High color rendering index lighting method and system based on multi-group LED adjustable spectrum

Abstract

The invention provides a high color rendering index lighting method and system based on multi-group LED adjustable spectrums, and relates to the technical field of intelligent lighting, and the method comprises the steps: obtaining lighting environment parameters and spectral space distribution characteristics, building a coupling relation equation of LED chip temperature and spectral characteristics, solving the equation through a deep learning optimization algorithm based on a multi-attention mechanism, and obtaining a high color rendering index lighting result. And constructing a multi-target loss function, updating equation parameters through a temperature self-adaptive online learning mode, obtaining real-time driving parameters of multiple groups of LEDs, generating a PWM driving signal sequence, and outputting a target mixed spectrum. According to the invention, high color rendering index illumination can be realized, the spectrum regulation and control precision and energy efficiency are improved, and different illumination environment requirements are met.

Description

Technical Field

[0001] The present invention relates to intelligent lighting technology, and in particular to a high color rendering index lighting method and system based on multiple groups of LEDs with adjustable spectra. Background Art

[0002] High color rendering index lighting not only accurately reproduces the true color of objects but also provides a comfortable visual experience. In recent years, tunable spectrum lighting technology based on multiple LED groups has received widespread attention. By adjusting the output intensity of LEDs with different wavelengths, flexible spectrum synthesis and high color rendering index lighting can be achieved.

[0003] However, existing multi-LED spectral controllable lighting technologies still face several challenges and limitations. First, most systems fail to fully consider the complex coupling relationship between LED chip temperature and spectral characteristics, resulting in unstable spectral output and difficulty in achieving precise control in practical applications. Second, existing technologies often use simple linear optimization algorithms to solve driving parameters, which struggles to address complex nonlinear spectral synthesis problems, especially when high color rendering index requirements are required. Finally, existing systems lack comprehensive consideration of the spatial distribution of the spectrum, making it difficult to ensure lighting uniformity and performing poorly in large-space lighting applications. Summary of the Invention

[0004] The embodiments of the present invention provide a high color rendering index lighting method and system based on multiple groups of LEDs with adjustable spectrum, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention, Acquire lighting environment parameters, including ambient brightness, ambient color temperature, and ambient color rendering index; use a spectral sensor to collect spectral data at different locations in the lighting space to obtain spectral spatial distribution characteristics; Based on the lighting environment parameters and the spectral spatial distribution characteristics, a coupling relationship equation between the LED chip temperature and the spectral characteristics is established. The coupling relationship equation includes a thermal resistance matrix between the LED arrays and a temperature-related quantum efficiency. The thermal resistance matrix is ​​used to calculate the real-time temperature of each LED chip, and the quantum efficiency is used to characterize the effect of temperature on the LED luminous efficiency. A deep learning optimization algorithm based on a multi-attention mechanism is used to solve the coupling relationship equation, construct a multi-objective loss function, and update the coupling relationship equation parameters through temperature-adaptive online learning to obtain real-time driving parameters of multiple groups of LEDs. The multi-objective loss function includes a color rendering index error term, a color temperature error term, a uniformity error term, and an energy efficiency error term. A PWM drive signal sequence is generated according to the real-time drive parameters, and a target mixed spectrum is output.

[0006] In an optional embodiment, Arranging a plurality of spectral sensors whose spatial coordinates satisfy uniform distribution on a hemispherical surface in the lighting space, collecting spectral data to obtain a spectral matrix, wherein the spectral matrix includes four dimensions: wavelength, radial distance, zenith angle, and azimuth angle; Get the simplified BRDF model parameters of the environment reflection surface, the simplified BRDF model is: ; where ρ d is the diffuse reflection coefficient, ρ s is the mirror reflection coefficient, α is the angle attenuation factor, θ i is the angle of incidence, θ r is the reflection angle, λ is the wavelength; Based on the spectral matrix and the simplified BRDF model parameters, the spectral power distribution, spatial intensity distribution and angular dependence are calculated, and a comprehensive feature vector including spectral moment characteristics, spatial correlation characteristics, uniformity characteristics and angular sensitivity is constructed; principal component analysis and feature reconstruction are performed on the comprehensive feature vector to obtain the spectral spatial distribution characteristics.

[0007] In an optional embodiment, Based on the lighting environment parameters and spectral spatial distribution characteristics, a coupling relationship equation between the LED chip temperature and spectral characteristics is established. The coupling relationship equation includes a thermal resistance matrix between LED arrays and a temperature-related quantum efficiency. The thermal resistance matrix is ​​used to calculate the real-time temperature of each LED chip. The quantum efficiency is used to characterize the effect of temperature on the LED luminous efficiency, including: Establishing a thermal coupling relationship for the LED array, describing the thermal coupling coefficient between the LED chips using a thermal resistance matrix, wherein the thermal coupling coefficient includes an ambient temperature term, a power consumption term, and an environmental impact term, wherein the environmental impact term is determined by color temperature and illumination, and calculating the junction temperature of each LED chip based on the elements of the thermal resistance matrix; Establish a temperature-illuminance coupling relationship for LEDs, including a temperature coefficient term and an illuminance influence term. The temperature coefficient term characterizes the attenuation characteristics of the LED luminous efficiency with temperature, and the illuminance influence term characterizes the correction effect of ambient illumination on quantum efficiency. Use an exponential function to describe the dynamic change of the LED luminous efficiency from the initial state to the steady state. The junction temperature output of the thermal coupling relationship and the efficiency calculation results of the temperature-illuminance coupling relationship are used as input parameters. The spectral characteristics of a single LED are described by a temperature-dependent Gaussian function. The spectra of multiple LEDs are weightedly superimposed based on the weight coefficients of junction temperature and ambient illumination, and the temperature-spectrum coupling relationship equation of the LED array is established.

[0008] In an optional embodiment, The spectral characteristics of a single LED are described by a temperature-dependent Gaussian function. The spectra of multiple LEDs are weighted and superimposed based on the weight coefficients of junction temperature and ambient illumination. The temperature-spectrum coupling relationship equation of the LED array is established, including: A temperature-dependent Gaussian function is used to describe the LED spectral characteristics. The Gaussian function includes a peak intensity term, a center wavelength term, and a standard deviation term. The peak intensity term and the center wavelength term are updated using a quadratic polynomial of the temperature variation, and the standard deviation term is updated using an exponential function. The update coefficients of the peak intensity term and the center wavelength term have a nonlinear relationship with the temperature variation and the ambient illumination, and the update coefficient of the standard deviation term has an exponential decay relationship with the temperature variation. Performing nonlinear superposition of the spectra of multiple LEDs, including a weighted term and a spectral interference term. The weighted term calculates the contribution weight of each LED using a composite weighting method based on junction temperature deviation and ambient illumination, where the ambient illumination affects the baseline value of the weight coefficient through a logarithmic function. The spectral interference term characterizes the spectral interaction between LEDs through a temperature-dependent interference coefficient, which decays exponentially with junction temperature. Based on the parameters of the Gaussian function and the coefficients of the nonlinear superposition, a complete coupling relationship equation is constructed.

[0009] In an optional embodiment, A deep learning optimization algorithm based on a multi-attention mechanism is used to solve the coupling relationship equation, a multi-objective loss function is constructed, and the coupling relationship equation parameters are updated through temperature-adaptive online learning to obtain real-time driving parameters of multiple groups of LEDs. The multi-objective loss function includes color rendering index error term, color temperature error term, uniformity error term, and energy efficiency error term. Constructing an input feature layer to perform feature encoding on the parameters of the coupling relationship equation; A multi-scale feature extraction network is established based on the feature encoding, and the dynamic temperature change characteristics are captured by the time attention module. The time attention module uses the product operation of the query matrix, the key matrix and the value matrix to obtain the time dimension weight; the spectral spatial distribution characteristics are extracted by the spatial attention module. The spatial attention module performs convolution operation after performing average pooling and maximum pooling on the feature map; and the channel attention module realizes adaptive weighting between feature channels; A bidirectional gated recurrent unit network is used to predict the real-time temperature of the LED chip; A temperature-sensitive adaptive multi-objective loss function is constructed, including a color rendering index error term based on spectral similarity, a dynamically weighted color temperature error term, a uniformity error term, and a temperature-sensitive energy efficiency error term. The color rendering index error term uses KL divergence to measure spectral distribution similarity. The weight coefficient of the color temperature error term is dynamically adjusted with temperature. The uniformity error term evaluates illumination uniformity at multiple spatial scales. The energy efficiency error term introduces a temperature-dependent weight function. An online learning strategy is designed based on the temperature state. The learning rate is dynamically adjusted through a temperature-adaptive learning rate regulation mechanism, which includes an exponential decay term for the temperature change. The sample importance weight is determined according to the deviation between the sample temperature and the average temperature. The parameters of the coupling relationship equation are optimized by adaptive adjustment of the momentum term based on the temperature gradient and gradient clipping under the temperature threshold, and the real-time driving parameters of multiple groups of LEDs are output.

[0010] In an optional embodiment, Constructing an adaptive multi-objective loss function includes: An adaptive weight function is constructed to weight the color rendering index error term, the color temperature error term, the uniformity error term, and the energy efficiency error term, wherein the adaptive weight function includes: a basic weight term for setting an initial weight of each error term; a temperature adjustment factor that is linearly related to the temperature when the temperature exceeds the temperature threshold; and a time adjustment factor that is dynamically updated during the optimization iteration process; The gradient of the loss function weighted by the adaptive weight function is calculated, and the product of the gradient value of each loss term and the temperature-related weight function in the energy efficiency error term is used as the target importance index to characterize the optimization priority of each error term under different temperature conditions; A dynamic balance strategy is adopted to update the weights of the adaptive multi-objective loss function under different temperature states, including: according to the deviation between the target importance index and the current weight and the loss change, combined with the learning rate adjustment mechanism and the momentum term based on the temperature gradient, the weight parameters of each error term are iteratively optimized to achieve dynamic balance of the weights of each loss function.

[0011] In an optional embodiment, Generating a PWM drive signal sequence according to the real-time drive parameters and outputting a target mixed spectrum includes: A discrete wavelet transform is performed on the real-time drive parameters to decompose the parameter signal into multiple frequency band sub-signals; a complexity coefficient is determined by calculating the energy entropy value and peak ratio of each frequency band, and the frequency bands are dynamically grouped into low-frequency, medium-frequency and high-frequency regions according to the complexity coefficient; a different quantization strategy is applied to each group of frequency band parameters, a PWM signal of a first current magnitude is used for the low-frequency group, a PWM signal of a second current magnitude is used for the medium-frequency group, and a PWM signal of a third current magnitude is used for the high-frequency group, wherein the first current magnitude is greater than the second current magnitude, and the second current magnitude is greater than the third current magnitude; and a multi-band control signal is recombined into a hybrid PWM drive sequence through a feedback-corrected wavelet reconstruction algorithm; A database of corresponding relationships between LED current and spectral output is established, the expected spectral output of the hybrid PWM drive sequence is calculated, and the error vector between the expected spectral output and the target spectrum is obtained. The correction values ​​of the PWM parameters are determined through an error back-propagation calculation method, and the duty cycle, current magnitude, and pulse edge shape of the hybrid PWM drive sequence are adjusted. Pulse density modulation and PWM modulation are combined to adjust the distribution density of the pulses in the time window, compensate for the nonlinear response characteristics of the LED, and output the target hybrid spectrum.

[0012] According to a second aspect of the embodiments of the present invention, Provides a high color rendering index lighting system based on multiple groups of LED adjustable spectrum, including: The first unit is configured to obtain lighting environment parameters, including ambient brightness, ambient color temperature, and ambient color rendering index; and to collect spectral data at different locations in the lighting space using a spectral sensor to obtain spectral spatial distribution characteristics. A second unit is configured to establish a coupled relationship equation between LED chip temperature and spectral characteristics based on the lighting environment parameters and spectral spatial distribution characteristics, wherein the coupled relationship equation includes a thermal resistance matrix between LED arrays and a temperature-related quantum efficiency, wherein the thermal resistance matrix is ​​used to calculate the real-time temperature of each LED chip, and the quantum efficiency is used to characterize the effect of temperature on the luminous efficiency of the LED; The third unit is used to solve the coupling relationship equation using a deep learning optimization algorithm based on a multi-attention mechanism, construct a multi-objective loss function, and update the coupling relationship equation parameters through temperature-adaptive online learning to obtain real-time driving parameters of multiple groups of LEDs. The multi-objective loss function includes a color rendering index error term, a color temperature error term, a uniformity error term, and an energy efficiency error term. The fourth unit is configured to generate a PWM drive signal sequence according to the real-time drive parameters and output a target mixed spectrum.

[0013] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0014] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0015] By acquiring lighting environment parameters and spectral spatial distribution characteristics, the present invention establishes a coupling relationship equation between LED chip temperature and spectral characteristics, which can accurately describe the thermal and optical characteristics of the LED array and lay the foundation for achieving high-precision spectral control.

[0016] The present invention adopts a deep learning optimization algorithm based on a multiple attention mechanism to solve the coupling relationship equation, constructs a multi-objective loss function, and updates the parameters through a temperature-adaptive online learning method, which can realize real-time optimization of LED driving parameters and effectively improve the accuracy and stability of spectral regulation.

[0017] The present invention generates a PWM drive signal sequence based on optimized real-time drive parameters, can output a target mixed spectrum, achieve LED lighting effects with high color rendering index, high color temperature accuracy, high uniformity, and high energy efficiency, and meet high-quality lighting needs in different scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the process of a high color rendering index lighting method based on multiple groups of LEDs with adjustable spectrum according to an embodiment of the present invention; Figure 2 This is a diagram of the temperature adaptive multi-objective loss function weight dynamic adjustment mechanism of the present invention; Figure 3 This is a temperature adaptive optimization diagram of LED driving parameters of the multiple attention mechanism of the present invention. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0020] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0021] Figure 1 FIG. 1 is a flow chart of a high color rendering index lighting method based on multiple groups of LEDs with adjustable spectrum according to an embodiment of the present invention. Figure 1 As shown, the method includes: Acquire lighting environment parameters, including ambient brightness, ambient color temperature, and ambient color rendering index; use a spectral sensor to collect spectral data at different locations in the lighting space to obtain spectral spatial distribution characteristics; Based on the lighting environment parameters and the spectral spatial distribution characteristics, a coupling relationship equation between the LED chip temperature and the spectral characteristics is established. The coupling relationship equation includes a thermal resistance matrix between the LED arrays and a temperature-related quantum efficiency. The thermal resistance matrix is ​​used to calculate the real-time temperature of each LED chip, and the quantum efficiency is used to characterize the effect of temperature on the LED luminous efficiency. A deep learning optimization algorithm based on a multi-attention mechanism is used to solve the coupling relationship equation, construct a multi-objective loss function, and update the coupling relationship equation parameters through temperature-adaptive online learning to obtain real-time driving parameters of multiple groups of LEDs. The multi-objective loss function includes a color rendering index error term, a color temperature error term, a uniformity error term, and an energy efficiency error term. A PWM drive signal sequence is generated according to the real-time drive parameters, and a target mixed spectrum is output.

[0022] In an optional embodiment, using a spectral sensor to collect spectral data at different locations in the lighting space to obtain spectral spatial distribution characteristics includes: Arranging a plurality of spectral sensors whose spatial coordinates satisfy uniform distribution on a hemispherical surface in the lighting space, collecting spectral data to obtain a spectral matrix, wherein the spectral matrix includes four dimensions: wavelength, radial distance, zenith angle, and azimuth angle; Get the simplified BRDF model parameters of the environment reflection surface, the simplified BRDF model is: ; where ρ d is the diffuse reflection coefficient, ρ s is the mirror reflection coefficient, α is the angle attenuation factor, θ i is the angle of incidence, θ r is the reflection angle, λ is the wavelength; Based on the spectral matrix and the simplified BRDF model parameters, the spectral power distribution, spatial intensity distribution and angular dependence are calculated, and a comprehensive feature vector including spectral moment characteristics, spatial correlation characteristics, uniformity characteristics and angular sensitivity is constructed; principal component analysis and feature reconstruction are performed on the comprehensive feature vector to obtain the spectral spatial distribution characteristics.

[0023] For example, multiple spectral sensors are arranged in the lighting space. The spatial coordinates of these sensors are evenly distributed across a hemisphere. These spectral sensors are used to collect spectral data, generating a spectral matrix. This spectral matrix contains four dimensions: wavelength, radial distance, zenith angle, and azimuth angle. For example, within the visible light range of 380-780 nm, a data point can be collected every 2 nm, for a total of 201 wavelength points. The radial distance can be set to 0.5 m, 1 m, 1.5 m, and 2 m. This results in a four-dimensional spectral matrix of 201 × 4 × 4 × 8.

[0024] The simplified BRDF model for environmental reflective surfaces includes three parameters: diffuse reflectance, specular reflectance, and angular attenuation. Based on the acquired spectral matrix and BRDF model parameters, the spectral power distribution, spatial intensity distribution, and angular dependence are calculated. The spectral power distribution can be obtained by summing the spectral matrix over the wavelength dimension. The spatial intensity distribution can be obtained by averaging the spectral matrix over the radial distance, zenith angle, and azimuth angle dimensions. Angular dependence can be characterized by calculating the rate of change of intensity at different angles.

[0025] The comprehensive feature vector includes spectral moment features, spatial correlation features, uniformity features, and angular sensitivity. Spectral moment features can be selected from the spectrum's mean, standard deviation, skewness, and kurtosis. Spatial correlation features can be characterized by the spatial autocorrelation function. Uniformity features can be expressed by the coefficient of variation. Angular sensitivity can be expressed as the slope of the intensity versus angle. Principal component analysis and feature reconstruction are performed on the comprehensive feature vector to extract the main characteristic components and obtain the final spectral spatial distribution characteristics. Principal components that explain 95% of the variance are retained, typically 3-5. These principal components are the key features of the spectral spatial distribution.

[0026] The present invention calculates the spectral power distribution, spatial intensity distribution and angular dependence, and constructs a comprehensive feature vector including spectral moment characteristics, spatial correlation characteristics, uniformity characteristics and angular sensitivity, thereby realizing a multi-dimensional characterization of the spectral characteristics of the lighting space and improving the integrity and accuracy of spectral feature extraction.

[0027] In an optional embodiment, based on the lighting environment parameters and the spectral spatial distribution characteristics, a coupling relationship equation between the LED chip temperature and the spectral characteristics is established. The coupling relationship equation includes a thermal resistance matrix between the LED arrays and a temperature-related quantum efficiency. The thermal resistance matrix is ​​used to calculate the real-time temperature of each LED chip. The quantum efficiency is used to characterize the effect of temperature on the LED luminous efficiency and includes: Establishing a thermal coupling relationship for the LED array, describing the thermal coupling coefficient between the LED chips using a thermal resistance matrix, wherein the thermal coupling coefficient includes an ambient temperature term, a power consumption term, and an environmental impact term, wherein the environmental impact term is determined by color temperature and illumination, and calculating the junction temperature of each LED chip based on the elements of the thermal resistance matrix; Establish a temperature-illuminance coupling relationship for LEDs, including a temperature coefficient term and an illuminance influence term. The temperature coefficient term characterizes the attenuation characteristics of the LED luminous efficiency with temperature, and the illuminance influence term characterizes the correction effect of ambient illumination on quantum efficiency. Use an exponential function to describe the dynamic change of the LED luminous efficiency from the initial state to the steady state. The junction temperature output of the thermal coupling relationship and the efficiency calculation results of the temperature-illuminance coupling relationship are used as input parameters. The spectral characteristics of a single LED are described by a temperature-dependent Gaussian function. The spectra of multiple LEDs are weightedly superimposed based on the weight coefficients of junction temperature and ambient illumination, and the temperature-spectrum coupling relationship equation of the LED array is established.

[0028] For example, the thermal coupling relationship of an LED array is constructed, and the thermal coupling coefficient between LED chips is described using a thermal resistance matrix. This thermal coupling coefficient consists of three components: ambient temperature, power consumption, and environmental impact. The ambient temperature term reflects the direct impact of the current ambient temperature on the LED chip temperature. For example, at an ambient temperature of 25°C, the ambient temperature term takes a value of 1.0. The power consumption term represents the relationship between the input power of the LED chip and the temperature rise. At an input power of 1W, the power consumption coefficient is approximately 0.3°C / W. The environmental impact term is determined by both color temperature and illumination. When the color temperature is 6500K and the illumination is 500lx, the environmental impact term takes a value of approximately 0.8. Taking a 4×4 LED array as an example, the thermal resistance matrix is ​​a 16×16 symmetric matrix. The diagonal elements represent the thermal resistance of the chip itself, ranging from 15-20°C / W. The off-diagonal elements represent the mutual thermal resistance between chips, typically 0.1-0.3 times the thermal resistance of the chip itself. Based on the element values ​​of the thermal resistance matrix, the real-time junction temperature of each LED chip can be calculated.

[0029] The temperature-illuminance coupling relationship of LEDs consists of two main components: the temperature coefficient term and the illuminance effect term. The temperature coefficient term characterizes the attenuation of LED luminous efficiency with temperature. Within the temperature range of 25-85°C, the luminous efficiency decreases by approximately 2-3% for every 10°C increase. The illuminance effect term characterizes the correction effect of ambient illuminance on quantum efficiency. For every 100lx increase in ambient illuminance, the quantum efficiency correction factor increases by approximately 0.5%. The dynamic change of LED luminous efficiency from initial state to steady state is described by an exponential function, with a time constant typically in the range of 100-200ms.

[0030] The efficiency calculation results of the junction temperature output and temperature-illuminance coupling relationship obtained from the above thermal coupling relationship are used as input parameters. A temperature-dependent Gaussian function is used to describe the spectral characteristics of a single LED. The weight coefficient is determined based on the junction temperature and ambient illumination, and the spectra of multiple LEDs are weighted and superimposed to establish the coupling relationship equation.

[0031] The present invention establishes the thermal coupling relationship and temperature-illuminance coupling relationship of the LED array and organically combines parameters such as the thermal resistance matrix, ambient temperature term, power consumption term and environmental influence term to achieve an accurate description of the coupling relationship between the temperature and spectral characteristics of the LED chip, providing a reliable theoretical basis for spectral optimization control. By introducing the influence of ambient illumination and establishing the temperature coefficient term and the illumination influence term, the coupling relationship equation can accurately reflect the dynamic change process of the LED luminous efficiency, significantly improving the accuracy and adaptability of the model in the actual application environment, and effectively solving the spectral deviation problem caused by the traditional method ignoring environmental factors. The coupling relationship equation is established in a hierarchical construction manner, which not only achieves an accurate description of the spectral characteristics of a single LED, but also realizes the accurate superposition of multiple LED spectra through weight coefficients, making the model have good scalability.

[0032] In an optional embodiment, the spectral characteristics of a single LED are described by a temperature-dependent Gaussian function, and the spectra of multiple LEDs are weighted and superimposed based on the weight coefficients of junction temperature and ambient illumination. The temperature-spectrum coupling relationship equation of the LED array is established, including: A temperature-dependent Gaussian function is used to describe the LED spectral characteristics. The Gaussian function includes a peak intensity term, a center wavelength term, and a standard deviation term. The peak intensity term and the center wavelength term are updated using a quadratic polynomial of the temperature variation, and the standard deviation term is updated using an exponential function. The update coefficients of the peak intensity term and the center wavelength term have a nonlinear relationship with the temperature variation and the ambient illumination, and the update coefficient of the standard deviation term has an exponential decay relationship with the temperature variation. The spectra of multiple LEDs are nonlinearly superimposed, including a weighted term and a spectral interference term. The weighted term calculates the contribution weight of each LED using a composite weighting method based on junction temperature deviation and ambient illumination, wherein the ambient illumination affects the baseline value of the weight coefficient through a logarithmic function. The spectral interference term characterizes the spectral interaction between LEDs through a temperature-dependent interference coefficient, which decays exponentially with junction temperature. A complete coupling relationship equation is constructed based on the parameters of the Gaussian function and the coefficient of the nonlinear superposition.

[0033] Exemplarily, a temperature-dependent Gaussian function is used to describe the spectral characteristics of a single LED. Taking a blue LED as an example, at a reference temperature of 25°C, the peak intensity is 1.0, the central wavelength is 455nm, and the half-width is 20nm. When the temperature changes, the peak intensity term and the central wavelength term are updated by a quadratic polynomial of the temperature change. Specifically, the peak intensity decreases by about 3.2% for every 10°C increase in temperature. In the range of 25-85°C, the rate of decrease accelerates with increasing temperature, and the rate of decrease reaches 4.5% / 10°C above 75°C. The central wavelength redshifts with increasing temperature, and the redshift is about 1.5nm for every 10°C increase. The amount of redshift increases slightly in the high temperature region, and the amount of redshift reaches 1.8nm / 10°C at 85°C.

[0034] The standard deviation term is updated using an exponential function to reflect the spectral broadening effect. Within the 25-85°C range, the standard deviation increases exponentially with temperature, initially at 0.8nm / 10°C. This rate decreases as the temperature rises, dropping to 0.5nm / 10°C above 75°C. The effect of ambient illumination on peak intensity and central wavelength is reflected through nonlinear correction factors. For every 200lx increase in illumination within the 0-1000lx range, the peak intensity correction factor increases by 0.01, and the central wavelength correction factor increases by 0.002.

[0035] For lighting systems composed of red, green, and blue LEDs, the spectra of multiple LEDs must be nonlinearly superimposed. Weighting terms are weighted using a composite method based on junction temperature deviation and ambient illuminance. Taking the standard operating temperature of 45°C as a benchmark, when the junction temperature of an LED deviates from the benchmark temperature, its weight coefficient decreases by 0.02 for every 5°C increase. Ambient illuminance influences the baseline value of the weight coefficient through a logarithmic function. As illuminance increases from 100 to 1000 lx, the baseline weight value increases from 0.8 to 1.0, with the rate of increase gradually decreasing as illuminance increases.

[0036] The spectral interference term characterizes the spectral interaction between LEDs. For example, in the spectral overlap region of a red and blue LED, interference occurs. The initial interference coefficient is 0.05 and decays exponentially with increasing junction temperature. As the junction temperature rises from 25°C to 85°C, the interference coefficient decreases from 0.05 to 0.01, and the decay rate significantly decreases above 65°C.

[0037] In practical applications, this approach was demonstrated using three-primary-color LEDs as an example. Initially, the ambient temperature was 25°C, the illumination was 300 lx, and the drive current for all three LEDs was 350 mA. After 10 minutes of operation, the junction temperature of the red LED rose to 58°C, resulting in a 12% decrease in peak intensity, a 4.5nm redshift in the center wavelength, and a 2.8nm increase in the full width at half maximum. The junction temperature of the green LED rose to 52°C, with a 9% decrease in peak intensity, a 3.8nm redshift in the center wavelength, and a 2.3nm increase in the full width at half maximum. The junction temperature of the blue LED rose to 48°C, with a 7% decrease in peak intensity, a 3.2nm redshift in the center wavelength, and a 2.0nm increase in the full width at half maximum. Based on these parameter variations, combined with the weight coefficients for each LED (0.28 for the red LED, 0.33 for the green LED, and 0.39 for the blue LED) and the spectral interference term, a complete coupling equation was constructed. This equation accurately predicts the spectral output characteristics of the LED array under different operating conditions, providing a basis for precise spectral control.

[0038] This method describes LED spectral characteristics using a temperature-dependent Gaussian function and updates various parameters using quadratic polynomials and exponential functions. This accurately describes the temperature-dependent changes in LED spectral characteristics and significantly improves the accuracy of spectral prediction. By introducing a composite weighting method based on junction temperature deviation and ambient illumination, combined with spectral interference terms, the spectra of multiple LEDs are nonlinearly superimposed, effectively addressing the problem of traditional linear superposition methods that ignores interactions between LEDs. A hierarchical parameter updating strategy enables the model to accurately reflect the dynamic changes in LED spectral characteristics.

[0039] In an optional embodiment, a deep learning optimization algorithm based on a multi-attention mechanism is used to solve the coupling relationship equation, a multi-objective loss function is constructed, and the coupling relationship equation parameters are updated through temperature-adaptive online learning to obtain real-time driving parameters of multiple groups of LEDs. The multi-objective loss function includes a color rendering index error term, a color temperature error term, a uniformity error term, and an energy efficiency error term, including: Constructing an input feature layer to perform feature encoding on the parameters of the coupling relationship equation; A multi-scale feature extraction network is established based on the feature encoding, and the dynamic temperature change characteristics are captured by the time attention module. The time attention module uses the product operation of the query matrix, the key matrix and the value matrix to obtain the time dimension weight; the spectral spatial distribution characteristics are extracted by the spatial attention module. The spatial attention module performs convolution operation after performing average pooling and maximum pooling on the feature map; and the channel attention module realizes adaptive weighting between feature channels; A bidirectional gated recurrent unit network is used to predict the real-time temperature of the LED chip; A temperature-sensitive adaptive multi-objective loss function is constructed, including a color rendering index error term based on spectral similarity, a dynamically weighted color temperature error term, a uniformity error term, and a temperature-sensitive energy efficiency error term. The color rendering index error term uses KL divergence to measure spectral distribution similarity. The weight coefficient of the color temperature error term is dynamically adjusted with temperature. The uniformity error term evaluates illumination uniformity at multiple spatial scales. The energy efficiency error term introduces a temperature-dependent weight function. An online learning strategy is designed based on the temperature state. The learning rate is dynamically adjusted through a temperature-adaptive learning rate regulation mechanism, which includes an exponential decay term for the temperature change. The sample importance weight is determined according to the deviation between the sample temperature and the average temperature. The parameters of the coupling relationship equation are optimized by adaptive adjustment of the momentum term based on the temperature gradient and gradient clipping under the temperature threshold, and the real-time driving parameters of multiple groups of LEDs are output.

[0040] For example, an input feature layer is constructed to encode the parameters of the coupling equation. Specifically, one-hot encoding or embedding encoding can be used to convert the discrete equation parameters into continuous vector representations. For example, parameters such as LED current and voltage can be normalized to the range of 0-1 and then mapped to a 128-dimensional feature vector through a fully connected layer.

[0041] Based on feature encoding, a multi-scale feature extraction network is established. The network includes a temporal attention module, a spatial attention module, and a channel attention module to capture the dynamic temperature change characteristics, the spectral spatial distribution characteristics, and realize the adaptive weighting between feature channels.

[0042] The temporal attention module uses the product of the query matrix, key matrix, and value matrix to obtain temporal weights. Specifically, the input feature sequence can be mapped into query, key, and value matrices through three independent fully connected layers, with the matrix dimensions being (sequence length, feature dimension). The dot product of the query matrix and the key matrix is ​​then calculated and softmax normalized to obtain the attention weights. Finally, the attention weights are multiplied by the value matrix to obtain the weighted temporal feature representation.

[0043] The spatial attention module performs average and max pooling on the feature map before performing a convolution operation. Specifically, global average pooling and max pooling are performed on the input 3D feature map (number of channels × height × width), resulting in two 2D feature maps. These two feature maps are concatenated along the channel dimension and then passed through a 7×7 convolutional layer and a sigmoid activation function to generate a spatial attention weight map. This weight map is then multiplied with the original feature map to achieve feature weighting in the spatial dimension.

[0044] The channel attention module implements adaptive weighting between feature channels. This can be achieved using a squeeze-and-excitation architecture. First, global average pooling is performed on the feature map to obtain channel descriptors. Then, two fully connected layers are used to learn the correlation between channels. Finally, a weight coefficient for each channel is obtained. This coefficient is multiplied by the original feature map to achieve feature recalibration along the channel dimension.

[0045] Based on feature extraction, a bidirectional gated recurrent unit (Bi-GRU) network is used to predict the real-time temperature of LED chips. The Bi-GRU consists of two GRU layers, one forward and one backward, which can simultaneously consider past and future contextual information. Specifically, the extracted feature sequence is input into the Bi-GRU, which outputs a hidden state at each time step. Finally, a fully connected layer maps the hidden state to the predicted temperature value.

[0046] A temperature-sensitive adaptive multi-objective loss function is constructed, including color rendering index error, color temperature error, uniformity error, and energy efficiency error. The color rendering index error term uses KL divergence to measure the similarity between the predicted spectrum and the target spectrum. The weight coefficient of the color temperature error term is dynamically adjusted with temperature. A temperature threshold can be set, and the weight of this term is increased when the temperature exceeds the threshold. The uniformity error term evaluates illumination uniformity at multiple spatial scales and can be quantified by calculating the variance of the illumination map at different scales. The energy efficiency error term introduces a temperature-dependent weight function; the higher the temperature, the greater the weight of the energy efficiency error term.

[0047] An online learning strategy is designed based on temperature conditions, dynamically adjusting the learning rate through a temperature-adaptive learning rate adjustment mechanism. This learning rate adjustment mechanism includes an exponential decay term based on the temperature change, which can be expressed as: new learning rate = base learning rate × exp(-temperature change × decay coefficient). The decay coefficient can be set to 0.01. The importance weight of a sample is determined based on the deviation of the sample temperature from the average temperature, with a higher weight being assigned based on the deviation.

[0048] The parameters of the coupling equation are optimized using adaptive adjustment of the momentum term based on the temperature gradient and gradient clipping under a temperature threshold. The momentum term adjustment can be expressed as: new momentum = original momentum × (1 + absolute value of the temperature gradient). During gradient clipping, the temperature threshold can be set to 80°C; when the temperature exceeds the threshold, the gradient value is clamped to the range [-1, 1]. Based on the optimized coupling equation parameters, real-time driving parameters for multiple sets of LEDs are output. For example, for a lighting system consisting of red, green, and blue LEDs, the current (in mA) and voltage (in V) values ​​for each LED can be output. At room temperature of 25°C, the driving parameters for the red LED might be 350mA / 2.2V, the green LED 300mA / 3.0V, and the blue LED 250mA / 3.3V. When the ambient temperature rises to 50℃, the optimized driving parameters may be adjusted to: red LED 320mA / 2.1V, green LED 280mA / 2.9V, blue LED 230mA / 3.2V.

[0049] Figure 2 This diagram illustrates the temperature-adaptive dynamic adjustment mechanism for the multi-objective loss function based on a multi-attention mechanism. This innovative mechanism automatically optimizes the weight distribution of different error terms based on temperature fluctuations, particularly demonstrating significant weight adjustments at the critical temperature thresholds of 45°C and 80°C. The benefits of this adaptive strategy include significantly improving system stability in high-temperature environments, increasing the color temperature error weight from 0.30 to 0.45, ensuring precise control of key visual parameters, and optimizing computational resources by reducing the weights of secondary parameters. This allows the system to effectively address the challenges posed by temperature fluctuations, extending the lifespan and performance stability of the LED lighting system in extreme environments.

[0050] Figure 3 This graph shows the temperature-adaptive optimization of LED driver parameters using the multi-attention mechanism of the present invention. This graph demonstrates the real-time optimization effect of the multi-attention mechanism and temperature-adaptive strategy on LED driver parameters. As the LED chip temperature rises from 25°C to 80°C, the system precisely adjusts the driving current of the three-color LEDs to ensure stable optical performance. This intelligent adjustment of the driving current significantly reduces performance degradation caused by high temperatures, extending the LED lifespan. Temperature prediction based on the bidirectional gated recurrent unit (Bi-GRU) enables the system to make parameter adjustments in advance, avoiding damage caused by thermal shock. This system achieves an optimal balance between energy efficiency and optical performance, improving energy savings by over 15% under high-temperature conditions. It also ensures the stability of color temperature and color rendering index, maintaining professional-grade lighting quality even in high-temperature environments of 80°C.

[0051] The present invention implements a deep learning optimization algorithm based on a multiple attention mechanism and temperature adaptation, which can effectively solve the coupling relationship equation of LED driving parameters and dynamically adjust the driving parameters according to real-time temperature changes, thereby ensuring high color rendering, accurate color temperature, uniformity and high energy efficiency of the LED lighting system.

[0052] In an optional embodiment, constructing an adaptive multi-objective loss function includes: An adaptive weight function is constructed to weight the color rendering index error term, the color temperature error term, the uniformity error term, and the energy efficiency error term, wherein the adaptive weight function includes: a basic weight term for setting an initial weight of each error term; a temperature adjustment factor that is linearly related to the temperature when the temperature exceeds the temperature threshold; and a time adjustment factor that is dynamically updated during the optimization iteration process; The gradient of the loss function weighted by the adaptive weight function is calculated, and the product of the gradient value of each loss term and the temperature-related weight function in the energy efficiency error term is used as the target importance index to characterize the optimization priority of each error term under different temperature conditions; A dynamic balance strategy is adopted to update the weights of the adaptive multi-objective loss function under different temperature states, including: according to the deviation between the target importance index and the current weight and the loss change, combined with the learning rate adjustment mechanism and the momentum term based on the temperature gradient, the weight parameters of each error term are iteratively optimized to achieve dynamic balance of the weights of each loss function.

[0053] Exemplarily, an adaptive weighting function is constructed to weight the color rendering index error term, color temperature error term, uniformity error term, and energy efficiency error term. The adaptive weighting function includes the following parts: Basic weight term: used to set the initial weight of each error term. For example, the initial weights of the color rendering index error term, color temperature error term, uniformity error term, and energy efficiency error term can be set to 0.3, 0.3, 0.2, and 0.2, respectively.

[0054] Temperature Adjustment Factor: When the temperature exceeds a preset threshold, this factor increases linearly with temperature. Specifically, you can set the temperature threshold to 45°C. When the temperature exceeds 45°C, the temperature adjustment factor increases linearly with the temperature, increasing by 0.02 for every 1°C increase. For example, when the temperature is 50°C, the temperature adjustment factor is 0.1.

[0055] Time Adjustment Factor: This factor is dynamically updated during the optimization iteration process. It can be set to an initial value of 1 and updated every 100 iterations. The update formula is: New Factor = Old Factor * (1 + Current Iteration Number / Total Iteration Number). Multiplying these three factors together yields the final adaptive weight function.

[0056] The gradient is calculated based on the loss function weighted by the adaptive weight function. The specific steps are as follows: Calculate the gradient of each error term. Taking the color rendering index error term as an example, its gradient can be calculated using the finite difference method: take a small perturbation near the current parameter value, calculate the change in error before and after the perturbation, and divide it by the perturbation amount to get the approximate gradient value.

[0057] The target importance index is obtained by multiplying the gradient of each error term by the temperature-dependent weight function in the energy efficiency error term. The temperature-dependent weight function can be defined as 1 when the temperature is below 35°C, linearly decreasing to 0.5 between 35°C and 45°C, and remaining unchanged at 0.5 when the temperature is above 45°C.

[0058] This objective importance metric is used to characterize the optimization priority of each error term under different temperature conditions. For example, at low temperatures, optimizing the color rendering index and color temperature may be more important, while at high temperatures, optimizing uniformity and energy efficiency may be more critical.

[0059] A dynamic balancing strategy is used to update the weights of an adaptive multi-objective loss function under different temperature conditions. The specific steps are as follows: Calculate the deviation between the target importance index and the current weight. For example, if the target importance index of the color rendering index error term is 0.4 and the current weight is 0.3, the deviation is 0.1. Calculate the loss change for each error term. This can be obtained by comparing the loss values ​​between two iterations. Combining a learning rate adjustment mechanism with a momentum term based on a temperature gradient, the weight parameters of each error term are iteratively optimized. The learning rate adjustment mechanism can adopt an annealing strategy, where the learning rate gradually decreases with increasing iterations. For example, the initial learning rate can be set to 0.01 and halved every 1000 iterations. The momentum term based on the temperature gradient can be defined as follows: when the temperature change rate exceeds a certain threshold (e.g., 1°C per minute), the momentum term increases; otherwise, it decreases. This helps to accelerate weight adjustment when the temperature changes rapidly, while maintaining relative weight stability when the temperature is stable.

[0060] Taking all the above factors into consideration, we can get the weight update formula: new weight = old weight + learning rate × (deviation between target importance index and current weight + loss change × momentum term).

[0061] Existing quality control methods for LED lighting systems primarily rely on fixed-weight optimization methods or simple linear adaptive strategies. This paper proposes an adaptive weight function based on multi-factor coupling, which organically combines basic weight terms, temperature adjustment factors, and time adjustment factors to achieve intelligent dynamic adjustment of weights. Furthermore, a target importance metric is introduced, accurately quantifying the optimization priority of each error term under different temperature conditions by multiplying the gradient value with the temperature weight function. A dynamic balancing strategy, combined with a temperature-adaptive learning rate mechanism and a momentum term based on the temperature gradient, achieves precise iterative optimization of weights.

[0062] In an optional embodiment, generating a PWM drive signal sequence according to the real-time drive parameters and outputting a target mixed spectrum includes: A discrete wavelet transform is performed on the real-time drive parameters to decompose the parameter signal into multiple frequency band sub-signals; a complexity coefficient is determined by calculating the energy entropy value and peak ratio of each frequency band, and the frequency bands are dynamically grouped into low-frequency, medium-frequency and high-frequency regions according to the complexity coefficient; a different quantization strategy is applied to each group of frequency band parameters, a PWM signal of a first current magnitude is used for the low-frequency group, a PWM signal of a second current magnitude is used for the medium-frequency group, and a PWM signal of a third current magnitude is used for the high-frequency group, wherein the first current magnitude is greater than the second current magnitude, and the second current magnitude is greater than the third current magnitude; and a multi-band control signal is recombined into a hybrid PWM drive sequence through a feedback-corrected wavelet reconstruction algorithm; A database of corresponding relationships between LED current and spectral output is established, the expected spectral output of the hybrid PWM drive sequence is calculated, and the error vector between the expected spectral output and the target spectrum is obtained. The correction values ​​of the PWM parameters are determined through an error back-propagation calculation method, and the duty cycle, current magnitude, and pulse edge shape of the hybrid PWM drive sequence are adjusted. Pulse density modulation and PWM modulation are combined to adjust the distribution density of the pulses in the time window, compensate for the nonlinear response characteristics of the LED, and output the target hybrid spectrum.

[0063] For example, a discrete wavelet transform is performed on the input real-time driving parameters. The complexity coefficient is determined by calculating the ratio of the energy entropy value to the peak value of each frequency band signal. For each frequency band signal, its Shannon entropy value is calculated as the energy entropy, and the peak value of the signal is also calculated. The complexity coefficient is defined as the ratio of the entropy value to the peak value. For example, if the entropy value of a frequency band signal is 2.5 and the peak value is 10, its complexity coefficient is 0.25.

[0064] Based on the calculated complexity coefficients, the five frequency bands were dynamically grouped into three regions: low-frequency, medium-frequency, and high-frequency. The specific grouping method is as follows: frequency bands with a complexity coefficient less than 0.2 are classified as low-frequency, those between 0.2 and 0.5 are classified as medium-frequency, and those greater than 0.5 are classified as high-frequency.

[0065] Different quantization strategies are applied to the three groups of frequency band parameters. The low-frequency group uses a PWM signal of the first current magnitude, with a current value of 700mA; the medium-frequency group uses a PWM signal of the second current magnitude, with a current value of 500mA; and the high-frequency group uses a PWM signal of the third current magnitude, with a current value of 300mA.

[0066] A feedback-corrected wavelet reconstruction algorithm reconstructs multi-band control signals into a hybrid PWM drive sequence. The algorithm first upsamples and filters the signals in each frequency band, then adds the processed signals together to produce a preliminary reconstruction. The reconstruction error is calculated by comparing the reconstructed signal with the original signal, and this error is fed back to each frequency band for correction. Iterative optimization is performed until the reconstruction error is less than a preset threshold (e.g., 0.1%).

[0067] To establish a database of the corresponding relationship between LED current and spectral output, the spectral output of the LED was measured and recorded at different current values ​​(from 100mA to 1000mA, with a step size of 50mA). A spectrometer was used to collect spectral data of the LED at each current value and store the data in the database.

[0068] Using the established database, the expected spectral output of a hybrid PWM drive sequence is calculated. The specific method is to decompose the PWM sequence into multiple time segments, each corresponding to a constant current value. The database is then queried to obtain the spectral data corresponding to each time segment. The spectral data for all time segments are then weighted averaged according to the time weight to obtain the expected spectral output.

[0069] The calculated expected spectrum output is compared with the target spectrum to obtain the error vector, which is defined as the difference between the expected spectrum and the target spectrum at each wavelength point.

[0070] The PWM parameter correction values ​​are determined using a reverse error propagation method. This method first reverse-maps the error vector to each frequency band. Then, based on the characteristics of each frequency band, the corresponding PWM parameter adjustment is calculated. For example, for the low-frequency group, the main adjustment is the duty cycle of the PWM signal; for the medium-frequency group, the current level is adjusted; and for the high-frequency group, the pulse edge shape is adjusted.

[0071] Based on the calculated correction values, the hybrid PWM drive sequence is adjusted. These adjustments include: duty cycle adjustment: fine-tuning the duty cycle of the low-frequency PWM signal within a ±5% range; current magnitude adjustment: adjusting the current value of the medium-frequency PWM signal within a ±50mA range; and pulse edge shape adjustment: adjusting the rising and falling slopes of the high-frequency PWM signal within a ±20% range.

[0072] Combining pulse density modulation (PDM) with PWM modulation techniques optimizes the pulse density within a time window. Specifically, the PWM sequence is divided into multiple time windows (e.g., 1ms). Within each window, the pulse density is adjusted based on the characteristics of the target spectrum. In spectral regions requiring a fast response, the pulse density is increased; in regions with slower changes, the pulse density is reduced.

[0073] The present invention realizes multi-scale decomposition and precise control of driving parameters through discrete wavelet transform and frequency band dynamic grouping strategy; innovatively combines pulse density modulation with PWM modulation to compensate for the nonlinear response characteristics of LEDs by adaptively adjusting the pulse distribution density; and adopts a complexity evaluation method based on frequency band energy entropy and a wavelet reconstruction algorithm with feedback correction to significantly improve system stability.

[0074] According to a second aspect of the embodiments of the present invention, Provides a high color rendering index lighting system based on multiple groups of LED adjustable spectrum, including: The first unit is configured to obtain lighting environment parameters, including ambient brightness, ambient color temperature, and ambient color rendering index; and to collect spectral data at different locations in the lighting space using a spectral sensor to obtain spectral spatial distribution characteristics. A second unit is configured to establish a coupled relationship equation between LED chip temperature and spectral characteristics based on the lighting environment parameters and spectral spatial distribution characteristics, wherein the coupled relationship equation includes a thermal resistance matrix between LED arrays and a temperature-related quantum efficiency, wherein the thermal resistance matrix is ​​used to calculate the real-time temperature of each LED chip, and the quantum efficiency is used to characterize the effect of temperature on the luminous efficiency of the LED; The third unit is used to solve the coupling relationship equation using a deep learning optimization algorithm based on a multi-attention mechanism, construct a multi-objective loss function, and update the coupling relationship equation parameters through temperature-adaptive online learning to obtain real-time driving parameters of multiple groups of LEDs. The multi-objective loss function includes a color rendering index error term, a color temperature error term, a uniformity error term, and an energy efficiency error term. The fourth unit is configured to generate a PWM drive signal sequence according to the real-time drive parameters and output a target mixed spectrum.

[0075] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0076] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0077] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0078] 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 it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high color rendering index lighting method based on multiple groups of LEDs with adjustable spectra, characterized in that: include: Acquire lighting environment parameters, wherein the lighting environment parameters include an ambient brightness value, an ambient color temperature value, and an ambient color rendering index value; Use spectral sensors to collect spectral data at different locations in the lighting space to obtain spectral spatial distribution characteristics; Based on the lighting environment parameters and the spectral spatial distribution characteristics, a coupling relationship equation between the LED chip temperature and the spectral characteristics is established. The coupling relationship equation includes a thermal resistance matrix between the LED arrays and a temperature-related quantum efficiency. The thermal resistance matrix is ​​used to calculate the real-time temperature of each LED chip, and the quantum efficiency is used to characterize the effect of temperature on the LED luminous efficiency. A deep learning optimization algorithm based on a multi-attention mechanism is used to solve the coupling relationship equation, construct a multi-objective loss function, and update the coupling relationship equation parameters through temperature-adaptive online learning to obtain real-time driving parameters of multiple groups of LEDs. The multi-objective loss function includes a color rendering index error term, a color temperature error term, a uniformity error term, and an energy efficiency error term. A PWM drive signal sequence is generated according to the real-time drive parameters, and a target mixed spectrum is output.

2. The method according to claim 1, characterized in that Spectral sensors are used to collect spectral data at different locations in the lighting space to obtain spectral spatial distribution characteristics including: Arranging a plurality of spectral sensors whose spatial coordinates satisfy uniform distribution on a hemispherical surface in the lighting space, collecting spectral data to obtain a spectral matrix, wherein the spectral matrix includes four dimensions: wavelength, radial distance, zenith angle, and azimuth angle; Get the simplified BRDF model parameters of the environment reflection surface, the simplified BRDF model is: ; where ρ d is the diffuse reflection coefficient, ρ s is the mirror reflection coefficient, α is the angle attenuation factor, θ i is the angle of incidence, θ r is the reflection angle, λ is the wavelength; Based on the spectral matrix and the simplified BRDF model parameters, the spectral power distribution, spatial intensity distribution and angular dependence are calculated, and a comprehensive feature vector including spectral moment characteristics, spatial correlation characteristics, uniformity characteristics and angular sensitivity is constructed; principal component analysis and feature reconstruction are performed on the comprehensive feature vector to obtain the spectral spatial distribution characteristics.

3. The method according to claim 1, characterized in that Based on the lighting environment parameters and spectral spatial distribution characteristics, a coupling relationship equation between the LED chip temperature and spectral characteristics is established. The coupling relationship equation includes a thermal resistance matrix between LED arrays and a temperature-related quantum efficiency. The thermal resistance matrix is ​​used to calculate the real-time temperature of each LED chip. The quantum efficiency is used to characterize the effect of temperature on the LED luminous efficiency, including: Establishing a thermal coupling relationship for the LED array, describing the thermal coupling coefficient between the LED chips using a thermal resistance matrix, wherein the thermal coupling coefficient includes an ambient temperature term, a power consumption term, and an environmental impact term, wherein the environmental impact term is determined by color temperature and illumination, and calculating the junction temperature of each LED chip based on the elements of the thermal resistance matrix; Establish a temperature-illuminance coupling relationship for LEDs, including a temperature coefficient term and an illuminance influence term. The temperature coefficient term characterizes the attenuation characteristics of the LED luminous efficiency with temperature, and the illuminance influence term characterizes the correction effect of ambient illumination on quantum efficiency. Use an exponential function to describe the dynamic change of the LED luminous efficiency from the initial state to the steady state. The junction temperature output of the thermal coupling relationship and the efficiency calculation results of the temperature-illuminance coupling relationship are used as input parameters. The spectral characteristics of a single LED are described by a temperature-dependent Gaussian function. The spectra of multiple LEDs are weightedly superimposed based on the weight coefficients of junction temperature and ambient illumination, and the temperature-spectrum coupling relationship equation of the LED array is established.

4. The method according to claim 3, characterized in that The spectral characteristics of a single LED are described by a temperature-dependent Gaussian function. The spectra of multiple LEDs are weighted and superimposed based on the weight coefficients of junction temperature and ambient illumination. The temperature-spectrum coupling relationship equation of the LED array is established, including: A temperature-dependent Gaussian function is used to describe the LED spectral characteristics. The Gaussian function includes a peak intensity term, a center wavelength term, and a standard deviation term. The peak intensity term and the center wavelength term are updated using a quadratic polynomial of the temperature variation, and the standard deviation term is updated using an exponential function. The update coefficients of the peak intensity term and the center wavelength term have a nonlinear relationship with the temperature variation and the ambient illumination, and the update coefficient of the standard deviation term has an exponential decay relationship with the temperature variation. Performing nonlinear superposition of the spectra of multiple LEDs, including a weighted term and a spectral interference term. The weighted term calculates the contribution weight of each LED using a composite weighting method based on junction temperature deviation and ambient illumination, where the ambient illumination affects the baseline value of the weight coefficient through a logarithmic function. The spectral interference term characterizes the spectral interaction between LEDs through a temperature-dependent interference coefficient, which decays exponentially with junction temperature. Based on the parameters of the Gaussian function and the coefficients of the nonlinear superposition, a complete coupling relationship equation is constructed.

5. The method according to claim 1, wherein A deep learning optimization algorithm based on a multi-attention mechanism is used to solve the coupling relationship equation, a multi-objective loss function is constructed, and the coupling relationship equation parameters are updated through temperature-adaptive online learning to obtain real-time driving parameters of multiple groups of LEDs. The multi-objective loss function includes color rendering index error term, color temperature error term, uniformity error term, and energy efficiency error term. Constructing an input feature layer to perform feature encoding on the parameters of the coupling relationship equation; A multi-scale feature extraction network is established based on the feature encoding, and the dynamic temperature change characteristics are captured by the time attention module. The time attention module uses the product operation of the query matrix, the key matrix and the value matrix to obtain the time dimension weight; the spectral spatial distribution characteristics are extracted by the spatial attention module. The spatial attention module performs convolution operation after performing average pooling and maximum pooling on the feature map; and the channel attention module realizes adaptive weighting between feature channels; A bidirectional gated recurrent unit network is used to predict the real-time temperature of the LED chip; A temperature-sensitive adaptive multi-objective loss function is constructed, including a color rendering index error term based on spectral similarity, a dynamically weighted color temperature error term, a uniformity error term, and a temperature-sensitive energy efficiency error term. The color rendering index error term uses KL divergence to measure spectral distribution similarity. The weight coefficient of the color temperature error term is dynamically adjusted with temperature. The uniformity error term evaluates illumination uniformity at multiple spatial scales. The energy efficiency error term introduces a temperature-dependent weight function. An online learning strategy is designed based on the temperature state. The learning rate is dynamically adjusted through a temperature-adaptive learning rate regulation mechanism, which includes an exponential decay term for the temperature change. The sample importance weight is determined according to the deviation between the sample temperature and the average temperature. The parameters of the coupling relationship equation are optimized by adaptive adjustment of the momentum term based on the temperature gradient and gradient clipping under the temperature threshold, and the real-time driving parameters of multiple groups of LEDs are output.

6. The method according to claim 5, characterized in that Constructing an adaptive multi-objective loss function includes: An adaptive weight function is constructed to weight the color rendering index error term, the color temperature error term, the uniformity error term, and the energy efficiency error term, wherein the adaptive weight function includes: a basic weight term for setting an initial weight of each error term; a temperature adjustment factor that is linearly related to the temperature when the temperature exceeds the temperature threshold; and a time adjustment factor that is dynamically updated during the optimization iteration process; The gradient of the loss function weighted by the adaptive weight function is calculated, and the product of the gradient value of each loss term and the temperature-related weight function in the energy efficiency error term is used as the target importance index to characterize the optimization priority of each error term under different temperature conditions; A dynamic balance strategy is adopted to update the weights of the adaptive multi-objective loss function under different temperature states, including: according to the deviation between the target importance index and the current weight and the loss change, combined with the learning rate adjustment mechanism and the momentum term based on the temperature gradient, the weight parameters of each error term are iteratively optimized to achieve dynamic balance of the weights of each loss function.

7. The method according to claim 1, characterized in that Generating a PWM drive signal sequence according to the real-time drive parameters and outputting a target mixed spectrum includes: A discrete wavelet transform is performed on the real-time drive parameters to decompose the parameter signal into multiple frequency band sub-signals; a complexity coefficient is determined by calculating the energy entropy value and peak ratio of each frequency band, and the frequency bands are dynamically grouped into low-frequency, medium-frequency and high-frequency regions according to the complexity coefficient; a different quantization strategy is applied to each group of frequency band parameters, a PWM signal of a first current magnitude is used for the low-frequency group, a PWM signal of a second current magnitude is used for the medium-frequency group, and a PWM signal of a third current magnitude is used for the high-frequency group, wherein the first current magnitude is greater than the second current magnitude, and the second current magnitude is greater than the third current magnitude; and a multi-band control signal is recombined into a hybrid PWM drive sequence through a feedback-corrected wavelet reconstruction algorithm; A database of corresponding relationships between LED current and spectral output is established, the expected spectral output of the hybrid PWM drive sequence is calculated, and the error vector between the expected spectral output and the target spectrum is obtained. The correction values ​​of the PWM parameters are determined through an error back-propagation calculation method, and the duty cycle, current magnitude, and pulse edge shape of the hybrid PWM drive sequence are adjusted. Pulse density modulation and PWM modulation are combined to adjust the distribution density of the pulses in the time window, compensate for the nonlinear response characteristics of the LED, and output the target hybrid spectrum.

8. A high color rendering index lighting system based on multiple groups of LEDs with adjustable spectrum, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to obtain lighting environment parameters, wherein the lighting environment parameters include an ambient brightness value, an ambient color temperature value, and an ambient color rendering index value; Use spectral sensors to collect spectral data at different locations in the lighting space to obtain spectral spatial distribution characteristics; A second unit is configured to establish a coupled relationship equation between LED chip temperature and spectral characteristics based on the lighting environment parameters and spectral spatial distribution characteristics, wherein the coupled relationship equation includes a thermal resistance matrix between LED arrays and a temperature-related quantum efficiency, wherein the thermal resistance matrix is ​​used to calculate the real-time temperature of each LED chip, and the quantum efficiency is used to characterize the effect of temperature on the luminous efficiency of the LED; The third unit is used to solve the coupling relationship equation using a deep learning optimization algorithm based on a multi-attention mechanism, construct a multi-objective loss function, and update the coupling relationship equation parameters through temperature-adaptive online learning to obtain real-time driving parameters of multiple groups of LEDs. The multi-objective loss function includes a color rendering index error term, a color temperature error term, a uniformity error term, and an energy efficiency error term. The fourth unit is configured to generate a PWM drive signal sequence according to the real-time drive parameters and output a target mixed spectrum.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images