Intelligent lighting control system and method based on multi-mode sensor

Abstract

The invention discloses an intelligent illumination control system and method based on a multi-modal sensor, and particularly relates to the field of data analysis, and the system comprises a multi-modal data collection module, a feature calculation module, a space-time coupling analysis module, a multi-modal verification module, a biological rhythm compensation module and an intelligent decision control module. According to the invention, through cooperative acquisition of the multi-modal sensor and space-time coupling modeling, accurate environment perception is realized, and the photo-thermal interaction effect is quantified to eliminate control deviation; an innovative biological rhythm compensation mechanism is fused with the ultraviolet dose and the human body activity data, and a dynamic safety protection boundary is established; and an LSTM-Transform hybrid architecture is combined with a rolling time domain optimization strategy, so that 200ms-level quick response and complex scene self-adaptive control are realized, and the dynamic performance and the security of the system are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and more specifically, to an intelligent lighting control system and method based on multi-modal sensors. Background Art

[0002] Traditional technical solutions have long been limited by a single-dimensional perception architecture, mainly relying on basic light intensity sensors or timing programs to achieve lighting control, lacking the overall perception ability of the environmental multi-physical fields. The current mainstream commercial systems generally adopt a photosensitive resistor array combined with a microcontroller architecture. By collecting local area illuminance data and comparing it with a preset threshold, PWM dimming technology is used to achieve linear brightness adjustment. Some mid- to high-end systems integrate passive infrared sensors to achieve human presence detection function, forming a serial control logic of "light detection - threshold judgment - dimming output".

[0003] The operation process of the existing technology shows a discrete data acquisition and linear decision-making mode: First, the illuminance value is obtained through a single-point light intensity sensor at a fixed period. The temperature compensation type system uses an independent thermistor for auxiliary measurement, and only simple moving average filtering is implemented in the data preprocessing link; the feature extraction stage is limited to the calculation of illuminance mean and fluctuation amplitude, and the control decision uses a look-up table method to match the preset brightness curve. In the output stage, the LED drive current is adjusted by a fixed duty cycle. Some systems integrating human sensing adopt a moving detection trigger mechanism, and directly switch the preset lighting scene after identifying the movement of the heat source.

[0004] The existing technology has three-dimensional defects: First, there is a serious dimensionality loss in the environmental parameter perception system. The independent processing of multi-modal data such as light intensity, temperature, and acoustics leads to misjudgment of the environmental state. In particular, the visual fatigue caused by the photo-thermal coupling effect cannot be quantitatively evaluated; Second, the control of ultraviolet exposure is completely decoupled from the human biological rhythm, lacking a safety protection mechanism based on the time-accumulated dose, and there are potential health hazards in long-term use; Third, the control strategy adopts a fixed-period sampling and linear response mode, which is difficult to adapt to the needs of dynamic changing scenarios, manifested as lagging adjustment of lighting parameters and stroboscopic phenomena, and obvious defects are exposed in the application of complex office environments and medical places. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the existing technology, the present invention provides an intelligent lighting control system and method based on multi-modal sensors. Through the following solutions, the three major defects of the fragmented processing of multi-dimensional environmental parameters, the disconnection between ultraviolet exposure and biological rhythm, and the lag of the fixed-period control strategy in the above-mentioned background art are solved, resulting in problems such as misjudgment of the environmental state, potential health hazards, and response delay.

[0006] To achieve the above object, the present invention provides the following technical solution: An intelligent lighting control system based on multi-modal sensors, comprising: Multi-modal data acquisition module: Composed of a heterogeneous sensor network, it realizes data acquisition by synchronously collecting multi-dimensional physical quantities of the environment. It integrates optical, thermal, acoustic, spectral, ultraviolet and human motion perception components, and pre-processes the original data in a standardized manner using a unified timing protocol; Feature calculation module: Performs feature extraction based on the original sensor data stream. By establishing a mathematical model, it converts the original signal into six quantitative parameters representing the dynamic characteristics of the environment, and uses sliding window processing and frequency domain analysis techniques to ensure the calculation accuracy; Spatio-temporal coupling analysis module: Models the spatio-temporal correlation between the light intensity change rate and the dynamic heat distribution index, constructs a photo-thermal coupling non-linear equation, and outputs the environmental dynamic index as the benchmark input for subsequent multi-modal verification; Multi-modal verification module: Verifies the environmental coordination using a dual-mode conditional architecture, triggers the acoustic-optical coordination or color difference protection mechanism based on the color deviation threshold, and uses the hyperbolic tangent function and inverse proportion constraint to achieve cross-verification and dynamic adjustment of multi-modal data; Biological rhythm compensation module: Integrates the ultraviolet cumulative exposure and human activity entropy parameters, constructs a dose-response dynamic balance model, and outputs a biological impact coefficient to simultaneously reflect the light demand intensity and the biological safety threshold, providing a basis for rhythm compensation for decision-making optimization; Intelligent decision-making and control module: Adopts a prediction model and a self-attention mechanism to construct a hybrid decision-making architecture, generates a fault tolerance compensation term through real-time data difference analysis, dynamically allocates the weights of feature parameters, outputs an environmental regulation index, and drives the lighting terminal to achieve multi-dimensional precise control.

[0007] Preferably, the multi-modal data acquisition module consists of six types of heterogeneous sensors to form a networked detection system. The visible light sensor array and the infrared thermal imaging component cooperate to obtain the environmental optical characteristics. The distributed temperature array constructs a spatial thermal field monitoring topology. The high-precision acoustic module captures the sound pressure characteristics of a specific frequency band. The multi-channel spectral sensor analyzes the spectral power distribution. The ultraviolet sensing unit continuously records the radiation dose. The pyroelectric infrared array quantifies the human motion trajectory. All sensing data is synchronously collected and pre-processed using a unified timing protocol.

[0008] Preferably, the multi-modal data acquisition module uses a TSL2591 high-precision optical sensor array, and 3 groups of sensor nodes are distributed in the space at the top. Each group of nodes includes visible light and infrared dual channels, synchronously collects illuminance data at a sampling rate of 1Hz, transmits the original data through the I2C bus, uses sliding window average filtering in the pre-processing stage, and uses the 3σ criterion to detect and remove mutation noise for outliers.

[0009] Preferably, the multi-modal data acquisition module uses an MLX90640 infrared thermal imaging sensor to construct a 5×5 temperature array. The node spacing is deployed in a grid pattern at a height of 2.5 m with a spacing of 0.5 m×0.5 m. Surface temperature data in the range of -40°C to 300°C is acquired at a sampling rate of 0.2 Hz. The BME280 environmental sensor is used to obtain air temperature and humidity compensation parameters. The improved Kriging method is adopted as the spatial interpolation algorithm to generate a continuous temperature field, and the dynamic calibration error is controlled within ±0.3°C.

[0010] Preferably, the multi-modal data acquisition module is configured with an INMP441 digital MEMS microphone array. Sound pressure waveform data is acquired at a sampling frequency of 48 kHz and a quantization accuracy of 24 bit. The hardware is pre-set with an A-weighting filter to eliminate low-frequency interference. A 5-second effective audio segment is intercepted every 30 seconds for FFT transformation. The frequency band analysis focuses on the 200 - 4000 Hz human voice characteristic frequency band, and the background noise baseline is dynamically calibrated by the minimum sound pressure level at night.

[0011] Preferably, the multi-modal data acquisition module deploys an AS7265x six-channel spectral sensor. Spectral irradiance at 18 characteristic wavelengths in the range of 415 - 940 nm is synchronously acquired with a period of 5 seconds. It is converted to XYZ tristimulus values through the CIE1931 color matching function. White balance calibration is performed with reference to the D65 standard light source. The McCamy approximation formula is used for color temperature calculation, and an automatic exposure compensation mechanism is enabled when the color coordinates are converted to the CIELAB uniform color space.

[0012] Preferably, the multi-modal data acquisition module uses a GUVA-S12SD ultraviolet sensitive element, configured with a 280 - 400 nm band-pass optical filter. The UV index in the range of 0 - 15 is recorded at 10-second intervals. The temperature compensation algorithm is enabled when establishing an exponential decay model for data accumulation. The periodic shading calibration method is used for dark current elimination, and a cubic polynomial fitting compensation curve is applied for non-linear correction.

[0013] Preferably, the multi-modal data acquisition module installs an AMG8833 pyroelectric infrared array sensor to construct a 16-zone dynamic detection grid with a 4×4 layout. The heat source intensity distribution in each area is output at a frequency of 0.5 Hz. The improved Hungarian algorithm is adopted for target matching in motion trajectory tracking. The speed vector is calculated based on the position difference between two adjacent frames. The stay determination threshold is set to a displacement less than 0.2 m for three consecutive frames, and Kalman filter smoothing processing is used for three-dimensional positioning error compensation.

[0014] Preferably, the feature parameters extracted by the feature calculation module include the light intensity change rate, dynamic heat distribution index, voiceprint activity, color deviation coefficient, cumulative ultraviolet exposure, and human activity entropy. The light intensity change rate is derived by constructing a composite model of the coefficient of variation and gradient change of the light intensity time series, quantifying the fluctuation intensity with the ratio of the standard deviation to the mean, and enhancing the instantaneous change feature by combining the root mean square of the differences between adjacent sampling points. The dynamic heat distribution index is based on the thermodynamic gradient theory, taking the ratio of the standard deviation of the temperature field to the sensor spacing to characterize the spatial heterogeneity, and calculating the normalized temperature gradient to reflect the change direction. Finally, the maximum value of the two is taken to ensure the capture of significant temperature features. The voiceprint activity is based on the sound energy density principle, performing band-pass integration on the 200 - 4000 Hz human voice frequency band, and suppressing environmental noise interference through frequency domain amplitude weighting and maximum frequency normalization. The color deviation coefficient uses the Euclidean distance to measure the color difference between the measured RGB and the reference value, and introduces the relevant color temperature ratio to compensate for the white light reference difference. Its calculation process implies the non-linear conversion of the CIELAB color space. The cumulative ultraviolet exposure is based on the radiation dosimetry principle, using an exponential decay integral model to quantify the time-varying ultraviolet intensity, and the decay coefficient corresponds to the dose memory effect of biological tissues. The human activity entropy integrates the information entropy theory and kinematic parameters, characterizing the randomness of motion through the position distribution probability entropy value, and enhancing the dynamic feature by combining the ratio of the maximum speed to the sampling interval for weighting.

[0015] Preferably, the light intensity change rate is specifically expressed as: , where LIVR represents the light intensity change rate, L is the instantaneous value sequence of the light sensor, σ is the light standard deviation, μ is the mean, i represents the sampling point serial number index of the light intensity data in the time series, and n represents the total number of sampling points included in the current analysis time window.

[0016] Preferably, the dynamic heat distribution index is specifically expressed as: , where THDI represents the dynamic heat distribution index, σ_T is the standard deviation of the temperature sensor array, ÑT is the temperature gradient, Dx is the sensor spacing, and T_avg is the average temperature.

[0017] Preferably, the voiceprint activity is specifically expressed as: , where SVA represents the voiceprint activity, S(f) is the sound pressure frequency domain signal, the integration range is 200 - 4000 Hz, and f_max is the highest effective frequency that can be resolved, specifically half of the sampling frequency.

[0018] Preferably, the color deviation coefficient is specifically expressed as: , where CDC represents the color deviation coefficient, RGB is the measured value, the subscript 0 is the reference value, and CCT is the relevant color temperature.

[0019] Preferably, the cumulative ultraviolet exposure is specifically expressed as: , UVIE represents the cumulative ultraviolet exposure, α is the attenuation coefficient corresponding to a 30-minute half-life, t_0 is the starting time point of ultraviolet monitoring, t represents the current time point, and UV(t) represents the instantaneous radiation intensity value measured by the ultraviolet sensor at time t.

[0020] Preferably, the α is determined according to the biological half-life model. Referring to the recommended limit of ultraviolet exposure by ACGIH, a 30-minute half-life is adopted. , through the formula α≈0.023 min is calculated. -1 .

[0021] Preferably, the human activity entropy is specifically expressed as: , HAE represents the human activity entropy, x_k is the position distribution probability of the infrared sensor, v_max is the maximum moving speed, Dt is the sampling time interval, and k represents the kth dynamic detection grid.

[0022] Preferably, the environmental dynamic index integrates the light intensity fluctuation and the thermal field distribution characteristics through a non-linear coupling model. First, the two-thirds power term of the light intensity change rate is used to strengthen the influence weight of high-intensity fluctuations, and the natural logarithm form of the dynamic thermal distribution index is introduced into the denominator term to suppress the over-sensitivity at the steady state of the thermal field. Then, the absolute value of the spatial gradient of the temperature field is calculated based on the thermodynamic gradient theory, and the influence of local sudden heat sources is compensated by an exponential decay function. Finally, the light intensity dominant term weighted by the optothermal coupling coefficient determined by principal component analysis is combined with the gradient decay coefficient derived from the thermal comfort model to adjust the temperature gradient term, realizing the normalized fusion of the light intensity time series fluctuation and the thermal field spatial heterogeneity.

[0023] Preferably, the environmental dynamic index is specifically expressed as: , EDI represents the environmental dynamic index, k_1 is the optothermal coupling coefficient, k_2 is the gradient decay coefficient, and ÑTHDI is the spatial gradient of the thermal distribution index.

[0024] Preferably, the multimodal verification module processes the environmental coordination index through a conditional judgment framework. The construction of the environmental coordination index is based on the interactive verification mechanism of the environmental dynamic index and multimodal parameters. When the color deviation coefficient does not exceed the preset threshold, the modulation intensity of the environmental dynamic index by the voiceprint activity is constrained by the hyperbolic tangent function, and the environmental coordination is reflected through the acousto-optic coupling effect; when the color deviation coefficient exceeds the threshold, the color difference protection mechanism is triggered, and the environmental dynamic index is dynamically attenuated by an inverse proportional function. The attenuation degree is determined by the ratio of the excess amplitude of the color deviation coefficient to the attenuation factor, and this threshold is set according to the CIE color tolerance standard.

[0025] Preferably, the environmental coordination index is specifically expressed as: , SCI represents the environmental coordination index, λ is the acousto-optic conversion threshold, δ is the color tolerance threshold, and γ is the color difference attenuation factor.

[0026] Preferably, the δ is set according to the color difference tolerance standard issued by CIE, λ refers to the frequency band weight parameter in the ANSI speech intelligibility test specification, and γ is determined by fitting the experimental data of the color difference comfort mapping in the laboratory environment.

[0027] Preferably, the biological impact coefficient is constructed by integrating the spatial comfort index, the cumulative ultraviolet exposure, and the human activity entropy. First, a dynamic balance model of ultraviolet dose and light demand is established, and the cumulative ultraviolet exposure is normalized using a dose threshold. When the cumulative exposure exceeds the safety threshold, the light intensity is suppressed by an inverse proportional function. Secondly, the sliding mean and variance of the human activity entropy are introduced to construct an adaptive adjustment term, and the Sigmoid function is used to constrain the deviation of the human activity entropy from the reference value, converting the randomness characteristics of the activity entropy into a standardized compensation coefficient. Finally, the light demand attenuation factor and the biological activity compensation term are fused in a multiplicative form to achieve the non-linear coupling between the biological safety limit of ultraviolet exposure and the light response of human dynamic behavior. The sliding mean is calculated using a time decay window mechanism, and the variance is dynamically updated through Kalman filtering.

[0028] Preferably, the biological impact coefficient is specifically expressed as: , where BIC represents the biological impact coefficient, η is the ultraviolet safety threshold, μ is the historical sliding mean of HAE, ν is the real-time variance of HAE, and σ is the sigmoid activation function.

[0029] Preferably, the environmental regulation index is constructed by integrating a feed-forward analysis and a feedback correction mechanism in a hybrid architecture. First, a long short-term memory network is used to perform a temporal prediction on the biological impact coefficient and historical characteristic parameters to generate a reference value. Then, a Transformer encoder is used to extract the spatial correlation characteristics of multi-modal data, and the self-attention mechanism is used to dynamically allocate the decision weights of each parameter. The biological impact coefficient is subjected to non-linear enhancement processing at the real-time calculation level, and an exponential term optimized by gradient descent is used to adjust the response intensity in different scenarios. At the same time, a fault tolerance compensation term is constructed, and the system confidence level is reflected by the product of the normalized difference between the real-time sensing data and the LSTM predicted value. A compensation factor derived from the Lyapunov stability theory is introduced to balance the influence of the prediction deviation. Finally, the feature enhancement term and the fault tolerance compensation term are linearly superimposed, and after Min-Max normalization, a 0-1 standardized regulation index is output. The environmental regulation index is mapped to a three-dimensional control vector of color temperature adjustment amount, light intensity modulation amount, and ultraviolet compensation intensity through a PID controller.

[0030] Preferably, the environmental regulation index is specifically expressed as: , ERI represents the environmental regulation index, w i is the feature self-attention weight, α i is the non-linear enhancement index, β is the fault tolerance compensation factor, x j is the real-time sensing parameter, is the parameter prediction value.

[0031] Preferably, the intelligent decision control module adopts a multi-objective optimization strategy to realize the dynamic regulation of lighting parameters. Based on the normalized output value of the environmental regulation index, the continuous index is discretized into three control variables of color temperature, illuminance, and ultraviolet compensation through a three-dimensional mapping function: the color temperature adjustment adopts the piecewise linear interpolation method to achieve the CIE1931 chromaticity coordinate matching in the range of 2700K - 6500K, the illuminance control constructs a logarithmic response curve based on the Weber-Fechner law to achieve the physiological brightness adaptation of 10 - 2000 lux, and the ultraviolet compensation dynamically adjusts the UV-LED drive current through a PID controller to maintain the set exposure threshold; at the same time, a dynamic priority mechanism is embedded, and when the human activity entropy exceeds the set critical value, the illuminance response rate is automatically increased, and the color temperature adjustment function is locked during the abnormal color deviation coefficient; the real-time update period of the control parameters is 200ms, and the rolling horizon optimization strategy is used to fuse the LSTM prediction result with the real-time environmental regulation index, and the fuzzy PID algorithm is used to eliminate the actuator lag effect.

[0032] Preferably, the intelligent lighting control method based on a multi-modal sensor includes: S1: Multi-modal data acquisition: It is composed of a heterogeneous sensor network, and data acquisition is realized by synchronously collecting multi-dimensional environmental physical quantities. It integrates optical, thermal, acoustic, spectral, ultraviolet, and human motion sensing components, and standardizes and preprocesses the original data using a unified timing protocol; S2: Feature calculation: Feature extraction is performed based on the original sensor data stream. By establishing a mathematical model, the original signal is converted into six quantization parameters representing the dynamic characteristics of the environment, and the sliding window processing and frequency domain analysis techniques are used to ensure the calculation accuracy; S3: Spatiotemporal coupling analysis: Model the spatiotemporal correlation between the light intensity change rate and the dynamic heat distribution index, construct a light-thermal coupling non-linear equation, and output the environmental dynamic index as the benchmark input for subsequent multi-modal verification; S4: Multi-modal verification: Adopt a dual-mode conditional architecture to verify the environmental coordination, trigger the sound-light coordination or color difference protection mechanism based on the color deviation threshold, and use the hyperbolic tangent function and inverse proportion constraint to achieve the cross-verification and dynamic adjustment of multi-modal data; S5: Biological rhythm compensation: Integrate the ultraviolet cumulative exposure amount and the human activity entropy parameter, construct a dose-response dynamic balance model, and output the biological impact coefficient to synchronously reflect the lighting demand intensity and the biological safety threshold, providing a rhythm compensation basis for decision optimization; S6: Intelligent decision-making control: A hybrid decision-making architecture is constructed using a prediction model and a self-attention mechanism. A fault-tolerant compensation term is generated through real-time data difference analysis, and the weights of feature parameters are dynamically allocated to output an environmental regulation index, driving the lighting terminal to achieve multi-dimensional precise control.

[0033] Technical effects and advantages of the present invention: The multi-modal environmental perception network constructed by the present invention breaks through the limitations of traditional single-dimensional detection. Through spatio-temporal synchronous acquisition of multiple sensors such as visible light arrays, infrared thermal imaging, and acoustic modules, combined with an improved Kriging algorithm, high-precision reconstruction of the temperature field is achieved, and a non-linear coupling model of the light intensity change rate and the dynamic heat distribution index is established, significantly improving the accuracy of environmental state analysis. The feature calculation module uses a composite analysis model of the coefficient of variation and gradient change to effectively capture the transient features of the lighting environment. Combined with the band-pass integration processing technology of voiceprint activity, the collaborative analysis of multi-physical field data is realized, fundamentally solving the problem of control inaccuracy caused by data dimension fragmentation in traditional systems. The biological rhythm dynamic compensation mechanism of the present invention completely revolutionizes the ultraviolet safety control paradigm. Through data fusion of ultraviolet-sensitive elements and human motion sensors, a dose threshold normalization model and a Sigmoid safety constraint function are constructed, and a protection boundary for ultraviolet exposure that conforms to the circadian rhythm is established. This mechanism real-time tracks the human activity trajectory and the cumulative dose of ultraviolet rays, and automatically injects a biological safety compensation amount during the color temperature adjustment process, overcoming the technical problem that it is difficult to balance lighting comfort and health safety in existing technologies, and showing unique advantages in sensitive scenarios such as medical lighting and elderly care. The intelligent decision-making module of the present invention deeply integrates LSTM time series prediction and Transformer self-attention mechanism, constructs a hybrid neural network architecture with the memory characteristics of environmental parameters, realizes the dynamic iteration of control strategies through a rolling horizon optimization strategy, and the fuzzy PID algorithm cooperates with a 200ms-level fast response mechanism to achieve precise matching of illuminance while ensuring the smooth transition of color temperature, significantly improving the control real-time performance in complex lighting environments, breaking through the traditional fixed-cycle control mode, and still maintaining stable parameter output in extreme scenarios such as rapid movement of people and sharp changes in lighting, providing a new generation of solutions for intelligent building lighting management. Brief description of the drawings

[0034] Figure 1 It is a schematic diagram of the module structure of the present invention.

[0035] Figure 2 It is a schematic diagram of the complete embodiment structure of the present invention.

[0036] Figure 3 It is a schematic diagram of the method structure of the present invention. Detailed implementation manners

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0038] Referring to Figure 1 - Figure 2 the intelligent lighting control system based on multi-modal sensors shown in the figure, which includes: Multi-modal data acquisition module: composed of a heterogeneous sensor network, it realizes data acquisition by synchronously collecting multi-dimensional physical quantities of the environment, integrates optical, thermal, acoustic, spectral, ultraviolet and human motion perception components, and preprocesses the raw data in a standardized manner using a unified timing protocol.

[0039] The multi-modal data acquisition module consists of six types of heterogeneous sensors to form a networked detection system. The visible light sensor array and the infrared thermal imaging component cooperate to obtain the optical characteristics of the environment. The distributed temperature array constructs the topology of the spatial thermal field monitoring. The high-precision acoustic module captures the sound pressure characteristics of specific frequency bands. The multi-channel spectral sensor analyzes the spectral power distribution. The ultraviolet sensing unit continuously records the radiation dose. The pyroelectric infrared array quantifies the human motion trajectory. All sensing data is synchronously collected and preprocessed through a unified timing protocol.

[0040] The multi-modal data acquisition module uses a TSL2591 high-precision optical sensor array, and 3 groups of sensor nodes are distributed in a distributed manner at the top of the space. Each group of nodes includes visible light and infrared dual channels, and synchronously collects illuminance data at a sampling rate of 1 Hz, and transmits the raw data through the I2C bus. In the preprocessing stage, a sliding window average filter is used, and the 3σ criterion is used for outlier detection to remove mutation noise.

[0041] The multi-modal data acquisition module uses an MLX90640 infrared thermal imaging sensor to construct a 5×5 temperature array. The node spacing is deployed in a grid pattern at a height of 2.5 m with a spacing of 0.5 m×0.5 m, and surface temperature data in the range of -40°C to 300°C is obtained at a sampling rate of 0.2 Hz. The BME280 environmental sensor is used to obtain air temperature and humidity compensation parameters. The improved Kriging method is used for the spatial interpolation algorithm to generate a continuous temperature field, and the dynamic calibration error is controlled within ±0.3°C.

[0042] The multi-modal data acquisition module is configured with an INMP441 digital MEMS microphone array to obtain sound pressure waveform data at a sampling frequency of 48 kHz and a quantization accuracy of 24 bits. An A-weighting filter is preset in the hardware to eliminate low-frequency interference. A 5-second effective audio segment is intercepted every 30 seconds for FFT transformation. The frequency band analysis focuses on the 200 - 4000 Hz human voice characteristic frequency band, and the background noise baseline is dynamically calibrated by the minimum sound pressure level at night.

[0043] The multi-modal data acquisition module deploys an AS7265x six-channel spectral sensor to synchronously collect the spectral irradiance of 18 characteristic wavelengths in the range of 415 - 940 nm with a period of 5 seconds. It is converted into XYZ tristimulus values through the CIE1931 color matching function, and white balance calibration is performed with reference to the D65 standard light source. The McCamy approximation formula is used for color temperature calculation, and an automatic exposure compensation mechanism is enabled when converting the color coordinates to the CIELAB uniform color space.

[0044] The multi-modal data acquisition module uses a GUVA-S12SD ultraviolet-sensitive element, configured with a 280 - 400 nm band-pass optical filter, records the UV index in the range of 0 - 15 at 10-second intervals, enables a temperature compensation algorithm when establishing an exponential decay model for data accumulation, uses a periodic shading calibration method for dark current elimination, and applies a cubic polynomial fitting compensation curve for non-linear correction.

[0045] The multi-modal data acquisition module installs an AMG8833 infrared pyroelectric array sensor, constructs a 16-zone dynamic detection grid with a 4×4 layout, outputs the heat source intensity distribution of each area at a frequency of 0.5 Hz, uses an improved Hungarian algorithm for target matching in motion trajectory tracking, calculates the velocity vector based on the position difference between two adjacent frames, sets the stay determination threshold to a displacement less than 0.2 m for three consecutive frames, and uses Kalman filter smoothing for three-dimensional positioning error compensation.

[0046] Feature calculation module: Performs feature extraction based on the original sensor data stream. By establishing a mathematical model, the original signal is converted into six quantization parameters representing the dynamic characteristics of the environment, and sliding window processing and frequency domain analysis techniques are used to ensure the calculation accuracy.

[0047] The feature parameters extracted by the feature calculation module include the light intensity change rate, dynamic heat distribution index, voiceprint activity, color deviation coefficient, cumulative ultraviolet exposure, and human activity entropy. The light intensity change rate is derived by constructing a composite model of the coefficient of variation and gradient change of the light intensity time series, quantifying the fluctuation intensity with the ratio of the standard deviation to the mean, and strengthening the instantaneous change feature by combining the root mean square of the differences between adjacent sampling points. The dynamic heat distribution index is based on the thermodynamic gradient theory, taking the ratio of the standard deviation of the temperature field to the sensor spacing to characterize the spatial heterogeneity, and calculating the normalized temperature gradient to reflect the change direction. Finally, the maximum value of the two is taken to ensure the capture of significant temperature features. The voiceprint activity is based on the sound energy density principle, performing band-pass integration on the 200 - 4000 Hz human voice frequency band, and suppressing environmental noise interference through frequency domain amplitude weighting and maximum frequency normalization. The color deviation coefficient uses the Euclidean distance to measure the color difference between the measured RGB and the reference value, and introduces the ratio of the correlated color temperature to compensate for the white light reference difference. Its calculation process implies the non-linear conversion of the CIELAB color space. The cumulative ultraviolet exposure is based on the radiation dosimetry principle, using an exponential decay integral model to quantify the time-varying ultraviolet intensity, and the decay coefficient corresponds to the dose memory effect of biological tissues. The human activity entropy combines the information entropy theory and kinematic parameters, characterizing the randomness of movement through the position distribution probability entropy value, and strengthening the dynamic feature by weighting the ratio of the maximum speed to the sampling interval.

[0048] The light intensity change rate is specifically expressed as: , where LIVR represents the light intensity change rate, L is the sequence of instantaneous values of the light sensor, σ is the light standard deviation, μ is the mean, i represents the sampling point index of the light intensity data in the time series, and n represents the total number of sampling points included in the current analysis time window.

[0049] The dynamic heat distribution index is specifically expressed as: , where THDI represents the dynamic heat distribution index, σ_T is the standard deviation of the temperature sensor array, ÑT is the temperature gradient, Dx is the sensor spacing, and T_avg is the average temperature.

[0050] The voiceprint activity is specifically expressed as: , where SVA represents the voiceprint activity, S(f) is the sound pressure frequency domain signal, the integration range is 200 - 4000 Hz, and f_max is the highest effective frequency that can be resolved, specifically half of the sampling frequency.

[0051] The color deviation coefficient is specifically expressed as: , where CDC represents the color deviation coefficient, RGB is the measured value, the subscript 0 is the reference value, and CCT is the correlated color temperature.

[0052] The cumulative ultraviolet exposure is specifically expressed as: , UVIE represents the cumulative ultraviolet exposure, α is the attenuation coefficient corresponding to a 30-minute half-life, t_0 is the starting time point of ultraviolet monitoring, t represents the current time point, and UV(t) represents the instantaneous radiation intensity value measured by the ultraviolet sensor at time t.

[0053] The α is determined based on the biological half-life model. Referring to the recommended limit of ultraviolet exposure by ACGIH, a 30-minute half-life is adopted. , through the formula α≈0.023 min is calculated. -1 .

[0054] The human activity entropy is specifically expressed as: , HAE represents the human activity entropy, x_k is the position distribution probability of the infrared sensor, v_max is the maximum moving speed, Dt is the sampling time interval, and k represents the k-th dynamic detection grid.

[0055] Space-time coupling analysis module: Model the spatio-temporal correlation between the light intensity change rate and the dynamic heat distribution index, construct a photo-thermal coupling non-linear equation, and output the environmental dynamic index as the benchmark input for subsequent multi-modal verification.

[0056] The environmental dynamic index integrates the light intensity fluctuation and the thermal field distribution characteristics through a non-linear coupling model. First, the three-halves power term of the light intensity change rate is used to strengthen the influence weight of high-intensity fluctuations, and the natural logarithm form of the dynamic heat distribution index is introduced in the denominator term to suppress the over-sensitivity at the thermal field steady state. Then, based on the thermodynamic gradient theory, the absolute value of the temperature field spatial gradient is calculated, and the influence of local sudden heat sources is compensated by an exponential decay function. Finally, the light intensity dominant term weighted by the photo-thermal coupling coefficient determined by principal component analysis is combined with the gradient decay coefficient derived from the thermal comfort model to adjust the temperature gradient term, realizing the normalized fusion of the light intensity time series fluctuation and the thermal field spatial heterogeneity.

[0057] The environmental dynamic index is specifically expressed as: , EDI represents the environmental dynamic index, k_1 is the photo-thermal coupling coefficient, k_2 is the gradient decay coefficient, and ÑTHDI is the spatial gradient of the heat distribution index.

[0058] Multi-modal verification module: Adopt a dual-mode conditional architecture to verify the environmental coordination, trigger the sound-light coordination or color difference protection mechanism based on the color deviation threshold, and use the hyperbolic tangent function and inverse proportion constraint to achieve cross-verification and dynamic adjustment of multi-modal data.

[0059] The multi-modal verification module processes the environmental coordination index through a conditional judgment framework. The construction of the environmental coordination index is based on an interactive verification mechanism of the environmental dynamic index and multi-modal parameters. When the color deviation coefficient does not exceed the preset threshold, the modulation intensity of the environmental dynamic index by the voiceprint activity is constrained by the hyperbolic tangent function, and the environmental coordination is reflected through the acoustic-optic coupling effect. When the color deviation coefficient exceeds the threshold, the color difference protection mechanism is triggered, and the environmental dynamic index is dynamically attenuated using an inverse function. The attenuation degree is determined by the ratio of the excess amplitude of the color deviation coefficient to the attenuation factor. This threshold is set according to the CIE color tolerance standard.

[0060] The specific expression of the environmental coordination index is as follows: , where SCI represents the environmental coordination index, λ is the acoustic-optic conversion threshold, δ is the color tolerance threshold, and γ is the color difference attenuation factor.

[0061] The δ is set according to the color difference tolerance standard published by CIE, λ refers to the frequency band weight parameter in the ANSI speech intelligibility test specification, and γ is determined by fitting the experimental data of the color difference comfort mapping in the laboratory environment.

[0062] Biological rhythm compensation module: Integrate the cumulative ultraviolet exposure and the human activity entropy parameter, construct a dose-response dynamic balance model, and output a biological impact coefficient to synchronously reflect the light demand intensity and the biological safety threshold, providing a rhythm compensation basis for decision optimization.

[0063] The biological impact coefficient is constructed by integrating the spatial comfort index, the cumulative ultraviolet exposure, and the human activity entropy. First, a dynamic balance model of ultraviolet dose and light demand is established, and the cumulative ultraviolet exposure is normalized using a dose threshold. When the cumulative exposure exceeds the safety threshold, the light intensity is suppressed by an inverse function. Secondly, the sliding mean and variance of the human activity entropy are introduced to construct an adaptive adjustment term, and the Sigmoid function is used to constrain the deviation of the human activity entropy from the reference value, converting the randomness feature of the activity entropy into a standardized compensation coefficient. Finally, the light demand attenuation factor and the biological activity compensation term are fused in a product form to achieve the non-linear coupling between the biological safety limit of ultraviolet exposure and the light response of human dynamic behavior. Among them, the sliding mean is calculated using a time decay window mechanism, and the variance is dynamically updated through Kalman filtering.

[0064] The specific expression of the biological impact coefficient is as follows: , where BIC represents the biological impact coefficient, η is the ultraviolet safety threshold, μ is the historical sliding mean of HAE, ν is the real-time variance of HAE, and σ is the sigmoid activation function.

[0065] Intelligent Decision Control Module: A hybrid decision-making architecture is constructed using a prediction model and self-attention mechanism. By analyzing the real-time data difference degree, a fault-tolerant compensation term is generated, and the weights of feature parameters are dynamically allocated to output an environmental regulation index, driving the lighting terminal to achieve multi-dimensional precise control.

[0066] The environmental regulation index is constructed by integrating a feed-forward analysis and a feedback correction mechanism through a hybrid architecture. First, a long short-term memory network is used to perform temporal prediction on the biological impact coefficient and historical feature parameters to generate a benchmark value. Then, the spatial correlation characteristics of multi-modal data are extracted by a Transformer encoder, and the decision weights of each parameter are dynamically allocated using the self-attention mechanism. Nonlinear enhancement processing is performed on the biological impact coefficient at the real-time calculation level, and an exponential term optimized by gradient descent is used to adjust the response intensity in different scenarios. At the same time, a fault-tolerant compensation term is constructed, and the system confidence level is reflected by the product of the normalized difference between the real-time sensing data and the LSTM prediction value. A compensation factor derived from the Lyapunov stability theory is introduced to balance the influence of prediction deviation. Finally, the feature enhancement term and the fault-tolerant compensation term are linearly superimposed, and after Min-Max normalization, a 0-1 standardized regulation index is output. The environmental regulation index is mapped to a three-dimensional control vector of color temperature adjustment amount, light intensity modulation amount, and ultraviolet compensation intensity through a PID controller.

[0067] The environmental regulation index is specifically expressed as: , where ERI represents the environmental regulation index, w i is the self-attention weight of the feature, α i is the nonlinear enhancement index, β is the fault-tolerant compensation factor, x j is the real-time sensing parameter, is the parameter prediction value.

[0068] The intelligent decision control module adopts a multi-objective optimization strategy to achieve dynamic regulation of lighting parameters. Based on the normalized output value of the environmental regulation index, the continuous index is discretized into three control quantities of color temperature, illuminance, and ultraviolet compensation through a three-dimensional mapping function: The color temperature adjustment uses piecewise linear interpolation to achieve CIE1931 chromaticity coordinate matching in the range of 2700K - 6500K. The illuminance control constructs a logarithmic response curve according to the Weber-Fechner law to achieve physiological brightness adaptation of 10 - 2000 lux. The ultraviolet compensation dynamically adjusts the UV-LED drive current through a PID controller to maintain the set exposure threshold. At the same time, a dynamic priority mechanism is embedded. When the human activity entropy exceeds the set critical value, the illuminance response rate is automatically increased, and the color temperature adjustment function is locked during abnormal color deviation coefficient. The real-time update period of the control parameters is 200ms. The rolling horizon optimization strategy is used to fuse the LSTM prediction results and the real-time environmental regulation index, and the fuzzy PID algorithm is used to eliminate the actuator lag effect.

[0069] Reference Figure 3, an intelligent lighting control method based on multi-modal sensors, comprising: S1: Multi-modal data acquisition: It is composed of a heterogeneous sensor network, and data acquisition is achieved by synchronously collecting multi-dimensional physical quantities of the environment. It integrates optical, thermal, acoustic, spectral, ultraviolet, and human motion sensing components, and performs standardized preprocessing on the original data using a unified timing protocol; S2: Feature calculation: Feature extraction is performed based on the original data stream of the sensor. By establishing a mathematical model, the original signal is converted into six quantization parameters representing the dynamic characteristics of the environment. The sliding window processing and frequency domain analysis techniques are used to ensure the calculation accuracy; S3: Spatiotemporal coupling analysis: Model the spatiotemporal correlation between the light intensity change rate and the dynamic heat distribution index, construct a light-thermal coupling nonlinear equation, and output the environmental dynamic index as the benchmark input for subsequent multi-modal verification; S4: Multi-modal verification: Adopt a dual-mode conditional architecture to verify the environmental coordination. Trigger the acoustic-optic coordination or color difference protection mechanism based on the color deviation threshold, and use the hyperbolic tangent function and inverse proportion constraint to achieve cross-verification and dynamic adjustment of multi-modal data; S5: Biological rhythm compensation: Integrate the ultraviolet cumulative exposure amount and the human activity entropy parameter, construct a dose-response dynamic balance model, and output the biological impact coefficient to synchronously reflect the lighting demand intensity and the biological safety threshold, providing a basis for rhythm compensation for decision-making optimization; S6: Intelligent decision-making control: Adopt a prediction model and a self-attention mechanism to construct a hybrid decision-making architecture. Generate a fault tolerance compensation term through real-time data difference analysis, dynamically allocate the weights of feature parameters, output the environmental regulation index, and drive the lighting terminal to achieve multi-dimensional precise control.

[0070] The present invention synchronously acquires multi-dimensional environmental physical quantities through a multi-modal data acquisition module. This module deploys a visible light sensor array, an infrared thermal imaging component, a high-precision acoustic module, a multi-channel spectral sensor, an ultraviolet sensitive element, and an infrared pyroelectric array, and uses a unified timing protocol for data synchronization acquisition and preprocessing. Among them, the optical sensor acquires data at a sampling rate of 1 Hz and uses a sliding window average filter. The temperature array obtains the spatial thermal field distribution at a frequency of 0.2 Hz and generates a continuous temperature field through an improved Kriging method. The acoustic module intercepts a 5-second audio segment every 30 seconds for FFT frequency band analysis. The spectral sensor converts the XYZ tristimulus values every 5 seconds and calculates the color temperature. The ultraviolet sensor records the UV index every 10 seconds and applies a temperature compensation algorithm. The human motion sensor tracks the heat source trajectory at a frequency of 0.5 Hz. Subsequently, the feature calculation module extracts six feature parameters, namely the light intensity change rate, the dynamic thermal distribution index, the voiceprint activity, the color deviation coefficient, the cumulative ultraviolet exposure, and the human activity entropy. Among them, the light intensity change rate is derived through a composite model of the coefficient of variation and gradient change. The dynamic thermal distribution index is calculated by combining the standard deviation and gradient of the temperature field. The voiceprint activity implements band-pass integration and noise suppression processing. The color deviation coefficient uses the Euclidean distance to measure the color difference and introduces color temperature compensation. The cumulative ultraviolet amount is calculated based on an exponential decay integral model. The human activity entropy fuses the information entropy and kinematic parameters. The spatio-temporal coupling analysis module constructs a photo-thermal coupling non-linear equation to generate an environmental dynamic index, strengthens the influence of light intensity fluctuations through the three-halves power, and realizes the normalized fusion of photo-thermal features by combining the temperature gradient attenuation coefficient. The multi-modal verification module triggers a dual-mode verification mechanism based on the color deviation threshold. When not exceeding the standard, the hyperbolic tangent function is used to constrain the acoustic-optical coupling effect. When exceeding the standard, a color difference protection mechanism with inverse proportional attenuation is activated. The biological rhythm compensation module integrates the ultraviolet dose and the human activity entropy to construct a biological influence coefficient, and realizes the biological safety limit through dose threshold normalization and Sigmoid function constraint. Finally, the intelligent decision-making control module generates an environmental regulation index through a hybrid architecture of an LSTM network and a Transformer encoder, combines long-term and short-term prediction with self-attention weight allocation, discretizes the index into color temperature, illuminance, and ultraviolet compensation control quantities through a three-dimensional mapping function, and dynamically adjusts the lighting parameters through a fuzzy PID algorithm to achieve multi-dimensional precise control based on real-time data difference analysis and rolling horizon optimization. The system updates the control parameters at a period of 200 ms, and automatically increases the response rate when human activities are abnormal to ensure that the lighting environment dynamically adapts to biological rhythms and spatial requirements.

[0071] Secondly, in the accompanying drawings of the disclosed embodiments of the present invention, only the structures related to the disclosed embodiments are involved. For other structures, reference can be made to the general design. Without conflict, the same embodiment and different embodiments of the present invention can be combined with each other; Finally, the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent lighting control system based on multimodal sensors, characterized in that: include: Multimodal data acquisition module: It is composed of a heterogeneous sensor network, which acquires data by synchronously collecting multi-dimensional physical quantities of the environment, integrates optical, thermal, acoustic, spectral, ultraviolet and human motion sensing components, and uses a unified timing protocol to standardize and pre-process the raw data; Feature calculation module: performs feature extraction based on the original data stream of the sensor. By establishing a mathematical model, the original signal is converted into six quantitative parameters that characterize the dynamic characteristics of the environment. The sliding window processing and frequency domain analysis technology are used to ensure the calculation accuracy. Spatiotemporal coupling analysis module: Model the spatiotemporal correlation between the light intensity change rate and the dynamic thermal distribution index, construct a nonlinear equation for light-thermal coupling, and output the environmental dynamic index as a benchmark input for subsequent multimodal verification; Multimodal verification module: A dual-mode conditional architecture is used to verify environmental coordination, triggering the sound-light coordination or color difference protection mechanism based on the color deviation threshold, and using the hyperbolic tangent function and inverse proportional constraints to achieve cross-validation and dynamic adjustment of multimodal data; Biorhythm compensation module: integrates the cumulative exposure to ultraviolet rays and the entropy parameters of human activity, builds a dose-response dynamic balance model, and outputs the biological impact coefficient to simultaneously reflect the intensity of light demand and the biological safety threshold, providing a basis for rhythm compensation for decision-making optimization; Intelligent decision-making control module: A hybrid decision-making architecture is constructed using a prediction model and a self-attention mechanism. Fault-tolerant compensation items are generated through real-time data difference analysis. The characteristic parameter weights are dynamically assigned to output the environmental control index, driving the lighting terminal to achieve multi-dimensional precise control.

2. The intelligent lighting control system based on multimodal sensors according to claim 1, characterized in that: The multimodal data acquisition module is composed of six types of heterogeneous sensors to form a networked detection system. The visible light sensor array and the infrared thermal imaging component cooperate to obtain the optical characteristics of the environment. The distributed temperature array constructs the space thermal field monitoring topology. The high-precision acoustic module captures the sound pressure characteristics of a specific frequency band. The multi-channel spectral sensor analyzes the spectral power distribution. The ultraviolet sensor unit continuously records the radiation dose. The infrared pyroelectric array quantifies the human body movement trajectory. All sensor data are synchronously collected and preprocessed through a unified timing protocol.

3. The intelligent lighting control system based on multimodal sensors according to claim 1, characterized in that: The characteristic parameters extracted by the characteristic calculation module include light intensity change rate, dynamic heat distribution index, voiceprint activity, color deviation coefficient, cumulative ultraviolet exposure and human activity entropy; the light intensity change rate is derived by constructing a composite model of the coefficient of variation and gradient change of the light intensity time series, quantifying the fluctuation intensity by the ratio of the standard deviation to the mean, and combining the root mean square of the difference between adjacent sampling points to enhance the instantaneous change characteristics; The dynamic heat distribution index is based on the thermodynamic gradient theory. It takes the ratio of the standard deviation of the temperature field to the sensor spacing to characterize the spatial heterogeneity. At the same time, it calculates the normalized temperature gradient to reflect the direction of change. Finally, the maximum value of the two is taken to ensure the capture of significant temperature features. The voiceprint activity is based on the principle of sound energy density. It implements bandpass integration on the 200-4000Hz human voice frequency band, and suppresses environmental noise interference through frequency domain amplitude weighting and maximum frequency normalization processing. The color deviation coefficient uses Euclidean distance to measure the color difference between the measured RGB and the reference value, and introduces the correlated color temperature ratio to compensate for the white light reference difference. Its calculation process implies the nonlinear transformation of CIELAB color space. The cumulative exposure to ultraviolet light is based on the principle of radiation dosimetry, and the exponential decay integral model is used to quantify the time-varying ultraviolet intensity. The attenuation coefficient corresponds to the dose memory effect of biological tissues. The entropy of human activity integrates information entropy theory and kinematic parameters, characterizes the randomness of movement through the position distribution probability entropy value, and strengthens the dynamic characteristics by weighting the ratio of maximum speed to sampling interval.

4. The intelligent lighting control system based on multimodal sensors according to claim 1, characterized in that: The environmental dynamic index integrates the light intensity fluctuation and thermal field distribution characteristics through a nonlinear coupling model. First, the square-third term of the light intensity change rate is used to strengthen the influence weight of high-intensity fluctuations. The natural logarithm form of the dynamic heat distribution index is introduced into the denominator to suppress the excessive sensitivity of the thermal field in steady state. Then, the absolute value of the spatial gradient of the temperature field is calculated based on the thermodynamic gradient theory, and the influence of local mutation heat sources is compensated by the exponential decay function. Finally, the light intensity dominant term is weighted by the light-heat coupling coefficient determined by principal component analysis, and the temperature gradient term is adjusted in combination with the gradient attenuation coefficient derived from the thermal comfort model, so as to realize the normalized fusion of light intensity time series fluctuations and thermal field spatial heterogeneity.

5. The intelligent lighting control system based on multimodal sensors according to claim 1, characterized in that: The multimodal verification module processes the environmental coordination index through a conditional judgment architecture. The construction of the environmental coordination index is based on the interactive verification mechanism of the environmental dynamic index and the multimodal parameters. When the color deviation coefficient does not exceed the preset threshold, the hyperbolic tangent function is used to constrain the modulation intensity of the voiceprint activity on the environmental dynamic index, and the environmental coordination is reflected through the acoustic-optical coupling effect. When the color deviation coefficient exceeds the threshold, the color difference protection mechanism is triggered, and the environmental dynamic index is dynamically attenuated using an inverse proportional function. The degree of attenuation is determined by the ratio of the color deviation coefficient exceeding the standard to the attenuation factor. The threshold is set according to the CIE color tolerance standard.

6. The intelligent lighting control system based on multimodal sensors according to claim 1, characterized in that: The biological impact coefficient is constructed by integrating the spatial comfort index, the cumulative exposure of ultraviolet rays and the entropy of human activity. First, a dynamic balance model of ultraviolet dose and light demand is established, and the cumulative exposure of ultraviolet rays is normalized by using a dose threshold. When the cumulative exposure exceeds the safety threshold, the light intensity is suppressed by an inverse proportional function. Secondly, the sliding mean and variance of human activity entropy are introduced to construct an adaptive adjustment term. The Sigmoid function is used to constrain the deviation of human activity entropy from the baseline value, and the random characteristics of activity entropy are converted into a standardized compensation coefficient. Finally, the light demand attenuation factor and the biological activity compensation term are fused in the form of a product to achieve the nonlinear coupling between the biosafety limit of ultraviolet exposure and the light response of human dynamic behavior. The sliding mean calculation adopts a time-attenuated window mechanism, and the variance is dynamically updated through Kalman filtering.

7. The intelligent lighting control system based on multimodal sensors according to claim 1, characterized in that: The environmental regulation index is constructed by integrating feedforward analysis and feedback correction mechanism through a hybrid architecture. First, a long short-term memory network is used to perform time series prediction on biological impact coefficients and historical characteristic parameters to generate benchmark values. Then, the spatial correlation characteristics of multimodal data are extracted through the Transformer encoder, and the decision weights of each parameter are dynamically allocated using the self-attention mechanism. In real-time computing, the biological impact coefficient is enhanced nonlinearly, and the exponential term optimized by gradient descent is used to adjust the response intensity in different scenarios. At the same time, a fault-tolerant compensation item is constructed. The system confidence level is reflected by the product of the normalized difference between the real-time sensor data and the LSTM prediction value. The compensation factor derived from the Lyapunov stability theory is introduced to balance the influence of the prediction deviation. Finally, the feature enhancement item and the fault-tolerant compensation item are linearly superimposed, and a 0-1 standardized control index is output after Min-Max normalization. The environmental control index is mapped into a three-dimensional control vector of color temperature adjustment amount, light intensity modulation amount and ultraviolet compensation intensity through the PID controller.

8. The intelligent lighting control system based on multimodal sensors according to claim 1, characterized in that: The intelligent decision-making control module adopts a multi-objective optimization strategy to realize dynamic control of lighting parameters. Based on the normalized output value of the environmental control index, the continuous index is discretized into three control quantities: color temperature, illumination, and ultraviolet compensation through a three-dimensional mapping function: the color temperature adjustment adopts piecewise linear interpolation to achieve CIE1931 chromaticity coordinate matching in the range of 2700K-6500K. The illumination control constructs a logarithmic response curve based on the Weber-Fechner law to achieve physiological brightness adaptation of 10-2000lux. The ultraviolet compensation dynamically adjusts the UV-LED drive current through the PID controller to maintain the set exposure threshold; at the same time, a dynamic priority mechanism is embedded. When the entropy of human activity exceeds the set critical value, the illumination response rate is automatically increased, and the color temperature adjustment function is locked during the abnormal color deviation coefficient. The real-time update period of the control parameters is 200ms. The rolling time domain optimization strategy is adopted to fuse the LSTM prediction results with the real-time environmental control index, and the fuzzy PID algorithm is used to eliminate the hysteresis of the actuator.

9. An intelligent lighting control method based on a multimodal sensor, used for using the intelligent lighting control system based on a multimodal sensor according to any one of claims 1 to 8, characterized in that: include: S1: Multimodal data acquisition: It consists of a heterogeneous sensor network, which acquires data by synchronously collecting multi-dimensional physical quantities of the environment, integrates optical, thermal, acoustic, spectral, ultraviolet and human motion sensing components, and uses a unified timing protocol to standardize the raw data preprocessing; S2: Feature calculation: Perform feature extraction based on the original data stream of the sensor. By establishing a mathematical model, the original signal is converted into six quantitative parameters that characterize the dynamic characteristics of the environment. The sliding window processing and frequency domain analysis technology are used to ensure the calculation accuracy. S3: Spatiotemporal coupling analysis: Model the spatiotemporal correlation between the light intensity change rate and the dynamic thermal distribution index, construct a nonlinear equation for light-thermal coupling, and output the environmental dynamic index as a benchmark input for subsequent multimodal verification; S4: Multimodal verification: A dual-mode conditional architecture is used to verify environmental coordination, triggering the sound-light coordination or color difference protection mechanism based on the color deviation threshold, and using the hyperbolic tangent function and inverse proportional constraints to achieve cross-validation and dynamic adjustment of multimodal data; S5: Biological rhythm compensation: Integrate the cumulative exposure to ultraviolet rays and the entropy parameters of human activity, build a dose-response dynamic balance model, output the biological impact coefficient to synchronously reflect the light demand intensity and the biological safety threshold, and provide a rhythm compensation basis for decision optimization; S6: Intelligent decision control: A hybrid decision architecture is constructed using a prediction model and a self-attention mechanism. Fault-tolerant compensation items are generated through real-time data difference analysis. The characteristic parameter weights are dynamically assigned to output the environmental control index, driving the lighting terminal to achieve multi-dimensional precise control.

Images