Simulation method for radiation flux density distribution of tower type photo-thermal power station receiver considering refraction effect

Abstract

The application discloses a simulation method for radiation flux density distribution of a tower type photo-thermal power station receiver considering refraction influence. Since a heliostat usually adopts a silver-coated glass reflector, the method considers the refraction effect of light on the glass layer, specifically comprising the following steps: obtaining heliostat field parameters, performing heliostat rotation focusing, and performing light incidence and shielding analysis; using a pre-trained full connection neural network to infer radiation flux density distribution parameters; after sampling the mirror field, a Monte Carlo ray tracing method considering refraction is used to simulate reflected light spots on the imaging plane; an elliptical Gaussian function with a cross term is used to fit the light spots and extract parameters, and a training set is constructed based on the sampling, simulation and fitting data to optimize the neural network; finally, the radiation flux density distribution of each heliostat on the imaging plane is calculated through the neural network inference parameters and the elliptical Gaussian function with the cross term, and is projected obliquely and parallelly to the receiver surface. The application combines high-precision simulation and neural network, and realizes efficient and accurate heliostat field light spot analysis.

Description

Technical Field

[0001] This invention relates to the field of solar thermal power generation technology, and in particular to a simulation method for the radiation flux density distribution of a receiver in a tower solar thermal power plant that takes into account the effects of refraction. Background Technology

[0002] Tower-type concentrated solar power (CSP) systems utilize a large-scale heliostat field to focus sunlight onto a receiver at the top of a central tower, converting solar energy into high-temperature heat energy to drive a steam turbine and generate electricity. These systems offer advantages such as high concentration ratio, high energy conversion efficiency, strong energy storage capacity, and low carbon emissions. In the research of tower-type CSP systems, the simulation of the radiation flux density distribution on the receiver surface is crucial, as it forms the basis for annual radiation energy estimation, heliostat focusing strategies, and heliostat field layout design.

[0003] Existing analytical simulation methods model the radiant flux density distribution of the heliostat's reflected light spot using probability density functions. For example, the HFLCAL model uses a circular Gaussian function, while the iHFLCAL and NEG models use an elliptic Gaussian function. The functions used in these models do not include cross terms, limiting their representational capabilities and making them unable to represent reflected light spots with principal axis deflection. Current analytical models generally suffer from inaccurate modeling functions, can only represent reflected light spots without principal axis deflection, and exhibit large simulation errors.

[0004] Existing analytical model simulation methods categorize analytical model parameter calculation into three types: fitting methods, direct calculation methods, and data-driven methods. Fitting methods calculate analytical model parameters by fitting measurement results or ray tracing simulation results. Substituting these parameters into the analytical model yields simulation results very close to the true values. However, since analytical model simulation is unnecessary when measurement or ray tracing results are available, this parameter calculation method is unsuitable for practical applications. Direct calculation methods substitute parameters from the heliostat field into empirical formulas to calculate analytical model parameters. However, the empirical formulas for the HFLCAL and iHFLCAL models do not include parameters of the heliostat glass layer, while the refractive effect of the glass layer on light is significant. Data-driven methods construct datasets and use pre-trained neural networks to calculate analytical model parameters. The NEG model, when constructing its dataset, samples parameters that do not include the heliostat glass layer, and the simulation uses refraction-ignoring tools (such as QMCRT) to generate reflected light spots. Currently, existing calculation methods generally suffer from incomplete consideration of influencing factors and neglect of the glass layer's refractive effect, resulting in large simulation errors. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a simulation method for the radiation flux density distribution of a tower-type solar thermal power plant receiver that takes into account the effects of refraction.

[0006] The objective of this invention is achieved through the following technical solution: a simulation method for the radiative flux density distribution of a tower-type solar thermal power plant receiver considering the effects of refraction, comprising: (1) Obtain parameters in the heliostat field, including the thickness and refractive index of the heliostat glass layer, the size of the heliostat, the normal perturbation of the micro-surface of the heliostat, the relative position of each heliostat and the receiver, the direction of sunlight at the current moment, the distribution of sunlight and the intensity of direct sunlight radiation. (2) Rotate each heliostat to focus on the designated position of the receiver, calculate the angle of incidence of sunlight on each heliostat surface, and the area of ​​each heliostat that is blocked by shadow, and calculate the area of ​​the effective reflection area of ​​the heliostat and the center of gravity shift. (3) Using a pre-trained fully connected neural network, the parameters of the radiation flux density distribution function are inferred; it includes an input layer, an output layer and three hidden layers. The input vector of the network includes the angle of incidence of sunlight on the heliostat, the distance from the heliostat to the receiver and the sunlight distribution parameters; the output vector of the network includes the total energy of the light spot reflected by the heliostat on the imaging plane, the standard deviation of the elliptic Gaussian function with cross terms and the correlation coefficient of the function variables. (4) Based on the area and centroid offset of the effective reflection region of the heliostat obtained in step (2), and combined with the total energy, standard deviation and correlation coefficient obtained in step (3), calculate the radiation flux density distribution on the imaging plane using an elliptic Gaussian function with cross terms, and project it obliquely onto the receiver surface.

[0007] Furthermore, the pre-trained fully connected neural network in step (3) is obtained through the following steps: (2.1) By performing multi-state sampling on the given mirror field to be simulated, and using a Monte Carlo ray tracing simulation tool that can simulate the refraction and total internal reflection of light in the glass layer of the heliostat, the reflected light spot of the heliostat on the imaging plane is calculated. (2.2) Taking into full account the refraction and total internal reflection effects generated by light propagating through the glass layer in the real world, and then using an elliptic Gaussian function with cross terms to fit each reflected light spot of the heliostat on the imaging plane to obtain the corresponding light spot model parameters; (2.3) Finally, the data obtained from sampling, simulation and fitting are used to construct a training dataset for supervised training of the fully connected neural network until the loss function value converges.

[0008] Further, in step (2.2), an elliptic Gaussian function with cross terms is used to fit each reflected light spot of the heliostat on the imaging plane to obtain the corresponding light spot model parameters; specifically, this is achieved through two steps: solving for the model parameters and optimizing the model parameters. Model parameter solution: The model parameters are solved by analyzing the reflected light spot and its marginal distribution. Model parameter optimization: Using the solution value as the initial value, the model parameters are optimized by defining an objective function and using an optimization algorithm; the resulting model parameters, combined with an elliptic Gaussian function with cross terms, can accurately fit the reflected light spot with the principal axis deflection shape.

[0009] Specifically, the expression for solving the model parameters is as follows: ; ; ; ; ; in This represents the total energy of the light spot reflected by the heliostat. and Let the standard deviation be the elliptic Gaussian function with cross terms. The correlation coefficient between the functional variables x and y; Here is the numerical matrix of the reflected light spot. The first of the numerical matrix Line number Column elements, and For the number of rows and columns of the numerical matrix, , ; and The length and width of a single pixel in the imaging plane when simulating using a Monte Carlo ray tracing simulation tool that takes refraction into account.

[0010] Specifically, to solve for the correlation coefficients of the model parameters, it is necessary to determine their signs. The sign determination criteria are as follows: ; in These represent the integrals of the numerical matrix of the reflected light spot in the first, second, third, and fourth quadrants, respectively; correlation coefficients with different signs correspond to different shapes of reflected light spots, and the shape of the reflected light spot determines... The symbol, therefore by The sign of the correlation coefficient can be determined by the following method: when Sometimes, ; when Sometimes, ; when Sometimes, At this point, there is no need to discuss the sign of the correlation coefficient.

[0011] Furthermore, the optimization of model parameters using an optimization algorithm by defining an objective function specifically involves using the Adam algorithm to optimize the objective function, the expression of which is as follows: ; in Here is the numerical matrix of the reflected light spot. For the first numerical matrix Line number Column elements, It is the maximum value in this numerical matrix; This is the numerical matrix of the elliptical Gaussian spot calculated based on the model parameters and the radiation flux density distribution function. This numerical matrix is ​​compared with... They have the same resolution and shape; For the first numerical matrix Line number Column elements, The maximum value in this numerical matrix. and For the number of rows and columns of the numerical matrix, , .

[0012] Further, in step (2.3), the fully connected neural network is trained under supervision until the loss function value converges; specifically, the fully connected neural network is pre-trained based on supervised learning, and the training of the fully connected neural network is completed by minimizing the loss function of the prediction error and using the gradient descent optimization algorithm and its hyperparameter configuration, thereby obtaining the pre-trained fully connected neural network.

[0013] Specifically, the imaging plane in step (3) is a virtual plane that passes through the focal point of the heliostat and is perpendicular to the direction of the reflected light from the center of the heliostat. The horizontal axis of the imaging plane is the direction in which the long side of the heliostat is projected onto the plane, and the vertical axis is perpendicular to the horizontal axis.

[0014] Furthermore, in step (4), the radiation flux density distribution on the imaging plane is calculated using an elliptic Gaussian function with cross terms, and the radiation flux density distribution of the heliostat reflected light spot on the imaging plane is... The definition is as follows: ; in This represents the total energy of the light spot reflected by the heliostat. and This represents the peak shift of the heliostat's reflected light spot under shading conditions, which is the centroid shift of the heliostat's effective reflective area. and Let the standard deviation be the elliptic Gaussian function with cross terms. Let x be the correlation coefficient between the functional variables x and y.

[0015] The beneficial effects of this invention are as follows: This invention considers the refraction and total internal reflection effects generated during the propagation of light through the glass layer of the heliostat, making the simulation results closer to the real situation. It uses an elliptic Gaussian function with cross terms to calculate the radiation flux density distribution. The cross terms can represent the reflected light spot with the principal axis deflection shape, making the simulation results more accurate and realizing efficient and precise heliostat field spot analysis. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] Figure 1 This is a flowchart illustrating the simulation method for the radiation flux density distribution of a receiver in a tower-type solar thermal power plant, taking into account the effects of refraction, which is provided for the present invention. Figure 2 This is a schematic diagram of the heliostat layout in the heliostat field to be simulated in this embodiment; Figure 3 This is a schematic diagram showing the curves of the light interception rate of all heliostats in the simulated heliostat field as a function of the side length of the imaging plane in this embodiment. Figure 4 The following is a schematic diagram of the radiation flux density distribution and its marginal distribution on the imaging plane obtained by using a simulation tool that considers refraction (such as RMCRT) in this embodiment; wherein, (a) is the marginal density distribution of the x-axis on the imaging plane, (b) is the marginal density distribution of the y-axis on the imaging plane, and (c) is the radiation flux density distribution on the imaging plane. Figure 5 This is a schematic diagram showing the division of the four quadrants of the radiation flux density distribution on the imaging plane in this embodiment; Figure 6 This is a schematic diagram showing the relationship between the morphology of the radiation flux density distribution on the imaging plane and the correlation coefficient in this embodiment; wherein, (a) is... The distribution of radiation flux density at that time is shown in Figure (b). The distribution of radiation flux density at that time is shown in Figure (c). The distribution pattern of radiation flux density at that time; Figure 7 This is a comparative schematic diagram showing the radiation flux density distribution of a heliostat simulated using different analytical models in this embodiment. Figure 8 This is a comparative schematic diagram showing the radiation flux density distribution of the heliostat field simulated using different analytical models in this embodiment. Detailed Implementation

[0018] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0019] The present invention will now be described in detail with reference to the accompanying drawings.

[0020] This invention provides a simulation method for the radiant flux density distribution of a receiver in a tower-type solar thermal power plant, taking into account the effects of refraction. Figure 1 As shown. For a given mirror field to be simulated and a given focusing strategy, the specific simulation method steps are as follows: S1: Obtain parameters in the heliostat field, such as the thickness and refractive index of the heliostat glass layer, the size of the heliostat, the normal perturbation of the micro-surface of the heliostat, the relative position of each heliostat and the receiver, the direction of sunlight at the current moment, the distribution of sunlight, and the intensity of direct solar radiation. S2: Following a given focusing strategy, each heliostat rotates and focuses on a designated position on the receiver. The angle of incidence of sunlight on each heliostat surface and the extent of shading on each heliostat are calculated. The heliostat surface is uniformly discretized into several sub-heliostats. The BMSS algorithm combined with the 3DDDA algorithm is used to calculate the shading of the heliostats, and the area of ​​the effective reflection region and the centroid shift of the heliostats are calculated.

[0021] S3: Using a pre-trained fully connected neural network, the parameters of the radiative flux density distribution function of each heliostat reflected light spot on the imaging plane are inferred from the data obtained in S1 and S2. The neural network consists of an input layer, an output layer, and three hidden layers. The network's input vector includes the angle of incidence of sunlight on the heliostat surface, the distance from the heliostat to the receiver, and the sunlight distribution parameters. The network's output vector includes the total energy of the heliostat reflected light spot on the imaging plane, the standard deviation of the elliptic Gaussian function with cross terms, and the correlation coefficient between the function variables x and y.

[0022] S4: Based on the area and centroid offset of the effective reflection region of the heliostat obtained in step S2, and the parameters of the radiation flux density distribution function obtained in step S3, the radiation flux density distribution of the heliostat reflected light spot on the imaging plane is calculated using an elliptic Gaussian function with cross terms. The radiation flux density distribution on the imaging plane is projected onto the receiver surface using the oblique parallel projection method, obtaining the radiation flux density distribution of a single-sided heliostat on the receiver. The radiation flux density distributions reflected from all heliostats to the receiver surface are summed to obtain the final simulation result.

[0023] Example 1: This invention provides a simulation method for the radiant flux density distribution of a receiver in a tower-type solar thermal power plant, taking into account the effects of refraction. Figure 1 As shown. For a given field of view to be simulated, the focusing strategy is for each heliostat to aim at the point on the equator closest to itself from the receiver. The specific simulation method steps are as follows: S1: Obtain parameters in the heliostat field, such as the thickness and refractive index of the heliostat glass layer, heliostat dimensions, normal perturbation of the heliostat microsurface, relative position of each heliostat and receiver, current solar radiation direction, solar radiation distribution, and direct solar radiation intensity. The data is shown in Table 1 below, and the heliostat layout is as follows. Figure 2 As shown: Table 1 ; S2: Following a given focusing strategy, each heliostat rotates and focuses on a designated position on the receiver. The angle of incidence of sunlight on each heliostat surface and the extent of shading on each heliostat are calculated. The heliostat surface is uniformly discretized into several sub-heliostats. The BMSS algorithm combined with the 3DDDA algorithm is used to calculate the shading of the heliostats, and the area of ​​the effective reflection region and the centroid shift of the heliostats are calculated.

[0024] S3: Using a pre-trained fully connected neural network, infer the parameters of the radiative flux density distribution function of each heliostat reflected light spot on the imaging plane from the data obtained in steps S1 and S2. The fully connected neural network consists of an input layer, an output layer, and three hidden layers. The network's input vector includes the angle of incidence of sunlight on the heliostat surface, the distance from the heliostat to the receiver, and the sunlight distribution parameters. The network's output vector includes the total energy of the heliostat reflected light spot on the imaging plane, the standard deviation of the elliptic Gaussian function with cross terms, and the correlation coefficients of the function variables x and y. Each hidden layer consists of a fully connected layer and a ReLU activation function, with the number of neurons in the three hidden layers being 512, 256, and 128, respectively.

[0025] The pre-trained fully connected neural network is obtained through the following steps: A Monte Carlo ray tracing simulation tool (such as RMCRT) considering refraction is used to simulate a given mirror field. This simulation tool is more accurate, comprehensively considering the refraction and total internal reflection effects of light propagating through the glass layer in the real world. The reflected light spot of the heliostat on the imaging plane is obtained, and the corresponding model parameters are obtained by fitting this reflected light spot. A dataset is constructed using the data obtained from sampling, simulation, and fitting, and used for supervised training of the fully connected neural network until the loss function converges. Specifically, the fully connected neural network is pre-trained based on supervised learning, and the training is completed by minimizing the loss function of the prediction error and using the gradient descent optimization algorithm and its hyperparameter configuration, thus obtaining the pre-trained fully connected neural network.

[0026] For the sampling of a given mirror field to be simulated, the following scene parameters are mainly considered: the angle of incidence of sunlight on the heliostat, the distance from the heliostat to the receiver, and the sunlight distribution parameters. The sampling range and step size of the scene parameters are summarized in Table 2 below. The dataset size used for subsequent training of the neural network is 1885 = 13 × 145 × 1.

[0027] Table 2 ; The angle of incidence of sunlight on the mirror is described by two angles, ω1 and ω2. ω1 is the angle between the incident light and the normal to the heliostat, and ω2 is the angle between the long side of the heliostat and the projection of the incident light onto the mirror surface. These two angles can equivalently represent the solar altitude angle, azimuth angle, and the orientation of the heliostat, and they also have a stronger correlation with the shape of the reflected light spot. The sampling of the angle of incidence of sunlight on the mirror is achieved by adjusting the sun's azimuth. The solar altitude angle is sampled uniformly in steps of 10° within the range [10°, 90°], and the solar azimuth angle is sampled uniformly in steps of 20° within the range [0°, 360°]. When the solar altitude angle is 90°, sampling of the solar azimuth angle is unnecessary.

[0028] The distance from the heliostat to the receiver is the distance between the center of the heliostat and the aiming point on the receiver. As the distance increases, the shape of the reflected light spot gradually diffuses. Distance sampling can be achieved by fixing the aiming point and adjusting the position of the heliostat within the field. The distance range in the simulated mirror field is [150m, 750m], and sampling is performed within this range in 50m increments. This range is large enough to cover all heliostat scenarios within the simulated mirror field.

[0029] The distribution of sunlight is described by the Buie solar shape model, where the parameter CSR (CircumSolar Ratio) describes the distribution of sunlight direction. The commonly used CSR range is 0 to 0.4; a larger value indicates a larger proportion of solar halo. Random, uniform sampling is performed on the CSR within the range [0, 0.4].

[0030] After sampling the scene parameters, these parameters can be used for RMCRT simulation to obtain the reflected light spot on the imaging plane. The RMCRT simulation tool fully considers the refraction and total internal reflection effects caused by light propagating through the glass layers, achieving Monte Carlo ray tracing along the entire optical path. When performing RMCRT simulation, the size of the imaging plane needs to be set. The imaging plane is an infinitely large ideal plane, but RMCRT cannot use an infinitely large imaging plane to store all the radiative flux density distribution data during simulation; therefore, a finite and suitable imaging plane size needs to be selected.

[0031] The RMCRT simulation tool was used to calculate the light interception rate of all heliostats on the imaging plane in the simulated heliostat field. For example... Figure 3 The curves showing the change in light interception rate with the side length of the imaging plane are presented. When the side length of the imaging plane is 20 meters, the light interception rate is 99.2%. The increase in light interception rate tends to level off after the side length of the imaging plane exceeds 20 meters. An imaging plane size of 20 meters is used as a reference. Simultaneously, the total energy counted on an imaging plane of this size is uniformly divided by 99.2% to compensate for the total energy loss caused by the finite size of the imaging plane.

[0032] The model parameters are obtained by fitting the reflected light spot. These parameters can be solved using the heliostat reflected light spot and its marginal distribution on the imaging plane. Figure 4 The diagram shows the radiant flux density distribution of the heliostat reflected light spot on the imaging plane, and the image is a heatmap. This distribution follows an elliptic Gaussian distribution with cross terms, and the marginal distribution is also Gaussian. Reddish areas indicate higher radiant flux density values, while bluish areas indicate lower radiant flux density values. The formulas for solving the model parameters are as follows: ; ; ; ; ; in This represents the total energy of the light spot reflected by the heliostat. and Let the standard deviation be the elliptic Gaussian function with cross terms. Let x be the correlation coefficient between the functional variables x and y. Here is the numerical matrix of the reflected light spot. The first of the numerical matrix Line number Column elements, and For the number of rows and columns of the numerical matrix, , ; and The length and width of a single pixel in the imaging plane when simulating using a Monte Carlo ray tracing simulation tool that takes refraction into account.

[0033] To determine the correlation coefficient of the model parameters, it is necessary to determine its sign. The sign determination criteria are as follows: ; in These are the integrals of the numerical matrix of the reflected light spot in the first quadrant, second quadrant, third quadrant, and fourth quadrant, respectively. The area corresponding to the reflected light spot is as follows Figure 5 As shown.

[0034] The correlation coefficient has a geometric meaning; different signs of the correlation coefficient correspond to different shapes of reflected light spots, such as... Figure 6 As shown. The shape of the reflected light spot determines... The symbol, therefore by The sign of the correlation coefficient can be determined by the following method: when Sometimes, ; when Sometimes, ; when Sometimes, At this point, there is no need to discuss the sign of the correlation coefficient.

[0035] After solving for the model parameters, using these initial values, the reflected light spot is fitted more accurately using parameter optimization methods to obtain the corresponding model parameters. The Adam algorithm is used to optimize the objective function, which is as follows: ; in Here is the numerical matrix of the reflected light spot. For the first numerical matrix Line number Column elements, This is the maximum value in the numerical matrix. This is the numerical matrix of the elliptical Gaussian spot calculated based on the model parameters and the radiation flux density distribution function. This numerical matrix is ​​compared with... They have the same resolution and shape. For the first numerical matrix Line number Column elements, The maximum value in this numerical matrix. and For the number of rows and columns of the numerical matrix, , .

[0036] A dataset is constructed using data obtained from sampling, simulation, and fitting, and a neural network is trained. Specifically, for a given mirror field to be simulated, the following scene parameters are mainly considered: the angle of incidence of sunlight on the heliostat surface, the distance from the heliostat to the receiver, and the sunlight distribution parameters. The heliostat-reflected light spot on the imaging plane is obtained through simulation of a set of scene parameters. The corresponding model parameters are obtained by fitting this reflected light spot. The scene parameters and model parameters are combined to construct a dataset for training the neural network. The fully connected neural network is pre-trained using supervised learning. The training is completed by minimizing the loss function of the prediction error and utilizing the gradient descent optimization algorithm and its hyperparameter configuration, thus obtaining a pre-trained fully connected neural network. The neural network uses mean squared error as the loss function, is trained using the Adam optimizer, with an initial learning rate of 0.01, a batch size of 1024, and 1000 training iterations.

[0037] S4: Based on the area and centroid shift of the heliostat's shadowed region obtained in S2, and the parameters of the radiation flux density distribution function obtained in S3, the radiation flux density distribution of the heliostat's reflected light spot on the imaging plane is calculated using an elliptic Gaussian function with cross terms. The calculation formula is as follows: ; in This represents the total energy of the reflected light spot from the heliostat. and This indicates the peak shift of the reflected light spot under shadow occlusion. and Let the standard deviation be the elliptic Gaussian function with cross terms. Let x be the correlation coefficient between the functional variables x and y.

[0038] Figure 7 This diagram illustrates the comparison of different analytical models in this embodiment. The comparison target is the radiation flux density distribution of the example heliostat in the simulated mirror field on the imaging plane. The dashed line represents the RMCRT simulation result, and the solid line represents the simulation result of each analytical model. Figure 7 It is known that HFLCAL, iHFLCAL, and NEG cannot represent the reflected light spot due to principal axis deflection, and although UNIZAR and Huang can represent the characteristics of principal axis deflection, they are still inaccurate. The method presented in this paper can accurately represent the reflected light spot due to principal axis deflection.

[0039] By using the oblique parallel projection method, the radiation flux density distribution on the imaging plane is projected onto the receiver surface, thus obtaining the radiation flux density distribution of a single heliostat on the receiver. The final simulation result is obtained by summing the radiation flux density distributions reflected from all heliostats to the receiver surface.

[0040] Figure 8This diagram illustrates the comparison of different analytical models in this embodiment. The comparison target is the radiative flux density distribution on the receiver surface in the simulated mirror field, i.e., the final simulation result. The dashed line represents the RMCRT simulation result, and the solid line represents the simulation result of each analytical model. Figure 8 It is evident that the simulation results from HFLCAL, iHFLCAL, NEG, UNIZAR, and Huang are still inaccurate. The simulation results from the method presented in this paper are more accurate.

[0041] Simulation accuracy can be characterized by total energy error, peak error, and mean absolute percentage error. The smaller the error, the higher the simulation accuracy. The formulas for calculating these three parameters are as follows.

[0042]

[0043]

[0044]

[0045] In the above formula, , and These represent the radiation flux density, peak value, and total energy obtained through RMCRT simulation, respectively. , and These represent the radiative flux density, peak value, and total energy obtained through analytical model simulation, respectively. 𝑚 and 𝑛 represent the number of rows and columns of pixels on the receiver. This indicates the receiver obtained through simulation. Line number The radiative flux density value at the column pixel.

[0046] Use the error calculation formula to calculate Figure 8 The errors are shown in Table 3 below.

[0047] Table 3 ; As shown in Table 3 above, compared with other methods, the present invention can achieve the minimum values ​​in total energy error, peak error, and mean absolute percentage error, and the simulation accuracy of the present invention is the highest.

[0048] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0049] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A simulation method for the radiative flux density distribution of a receiver in a tower-type solar thermal power plant, considering the effects of refraction, characterized in that, include: (1) Obtain parameters in the heliostat field, including the thickness and refractive index of the heliostat glass layer, the size of the heliostat, the normal perturbation of the micro-surface of the heliostat, the relative position of each heliostat and the receiver, the direction of sunlight at the current moment, the distribution of sunlight and the intensity of direct sunlight radiation. (2) Rotate each heliostat to focus on the designated position of the receiver, calculate the angle of incidence of sunlight on each heliostat surface, and the area of ​​each heliostat that is blocked by shadow, and calculate the area of ​​the effective reflection area of ​​the heliostat and the center of gravity shift. (3) Using a pre-trained fully connected neural network, the parameters of the radiation flux density distribution function are inferred; it includes an input layer, an output layer and three hidden layers. The input vector of the network includes the angle of incidence of sunlight on the heliostat, the distance from the heliostat to the receiver and the sunlight distribution parameters; the output vector of the network includes the total energy of the light spot reflected by the heliostat on the imaging plane, the standard deviation of the elliptic Gaussian function with cross terms and the correlation coefficient of the function variables. (4) Based on the area and centroid offset of the effective reflection region of the heliostat obtained in step (2), and combined with the total energy, standard deviation and correlation coefficient obtained in step (3), calculate the radiation flux density distribution on the imaging plane using an elliptic Gaussian function with cross terms, and project it obliquely onto the receiver surface.

2. The simulation method for radiant flux density distribution of a tower-type solar thermal power plant receiver considering the effect of refraction, as described in claim 1, is characterized in that... The pre-trained fully connected neural network in step (3) is obtained through the following steps: (2.1) By performing multi-state sampling on the given mirror field to be simulated, and using a Monte Carlo ray tracing simulation tool that can simulate the refraction and total internal reflection of light in the glass layer of the heliostat, the reflected light spot of the heliostat on the imaging plane is calculated. (2.2) Taking into full account the refraction and total internal reflection effects generated by light propagating through the glass layer in the real world, and then using an elliptic Gaussian function with cross terms to fit each reflected light spot of the heliostat on the imaging plane to obtain the corresponding light spot model parameters; (2.3) Finally, the data obtained from sampling, simulation and fitting are used to construct a training dataset for supervised training of the fully connected neural network until the loss function value converges.

3. The simulation method for radiant flux density distribution of a tower-type solar thermal power plant receiver considering the effect of refraction, as described in claim 2, is characterized in that... In step (2.2), an elliptic Gaussian function with cross terms is used to fit each reflected spot of the heliostat on the imaging plane to obtain the corresponding spot model parameters; specifically, this is achieved through two steps: solving for the model parameters and optimizing the model parameters. Model parameter solution: The model parameters are solved by analyzing the reflected light spot and its marginal distribution. Model parameter optimization: Using the solution value as the initial value, the model parameters are optimized by defining an objective function and using an optimization algorithm; the resulting model parameters, combined with an elliptic Gaussian function with cross terms, can accurately fit the reflected light spot with the principal axis deflection shape.

4. The simulation method for radiant flux density distribution of a tower-type solar thermal power plant receiver considering the effect of refraction, as described in claim 3, is characterized in that... The expression for solving the model parameters is as follows: ； ； ； ； ； in This represents the total energy of the light spot reflected by the heliostat. and Let the standard deviation be the elliptic Gaussian function with cross terms. The correlation coefficient between the functional variables x and y; Here is the numerical matrix of the reflected light spot. The first of the numerical matrix Line number Column elements, and For the number of rows and columns of the numerical matrix, , ; and The length and width of a single pixel in the imaging plane when simulating using a Monte Carlo ray tracing simulation tool that takes refraction into account.

5. The simulation method for radiant flux density distribution of a tower-type solar thermal power plant receiver considering the effect of refraction, as described in claim 4, is characterized in that... To determine the correlation coefficient of the model parameters, it is necessary to determine its sign. The sign determination criteria are as follows: ； in These represent the integrals of the numerical matrix of the reflected light spot in the first, second, third, and fourth quadrants, respectively; correlation coefficients with different signs correspond to different shapes of reflected light spots, and the shape of the reflected light spot determines... The symbol, therefore by The sign of the correlation coefficient can be determined by the following method: when Sometimes, ; when Sometimes, ; when Sometimes, At this point, there is no need to discuss the sign of the correlation coefficient.

6. The simulation method for radiant flux density distribution of a tower-type solar thermal power plant receiver considering the effect of refraction, as described in claim 3, is characterized in that... The process involves optimizing the model parameters using an optimization algorithm by defining an objective function; specifically, the Adam algorithm is used to optimize the objective function, the expression of which is as follows: ； in Here is the numerical matrix of the reflected light spot. For the first numerical matrix Line number Column elements, It is the maximum value in this numerical matrix; This is the numerical matrix of the elliptical Gaussian spot calculated based on the model parameters and the radiation flux density distribution function. This numerical matrix is ​​compared with... They have the same resolution and shape; For the first numerical matrix Line number Column elements, The maximum value in this numerical matrix. and For the number of rows and columns of the numerical matrix, , .

7. The simulation method for radiant flux density distribution of a tower-type solar thermal power plant receiver considering the effect of refraction, as described in claim 2, is characterized in that... In step (2.3), the fully connected neural network is trained under supervision until the loss function value converges. Specifically, the fully connected neural network is pre-trained based on supervised learning, and the training of the fully connected neural network is completed by minimizing the loss function of the prediction error and using the gradient descent optimization algorithm and its hyperparameter configuration, thereby obtaining the pre-trained fully connected neural network.

8. The simulation method for radiant flux density distribution of a tower-type solar thermal power plant receiver considering the effect of refraction, as described in claim 1, is characterized in that... The imaging plane in step (3) is a virtual plane that passes through the focal point of the heliostat and is perpendicular to the direction of the reflected light from the center of the heliostat. The horizontal axis of the imaging plane is the direction in which the long side of the heliostat is projected onto the plane, and the vertical axis is perpendicular to the horizontal axis.

9. The simulation method for radiant flux density distribution of a tower-type solar thermal power plant receiver considering the effect of refraction, as described in claim 1, is characterized in that... In step (4), the radiation flux density distribution on the imaging plane is calculated using an elliptic Gaussian function with cross terms, and the radiation flux density distribution of the heliostat reflected light spot on the imaging plane is also calculated. The definition is as follows: ； in This represents the total energy of the light spot reflected by the heliostat. and This represents the peak shift of the heliostat's reflected light spot under shading conditions, which is the centroid shift of the heliostat's effective reflective area. and Let the standard deviation be the elliptic Gaussian function with cross terms. Let x be the correlation coefficient between the functional variables x and y.

Images