of the ith battery performance predicted by the battery performance prediction neural network, m is the total number of layers of the laminated solar battery to be optimized,
tjand the structure information of the j layer corresponding to the T is Lagrange constraint factor, and the value is 1.
6. The method according to claim 4, wherein the selection of the continuation or termination iteration based on the current iteration number is specifically: if the current iteration times are more than five hundred times and no more optimal objective function value is found in the last three hundred iterations, the optimization is finished, and the iteration is terminated.
7. The method of any one of claims 1 to 6, wherein the neural network comprises: the system comprises a serial input module consisting of a batch normalization layer and a plurality of full connection layers which are sequentially connected in series, a parallel hidden module consisting of a plurality of full connection layer serial branches which are connected in parallel, and a serial output module consisting of a connection layer and a plurality of full connection layers which are sequentially connected in series;
wherein each full-link layer is activated by an activation function, and a plurality of all full-link layers are activated by L1And/or L2Regularization constraints; and the output of the series input module is respectively input to each full-connection layer series branch, and the output of all the full-connection layer series branches is converged to the connection layer.
8. The method for optimizing the structure of the laminated solar cell according to claim 7, wherein the number of the fully-connected layer serial branches is three, and each branch comprises five fully-connected layers which are respectively 12, 10, 8 and 5 in the number of the neurons which are sequentially connected in series; the total connection layer in the series connection output module has five layers, the number of neurons in each layer is 20,12,8, 5 and 3 in sequence, and the number of neurons in the connection layer is 18; the total connection layer in the series input module has three layers, and the number of neurons in each layer is 5, 10 and 12 in sequence.
9. The method of claim 8, wherein the first fully-connected layer of the series input module is L1And L2Regularized constraint operation, L1And L2The normalization coefficients are 0.001 and 0.002 respectively;
l is respectively adopted in the last two full-connection layers in the series output module2The constraint operation is regularized, and the L2 normalization coefficients are all 0.005.
10. A machine-readable storage medium having stored thereon machine-executable instructions which, when invoked and executed by a processor, cause the processor to carry out a method of stacked solar cell structure optimization according to any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010473536.0A CN111625998A (en) | 2020-05-29 | 2020-05-29 | Method for optimizing structure of laminated solar cell |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010473536.0A CN111625998A (en) | 2020-05-29 | 2020-05-29 | Method for optimizing structure of laminated solar cell |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111625998A true CN111625998A (en) | 2020-09-04 |
Family
ID=72259394
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010473536.0A Pending CN111625998A (en) | 2020-05-29 | 2020-05-29 | Method for optimizing structure of laminated solar cell |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111625998A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113591298A (en) * | 2021-07-27 | 2021-11-02 | 武汉理工大学 | Optical structure optimization design method based on deep neural network |
| CN113919576A (en) * | 2021-10-15 | 2022-01-11 | 厦门大学 | Method for predicting structural performance of solar cell |
| CN114581446A (en) * | 2022-05-06 | 2022-06-03 | 浙江双元科技股份有限公司 | Battery core abnormity detection method and system of laminated battery |
| WO2023060579A1 (en) * | 2021-10-15 | 2023-04-20 | 厦门大学 | Solar cell structure performance prediction method |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2002003716A1 (en) * | 2000-06-30 | 2002-01-10 | British Telecommunications Public Limited Company | A memetic method for multiobjective optimisation |
| US20130317667A1 (en) * | 2010-09-30 | 2013-11-28 | Empire Technology Development Llc | Automatic flight control for uav based solid modeling |
| CN104991997A (en) * | 2015-06-11 | 2015-10-21 | 北京航空航天大学 | Generalized rate related P-I hysteresis model-establishing method of adaptive difference evolutionary algorithm optimization |
| CN106300418A (en) * | 2016-08-30 | 2017-01-04 | 合肥工业大学 | Photovoltaic DC-to-AC converter based on adaptive differential evolution algorithm controls the discrimination method of parameter |
| CN109635506A (en) * | 2019-01-10 | 2019-04-16 | 浙江大学 | The photovoltaic cell model parameter identification method of adaptive chaos tree and seed algorithm |
| CN110298495A (en) * | 2019-06-11 | 2019-10-01 | 天津大学 | The array optimization method of wave-power device based on modified differential evolution algorithm |
| CN110474904A (en) * | 2019-08-16 | 2019-11-19 | 武汉思普崚技术有限公司 | A kind of Situation Awareness method and system improving prediction |
| CN110490326A (en) * | 2019-07-17 | 2019-11-22 | 长沙学院 | A kind of differential evolution method and system of adaptive multi-Vari strategy |
| CN110765956A (en) * | 2019-10-28 | 2020-02-07 | 西安电子科技大学 | Double-person interactive behavior recognition method based on component characteristics |
| CN110764255A (en) * | 2019-10-21 | 2020-02-07 | 天津大学 | Design and optimization method of waveguide display sub-wavelength diffraction grating |
| CN111090941A (en) * | 2019-12-17 | 2020-05-01 | 哈尔滨工业大学 | Spacecraft optimal Lambert orbit rendezvous method based on multi-objective optimization algorithm |
-
2020
- 2020-05-29 CN CN202010473536.0A patent/CN111625998A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2002003716A1 (en) * | 2000-06-30 | 2002-01-10 | British Telecommunications Public Limited Company | A memetic method for multiobjective optimisation |
| US20130317667A1 (en) * | 2010-09-30 | 2013-11-28 | Empire Technology Development Llc | Automatic flight control for uav based solid modeling |
| CN104991997A (en) * | 2015-06-11 | 2015-10-21 | 北京航空航天大学 | Generalized rate related P-I hysteresis model-establishing method of adaptive difference evolutionary algorithm optimization |
| CN106300418A (en) * | 2016-08-30 | 2017-01-04 | 合肥工业大学 | Photovoltaic DC-to-AC converter based on adaptive differential evolution algorithm controls the discrimination method of parameter |
| CN109635506A (en) * | 2019-01-10 | 2019-04-16 | 浙江大学 | The photovoltaic cell model parameter identification method of adaptive chaos tree and seed algorithm |
| CN110298495A (en) * | 2019-06-11 | 2019-10-01 | 天津大学 | The array optimization method of wave-power device based on modified differential evolution algorithm |
| CN110490326A (en) * | 2019-07-17 | 2019-11-22 | 长沙学院 | A kind of differential evolution method and system of adaptive multi-Vari strategy |
| CN110474904A (en) * | 2019-08-16 | 2019-11-19 | 武汉思普崚技术有限公司 | A kind of Situation Awareness method and system improving prediction |
| CN110764255A (en) * | 2019-10-21 | 2020-02-07 | 天津大学 | Design and optimization method of waveguide display sub-wavelength diffraction grating |
| CN110765956A (en) * | 2019-10-28 | 2020-02-07 | 西安电子科技大学 | Double-person interactive behavior recognition method based on component characteristics |
| CN111090941A (en) * | 2019-12-17 | 2020-05-01 | 哈尔滨工业大学 | Spacecraft optimal Lambert orbit rendezvous method based on multi-objective optimization algorithm |
Non-Patent Citations (3)
| Title |
|---|
| 刘树强等: "一种求解动态优化问题的改进自适应差分进化算法", 《计算机工程》, vol. 47, no. 4, 3 April 2020 (2020-04-03), pages 1 - 9 * |
| 姚敏立等: "基于动态参数差分进化算法的多约束稀布矩形面阵优化", 《电子与信息学报》, vol. 42, no. 5, 15 May 2020 (2020-05-15), pages 1 - 7 * |
| 石佳佳: "差分进化算法在阵列天线波束赋形中的应用研究", 《中国优秀硕士学位论文全文数据库（电子期刊）信息科技辑》, vol. 136, 15 December 2019 (2019-12-15), pages 1 - 72 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113591298A (en) * | 2021-07-27 | 2021-11-02 | 武汉理工大学 | Optical structure optimization design method based on deep neural network |
| CN113591298B (en) * | 2021-07-27 | 2024-03-08 | 武汉理工大学 | Optical structure optimization design method based on deep neural network |
| CN113919576A (en) * | 2021-10-15 | 2022-01-11 | 厦门大学 | Method for predicting structural performance of solar cell |
| WO2023060579A1 (en) * | 2021-10-15 | 2023-04-20 | 厦门大学 | Solar cell structure performance prediction method |
| CN114581446A (en) * | 2022-05-06 | 2022-06-03 | 浙江双元科技股份有限公司 | Battery core abnormity detection method and system of laminated battery |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111625998A (en) | Method for optimizing structure of laminated solar cell | |
| CN112116144B (en) | Regional power distribution network short-term load prediction method | |
| CN111260211A (en) | Intelligent energy system evaluation method and device based on AHP-improved entropy weight method-TOPSIS | |
| CN109635506A (en) | The photovoltaic cell model parameter identification method of adaptive chaos tree and seed algorithm | |
| CN111242287A (en) | Neural network compression method based on channel L1 norm pruning | |
| CN112036512A (en) | Image classification neural network architecture searching method and device based on network clipping | |
| CN108038507A (en) | Local receptor field extreme learning machine image classification method based on particle group optimizing | |
| CN109193756B (en) | Scene decoupling dynamic economic dispatching model solving method of wind power access system | |
| CN110837915B (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
| CN113762387B (en) | Multi-element load prediction method for data center station based on hybrid model prediction | |
| CN113485826A (en) | Load balancing method and system for edge server | |
| CN112287990A (en) | Model optimization method of edge cloud collaborative support vector machine based on online learning | |
| CN113972667A (en) | Power distribution network reactive power optimization method based on one-dimensional convolutional neural network | |
| CN113570039A (en) | Optimized consensus block chain system based on reinforcement learning | |
| CN112487938A (en) | Method for realizing garbage classification by utilizing deep learning algorithm | |
| Li et al. | Short term prediction of photovoltaic power based on FCM and CG-DBN combination | |
| CN113537365A (en) | Multitask learning self-adaptive balancing method based on information entropy dynamic weighting | |
| CN114612761A (en) | Network architecture searching method for image recognition | |
| CN115577864A (en) | Distribution network operation optimization scheduling method based on multi-model combined operation | |
| CN112036651A (en) | Electricity price prediction method based on quantum immune optimization BP neural network algorithm | |
| CN104219682A (en) | Method and system for constructing network layers of hybrid power supply wireless sensor networks of farmlands | |
| CN113473580B (en) | User association joint power distribution method based on deep learning in heterogeneous network | |
| CN113722980A (en) | Ocean wave height prediction method, system, computer equipment, storage medium and terminal | |
| CN113177673A (en) | Air conditioner cold load prediction optimization method, system and equipment | |
| CN111985598A (en) | Configuration method of distributed power supply |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |