CN-121985449-A - Human factor illumination self-adaptive adjusting method based on multi-biological feature fusion and dynamic learning

CN121985449ACN 121985449 ACN121985449 ACN 121985449ACN-121985449-A

Abstract

The invention relates to the technical field of intelligent illumination control and artificial engineering, in particular to an artificial illumination self-adaptive adjusting method based on multi-biological feature fusion and dynamic learning, the method utilizes a fusion network dynamic weighting based on an attention mechanism to generate accurate user state estimation by synchronously collecting multisource biological characteristics such as brain electricity, eye movement, heart rate variability, facial thermal imaging and the like of a user. On the basis, the system maintains and continuously updates an individual illumination response model library for each user, and combines the reinforcement learning framework to perform online exploration optimization while utilizing the historical optimal illumination strategy, thereby generating highly personalized illumination adjustment parameters. After adjustment, the model is continuously improved through closed-loop verification, and quick start and long-term rhythm adaptation of a new user are realized by meta learning. The intelligent light environment adjusting device solves the problems of single perception, lack of individuation and self-evolution capability in the prior art, and achieves intelligent light environment adjustment of comprehensive perception, continuous optimization and deep fitting of human bodies.

Inventors

YU XIN
HE JIANFENG
XU JIE
GUO LANHUI

Assignees

东华理工大学

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. The artificial illumination self-adaptive adjusting method based on multi-biological feature fusion and dynamic learning is characterized by comprising the following steps of: S1, synchronously collecting at least three different types of biological feature original data of a target user, and preprocessing and feature extraction to obtain a multi-mode biological feature vector with aligned time sequences; s2, carrying out multi-level dynamic feature fusion on the multi-mode biological feature vector to generate a dynamically weighted user state vector, and mapping the dynamically weighted user state vector into quantized estimation of the current multi-dimensional physiological and psychological state of the user; S3, generating a personalized lighting strategy based on a dynamic learning mechanism, namely inquiring a personal lighting response model library of a user to acquire a historical optimal strategy base according to the current user state, and introducing exploration based on reinforcement learning, carrying out online optimization by utilizing a weighing mechanism to generate a final lighting parameter instruction; S4, adjusting the lighting equipment according to the lighting parameter instruction, collecting biological characteristic data after adjustment, performing closed-loop verification, and feeding back and updating a verification result to the personal lighting response model library; S5, analyzing cross-user commonality rules and long-term rhythm changes of the users based on the meta learning model, and being used for cold start recommendation of new users and fine adjustment of long-term illumination adjustment targets.
2. The method of claim 1, wherein in step S1, the biometric types include at least three of physiological signals reflecting autonomic nervous system activity, electroencephalogram EEG features reflecting central nervous state, eye movement features reflecting visual fatigue and cognitive load, and facial thermal imaging features.
3. The method according to claim 1, wherein in step S2, the multi-level dynamic feature fusion includes performing primary feature level fusion on the same type of features, performing intermediate level decision level fusion on different types of features by using a network based on an attention mechanism, dynamically learning contribution weights of the features to specific state inference, and performing high-order state mapping on the fused feature vectors to output quantized state estimation.
4. A method according to claim 3, wherein the attention mechanism network is capable of adaptively adjusting feature weight allocation based on environmental context information.
5. The method according to claim 1, wherein in step S3, the reinforcement learning framework uses a user state vector as a state space, uses an adjustment amount of an illumination parameter as an action space, and uses subjective feedback of a user or automatic evaluation based on improvement of a biometric state as a return signal.
6. The method of claim 1 or 5, wherein the reinforcement learning framework employs a near-end policy optimization or deep Q network algorithm.
7. The method of claim 1, further comprising the step of starting a safety illumination mode, adjusting the light to a preset relaxation mode and issuing a reminder when the fused biometric indicates that the user is in a preset abnormal state.
8. A system for implementing the method of any one of claims 1-7, comprising: the multi-biological feature perception module is used for acquiring various biological feature original data; The data processing and fusion computing module is used for executing feature extraction, multi-level dynamic feature fusion and user state mapping; the dynamic learning and strategy engine module is used for maintaining and updating the personal illumination response model library and executing illumination strategy generation and optimization based on reinforcement learning; the lighting control execution module is used for controlling the lighting equipment according to the lighting strategy instruction; the user interaction and feedback module is used for receiving subjective feedback of a user; And the closed loop verification module is used for verifying the lighting adjustment effect and feeding back the lighting adjustment effect to the dynamic learning and strategy engine module.
9. The system of claim 8, further comprising a meta-learning server for performing cross-user commonality analysis and long-term rhythm prediction and data interaction with the dynamic learning and policy engine module.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any of claims 1-7.

Description

Human factor illumination self-adaptive adjusting method based on multi-biological feature fusion and dynamic learning Technical Field The invention relates to the technical field of intelligent illumination control and artificial engineering, in particular to an artificial illumination self-adaptive adjusting method based on multi-biological feature fusion and dynamic learning. Background Along with the development of health illumination and intelligence, human-induced illumination (Human-CENTRICLIGHTING) aims to dynamically adjust light environment (such as color temperature, illumination and spectrum) to meet physiological rhythm, psychological state and operation requirements of people, so that health, comfort and efficiency are improved. The prior art is mostly based on single or simple parameters, such as: The preset mode based on time/scene is to switch the illumination mode according to a time schedule or a fixed scene (such as reading and resting), and lack response to the real-time state of the individual. The illumination is automatically adjusted according to the ambient light sensor based on the adjustment of the ambient sensor, but the actual physiological and psychological feelings of the person are not considered. Based on preliminary feedback of a single biological signal, the "stress" or "relaxed" state is roughly judged, for example using Heart Rate Variability (HRV) or Galvanic Skin Response (GSR), and the light is adjusted. The method has obvious defects of single signal, easy interference, rough state judgment, lack of personalized learning, incapability of adapting to the difference of illumination response of different individuals, static solidification of an adjustment strategy and incapability of evolving along with habit or long-term rhythm change of a user. Therefore, there is a need for an intelligent illumination adjustment method that can more accurately and fully sense the status of a user, and can continuously learn and personally evolve. Disclosure of Invention In order to overcome the defects in the prior art, the invention provides a human factor illumination self-adaptive adjustment method based on multi-biological feature fusion and dynamic learning, so as to solve the problems in the background art. The invention provides a human factor illumination self-adaptive adjusting method based on multi-biological feature fusion and dynamic learning, which comprises the following steps: S1, synchronously collecting at least three different types of biological feature original data of a target user, and preprocessing and feature extraction to obtain a multi-mode biological feature vector with aligned time sequences; s2, carrying out multi-level dynamic feature fusion on the multi-mode biological feature vector to generate a dynamically weighted user state vector, and mapping the dynamically weighted user state vector into quantized estimation of the current multi-dimensional physiological and psychological state of the user; S3, generating a personalized lighting strategy based on a dynamic learning mechanism, namely inquiring a personal lighting response model library of a user to acquire a historical optimal strategy base according to the current user state, and introducing exploration based on reinforcement learning, carrying out online optimization by utilizing a weighing mechanism to generate a final lighting parameter instruction; S4, adjusting the lighting equipment according to the lighting parameter instruction, collecting biological characteristic data after adjustment, performing closed-loop verification, and feeding back and updating a verification result to the personal lighting response model library; S5, analyzing cross-user commonality rules and long-term rhythm changes of the users based on the meta learning model, and being used for cold start recommendation of new users and fine adjustment of long-term illumination adjustment targets. Further, in step S1, the biometric types include at least three of a physiological signal reflecting an autonomic nervous system activity, an electroencephalogram EEG feature reflecting a central nervous state, an eye movement feature reflecting visual fatigue and cognitive load, and a facial thermal imaging feature. Further, in step S2, the multi-level dynamic feature fusion comprises primary feature level fusion of the same type of features, intermediate decision level fusion of different types of features by using a network based on an attention mechanism, dynamic learning of contribution weights of the features to specific state inference, and high-order state mapping of the fused feature vectors to output quantized state estimation. Further, the attention mechanism network can adaptively adjust feature weight allocation according to environmental context information. Further, in step S3, the reinforcement learning framework uses the user state vector as a state space, uses the adjustment amount of the illumination parameter