Mixed light LED optical characteristic prediction method based on deep learning

Abstract

The invention discloses a mixed light LED optical characteristic prediction method based on deep learning, and relates to the field of photoelectron technology and artificial intelligence cross technology, and the method comprises the steps: adjusting the spectrum output through white-red LED configuration equipment, collecting time-related optical and chromaticity measurement data, and carrying out the prediction of the optical characteristics of a mixed light LED according to a spectrum power distribution model, constructing spectral power distribution of the white-red LED configuration equipment to obtain a steady data set of spectral power distribution measurement values; and training a deep learning algorithm based on the robust data set of the spectral power distribution measurement value, and predicting to obtain the spectral power distribution and optical and chromaticity parameters of the white-red LED configuration equipment. Therefore, by adopting the mixed light LED optical characteristic prediction method based on deep learning, a robust data set of spectral power distribution measurement values can be constructed, optical and chromaticity parameters can be predicted by a deep learning algorithm, and a reference basis is provided for optimizing LED spectral performance.

Description

Technical Field

[0001] The present invention relates to the field of the intersection of optoelectronics technology and artificial intelligence, and in particular to a method for predicting the optical characteristics of mixed-light LEDs based on deep learning. Background Art

[0002] White light-emitting diodes (LEDs) are widely used in various fields due to their energy efficiency, durability, and excellent color rendering. The color quality and performance of white LEDs are primarily influenced by their spectral characteristics, which are determined by the combination of phosphor materials and blue light-emitting chips. Therefore, ensuring the stability and consistency of the spectral power distribution (SPD) is particularly important for modern lighting, especially in LED devices.

[0003] However, traditional white light LEDs have problems such as low color rendering index, poor color temperature (CCT) stability, and uneven light spot. To solve the problems of traditional white light LEDs, existing technologies have adopted many methods for research. For example, TXLee et al. combined near-field hyperspectral images and far-field spectral angle distribution to study the spatial light intensity and color distribution, and provided an accurate and convenient method for developing white light LED optical models; MNKhan measured and analyzed the color stability of white light LED lamps over time based on thermal management solutions; CCSun et al. proposed a new modeling algorithm for phosphorescent conversion white light emitting diodes (pcWLEDs) to accurately simulate color appearance.

[0004] While existing technologies can partially address the challenges of traditional white LEDs, the optical properties of white LEDs are affected by a complex array of factors. Existing technologies often rely on physical models or experimental measurements to eliminate the influence of irrelevant factors. This leads to high computational complexity and sensitivity to slight changes in experimental conditions. This makes it difficult to efficiently handle the effects of multiple variables, limiting their adaptability in practical applications. Therefore, there is a need for a more efficient and scalable deep learning approach to improve the accuracy and reliability of LED spectral analysis and optimization. Summary of the Invention

[0005] The purpose of this invention is to provide a method for predicting the optical characteristics of mixed-light LEDs based on deep learning, which can adjust the spectral output by configuring white and red LED devices, construct a robust dataset of spectral power distribution measurements, and realize the prediction of optical and chromaticity parameters using deep learning algorithms.

[0006] To achieve the above objectives, the present invention provides a method for predicting the optical characteristics of mixed light LEDs based on deep learning, comprising the following steps:

[0007] S1. Regulate the spectral output of a white-red LED configuration device and collect time-correlated optical and colorimetric measurement data. Then, based on the spectral power distribution model, construct the spectral power distribution of the white-red LED configuration device to obtain a robust dataset of spectral power distribution measurements.

[0008] S2. Based on a robust dataset of spectral power distribution measurements, a deep learning algorithm is trained to predict the spectral power distribution and optical and chromaticity parameters of devices with white and red LED configurations.

[0009] Preferably, in step S1, the white-red LED configuration device includes a red LED and a white LED, and the red phosphor and environmental changes are simulated by adjusting the currents of the white LED and the red LED and controlling the heat sink temperature.

[0010] Preferably, the red LED and the white LED are powered by a RIGOL DP831 programmable DC linear power supply, managed by data acquisition software, and placed in the center of an integrating sphere coated with a highly reflective coating; at the same time, time-dependent optical and colorimetric measurement data are recorded by a HASS-2000 spectroradiometer and integrating sphere.

[0011] Preferably, the optical and colorimetric measurement data include color temperature, chromaticity coordinates and luminous flux.

[0012] Preferably, in step S1, the spectral power distribution of the white-red LED configuration device is constructed as follows:

[0013]

[0014] in,

[0015]

[0016] Where, P λ The spectral power distribution of the device configured for white-red LED, P λ,m is the mth spectral power distribution, P opt,m is the optical power of the mth spectral power distribution, Δλ m is the half-maximum width of the spectral power distribution, λ peak,m represents the peak wavelength of the mth spectral power distribution, h is Planck's constant, which is equal to 6.62607015×10 -34 J·s, c is the speed of light, c is equal to 3×10 8 m / s, λ1 and λ2 represent the boundary wavelengths of the spectral region.

[0017] Preferably, the peak wavelength and half-height width of the spectral power distribution are affected by temperature, and their expressions are as follows:

[0018] λ peak,m (Tj )=k peak,m (T j -T0)+λ peak,m,r ;

[0019] Δλ m (T j )=k peak,m (T j -T0)+Δλ m,r ;

[0020] Where k peak,m is the temperature coefficient that affects the peak wavelength, λ peak,m,r is the reference peak wavelength, Δλ m is the half-height width of the mth SPD, k Δλ,m is the influence coefficient of temperature on half-height width, Δλ m,r is the reference half-height width, T j is the temperature under a specific state, and T0 is the initial temperature.

[0021] Preferably, in step S2, the currents of the white LED and the red LED and the temperature of the controlled heat sink are used as input features of the deep learning algorithm to predict the spectral power distribution and optical and chromaticity parameters of the white-red LED configuration device.

[0022] Preferably, in step S2, the deep learning algorithm includes LSTM, CNN, AE and BP-NN.

[0023] Therefore, the present invention adopts the above-mentioned mixed light LED optical characteristics prediction method based on deep learning, which has the following technical effects:

[0024] (1) The present invention adjusts the spectral output by configuring a white-red LED device. This method can effectively simulate the change in the red phosphor ratio commonly found in phosphorescent white light LEDs, thereby achieving precise control of the LED spectral output.

[0025] (2) The present invention uses the heat sink temperature, white light LED current and red light LED current as input features of the deep learning algorithm, and realizes the prediction of the spectral power distribution and luminous flux, optical power, color temperature and chromaticity coordinates of white and red LED configuration devices under different working conditions, providing a reference basis for optimizing LED spectral performance in advanced lighting applications.

[0026] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic diagram illustrating the interrelationships among data collection, regression analysis, and experiments in an embodiment of a method for predicting optical characteristics of mixed-light LEDs based on deep learning;

[0028] Figure 2 In an embodiment of a method for predicting optical characteristics of a hybrid light LED based on deep learning, measured data of a white and red LED configuration device when the heat sink temperature is 20°C, wherein (a) is the luminous flux, (b) is the chromaticity coordinate x, (c) is the chromaticity coordinate y, and (d) is the color temperature CCT;

[0029] Figure 3 In an embodiment of a method for predicting optical characteristics of a hybrid light LED based on deep learning, measured data of a white and red LED configuration device when the heat sink temperature is 80°C, where (a) is the luminous flux, (b) is the chromaticity coordinate x, (c) is the chromaticity coordinate y, and (d) is the color temperature CCT;

[0030] Figure 4 This is an embodiment of a method for predicting optical properties of mixed-light LEDs based on deep learning, in which the SPD of a white-red LED configuration device under different working conditions is predicted by a deep learning algorithm, wherein (a) is the prediction result of a BP-NN model, (b) is the prediction result of an LSTM model, (c) is the prediction result of a CNN model, and (d) is the prediction result of an AE model;

[0031] Figure 5 The embodiment of the method for predicting optical characteristics of hybrid light LEDs based on deep learning is to measure and predict the optical and chromaticity parameter errors under different working conditions by BP-NN;

[0032] Figure 6 In an embodiment of a method for predicting optical properties of a hybrid light LED based on deep learning, the optical and chromaticity parameter errors measured and predicted under different working conditions are obtained by using LSTM;

[0033] Figure 7 In an embodiment of a method for predicting optical characteristics of a hybrid light LED based on deep learning, the optical and colorimetric parameter errors measured and predicted under different working conditions by CNN are analyzed;

[0034] Figure 8 The invention relates to an embodiment of a method for predicting optical characteristics of mixed light LEDs based on deep learning, which measures and predicts optical and chromaticity parameter errors under different working conditions through AE. DETAILED DESCRIPTION

[0035] The present invention can be explained in more detail by the following examples. The purpose of disclosing the present invention is to protect all changes and improvements within the scope of the present invention. The present invention is not limited to the following examples.

[0036] Example 1

[0037] See Figure 1, the present invention provides a method for predicting the optical properties of mixed light LEDs based on deep learning, including that the LED is powered by a RIGOLDP831 programmed DC linear power supply, managed by dedicated data acquisition software, and placed in the center of an integrating sphere coated with a highly reflective coating. At the same time, the HASS-2000 spectroradiometer and integrating sphere record time-related optical and colorimetric measurement data, wherein the integrating sphere collects light evenly from all angles, eliminates directional deviations, and ensures that the measured spectral power distribution (SPD) accurately represents the overall luminescence, rather than the luminescence of a limited angle segment. Such uniformity is crucial for determining chromaticity coordinates and color temperature (CCT). By integrating the light throughout the integrating sphere, the measurement of luminous flux and optical power can be made more consistent and repeatable, and this consistency helps to calibrate the system and obtain high-quality, reliable data, providing effective data support for deep learning models.

[0038] In actual production, the chemical composition, concentration, and coating uniformity of the phosphor may affect spectral stability. To this end, this embodiment adjusts the spectral output through a white-red LED configuration device. This method can effectively simulate the changes in the red phosphor ratio commonly found in phosphorescent white light LEDs, thereby achieving precise control of the LED spectral output. By adjusting the output of the white-red LEDs, key operating parameters (such as heat sink temperature, white LED current, and red LED current) can be systematically varied; at the same time, these parameters directly affect the spectral characteristics, including luminous flux, color temperature (CCT), and chromaticity coordinates.

[0039] The chromaticity characteristics of LEDs can theoretically be effectively characterized by their spectral power distribution (SPD). Generally, the asymmetric spectral power distribution of a monochromatic LED can be modeled using a Gaussian function as follows:

[0040]

[0041] Here, σ describes the spectral width of the emitted light, which is expressed as:

[0042]

[0043] Where, P opt is the optical power of the SPD, λ peak is the peak wavelength, Δλ is the half-height width of the SPD, λ is the specific wavelength, P λ is the spectral power distribution, h represents the Planck constant, which is equal to 6.62607015×10 -34 J·s, c represents the speed of light, c is equal to 3×10 8 m / s, λ1 and λ2 represent the boundary wavelengths of the spectral region, and ΔE represents the energy difference.

[0044] For mixed light LED, the corresponding spectral power distribution is given by Calculated. In addition, the peak wavelength λ peak The half-height width Δλ is affected by temperature, and its expression is as follows:

[0045] λ peak,m (T j )=k peak,m (T j -T0)+λ peak,m,r ;

[0046] Δλ m (T j )=k peak,m (T j -T0)+Δλ m,r ;

[0047] Where λ peak,m represents the peak wavelength of the mth SPD, k peak,m is the temperature coefficient that affects the peak wavelength, λ peak,m,r is the reference peak wavelength, Δλ m is the half-height width of the mth SPD, k Δλ,m is the influence coefficient of temperature on half-height width, Δλ m,r is the reference full width at half maximum.

[0048] The spectral power distribution of the white and red LED configuration device can be constructed by the above SPD model. Due to the changes in injection current and temperature, the optical power P of the mth SPD is opt,m There is a relationship between temperature and current. This embodiment explores the influence of white LED current and red LED current on spectral characteristics at different heat sink temperatures.

[0049] See Figure 2 and Figure 3 , when the white LED current ranges from 260mA to 270mA and the red LED current ranges from 20mA to 420mA (when the heat sink temperature is 80°C), the luminous flux of the combined system ranges from 94.3lm to 149.1lm. The injected current directly affects the carrier concentration in the active region of the LED, thereby controlling the radiative recombination rate and light emission intensity; and a higher current leads to an increase in light output (such as luminous flux). However, the sensitivity of the human eye reaches its peak in the green-yellow region, so the luminous efficacy of white light LEDs (which have a wider spectrum) is generally higher than that of red light LEDs. As the current of the red LED increases, its contribution to the total luminescence (which has a lower photopic response) becomes more significant, affecting the perceived brightness.

[0050] The chromaticity coordinates vary within the ranges of x: 0.32-0.46 and y: 0.308-0.313. Increasing the red LED current tends to shift the coordinates toward the red region of the CIE diagram, while increasing the proportion of white LED light shifts the coordinates toward the cooler (bluer) region. The chromaticity of the mixed light used in this embodiment is determined by the additive combination of the spectral outputs of the white and red LEDs. As the relative intensity of the red component increases, the overall light shifts toward warmer hues along the color axis. This follows the principle of additive color mixing, where the final perceived color is a function of the weighted contribution of each LED spectrum.

[0051] The color temperature (CCT) of the mixed light ranges from 1836K to 6026K, reflecting the spectral energy distribution of the light source. A higher contribution from red LEDs results in a lower CCT value (warmer light), while a predominant white LED configuration results in a higher CCT value (cooler light). Increasing the red LED current enriches the long-wavelength (red) portion of the spectrum, lowering the color temperature. Conversely, a higher contribution from white LEDs, including a greater proportion of blue and green wavelengths, results in a higher color temperature.

[0052] Based on the above theoretical foundations, this embodiment can simulate the impact of red phosphor and environmental changes on key spectral properties (including color temperature, chromaticity coordinates, and luminous flux) by adjusting the current of white LEDs and red LEDs and controlling the heat sink temperature, thereby constructing a robust dataset of high-quality spectral power distribution (SPD) measurements, laying a solid foundation for subsequent deep learning algorithm prediction of SPD.

[0053] Example 2

[0054] The present invention also trains an existing deep learning algorithm based on a robust dataset of spectral power distribution (SPD) measurements to verify the effectiveness of the robust dataset.

[0055] In this embodiment, four deep learning models, LSTM, CNN, AE and BP-NN, are selected to calculate the heat sink temperature T, the white LED driving current I W and the red LED drive current I R As input features, the SPD curve is predicted. During the training process, all four models are optimized using the Adam optimizer, with a learning rate of 0.001, a maximum number of iterations of 2000, and an early stopping technique to ensure training consistency.

[0056] The robust dataset of spectral power distribution measurements contains a total of 5166 sets of SPD data. In this example, data between 20°C and 68°C (4182 sets of data) are selected as the training set, and data between 71°C and 80°C (984 sets of data) are selected as the test set. Figure 4, CNN has the highest accuracy among the four deep learning algorithms, and the average determination coefficient (R 2 ) is 0.9985, and the mean square error (MSE) is 0.0030; the average R 2 The values ​​are 0.9980, 0.9978, and 0.9933, respectively, with average MSE values ​​of 0.0037, 0.0046, and 0.0163, respectively. This indicates that the precisely adjusted operating parameters in this embodiment, such as heat sink temperature, white LED current, and red LED current, are directly correlated with changes in the LED spectral output, providing a reference for optimizing LED spectral performance in advanced lighting applications.

[0057] In addition, this embodiment also uses the above four deep learning algorithms to predict the luminous flux Φ, color temperature (CCT) and chromaticity coordinates (x, y) of the white and red LED configuration device under different working conditions. Figures 5 to 8 It is a color map of the relative errors of the four algorithms in predicting the four optical parameters on the prediction set. Figures 5 to 8 The errors in the measured and predicted optical and colorimetric parameters of the white-red LED configuration—luminous flux Φ, chromaticity coordinates (x, y), and color temperature (CCT)—differed between the four deep learning models under various operating conditions (temperature, white LED current, red LED current). Specifically, the BP-NN model achieved an average relative error of 0.878% (Φ), 0.399% (x), 0.0368% (y), and 1.657% (CCT). In comparison, the LSTM model achieved an error of 0.432% (Φ), 0.285% (x), 0.0204% (y), and 0.472% (CCT), demonstrating its ability to capture the complex temporal dynamics that affect LED performance. The errors of the CNN model are 0.374% (Φ), 0.320% (x), 0.0161% (y), and 0.753% (CCT), while the errors of the AE model are 0.688% (Φ), 0.641% (x), 0.0275% (y), and 0.597% (CCT).

[0058] according to Figures 5 to 8, under various operating conditions (temperature, white LED current, red LED current), the errors in the measured and predicted optical and colorimetric parameters - luminous flux Φ, chromaticity coordinates (x,y), and color temperature (CCT) - of the white-red LED configuration differed between the four deep learning models. Specifically, the average relative errors of BP-NN, LSTM, CNN, and AE in predicting optical and colorimetric parameters were 0.743%, 0.302%, 0.366%, and 0.488%, respectively. Overall, LSTM performed best in predicting colorimetric and optical parameters. The superior performance of LSTM, CNN, and AE over BP-NN can be attributed to their enhanced ability to model the complex nonlinear relationship between operating conditions and LED optical properties. At the same time, the superior performance of LSTM, CNN, and AE over BP-NN can be attributed to their ability to capture the complex nonlinear relationship between LED optical properties, thereby establishing a direct link between the observed spectral behavior and its underlying physical causes. For example, the regulation of red LED output simulates the effect of varying the red phosphor ratio, while adjustments to the heat sink temperature and drive current reflect the impact of thermal and electrical conditions on the LED's spectral output. This integrated approach not only improves prediction accuracy but also deepens our understanding of the interplay between operating parameters and the optical performance of LED devices.

[0059] Therefore, the present invention adopts the above-mentioned deep learning-based mixed light LED optical property prediction method to construct a robust dataset of spectral power distribution measurement values, which can predict optical and chromaticity parameters through deep learning algorithms, providing a reference basis for optimizing LED spectral performance.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting optical properties of mixed light LEDs based on deep learning, characterized in that: The following steps are involved: S1. Regulate the spectral output of a white-red LED configuration device and collect time-correlated optical and colorimetric measurement data. Then, based on the spectral power distribution model, construct the spectral power distribution of the white-red LED configuration device to obtain a robust dataset of spectral power distribution measurements. S2. Based on a robust dataset of spectral power distribution measurements, a deep learning algorithm is trained to predict the spectral power distribution and optical and chromaticity parameters of devices with white and red LED configurations.

2. The method for predicting optical characteristics of mixed light LEDs based on deep learning according to claim 1, characterized in that: In step S1 , a white-red LED configuration device includes a red LED and a white LED, and adjusts the currents of the white LED and the red LED, and controls the temperature of the heat sink to simulate red phosphor and environmental changes.

3. The method for predicting optical characteristics of mixed light LEDs based on deep learning according to claim 2, characterized in that: The red and white LEDs were powered by a RIGOL DP831 programmable DC linear power supply, managed by data acquisition software, and placed in the center of an integrating sphere coated with a highly reflective coating. Meanwhile, time-correlated optical and colorimetric measurement data were recorded by a HASS-2000 spectroradiometer and integrating sphere.

4. The method for predicting optical characteristics of mixed light LEDs based on deep learning according to claim 1, characterized in that: The optical and colorimetric measurement data include color temperature, chromaticity coordinates and luminous flux.

5. The method for predicting optical characteristics of mixed light LEDs based on deep learning according to claim 1, characterized in that: In step S1, the spectral power distribution of the white and red LED configuration device is constructed as follows: in, Where, P λ The spectral power distribution of the device configured for white-red LED, P λ,m is the mth spectral power distribution, P opt,m is the optical power of the mth spectral power distribution, Δλ m is the half-maximum width of the spectral power distribution, λ peak,m represents the peak wavelength of the mth spectral power distribution, h is Planck's constant, c is the speed of light, and λ1 and λ2 are the boundary wavelengths of the spectral region.

6. The method for predicting optical characteristics of mixed light LEDs based on deep learning according to claim 5, characterized in that: The peak wavelength and half-height width of the spectral power distribution are affected by temperature, and their expressions are as follows: λ peak,m (T j )=k peak,m (T j -T0)+λ peak,m,r ; Dl m (T j )=k peak,m (T j -T0)+Dl m,r ; Where k peak,m is the temperature coefficient that affects the peak wavelength, λ peak,m,r is the reference peak wavelength, Δλ m is the half-height width of the mth SPD, k Δλ,m is the influence coefficient of temperature on half-height width, Δλ m,r is the reference half-height width, T j is the temperature under a specific state, and T0 is the initial temperature.

7. The method for predicting optical characteristics of mixed light LEDs based on deep learning according to claim 1 or claim 2, characterized in that: In step S2, the currents of the white LED and the red LED and the temperature of the control heat sink are used as input features of the deep learning algorithm to predict the spectral power distribution and optical and chromaticity parameters of the white-red LED configuration device.

8. The method for predicting optical characteristics of mixed light LEDs based on deep learning according to claim 1, characterized in that: In step S2, the deep learning algorithms include LSTM, CNN, AE and BP-NN.

Images