A method, system and device for performance optimization of a nanofluid photovoltaic and thermal system

Abstract

The application discloses a kind of nanofluid photovoltaic and thermal system performance optimization method, system and device, it is related to solar comprehensive utilization and artificial intelligence cross technical field, this method includes obtaining nanofluid photovoltaic / thermal system high-throughput dataset based on light-electricity-heat coupling model;With the geometric structure parameter of nanofluid system as input, with the performance evaluation index corresponding to nanofluid photovoltaic / thermal system as output, train deep neural network DNN forward prediction model;DNN forward prediction model is used as the fitness function of genetic algorithm, with geometric structure parameter as optimization variable, with the maximum performance evaluation index as optimization goal to carry out iterative optimization;The optimal geometric structure parameter combination that makes performance evaluation index maximization is output;The method can quickly and accurately predict the performance of nanofluid PV / T system, and efficiently realize multi-parameter global optimization.

Description

Technical Field

[0001] This invention relates to the field of integrated solar energy utilization and artificial intelligence, specifically to a method, system, and device for performance optimization of a nanofluid photovoltaic and thermal system. Background Technology

[0002] Photovoltaic (PV) technology can directly convert solar radiation into electrical energy and is currently the most widely used form of solar energy utilization. However, traditional photovoltaic cells can only generate electricity using specific wavelengths of the solar spectrum, while the radiation energy in other wavelengths is converted into heat, causing the cell's operating temperature to rise. This not only wastes energy but also significantly reduces photoelectric conversion efficiency, a problem that limits the further development of photovoltaic systems.

[0003] Photovoltaic / thermal (PV / T) systems integrate photovoltaic modules with solar collectors, generating both electricity and heat simultaneously, thus significantly improving the overall utilization rate of solar energy. In particular, nanofluid-based spectral PV / T systems, by controlling parameters such as the type, size, and volume fraction of nanoparticles, can precisely design the radiation characteristics of the nanofluids. This allows them to selectively absorb long-wave solar radiation that is ineffective against photovoltaic cells and convert it into heat, while efficiently transmitting short-wave solar radiation to power the cells, thereby achieving efficient utilization of the entire spectrum. Despite the significant advantages of nanofluid PV / T systems, their performance optimization faces considerable challenges. A complex nonlinear coupling relationship exists between the system's photoelectric conversion efficiency and photothermal conversion efficiency, and it is influenced by multiple parameters, including nanoparticle radius, volume fraction, and fluid layer thickness.

[0004] For performance optimization of nanofluidic PV / T systems, some studies have established experimental platforms to obtain photoelectric and photothermal performance under different parameter combinations. However, experimental methods require verification of each parameter combination individually, resulting in long research cycles and high costs, which significantly slows down product iteration in practical applications. In terms of numerical simulation, current research has established optical and thermal loss models or two-dimensional-three-dimensional coupled numerical models for evaluation. However, model simplification and assumptions cannot fully capture the multi-parameter coupling interactions, causing the system performance under actual operating conditions to deviate from the design expectations and resulting in energy waste. In addition, some researchers have attempted to use genetic algorithms or optimization methods based on Mie scattering theory for parameter optimization, but these works mostly remain at the parameter research level and fail to systematically explore the full range of parameter combinations. The optimality of the obtained parameters is difficult to guarantee, and in practical engineering, the true global optimal design point may be missed, limiting the improvement of full-spectrum utilization efficiency.

[0005] In summary, existing performance optimization methods for nanofluidic PV / T systems struggle to accurately capture multi-parameter coupling, leading to actual performance deviating from design expectations. Furthermore, traditional optimization algorithms are prone to getting trapped in local optima, making it difficult to fully explore high-dimensional parameter spaces and ensuring the reliability of the global optimal solution, resulting in low energy utilization. Summary of the Invention

[0006] To address the shortcomings of existing technologies, such as low efficiency, difficulty in fully exploring high-dimensional parameter spaces, and inability to guarantee obtaining the global optimal solution, this invention proposes a performance optimization method, system, and device for nanofluidic photovoltaic and thermal systems. It proposes a DNN-GA framework that can accurately find the "sweet spot" parameters that maximize the overall performance of the system based on the intrinsic properties of different materials, thereby solving the problems existing in the prior art.

[0007] A method for performance optimization of a nanofluid photovoltaic and thermal system includes the following steps: A high-throughput dataset of nanofluid photovoltaic and thermal systems based on an optical-electrical-thermal coupling model was obtained; the high-throughput dataset includes the geometric structural parameters of the nanofluid system and the corresponding performance evaluation indicators. Using the geometric structural parameters of the nanofluid system as input and the performance evaluation index corresponding to the nanofluid photovoltaic and thermal system as output, a deep neural network (DNN) positive prediction model is trained to construct a nonlinear mapping relationship between the geometric structural parameters and the performance evaluation index. The genetic algorithm uses the nonlinear mapping relationship between geometric structure parameters and performance evaluation indicators as its fitness function, with geometric structure parameters as the variables to be optimized and maximizing the performance evaluation indicator as the optimization objective. Specifically, it involves: randomly generating an initial population within a given range of geometric structure parameters, where each individual is encoded as a combination of geometric structure parameters; inputting each individual into a trained DNN forward prediction model, obtaining the corresponding predicted performance evaluation indicator value based on the nonlinear mapping relationship, and using this predicted value as the individual's fitness; iteratively performing selection, crossover, mutation, and fitness evaluation operations based on the individual's fitness value until the convergence condition is met, outputting the optimal combination of geometric structure parameters that maximizes the performance evaluation indicator. The optimal combination of geometric parameters is used to maximize the performance of nanofluid photovoltaic and thermal systems.

[0008] Furthermore, the establishment of a high-throughput dataset for nanofluidic photovoltaic and thermal systems based on a light-electricity-thermal coupling model specifically includes the following steps: The physical model and key geometric parameters of the nanofluidic photovoltaic / thermal system were determined. The physical model, from top to bottom, consists of a nanofluidic layer, a glass cover, a silicon photovoltaic cell, and a bottom water-cooling channel. The key geometric parameters include the radius of the nanoparticles.r Volume fraction of nanoparticles f v and the thickness of the nanofluid layer h ; Based on the radius and complex refractive index of the nanoparticles and the complex refractive index of the base fluid, the optical properties of a single particle are calculated using the Lorenz-Mie theory. Combined with the volume fraction of the nanoparticles, the effective radiation characteristic parameters of the nanofluid system are calculated based on the independent scattering approximation. Based on the effective radiation characteristic parameters of the nanofluid system, the thickness of the nanofluid layer, and the refractive index of each layer of the nanofluid photovoltaic and thermal system, the improved Monte Carlo method MCML is used to solve the radiation transfer equation RTE by tracking the transmission of photon packets in the multilayer medium, and the spectral transmittance, spectral absorptivity, and spectral reflectance of the nanofluid photovoltaic / thermal system within a set wavelength range are calculated. The spectral transmittance, spectral absorptivity, and spectral reflectance are input into a coupled photo-electric-thermal model to calculate the performance indicators of the nanofluid photovoltaic and thermal system. The performance indicators include photoelectric conversion efficiency and photothermal conversion efficiency. A comprehensive performance evaluation index MF is introduced to balance the photoelectric and photothermal outputs of the nanofluid photovoltaic and thermal system. By iteratively changing the nanoparticle radius, volume fraction, and fluid layer thickness, high-throughput theoretical calculations are performed to construct a high-throughput dataset of nanofluid photovoltaic and thermal systems that includes geometric structural parameters and corresponding performance evaluation indicators.

[0009] Furthermore, the comprehensive performance index MF is expressed as follows: ; In the formula, or pv,unfiltered The efficiency of the bare cell without nanofluid filtration. or th For photothermal conversion efficiency, or pv For photoelectric conversion efficiency, w This is the weighting factor for electrical energy relative to thermal energy.

[0010] Furthermore, the training of the deep neural network (DNN) forward prediction model specifically involves updating the weights and biases in the model using the backpropagation algorithm and gradient descent optimizer to minimize the loss function; the loss function is the mean squared error (MSE) between the model's predicted values ​​and the actual values ​​in the dataset.

[0011] Furthermore, the iterative selection, crossover, mutation, and fitness evaluation operations based on the individual's fitness value specifically include the following steps: Based on the individual's fitness value, roulette wheel selection or tournament selection methods are used to select individuals with high fitness from the current population as parents; The selected parent individuals are crossbred according to a preset crossover probability to generate offspring individuals; The gene loci in the offspring individuals are mutated according to a preset mutation probability to introduce new gene information; The offspring individuals generated through selection, crossover, and mutation operations are merged with the parent individuals to generate a new generation of population.

[0012] Furthermore, the convergence condition is reaching a preset maximum number of generations or the change in the fitness value of the best individual in a population across multiple consecutive generations being less than a preset threshold.

[0013] Furthermore, the individual coding method adopts real number coding, using the values ​​of geometric structural parameters as gene loci.

[0014] The present invention also includes a performance optimization system for a nanofluid photovoltaic and thermal system, comprising: The acquisition module is used to acquire a high-throughput dataset of nanofluid photovoltaic and thermal systems based on the photo-electric-thermal coupling model; the high-throughput dataset includes the geometric structural parameters of the nanofluid system and the corresponding performance evaluation indexes; The model training module is used to train a deep neural network (DNN) positive prediction model by taking the geometric structural parameters of the nanofluid system as input and the performance evaluation index corresponding to the nanofluid photovoltaic and thermal system as output, so as to construct a nonlinear mapping relationship between the geometric structural parameters and the performance evaluation index. The optimization module uses the nonlinear mapping relationship between geometric structure parameters and performance evaluation indicators as the fitness function of the genetic algorithm. Geometric structure parameters are used as the variables to be optimized, and maximizing the performance evaluation indicator is the optimization objective. Specifically, it includes: randomly generating an initial population within a given range of geometric structure parameters, where each individual is encoded as a combination of geometric structure parameters; inputting each individual into the trained DNN forward prediction model, obtaining the corresponding predicted performance evaluation indicator value based on the nonlinear mapping relationship, and using this predicted value as the individual's fitness; iteratively performing selection, crossover, mutation, and fitness evaluation operations based on the individual's fitness value until the convergence condition is met, outputting the optimal combination of geometric structure parameters that maximizes the performance evaluation indicator. The optimization module is used to maximize the performance of nanofluid photovoltaic and thermal systems based on the optimal combination of geometric parameters.

[0015] The present invention also includes a computer device for performance optimization of a nanofluid photovoltaic and thermal system, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the performance optimization method for the nanofluid photovoltaic and thermal system.

[0016] The present invention also includes a readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, are used to perform the steps of the performance optimization method for the nanofluid photovoltaic and thermal system.

[0017] This invention provides a method for performance optimization of nanofluid photovoltaic and thermal systems, which has the following beneficial effects: This invention constructs a deep neural network forward prediction model, using nanoparticle radius, volume fraction, and fluid layer thickness as inputs, and photoelectric conversion efficiency, photothermal conversion efficiency, and comprehensive performance indicators as outputs. This achieves high-precision and rapid prediction of the performance of nanofluid photovoltaic / thermal systems, avoiding the problems of low computational efficiency and high experimental costs caused by repeated complex physical simulations or experimental tests in traditional methods. Furthermore, this invention collaboratively optimizes the trained deep neural network with a genetic algorithm, using the deep neural network as the fitness function evaluator and maximizing the comprehensive performance indicator as the optimization objective. The genetic algorithm's global search capability efficiently optimizes in a complex high-dimensional nonlinear parameter space, overcoming the shortcomings of traditional optimization methods that rely on trial and error and are prone to getting trapped in local optima. It reliably obtains the globally optimal combination of geometric parameters that maximizes the system's comprehensive performance. Finally, based on the optimal combination of geometric parameters, a nanofluid photovoltaic / thermal system can be prepared or controlled, realizing a complete closed loop from performance prediction and global optimization to practical application. This provides an efficient, accurate, and universally applicable intelligent solution for the rational design of nanofluid photovoltaic / thermal systems. Attached Figure Description

[0018] Figure 1 This is a flowchart of a performance optimization method for nanofluid photovoltaic and thermal systems based on deep neural networks and genetic algorithms, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of a deep neural network (DNN) positive prediction model in an embodiment of the present invention; Figure 3 This is a two-dimensional performance cloud map of different nanofluids (Ag, Au, Al) under the influence of complex multi-parameters in the embodiments of the present invention; Figure 4 This is a comparison chart of the accuracy of the PV / T system model used for model validation in an embodiment of the present invention; Figure 5This is a comparison chart of the prediction effects of deep neural network model (DNN), decision tree (DT), and random forest (RF) models on the comprehensive performance (MF) of nanofluids, taking Ag nanofluids as an example in this embodiment of the invention. Figure 6 A radar chart comparing six evaluation metrics of three machine learning models in this embodiment of the invention; Figure 7 This is a transmission and absorption spectrum of Ag, Au, and Al nanofluids under the optimal parameter combination after optimization by the DNN-GA framework in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] This invention proposes a performance optimization method for nanofluid photovoltaic and thermal systems. Taking a spectroscopic PV / T system based on three nanofluids, Ag, Au, and Al, as an example, it is also applicable to other types of nanoparticle materials (such as Cu, carbon nanotubes, and mixed nanofluids).

[0021] like Figure 1 As shown, the method specifically includes the following steps: S1. Establish a high-throughput dataset for nanofluidic PV / T systems based on an optical-electrical-thermal coupling model.

[0022] S1.1: Determine the physical model and key geometric parameters of the nanofluidic PV / T system under study. The system model, from top to bottom, consists of: a nanofluidic layer, a glass cover plate, a silicon photovoltaic cell, and a bottom water-cooling channel. Key geometric parameters include: the radius of the nanoparticles. r Volume fraction of nanoparticles f v and the thickness of the nanofluid layer h ;

[0023] S1.2: To construct a high-throughput dataset, it is necessary to calculate the system performance indicators under different combinations of geometric parameters using a theoretical model. The specific calculation process is as follows. Based on the radius of the nanoparticles... r and complex refractive index m p = n p + I p ( i (where the imaginary unit is used), and the complex refractive index of the base fluid (water). m host = nm + I m The optical properties of a single particle were calculated based on the Lorenz-Mie theory. For a radius of... r spherical particles with extinction cross section C ext scattering cross section C sca and scattering phase function p Through Mie scattering coefficient a n and b n The calculation yields:

[0024] (1) (2) (3) In the formula, k 1 = 2π n m / l The wavenumber in the base fluid. l The incident light wavelength, n m The real part of the refractive index of the base liquid; i The scattering angle is the angle between the direction of the incident light and the direction of the scattered light. a n , b n The Mie scattering coefficient is given by the Riccati-Bessel function; S 11 , S 22 This is the scattering amplitude function.

[0025] Mie scattering coefficient a n and b n The expression is: (4) (5) in, x = k 1 r For particle size parameters, m = m p / m host The relative complex refractive index, ψ n ( x )and xn ( x ) are the Bessel function and the Hankel function for the first type of sphere, respectively.

[0026] Further combining the volume fraction of particles f v Based on the independent scattering approximation, the effective radiation characteristic parameters of the nanofluid system, including the attenuation coefficient, were calculated. scattering coefficient absorption coefficient and scattering phase function: (7) (8) (9) (10) In the formula, < V > r The average volume of the nanoparticles. k m This represents the imaginary part of the complex refractive index of the base liquid.

[0027] S1.3: Known optical properties of nanofluid layers ( ),thickness h The refractive indices of each layer in the system are considered, and the radiative transfer equation (RTE) is solved using a modified Monte Carlo method (MCML). This improved algorithm can accurately handle the reflection and refraction problems at the interfaces between absorbing media. The radiative transfer equation is as follows:

[0028] (11) In the formula, I Let be the spectral radiance along the direction Ω. s This represents the path length.

[0029] By tracking the transport of a large number of photon packets (e.g., 1 × 10⁶) in a multilayer medium, the spectral transmittance of the nanofluidic PV / T system in the wavelength range of 300–2500 nm was calculated. T ( l ), spectral absorbance A ( l and spectral reflectance R ( l As shown in equations (12)-(14); these spectral data are the basic inputs for subsequent calculations of the photoelectric and photothermal properties of the system: (12) (13) (14) In the formula, N 0 represents the total number of incident photons. N t , N r These represent the number of photons transmitted and reflected, respectively. Thus, step S1.3 completes the forward simulation from nanofluid geometry parameters to the system's spectral characteristics, and the obtained... T ( l ), A ( l )and R ( l This will be directly used in the calculation of photoelectric conversion efficiency and photothermal conversion efficiency in step S1.4.

[0030] When dealing with absorbing media interfaces, we first consider the case of perpendicular incidence. In this case, the reflection coefficient can be expressed using the complex refractive index as:

[0031] (15) For obliquely incident radiation, the reflection coefficients of the polarization components parallel and perpendicular to the plane of incidence are defined by the following equation: (16) (17) Reflection component of electric field vector – parallel component r ∥ and vertical components r ⊥ It can be represented as: (18) In the formula, total Fresnel reflection is the average value of the reflections of the two polarization components. r =( r ∥ + r ⊥ ) / 2. Simultaneously, the absorption index of the absorbing medium affects the calculation of the angle of refraction, leading to the generation of a complex angle of refraction. When both sides of the interface are absorbing media, the angle of refraction... i 2 can be calculated using the generalized Snell's law:

[0032] (19) in, p Defined by the following formula: (20) S1.4: Input the spectral data obtained in step S1.3 into the coupled optical-electrical-thermal model to calculate the system's performance indicators. First, based on spectral transmittance... T ( l ) and the spectral response of silicon cells SR ( l ), calculate the short-circuit current density J sc :

[0033] (twenty one) In the formula, G ( l () represents the AM1.5D solar irradiance spectrum.

[0034] Dark saturation current density of the battery J 0 is: (twenty two) in, K ′ is an empirical constant. T c The battery temperature is 298.15 K. E g For silicon band gap, k B is the Boltzmann constant.

[0035] Open circuit voltage V oc It is given by the following formula: (twenty three) In the formula, n d For diode ideality factor, q It represents the elementary charge.

[0036] The fill factor FF is: (twenty four) in V m = kV oc It is the voltage at the maximum power point, where k It's usually between 0.7 and 0.8; here, we've set its value to 0.78.

[0037] Combined with the battery's dark saturation current density J 0. Open circuit voltage V oc and fill factor FF The photoelectric conversion efficiency of the system is obtained. or pv : (25) System photothermal conversion efficiency or th Then it is calculated using the following formula: (26) In the formula, or collector The collector efficiency is taken as 0.67.

[0038] S1.5: Introduce the comprehensive performance evaluation index MF (Merit Function) to balance the photoelectric and photothermal outputs of the system. MF is defined as follows:

[0039] (27) In the formula, or pv,unfiltered The bare cell efficiency without nanofluid filtration is 15.89%. w This is the weighting factor for electrical energy relative to thermal energy (taken as 3).

[0040] Thus, through steps S1.2 to S1.5, the process of obtaining geometric parameters ( r , f v , h ) to performance metrics ( or pv , or th The forward computation of (MF) is performed. By repeating the above steps and changing the values ​​of the geometric parameters, a high-throughput dataset for subsequent machine learning training can be constructed.

[0041] S1.6: By changing the radius of nanoparticles r Volume fraction f v and fluid layer thickness h Repeat steps 1.2 to 1.5 to perform high-throughput theoretical calculations and construct a large dataset for machine learning training. In this embodiment, the parameter ranges are shown in Table 1. Ultimately, the dataset contains the geometric parameters of the nanofluidic system (…). r , f v , h ) and corresponding performance indicators ( or pv , or th The dataset is divided into training, validation, and test sets according to a certain ratio (e.g., 8:1:1).

[0042] Table 1. Data sets and parameter descriptions for three types of nanofluids S2. Establish and train a deep neural network (DNN) positive prediction model to achieve fast and accurate prediction of system performance given geometric parameters.

[0043] S2.1: Build a deep neural network model, the architecture of which is as follows: Figure 2As shown. This network has a nanoparticle radius r Volume fraction f v and fluid layer thickness h As the input layer, the system's photoelectric conversion efficiency is considered. or pv Photothermal conversion efficiency or th The output layer is defined by the comprehensive performance index MF. The network contains multiple fully connected hidden layers (preferably 8 layers in this embodiment), each containing multiple neurons (preferably 2000 neurons in this embodiment). A nonlinear transformation is introduced after each hidden layer using the ReLU activation function to capture complex input-output mapping relationships. The forward propagation process of the network can be described as follows:

[0044] (28) (29) In the formula, the superscript 1 indicates the first... l layer, w ir l As weight, b i l For bias, ϕ This is the ReLU activation function.

[0045] The loss function is defined as the mean squared error (MSE) between the network's predicted values ​​and the actual values ​​in the dataset: (30) In the formula, N For the sample size, y k For the true value, ŷ k These are predicted values.

[0046] It should be noted that the deep neural network used in this embodiment is a standard multilayer fully connected feedforward neural network. Its network structure (including the number of hidden layers, the number of neurons per layer, activation functions, etc.) was determined through experimental comparison and optimization. However, the network itself does not introduce new structural units or special connection methods, and belongs to a network architecture known in the art. The innovation of this invention lies in the synergy between the trained DNN model and a genetic algorithm to form a closed-loop "prediction-optimization" framework for performance optimization of nanofluidic PV / T systems. Specifically, the DNN model is used as a high-precision fitness function evaluator to replace traditional time-consuming physical simulations or experiments. Combined with the global search capability of the genetic algorithm, efficient optimization of high-dimensional nonlinear parameter spaces is achieved.

[0047] S2.2: Train the DNN model using the training set data. Continuously update the weights and biases in the network using the backpropagation algorithm and gradient descent optimizer to minimize the loss function. Monitor the model training process using the validation set data to prevent overfitting and adjust hyperparameters (such as the learning rate).

[0048] S2.3: Evaluate the prediction accuracy and generalization ability of the trained DNN model using test set data. Multiple metrics, including the coefficient of determination (R²), root mean square error (RMSE), and mean absolute percentage error (MAPE), are used to comprehensively evaluate model performance. A high-precision DNN positive prediction model is ultimately obtained for subsequent optimization.

[0049] S3. Establish a closed-loop "prediction-optimization" framework that combines deep neural networks (DNN) and genetic algorithms (GA) to achieve global optimization of system performance and reverse design of structural parameters.

[0050] S3.1: The DNN positive prediction model trained in S2 is used as the fitness function evaluator and combined with a genetic algorithm. The optimization objective is to maximize the overall performance index MF of the system.

[0051] S3.2: Initialize the genetic algorithm parameters, including population size, crossover probability, mutation probability, and maximum number of generations. Set the nanoparticle radius... r Volume fraction f v and fluid layer thickness h As a decision variable to be optimized, it is encoded within a given range.

[0052] S3.3: Run the genetic algorithm for iterative optimization. In each generation, the individuals in the population (i.e., a set of geometric parameters) are input into the DNN forward prediction model, and the corresponding MF value is calculated as the fitness of that individual. Based on the fitness, a new generation of the population is generated through selection, crossover, and mutation operations.

[0053] S3.4: Repeat S3.3 until the convergence condition is met (e.g., reaching the maximum number of generations or the fitness no longer increases). Finally, the optimal individual output by the genetic algorithm is the globally optimal combination of geometric parameters that maximizes the overall system performance (MF).

[0054] The deep neural network model constructed in this invention can accurately learn and map the geometric parameters of nanofluids and the performance of PV / T systems. or pv , or thThe model demonstrates a complex nonlinear relationship between three different nanofluid performance indices (e.g., MF). Examples show that the model achieves a prediction accuracy of over 99.48% for all three indices, far exceeding traditional machine learning models such as Random Forest (RF) and Decision Tree (DT). This invention combines DNN with Genetic Algorithm (GA) to form a closed-loop "prediction-optimization" framework. GA, as a powerful global search algorithm, can effectively address the complex nonlinear relationships between these indices and the high-dimensional, nonlinear, multi-peak performance spaces constructed by DNN (e.g., MF). Figure 3 The method (shown) efficiently explores and overcomes the drawback of traditional optimization methods that are prone to getting trapped in local optima, reliably converging to the global optimum. The framework of this invention is driven by high-throughput data generated from a physical model and does not depend on a specific material system. The successful optimization of three different nanofluids (Ag, Au, and Al) in the examples verifies that the method has good universality and transferability, providing a novel streamlined solution for the rapid screening and design of other novel nanofluids or solar energy devices.

[0055] Based on the above methods, the present invention proposes an embodiment, which specifically includes the following steps: S1. Construct a high-throughput dataset: First, high-throughput theoretical calculations were performed for each nanofluid based on the parameter ranges and step sizes listed in Table 1. Nanoparticle radius r Value range 2.5-50 nm, step size 2.5 nm; fluid layer thickness h Value range 2-20 mm, step size 2 mm; volume fraction f v In the low concentration region (5×10) -7 Up to 1×10 -5 ) Using 5×10 -7 Step size, in the high concentration area (1×10) -5 Up to 1×10 -4 ) Using 5×10 -6 Step size. 7400 data samples were generated for each nanofluid. All samples were randomly divided into a training set (5920 sets), a validation set (740 sets), and a test set (740 sets) in an 8:1:1 ratio.

[0056] To ensure the reliability of the model, the Monte Carlo radiative transfer model and the PV / T system coupling model in step 1 were first verified. For example... Figure 4 As shown, for Ag-water nanofluids (particle size 49.9 nm, mass fractions 31.8 ppm and 5.3 ppm, optical path 1 cm), the simulated transmittance of this invention is in high agreement with the experimental results of Han et al. Table 2 compares the performance of the PV / T system, and the results show that the maximum relative error between the simulated electrical efficiency and thermal efficiency and the experimental values ​​is less than 10%, proving the accuracy of the physical model established in this invention.

[0057] Table 2 Comparison of simulation results and experimental results of PV / T system model S2. Establish and train a DNN positive prediction model: Building such Figure 2 The deep neural network shown has 3 nodes in its input layer. r , f v , h The output layer has 3 nodes. or pv , or th The network contains 8 hidden layers, each with 2000 neurons, and the activation function is ReLU. The Adam optimizer was used, with an initial learning rate of 0.001 and a loss function defined by Equation (30) as mean squared error (MSE). The model was trained using the training set, with 3000 training epochs. During training, the loss functions of both the training and validation sets steadily decreased and eventually converged, indicating that the model was well-trained and did not exhibit overfitting.

[0058] To highlight the advantages of DNNs, this invention also constructs decision tree (DT) and random forest (RF) models for comparison. The prediction performance of the three models on the test set is compared as follows: Figure 5 As shown in the figure, it can be clearly seen that the predicted values ​​(scatter points) of the DNN model almost perfectly fit the diagonal (ideal prediction line), while the scatter points of the DT and RF models exhibit varying degrees of dispersion. Combined with the radar chart ( Figure 6 Quantitative analysis of Ag nanofluids, R0 of the DNN model. 2 With an accuracy as high as 0.9864, an RMSE as low as 0.0117, and a MAPE of 0.09182, DNN outperforms both RF and DT models across the board. For Au and Al nanofluids, DNN also demonstrates the highest prediction accuracy (both exceeding 99.48%). This fully demonstrates the superiority of DNN in capturing the nonlinear mapping relationships of such complex physical problems.

[0059] S3. Global system performance optimization based on the DNN-GA framework: The high-precision DNN model trained in step 2 was used as the fitness function and coupled with a genetic algorithm to globally optimize the geometric parameters of Ag, Au, and Al nanofluids. The genetic algorithm parameters were set as follows: population size 200, maximum number of generations 500, crossover probability 0.8, and mutation probability 0.1.

[0060] The optimization process is in Figure 3The algorithm operates within a highly complex, multi-peak performance space. Traditional parameter scanning or gradient descent methods are prone to getting trapped in local optima, while genetic algorithms, with their global search capabilities, can effectively explore the entire space. The globally optimal parameter combinations and their corresponding system performance are shown in Table 3.

[0061] As shown in Table 3, the optimal parameter combination for Ag nanofluids is: r =4.02nm, h =9.91mm, f v =9.45×10 -5 At this point, the overall system performance MF reaches its highest value of 1.3603. The optimal parameters for Au nanofluids are... r =2.63nm, h =10.04mm, f v =9.67×10 -5 This corresponds to MF=1.3190. The optimal parameters for Al nanofluids are... r =3.91nm, h =9.86mm, f v =9.72×10 -5 The corresponding MF = 1.2978.

[0062] Table 3. Combinations of structural parameters and performance indicators after optimization by genetic algorithm Figure 7 The spectral transmittance and absorptivity of three nanofluids under their respective optimal parameter combinations are presented. The Ag nanofluid maintains high transmittance in the main wavelength range of photovoltaic cell response (approximately 500-1100 nm) while exhibiting significant absorption in specific wavelength ranges, achieving ideal spectral dispersion. The Au and Al nanofluids also exhibit spectral characteristics consistent with their material properties. These results demonstrate that the proposed DNN-GA framework can accurately identify the "sweet spot" parameters that maximize the overall performance of the system based on the intrinsic properties of different materials, providing clear guidance for the rational design of high-performance nanofluidic PV / T systems.

[0063] In summary, the method provided by this invention can not only quickly and accurately predict the performance of nanofluidic PV / T systems, but also efficiently achieve global optimization of multiple parameters, solving the problems of low efficiency and easy getting trapped in local optima in traditional methods. It has important theoretical significance and broad application prospects in the field of high-efficiency solar energy utilization.

[0064] Based on the same inventive concept, this invention also proposes a performance optimization system for nanofluidic photovoltaic and thermal systems, comprising: The acquisition module is used to acquire a high-throughput dataset of nanofluid photovoltaic and thermal systems based on the photo-electric-thermal coupling model. The high-throughput dataset includes the geometric structural parameters of the nanofluid system and the corresponding performance evaluation indicators.

[0065] The model training module is used to train a deep neural network (DNN) positive prediction model by taking the geometric structural parameters of the nanofluid system as input and the performance evaluation index corresponding to the nanofluid photovoltaic and thermal system as output, so as to construct a nonlinear mapping relationship between the geometric structural parameters and the performance evaluation index.

[0066] The optimization module uses the nonlinear mapping relationship between geometric structure parameters and performance evaluation indicators as the fitness function of the genetic algorithm. Geometric structure parameters are the variables to be optimized, and maximizing the performance evaluation indicator is the optimization objective. Specifically, it includes: randomly generating an initial population within a given range of geometric structure parameters, where each individual is encoded as a combination of geometric structure parameters; inputting each individual into the trained DNN forward prediction model, obtaining the corresponding predicted performance evaluation indicator value based on the nonlinear mapping relationship, and using this predicted value as the individual's fitness; iteratively performing selection, crossover, mutation, and fitness evaluation operations based on the individual's fitness value until the convergence condition is met, outputting the optimal combination of geometric structure parameters that maximizes the performance evaluation indicator.

[0067] The optimization module is used to maximize the performance of nanofluid photovoltaic and thermal systems based on the optimal combination of geometric parameters.

[0068] The present invention also proposes a computer device for performance optimization of nanofluid photovoltaic and thermal systems, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the performance optimization method for nanofluid photovoltaic and thermal systems.

[0069] The present invention also proposes a readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, are used to perform steps of a performance optimization method for a nanofluid photovoltaic and thermal system.

[0070] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for performance optimization of a nanofluid photovoltaic and thermal system, characterized in that, Includes the following steps: A high-throughput dataset of nanofluid photovoltaic and thermal systems based on an optical-electrical-thermal coupling model was obtained; the high-throughput dataset includes the geometric structural parameters of the nanofluid system and the corresponding performance evaluation indicators. Using the geometric structural parameters of the nanofluid system as input and the performance evaluation index corresponding to the nanofluid photovoltaic and thermal system as output, a deep neural network (DNN) positive prediction model is trained to construct a nonlinear mapping relationship between the geometric structural parameters and the performance evaluation index. The genetic algorithm uses the nonlinear mapping relationship between geometric structure parameters and performance evaluation indicators as its fitness function, with geometric structure parameters as the variables to be optimized and maximizing the performance evaluation indicator as the optimization objective. Specifically, it involves: randomly generating an initial population within a given range of geometric structure parameters, where each individual is encoded as a combination of geometric structure parameters; inputting each individual into a trained DNN forward prediction model, obtaining the corresponding predicted performance evaluation indicator value based on the nonlinear mapping relationship, and using this predicted value as the individual's fitness; iteratively performing selection, crossover, mutation, and fitness evaluation operations based on the individual's fitness value until the convergence condition is met, outputting the optimal combination of geometric structure parameters that maximizes the performance evaluation indicator. The optimal combination of geometric parameters is used to maximize the performance of nanofluid photovoltaic and thermal systems.

2. The performance optimization method for a nanofluid photovoltaic and thermal system according to claim 1, characterized in that, The establishment of a high-throughput dataset for nanofluidic photovoltaic and thermal systems based on a photo-electric-thermal coupling model specifically includes the following steps: The physical model and key geometric parameters of the nanofluidic photovoltaic / thermal system were determined. The physical model, from top to bottom, consists of a nanofluidic layer, a glass cover, a silicon photovoltaic cell, and a bottom water-cooling channel. The key geometric parameters include the radius of the nanoparticles. r Volume fraction of nanoparticles f v and the thickness of the nanofluid layer h ; Based on the radius and complex refractive index of the nanoparticles and the complex refractive index of the base fluid, the optical properties of a single particle are calculated using the Lorenz-Mie theory. Combined with the volume fraction of the nanoparticles, the effective radiation characteristic parameters of the nanofluid system are calculated based on the independent scattering approximation. Based on the effective radiation characteristic parameters of the nanofluid system, the thickness of the nanofluid layer, and the refractive index of each layer of the nanofluid photovoltaic and thermal system, the improved Monte Carlo method MCML is used to solve the radiation transfer equation RTE by tracking the transmission of photon packets in the multilayer medium, and the spectral transmittance, spectral absorptivity, and spectral reflectance of the nanofluid photovoltaic / thermal system within a set wavelength range are calculated. The spectral transmittance, spectral absorptivity, and spectral reflectance are input into a coupled photo-electric-thermal model to calculate the performance indicators of the nanofluid photovoltaic and thermal system. The performance indicators include photoelectric conversion efficiency and photothermal conversion efficiency. A comprehensive performance evaluation index MF is introduced to balance the photoelectric and photothermal outputs of the nanofluid photovoltaic and thermal system. By iteratively changing the nanoparticle radius, volume fraction, and fluid layer thickness, high-throughput theoretical calculations are performed to construct a high-throughput dataset of nanofluid photovoltaic and thermal systems that includes geometric structural parameters and corresponding performance evaluation indicators.

3. The performance optimization method for a nanofluid photovoltaic and thermal system according to claim 2, characterized in that, The comprehensive performance index MF is expressed as follows: In the formula, η pv,unfiltered The efficiency of the bare cell without nanofluid filtration. η th For photothermal conversion efficiency, η pv For photoelectric conversion efficiency, w This is the weighting factor for electrical energy relative to thermal energy.

4. The performance optimization method for a nanofluid photovoltaic and thermal system according to claim 1, characterized in that, The training of the deep neural network (DNN) forward prediction model specifically involves updating the weights and biases in the model using the backpropagation algorithm and gradient descent optimizer to minimize the loss function; the loss function is the mean squared error (MSE) between the model's predicted values ​​and the actual values ​​in the dataset.

5. The performance optimization method for a nanofluid photovoltaic and thermal system according to claim 1, characterized in that, The iterative selection, crossover, mutation, and fitness evaluation operations based on the fitness value of an individual specifically include the following steps: Based on the individual's fitness value, roulette wheel selection or tournament selection methods are used to select individuals with high fitness from the current population as parents; The selected parent individuals are crossbred according to a preset crossover probability to generate offspring individuals; The gene loci in the offspring individuals are mutated according to a preset mutation probability to introduce new gene information; The offspring individuals generated through selection, crossover, and mutation operations are merged with the parent individuals to generate a new generation of population.

6. The performance optimization method for a nanofluid photovoltaic and thermal system according to claim 1, characterized in that, The convergence condition is to reach a preset maximum number of generations or for the fitness value of the best individual in a population to change less than a preset threshold over multiple generations.

7. The performance optimization method for a nanofluid photovoltaic and thermal system according to claim 1, characterized in that, The individual coding method uses real number coding, with the values ​​of geometric structural parameters as gene loci.

8. A performance optimization system for a nanofluid photovoltaic and thermal system, characterized in that, include: The acquisition module is used to acquire a high-throughput dataset of nanofluidic photovoltaic and thermal systems based on the optical-electrical-thermal coupling model; The high-throughput dataset includes the geometric parameters of the nanofluidic system and the corresponding performance evaluation metrics. The model training module is used to train a deep neural network (DNN) positive prediction model by taking the geometric structural parameters of the nanofluid system as input and the performance evaluation index corresponding to the nanofluid photovoltaic and thermal system as output, so as to construct a nonlinear mapping relationship between the geometric structural parameters and the performance evaluation index. The optimization module uses the nonlinear mapping relationship between geometric structure parameters and performance evaluation indicators as the fitness function of the genetic algorithm. Geometric structure parameters are used as the variables to be optimized, and maximizing the performance evaluation indicator is the optimization objective. Specifically, it includes: randomly generating an initial population within a given range of geometric structure parameters, where each individual is encoded as a combination of geometric structure parameters; inputting each individual into the trained DNN forward prediction model, obtaining the corresponding predicted performance evaluation indicator value based on the nonlinear mapping relationship, and using this predicted value as the individual's fitness; iteratively performing selection, crossover, mutation, and fitness evaluation operations based on the individual's fitness value until the convergence condition is met, outputting the optimal combination of geometric structure parameters that maximizes the performance evaluation indicator. The optimization module is used to maximize the performance of nanofluid photovoltaic and thermal systems based on the optimal combination of geometric parameters.

9. A computer device for performance optimization of nanofluid photovoltaic and thermal systems, characterized in that, include: A memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the performance optimization method for the nanofluid photovoltaic and thermal system according to any one of claims 1-7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program, which includes program instructions that, when executed by a processor, are used to perform the steps of the performance optimization method for the nanofluid photovoltaic and thermal system according to any one of claims 1-7.

Images