CN-115510746-B - Sky color modeling method based on sunny spectral information and BP neural network method

Abstract

The invention discloses a sky color modeling method based on sunny spectrum information, which solves the problems that the existing sky brightness model can not restore sky color information and color deviation exists when the correlated color temperature is used for representing colors. The method comprises the steps of measuring spectrum data of 145 sky primitives of a hemispherical surface by utilizing a spectrum sky scanner, solving color coordinates (x, y) of each sky primitive on the basis of the spectrum data, establishing a zenith color model on the basis of direct and scattered data, constructing a BP neural network model frame, and setting an input layer as a sky primitive altitude, a sky primitive azimuth angle, a solar altitude angle, a solar azimuth angle and a zenith color coordinate value to obtain a final model. The method establishes a model with color information to perfect the application of the sky model in the field of colorimetry. The invention provides a model to estimate the real-time sky spectrum data change rule, and provides theoretical reference for the lighting simulation of the building with spectrum selective glass. The color coordinates provided by the invention are more scientific and accurate than the correlated color temperature.

Inventors

• XUE PENG
• ZHANG YUWEI
• ZHAO YIFAN
• WANG HE
• LUO TAO

Assignees

• 北京工业大学

Dates

Publication Date

20260508

Application Date

20220922

Claims (1)

1. 1. The sky color modeling method based on the sunny spectral information is characterized by comprising the following steps of: step S10, calculating color coordinate values of 145 sky surface elements based on the measured spectrum data; step S20, establishing a color coordinate nonlinear regression model of sky surface elements where zenith is located Step S30, building and training a BP neural network; Step S40, predicting color coordinate values of 145 sky primitives; the method specifically comprises the following steps: step 1, calculating color coordinates of 145 sky surface elements based on measured spectrum data; for one sky scanning, the color coordinates of each sky surface source are obtained through spectrum measurement, light source tristimulus value calculation and color coordinate calculation; Step 1.1, utilizing a sky scanner to obtain spectrum data of 145 sky surface elements; Step 1.2, calculating a color coordinate value of each day of null surface element, calculating the color coordinate by taking a spectrum as a calculation input value according to the following formula: (1); (2); Wherein K m is the maximum spectral luminous efficacy of human eyes, lambda is the radiation wavelength value, 380-780 nm is the wave band capable of forming human eye luminosity and colorimetry perception, X, Y, Z are tristimulus values, P # ) The relative spectral power distribution of the sky test points; ( ), ( ), ( ) A color matching function is given for CIE through experiments, x, y and z are color coordinates, x+y+z=1, and each time of scanning, the color coordinates of 145 sky primitives are obtained; step2, establishing a color coordinate model of a sky surface element where a zenith is located; establishing a model of zenith color coordinates by using direct radiation and scattered radiation data; step 2.1, obtaining the direct solar irradiance and the sky scattering irradiance which are at the same time as sky scanning data by using a radiometer; Step 2.2, constructing a nonlinear calculation model of sky meteorological parameters based on direct solar irradiance and sky scattering irradiance, and estimating model parameter values by using a least square method, wherein the model parameter values are shown in the following formula: (3); Wherein E d is sky scattering irradiance, E s is direct solar irradiance, gamma s is zenith solar angle, and w is sky meteorological index; and 2.3, establishing a calculation model of the color coordinates of the sky-ground surface element where the zenith is positioned by using the sky meteorological parameters obtained in the step 2.2, and estimating the parameter values of the model by using a least square method, wherein the formula is as follows: (4); (5); step 3, building and training a sky surface element color coordinate neural network; Building and training a BP algorithm-based neural network, wherein the inputs of the BP algorithm-based neural network are sky-ground surface element altitude, sky-ground surface element azimuth, sun altitude, sun azimuth and zenith color coordinate values obtained in the step 2, and the X and Y color coordinate values obtained in the step 1 are output respectively; Step 3.1, building a BP neural network structure, setting a hidden layer to be a 2-layer full-connection structure, setting 13 nodes of each hidden layer, connecting the hidden layer with a full-connection layer with an output of 1, setting an activation function in the network to be a leakage ReLU, and setting a Loss function to be a mean square error MSE Loss; Step 3.2, preparing training data, and for calculating the used whole data, randomly selecting 90% of the data as a training set and the rest 10% of the data as a verification set; step 3.3, training the BP neural network, training by using a loss function, wherein the training frequency is 500 times at maximum, the learning rate is set to be 0.005, and training is stopped and model parameters are reserved when the maximum training frequency or the loss function convergence is reached; step 4, predicting the color coordinates of 145 sky surface elements, and establishing a sky color model; Calculating the x, y color coordinates of 145 sky surface elements in real time by utilizing time and radiation data; step 4.1, calculating the color coordinates of the sky surface element where the zenith is positioned in real time according to the step 2; Step 4.2, calculating a real-time solar altitude and solar azimuth by using the local time and the geographic position; Step 4.3, calculating the color coordinates of each sky surface element by using the BP neural network trained in the step 3; Step 4.3, calculating the color coordinates of each sky surface element by using the BP neural network trained in the step 3; step 4.4, it is known from the above analysis that the primary influencing factors of the color coordinates of each sky surface are the sky altitude angle and the sky square The five variables are used as input layer neurons, and the color coordinate x is used as output layer neurons to establish a neural network model; And 4.5, performing error analysis on the five-variable model input by the calculated zenith color coordinates, performing error analysis on the obtained model by using a medium normalized relative error formula NMBE and a root mean square error formula CVRMSE given in ASHRAE specifications, wherein the two indexes are suitable for error analysis of model calculation and measured values, and the specific error values are NMBEx =0.52%, CVRMSEx =3.54%, NMBEy =0.67% and CVRMSEx =3.55% after calculation.

Description

Sky color modeling method based on sunny spectral information and BP neural network method Technical Field The invention relates to a sky color modeling method based on spectrum information, and belongs to the technical field of natural lighting simulation of buildings. Background Natural light not only can bring brightness and visual effect, but also can improve the working efficiency and the health level of people. In addition, the energy-saving building material has great energy-saving potential as an important component of a green building. Quantification of natural lighting is typically performed by static parameters (illuminance and lighting coefficient) and dynamic parameters (natural lighting time percentage, effective lighting intensity, etc.). The calculation of the parameters is realized by means of numerical simulation, and the premise of the simulation is to establish a sky brightness distribution model, wherein a cloudy sky model, a sunny sky model and a full meteorological sky model are commonly used. However, natural light has not only the characteristic of brightness, but also color information, which has a very significant effect on human vision, non-vision and perception. Color studies often rely on colorimetry metrics such as color rendering index, associated color temperature and color coordinates, and the like. The color of the sky changes with time and weather conditions, which has important effects on color research, such as evaluating the performance of spectrum selective windowpanes, increasing the authenticity of outdoor scene rendering, assisting in developing a dimming and toning lighting control system, building a more realistic artificial dome experimental system, and the like. Therefore, it is necessary to build a sky color model (a mathematical model capable of reflecting the law of change in sky color). The invention discloses a sky color modeling method based on sunny spectral information, which mainly solves the problems that the existing sky brightness model can not restore sky color information and color deviation exists when the correlated color temperature is used for representing colors. The realization scheme of the sky color model is that a spectrum sky scanner is used for measuring spectrum data of 145 sky surface elements of a hemispherical surface, color coordinates (x, y) of each sky surface element are solved based on the spectrum data, a zenith color model is built based on direct and scattered data, a BP neural network model frame is built, and an input layer is set to be a sky surface element altitude angle, a sky surface element azimuth angle, a sun altitude angle, a sun azimuth angle and zenith color coordinate values to obtain the sky color model. Based on spectral information, the invention adopts BP neural network method analysis to realize sky color modeling. Disclosure of Invention Aiming at the problem that the existing sky model cannot represent sky colors and has color deviation, the invention provides a sky color modeling method based on sunny spectral information and a BP neural network method, which mainly takes color coordinates calculated based on measured spectra as an output layer, takes parameters such as sky surface element positions, sun positions, zenith color coordinates and the like as an input layer, establishes a model capable of representing sky color distribution through the BP neural network method, and comprises the following steps: Step 1, calculating color coordinates of 145 sky primitives based on the measured spectrum data. For one sky scan, the color coordinates of each sky surface element are obtained through spectrum measurement, light source tristimulus value calculation and color coordinate calculation. Step 1.1, utilizing an SP400 sky scanner to obtain spectrum data of 145 sky surface elements; Step 1.2, calculating a color coordinate value of each day of null surface element, calculating the color coordinate by taking a spectrum as a calculation input value according to the following formula: wherein K m is the maximum spectral luminous efficacy of human eyes, lambda is a radiation wavelength value, 380-780 nm is a wave band capable of forming human eye luminosity and colorimetry perception, X, Y, Z are tristimulus values, and P (lambda) is the relative spectral power distribution of sky test points; Given a color matching function for CIE through experiments, x, y, z are color coordinates, x+y+z=1, and each scan yields color coordinates of 145 sky primitives. And 2, establishing a color coordinate model of a sky surface element where the zenith is located. And modeling zenith color coordinates using the direct radiation and scattered radiation data. Step 2.1, obtaining the direct solar irradiance and the sky scattering irradiance which are at the same time as sky scanning data by using a radiometer; Step 2.2, constructing a nonlinear calculation model of sky meteorological parameters based on direct solar irradiance