An ai dynamic spectrum lighting system and control method simulating natural light rhythm

Abstract

The application relates to an AI dynamic spectrum lighting system and control method simulating natural light rhythm, which realizes double precise regulation and control of indoor light environment visual quality and rhythm effect by combining a biological rhythm model and a spectrum-chroma coupled deep neural network. The system generates a double target parameter set containing visual color coordinates and rhythm stimulation values based on user parameters and environmental sensor data, through melanopsin equivalent illuminance calculation and lens transmittance correction; through a pre-trained neural network combined with a 480nm cyan light channel adjustment strategy, optimal PWM driving signals of a five-channel LED are output while maintaining stable color coordinates; through real-time feedback of a spectrometer, PID closed-loop control and thermal wavelength drift compensation, combined with time domain dithering and logarithmic gradual change algorithm, smooth transition of light and color and dynamic optimization of rhythm effect are realized. While ensuring visual comfort, the application can precisely regulate non-visual biological effects of the human body according to natural rhythm, and is suitable for intelligent light environment scenes such as healthy lighting, medical rehabilitation and the like.

Description

Technical Field

[0001] This invention relates to the field of artificial lighting technology, specifically to an AI dynamic spectral lighting system and control method that simulates the rhythm of natural light. Background Technology

[0002] Modern photobiology research shows that light not only produces visual effects through cone and rod cells, but also influences non-visual biological effects such as circadian rhythms, hormone secretion, and alertness through intrinsically photosensitive retinal ganglion cells (ipRGCs). With the popularization of the concept of healthy lighting, constructing a lighting environment that can simultaneously meet the needs of visual tasks and maintain physiological circadian rhythms has become a core development direction in the field of intelligent lighting.

[0003] Currently, typical adjustable color temperature lighting technologies mainly employ a dual-channel hybrid scheme using cool white temperature LEDs and warm white temperature LEDs. While adjusting the color temperature (CCT), this technology also linearly changes its spectral energy distribution (SPD), resulting in a strong coupling relationship between color temperature and circadian rhythm stimulation values ​​(such as melanopsin equivalent illuminance m-EDI). In other words, "high color temperature inevitably leads to high circadian stimulation (energizing), and low color temperature inevitably leads to low circadian stimulation (relaxing)."

[0004] Existing control technologies are mostly based on look-up tables or simple polynomial fitting, relying on the calibration data of the LED light source at the factory. However, since visual parameters (color coordinates, illuminance) and non-visual parameters (m-EDI, CS value) are nonlinearly related, and multi-channel spectral mixing has infinitely many solutions (metamery), traditional simple linear control cannot independently and significantly adjust the intensity of rhythmic stimuli while maintaining the visual color.

[0005] Furthermore, existing technologies face significant challenges in long-term operation. Due to the inconsistency between the temperature characteristics and light decay curves of different colored LED chips (differential aging), open-loop control causes the mixed light color to drift over time, destroying the preset metamerism effect. At the same time, the lack of consideration for individual user differences (such as age-related changes in lens transmittance) results in significant deviations in the biological effects of the same light formula on different populations.

[0006] Therefore, there is an urgent need for an intelligent lighting system that can decouple visual effects from non-visual effects and accurately regulate the intensity of biorhythmic stimuli while maintaining stable visual parameters, so as to meet the diverse health lighting needs of users of different ages, occupations and scenarios. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies by providing an AI dynamic spectral lighting system and control method that simulates natural light rhythms. Based on the principle of metamerism, this method utilizes deep learning and multi-channel LED light sources to achieve precise dynamic regulation of non-visual biological rhythm effects while maintaining constant visual light color, thereby providing users with a visually comfortable and physiologically healthy intelligent light environment.

[0008] To achieve the above objectives, the present invention provides an AI dynamic spectral illumination system and control method that simulates natural light rhythms, comprising the following steps: Based on static parameters input by the user and real-time parameters collected by environmental sensors, the equivalent illuminance of the target melanin is calculated through a biological rhythm model and a lens transmittance correction coefficient is introduced for compensation. A dual target vector is constructed by combining the target chromaticity parameters to generate a dual target parameter set containing visual chromaticity coordinates and rhythmic stimulus values. Based on the dual target parameter set, it is input into the pre-trained spectral-chromaticity coupled deep neural network. Optical quality constraint judgment is performed through the hidden layer constraint mechanism, and the energy ratio of the 480nm band is adjusted while maintaining the constant chromaticity coordinates through the 480nm cyan channel adjustment strategy, and the five-channel optimal driving current duty cycle signal is output. Based on the five-channel optimal drive current duty cycle signal, the five-channel LED module is driven by a high-frequency PWM dimming driver to perform light and color mixing. The actual spectral power distribution and color coordinates are collected in real time by a spectrometer. The actual color temperature, brightness and black pixel equivalent illuminance value are calculated to obtain the current actual light environment parameters. Based on the deviation between the current actual light environment parameters and the dual target parameter set, the PID closed-loop feedback algorithm is used to fine-tune the PWM duty cycle of each channel, and the compensation coefficients of the red and cyan light channels are adjusted to address the wavelength drift caused by LED thermal effects. The time-domain jitter and logarithmic gradient algorithms are applied to generate an intermediate transition frame sequence and drive the light source in sequence, resulting in an intelligent light environment with smooth and constant visual effects and precise dynamic adjustment of rhythm effects.

[0009] Furthermore, based on user-input static parameters and real-time parameters collected by environmental sensors, the equivalent illuminance of the target melanopsin is calculated using a biorhythm model, and a lens transmittance correction coefficient is introduced for compensation. A dual target vector is constructed by combining the target chromaticity parameters, generating a dual target parameter set containing visual chromatic coordinates and rhythmic stimulus values, including: Based on static parameters input by the user and real-time parameters collected by environmental sensors, the equivalent illuminance of the target melanops is calculated through a circadian rhythm model deployed in the cloud or locally. Based on the user's age parameter, the lens transmittance correction coefficient is calculated according to the lens spectral transmittance formula, and the equivalent illuminance of the target melanin is compensated to obtain the compensated equivalent illuminance of the target melanin. Based on the user-defined target chromaticity parameters or scene mode, and combined with the compensated target black visual equivalent illuminance, a dual target vector is constructed to generate a dual target parameter set containing visual chromaticity coordinates and rhythmic stimulus values.

[0010] Furthermore, based on the user's age parameter, a lens transmittance correction coefficient is calculated according to the lens spectral transmittance formula, and compensation calculations are performed on the equivalent illuminance of the target melanoplasm to obtain the compensated equivalent illuminance of the target melanoplasm, including: Based on the user's age parameter, the user's lens transmittance is calculated according to the lens spectral transmittance formula; Based on the lens transmittance of the user and the lens transmittance of the standard observer, a lens transmittance correction factor is calculated. Based on the lens transmittance correction coefficient, the equivalent illuminance of the target melanocyte is compensated and calculated to generate the compensated equivalent illuminance of the target melanocyte.

[0011] Furthermore, the pre-trained spectral-chromaticity coupled deep neural network is obtained through the following steps: A basic spectral library was established based on the spectral power distribution of a five-channel LED under different driving currents and operating temperatures. Based on the aforementioned basic spectral library, the Monte Carlo mixing simulation method is used to randomly generate a combination of duty cycles for five channels, and the mixed spectrum is synthesized according to the principle of spectral superposition. Based on the mixed spectrum, the corresponding color coordinates, illuminance, color rendering index and melanopsine equivalent illuminance are calculated, and effective samples that meet the requirements of color rendering index and color tolerance are screened to construct a training set. Based on the training set, a lightweight fully connected regression network is trained to obtain the pre-trained spectral-chromaticity coupled deep neural network.

[0012] Furthermore, optical quality constraint determination is performed through a hidden layer constraint mechanism, including: Based on the hidden layer constraint mechanism of the deep neural network, the color rendering index and color tolerance of the current spectral combination are calculated. Based on the color rendering index and the color tolerance, determine whether the optical quality constraints are met. If the color rendering index is less than 90 or the color tolerance is greater than 2 SDCM, the penalty mechanism of the loss function is triggered, forcing the deep neural network to re-optimize in the metamerism solution space until it outputs a five-channel optimal drive current duty cycle signal that satisfies the optical quality constraints.

[0013] Furthermore, by employing a 480nm cyan channel adjustment strategy, the energy proportion of the 480nm band is adjusted while maintaining constant color coordinates, including: Based on the deep neural network, by increasing or decreasing the weight of the 480nm cyan channel and correspondingly adjusting the duty cycle of the blue and green channels, the energy proportion of the 480nm band in the spectrum is adjusted while keeping the mixed color coordinates constant within the McAdam ellipse range, and the adjusted five-channel optimal drive current duty cycle signal is output.

[0014] Furthermore, the five-channel LED module includes a red light channel, a green light channel, a blue light channel, a cyan light channel, and a warm white light channel. The peak wavelength of the cyan light channel is 478nm to 482nm, the peak wavelength of the red light channel is 660nm, the peak wavelength of the green light channel is 525nm, the peak wavelength of the blue light channel is 450nm, and the correlated color temperature of the warm white light channel is 2700K.

[0015] Furthermore, the high-frequency PWM dimming driver employs mixed-signal dimming technology to drive the five-channel LED module: In the high brightness range, a constant current amplitude modulation method is used to directly adjust the reference voltage of the DC-DC converter to change the driving current amplitude and drive the five-channel LED module. In the low brightness range or high-precision color mixing range, an ultra-high frequency PWM modulation method is adopted, with the PWM frequency set from 4kHz to 10kHz. A 16-bit PWM generator is used to provide 65,536 levels of grayscale adjustment capability to drive the five-channel LED module.

[0016] Furthermore, the PID closed-loop feedback algorithm is used to fine-tune the PWM duty cycle of each channel, and the compensation coefficients of the red and cyan channels are adjusted to address the wavelength drift caused by LED thermal effects, including: Based on the current actual light environment parameters and the dual target parameter set, calculate the color coordinate drift and rhythmic stimulation error. Based on the color coordinate drift and the rhythmic stimulation error, the PWM duty cycle adjustment of each channel is calculated using a PID closed-loop feedback algorithm. The PWM duty cycle of each channel is fine-tuned based on the PWM duty cycle adjustment amount. Based on the LED module temperature and temperature-luminous flux normalization curve collected by the NTC thermistor, the temperature compensation coefficients of the red and cyan channels are calculated, and the drive current is dynamically adjusted to obtain the compensated PWM duty cycle.

[0017] Furthermore, a sequence of intermediate transition frames is generated using temporal dithering and logarithmic gradation algorithms, and these frames are then used to drive the light source sequentially, including: Determine the initial spectral state and the target spectral state based on mode switching requirements or parameter adjustment requirements; Based on the initial spectral state and the target spectral state, the spectral parameters of the intermediate transition state are calculated using a time-domain dithering and logarithmic gradient algorithm to generate an intermediate transition frame sequence. Based on the intermediate transition frame sequence, the light source is driven sequentially to achieve a smooth transition from the initial spectral state to the target spectral state.

[0018] A second aspect of the present invention provides an AI dynamic spectral illumination system that simulates natural light rhythms, including a memory, a processor, and a five-channel light-emitting diode (LED) module, wherein the processor executes the above-described AI dynamic spectral illumination control method that simulates natural light rhythms. Attached Figure Description

[0019] Figure 1 This is a general block diagram of the system architecture of the present invention.

[0020] Figure 2 This is a schematic diagram of the deep neural network structure in this invention. Detailed Implementation

[0021] To facilitate understanding of this application, a more complete description will be provided below. This application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0022] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of the application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. In the description of this application, "several" means at least one, such as one, two, etc., unless otherwise explicitly specified.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] In this application, the technical features described in an open-ended manner include both closed technical solutions consisting of the listed features and open technical solutions that include the listed features.

[0025] In this application, numerical ranges are referred to as continuous unless otherwise specified, and include the minimum and maximum values ​​of the range, as well as every value between the minimum and maximum values. Furthermore, when the range refers to integers, it includes every integer between the minimum and maximum values ​​of the range. Additionally, when multiple ranges are provided to describe a feature or characteristic, the ranges may be merged. In other words, unless otherwise specified, all ranges disclosed herein should be understood to include any and all subranges to which they are incorporated.

[0026] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0027] like Figure 1 As shown, the AI ​​dynamic spectral illumination system and control method for simulating natural light rhythms provided by this invention has an overall architecture including a data acquisition layer, an algorithm processing layer, a drive control layer, and a light source output layer. This system, through AI deep neural networks and multi-channel LED mixing technology, achieves the function of precisely adjusting the intensity of biorhythmic stimuli while maintaining constant visual color coordinates. Example

[0028] This embodiment details the steps of calculating the equivalent illuminance of the target melanin using a biological rhythm model based on static parameters input by the user and real-time parameters collected by environmental sensors, and compensating by introducing a lens transmittance correction coefficient. It also describes the steps of constructing a dual target vector by combining the target chromaticity parameters, and generating a dual target parameter set containing visual chromaticity coordinates and rhythmic stimulus values.

[0029] Step S1: Based on the static parameters input by the user and the real-time parameters collected by the environmental sensors, calculate the equivalent illuminance of the target melanops through a biological rhythm model deployed in the cloud or locally.

[0030] Specifically, the system first receives static parameters input by the user, including the user's age and occupation (such as office worker, medical staff, student, etc.). Simultaneously, environmental sensors collect real-time parameters, including current time, ambient illuminance, and user activity status. The system then inputs these parameters into a cloud-based or locally deployed circadian rhythm model, based on the CIES026 standard model, to calculate the target melanopsightine equivalent illuminance (Target_mEDI) required by the user at the current moment.

[0031] The formula for calculating melanopsine equivalent illuminance (mEDI) is as follows: mEDI=∫[380nm, 780nm]S(λ)·M(λ)·dλ Where S(λ) represents the spectral power distribution of the light source, and M(λ) is the spectral sensitivity function of melanopsin, ranging from 380 nm to 780 nm wavelength. According to the CIES026 standard, M(λ) reaches its peak at approximately 480 nm.

[0032] For example, during the working hours from 9:00 AM to 12:00 PM, the target melanopsine equivalent illuminance calculated by the model for office workers may be 350-450 mEDI units to provide appropriate alertness stimulation; while after 8:00 PM, the target melanopsine equivalent illuminance may be reduced to below 100 mEDI units to promote melatonin secretion and prepare for sleep.

[0033] Step S2: Based on the user's age parameter, calculate the user's lens transmittance according to the lens spectral transmittance formula.

[0034] Specifically, the system calculates the transmittance of the user's lens to different wavelengths of light based on the user's age parameter and using the lens spectral transmittance formula provided in the CIE 203:2012 standard. As people age, the human eye lens yellows, and the transmittance to short-wavelength light (especially the ipRGCs sensitive area around 480nm) decreases significantly.

[0035] The formula for calculating lens transmittance T_lens(λ, age) is as follows: T_lens(λ,age)=10^[-(A_L1(λ)·age+A_L0(λ))] Where A_L1(λ) and A_L0(λ) are wavelength-dependent coefficients, which can be obtained from the CIE203:2012 standard.

[0036] Step S2.1: Calculate the lens transmittance correction factor based on the lens transmittance of the user and the lens transmittance of the standard observer.

[0037] Specifically, the system compares the calculated lens transmittance of the user with that of a standard observer (typically 32 years old) to calculate a lens transmittance correction factor. This correction factor reflects the difference in the actual number of photons reaching the retina when the user and the standard observer receive the same light source.

[0038] The formula for calculating the correction factor C_age is as follows: C_age=∫[380nm, 780nm]S(λ)·M(λ)·T_lens(λ, 32)·dλ / ∫[380nm, 780nm]S(λ)·M(λ)·T_lens(λ, age)·dλ This formula takes into account the difference in effective melanopsin stimulation between a standard observer and a user of a specific age under a given light source S(λ).

[0039] The system pre-calculates and stores correction factor curves corresponding to different ages. For example, for a 65-year-old user, the transmittance of their lens at a wavelength of 480nm may only be 60% of that of a 32-year-old standard observer, so the calculated correction factor is 1 / 0.6≈1.67.

[0040] Step S2.2: Based on the lens transmittance correction coefficient, calculate the compensation for the equivalent illuminance of the target melanocyte to generate the compensated equivalent illuminance of the target melanocyte.

[0041] Specifically, the system multiplies the target melanoplasmic equivalent illuminance calculated in step S1 by the correction coefficient in step S2.1 to obtain the compensated target melanoplasmic equivalent illuminance: Final_mEDI = Target_mEDI * C_age This ensures that users of different ages receive the same intensity of effective melanin stimulation reaching their retina when using the system, thus achieving personalized circadian rhythm regulation.

[0042] For example, if a 65-year-old user's target mEDI is 200, after considering transmittance correction, the actual driving mEDI value set internally by the system should be 200*1.67≈334 to compensate for the filtering effect of their lens on short-wavelength blue light.

[0043] Step S3: Based on the user-defined target chromaticity parameters or scene mode, and combined with the compensated target black visual equivalent illuminance, construct a dual target vector to generate a dual target parameter set containing visual chromaticity coordinates and rhythmic stimulus values.

[0044] Specifically, the system receives target color parameters set by the user through the interface, such as the desired color temperature (e.g., 3000K, 4000K, 5000K, etc.), brightness (e.g., 300lux, 500lux, etc.), or the directly selected scene mode (e.g., "reading mode", "work mode", "rest mode", etc.). The system converts these visual parameters into target color coordinates (Target_x, Target_y) and illuminance values ​​(Target_Y).

[0045] The conversion from color temperature to color coordinates uses the following formula (McCamy approximation formula): If the color temperature (CCT) is within the range of [1000K, 15000K]: n=10^6 / CCT x=-0.2661239*10^9 / CCT^3-0.2343589*10^6 / CCT^2+0.8776956*10^3 / CCT+0.179910 y=-1.1063814*x^3-1.34811020*x^2+2.18555832*x-0.20219683 Subsequently, the system combines these visual parameters with the compensated target melanopyroid equivalent illuminance (Final_mEDI) obtained in step S2.2 to construct a dual target vector: Target_Vector=[Target_x, Target_y, Target_Y, Final_mEDI] This parameter set includes both visual requirements (visual color coordinates) and non-visual biological effect requirements (rhythmic stimulus values), serving as control targets for subsequent spectral synthesis. Example

[0046] This embodiment details the steps of inputting a dual-target parameter set into a pre-trained spectral-chromaticity coupled deep neural network, performing optical quality constraint judgment through a hidden layer constraint mechanism, and adjusting the energy ratio of the 480nm band while maintaining constant chromatic coordinates through a 480nm cyan channel adjustment strategy, and outputting a five-channel optimal drive current duty cycle signal.

[0047] Step S4: Based on the dual target parameter set, input it into the pre-trained spectral-chromaticity coupled deep neural network.

[0048] like Figure 2 As shown, the system inputs the dual objective parameter set generated in Example 1 into a pre-trained spectral-chromaticity coupled deep neural network. This network is a lightweight neural network specifically trained for spectral-chromaticity relationships, and its structure includes: Input layer (4 neurons): Receives the target color coordinates (x, y), brightness (Y), and mEDI value; Hidden layer 1 (64 neurons, ReLU activated); Hidden layer 2 (128 neurons, ReLU activation, Dropout=0.2 to prevent overfitting); Hidden layer 3 (64 neurons, ReLU activated); Output layer (5 neurons, Sigmoid activation): corresponding to 5 PWM duty cycles (0.0-1.0).

[0049] The training process for this network is as follows: First, a basic spectral library S_i(λ, I, T) was established based on the spectral power distribution of a five-channel LED (red, green, blue, cyan, and warm white) under different driving currents and operating temperatures. These spectral data were obtained through laboratory measurements, ensuring the physical accuracy of the model.

[0050] Then, based on these fundamental spectra, a combination of duty cycles (d_1, d_2, d_3, d_4, d_5) of five channels is randomly generated using the Monte Carlo mixing simulation method, and a mixed spectrum is synthesized according to the principle of spectral superposition. S_mix(λ)=Σ[i=1to5]d_i·S_i(λ, T_ref) Next, for each mixed spectrum, the corresponding color coordinates (x, y), illuminance Y, color rendering index Ra, and rhythm factor mEDI are calculated: [x, y]=(Σ[λ=380nmto780nm]S_mix(λ)·[x̄(λ),ȳ(λ)]) / (Σ[λ=380nmto780nm]S_mix(λ)·ȳ(λ)) Y=k_m·Σ[λ=380nmto780nm]S_mix(λ)·ȳ(λ) mEDI=k_mel·Σ[λ=380nmto780nm]S_mix(λ)·M(λ) Where x̄(λ) and ȳ(λ) are the CIE1931 standard chromaticity observer functions, k_m is the maximum luminous efficiency (683 lm / W), and k_mel is the melanopsin normalization constant.

[0051] Valid samples (approximately 2 million sets) meeting the criteria of color rendering index Ra > 90 and color tolerance less than 2 SDCM were selected as the training set for the neural network. During training, the input was (Target_x, Target_y, Target_Y, Target_mEDI), and the labels were (d_1, d_2, d_3, d_4, d_5). The loss function used was the weighted mean squared error (MSE). Loss=α·MSE(x,y)+β·MSE(Y)+γ·MSE(mEDI)+δ·(Max(0,90-Ra)+Max(0,SDCM-2)) The last two terms are penalty terms for optical quality constraints, with weighting coefficients α, β, γ, and δ of 1.0, 0.5, 0.8, and 2.0, respectively, to balance the importance of each objective.

[0052] After training, the network can quickly predict the optimal channel duty cycle combination under given target parameters, reducing the traditional computational latency of tens of milliseconds to the microsecond level and achieving real-time response.

[0053] Step S5: Based on the hidden layer constraint mechanism of the deep neural network, calculate the color rendering index and color tolerance of the current spectral combination.

[0054] Specifically, during the forward propagation process, the neural network calculates the spectral quality parameters corresponding to the currently predicted channel duty cycle combination through specific hidden layer units, including the color rendering index (Ra and R9) and the color tolerance (expressed in SDCM units) relative to the target color coordinates.

[0055] The calculation of the color rendering index Ra follows the CIE 13.3-1995 standard and requires calculating the color difference of eight standard test color samples under a reference illuminant and a test illuminant: R_i = 100 - 4.6·ΔE_i Ra = (1 / 8)Σ[i=1 to 8]R_i The calculation of color tolerance SDCM is based on the relationship between the Euclidean distance in the u'v' color space and the MacAdam ellipse: SDCM=sqrt((u'_actual-u'_target)^2+(v'_actual-v'_target)^2) / SDCM_unit(CCT) Where SDCM_unit(CCT) is the distance corresponding to 1 SDCM unit at a specific color temperature.

[0056] Step S5.1: Based on the color rendering index and the color tolerance, determine whether the optical quality constraints are met.

[0057] Specifically, the system determines whether the calculated color rendering index is greater than 90 and whether the color tolerance is less than 2 SDCM. These two indicators respectively ensure the color reproduction of the mixed light and its accurate matching with the target color coordinates.

[0058] Step S5.2: If the color rendering index is less than 90 or the color tolerance is greater than 2 SDCM, the penalty mechanism of the loss function is triggered, forcing the deep neural network to re-optimize in the metamerism solution space until it outputs a five-channel optimal driving current duty cycle signal that satisfies the optical quality constraints.

[0059] Specifically, if the prediction result does not meet the quality constraints, the system will trigger the penalty term in the loss function: L_penalty=α·max(0,90-Ra)+β·max(0,SDCM-2) This penalty mechanism increases the loss value for solutions that do not meet the conditions, forcing the network to search the solution space that meets the constraints during gradient descent until a solution that simultaneously satisfies the requirements for color coordinate accuracy and spectral quality is found.

[0060] Step S6: Based on the deep neural network, by increasing or decreasing the weight of the 480nm cyan channel and correspondingly adjusting the duty cycle of the blue and green channels, while keeping the mixed color coordinates constant within the range of the McAdam ellipse, the energy proportion of the 480nm band in the spectrum is adjusted, and the adjusted five-channel optimal drive current duty cycle signal is output.

[0061] In a specific embodiment, the system identifies 480nm wavelength as the sensitive peak of melanopsin. At the same time, the visual efficiency function V(λ) is relatively low at this wavelength. This characteristic makes increasing or decreasing the energy of this wavelength have a significant impact on non-visual biological effects, but a relatively small impact on perceived color.

[0062] The system introduces a special "480nm cyan balance operator" in the last layer of the neural network. This operator allows the output proportion of the 480nm channel to be adjusted while maintaining the overall color coordinates. When it is necessary to increase the mEDI value, the system increases the weight of the 480nm cyan channel and decreases the weight of the 450nm blue channel; when it is necessary to decrease the mEDI value, the operation is reversed.

[0063] The specific implementation of the metamerism adjustment algorithm is as follows: 1. Given an initial five-channel duty cycle vector d_vec=[d_red, d_green, d_blue, d_cyan, d_warm], 2. While keeping the target color coordinates constant, calculate the change in duty cycle Δd_cyan of the cyan channel: To increase mEDI: Δd_cyan = min(0.3, (Target_mEDI - Current_mEDI) / k); To reduce mEDI: Δd_cyan = max(-0.2, (Target_mEDI - Current_mEDI) / k); Where k is an empirical coefficient, which is related to the light source configuration.

[0064] 3. To maintain the color coordinates constant, calculate the corresponding adjustments for the blue and green light channels: Δd_blue=-Δd_cyan·(C_b / C_c), Δd_green=-Δd_cyan·(C_g / C_c)·((Current_x-Blue_x) / (Green_x-Current_x)), Where C_b, C_g, and C_c are the normalization coefficients for the blue, green, and cyan channels, respectively, and Blue_x and Green_x are the x-coordinates of the corresponding color points for each channel.

[0065] 4. Update the duty cycle vector: d_vec_new=[d_red, d_green+Δd_green, d_blue+Δd_blue, d_cyan+Δd_cyan, d_warm].

[0066] 5. Calculate the chromaticity coordinates and mEDI under the new duty cycle, and verify whether the following conditions are met: Color coordinate offset <0.001 (approximately 1 SDCM); |New mEDI-Target_mEDI| < Error threshold.

[0067] 6. If not satisfied, return to step 2 to adjust the parameters; if satisfied, output the final duty cycle vector.

[0068] This adjustment process is constrained by the range of the McAdam ellipse, ensuring that color coordinate changes remain within a range imperceptible to the human eye (typically within 1 SDCM). In this way, the system can achieve a maximum mEDI value adjustment range of ±40% without altering visual color perception.

[0069] Finally, the system outputs a five-channel optimal drive current duty cycle signal that satisfies all constraints, i.e., the PWM duty cycle value (0%-100%) of each channel. Example

[0070] This embodiment details the steps of driving a five-channel LED module to perform light and color mixing based on the five-channel optimal drive current duty cycle signal, using a high-frequency PWM dimming driver, acquiring the actual spectral power distribution and color coordinates in real time using a spectrometer, calculating the actual color temperature, brightness, and black pixel equivalent illuminance value, and obtaining the current actual light environment parameters.

[0071] Step S7: Based on the five-channel optimal drive current duty cycle signal, drive the five-channel LED module to perform light color mixing through a high-frequency PWM dimming driver.

[0072] The system employs a light mixing module composed of five different types of LEDs, including: 1. Cyan channel: peak wavelength of 478nm to 482nm, corresponding to the melanopsin absorption peak of ipRGCs; 2. Red channel: Peak wavelength is 660nm, used to compensate for the increase in color temperature caused by the addition of cyan light and to supplement the color rendering of R9; 3. Green channel: Peak wavelength of 525nm, used to adjust light performance (brightness balance); 4. Blue channel: Peak wavelength of 450nm, used for regular color temperature adjustment; 5. Warm White Channel: With a correlated color temperature of 2700K, it provides a continuous spectral base, fills the "spectral holes" caused by the narrow band spectrum of RGB, and ensures a high color rendering index.

[0073] This specific five-channel combination design is intended to maximize the "visual-non-visual" decoupling capability and precisely match the difference in sensitivity curves between human eye cone cells (L, M, S cones) and photoreceptor ganglion cells (ipRGCs).

[0074] The system's high-frequency PWM dimming driver employs hybrid analog-digital dimming technology and operates in three different zones: 1. High brightness range (100%-32%) (71): A constant current amplitude modulation method (AnalogDimming) is used to directly adjust the reference voltage of the DC-DC converter and change the current amplitude. This method has high luminous efficiency and no flicker problem. Specifically, the control unit (1) adjusts the reference voltage through the DAC output, and the constant current driver (2) controls the current value of each channel LED.

[0075] 2. Low brightness / high precision color mixing range (28%-0.1%) (72): The ultra-high frequency PWM modulation method is adopted, and the frequency is set from 4kHz to 10kHz, which is far higher than the IEEE1789-2015 standard for the definition of high-risk flicker (>3000Hz is considered risk-free). The system adopts a 16-bit PWM generator, which provides 65536 levels of grayscale adjustment capability to ensure smooth color transition without stepping when fine-tuning the proportion of each channel. Specifically, the control unit (1) generates PWM signals to drive MOSFET switches (MOS1-MOS5) to control the LED current path.

[0076] 3. Overlap Buffer (28%-32%) (73): Within this interval, both amplitude modulation and PWM modulation are used simultaneously. The weighting coefficient is calculated according to the following formula: W_PWM=(32%-Brightness) / (32%-28%); W_Analog=1-W_PWM.

[0077] The final driving strength is: Drive=W_Analog·Analog_Drive+W_PWM·PWM_Drive; This weighted blending method ensures a smooth transition between the two dimming modes, avoiding abrupt changes in brightness.

[0078] The specific implementation process of PWM / DC switching is as follows: 1. The system detects the current required brightness percentage level; 2. If level > 32%, use DC dimming exclusively, set DC_Level = level, PWM_Level = 0; 3. If level < 28%, use PWM dimming exclusively, set PWM_Level = level, DC_Level = 0; 4. If 28% ≤ level ≤ 32% (buffer zone), calculate the mixed weights: weight_PWM=(32%-level) / (32%-28%); weight_DC = 1 - weight_PWM; Set DC_Level = 32% * weight_DC; Set PWM_Level=(level / 32%)*weight_PWM*100%.

[0079] 5. Output DC_Level through the DAC and PWM_Level through the PWM generator.

[0080] Step S8: Collect the actual spectral power distribution and color coordinates in real time using a spectrometer, calculate the actual color temperature, brightness, and melanopsin equivalent illuminance value, and obtain the current actual light environment parameters.

[0081] Specifically, the system integrates a miniature spectrometer or multi-channel color sensor (XYZ+Cyan) inside the luminaire to collect the actual spectral power distribution (SPD) and the actual color coordinates of the light-emitting surface after mixing in real time. The acquisition frequency is 10Hz to ensure that the system can capture changes in light output in a timely manner.

[0082] Based on the acquired actual spectral data, the system calculates the following parameters in real time: 1. Actual chromatic coordinates (x, y): calculated using standard colorimetry; 2. Actual correlated color temperature (CCT): Calculated using the McCamy formula or by looking up a table; 3. Actual illuminance (lux): Calculated by integration; 4. Actual color rendering index (Ra): The color rendering properties of the eight test samples were calculated using the CIE method; 5. Actual mEDI value: Calculated by integrating the spectrum with the melanopsin action function.

[0083] These parameters constitute the current set of actual optical environment parameters, which will be used for subsequent closed-loop control. Example

[0084] This embodiment details the steps of using a PID closed-loop feedback algorithm to fine-tune the PWM duty cycle of each channel based on the deviation between the current actual light environment parameters and the dual target parameter set, adjusting the compensation coefficients of the red and cyan light channels to address the wavelength drift caused by LED thermal effects, and applying time-domain jitter and logarithmic gradient algorithms to generate an intermediate transition frame sequence and drive the light source sequentially, resulting in an intelligent light environment with a smooth and constant visual effect and precise dynamic adjustment of the rhythm effect.

[0085] Step S9: Based on the current actual light environment parameters and the dual target parameter set, calculate the color coordinate drift and rhythmic stimulation error.

[0086] The system compares the real-time acquired actual light environment parameters in Example 3 with the dual target parameter set generated in Example 1, and calculates the deviation: 1. Color coordinate shift: Δuv=sqrt((u'_actual-u'_target)^2+(v'_actual-v'_target)^2); 2. Rhythmic stimulus error: ΔmEDI = mEDI_actual - mEDI_target.

[0087] Step S10: Based on the color coordinate drift and the rhythmic stimulation error, calculate the PWM duty cycle adjustment of each channel using a PID closed-loop feedback algorithm.

[0088] Specifically, the system employs two parallel PID controllers to address color coordinate drift and rhythmic stimulus errors: 1. Color coordinate PID controller: Calculates the adjustment values ​​for the red, green, and blue channels, and is mainly responsible for color coordinate correction; 2. Rhythmic Stimulation PID Controller: Calculates the differential adjustment amount of the cyan and blue light channels, and adjusts the mEDI value while keeping the color coordinates unchanged.

[0089] The specific implementation of the PID controller is as follows: ΔPWM_i(t)=K_p·e(t)+K_i·∫[0,t]e(τ)dτ+K_d·de(t) / dt, Where e(t) is the current error (which can be either chromaticity coordinate error or mEDI error), and K_p, K_i, and K_d are the proportional, integral, and derivative coefficients, respectively. For chromaticity coordinate PID, typical parameters are set to K_p=0.5, K_i=0.1, and K_d=0.05; for mEDIPID, typical parameters are set to K_p=0.3, K_i=0.08, and K_d=0.02.

[0090] The output of the PID controller is the PWM duty cycle adjustment ΔPWM_i for each channel.

[0091] Step S11: Fine-tune the PWM duty cycle of each channel based on the PWM duty cycle adjustment amount, and calculate the temperature compensation coefficients of the red light channel and the cyan light channel based on the LED module temperature and temperature-luminous flux normalization curve collected by the NTC thermistor, dynamically adjust the drive current, and obtain the compensated PWM duty cycle.

[0092] Specifically, the system first performs basic adjustments to the PWM duty cycle of each channel: PWM_new_i=PWM_old_i+ΔPWM_i; Then, the system monitors the LED junction temperature in real time using an NTC thermistor (5) arranged on the LED aluminum substrate. Based on the temperature data and the pre-stored temperature-luminous flux normalization curve, the system calculates the thermal compensation coefficient.

[0093] The temperature-luminous flux normalization curve uses a polynomial model: Φ(T) / Φ(25℃)=1+a_1·(T-25)+a_2·(T-25)^2; Among them, the temperature coefficients of different LED materials vary significantly, for example: 1. Red LED (AlGaInP): a_1=-0.009, a_2=-0.0001, luminous flux drops to 60% of that at 25℃ at 85℃; 2. Blue LED (InGaN): a_1=-0.004, a_2=-0.00005, the luminous flux drops to 85% of that at 25℃ at 85℃; 3. Blue LED (InGaN): a_1=-0.0035, a_2=-0.00004, temperature characteristics are similar to those of cyan LED; 4. Green LED (InGaN): a_1=-0.005, a_2=-0.00006, temperature sensitivity is between that of cyan and red light; 5. Warm white LED (blue light + yellow phosphor): a_1=-0.003, a_2=-0.00003, with the best temperature stability.

[0094] The system calculates the temperature compensation coefficient α_i for temperature-sensitive channels (especially red and cyan light): α_i(T)=1 / (Φ_i(T) / Φ_i(25℃)) And applied to the drive current: I_drive_i=I_nominal_i*α_i(T) For example, for a red LED at 85°C, α_red = 1 / 0.6 ≈ 1.67, which means that the system will increase the red light driving current by about 67% to compensate for the decrease in light output caused by heat.

[0095] Finally, the system combines PID adjustment and thermal compensation to obtain the compensated PWM duty cycle for each channel.

[0096] Step S12: Determine the initial spectral state and the target spectral state based on mode switching requirements or parameter adjustment requirements.

[0097] Specifically, the system identifies user input requests for mode switching (such as switching from "working mode" to "rest mode") or parameter adjustment (such as adjusting the mEDI value), and determines the current initial spectral state (including the PWM duty cycle of each channel) and the target spectral state (adjusted PWM duty cycle).

[0098] Step S13: Based on the initial spectral state and the target spectral state, apply the temporal dithering and logarithmic gradient algorithm to calculate the spectral parameters of the intermediate transition state and generate an intermediate transition frame sequence.

[0099] Instead of directly transitioning from the initial state to the target state, the system calculates a series of intermediate states to form a smooth transition sequence. Compared to linear gradual transitions, this system employs a logarithmic gradual transition algorithm, ensuring that parameter changes conform to the Weber-Fechner law as perceived by the human eye.

[0100] For any parameter P (which can be brightness, color temperature, or mEDI value), the gradual change value at time t is calculated using the following formula: P(t)=P_start·(P_target / P_start)^(t / T) Where t is the current time point (0≤t≤T), and T is the total transition time, which is usually set to 0.5-2 seconds, depending on the magnitude of the change.

[0101] For subtle changes, the system also applies a time-domain jitter algorithm, which alternates between two similar states at a high frequency between adjacent frames, so that the human eye perceives the average effect of the two, thereby achieving a transition smoothness that exceeds the physical resolution of PWM.

[0102] Specifically, for each required intermediate duty cycle value d_i, if d_i falls between two adjacent PWM quantization levels, a time-domain jitter sequence is generated as follows: 1. Let the two closest quantization levels be d_low and d_high, then d_i = d_low + k * (d_high - d_low), where 0 <k<1; 2. In N consecutive frames, there are k·N frames outputting d_high and (1-k)·N frames outputting d_low; 3. Arrange these frames to distribute them as evenly as possible, avoiding low-frequency modulation; For example, if a 45.3% duty cycle is required, but the PWM resolution can only represent 45% and 46%, then in a 10-frame sequence, 3 frames are arranged for 46% and 7 frames for 45%, with the frame arrangement being [45, 46, 45, 45, 46, 45, 45, 46, 45, 45].

[0103] Step S14: Based on the intermediate transition frame sequence, drive the light source sequentially to achieve a smooth transition from the initial spectral state to the target spectral state.

[0104] Specifically, the system loads the PWM duty cycle values ​​from the intermediate transition frame sequence into the PWM dimming driver according to a frame rate of 60-120Hz, driving the five-channel LED module to emit light. This gradual transition process is imperceptible to the user, avoiding discomfort caused by abrupt changes.

[0105] For adjustments to non-visual effects (such as adjusting only mEDI while keeping the color coordinates unchanged), the system ensures that the changes in color coordinates always remain within the range of the McAdam ellipse, so that users cannot visually perceive changes in light color, and only physiological effects are adjusted.

[0106] Table 1: Comparison of mEDI values ​​and flicker index between the system of the present invention and the traditional dual-channel LED lighting system at different time periods.

[0107] Note: PstLM is a short-term flicker visibility index; the smaller the value, the lower the flicker risk. The ultra-high frequency PWM technology of this invention makes the PstLM value significantly lower than that of traditional systems, and this invention can adjust the mEDI value while maintaining the same color temperature, which traditional systems cannot achieve.

[0108] Through the above steps, the system ultimately outputs an intelligent light environment with a smooth and constant visual effect and a precise dynamic adjustment of the rhythm effect, achieving the goal of healthy lighting with "constant light color and variable rhythm".

[0109] Example 5: Application of Metamerism Adjustment Strategy This embodiment provides a specific example of a metamerism control strategy, detailing how to achieve independent adjustment of rhythmic stimulus values ​​at a fixed color temperature.

[0110] Taking a color temperature of 4000K as an example, the system achieves metamerism adjustment through the following steps: Step 1: Based on the target color temperature of 4000K, calculate the standard CIE color coordinates (x, y) = (0.3805, 0.3769). These coordinates will be used as a constant target for visual control.

[0111] Step 2: Set different rhythmic stimulus targets according to different time periods and needs: Morning high alert mode: target mEDI=380; Standard neutral mode: target mEDI=280 (corresponding to a standard 4000K light source); Evening relaxation mode: target mEDI=180.

[0112] Step 3: For the morning high alert mode (high mEDI), the system performs the following channel adjustments: Increase the proportion of the cyan channel (478nm) to 60%; Reduce the proportion of the blue light channel (450nm) to 30%; Increase the proportion of the red light channel (660nm) to 45% to compensate for the color temperature shift caused by the increase in cyan light; Maintain the proportion of the green light channel (525nm) at 40%; Maintain the warm white channel (2700K) at 35%.

[0113] Step 4: For standard neutral mode, the system performs the following channel adjustments: The cyan channel is set to account for 40%; The Blu-ray channel is set to account for 45%; The red light channel is set to account for 40%; The proportion of green channels is set at 40%; The warm white channel is set to account for 35%.

[0114] Step 5: For the evening relaxation mode (low mEDI), the system performs the following channel adjustments: Reduce the proportion of glaucoma channels to 20%; Increase the proportion of blue light channels to 60%; Increase the proportion of red light channels to 50%; Maintain the green light channel ratio at 40%; Increase the proportion of warm white channels to 40%.

[0115] The actual measured spectral parameters for each mode are shown in the table below: Table 2: Actual measured spectral parameters under each mode

[0116] The spectral power distribution (SPD) of the three modes differs significantly, especially in the energy proportion near 480 nm. However, their visual color coordinates are almost identical, and their color tolerances are all less than 0.5 SDCM, far below the threshold that the human eye can distinguish (approximately 1 SDCM). This confirms that the present invention has successfully achieved precise decoupling between visual parameters and non-visual biological effects.

[0117] Example 6: Application Scenarios This embodiment illustrates specific implementations of the present invention in different application scenarios.

[0118] Scenario 1: "Warm Light High Alertness" Anti-Fatigue Mode for Shift Workers For night shift medical staff or control room operators, the system is set to a "warm color temperature (3000K) + high mEDI" mode. By significantly increasing the ratio of 480nm cyan and red light while reducing conventional 450nm blue light, the lighting environment presents a warm color tone with low glare, reducing psychological stress and eye fatigue. At the physiological level, it provides a melatonin-suppressing effect equivalent to 6000K cool white light, maintaining the staff's high alertness and focus, thus solving the pain point of not being able to achieve both comfortable lighting and alertness at night.

[0119] In practice, the system is set with the following target parameters: color temperature 3000K, illuminance 500 lux, and mEDI value 350 (equivalent to the stimulation intensity of 6000K ordinary white light). The neural network will automatically calculate the optimal combination of the five-channel duty cycles, which typically results in a higher proportion of the 480nm cyan channel, a corresponding increase in the red channel to neutralize the color temperature, and a lower proportion of the 450nm blue channel.

[0120] Scenario 2: A cumulative management system for "light dose" based on natural light deficiency compensation The system integrates a personal light dosimeter (based on wearable device or mobile phone sensor data). When the system detects insufficient natural light intake (Lux·h) during the day, it automatically triggers a "light supplement" mode in the evening (e.g., 5:00 PM - 7:00 PM). This mode subtly increases the mEDI value to the maximum threshold while maintaining the user's preferred indoor color temperature, quickly supplementing the rhythmic stimulation signals missing during the day and preventing circadian rhythm phase delay. Once the preset light dose is reached, it automatically and seamlessly transitions to a low mEDI sleep-aid mode.

[0121] In practice, the system receives the user's light exposure data through an API interface. If it detects that the daytime natural light intake is less than 70% of the target value (e.g., 5000 Lux·h), it activates the compensation mode in the evening. At this time, the system will increase the mEDI by 40% while maintaining the user-set color temperature (e.g., 4000K) until the cumulative light dose reaches the target or for a maximum of 2 hours, and then smoothly transition to the standard night mode.

[0122] Scenario 3: Intervention program for "Sunset Syndrome" in Alzheimer's patients For elderly individuals with cognitive impairment, the system utilizes AI to identify periods of abnormal agitation (typically at dusk). During these periods, the system generates a specific "40Hz gamma banding spectrum," which involves subtly modulating a high mEDI component at a frequency of 40Hz, imperceptible to the human eye, while maintaining stable background lighting. This approach aims to improve patients' cognitive status and alleviate anxiety during dusk by inducing gamma oscillations in the visual pathway, while simultaneously enhancing circadian rhythm synchronization through high mEDI to improve nighttime sleep quality.

[0123] In practice, the system sets the light source to a warm and comfortable color temperature of 2700K between 16:00 and 19:00, while maintaining a relatively high mEDI value (approximately 250). Simultaneously, a special PWM modulation mode is activated, causing the cyanosis channel to fluctuate within ±10% at a frequency of 40Hz, while other channels are adjusted accordingly to maintain constant color coordinates. This mode combines cognitive intervention and rhythm regulation functions, and considering the yellowing of the lens in older adults, the system automatically applies a transmittance compensation coefficient of 1.8 times based on the user's age (e.g., 85 years old).

[0124] Through the detailed description of the above embodiments, the AI ​​dynamic spectral lighting system and control method for simulating natural light rhythms provided by the present invention achieves the goal of precisely regulating non-visual biological rhythm effects while maintaining constant visual color perception. This system fully considers individual user differences, environmental factors, and application scenario requirements, and through AI deep learning and multi-channel LED mixing technology, provides different users with a visually comfortable and physiologically healthy intelligent light environment in different scenarios.

[0125] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. An AI dynamic spectral illumination system and control method simulating natural light rhythms, characterized in that, Includes the following steps: Based on static parameters input by the user and real-time parameters collected by environmental sensors, the equivalent illuminance of the target melanin is calculated through a biological rhythm model and a lens transmittance correction coefficient is introduced for compensation. A dual target vector is constructed by combining the target chromaticity parameters to generate a dual target parameter set containing visual chromaticity coordinates and rhythmic stimulus values. Based on the dual target parameter set, it is input into the pre-trained spectral-chromaticity coupled deep neural network. Optical quality constraint judgment is performed through the hidden layer constraint mechanism, and the energy ratio of the 480nm band is adjusted while maintaining the constant chromaticity coordinates through the 480nm cyan channel adjustment strategy, and the five-channel optimal driving current duty cycle signal is output. Based on the five-channel optimal drive current duty cycle signal, the five-channel LED module is driven by a high-frequency PWM dimming driver to perform light and color mixing. The actual spectral power distribution and color coordinates are collected in real time by a spectrometer. The actual color temperature, brightness and black pixel equivalent illuminance value are calculated to obtain the current actual light environment parameters. Based on the deviation between the current actual light environment parameters and the dual target parameter set, the PID closed-loop feedback algorithm is used to fine-tune the PWM duty cycle of each channel, and the compensation coefficients of the red and cyan light channels are adjusted to address the wavelength drift caused by LED thermal effects. The time-domain jitter and logarithmic gradient algorithms are applied to generate an intermediate transition frame sequence and drive the light source in sequence, resulting in an intelligent light environment with smooth and constant visual effects and precise dynamic adjustment of rhythm effects.

2. The AI ​​dynamic spectral illumination control method for simulating natural light rhythms according to claim 1, characterized in that, Based on static parameters input by the user and real-time parameters collected by environmental sensors, the equivalent illuminance of the target melanin is calculated using a biorhythm model, and a lens transmittance correction coefficient is introduced for compensation. A dual target vector is constructed by combining the target chromaticity parameters, generating a dual target parameter set containing visual chromatic coordinates and rhythmic stimulus values, including: Based on static parameters input by the user and real-time parameters collected by environmental sensors, the equivalent illuminance of the target melanops is calculated through a circadian rhythm model deployed in the cloud or locally. Based on the user's age parameter, the lens transmittance correction coefficient is calculated according to the lens spectral transmittance formula, and the equivalent illuminance of the target melanin is compensated to obtain the compensated equivalent illuminance of the target melanin. Based on the user-defined target chromaticity parameters or scene mode, and combined with the compensated target black visual equivalent illuminance, a dual target vector is constructed to generate a dual target parameter set containing visual chromaticity coordinates and rhythmic stimulus values.

3. The AI ​​dynamic spectral illumination control method for simulating natural light rhythms according to claim 2, characterized in that, Based on the user's age parameter, a lens transmittance correction coefficient is calculated according to the lens spectral transmittance formula. This correction coefficient is then used to compensate for the equivalent illuminance of the target melanin, resulting in the compensated equivalent illuminance of the target melanin, including: Based on the user's age parameter, the user's lens transmittance is calculated according to the lens spectral transmittance formula; Based on the lens transmittance of the user and the lens transmittance of the standard observer, a lens transmittance correction factor is calculated. Based on the lens transmittance correction coefficient, the equivalent illuminance of the target melanocyte is compensated and calculated to generate the compensated equivalent illuminance of the target melanocyte.

4. The AI ​​dynamic spectral illumination control method for simulating natural light rhythms according to claim 1, characterized in that, The pre-trained spectral-chromaticity coupled deep neural network is obtained through the following steps: A basic spectral library was established based on the spectral power distribution of a five-channel LED under different driving currents and operating temperatures. Based on the aforementioned basic spectral library, the Monte Carlo mixing simulation method is used to randomly generate a combination of duty cycles for five channels, and the mixed spectrum is synthesized according to the principle of spectral superposition. Based on the mixed spectrum, the corresponding color coordinates, illuminance, color rendering index and melanopsine equivalent illuminance are calculated, and effective samples that meet the requirements of color rendering index and color tolerance are screened to construct a training set. Based on the training set, a lightweight fully connected regression network is trained to obtain the pre-trained spectral-chromaticity coupled deep neural network.

5. The AI ​​dynamic spectral illumination control method for simulating natural light rhythms according to claim 1, characterized in that, Optical quality constraint determination is performed through a hidden layer constraint mechanism, including: Based on the hidden layer constraint mechanism of the deep neural network, the color rendering index and color tolerance of the current spectral combination are calculated. Based on the color rendering index and the color tolerance, determine whether the optical quality constraints are met. If the color rendering index is less than 90 or the color tolerance is greater than 2 SDCM, the penalty mechanism of the loss function is triggered, forcing the deep neural network to re-optimize in the metamerism solution space until it outputs a five-channel optimal drive current duty cycle signal that satisfies the optical quality constraints.

6. The AI ​​dynamic spectral illumination control method for simulating natural light rhythms according to claim 1, characterized in that, The energy proportion of the 480nm band is adjusted by using a 480nm cyan channel modulation strategy while maintaining constant color coordinates. This includes: Based on the deep neural network, by increasing or decreasing the weight of the 480nm cyan channel and correspondingly adjusting the duty cycle of the blue and green channels, the energy proportion of the 480nm band in the spectrum is adjusted while keeping the mixed color coordinates constant within the McAdam ellipse range, and the adjusted five-channel optimal drive current duty cycle signal is output.

7. The AI ​​dynamic spectral illumination control method for simulating natural light rhythms according to claim 1, characterized in that, The five-channel LED module includes a red light channel, a green light channel, a blue light channel, a cyan light channel, and a warm white light channel. The peak wavelength of the cyan light channel is 478nm to 482nm, the peak wavelength of the red light channel is 660nm, the peak wavelength of the green light channel is 525nm, the peak wavelength of the blue light channel is 450nm, and the correlated color temperature of the warm white light channel is 2700K.

8. The AI ​​dynamic spectral illumination control method for simulating natural light rhythms according to claim 1, characterized in that, The high-frequency PWM dimming driver uses hybrid digital-analog dimming technology to drive the five-channel LED module. In the high brightness range, a constant current amplitude modulation method is used to directly adjust the reference voltage of the DC-DC converter to change the driving current amplitude and drive the five-channel LED module. In the low brightness range or high-precision color mixing range, an ultra-high frequency PWM modulation method is adopted, with the PWM frequency set from 4kHz to 10kHz. A 16-bit PWM generator is used to provide 65,536 levels of grayscale adjustment capability to drive the five-channel LED module.

9. The AI ​​dynamic spectral illumination control method for simulating natural light rhythms according to claim 1, characterized in that, The PID closed-loop feedback algorithm is used to fine-tune the PWM duty cycle of each channel, and the compensation coefficients of the red and cyan channels are adjusted to address the wavelength drift caused by LED thermal effects, including: Based on the current actual light environment parameters and the dual target parameter set, calculate the color coordinate drift and rhythmic stimulation error. Based on the color coordinate drift and the rhythmic stimulation error, the PWM duty cycle adjustment of each channel is calculated using a PID closed-loop feedback algorithm. The PWM duty cycle of each channel is fine-tuned based on the PWM duty cycle adjustment amount. Based on the LED module temperature and temperature-luminous flux normalization curve collected by the NTC thermistor, the temperature compensation coefficients of the red and cyan channels are calculated, and the drive current is dynamically adjusted to obtain the compensated PWM duty cycle.

10. An AI dynamic spectrum illumination system simulating natural light rhythms, comprising a memory, a processor, and a five-channel light-emitting diode module, characterized in that, The processor executes the AI ​​dynamic spectral illumination control method for simulating natural light rhythms as described in any one of claims 1 to 9.

Images