Natural-based healing space and light environment construction method

Abstract

The invention discloses a natural light shadow-based healing space and light environment construction method, and belongs to the technical field of space environment engineering. According to the method, environment, physiology, interaction and photovoltaic energy storage data are collected through a multi-source sensor, the data are input into a multi-target collaborative decision-making model after preprocessing and feature extraction, and light and shadow control, energy scheduling and multi-sensory collaborative instructions are generated; the execution layer is linked with components such as a photovoltaic ceramic top and the like for execution, and the feedback layer monitors in real time and dynamically optimizes model parameters. According to the invention, cooperation of emotion adaptation, energy efficiency and personalized multi-sensory experience is realized, the healing effect and the energy utilization efficiency are remarkably improved, and the method is suitable for light environment construction of various healing spaces.

Description

Technical Field

[0001] This invention belongs to the field of space environment engineering technology, specifically relating to a method for constructing a nature-based healing space and light environment. Background Technology

[0002] As people place increasing importance on mental health, healing spaces have been widely used in medical rehabilitation, relaxation, and stress reduction. The lighting environment, as a core element of the healing experience, directly determines the therapeutic effect through its rational design. However, existing lighting environment construction technologies for healing spaces have significant shortcomings: most use fixed lighting parameters and rely on manually preset modes. This makes it impossible to capture emotional fluctuations in real time based on users' physiological signals and dynamically adjust the lighting scheme, and it also lacks precise adaptation to users' personalized preferences. Consequently, users with different emotional states and preferences find it difficult to obtain a targeted healing experience.

[0003] Meanwhile, existing technologies lack a coordinated mechanism between photovoltaic energy storage systems and light environment control, often relying on a single energy supply model. This fails to fully utilize the clean characteristics of photovoltaic energy and neglects to scientifically balance energy supply and load demand, frequently resulting in energy waste or diminished therapeutic effects due to insufficient energy. Furthermore, the lack of a unified synchronous control mechanism for multi-sensory modules such as light, sound, and aromatherapy leads to hardware response delays and clock asynchrony, causing discontinuities in the sensory experience and hindering the creation of an immersive therapeutic atmosphere. These combined problems prevent existing technologies from simultaneously achieving targeted therapeutic effects, efficient energy utilization, and immersive sensory experiences, severely limiting the application value and development potential of therapeutic spaces. Summary of the Invention

[0004] To address the aforementioned problems in the existing technology, this invention provides a method for constructing a nature-based healing space and light environment.

[0005] The objective of this invention can be achieved through the following technical solutions: A method for constructing a nature-based healing space and light environment, characterized in that the healing space includes a perception layer, a decision-making unit, an execution layer, and a feedback layer, and the light environment construction method includes: Ambient light quality data, user physiological data, user interaction data, and photovoltaic energy storage system status data are collected through multi-source sensors in the perception layer. After the algorithm decision unit preprocesses and extracts features from the collected data, it inputs the data into the multi-objective collaborative decision model to generate a set of control instructions that integrates light and shadow control, energy scheduling, and multi-sensory collaboration. The multi-objective collaborative decision model includes an emotion-light and shadow adaptation module, an energy load collaborative scheduling module, and a personalization and cross-modal synchronization module, and the three modules operate in conjunction with each other. Based on the control instruction set, the execution layer coordinates the photovoltaic ceramic top, the carved filter device, the energy storage system, and the multi-sensory module to perform light environment construction, energy scheduling, and multi-sensory collaborative control. The feedback layer monitors the operational status and effect data of the healing space in real time through sensors, and transmits the monitoring data back to the algorithm decision-making unit to dynamically optimize the parameters of the multi-objective collaborative decision-making model, thereby realizing the closed-loop iteration of the light environment.

[0006] In a further embodiment of the present invention, in the data acquisition step: Ambient light quality data includes illuminance, color temperature, spectral distribution, temperature and humidity, ambient noise, and weather parameters; User physiological data includes heart rate, brain waves, pupil diameter, dwell time, and posture change data; Photovoltaic energy storage system status data includes real-time power generation, energy storage battery state of charge, load power consumption, and power generation efficiency data; User interaction data includes user preferences and feedback data regarding lighting, audio, and fragrance. The data preprocessing process in the algorithm decision-making step includes: removing outliers, filling missing values, and aligning the time sequence of the collected multi-source data, and then mapping the data to the [0, 1] interval through Z-score standardization to eliminate dimensional differences.

[0007] In a further embodiment of the present invention, the feature extraction process in the algorithm decision step includes: Extract time-domain and frequency-domain features from user physiological data; Statistical features were extracted from ambient light quality data; Extracting trend features from the status data of photovoltaic energy storage systems; The emotion-based lighting and shadow adaptation module uses a pre-set emotion-based lighting and shadow mapping library and extracted user physiological feature data to classify emotions through a BP neural network model. Based on the classification results, it generates a basic lighting and shadow recipe that adapts to the user's current emotional state. The basic lighting and shadow recipe includes the illuminance range, color temperature range, and change period parameters of the lighting and shadow. The basic lighting and shadow recipe serves as the priority determination basis for the energy load collaborative scheduling module and the core reference parameter for the personalized and cross-modal synchronization module.

[0008] In a further embodiment of the present invention, the energy load coordinated scheduling module dynamically allocates energy based on a time-series power generation prediction model and a load demand model: The time-series power generation prediction model uses an LSTM-GRU hybrid neural network. It takes preprocessed historical power generation data, weather parameters and photovoltaic module status data as input, optimizes the model parameters through sliding window training and five-fold cross-validation, and outputs the predicted power generation value for the next 1-6 hours. The load demand model is based on the gradient boosting tree algorithm. It predicts the current load demand by learning the mapping relationship between historical load data and light and shadow control parameters and multi-sensory module operating parameters. When the predicted power generation cannot meet the current load demand, the energy load collaborative scheduling module uses a multi-objective optimization algorithm to solve the Pareto optimal solution for the energy supply sufficiency rate and the healing effect achievement rate, and outputs the light and shadow parameter adjustment command to the emotional light and shadow adaptation module to balance energy supply and healing effect.

[0009] In a further embodiment of the present invention, the personalization and cross-modal synchronization module specifically includes: The user interaction data is analyzed by a user preference learning model. The model integrates a user-based collaborative filtering algorithm and a Transformer time series model to extract user behavior sequence features and generate personalized preference vectors, thereby obtaining personalized optimization coefficients and adjusting the basic lighting and shadow recipe. By using a unified timestamp synchronization mechanism and a timestamp calibration algorithm to achieve clock synchronization, and then using a delay compensation algorithm to correct hardware response delay, changes in light and shadow parameters are synchronized with adjustments to audio and fragrance parameters. Synchronization errors are dynamically corrected using a PID control algorithm.

[0010] In a further embodiment of the present invention, the photovoltaic ceramic top of the execution layer integrates a photovoltaic power generation module and an electronically controlled dimming film, and adjusts the transmittance and spectral distribution by receiving light and shadow control commands; the carved filter device achieves speed adjustment through motor drive, and forms a dynamic natural light and shadow effect in conjunction with the photovoltaic ceramic top; the generation process of the light and shadow control commands includes: based on the basic light and shadow formula and personalized optimization coefficients, mapping emotional features and environmental features into specific transmittance adjustment values ​​and spectral distribution parameters through a fuzzy control algorithm.

[0011] In a further embodiment of the present invention, the energy storage system includes an energy storage battery pack and an energy distribution unit, which receives energy dispatch instructions to realize power generation storage and load power distribution; the multi-sensory module includes an audio adjustment module, an aroma release module and a noise suppression module, which respectively receive corresponding control instructions to adjust the background music BPM, aroma release intensity and noise suppression level; the generation process of the energy dispatch instructions includes: based on the time-series power generation forecast value and the load demand forecast value, allocating the energy storage charging power, the load power supply power and the backup power start-up threshold through an integer linear programming algorithm.

[0012] In a further embodiment of the present invention, the feedback optimization step includes short-term optimization and long-term optimization: Short-term optimization is based on real-time monitoring of user physiological data and light and shadow effect data. Gradient descent method is used to fine-tune light and shadow parameters, multi-sensory synchronization accuracy and energy allocation ratio. The optimization objective function is healing effect score + energy self-sufficiency rate - synchronization error. The optimization cycle is in minutes. Long-term optimization is based on accumulated user interaction data and system operation data. Incremental training algorithms are used to update user preference profiles, transfer learning algorithms are used to optimize the emotion and light-shadow mapping library, and adaptive learning rates are used to adjust the parameters of the time-series power generation prediction model. The optimization cycle is daily or weekly.

[0013] In a further embodiment of the present invention, the unified timestamp synchronization mechanism uses the synchronization clock signal output by the algorithm decision unit, combined with the delay prediction model to predict the hardware response delay, so that the synchronization error of light and shadow parameter adjustment, audio BPM change and fragrance release intensity adjustment does not exceed the preset threshold; the user preference learning model updates the user preference weight in real time through an online learning algorithm, and when new user interaction data is added, a forgetting factor is used to balance the weight ratio of new and old data.

[0014] In a further embodiment of the present invention, the multi-source sensor in the sensing layer includes ambient light quality, physiological sensing, interactive acquisition, and photovoltaic status sensor: the ambient light quality sensor is distributed and transmits data through a designated bus; the physiological sensor supports both contact and non-contact dual-mode acquisition, and improves accuracy through data fusion; the interactive acquisition module is deployed in key areas, taking into account both active input and passive capture; the photovoltaic status sensor is attached to the photovoltaic module; all sensors are equipped with an automatic calibration mechanism, which corrects deviations by comparing with standard reference data, and triggers an alarm when the deviation exceeds the standard.

[0015] This invention has at least the following beneficial effects: 1. By using a backpropagation neural network to classify user emotions and generate appropriate basic lighting and shadow formulas, user emotions can be effectively guided towards a positive direction, thus improving the targeted nature of the therapy.

[0016] 2. Based on the LSTM-GRU hybrid model and multi-objective optimization algorithm, the energy supply and load demand are dynamically balanced to achieve Pareto optimality of energy self-sufficiency rate and therapeutic effect achievement rate. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart of a method provided in one embodiment of the present invention. Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0020] Please refer to Figure 1 In one embodiment, the present invention provides a method for constructing a nature-based healing space and light environment. The healing space includes a perception layer, a decision-making unit, an execution layer, and a feedback layer. The method for constructing the light environment includes: Ambient light quality data, user physiological data, user interaction data, and photovoltaic energy storage system status data are collected through multi-source sensors in the perception layer. After the algorithm decision unit preprocesses and extracts features from the collected data, it inputs the data into the multi-objective collaborative decision model to generate a set of control instructions that integrates light and shadow control, energy scheduling, and multi-sensory collaboration. The multi-objective collaborative decision model includes an emotion-light and shadow adaptation module, an energy load collaborative scheduling module, and a personalization and cross-modal synchronization module, and the three modules operate in conjunction with each other. Based on the control instruction set, the execution layer coordinates the photovoltaic ceramic top, the carved filter device, the energy storage system, and the multi-sensory module to perform light environment construction, energy scheduling, and multi-sensory collaborative control. The feedback layer monitors the operational status and effect data of the healing space in real time through sensors, and transmits the monitoring data back to the algorithm decision-making unit to dynamically optimize the parameters of the multi-objective collaborative decision-making model, thereby realizing the closed-loop iteration of the light environment.

[0021] In this embodiment, the perception layer serves as the core of the entire system's data input, and its design focuses on achieving comprehensive and accurate acquisition of multi-source information. Specialized sensing devices are employed to address different types of data acquisition needs: ambient light quality sensors capture key physical parameters affecting the lighting environment experience, comprehensively reflecting the overall state of the spatial lighting environment; physiological sensors focus on users' physiological signals, indirectly mapping their emotional state and comfort level, providing objective evidence for emotional adaptation; interactive acquisition sensors capture users' active operations and passive behavioral feedback, uncovering users' subjective preferences; and photovoltaic status sensors monitor the real-time operating status of the energy supply system, providing data support for energy dispatch. Through collaborative work, these various sensors overcome the limitations of single data dimensions, achieving comprehensive perception of the three core dimensions of environment, users, and energy, ensuring sufficient and comprehensive data sources for subsequent decision-making processes.

[0022] The algorithm decision-making unit is the brain of the system. First, the data preprocessing stage addresses the heterogeneity, noise, and incompleteness of multi-source data. Through a series of standardized processing procedures, it removes abnormal data caused by equipment interference and environmental fluctuations during the data collection process, fills in any missing values ​​that may have appeared during data collection, and aligns data collected at different time series to ensure data consistency and integrity. Based on this, standardization processes map data of different dimensions and scales to a unified interval, eliminating the interference of dimensional differences on subsequent model calculations and laying the foundation for feature extraction and model input.

[0023] The feature extraction stage employs targeted extraction methods based on the characteristics of different data types: the time-domain and frequency-domain features of user physiological data reflect the dynamic changes and inherent rhythms of physiological signals, serving as a key basis for judging the user's emotional state; the statistical features of ambient light quality data summarize the overall characteristics and changing trends of the light environment; and the trend features of photovoltaic energy storage systems help predict changes in energy supply. By accurately extracting the core features of various data types, the raw data is transformed into feature vectors with decision-making value, improving the efficiency and accuracy of subsequent model calculations.

[0024] The three modules of the multi-objective collaborative decision-making model follow a logic of linkage, coordination, and mutual constraint: The emotion-light-shadow adaptation module serves as the core foundation, based on a pre-set emotion-light-shadow mapping library. This library, constructed through extensive theoretical research and practical verification, establishes a scientific correspondence between emotional states and light-shadow parameters. Combined with extracted user physiological characteristics, it uses a neural network model to accurately classify the user's current emotional state, thereby generating a basic light-shadow formula adapted to that emotional state. This formula provides a benchmark for subsequent energy scheduling and personalized optimization. The energy load collaborative scheduling module, based on the basic light-shadow formula, dynamically balances energy efficiency and therapeutic effects by combining the predicted results of energy supply and load demand. When energy supply cannot meet the current load demand, it adjusts the light-shadow parameters through optimization algorithms to ensure that the core therapeutic effect can still be maintained under energy constraints. The personalized and cross-modal synchronization module, based on user interaction data, mines user personalized preferences, optimizes and adjusts the basic light-shadow formula, and achieves collaborative operation of light-shadow with multi-sensory modules such as audio and fragrance through a synchronization mechanism, enhancing the user's immersion and experience. The three modules do not operate independently, but rather interact through data to ensure that the decision-making results can simultaneously meet the multi-dimensional needs of emotional adaptation, energy efficiency, personalization, and multi-sensory synchronization.

[0025] The core function of the execution layer is to translate the control commands output by the decision-making unit into actual physical actions. Its design hinges on the coordinated linkage and precise response of each component. The photovoltaic ceramic roof, as a core component in constructing the light environment, integrates both energy harvesting and light and shadow adjustment. By receiving light and shadow control commands, it adjusts its own light transmission characteristics and spectral distribution to actively regulate natural light. The carved filter device works in synergy with the photovoltaic ceramic roof, dynamically adjusting the spatial distribution of light to simulate the dynamic effects of natural light and shadow, creating a more immersive light environment. The energy storage system, as the core of energy supply, stores and distributes energy according to energy dispatch commands. It stores excess energy when supply is sufficient and releases stored energy when supply is insufficient, ensuring the stable operation of each execution component. The multi-sensory module adjusts parameters such as audio, fragrance, and noise suppression according to corresponding control commands, working in synergy with the light and shadow environment to construct a multi-dimensional, immersive healing scene. Each execution component, through precise response to control commands, achieves coordinated linkage between light and shadow, energy, and multi-sensory experience, translating the optimization goals of the decision-making unit into actual spatial experiences.

[0026] The feedback layer is designed to achieve dynamic iterative optimization of the system. Sensors monitor the system's operational status and therapeutic effects in real time, including actual parameters of the light environment, real-time physiological feedback from the user, and energy consumption of each component. This monitoring data is transmitted back to the algorithm decision-making unit in real time. The decision-making unit compares the monitored data with the expected goals to determine the deviation between the current system operating state and the optimal state. Based on the deviation, the parameters of the multi-objective collaborative decision-making model are dynamically adjusted, including the emotional light-shadow mapping relationship, energy scheduling strategies, and personalized preference weights, so that the model's output control commands better match actual needs. This closed-loop optimization mechanism ensures that the system can continuously learn data patterns during operation, constantly adjust its parameters, and continuously improve the effect of the light environment construction, adapting to dynamic changes in user state and environmental conditions.

[0027] In a further embodiment of the present invention, in the data acquisition step: Ambient light quality data includes illuminance, color temperature, spectral distribution, temperature and humidity, ambient noise, and weather parameters; User physiological data includes heart rate, brain waves, pupil diameter, dwell time, and posture change data; Photovoltaic energy storage system status data includes real-time power generation, energy storage battery state of charge, load power consumption, and power generation efficiency data; User interaction data includes user preferences and feedback data regarding lighting, audio, and fragrance. The data preprocessing process in the algorithm decision-making step includes: removing outliers, filling missing values, and aligning the time sequence of the collected multi-source data, and then mapping the data to the [0, 1] interval through Z-score standardization to eliminate dimensional differences.

[0028] In this embodiment, the collection of ambient light quality data revolves around the core physical characteristics of the light environment and surrounding environmental conditions. These parameters collectively determine the basic state of the light environment and are an important basis for light and shadow adjustment. At the same time, surrounding environmental parameters also indirectly affect the user's comfort and therapeutic experience, so comprehensive capture is necessary. The collection of user physiological data focuses on core physiological indicators that reflect the user's emotional state and physical comfort level. These indicators are key to judging the user's current state and achieving emotional adaptation. By analyzing this data, the user's degree of adaptation to the light environment and emotional changes can be indirectly perceived, providing objective support for adjusting the light and shadow formula.

[0029] The collection of status data from photovoltaic energy storage systems aims to monitor the real-time operational status and capacity of the energy supply system, including parameters of key aspects such as energy generation, storage, and consumption. This data forms the basis for coordinated energy load scheduling, helping the system predict the matching of energy supply and demand and adjust scheduling strategies in advance. The collection of user interaction data considers both proactive and reactive user feedback. Proactive feedback directly reflects user preferences, while reactive behavior indirectly reflects potential user needs. Analyzing this data allows for precise identification of personalized user preferences, providing a basis for personalized optimization. The collection of various data types is not conducted in isolation but rather based on the inherent relationships between different data dimensions, achieving collaborative collection of multi-dimensional data. This ensures that data corroborates and complements each other, providing a comprehensive and high-value data source for subsequent processing and decision-making.

[0030] The core objective of data preprocessing is to improve data quality, eliminate interfering factors, establish correlations and consistency between data points, and lay the foundation for subsequent feature extraction and model input. Specifically, this includes: The core principle of outlier removal is to identify and remove abnormal data that deviates from the normal data distribution range. These abnormal data are mostly caused by non-systematic factors such as sensor malfunctions and sudden environmental interference, and do not reflect the true state changes. By employing identification methods based on data distribution patterns and statistical characteristics, it is determined whether the data conforms to the normal trend and fluctuation range. Data exceeding the reasonable range is identified as outliers and removed, thus avoiding interference from abnormal data in subsequent model training and decision-making, and ensuring the authenticity of the data.

[0031] The missing value imputation stage addresses data loss issues that may arise during data acquisition due to signal interruptions or temporary equipment malfunctions. Its core principle is based on the temporal characteristics and correlations of the data, using appropriate interpolation or prediction methods to fill in missing data points. The imputation process fully considers the time-series patterns of the data and its correlation with other related data to ensure that the imputed data conforms to the overall data trend, maintaining data integrity and consistency, and avoiding incomplete feature extraction or model training bias caused by missing data.

[0032] In the time-series alignment stage, differences in the acquisition frequency and response speed of different types of sensors can lead to inconsistent acquisition time points for data from different dimensions. Directly performing joint analysis can result in data correlation errors. Therefore, the core principle of time-series alignment is to adjust the acquired data from different time series based on a unified time benchmark, ensuring that various data at the same time point can accurately correspond and establish temporal correlations between data. Through time-series alignment, it is ensured that in subsequent analysis, various types of data can be correlated within the same time dimension, accurately uncovering the inherent patterns between data and improving the accuracy of model decision-making.

[0033] The Z-score standardization process addresses the issue of dimensional differences in multi-source data. Its core principle is based on the statistical characteristics of the data, transforming data of different dimensions and scales into standardized data under a unified standard. By calculating the statistical features of the data, the original data is mapped to a fixed interval, eliminating the influence of data dimensions and making different types of data comparable and additive. This standardization process not only avoids unreasonable weight allocation in the model due to differences in data scale, but also improves the model's training efficiency and convergence speed, ensuring that the model can fairly and accurately utilize various types of data for decision-making.

[0034] In a further embodiment of the present invention, the feature extraction process in the algorithm decision step includes: Extract time-domain and frequency-domain features from user physiological data; Statistical features were extracted from ambient light quality data; Extracting trend features from the status data of photovoltaic energy storage systems; The emotion-based lighting and shadow adaptation module uses a pre-set emotion-based lighting and shadow mapping library and extracted user physiological feature data to classify emotions through a BP neural network model. Based on the classification results, it generates a basic lighting and shadow recipe that adapts to the user's current emotional state. The basic lighting and shadow recipe includes the illuminance range, color temperature range, and change period parameters of the lighting and shadow. The basic lighting and shadow recipe serves as the priority determination basis for the energy load collaborative scheduling module and the core reference parameter for the personalized and cross-modal synchronization module.

[0035] In this embodiment, the core value of user physiological data lies in reflecting the dynamic changes in the user's emotional and physical states. This type of data has obvious temporal and rhythmic characteristics; therefore, feature extraction focuses on time-domain and frequency-domain features. Time-domain features can reflect the changing patterns of physiological signals over time, such as the mean, variance, and trend changes of the signal, directly reflecting the stability and fluctuations of the physiological state. Frequency-domain features, on the other hand, mine the frequency distribution characteristics of the signal by performing frequency domain transformation on the physiological signal. Under different emotional states, the frequency components of the physiological signal will exhibit specific changing patterns, and these frequency-domain features can indirectly map the user's emotional changes.

[0036] The core value of photovoltaic energy storage system status data lies in predicting changes in energy supply trends and providing support for energy dispatch. The key to this type of data is its changing patterns over time; therefore, feature extraction focuses on trend characteristics. By analyzing the growth or decline trends, rates of change, and fluctuation amplitudes of the data, we can accurately grasp the dynamic changes in energy supply, providing crucial references for the energy load collaborative dispatch module and ensuring the foresight and rationality of energy dispatch strategies.

[0037] The emotion classification process employs a backpropagation (BP) neural network model, which possesses powerful nonlinear mapping capabilities and can effectively uncover the complex correlation between physiological characteristics and emotional states. The model is trained based on a pre-defined emotional light and shadow mapping library, constructed through extensive theoretical research, physiological and psychological principles, and practical data validation. This library contains a scientific correspondence between different emotional states and corresponding light and shadow parameters. During operation, the model uses extracted user physiological characteristic data as input and, through multi-layer iterative calculations of the neural network, accurately classifies the user's current emotional state. The classification result directly determines the core direction of the light and shadow formulation.

[0038] In the basic lighting and shadow formula generation stage, the core basis is the emotion classification results, combined with the scientific correspondence in the emotion lighting and shadow mapping library, to generate a basic lighting and shadow parameter combination adapted to the user's current emotional state. The basic lighting and shadow formula contains core lighting and shadow parameters designed according to physiological and psychological principles. In different emotional states, the human body's comfort perception and healing needs for lighting and shadow differ. For example, the lighting and shadow parameters needed to soothe emotions differ significantly in their core characteristics from those needed to stimulate emotions. By precisely setting these parameters, the basic lighting and shadow formula ensures a positive guiding and regulating effect on the user's emotions.

[0039] Furthermore, the design of the basic lighting formula fully considers the operational needs of subsequent modules, clearly defining its functional positioning as the priority determination basis for the energy load collaborative scheduling module and the core reference parameter for the personalized and cross-modal synchronization modules. For the energy load collaborative scheduling module, the basic lighting formula represents the core requirement of emotional adaptation and is a priority target that needs to be guaranteed during energy scheduling. When energy supply and healing effects conflict, adjustments are made based on the core requirements of the basic lighting formula. For the personalized and cross-modal synchronization modules, the basic lighting formula is the benchmark framework for personalized optimization. Personalized adjustments must be made without deviating from the core requirement of emotional adaptation. At the same time, cross-modal synchronization also needs to be centered on the changing rhythm of the lighting formula to ensure the synergy of multi-sensory experiences.

[0040] The emotion-based lighting and shadow adaptation module does not operate independently, but rather works closely with the other two modules through a basic lighting and shadow recipe. The generated basic lighting and shadow recipe provides a core benchmark for the entire multi-objective collaborative decision-making model. The energy load collaborative scheduling module prioritizes this recipe and dynamically adjusts its operations under energy constraints to ensure that emotion adaptation needs are met to the greatest extent possible. The personalization and cross-modal synchronization module, based on this recipe, optimizes it in conjunction with user preferences and achieves multi-sensory synchronization to enhance the user experience. This collaborative mechanism ensures that emotion adaptation remains the core objective of the entire decision-making process, with the achievement of other objectives revolving around this core, thus avoiding a one-sided approach to the decision-making process.

[0041] In a further embodiment of the present invention, the energy load coordinated scheduling module dynamically allocates energy based on a time-series power generation prediction model and a load demand model: The time-series power generation prediction model uses an LSTM-GRU hybrid neural network. It takes preprocessed historical power generation data, weather parameters and photovoltaic module status data as input, optimizes the model parameters through sliding window training and five-fold cross-validation, and outputs the predicted power generation value for the next 1-6 hours. The load demand model is based on the gradient boosting tree algorithm. It predicts the current load demand by learning the mapping relationship between historical load data and light and shadow control parameters and multi-sensory module operating parameters. When the predicted power generation cannot meet the current load demand, the energy load collaborative scheduling module uses a multi-objective optimization algorithm to solve the Pareto optimal solution for the energy supply sufficiency rate and the healing effect achievement rate, and outputs the light and shadow parameter adjustment command to the emotional light and shadow adaptation module to balance energy supply and healing effect.

[0042] In this embodiment, the input data of the model undergoes standardization and time-series alignment during preprocessing to ensure data consistency and validity. Historical power generation data reflects the inherent power generation characteristics and long-term trends of the photovoltaic energy storage system. Weather parameters directly affect the power generation efficiency of photovoltaic modules and are key external factors causing fluctuations in power generation. Photovoltaic module status data reflects the impact of changes in the equipment's own performance on power generation capacity. By inputting these three types of core data into the model, the combined effects of internal characteristics, external environment, and equipment status on power generation can be comprehensively considered.

[0043] During model training, a sliding window training method is employed to adapt to dynamic changes in data distribution, ensuring the model learns the latest data patterns in a timely manner. Simultaneously, five-fold cross-validation is used to optimize model parameters, effectively preventing overfitting and improving the model's generalization ability and prediction accuracy. The model's output provides a predicted power generation value for a future period, offering a clear forward-looking reference for energy dispatch and enabling energy dispatch to shift from passive response to proactive prediction.

[0044] The core design goal of the load demand model is to accurately predict the current energy load demand of the healing space and establish a scientific correlation between load demand and system operating parameters. This model is based on the gradient boosting tree algorithm, which has powerful feature interaction capture and nonlinear fitting capabilities, effectively uncovering the complex mapping relationship between multiple input parameters and load demand.

[0045] The core of the model's training lies in learning the intrinsic relationship between historical load data and lighting control parameters, as well as the operating parameters of multi-sensory modules. Lighting control parameters directly determine the operating status of lighting modules such as photovoltaic ceramic roofs and carved filter devices; different lighting adjustment needs correspond to different energy consumption levels. Multi-sensory module operating parameters determine the operating intensity of modules such as audio, fragrance, and noise suppression, which are also key factors affecting load demand. Through deep learning of these parameters and historical load data, the model can establish accurate mapping relationships, enabling scientific prediction of current load demand.

[0046] When the time-series power generation forecasting model predicts that future power generation cannot meet the current load demand forecast, the multi-objective optimization balance mechanism is activated. Its core principle is to use scientific algorithms to solve for the Pareto optimal solution of energy supply sufficiency rate and healing effect achievement rate, so as to achieve a dynamic balance between the two.

[0047] The core design of the multi-objective optimization algorithm lies in using energy supply sufficiency rate and healing effect achievement rate as two core optimization objectives to construct a reasonable objective function. Energy supply sufficiency rate reflects the degree to which energy supply meets load demand, directly affecting the stability of system operation; healing effect achievement rate reflects the degree to which the lighting and multi-sensory experience adapts to the user's emotions, representing the core value of the healing space. Because there is a certain coupling and constraint relationship between the two objectives, they cannot both achieve absolute optimality simultaneously. Therefore, the multi-objective optimization algorithm is used to find the Pareto optimal solution, that is, without significantly reducing the performance of one objective, it is impossible to further improve the performance of the other objective, thus forming the optimal decision set.

[0048] The energy load collaborative scheduling module and the emotion-based lighting and shadow adaptation module are linked through dynamic command interaction. The basic lighting and shadow recipes generated by the emotion-based lighting and shadow adaptation module provide a core reference for load demand forecasting, while the adjustment commands from the energy load collaborative scheduling module provide energy constraints for optimizing the lighting and shadow recipes. This two-way linkage mechanism ensures that the generation of lighting and shadow recipes considers both emotion adaptation needs and energy supply capacity, forming a closed loop in the decision-making process of the entire system and avoiding the one-sidedness of decisions made by a single module.

[0049] In a further embodiment of the present invention, the personalization and cross-modal synchronization module specifically includes: The user interaction data is analyzed by a user preference learning model. The model integrates a user-based collaborative filtering algorithm and a Transformer time series model to extract user behavior sequence features and generate personalized preference vectors, thereby obtaining personalized optimization coefficients and adjusting the basic lighting and shadow recipe. By using a unified timestamp synchronization mechanism and a timestamp calibration algorithm to achieve clock synchronization, and then using a delay compensation algorithm to correct hardware response delay, changes in light and shadow parameters are synchronized with adjustments to audio and fragrance parameters. Synchronization errors are dynamically corrected using a PID control algorithm.

[0050] In this embodiment, the core design goal of the user preference learning model is to accurately mine users' personalized preferences, providing a scientific basis for optimizing lighting and shadow recipes. This model integrates a user-based collaborative filtering algorithm with a Transformer time-series model, fully leveraging the complementary advantages of both algorithms to achieve in-depth analysis of user interaction data.

[0051] The core principle of user-based collaborative filtering algorithms is to predict the preferences of target users by leveraging the behavioral similarities of user groups. By analyzing the preferences of user groups with similar interactive behaviors for light and shadow, audio, and fragrance, it is possible to uncover group patterns in user preferences, providing a reference for predicting the preferences of target users. This algorithm can effectively utilize collective intelligence to compensate for preference identification biases caused by insufficient interaction data from individual users, thereby improving the stability of preference prediction.

[0052] The Transformer temporal model possesses powerful temporal feature extraction capabilities, effectively capturing the time-series patterns of user interaction data. User preference choices are not isolated events but exhibit temporal correlations, with inherent logical connections between interactive behaviors at different time points. Through its self-attention mechanism, the Transformer model automatically focuses on key nodes and related features in the user interaction sequence, extracting temporal patterns of user behavior and generating preference feature vectors with a time dimension.

[0053] Meanwhile, the fusion of the two algorithms is not a simple superposition, but rather an organic combination of the group preference features mined by the collaborative filtering algorithm and the temporal preference features extracted by the Transformer model through a feature fusion mechanism, generating a comprehensive and accurate personalized preference vector. This preference vector can fully reflect the user's individual preferences, potential needs, and behavioral patterns. Based on this, the obtained personalized optimization coefficients can accurately adjust the basic lighting and shadow recipe generated by the emotion lighting and shadow adaptation module, ensuring that the lighting and shadow recipe not only matches the user's current emotional state but also aligns with the user's personalized preferences.

[0054] The core design goal of the unified timestamp synchronization mechanism is to achieve precise synchronization of parameter adjustments for multi-sensory modules such as light and shadow, audio, and fragrance, eliminating latency and errors and creating an immersive healing atmosphere. This mechanism ensures the continuity and synergy of the multi-sensory experience through a three-tiered linkage process of clock synchronization, latency compensation, and error correction.

[0055] Timestamp calibration algorithms are fundamental to clock synchronization. Because different modules' hardware and control chips have clock differences, directly using their respective local clocks for parameter adjustments will inevitably lead to synchronization deviations. By having the algorithm decision unit output a unified synchronization clock signal, each module performs clock calibration based on this signal, ensuring that all modules' clocks are consistent and laying the foundation for synchronized parameter adjustments. This unified clock reference design fundamentally reduces synchronization errors caused by clock differences.

[0056] The core function of delay compensation algorithms is to correct hardware response latency and signal transmission latency. In actual operation, after receiving control commands, hardware devices need a certain amount of time to complete state adjustments, and delays also occur during signal transmission. These delays can cause the parameter adjustments of multi-sensor modules to be out of sync. Delay compensation algorithms predict the latency characteristics of different modules, sending control commands in advance to modules with larger delays, or dynamically adjusting the command execution time, ensuring precise alignment of parameter adjustments across modules. This algorithm effectively offsets existing latency factors and improves synchronization accuracy.

[0057] PID control algorithms are used to dynamically correct synchronization errors. Despite clock calibration and delay compensation, minor synchronization errors may still occur during actual operation due to environmental interference, changes in equipment status, and other factors. The PID control algorithm dynamically calculates the adjustment amount by monitoring the deviation between the actual execution time of parameter adjustments for each module and the target synchronization time in real time. This allows for fine-tuning of subsequent command transmission times or module response parameters, forming a closed-loop error correction mechanism. This dynamic correction continuously optimizes synchronization accuracy, ensuring that synchronization errors are always controlled within a reasonable range and guaranteeing the continuity of the multi-sensory experience.

[0058] Personalized adjustments and multi-sensory synchronization are coordinated through a unified control logic. After the personalized optimization coefficients adjust the basic lighting and shadow formula, the generated personalized lighting and shadow parameters are synchronously transmitted to a unified timestamp synchronization mechanism. This mechanism takes the change rhythm of personalized lighting and shadow parameters as its core and formulates corresponding audio and fragrance parameter adjustment schemes and synchronization execution plans.

[0059] During parameter adjustments, a synchronization mechanism ensures that changes in audio BPM, fragrance release intensity, and lighting parameters are highly synchronized in terms of timing and rhythm, creating a synergistic effect across the multi-sensory experience. For example, the precise synchronization between the gradual changes in lighting parameters, the rhythmic variations in audio, and the progressive increase in fragrance concentration enhances the user's sensory immersion and improves the therapeutic effect. This synergistic mechanism avoids a disconnect between personalized adjustments and multi-sensory synchronization, ensuring that the personalized experience is fully realized within an immersive multi-sensory atmosphere.

[0060] In a further embodiment of the present invention, the photovoltaic ceramic top of the execution layer integrates a photovoltaic power generation module and an electronically controlled dimming film, and adjusts the transmittance and spectral distribution by receiving light and shadow control commands; the carved filter device achieves speed adjustment through motor drive, and forms a dynamic natural light and shadow effect in conjunction with the photovoltaic ceramic top; the generation process of the light and shadow control commands includes: based on the basic light and shadow formula and personalized optimization coefficients, mapping emotional features and environmental features into specific transmittance adjustment values ​​and spectral distribution parameters through a fuzzy control algorithm.

[0061] In a further embodiment of the present invention, the energy storage system includes an energy storage battery pack and an energy distribution unit, which receives energy dispatch instructions to realize power generation storage and load power distribution; the multi-sensory module includes an audio adjustment module, an aroma release module and a noise suppression module, which respectively receive corresponding control instructions to adjust the background music BPM, aroma release intensity and noise suppression level; the generation process of the energy dispatch instructions includes: based on the time-series power generation forecast value and the load demand forecast value, allocating the energy storage charging power, the load power supply power and the backup power start-up threshold through an integer linear programming algorithm.

[0062] In a further embodiment of the present invention, the feedback optimization step includes short-term optimization and long-term optimization: Short-term optimization is based on real-time monitoring of user physiological data and light and shadow effect data. Gradient descent method is used to fine-tune light and shadow parameters, multi-sensory synchronization accuracy and energy allocation ratio. The optimization objective function is healing effect score + energy self-sufficiency rate - synchronization error. The optimization cycle is in minutes. Long-term optimization is based on accumulated user interaction data and system operation data. Incremental training algorithms are used to update user preference profiles, transfer learning algorithms are used to optimize the emotion and light-shadow mapping library, and adaptive learning rates are used to adjust the parameters of the time-series power generation prediction model. The optimization cycle is daily or weekly.

[0063] In this embodiment, the core design objective of short-term optimization is to quickly respond to real-time changes, correct immediate deviations during system operation, and ensure real-time adaptability to the lighting environment. This optimization process is based on real-time monitoring data and uses gradient descent to fine-tune parameters, resulting in a short optimization cycle and fast response speed.

[0064] The input data for short-term optimization comes from real-time monitoring in the feedback layer, including user physiological data and lighting effect data. User physiological data directly reflects the user's current adaptation to the light environment and emotional changes, serving as the core basis for judging the therapeutic effect; lighting effect data reflects the deviation between the actual parameters of the current light environment and the preset target. This real-time data is continuously transmitted to the algorithm decision-making unit, providing immediate basis for short-term optimization.

[0065] The optimization algorithm employs gradient descent, a highly efficient parameter optimization tool that progressively adjusts parameters along the gradient of the objective function to minimize or maximize it. The objective function for short-term optimization is designed as healing effect score + energy self-sufficiency rate - synchronization error. This function comprehensively considers three core indicators: healing effect, energy efficiency, and multi-sensory synchronization accuracy, avoiding performance imbalances caused by optimizing a single objective. The healing effect score is calculated based on user physiological data, the energy self-sufficiency rate reflects energy utilization efficiency, and the synchronization error reflects the accuracy of multi-sensory coordination.

[0066] By fine-tuning lighting parameters, multi-sensory synchronization accuracy, and energy allocation ratios using gradient descent, deviations occurring during real-time operation can be quickly corrected. For example, when user physiological data indicates that the therapeutic effect is not as expected, the algorithm adjusts lighting parameters such as illuminance and color temperature; when synchronization errors increase, the time base for multi-sensory synchronization is fine-tuned; and when energy self-sufficiency decreases, the energy allocation ratio is optimized. This fine-tuning process is fast and accurate, enabling the system to return to its optimal operating state in a short time, ensuring the stability of the real-time experience.

[0067] The core design goal of long-term optimization is to iteratively upgrade the system model based on accumulated data, adapting to long-term changes in user preferences and environmental conditions, and achieving continuous improvement in system performance. This optimization process is based on accumulated historical data, employs various scientific algorithms for model updates, has a long optimization cycle, and emphasizes accumulation and upgrading.

[0068] The input data for long-term optimization includes accumulated user interaction data and system operation data. User interaction data reflects long-term trends in user preferences, while system operation data contains key information such as long-term patterns in environmental conditions and the effectiveness of model parameters. This accumulated data provides a rich basis for model optimization and user profile updates.

[0069] Long-term optimization achieves multi-dimensional upgrades through multi-algorithm collaboration: First, incremental training algorithms are used to update user preference profiles. Incremental training algorithms can update existing preference profiles using new data without retraining the entire model, saving computational resources and capturing changing trends in user preferences in a timely manner, ensuring the long-term accuracy of preference profiles. Second, the emotion-light-shadow mapping library is optimized through transfer learning algorithms. Transfer learning algorithms can utilize existing model knowledge and data patterns to quickly adapt to new application scenarios or changes in data distribution. By combining accumulated operational data with the initial mapping library, the correspondence between emotion and light-shadow parameters is optimized, improving the accuracy of emotion matching. Third, adaptive learning rates are used to adjust the parameters of the time-series power generation prediction model. Adaptive learning rates dynamically adjust the learning step size based on the model's training performance. When the model's prediction accuracy is high, the learning rate is reduced to stabilize the model; when the prediction accuracy decreases, the learning rate is increased to quickly adjust parameters, ensuring that the time-series power generation prediction model can maintain high prediction accuracy over the long term.

[0070] The core advantage of long-term optimization lies in its ability to leverage data accumulation to iteratively upgrade system capabilities, enabling the system to not only adapt to short-term dynamic changes but also continuously adapt to long-term trend changes, achieving long-term stable improvement in core indicators such as healing effects and energy efficiency.

[0071] Short-term optimization and long-term optimization do not operate independently, but rather form a synergistic relationship of immediate correction and long-term accumulation. Short-term optimization addresses real-time deviations through rapid fine-tuning, ensuring that the system maintains good performance at each stage of operation and accumulating high-quality operational data for long-term optimization. Long-term optimization, based on the data accumulated from short-term optimization, performs in-depth optimization of the model and parameters, enhancing the system's core capabilities, and feeds the optimized model parameters back into the short-term optimization process, improving the accuracy and effectiveness of short-term optimization.

[0072] This collaborative mechanism creates a closed loop in the optimization process. Short-term optimization responds to immediate changes, while long-term optimization accumulates core capabilities. The two support and promote each other, ensuring both the stability of the real-time experience and the long-term improvement of system performance. The optimization cycle is designed to adapt to the changing characteristics: short-term optimization cycles match the rapid fluctuations in user emotions and environmental conditions, while long-term optimization cycles match the long-term changes in user preferences and system operating patterns, ensuring the targeting and effectiveness of the optimization strategy.

[0073] In a further embodiment of the present invention, the unified timestamp synchronization mechanism uses the synchronization clock signal output by the algorithm decision unit, combined with the delay prediction model to predict the hardware response delay, so that the synchronization error of light and shadow parameter adjustment, audio BPM change and fragrance release intensity adjustment does not exceed the preset threshold; the user preference learning model updates the user preference weight in real time through an online learning algorithm, and when new user interaction data is added, a forgetting factor is used to balance the weight ratio of new and old data.

[0074] In this embodiment, the core function of the delay prediction model is to predict hardware response latency, which is based on in-depth analysis of historical synchronization data and hardware status data. Historical synchronization data contains the response latency patterns of different hardware modules during past operations, while hardware status data reflects the current operating status of the device. These data together constitute the basis for delay prediction. By learning the mapping relationship between hardware status and response latency, the model can predict the response latency of each module before control commands are sent, providing a reference for adjusting the synchronization clock signal.

[0075] Based on the delay prediction results, the synchronization clock signal output by the algorithm decision unit is adjusted accordingly. For modules with predicted large delays, control commands are sent earlier; for modules with predicted small delays, commands are sent at the normal time. This predictive adjustment reduces synchronization errors caused by hardware response delays at their source. This proactive prediction works synergistically with traditional delay compensation algorithms and PID control algorithms: the delay prediction model enables proactive avoidance, the delay compensation algorithm enables in-process adjustment, and the PID control algorithm enables post-event correction. These three components constitute a three-level error control system, significantly improving synchronization accuracy.

[0076] The threshold setting for synchronization error is based on the sensitivity of the user's sensory experience, ensuring that the error remains within a range imperceptible to the user. The three-level error control system continuously monitors the synchronization error. When the error approaches or exceeds the threshold, it dynamically adjusts the prediction parameters, compensation coefficients, and PID correction parameters to ensure that the synchronization error is always kept below the threshold, guaranteeing the continuity and immersion of the multi-sensory experience.

[0077] The core of optimizing the user preference learning model lies in introducing online learning algorithms and forgetting factors to achieve real-time updates of preference weights and dynamic balance between new and old data, ensuring the timeliness and accuracy of personalized adaptation.

[0078] The core advantage of online learning algorithms lies in their ability to absorb new user interaction data in real time, dynamically updating user preference weights without retraining the entire model. When a user generates a new interaction, the data is transmitted to the model in real time. The online learning algorithm integrates the features of the new data into the existing preference vector through incremental updates, quickly adjusting the preference weights. This ensures that the personalized optimization coefficients can promptly reflect the user's latest preferences, avoiding the lag problem caused by static learning.

[0079] The introduction of the forgetting factor aims to balance the weighting of new and old data, preventing older data from excessively interfering with users' current preferences. The forgetting factor is a coefficient between 0 and 1; during model updates, older data is assigned a smaller weight, while newer data is assigned a larger weight. Over time, the weight of earlier user interaction data gradually diminishes, while the weight of more recent interaction data becomes increasingly prominent, enabling the preference model to dynamically track changing trends in user preferences. This design preserves the core information reflecting users' long-term stable preferences from older data while also responding promptly to new preference changes, achieving a balance between stability and flexibility in preference identification.

[0080] The model update process works in synergy with short-term and long-term optimization: real-time updates of the online learning algorithm support personalized adjustments for short-term optimization, ensuring the adaptability of the immediate experience; weight balancing dominated by the forgetting factor provides a foundation for updating user preference profiles in long-term optimization, enabling long-term optimization to more accurately capture long-term trends in user preferences and achieve continuous optimization of personalized experience.

[0081] In a further embodiment of the present invention, the multi-source sensor in the sensing layer includes ambient light quality, physiological sensing, interactive acquisition, and photovoltaic status sensor: the ambient light quality sensor is distributed and transmits data through a designated bus; the physiological sensor supports both contact and non-contact dual-mode acquisition, and improves accuracy through data fusion; the interactive acquisition module is deployed in key areas, taking into account both active input and passive capture; the photovoltaic status sensor is attached to the photovoltaic module; all sensors are equipped with an automatic calibration mechanism, which corrects deviations by comparing with standard reference data, and triggers an alarm when the deviation exceeds the standard.

[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for constructing a nature-based healing space and lighting environment, characterized in that, The healing space includes a perception layer, a decision-making unit, an execution layer, and a feedback layer; the method for constructing the light environment includes: Ambient light quality data, user physiological data, user interaction data, and photovoltaic energy storage system status data are collected through multi-source sensors in the perception layer. After the algorithm decision unit preprocesses and extracts features from the collected data, it inputs the data into the multi-objective collaborative decision model to generate a set of control instructions that integrates light and shadow control, energy scheduling, and multi-sensory collaboration. The multi-objective collaborative decision model includes an emotion-light and shadow adaptation module, an energy load collaborative scheduling module, and a personalization and cross-modal synchronization module, and the three modules operate in conjunction with each other. Based on the control instruction set, the execution layer coordinates the photovoltaic ceramic top, the carved filter device, the energy storage system and the multi-sensory module to perform light environment construction, energy scheduling and multi-sensory collaborative control. The feedback layer monitors the operational status and effect data of the healing space in real time through sensors, and transmits the monitoring data back to the algorithm decision-making unit to dynamically optimize the parameters of the multi-objective collaborative decision-making model, thereby realizing the closed-loop iteration of the light environment.

2. The method for constructing a nature-based healing space and light environment according to claim 1, characterized in that, In the data acquisition steps: Ambient light quality data includes illuminance, color temperature, spectral distribution, temperature and humidity, ambient noise, and weather parameters; User physiological data includes heart rate, brain waves, pupil diameter, dwell time, and posture change data; Photovoltaic energy storage system status data includes real-time power generation, energy storage battery state of charge, load power consumption, and power generation efficiency data; User interaction data includes user preferences and feedback data regarding lighting, audio, and fragrance. The data preprocessing process in the algorithm decision-making step includes: removing outliers, filling missing values, and aligning the time sequence of the collected multi-source data, and then mapping the data to the [0, 1] interval through Z-score standardization to eliminate dimensional differences.

3. The method for constructing a nature-based healing space and light environment according to claim 1, characterized in that, The feature extraction process in the algorithm decision-making step includes: Extract time-domain and frequency-domain features from user physiological data; Statistical features were extracted from ambient light quality data; Extracting trend features from the status data of photovoltaic energy storage systems; The emotion-based lighting and shadow adaptation module uses a pre-set emotion-based lighting and shadow mapping library and extracted user physiological feature data to classify emotions through a BP neural network model. Based on the classification results, it generates a basic lighting and shadow recipe that adapts to the user's current emotional state. The basic lighting and shadow recipe includes the illuminance range, color temperature range, and change period parameters of the lighting and shadow. The basic lighting and shadow recipe serves as the priority determination basis for the energy load collaborative scheduling module and the core reference parameter for the personalized and cross-modal synchronization module.

4. The method for constructing a nature-based healing space and light environment according to claim 1, characterized in that, The energy load coordinated scheduling module dynamically allocates energy based on a time-series power generation prediction model and a load demand model. The time-series power generation prediction model uses an LSTM-GRU hybrid neural network. It takes preprocessed historical power generation data, weather parameters and photovoltaic module status data as input, optimizes the model parameters through sliding window training and five-fold cross-validation, and outputs the predicted power generation value for the next 1-6 hours. The load demand model is based on the gradient boosting tree algorithm. It predicts the current load demand by learning the mapping relationship between historical load data and light and shadow control parameters and multi-sensory module operating parameters. When the predicted power generation cannot meet the current load demand, the energy load collaborative scheduling module uses a multi-objective optimization algorithm to solve the Pareto optimal solution for the energy supply sufficiency rate and the healing effect achievement rate, and outputs the light and shadow parameter adjustment command to the emotional light and shadow adaptation module to balance energy supply and healing effect.

5. The method for constructing a nature-based healing space and light environment according to claim 1, characterized in that, The personalized and cross-modal synchronization module specifically includes: The user interaction data is analyzed by a user preference learning model. The model integrates a user-based collaborative filtering algorithm and a Transformer time series model to extract user behavior sequence features and generate personalized preference vectors, thereby obtaining personalized optimization coefficients and adjusting the basic lighting and shadow recipe. By using a unified timestamp synchronization mechanism and a timestamp calibration algorithm to achieve clock synchronization, and then using a delay compensation algorithm to correct hardware response delay, changes in light and shadow parameters are synchronized with adjustments to audio and fragrance parameters. Synchronization errors are dynamically corrected using a PID control algorithm.

6. The method for constructing a nature-based healing space and light environment according to claim 1, characterized in that, The photovoltaic ceramic top of the execution layer integrates photovoltaic power generation components and electronically controlled dimming film, and adjusts the transmittance and spectral distribution by receiving light and shadow control commands; the carved filter device achieves speed adjustment through motor drive, and works with the photovoltaic ceramic top to form a dynamic natural light and shadow effect; the generation process of the light and shadow control commands includes: based on the basic light and shadow formula and personalized optimization coefficients, using a fuzzy control algorithm to map emotional features and environmental features into specific transmittance adjustment values ​​and spectral distribution parameters.

7. The method for constructing a nature-based healing space and light environment according to claim 1, characterized in that, The energy storage system includes an energy storage battery pack and an energy distribution unit, which receives energy dispatch instructions to realize power generation storage and load power distribution; the multi-sensory module includes an audio adjustment module, a fragrance release module and a noise suppression module, which respectively receive corresponding control instructions to adjust the background music BPM, fragrance release intensity and noise suppression level. The process of generating the energy dispatch instructions includes: allocating energy storage charging power, load power supply power, and backup power start-up threshold based on time-series power generation forecasts and load demand forecasts using an integer linear programming algorithm.

8. The method for constructing a nature-based healing space and light environment according to claim 1, characterized in that, The feedback optimization steps include short-term optimization and long-term optimization: Short-term optimization is based on real-time monitoring of user physiological data and light and shadow effect data. Gradient descent method is used to fine-tune light and shadow parameters, multi-sensory synchronization accuracy and energy allocation ratio. The optimization objective function is healing effect score + energy self-sufficiency rate - synchronization error. The optimization cycle is in minutes. Long-term optimization is based on accumulated user interaction data and system operation data. Incremental training algorithms are used to update user preference profiles, transfer learning algorithms are used to optimize the emotion and light-shadow mapping library, and adaptive learning rates are used to adjust the parameters of the time-series power generation prediction model. The optimization cycle is daily or weekly.

9. A method for constructing a nature-based healing space and light environment according to claim 1, characterized in that, The unified timestamp synchronization mechanism uses the synchronization clock signal output by the algorithm decision unit, combined with the delay prediction model to predict hardware response delay, so that the synchronization error of light and shadow parameter adjustment, audio BPM change and fragrance release intensity adjustment does not exceed the preset threshold; the user preference learning model updates user preference weights in real time through online learning algorithm, and when new user interaction data is added, a forgetting factor is used to balance the weight ratio of new and old data.

10. A method for constructing a nature-based healing space and light environment according to claim 1, characterized in that, The sensing layer comprises multiple source sensors, including ambient light quality sensors, physiological sensors, interactive acquisition sensors, and photovoltaic status sensors. Ambient light quality sensors are distributed and transmit data via a designated bus. Physiological sensors support both contact and non-contact dual-mode acquisition, with accuracy improved through data fusion. Interactive acquisition modules are deployed in key areas, accommodating both active input and passive capture. Photovoltaic status sensors are integrated with photovoltaic modules. All sensors are equipped with an automatic calibration mechanism, correcting deviations by comparing with standard reference data, and triggering alarms when deviations exceed the limits.

Images