Real-time simulation method and system for radiation energy density distribution of tower solar power generation system based on convolution model

By adopting a real-time simulation method based on convolutional model in tower solar power generation systems, combining GPU parallel processing and bounding box removal algorithm, the calculation efficiency of radiation energy density distribution is optimized, and the high computational complexity and error in the existing technology are solved The major problem has achieved efficient simulation of radiation energy density distribution.

CN118627401BActive Publication Date: 2024-11-19ZHEJIANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411103993.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-11-19
Estimated Expiration
2044-08-13

AI Technical Summary

Technical Problem

In the existing tower solar power generation system, the convolutional model has complex numerical integral calculations when calculating the radiation energy density distribution, which makes it large and difficult to solve efficiently, especially when the helioscopic is close to the receiver or is partially shadowed. When blocking, the simulation results have a large error.

Method used

Using a real-time simulation method based on the convolution model, the shadow occlusion distribution of the heliostat mirror is calculated, the analytical model is established under the local coordinate system, and the radiation energy density of the light spot on the imaging plane is calculated using the fast numerical integration method, and the radiation energy density of the light spot on the imaging plane is obliquely parallel. Projected to the receiver plane, combining GPU parallel processing and bounding box removal algorithm to optimize computing efficiency.

Benefits of technology

It realizes the speed of radiation energy density distribution simulation without reducing the calculation accuracy, which is suitable for real-time simulation and optimization problems of mirror field.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118627401B_ABST
    Figure CN118627401B_ABST
Patent Text Reader

Abstract

The present invention relates to a real-time simulation method and system for the radiation energy density distribution of a tower solar power generation system based on a convolution model, and belongs to the technical field of tower solar mirror field simulation. The method comprises: 1) calculating the shadow shielding distribution of the heliostat according to the mirror field layout and the light direction; 2) using a neural network or a parameter calculation method in an existing model to determine the radiation energy density distribution function parameters, and establishing an analytical model in a local coordinate system; 3) using a fast numerical integration method to efficiently calculate the radiation energy density of a light spot in a local coordinate system; 4) projecting the analytical model in the local coordinate system obliquely parallel to the light reflection direction to a receiver plane, and establishing a receiver plane radiation energy density distribution model, which can efficiently simulate the radiation energy density of the light spot and ensure the accuracy of the result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of tower solar mirror field simulation, and in particular to a real-time simulation method and system for simulating the radiation energy density distribution of a tower solar power generation system based on a convolution model, which is used to simulate the radiation energy density distribution of a plane light spot of a tower solar mirror field receiver. Background Art

[0002] Tower solar thermal power generation technology has attracted widespread attention from the industry due to its low cost and stable power generation. In this system, radiation energy is concentrated on the receiver surface and further stored and converted. Therefore, it is crucial to accurately and efficiently model and simulate the radiation energy density distribution on the receiver surface.

[0003] The research on the distribution of radiant energy density in tower solar thermal power generation systems is mainly divided into ray tracing and analytical models. Ray tracing originates from the realistic graphics rendering algorithm in computer graphics. It generates a distribution diagram of the radiant energy density on the surface of the receiver by simulating the propagation process of light emitted from the sun, reflected by the heliostat and reaching the surface of the receiver. Unlike the numerical vector used to represent color in realistic graphics rendering, in tower solar energy systems, the final result of the ray tracing method is the radiant energy density at each point on the surface of the receiver. The advantage of the ray tracing method is that it is applicable to various types of heliostats and receivers, and can perform accurate simulation calculations under different optical conditions. In addition, the high accuracy of the ray tracing method makes it often used as the true value for fitting and comparison of analytical models. However, although increasing the number of ray sampling can improve the simulation accuracy, it will also significantly increase the consumption of computing resources. In 2019, Duan Xiaoyue et al. proposed a GPU-based Quasi-Monte Carlo Ray Tracing algorithm (Quasi-Monte Carlo Ray Tracing, referred to as QMCRT) (DUAN X, HE C, LIN X, et al. Quasi-Monte Carlo ray tracing algorithm for radiative flux distribution simulation [J]. Solar Energy, 2020, 211: 167-182.), which uses GPU to process a large number of rays in parallel and achieves efficient simulation of tracing 200 million rays in 1.5 seconds. In 2022, Lin Xiaoxia et al. improved the QMCRT algorithm and proposed the RMCRT algorithm (LIN X, HE C, HUANG W, et al. GPU-based Monte Carlo ray tracing simulation considering refraction for central receiver system [J]. Renewable Energy, 2022, 193: 367-382.) that considers glass refraction and internal reflection. While maintaining good simulation accuracy, the simulation speed is increased to 16 times that of QMCRT. Despite this, the computational efficiency of ray tracing is still not as good as that of analytical models.

[0004] The imaging plane is a virtual plane that passes through the focal point of the heliostat and is perpendicular to the main direction of the light reflected by the heliostat. The analytical model can be defined on this imaging plane or on the plane where the heliostat mirror is located, and the numerical model is used to approximate the radiant energy density distribution of the reflected light spot. Modeling the radiant energy density distribution on the receiver surface includes three key steps: calculating the shadow shielding between the heliostats, defining the analytical model, and projecting the radiant energy density distribution model on the defined plane onto the receiver surface (SÁNCHEZ-GONZÁLEZ A, SANTANA D. Solar flux distribution on central receivers: A projection method from analytic function[J]. RenewableEnergy, 2015, 74: 576-587.). The analytical model can be divided into a convolution model and a simplified model based on Gaussian distribution according to its assumptions on the radiant energy density distribution of the heliostat reflected light spot. These two models have their own advantages and disadvantages in different application scenarios. The convolution model is usually more accurate when dealing with complex light spot shapes, while the simplified model has higher computational efficiency.

[0005] The convolution model regards the radiation energy density distribution generated by a single-sided heliostat as the convolution result of the light spot reflected by the micro-element of the heliostat. When the heliostat is partially blocked by a shadow, the area not blocked by the shadow is called the effective reflection area. The convolution model can be defined on the imaging plane or the plane where the heliostat mirror is located, using the energy density distribution generated by the light reflected by the micro-element of the heliostat as the convolution kernel, and the convolution area is the projection of the effective reflection area on the defined plane. When implementing the convolution model, complex integral calculations are involved, and numerical integration methods are usually required to solve it (HUANG W, YU L. Development of a new flux density function for a focusing heliostat[J]. Energy, 2018, 151: 358-375.). When the heliostat is close to the receiver and partially blocked by a shadow, the radiation energy density distribution obtained by the convolution model simulation is closer to the actual light spot distribution in shape. This method can more accurately reflect the real situation and improve the accuracy of the simulation results.

[0006] The existing analytical model fast mirror field simulation algorithm is only applicable to simplified models, with large simulation errors and large computational complexity: He Caitou (HE C, ZHAO Y, FENG J. An improved flux density distribution modelfor a flat heliostat (iHFLCAL) compared with HFLCAL[J]. Energy, 2019, 189:116239.) et al. proposed a fast simulation method based on GPU drawing pipeline, taking the flat receiver panel as the drawing target, and mapping the analytical model on the heliostat plane to the flat receiver through affine transformation. The calculation speed of this method is nearly two orders of magnitude faster than the ray tracing method, but this method uses a simplified model and cannot accurately draw the light spot of the heliostat close to the receiver. In addition, this method adopts a direct clipping method for the shadow-blocked area on the heliostat, and does not accurately draw the impact of the shadow-blocking on the light spot, resulting in a large calculation error.

[0007] In view of the radiant energy spot produced by the focusing polygonal heliostat, Collado proposed the UNIZAR model (COLLADO FJ, GOMEZ AP, TURÉGANO J A. Ananalytic function for the flux density due to sunlight reflected from aheliostat[J]. Solar Energy, 1986, 37(3): 215-234.) which uses the circular Gaussian distribution density function as the convolution kernel. The model defines the convolution kernel function on the imaging plane, and the convolution domain is the projection of the effective reflection area on the imaging plane. The direct calculation formula of the convolution kernel parameters (the standard deviation of the circular Gaussian distribution) of the UNIZAR model involves the solar shape error, the normal perturbation of the micro-surface of the heliostat, and the tracking error. The tracking error describes the deviation of the actual focus point of the heliostat from the ideal focus point. Alberto et al. (Sánchez-González, A. Analytic function for heliostat flux mapping with astigmatism and defocus[J]. Solar Energy, 2022, 241: 24-38.) further considered the effects of off-axis incidence and scattering errors of sunlight on the radiant energy spot in the UNIZAR model, especially when the focal length of the heliostat is small or the solar incidence angle is large, the simulation results are more accurate.

[0008] For rectangular and circular focusing heliostats, Huang Weidong et al. proposed a method of using an elliptical Gaussian distribution function as a convolution kernel on the imaging plane (HUANG W, YU L. Development of a new flux density function for a focusing heliostat[J]. Energy, 2018, 151: 358-375.). The scene parameters involved in the calculation formula of the two standard deviation parameters of the convolution kernel include the solar shape error, the normal perturbation of the microsurface of the heliostat, and the tracking error. Unlike the UNIZAR model, which directly sets the solar shape error as a constant, Huang Weidong et al. calculated the solar shape error by using the direct normal irradiance (DNI). When using Gaussian distribution as the convolution kernel, it is necessary to use the numerical integration method to solve the convolution model calculation formula, which makes the calculation process relatively complicated.

[0009] In order to solve the above problems, He Caitou et al. proposed a Cauchy model for rectangular flat heliostats, which uses quasi-Cauchy function as the convolution kernel (HE C, DUAN X, ZHAO Y, et al. An analytical flux density distribution model with a closed-form expression for a flatheliostat[J]. Applied Energy, 2019, 251: 113310.). The Cauchy model is defined on the plane of the heliostat, and the effective reflection area of ​​the heliostat is used as the convolution domain. Through the Cauchy model, the closed-form analytical solution of the radiation energy density distribution of each point on the plane where the heliostat mirror is located can be directly obtained, thus eliminating the need for numerical integration. He Caitou also proposed a method to fit the convolution kernel parameters of the Cauchy model through actual measurement results or ray tracing results. However, since the model parameters cannot be directly calculated through scene parameters, the Cauchy model cannot be applied to actual simulations. In order to solve the problem of increasing calculation error of the Cauchy model when the incident angle increases, He et al. proposed the iCauchy model in 2021, which is defined on the imaging plane (HE C, ZHAO H, HE Q, et al. Analytical radiative flux model via convolution integral and image plane mapping [J]. Energy, 2021, 222: 119937.). The iCauchy model projects the effective reflection area of ​​the heliostat onto the imaging plane as the convolution area, performs convolution calculation with the quasi-Cauchy kernel function, and maps the result to the receiver surface through oblique parallel projection, thereby improving the accuracy of the analytical model. Although the iCauchy model is more accurate than the Cauchy model, its parameters also need to be determined by fitting methods, so it still cannot be applied to actual simulations.

[0010] The convolution model uses the effective reflection area of ​​the heliostat or its projection area on the imaging plane as the convolution domain, and fully considers the geometric information of these areas. In addition, the model simulates the radiant energy density distribution of the micro-element reflection spot of the heliostat through the kernel function, which is not limited by the shape and size of the heliostat. However, the convolution model involves complex numerical integration calculations and has a large amount of calculations. Since not all functions can be solved to the original function, and there are not many cases where the original function can be expressed by elementary functions. In the convolution model, models based on Gaussian convolution kernels, such as the UNIZAR model and the Huang model, involve double integrals of two-dimensional Gaussian distributions in calculations. The usual implementation of such models is to use the Gauss-Legendre method for numerical integration for each receiver pixel visible to the micro-heliostat. The accuracy of this type of integration is affected by the integration points and weights. The more integration points and the more accurate the weights, the higher the accuracy of the integration. However, too many integration points will significantly slow down the speed of the integration calculation, and may also cause underflow of floating-point operations, resulting in a decrease in the accuracy of the result. At the same time, the complex integral calculation formula is not conducive to solving other optimization problems that rely on gradients. The more complex the integral formula, the more difficult it is to solve the gradient. How to efficiently solve the radiation energy density distribution of the convolution model without reducing the accuracy of the numerical integration method is a challenging problem. Summary of the invention

[0011] The purpose of the present invention is to provide a real-time simulation method and system for the radiation energy density distribution of a tower solar power generation system based on a convolution model. The method can quickly and efficiently simulate the radiation energy density of a light spot and ensure the accuracy of the result.

[0012] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0013] A real-time simulation method for radiation energy density distribution of a tower solar power generation system based on a convolution model comprises the following steps:

[0014] 1) Calculate the shadow distribution of the heliostat according to the given mirror field layout and light direction;

[0015] 2) Determine the parameters of the radiation energy density distribution function and establish an analytical model in the local coordinate system based on the convolution model;

[0016] 3) Calculating the radiation energy density of the light spot on the imaging plane in the local coordinate system using a fast numerical integration method based on the analytical model;

[0017] 4) The analytical model in the local coordinate system is projected obliquely parallel to the direction of light reflection onto the receiver plane, and the radiation energy density distribution on the receiver plane is calculated.

[0018] In the above technical solution, in order to ensure the accuracy and efficiency of the radiation energy density simulation, further, step 1) includes:

[0019] 1-1. For each heliostat in the mirror field, determine a set of related heliostats that may cause shadows and obstructions to the heliostat;

[0020] 1-2. Evenly discretize the heliostat plane into a number of sub-heliostats;

[0021] 1-3. Calculate the shadow and shielding conditions of each sub-heliostat respectively, and obtain the effective reflection area of ​​the heliostat that is not shadowed and shielded.

[0022] Furthermore, in order to improve the efficiency of determining related heliostats, in step 1-1, the GPU parallel light column traversal algorithm and bounding box culling algorithm are used to determine the set of related heliostats that shadow or block the current heliostat; in order to improve the accuracy of the radiation energy density simulation, in step 1-3, a ray is emitted from the center of the sub-heliostat along the directions of the incident ray and the reflected ray respectively, and the voxels they pass through in the mirror field are detected by the 3D-DDA algorithm, and the intersection judgment is performed with the sub-heliostat in the voxel to obtain the effective reflection area that is not shadowed or blocked.

[0023] Furthermore, in order to improve the accuracy of the radiation energy density simulation, step 2) specifically includes:

[0024] 2-1. Establish the origin passing through the focal point of the heliostat and perpendicular to the reflected light The imaging plane is the unit vector of the projection direction of the long side of the heliostat on the imaging plane. As the imaging plane coordinate system basis vector, for the imaging plane another basis vector ,have ;

[0025] 2-2. Project all sub-heliostats of the current heliostat in parallel and oblique manner onto the imaging plane;

[0026] 2-3. Obtain scene parameters, including: heliostat area, heliostat aspect ratio, distance from heliostat to receiver, heliostat glass thickness, incident light altitude angle, incident light azimuth angle, heliostat glass layer normal disturbance, CSR (Buie sunlight distribution model parameter), glass refractive index;

[0027] 2-4. Calculate the energy density of the heliostat , heliostat , Total error in direction , , and the length and width of the rectangle obtained by applying the inverse transformation of the shear transformation after the heliostat is projected onto the imaging plane; where,

[0028] 2-5. Calculate the inverse of the slope of the hypotenuse of the parallelogram after the heliostat is projected onto the imaging plane;

[0029] 2-6. Based on the convolution model, the radiation energy density distribution function on the imaging plane in the local coordinate system is obtained:

[0030]

[0031]

[0032] in is the coordinate of the sub-heliostat projected on the imaging plane, are the coordinates of a point on the imaging plane, is the integration area.

[0033] Furthermore, in order to improve the speed of the radiation energy density simulation, in step 3), the radiation energy density distribution function is deformed according to the shape of the integral region to reduce the calculation complexity and improve the calculation speed while ensuring the calculation accuracy. Specifically:

[0034] When the integral area When is a rectangle, we have:

[0035]

[0036] According to the formula:

[0037]

[0038]

[0039] Among them, A, is a constant.

[0040] have:

[0041]

[0042] in , , , is a portion of the coordinates of the four vertices of the rectangular integration region.

[0043] When the integral area When the region is a parallelogram, the inverse transformation of the shear transformation is applied to the integral region to obtain a new rectangular integral region. :

[0044]

[0045]

[0046] in is the reciprocal of the slope of the hypotenuse in the parallelogram integration region, The rectangular integration region after applying the shear transformation to the parallelogram integration region, , , , for The four vertices of

[0047] The shear transformation formula is:

[0048]

[0049] is a rectangular integration region The horizontal coordinate of the previous point The new coordinates after applying the shear transformation. Similarly, The vertical coordinate Corresponding to the new coordinates;

[0050] From the shear transformation formula:

[0051]

[0052] In this step, if the computing device supports fast To calculate, the following transformation can be used:

[0053]

[0054] In modern computing devices, multiplication and addition can be fused into one FMA instruction (Fused Multiply-Add) for calculation, that is,

[0055]

[0056] From the above formula we can get:

[0057]

[0058] On NVIDIA GPUs with compute capability 7.5 or higher, The calculation can be completed using a single assembly instruction "tanh.approx.f32". It only takes about 4 GPU instructions to complete the calculation quickly. Calculation.

[0059] Furthermore, in step 3), for the heliostats with shadow-blocked areas, the following method is used to quickly count the shadow-blocked areas:

[0060] The shadow occlusion area statistics are divided into 3 levels:

[0061] Level 1. Read the shadow shading results of four consecutive rows of sub-heliostats at a time. If none of the four rows of sub-heliostats are shaded, skip the shadow shading area statistics of this part. If there are shadowed sub-heliostats in the four rows of sub-heliostats, check the shadow shading of one row of sub-heliostats at a time.

[0062] Level 2. Read the shadow shading results of one row of sub-heliostats at a time. If all the sub-heliostats in this row are not shaded, skip the shadow shading area statistics of this part. If there are sub-heliostats shaded in this row, check the shadow shading of four consecutive sub-heliostats in this row at a time.

[0063] Level 3. Read the shadow shading results of four consecutive sub-heliostats at a time. If none of the four sub-heliostats are shaded, skip the shadow shading area statistics. If there are shadowed sub-heliostats among the four sub-heliostats, count their shadow shading situations respectively.

[0064] When counting the shadow occlusion area, proceed in order of level 1-level 2-level 3.

[0065] Furthermore, for each sub-heliostat in the shadowed area, the following method is used to quickly calculate its influence on the radiant energy density distribution on the imaging plane:

[0066] In step 3), the rectangular integral region after applying the shear transformation to the parallelogram integral region is obtained. and part of the coordinates of its four vertices , , , ,but The length is , width is Since the sub-heliostats are evenly divided by the rectangular heliostats, they are also rectangular, and their projections on the imaging plane are also parallelograms; assuming that the heliostats are evenly divided into Heliostat, set The sub-heliostat in the shadow area is located in the sub-heliostat grid of the heliostat where it is located. Line The projection area on the imaging plane is , then the rectangular integral area after the shear transformation is applied to the area is , Part of the coordinates of the four vertices is , , , , Length ,Width ,have:

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074] but:

[0075]

[0076] in, is the number of sub-heliostats located in the shadowed area, is the radiation energy density value blocked by the heliostat shadow.

[0077] Furthermore, for heliostats with shadowed areas, the following formula is used to calculate a point on the imaging plane: The radiation energy density distribution at:

[0078]

[0079] Integration area is the portion of the heliostat that is not obscured by the shadow. When shadow occlusion is not considered The radiation energy density at , the integration area is , which is the projection of the entire heliostat on the imaging plane.

[0080] Furthermore, in order to improve the accuracy of the radiation energy density simulation, step 4) specifically includes:

[0081] 4-1. Discretize the receiver surface into a uniform pixel grid and record the center position of each pixel and the receiver surface at Normal ;

[0082] 4-2. Judgment Whether it is visible to the heliostat, set the direction of the reflected light on the heliostat ,like , then it is believed that Visible to the heliostat; if visible, along Will Projected onto the heliostat imaging plane, the projection point The coordinates in the imaging plane coordinate system are ;

[0083] 4.3. Calculate the projection point according to the analytical model Radiant energy density value at the imaging plane ;

[0084] 4.4. Reflected light spot The radiation energy density at .

[0085] A real-time simulation system for the radiation energy density distribution of a tower solar power generation system based on a convolution model is used to implement any of the methods described above.

[0086] Compared with the prior art, the present invention is beneficial in that:

[0087] The present invention adopts a real-time simulation algorithm for the radiation energy density distribution based on a Gaussian convolution kernel convolution model, which can effectively and significantly improve the calculation efficiency without reducing the calculation accuracy of the existing Gaussian convolution kernel convolution model, so that it can be applied to real-time simulation of the mirror field, and can be used to solve the problems of mirror field layout optimization, receiver spot optimization, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 It is a rectangular area schematic diagram of the radiation energy density distribution simulation method of the plane light spot of the tower solar mirror field receiver in an embodiment of the present invention.

[0089] Figure 2 Schematic diagram of the world coordinate system and the local coordinate system in the heliostat field in an embodiment of the present invention;

[0090] Figure 3 Schematic diagram of the oblique parallel projection method in an embodiment of the present invention;

[0091] Figure 4 Schematic diagram of shadow shielding of heliostats in a mirror field according to an embodiment of the present invention;

[0092] Figure 5 Schematic diagram of the layout of the experimental heliostat field in an embodiment of the present invention. DETAILED DESCRIPTION

[0093] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described below in conjunction with the embodiments and the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the described embodiments, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0094] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the usual meanings understood by persons with ordinary skills in the field to which the present invention belongs. The words "include" or "comprise" and the like used in the present invention mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. The words "connect" or "connected" and the like are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0095] Example

[0096] The real-time simulation method of the radiation energy density distribution of a tower solar power generation system based on the convolution model in this embodiment includes the following steps:

[0097] S1, calculate the shadow distribution of the heliostat according to the mirror field layout and light direction. The details are as follows:

[0098] Given the mirror field layout and light direction, firstly, the GPU parallel light column traversal algorithm and bounding box culling algorithm are used to determine the relevant heliostat set that produces shadows or occlusions on the current heliostat. Then, the heliostat plane is evenly discretized into several sub-heliostats, and a ray is emitted from the center of the sub-heliostat along the direction of the incident ray and the reflected ray respectively. The voxels they pass through in the mirror field are detected by the 3D-DDA algorithm, and the intersection judgment is performed with the heliostat in the voxel, that is, whether the ray intersects with the rectangular area of ​​the heliostat mirror surface in the voxel. If it intersects, it is considered that the entire area of ​​the sub-heliostat is shadowed or occluded. Finally, the shadows and occlusions of the sub-heliostats are counted to obtain the effective reflection area that is not shadowed or occluded.

[0099] S2, using the neural network or the parameter calculation method in the existing model to determine the parameters of the radiation energy density distribution function and establish an analytical model in the local coordinate system. Specifically including:

[0100] Establish the origin through the focal point of the heliostat and perpendicular to the reflected light The imaging plane is the unit vector of the projection direction of the long side of the heliostat on the imaging plane. As the basis vector of the imaging plane coordinate system. For another basis vector of the imaging plane ,have ;

[0101] Projecting all sub-heliostats of the current heliostat in parallel and oblique directions onto the imaging plane;

[0102] Obtain scene parameters such as heliostat area, heliostat aspect ratio, distance from heliostat to receiver, heliostat glass thickness, incident light altitude angle, incident light azimuth angle, heliostat glass layer normal disturbance, CSR (parameters of Buie sunlight distribution model), and glass refractive index;

[0103] Calculate the energy density on the heliostat using a neural network or an existing convolutional model such as UNIZAR , Heliostat or Total error in direction , , and the length and width of the rectangle obtained by applying the inverse transformation of the shear transformation after the heliostat is projected onto the imaging plane;

[0104] Calculate the inverse of the slope of the hypotenuse of the parallelogram after the heliostat is projected onto the imaging plane;

[0105] Based on the convolution model, an analytical model of the radiation energy density distribution function on the imaging plane is constructed:

[0106]

[0107]

[0108] in is the coordinate of the micro-heliostat projected on the imaging plane, are the coordinates of a point on the imaging plane, is the integral area, i.e., the projection of the part of the heliostat not blocked by the shadow on the imaging plane.

[0109] S3 uses fast numerical integration method to efficiently calculate the radiation energy density of the spot in the local coordinate system. Specifically including:

[0110] 3-1. When the integral area When is a rectangle, we have:

[0111]

[0112] According to the formula:

[0113]

[0114]

[0115] Among them, A, is a constant.

[0116] have:

[0117]

[0118] The integral area is a rectangular area, , , , is a part of the coordinates of the four vertices of the rectangular integration area, A point on the imaging plane The radiation energy density distribution at .

[0119] When the integral area When the region is a parallelogram, the inverse transformation of the shear transformation is applied to the integral region to obtain a new rectangular integral region. :

[0120]

[0121]

[0122] in is the reciprocal of the slope of the hypotenuse in the parallelogram integration region, The rectangular integration region after applying the shear transformation to the parallelogram integration region, , , , for The shear transformation formula is:

[0123]

[0124] From the shear transformation formula we can get:

[0125]

[0126] 3-2. If the computing device supports fast To calculate, the following transformation can be used:

[0127]

[0128] In modern computing devices, multiplication and addition can be fused into one FMA instruction (Fused Multiply-Add) for calculation, that is,

[0129]

[0130] From the above formula we can get:

[0131]

[0132] On NVIDIA GPUs with compute capability 7.5 or higher, The calculation can be completed using a single assembly instruction "tanh.approx.f32". It only takes about 4 GPU instructions to complete the calculation quickly. Calculation.

[0133] 3-3. For heliostats with shadowed areas, use the following method to quickly count the shadowed areas:

[0134] Level 1. Read the shadow shading results of 4 consecutive rows of micro-heliostats at a time. If none of the 4 rows of micro-heliostats are shaded, skip the shadow shading area statistics of this part. If there are shadowed micro-heliostats in the 4 rows of micro-heliostats, check the shadow shading of one row of micro-heliostats at a time.

[0135] Level 2. Read the shadow shading results of one row of micro-heliostats at a time. If the micro-heliostats in this row are not shaded, skip the shadow shading area statistics of this part. If there are shadowed micro-heliostats in this row, check the shadow shading of four consecutive micro-heliostats in this row at a time.

[0136] Level 3. Read the shadow shading results of four consecutive micro-heliostats at one time. If none of the four micro-heliostats are shaded, skip the shadow shading area statistics. If there are shadowed micro-heliostats among the four micro-heliostats, count their shadow shading situations respectively.

[0137] When counting the shadow occlusion area, proceed in the order of level 1 -> level 2 -> level 3.

[0138] 3-4. For each sub-heliostat in the shadow area, the following method is used to quickly calculate its impact on the radiant energy density distribution on the imaging plane:

[0139] In step 3-1, the rectangular integral region after the shear transformation is applied to the parallelogram integral region is obtained. and part of the coordinates of its four vertices , , , ,but The length is , width is Since the sub-heliostats are evenly divided by the rectangular heliostats, they are also rectangular, and their projection on the imaging plane is also a parallelogram. Assume that the heliostats are evenly divided into Heliostat, set The sub-heliostat in the shadow area is located in the sub-heliostat grid of the heliostat where it is located. Line List( , The serial numbers are from The projection area on the imaging plane is , then the rectangular integral area after the shear transformation is applied to the area is , Part of the coordinates of the four vertices is , , , , Length ,Width ,have:

[0140]

[0141]

[0142]

[0143]

[0144]

[0145]

[0146]

[0147]

[0148] 3-5. For heliostats with shadowed areas, the following formula is used to calculate the The radiation energy density distribution at:

[0149]

[0150] Integration area is the portion of the heliostat that is not obscured by the shadow. When shadow occlusion is not considered The radiation energy density at , the integration area is , which is the projection of the entire heliostat on the imaging plane.

[0151] S4, projecting the analytical model in the local coordinate system obliquely parallel to the light reflection direction onto the receiver plane. Specifically, it includes the following:

[0152] Discretize the receiver surface into a uniform pixel grid and record the center position of each pixel and the receiver surface at Normal .

[0153] judge Whether it is visible to the heliostat, set the direction of the reflected light on the heliostat ,like , then it is believed that Visible to the heliostat; if visible, along Will Projected onto the heliostat imaging plane, the projection point The coordinates in the imaging plane coordinate system are .

[0154] Substitute the analytical model calculation formula to solve .

[0155] Reflected light spot The radiation energy density at

[0156] Algorithm complexity analysis:

[0157] In the implementation of the radiation energy density distribution simulation of the general convolution model, the Gauss-Legendre integral is used.

[0158] For the standard integration interval The Gauss-Legendre integral on is given by:

[0159]

[0160] The Gauss-Legendre integration points and weight coefficients can be obtained by the following formula:

[0161]

[0162]

[0163] is a classical orthogonal polynomial, defined as follows:

[0164]

[0165] For the Gauss-Legendre integral over a two-dimensional standard rectangular region, the formula is as follows:

[0166]

[0167] in, and are the number of integral points in two directions respectively.

[0168] If the integration region is not rectangular, it is necessary to map the quadrilateral to a standard rectangle through coordinate transformation so that the points in the two correspond one to one. The coordinate mapping can be achieved by the following formula:

[0169]

[0170] For the quadrilateral The coordinates of the vertices, is the shape function, defined as:

[0171]

[0172] In summary, the double Gauss-Legendre integral over the quadrilateral region can be expressed as

[0173]

[0174]

[0175] Unlike the Gaussian model (simplified model) which directly spreads the effect of shadows and shading on the radiant energy density over the entire heliostat, the convolution model needs to accurately reflect the effect of shadows and shading on the radiant energy density. In this embodiment, the heliostat is evenly divided into sub-heliostats (discrete grids), and the integral area used to calculate the radiant energy density is usually irregular. In the usual implementation, for a point on the imaging plane The radiation energy density of each sub-heliostat in the integration area needs to be calculated separately and accumulated, that is:

[0176]

[0177] The integral area It is the area on the heliostat that is not blocked by shadow.

[0178] In this embodiment, the following method is used to calculate a point on the imaging plane: The radiant energy density is:

[0179]

[0180] When shadow occlusion is not considered The radiation energy density at , the integration area is , which is the projection of the entire heliostat on the imaging plane.

[0181] For a well-planned heliostat field, the area blocked by shadows is approximately By using the method of this embodiment to distribute the calculation integral, the remaining steps are the same as those in the usual implementation, which can reduce about The amount of calculation.

[0182] In the following comparisons, add, subtract, multiply and divide are counted as one operation each. (For calculation )and Each is 5 operations.

[0183] In common implementations, the Gauss-Legendre integration points and weight coefficients are generally pre-calculated and not performed in the simulation.

[0184] In a typical implementation, the calculation

[0185]

[0186] need calculation (coefficient part + part + index part)

[0187] In the usual implementation, the Gauss–Legendre method is used to compute:

[0188]

[0189] The number of integration points in both directions is 6.

[0190] The computational cost of the double Gauss-Legendre integral on a single parallelogram area is (Shape function calculation + coefficients, , Multiplication calculation + Compute+ Calculate, here Right now Since the number of integration points in both directions is 6, a total of sub-integral point calculation).

[0191] In this embodiment, the following formula is used for calculation:

[0192]

[0193]

[0194]

[0195] The integral calculation amount over the parallelogram area is ( Calculation + parameter multiplication + erf and its parameter calculation + erf), where, due to , so the calculation of this parameter is ignored.

[0196] In the integral calculation part, the amount of calculation of the method in this embodiment is the amount of calculation of the usual implementation. .

[0197] The method in this embodiment is the same as the usual implementation, and the integral area is the projection of the part of the heliostat that is not blocked by the shadow on the imaging plane, that is, the projection of the entire heliostat on the imaging plane minus the projection of the part blocked by the shadow on the imaging plane. Before executing this step, it is necessary to count the shadow-blocked area on the heliostat.

[0198] In the usual implementation, the program needs to determine whether the sub-heliostat is in the shadow area one by one. sub-heliostats (n rows and m columns), the program needs to execute Second judgment.

[0199] In this embodiment, the judgment process is divided into three levels. Assume that a heliostat is divided into Micro-heliostats, for level 1, the best case is to conduct The worst case is to make an if judgment. If judgment; for level 2, the best case is to perform The worst case is to make an if judgment. If judgment; for level 3, the best case is to perform The worst case is to make an if judgment. Second if judgment.

[0200] For the actual shadow occlusion in the mirror field, see Figure 4 , for heliostat C, it is shaded by heliostats B and D. The shaded area is located below heliostat C and is relatively small. Therefore, in actual calculations, the above-mentioned level 1 is suitable for most cases of shadow shaded area statistics. If the number of judgments is close to the best case of level 1, that is, Second if judgment.

[0201] If the heliostat is divided into sub-heliostats, then in the usual implementation, The method of this embodiment requires executing about The computational effort is about the usual implementation of .

[0202] Experimental results:

[0203] After testing, the method of this embodiment is used for a mirror field of 6282 heliostats. Figure 5As shown, the size of the heliostat is 4.0m×3.2m, the height of the cylindrical receiver is 20m, the radius is 10m, and the height of the receiving tower is 180m. For the following comparative experiments, the model parameters are calculated using the formula of the UNIZAR model, and all scenario parameters and simulation parameters in the comparative experiments are the same.

[0204] In the UNIZAR model,

[0205]

[0206]

[0207]

[0208] in is the distance from the center of the heliostat to the focusing point; is the sun shape error, usually taken as a constant of 4.65mrad; is the slope error, caused by the uneven surface of the heliostat; is the tracking error, which is not considered in the simulation and is set to 0; is the incident angle of sunlight on the heliostat; is the direct normal irradiation intensity (DNI), which is usually taken as a constant of 1000; is the atmospheric transmission efficiency; is the heliostat reflectivity; is the total area of ​​heliostats; is the mirror area of ​​the heliostat; in the mirror field of 6282 heliostats, all heliostats are composed of only a single mirror, so .

[0209] The following experiment uses this loss function to analyze the error:

[0210]

[0211] in, is the number of pixels in the simulation results, is the value of the i-th pixel in the simulation result obtained by using the method of this embodiment, is the value of the i-th pixel in the simulation result obtained using the usual implementation method.

[0212] In the experiment, a heliostat in a field of 6282 heliostats was selected to verify the correctness of the fast erf(x) calculation in this embodiment.

[0213]

[0214] The accuracy comparison between the method of this embodiment and the conventional method (using Gauss-Legendre numerical integration) is as follows:

[0215]

[0216] It can be seen that the result of the fast calculation method proposed in this embodiment is correct. Although the fast erf calculation proposed in this embodiment will bring new errors, the error is still at a very small level ( ). If high-precision results are required, erf in the standard library can be used instead of the fast erf in this embodiment.

[0217] The comparison of the simulation speed of the mirror field composed of 6282 mirrors using the method of this embodiment and the conventional method is as follows:

[0218]

[0219] For the simulation parameters where the receiver pixel side length is 0.05 m and the heliostat is divided into 100x100, the error between the method of this embodiment and the conventional method is (Use the fast erf calculation proposed in this example).

[0220] It can be seen from the above experimental results that the real-time simulation method proposed in this embodiment can greatly improve the simulation speed of the radiation energy density distribution in the tower solar power generation system while ensuring the calculation accuracy.

Claims

1. A real-time simulation method for the radiation energy density distribution of a tower solar power generation system based on a convolution model, characterized in that: The following steps are involved: 1) Calculate the shadow distribution of the heliostat according to the given mirror field layout and light direction; 2) Determine the parameters of the radiation energy density distribution function and establish an analytical model in the local coordinate system based on the convolution model; specifically include: 2-1. Establish the origin passing through the focal point of the heliostat and perpendicular to the reflected light The imaging plane is the unit vector of the projection direction of the long side of the heliostat on the imaging plane. As the imaging plane coordinate system basis vector, for the imaging plane another basis vector ,have ; 2-2. Project all sub-heliostats of the current heliostat in parallel and oblique manner onto the imaging plane; 2-3. Obtain scene parameters, including: heliostat area, heliostat aspect ratio, distance from heliostat to receiver, heliostat glass thickness, incident light altitude angle, incident light azimuth angle, heliostat glass layer normal disturbance, Buie sunlight distribution model parameter CSR, and glass refractive index; 2-4. Calculate the energy density of the heliostat , heliostat , Total error in direction , , and the length and width of the rectangle obtained by applying the inverse transformation of the shear transformation after the heliostat is projected onto the imaging plane; where, 2-5. Calculate the inverse of the slope of the hypotenuse of the parallelogram after the heliostat is projected onto the imaging plane; 2-6. Based on the convolution model, the radiation energy density distribution function on the imaging plane in the local coordinate system is obtained: ; ; in is the coordinate of the sub-heliostat projected on the imaging plane, are the coordinates of a point on the imaging plane, is the integration area; 3) Based on the analytical model, a fast numerical integration method is used to calculate the radiation energy density of the light spot on the imaging plane in the local coordinate system; wherein the radiation energy density distribution function is deformed according to the shape of the integral area to reduce the calculation complexity and improve the calculation speed while ensuring the calculation accuracy, specifically: When the integral area When is a rectangle, we have: ; According to the formula: ; ; Among them, A, is a constant; have: ; in , , , is part of the coordinates of the four vertices of the rectangular integration area; When the integral area When the region is a parallelogram, the inverse transformation of the shear transformation is applied to the integral region to obtain a new rectangular integral region. : ; ; in is the reciprocal of the slope of the hypotenuse in the parallelogram integration region, The rectangular integration region after applying the shear transformation to the parallelogram integration region, , , , for The four vertices of The shear transformation formula is: ; is a rectangular integration region The horizontal coordinate of the previous point The new coordinates after applying the shear transformation. Similarly, The vertical coordinate Corresponding to the new coordinates; From the shear transformation formula: ; 4) The analytical model in the local coordinate system is projected obliquely parallel to the direction of light reflection onto the receiver plane, and the radiation energy density distribution on the receiver plane is calculated.

2. The real-time simulation method for radiation energy density distribution of a tower solar power generation system based on a convolution model according to claim 1 is characterized in that: Step 1) includes: 1-1. For each heliostat in the mirror field, determine a set of related heliostats that may cause shadows and obstructions to the heliostat; 1-2. Evenly discretize the heliostat plane into a number of sub-heliostats; 1-3. Calculate the shadow and shielding conditions of each sub-heliostat respectively, and obtain the effective reflection area of ​​the heliostat that is not shadowed and shielded.

3. The real-time simulation method for radiation energy density distribution of a tower solar power generation system based on a convolution model according to claim 2 is characterized in that: In step 1-1, the GPU parallel light column traversal algorithm and bounding box culling algorithm are used to determine the set of related heliostats that shadow or block the current heliostat; in step 1-3, a ray is emitted from the center of the sub-heliostat along the direction of the incident ray and the reflected ray respectively, and the voxels they pass through in the mirror field are detected by the 3D-DDA algorithm, and the intersection judgment is performed with the sub-heliostat in the voxel to obtain the effective reflection area that is not shadowed or blocked.

4. The real-time simulation method for radiation energy density distribution of a tower solar power generation system based on a convolution model according to claim 1 is characterized in that: In step 3), for heliostats with shadowed areas, the following method is used to quickly count the shadowed areas: The shadow occlusion area statistics are divided into 3 levels: Level 1. Read the shadow shading results of four consecutive rows of sub-heliostats at a time. If none of these four rows of sub-heliostats are shaded, skip the statistics of the shadow shading area of ​​this part. If there are sub-heliostats in the four rows that are shaded by shadows, then the shaded conditions of one row of sub-heliostats are checked at a time; Level 2. Read the shadow shading results of one row of sub-heliostats at a time. If all the sub-heliostats in this row are not shaded, skip the shadow shading area statistics of this part. If there are sub-heliostats shaded in this row, check the shadow shading of four consecutive sub-heliostats in this row at a time. Level 3. Read the shadow shading results of four consecutive sub-heliostats at a time. If none of the four sub-heliostats are shaded, skip the shadow shading area statistics. If there are shadowed sub-heliostats among the four sub-heliostats, count their shadow shading situations respectively. When counting the shadow occlusion area, proceed in order of level 1-level 2-level 3.

5. The real-time simulation method for radiation energy density distribution of a tower solar power generation system based on a convolution model according to claim 1 is characterized in that: For each sub-heliostat in the shadow area, the following method is used to quickly calculate its impact on the radiant energy density distribution on the imaging plane: Since the sub-heliostats are evenly divided by the rectangular heliostats, they are also rectangular, and their projections on the imaging plane are also parallelograms; assuming that the heliostats are evenly divided into Heliostat, set The sub-heliostat in the shadow area is located in the sub-heliostat grid of the heliostat where it is located. Line The projection area on the imaging plane is , then the rectangular integral area after the shear transformation is applied to the area is , Part of the coordinates of the four vertices is , , , , Length ,Width ,have: ; ; ; ; ; ; ; but: ; in, is the number of sub-heliostats located in the shadowed area, is the radiation energy density value blocked by the heliostat shadow.

6. The real-time simulation method for radiation energy density distribution of a tower solar power generation system based on a convolution model according to claim 1, characterized in that: For heliostats with shadowed areas, the following formula is used to calculate the value of a point on the imaging plane. The radiation energy density distribution at: ; Integration area is the portion of the heliostat that is not obscured by the shadow. When shadow occlusion is not considered The radiation energy density at , the integration area is , which is the projection of the entire heliostat on the imaging plane.

7. The real-time simulation method for radiation energy density distribution of a tower solar power generation system based on a convolution model according to claim 1, characterized in that: Step 4) specifically includes: 4-1. Discretize the receiver surface into a uniform pixel grid and record the center position of each pixel and the receiver surface at Normal ; 4-2. Judgment Whether it is visible to the heliostat, set the direction of the reflected light on the heliostat ,like , then it is believed that Visible to the heliostat; if visible, along Will Projected onto the heliostat imaging plane, the projection point The coordinates in the imaging plane coordinate system are ; 4.

3. Calculate the projection point according to the analytical model Radiant energy density value at the imaging plane ; 4.

4. Reflected light spot The radiation energy density at .

8. A real-time simulation system for the radiation energy density distribution of a tower solar power generation system based on a convolution model, characterized in that: Used to implement the method according to any one of claims 1 to 7.