Intelligently-driven dynamic energy-saving adjustment method for sensing illumination

Abstract

The invention relates to the technical field of intelligent lighting, in particular to an intelligent-driven sensing lighting dynamic energy-saving adjusting method, which comprises the following steps of: sensing and reconstructing time-space distribution of outdoor natural light by calling outdoor and indoor environment sensors; in combination with the occupancy state of the office station and the activity type of the user, establishing a lighting demand constraint; activating a linkage acquisition device, and generating a user comfort vector; light comfort perception, illumination demand constraint and a user comfort vector serve as input and are sent into a multi-target cooperative adjustment controller, and adjustment decision vector optimization is carried out with user comfort maintenance, energy consumption minimization and natural light fusion matching as optimization targets; and finally, parameters such as brightness and color temperature of the LED lighting system are dynamically adjusted according to an optimization result, and collaborative improvement of the light environment quality and the energy-saving effect is realized. According to the method, accurate response to natural light changes and self-adaptive adjustment of personalized requirements of users are achieved, and the intelligent level of office lighting and the energy utilization efficiency are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of smart lighting technology, and in particular to a smart-driven, sensing-based dynamic energy-saving adjustment method for lighting. Background Technology

[0002] With the continuous advancement of green building and smart office concepts, the intelligentization and energy conservation of lighting systems have become important directions for optimizing modern office environments. Traditional LED office lighting systems mostly adopt timed control or on / off dimming strategies based on simple illumination thresholds, lacking precise response to dynamic changes in outdoor natural light. This makes it difficult to achieve harmonious integration with the natural light environment, easily leading to over- or under-illumination, affecting employee visual comfort and work efficiency, and resulting in energy waste. In recent years, some intelligent lighting systems have begun to introduce ambient light sensors to achieve a certain degree of adaptive dimming, but their adjustment logic is still limited to local illumination intensity feedback, failing to fully consider the comprehensive impact of individual user differences, actual activity needs, and multi-dimensional environmental factors, making it difficult to balance energy-saving effects and user experience.

[0003] Furthermore, current technologies for utilizing natural light largely remain at the level of static compensation, failing to model and predict the spatiotemporal distribution characteristics of outdoor light. This is especially problematic in scenarios with drastic changes in light, such as sudden shifts in weather or sunrise and sunset, where system response is sluggish and adjustment smoothness is poor. Simultaneously, user comfort perception relies heavily on subjective settings or fixed patterns, lacking a real-time emotion and comfort state recognition mechanism based on multi-source data such as physiological and behavioral data, making it difficult to achieve human-centered, refined lighting control.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide an intelligent-driven, sensing-based dynamic energy-saving adjustment method for lighting, aiming to solve the technical problem that traditional and existing intelligent LED office lighting systems are unable to balance energy-saving effects and user light comfort due to insufficient response to dynamic changes in outdoor natural light and failure to integrate individual user differences and multi-dimensional environmental factors.

[0006] To achieve the above objectives, the present invention provides an intelligent-driven, sensing-based dynamic energy-saving adjustment method for lighting, the method comprising:

[0007] The system uses outdoor light sensors and indoor environmental sensors to perceive the ambient light field, reconstructs the spatiotemporal distribution of outdoor natural light, and uses the spatiotemporal distribution to establish a perception of light comfort.

[0008] Extract the occupancy status and user activity type of the office workstations, and construct lighting requirement constraints based on the occupancy status and user activity type;

[0009] Activate the linkage acquisition device, perform multi-dimensional data perception of the user based on the linkage acquisition device, and establish a user comfort vector using the multi-dimensional data perception results;

[0010] After establishing the adjustment decision vector, the light comfort perception, the lighting demand constraint, and the user comfort vector are used as input data and sent to the multi-objective collaborative adjustment controller to perform adjustment decision vector optimization. The collaborative objectives of the multi-objective collaborative adjustment controller include user comfort maintenance objective, energy consumption minimization objective, and natural light fusion matching objective.

[0011] The energy consumption of the LED office workstation lighting system is dynamically adjusted based on the optimization results of the adjustment decision vector.

[0012] Optionally, establishing light comfort perception using the spatiotemporal distribution includes:

[0013] Perform a historical environment data call to obtain the historical environment dataset;

[0014] After performing weather similarity clustering on the historical environmental dataset, abrupt change time points are predicted based on the time identifiers of the similarity clusters, and abrupt change node prediction result is established.

[0015] Obtain the location data of the office workstation, perform online weather reading based on the location data, and establish online weather reading results;

[0016] The light comfort perception is compensated by using the predicted results of the mutation nodes and the network weather reading results.

[0017] Optionally, the step of compensating for light comfort perception using the mutation node prediction results and the networked weather reading results includes:

[0018] Temporal extraction of ambient light field perception is performed to establish a temporal distribution dataset of outdoor natural light spatiotemporal distribution;

[0019] Using the aforementioned temporal distribution dataset, temporal prediction is performed to establish a perception of light comfort.

[0020] Based on the predicted mutation node results, conflict identification is performed for time-series prediction, and the conflict identification results are used to establish a first compensation feedback.

[0021] The impact analysis of time-series forecasting is performed using the networked weather reading results, and a second compensation feedback is established.

[0022] The first compensation feedback and the second compensation feedback are used to compensate for the perceived light comfort.

[0023] Optionally, the adjustment decision vector includes brightness adjustment amount, color temperature adjustment factor, switching timing offset, dimming smoothness factor, and energy-saving mode migration factor.

[0024] Optionally, the step of sending the control decision vector to the multi-objective coordinated control controller for optimization includes:

[0025] Based on the occupancy status, user activity type, and spatiotemporal distribution of outdoor natural light, workstation scene recognition is performed, and scene recognition results are established.

[0026] The search space is trimmed using the scene recognition results, and the decision vector optimization management is performed based on the trimmed search space.

[0027] Optionally, the adjustment of decision vector optimization management based on the pruned search space includes:

[0028] In each iteration, after initializing the current adjustment decision vector, iterative optimization is performed using a lightweight gradient descent device.

[0029] Perform multiple rounds of iterative evaluation to generate continuous iterative evaluation results;

[0030] If the continuous iterative evaluation results cannot meet the preset convergence threshold, then an auxiliary optimization instruction is generated;

[0031] The auxiliary optimization command is used to invoke the historical scenario energy-saving template for enhanced optimization.

[0032] Optionally, establishing light comfort perception using the spatiotemporal distribution includes:

[0033] Feature extraction is performed using the aforementioned spatiotemporal distribution to establish a feature set. The extracted features include light intensity distribution, color temperature gradient, brightness fluctuation rate, incident azimuth and elevation angles, light spot uniformity, and natural light ratio.

[0034] Establish a comfort label mapping based on the existing office environment comfort dataset;

[0035] Based on the comfort label mapping, a sliding time window aggregation and matching of feature sets is performed to establish light comfort perception.

[0036] Optionally, the dynamic energy consumption adjustment of the LED office workstation lighting system based on the optimization result of the adjustment decision vector includes:

[0037] Establish a dataset of users' physiological states and configure the energy tolerance factor for each physiological state;

[0038] An energy consumption optimization function is constructed using the energy consumption tolerance factor, and energy consumption optimization is performed based on the energy consumption optimization function to adjust the optimization results of the decision vector.

[0039] Dynamic adjustment of LED office workstation lighting system based on energy consumption optimization results.

[0040] Optionally, the linkage data acquisition device includes a workstation camera, a sound acquisition device, and a wearable physiological data acquisition device.

[0041] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing an intelligent-driven sensing lighting dynamic energy-saving adjustment program, wherein when the intelligent-driven sensing lighting dynamic energy-saving adjustment program is executed by a processor, it implements the steps of the intelligent-driven sensing lighting dynamic energy-saving adjustment method as described above.

[0042] This invention provides an intelligent-driven, sensing-based dynamic energy-saving adjustment method for lighting. The method integrates the perception of outdoor natural light spatiotemporal distribution, user activity status recognition, and multi-dimensional physiological behavior data to construct a light comfort perception and lighting demand constraint, achieving comprehensive dynamic perception of the lighting environment and personalized demand modeling. It introduces a user comfort vector and a multi-objective collaborative adjustment mechanism to achieve intelligent trade-offs and optimization decisions between maintaining user comfort, minimizing energy consumption, and matching natural light fusion, significantly improving the personalized service level and energy utilization efficiency of the lighting system. Through optimization of the adjustment decision vector and dynamic dimming execution, the system's responsiveness and adjustment smoothness to complex lighting changes are enhanced, effectively avoiding over-illumination or insufficient illuminance. This ensures visual comfort and work efficiency in office scenarios while achieving refined, adaptive energy-saving control, demonstrating promising application prospects and promotional value. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating an embodiment of the intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting according to the present invention.

[0044] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0045] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0046] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting according to the present invention. An embodiment of the intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting according to the present invention is presented.

[0047] In one embodiment, the intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting includes:

[0048] Step S100: Use outdoor light sensor and indoor environment sensor to perceive the ambient light field, reconstruct the spatiotemporal distribution of outdoor natural light, and use the spatiotemporal distribution to establish light comfort perception.

[0049] The outdoor light sensor can be a sensing device used to collect parameters such as outdoor natural light intensity, color temperature, and incident angle, providing raw data input for reconstructing the spatiotemporal distribution of outdoor natural light. In this embodiment, the outdoor light sensor can be deployed on the building facade or in the lighting area, using photoelectric detection elements to sense solar radiation and diffuse skylight characteristics in real time. For example, the outdoor light sensor can include, but is not limited to, one or more of illuminance sensors, color temperature sensors, and solar azimuth angle sensors. The indoor environment sensor can be a sensing device deployed in the indoor space to sense environmental parameters such as local illumination, temperature, and humidity, assisting in correcting the actual projection effect of the outdoor natural light model indoors and improving the accuracy of light field perception. Furthermore, the indoor environment sensor can be distributed around office workstations to collect indicators such as indoor illuminance, reflected light, and glare. In an exemplary embodiment, the indoor environment sensor can include, but is not limited to, one or more of illuminance meters, glare sensors, and environmental spectrometers. The spatiotemporal distribution of outdoor natural light can describe the dynamic distribution characteristics of natural light in both temporal (e.g., diurnal variation, sudden weather changes) and spatial (e.g., window orientation, distribution of obstructions) dimensions. It can serve as a fundamental input for constructing light comfort perception, reflecting the dynamic impact of natural light on indoor lighting needs. In one specific embodiment, the spatiotemporal distribution of outdoor natural light can be modeled and extrapolated based on outdoor light sensors and building geometry information, combined with meteorological data. Light comfort perception can be a quantitative representation of human visual comfort based on a comprehensive assessment of the spatiotemporal distribution of natural light and the indoor light environment. It can be used to provide comfort constraints that synergize with natural light for adjustment decisions. Furthermore, light comfort perception can be calculated using multidimensional weighted calculations that integrate indicators such as illuminance uniformity, glare index, and color temperature matching. In an exemplary embodiment, light comfort perception may include, but is not limited to, one or more of the following: visual fatigue index, illuminance uniformity score, and color temperature harmony.

[0050] Ambient light field perception can be achieved by simultaneously reading data streams from both outdoor and indoor sensors to construct a joint observation matrix for the indoor and outdoor light environment. Furthermore, this operation can overcome the limitations of a single sensor's perspective and improve the completeness of light field modeling. Reconstructing the spatiotemporal distribution of outdoor natural light can be done by inferring the spatiotemporal projection path of natural light indoors based on sensor data and building BIM models. Further, this reconstruction can be achieved by using ray tracing algorithms to simulate the propagation and reflection of sunlight inside buildings, or by combining historical meteorological data with real-time sensor input to construct a time-series prediction model, thus enabling proactive modeling of dynamic changes in natural light and supporting advance adjustments. Establishing light comfort perception using the spatiotemporal distribution can be achieved by inputting the reconstructed natural light distribution into a comfort evaluation function to output a comprehensive comfort score. Further, this operation can be achieved by weighted calculation based on the CIE glare index and illuminance uniformity formula, or by training a neural network model to directly regress comfort labels from the light distribution image, thereby transforming the physical light environment into human factors engineering-understandable comfort indicators.

[0051] Step S200: Extract the occupancy status and user activity type of the office workstations, and construct lighting requirement constraints based on the occupancy status and user activity type.

[0052] The occupancy status of an office workstation can be binary or continuous status information representing whether a specific workstation is being used. This can be used as a trigger condition for lighting demand constraints, avoiding ineffective lighting in unoccupied areas. In this embodiment, the occupancy status of an office workstation can identify the presence of personnel through means such as infrared pyroelectric sensors, pressure pads, or Wi-Fi positioning. For example, the occupancy status of an office workstation can include, but is not limited to, one or more of the following: static occupancy detection, dynamic entry and exit recognition, and long-term absence judgment. User activity type can be a classification and identification result of the type of task currently being performed by the user. This can be used to distinguish between high-illuminance demand (such as drawing) and low-illuminance demand (such as meetings) scenarios, refining lighting demand constraints. Furthermore, user activity type can be based on cluster analysis of behavioral data such as screen usage status, keyboard and mouse operation frequency, and posture recognition. In a specific embodiment, user activity type can include, but is not limited to, one or more of the following: document reading, video conferencing, and creative design. Lighting demand constraints can be upper and lower limit boundary conditions for the output of the lighting system, jointly determined by the occupancy status and activity type. This can be used to ensure that the lighting output meets the functional requirements of the actual use scenario. In one exemplary embodiment, lighting demand constraints can map activity types to preset illuminance-color temperature demand ranges and enable or disable these ranges based on occupancy status. For example, lighting demand constraints may include, but are not limited to, one or more of minimum illuminance thresholds, maximum color temperature limits, and dynamic response delay limits.

[0053] Extracting the occupancy status and user activity type of office workstations can be achieved by parsing raw signals from the occupancy sensor and behavior recognition module, outputting structured status labels. Furthermore, this operation can achieve precise alignment between lighting requirements and actual usage behavior. Constructing lighting requirement constraints based on occupancy status and user activity type can be done by mapping status labels to a predefined lighting parameter boundary table, generating a dynamic constraint set. Further, this operation can be implemented by matching activity types to ISO / CIE recommended illuminance standards using a lookup table method, or by learning user preferences online to dynamically update constraint boundaries, thereby preventing lighting output from deviating from functional requirements.

[0054] Step S300: Activate the linkage acquisition device, perform multi-dimensional data perception of the user based on the linkage acquisition device, and establish a user comfort vector using the multi-dimensional data perception results.

[0055] The linked data acquisition device can be a multimodal physiological and behavioral data acquisition terminal that works in conjunction with the lighting system, providing individualized physiological and behavioral basis for constructing user comfort vectors. In this embodiment, the linked data acquisition device can integrate wearable devices, desktop cameras, biosignal sensors, etc., and be activated on demand to reduce privacy risks. For example, the linked data acquisition device can include, but is not limited to, one or more of smart bracelets, non-contact heart rate monitors, and eye-tracking cameras. The user's multidimensional data can be a multi-source heterogeneous data set reflecting the user's physiological state and behavioral patterns, which can be used to support individualized comfort state recognition, going beyond subjective questionnaires or fixed settings. Furthermore, the user's multidimensional data can be simultaneously acquired through the linked data acquisition device, including indicators such as heart rate variability, sitting posture angle, and screen viewing time. In a specific embodiment, the user's multidimensional data can include, but is not limited to, one or more of physiological indicator data, behavioral interaction data, and attention concentration data.

[0056] A user comfort vector can be a numerical vector that maps multidimensional data to represent an individual's current comfort state, and can be used as a personalized comfort objective input in multi-objective optimization. In an exemplary embodiment, the user comfort vector can be reduced in dimensionality and standardized into a unified vector space representation using a machine learning model. For example, the user comfort vector may include, but is not limited to, one or more components such as emotional comfort, visual fatigue, and attentional fit. Activating the linked data collection device can be done by starting the physiological and behavioral data collection module as needed, with user authorization. Furthermore, this operation can balance data acquisition and privacy protection, collecting sensitive information only when necessary.

[0057] Multidimensional data perception of users is performed based on linked acquisition devices. This can involve simultaneously acquiring multi-channel signals such as heart rate, posture, and screen interaction, and aligning them over time. Furthermore, this operation can construct a high-dimensional user state representation, supporting individualized modeling. The user comfort vector is then established using the multidimensional data perception results. This can be achieved by compressing the raw data into a fixed-dimensional vector through feature engineering or embedding models. Further, this operation can be implemented through dimensionality reduction and standardization using principal component analysis (PCA), or by learning a low-dimensional latent representation through an autoencoder, thereby transforming raw physiological behavior data into optimizable targets.

[0058] In step S400, after establishing the adjustment decision vector, the light comfort perception, lighting demand constraints, and user comfort vector are sent as input data to the multi-objective collaborative adjustment controller to perform adjustment decision vector optimization. The collaborative objectives of the multi-objective collaborative adjustment controller include user comfort maintenance objective, energy consumption minimization objective, and natural light fusion matching objective.

[0059] The adjustment decision vector can be a set of control instructions containing dimming parameters of each LED lamp (such as brightness, color temperature, and on / off status), which can be used to directly drive the LED office workstation lighting system to perform dynamic dimming. In this embodiment, the adjustment decision vector can be generated by a multi-objective collaborative adjustment controller through an optimization algorithm. For example, the adjustment decision vector can include, but is not limited to, one or more of the following: brightness adjustment component, color temperature adjustment component, and zone on / off component. The multi-objective collaborative adjustment controller can be a computing unit that executes a multi-objective optimization algorithm to generate the adjustment decision vector, which can be used to achieve a dynamic trade-off between comfort, energy saving, and natural light integration. Further, the multi-objective collaborative adjustment controller can receive three types of inputs: light comfort perception, lighting demand constraints, and user comfort vector, and use the Pareto front search method to solve for the optimal solution set. In a specific embodiment, the multi-objective collaborative adjustment controller can include, but is not limited to, one or more of the following: an NSGA-II-based optimizer, a model predictive controller, and a reinforcement learning decision engine. The user comfort maintenance objective can be an optimization objective that keeps the user comfort vector no lower than an individualized threshold during the adjustment process, which can be used to prevent a decrease in comfort due to excessive energy saving. For example, user comfort maintenance targets may include, but are not limited to, one or more of the following: physiological comfort baseline, behavioral adaptation tolerance, and emotional stability range. Energy consumption minimization targets may be optimization objectives that minimize the total power consumption of the system while satisfying other constraints, and can be used to improve energy utilization efficiency and reduce ineffective power consumption. In an exemplary embodiment, energy consumption minimization targets may include, but are not limited to, one or more of the following: instantaneous power minimization, daily cumulative energy consumption minimization, and peak load reduction.

[0060] The natural light fusion matching degree objective can be an optimization objective that measures the degree of coordination and consistency between artificial lighting and natural light in terms of illuminance, color temperature, and directionality, and can be used to enhance the naturalness and visual continuity of the lighting environment. Furthermore, the natural light fusion matching degree objective can include, but is not limited to, one or more of the following: color temperature consistency score, illuminance gradient smoothness, and light and shadow direction matching degree. Establishing the adjustment decision vector can be done by initializing the set of control parameters of the luminaires to be optimized as optimization variables. Furthermore, this operation can provide an operable decision space for multi-objective optimization. Sending light comfort perception, lighting demand constraints, and user comfort vectors as input data to the multi-objective collaborative adjustment controller can be done by transmitting these three types of structured data to the optimization calculation unit via an internal communication bus. Furthermore, this operation can complete multi-source information fusion, providing complete input for collaborative optimization. The multi-objective collaborative adjustment controller performs adjustment decision vector optimization, which can be done by searching for a Pareto optimal solution set while satisfying lighting demand constraints. Furthermore, this operation can be achieved by using the NSGA-II algorithm to solve a multi-objective optimization problem, or by using a weighted summation method to transform the multi-objective problem into a single objective for gradient descent, thereby achieving a dynamic balance between comfort, energy efficiency, and the integration of natural light.

[0061] Step S500: Dynamically adjust the energy consumption of the LED office workstation lighting system based on the optimization results of the adjustment decision vector.

[0062] The LED office workstation lighting system can be a cluster of independently dimmable LED luminaires deployed at office workstations, enabling refined and localized dynamic lighting output. In this embodiment, the LED office workstation lighting system can receive adjustment decision vectors and then perform brightness and color temperature adjustments via DALI or Zigbee protocols. For example, the LED office workstation lighting system can include, but is not limited to, one or more of desktop task lights, ceiling-mounted integrated panel lights, and adjustable color temperature downlights. Dynamic energy consumption adjustment of the LED office workstation lighting system based on the optimization results of the adjustment decision vectors can be achieved by parsing the optimized decision vectors into specific luminaire control commands and issuing them for execution. Furthermore, this operation can achieve smooth, precise, and personalized dynamic dimming, avoiding abrupt changes and over-adjustment.

[0063] Taking a cloudy afternoon office scenario as an example, the intelligent-driven dynamic energy-saving adjustment method for sensing lighting in this embodiment can be as follows: an outdoor light sensor detects that the rapid thickening of cloud cover causes a sudden drop in natural light, while an indoor environmental sensor confirms that the illuminance at the workstation near the window is below 300 lux; the system reconstructs the spatiotemporal distribution of outdoor natural light and determines that this change will last for more than 15 minutes; the light comfort sensing module outputs a warning of decreased comfort; simultaneously, the linked acquisition device detects that a user's heart rate variability has decreased and screen viewing time has increased, and the user comfort vector shows an increase in visual fatigue; the occupancy status is "in The task type is identified as "document proofreading," and the minimum illuminance is set at 500 lux. The multi-objective collaborative adjustment controller integrates three types of inputs and generates an adjustment decision vector under the premise of maintaining the user comfort vector at a threshold, meeting the 500 lux requirement, and minimizing energy consumption. This vector increases the brightness of the LED panel light at the window workstation from 30% to 70% and adjusts the color temperature from 4000K to 5000K to compensate for the color temperature shift of natural light. The system smoothly transitions, avoiding sudden changes in illuminance, ensuring visual clarity for the proofreading task while avoiding excessive supplemental lighting that would waste energy.

[0064] In one embodiment, the perception of light comfort is established using spatiotemporal distribution, including:

[0065] Perform a historical environment data call to obtain the historical environment dataset;

[0066] The historical environment dataset can be a time-series record of past outdoor and indoor lighting, weather, time, and other multi-dimensional environmental parameters stored in the system. It can provide a data foundation for weather similarity clustering and abrupt change prediction, supporting the learning and reproduction of natural light change patterns. In this embodiment, the historical environment dataset can be formed through continuous collection and structured storage of data from outdoor light sensors, indoor environmental sensors, and meteorological interfaces over a long period. Furthermore, the historical environment dataset can include, but is not limited to, one or more of the following: sunny daytime lighting sequences, alternating sunny and cloudy periods, and sunset decay curves. Retrieving the historical environment dataset by executing a historical environment data retrieval can be achieved by searching historical environmental records matching the current season, time period, and weather type from a local database or edge storage. Furthermore, the historical environment dataset can be retrieved by filtering based on time windows, meteorological conditions, or spatial regions, thereby achieving the technical effect of providing high-quality training and comparison samples for subsequent clustering and prediction.

[0067] After performing weather similarity clustering on the historical environmental dataset, abrupt change time points are predicted based on the time identifiers of the similarity clusters, and abrupt change node prediction result is established.

[0068] Weather similarity clustering can be an unsupervised grouping based on the light evolution characteristics in historical environmental datasets, classifying weather segments with similar dynamic patterns into one category. This can be used to identify reusable natural light change patterns and infer the possible evolution path of the current weather. In an exemplary embodiment, weather similarity clustering can employ time-series clustering algorithms to cluster features such as light intensity, rate of change, and color temperature trajectory. For example, weather similarity clustering can include, but is not limited to, one or more of the following: stable sunny day clustering, rapid cloud cover clustering, and dusk dimming clustering. Time stamps can be timestamps marking the occurrence of key events in historical weather segments, which can be used as a reference for predicting future abrupt change times and establishing a time mapping relationship between historical and current scenarios. In a specific embodiment, time stamps can extract the time coordinates corresponding to the start point, peak, or inflection point of light abrupt changes from the clustered weather segments. Further, time stamps can include, but are not limited to, one or more of the following: cloud cover start time, sunset critical point, and light recovery time.

[0069] The prediction of abrupt change timing can be based on matching the current environmental state with historical clustering results to deduce the specific time when drastic changes in illumination may occur in the future. This can be used to predict the timing of natural light abrupt changes and reserve an adjustment window for the lighting system. In this embodiment, the prediction of abrupt change timing can compare the current illumination trajectory with each cluster center, select the most similar class, and predict the abrupt change point based on its time identifier offset. For example, the prediction of abrupt change timing can include, but is not limited to, one or more of the following: cloud cover prediction point, critical point of solar incidence angle, and point of dominance of diffuse light. The prediction result of abrupt change node can be a structured output containing one or more predicted abrupt change timing points and their confidence levels. This can be used as a forward-looking input to compensate for the perception of light comfort and improve the system's response lead. In a specific embodiment, the prediction result of abrupt change node can integrate the abrupt change timing prediction with the clustering matching degree to generate a weighted list of event predictions. Furthermore, the prediction result of abrupt change node can include, but is not limited to, one or more of the following: high-confidence abrupt change warning, low-probability disturbance indication, and multi-peak abrupt change sequence.

[0070] Weather similarity clustering of historical environmental datasets can be achieved by extracting illumination time-series feature vectors and merging data fragments with similar evolution patterns using clustering algorithms. Furthermore, weather similarity clustering of historical environmental datasets can be implemented by using Dynamic Time Warping (DTW) distance metrics combined with K-Means clustering or by using autoencoder embedding followed by spectral clustering, thereby abstracting typical natural light change patterns and supporting pattern-matching prediction. Predicting abrupt change times based on the time identifiers of similar clusters can be achieved by aligning the current illumination trajectory to the center of the most similar cluster and extrapolating the future abrupt change time based on its historical time identifier. Further, abrupt change time prediction based on the time identifiers of similar clusters can be achieved through linear time scaling mapping or state transition prediction based on Hidden Markov Models, thereby enabling early prediction of illumination abrupt events and reducing system response latency. Establishing abrupt change node prediction result can be achieved by integrating the prediction results of multiple candidate clusters, weighting them to generate a list of abrupt change events with confidence levels, thereby providing a structured and interpretable abrupt change early warning input.

[0071] Obtain the location data of the office workstation, read the weather data online based on the location data, and establish the online weather reading results;

[0072] The location data of office workstations can be the geographical coordinates or logical location information of a specific workstation within the building. This data can be used to accurately retrieve external weather forecasts for the corresponding area, improving the spatial relevance of weather readings. In this embodiment, the location data of office workstations can be obtained through BIM model binding, Wi-Fi fingerprint positioning, or user profiles. For example, the location data of office workstations can include, but is not limited to, one or more of latitude and longitude coordinates, floor-room codes, and lighting area classification labels. Networked weather reading can be the act of actively accessing external meteorological service interfaces based on workstation location data to obtain forecast information. This can be used to introduce external prior knowledge to compensate for the lag in local sensor observations. In an exemplary embodiment, networked weather reading can call a third-party API to obtain high-resolution meteorological data for the next 1-2 hours. Furthermore, networked weather reading can include, but is not limited to, one or more of real-time cloud cover queries, solar irradiance forecasts, and sky diffuse ratio predictions.

[0073] Networked weather readings can be structured weather forecast data related to the workstation location returned from external meteorological services. This data can be used to supplement macro-weather trends for light comfort perception, enhancing model robustness. In one specific embodiment, networked weather readings can parse API responses to extract key fields such as cloud cover, solar altitude angle, and direct / diffuse irradiance ratio. For example, networked weather readings can include, but are not limited to, one or more of short-term cloud image forecasts, solar trajectory predictions, and atmospheric transparency indices. Obtaining the location data of the workstation can be achieved by extracting the workstation space identifier from user configuration, building information systems, or equipment registration information, thereby ensuring that external weather information is consistent with actual lighting conditions. Networked weather readings based on location data can be performed by constructing a request containing latitude, longitude, or regional codes to call a meteorological API to obtain short-term forecasts. Furthermore, networked weather readings based on location data can be implemented through periodic polling to update the 60-minute forecast or event-triggered readings, thereby introducing high spatiotemporal resolution external meteorological priors and enhancing forecast coverage. Establishing networked weather reading results can involve parsing JSON / XML data returned by APIs, extracting meteorological parameters related to lighting, and standardizing them. This can achieve the technical effect of forming an external environment representation that can be integrated with local sensing.

[0074] The results of mutation node prediction and online weather readings are used to compensate for the perceived light comfort.

[0075] Compensating for light comfort perception using abrupt change predictions and online weather readings can be achieved by using predicted abrupt change points and external weather trends as correction factors to dynamically adjust the output weights or thresholds of the light comfort perception model. Furthermore, this compensation can be implemented by introducing a predicted abrupt change countdown decay factor into the comfort evaluation function or by constructing a dual-channel fusion network. This allows for a more forward-looking approach to light comfort perception, avoiding the lag caused by relying solely on current measured values.

[0076] For example, in an office setting before a sudden afternoon thunderstorm, the intelligent-driven sensing-based dynamic energy-saving lighting adjustment method of this embodiment could be as follows: The system detects that the current light intensity is slowly decreasing, calls up historical environmental datasets and performs weather-similar clustering, identifying clusters that highly match the "cloud accumulation before a summer afternoon thunderstorm" pattern; the historical time stamp of this cluster shows that it takes an average of 18 minutes from the current light level to complete shading; based on this, the abrupt change point is predicted to be 15 minutes later; simultaneously, the system acquires the location data of the workstation near the window (39.9°N, 116°E). (0.4°), the system reads the short-term forecast from the meteorological bureau and confirms that a strong convective cloud cluster will pass through within 30 minutes, and the cloud cover will increase from 30% to 90%. The sudden change node prediction result and the network weather reading result jointly trigger the light comfort perception compensation mechanism, which lowers the comfort score originally calculated based on the current illuminance of 500 lux, and activates the LED supplementary lighting plan in advance. 10 minutes later, the natural light begins to drop sharply, and the system has smoothly increased the artificial lighting to 600 lux to maintain the total illuminance stable, so as to avoid visual discomfort for employees due to sudden changes in illuminance, and at the same time avoids excessively early lighting to avoid energy waste.

[0077] In one embodiment, compensation for perceived light comfort is achieved using mutation node prediction results and networked weather readings, including:

[0078] Temporal extraction of ambient light field perception is performed to establish a temporal distribution dataset of outdoor natural light spatiotemporal distribution;

[0079] The temporal distribution dataset can be a structured data set that reflects the continuous evolution of outdoor natural light in the spatiotemporal dimension, formed by organizing ambient light field sensing data in chronological order. It can be used as training and inference input for temporal prediction models, supporting dynamic light comfort modeling. In this embodiment, the temporal distribution dataset can be formed into a multidimensional time series by performing sliding window sampling and timestamp alignment on the ambient light field sensing output. For example, the temporal distribution dataset can include, but is not limited to, one or more of the following: illuminance-time series, color temperature-azimuth time series, and spatial gradient evolution series. Temporal extraction of ambient light field sensing data to establish a temporal distribution dataset of the spatiotemporal distribution of outdoor natural light can be achieved by extracting feature vectors from the continuous ambient light field sensing stream at a fixed time granularity, constructing a timestamped multivariate sequence. Furthermore, this operation can be achieved by setting a unified sampling frequency and spatial mapping rules, thereby preserving the complete trajectory of the dynamic evolution of natural light and providing structured input for subsequent temporal modeling.

[0080] Temporal prediction is performed using a time-series distributed dataset to establish a perception of light comfort.

[0081] Utilizing time-series distributed datasets for time-series prediction to establish light comfort perception can be achieved by using time-series models (such as LSTM or Transformer) to predict short-term changes in natural light and mapping them to comfort indices. In an exemplary embodiment, this operation can be implemented by using a recurrent neural network to predict the illuminance-color temperature trajectory for the next 5 minutes and then inputting it into a comfort evaluation function; or by training the time-series model end-to-end to directly output the predicted light comfort value. Furthermore, this operation enables the system to achieve dynamic comfort prediction based on local historical data, but it is susceptible to interference from sudden weather events.

[0082] Conflict identification is performed based on the prediction results of mutation nodes for time series prediction, and the first compensation feedback is established using the conflict identification results.

[0083] Conflict identification can involve comparing the predicted results of mutation nodes with the temporal prediction output to detect inconsistencies in the timing or intensity of the mutation. This can be used to identify prediction biases in local temporal models caused by a lack of prior mutation information. In one specific embodiment, conflict identification can calculate the time difference of predicted events, the difference in rates of change, or the overlap of confidence intervals, and set a threshold to determine whether a conflict exists. For example, conflict identification can include, but is not limited to, one or more of the following: time-shifted conflict, intensity underestimation conflict, and event missing conflict. The conflict identification result can be structured judgment information output from the conflict identification process, including conflict type, location, and temporal impact range. This information can be used to drive the generation of the first compensation feedback to correct local inaccuracies in light comfort perception. Furthermore, the conflict identification result can be obtained by encapsulating the conflict detection index into a weighted correction signal. For example, the conflict identification result can include, but is not limited to, one or more of the following: high-confidence conflict markers, edge-blurred conflict prompts, and multi-point concurrent conflict sequences.

[0084] The first compensation feedback can be an internal correction signal generated based on the conflict identification results to correct temporal prediction biases. It can be used to improve the system's responsiveness to abrupt events not captured by the temporal model. In this embodiment, the first compensation feedback can be obtained by introducing a weighted attenuation of the conflict region or a mutation enhancement factor into the light comfort perception model. For example, the first compensation feedback can include, but is not limited to, one or more of the following: pre-mutation warning gain, prediction lag compensation term, and local smoothing suppression factor. Conflict identification in temporal prediction based on mutation node prediction results can involve aligning and comparing the predicted time point of the mutation node with the derivative extremum point of the current temporal prediction to determine if there is a significant deviation. Further, this operation can measure the similarity of the event sequences by calculating the Hausdorff distance between the two, or constructing a binary classifier to determine whether a prediction conflict has occurred, thereby identifying abrupt events that the local model failed to foresee and triggering the internal correction mechanism. Establishing the first compensation feedback using the conflict identification results can transform the conflict location and intensity into local correction weights for the light comfort perception output. Further, this operation enables the system to specifically compensate for comfort misjudgments caused by the failure to model mutations.

[0085] Impact analysis of time-series forecasts using online weather readings is conducted to establish a second compensation feedback mechanism.

[0086] The impact analysis can assess the degree of disturbance of the networked weather readings to the implicit illumination trend in the current time-series forecast. It can be used to quantify the potential impact of external weather trends on comfort perception, providing a basis for the second compensation. In one specific embodiment, the impact analysis can be implemented by mapping external meteorological parameters (such as cloud cover change rate and solar altitude angle derivative) to an illuminance change sensitivity function. For example, the impact analysis can include, but is not limited to, one or more of irradiance disturbance assessment, color temperature drift risk analysis, and spatial uniformity disruption prediction. The second compensation feedback can be a global correction signal generated based on the impact analysis results and incorporating prior external meteorological information. It can be used to enhance the model's adaptability to changes in external macro-weather conditions and avoid local perception blind spots. In this embodiment, the second compensation feedback can be obtained by dynamically adjusting the baseline threshold or response sensitivity of light comfort perception using meteorological disturbance scores as adjustment factors. For example, the second compensation feedback can include, but is not limited to, one or more of trend advance response gain, long-term attenuation compensation term, and multi-source consistency correction factor.

[0087] Impact analysis of time-series forecasts using online weather data can involve substituting external meteorological parameters into physics or data-driven models to assess the magnitude of their disturbance to the current forecast path. Furthermore, this operation can be achieved by simulating the impact of cloud cover changes on illuminance using radiative transfer models, or by estimating the correlation between meteorological variables and illuminance residuals using regression models. This quantifies the potential interference of external weather trends on local forecasts, supporting global correction. Establishing a second compensation feedback mechanism can involve encoding the impact analysis results into continuous adjustment factors and injecting them into the light comfort perception calculation process. This further enables the system to integrate macro-meteorological priors, improving the model's robustness under complex weather conditions.

[0088] The perception of light comfort is compensated by using the first compensation feedback and the second compensation feedback.

[0089] Compensating for perceived light comfort using first and second compensation feedback can be achieved by applying a weighted superposition or gated fusion of the two types of compensation feedback to the original perceived light comfort output. In an exemplary embodiment, this operation can be implemented by dynamically allocating the contribution weights of the two types of feedback using an attention mechanism, or by introducing local correction terms and global offset terms into the comfort evaluation function, thereby achieving synergistic optimization of intrinsic mutation correction and extrinsic trend fusion, significantly improving prediction accuracy and regulatory foresight.

[0090] For example, in an office setting with cloudy skies before sunset, the intelligent-driven dynamic energy-saving lighting adjustment method of this embodiment could be as follows: The system constructs a time-series distribution dataset based on ambient light field perception and predicts using an LSTM model that the illuminance will slowly decrease to 300 lux in the next 10 minutes, corresponding to light comfort perception remaining within the comfortable range; however, the abrupt change node prediction module identifies the current lighting pattern as matching the "sunset + rapid dissipation of thin clouds" category based on historical clustering, predicting a brief rebound in illuminance after 15 minutes; this abrupt change node conflicts with the monotonic decreasing trend predicted by the LSTM, and the conflict identification module determines it as an "event missing conflict". The system generates a first compensation feedback, injecting a weak upward pulse into the prediction curve. Simultaneously, the online weather readings show that the solar altitude angle will drop below 5° in 20 minutes, with no new cloud systems entering. The impact analysis module assesses that the sunset decay will be steeper than the model predicts, generating a second compensation feedback and lowering the overall comfort benchmark. With the combined effect of these two types of feedback, the light comfort perception predicts in advance that the illuminance will first rise slightly and then drop sharply. Based on this, the multi-objective collaborative adjustment controller slightly reduces artificial lighting in advance to avoid overcompensation when natural light recovers. Subsequently, during the sunset acceleration phase, the LED output is smoothly increased, maintaining a stable total illuminance throughout, which is both energy-saving and ensures visual continuity.

[0091] In one embodiment, the adjustment decision vector includes brightness adjustment amount, color temperature adjustment factor, on / off timing offset, dimming smoothness factor, and energy-saving mode migration factor.

[0092] The adjustment decision vector can be a set of control commands containing multiple independently adjustable lighting parameter dimensions, used to drive the LED office workstation lighting system to achieve multi-dimensional dynamic dimming. In this embodiment, the adjustment decision vector serves as the optimized output of a multi-objective collaborative adjustment controller, supporting a refined trade-off between comfort, energy efficiency, and natural light integration. The brightness adjustment amount can be a continuous variable characterizing the adjustment range of the LED luminous flux output, used to accurately match the current natural light level with the illuminance required by the user's task, avoiding over-illumination or insufficient illuminance. In an exemplary embodiment, the brightness adjustment amount may include, but is not limited to, global brightness offset, local area gain, and dynamic illuminance compensation coefficient. The color temperature adjustment factor can be a parameter controlling the change of the correlated color temperature (CCT) of the light source, reflecting the degree of adjustment of the ratio of warm and cool light, used to dynamically adapt the color temperature according to the circadian rhythm phase or task type, improving the synchronicity of physiological rhythms and cognitive focus. Furthermore, the color temperature adjustment factor may include, but is not limited to, a daytime high color temperature enhancement factor, a nighttime low color temperature soothing factor, and a task-oriented color temperature offset.

[0093] The switching timing offset can be a predictive offset value of the lighting system's start-up and stop times relative to the user's actual arrival or departure time. It is used to achieve a seamless lighting experience by turning on the system earlier or delaying its shutdown, while reducing ineffective energy consumption during unoccupied periods. In one specific embodiment, the switching timing offset may include, but is not limited to, pre-start advance, delayed shutdown hysteresis, and activity prediction window offset. The dimming smoothness factor can be a damping parameter controlling the rate of change in brightness or color temperature. It is used to suppress abrupt changes and flicker effects, maintaining visual continuity and adjustment stability in scenarios with drastic fluctuations in natural light (such as rapidly moving clouds). For example, the dimming smoothness factor may include, but is not limited to, brightness change slope limits, color temperature transition time constants, and multi-lamp synchronization smoothing coefficients. The energy-saving mode migration factor can be a discrete or continuous variable characterizing the system's tendency to switch between different energy efficiency-comfort strategy modes. It is used to dynamically select operating strategies (such as high comfort mode / ultimate energy-saving mode) based on overall energy constraints and real-time comfort status. Furthermore, the energy-saving mode migration factor may include, but is not limited to, comfort priority weight, energy-saving aggressiveness coefficient, and mode switching threshold offset. Combining brightness adjustment, color temperature adjustment factor, switching timing offset, dimming smoothness factor, and energy-saving mode migration factor into an adjustment decision vector can encapsulate five independent control dimensions into a unified high-dimensional vector structure, which can then be input to the controller as optimization variables. Furthermore, this operation can support refined, multi-dimensional, coordinated control of lighting output by constructing a multi-degree-of-freedom adjustment space.

[0094] The multi-objective collaborative adjustment controller optimizes the adjustment decision vector, which includes brightness adjustment, color temperature adjustment factor, switching timing offset, dimming smoothness factor, and energy-saving mode migration factor. This optimization can be achieved by performing a Pareto optimal search on the five-dimensional decision vector while satisfying lighting demand constraints and user comfort vector thresholds. Furthermore, this operation can handle high-dimensional nonlinear optimization problems using multi-objective evolutionary algorithms (such as NSGA-III) or by introducing a hierarchical optimization strategy: first determining the energy-saving mode migration factor, then optimizing the remaining continuous variables, thereby achieving the synergistic optimal configuration of multi-dimensional adjustment parameters among the three objectives, transcending the limitations of single brightness adjustment. Dynamic dimming control of the LED office workstation lighting system is then executed based on the optimized adjustment decision vector. This can involve analyzing each component of the five-dimensional vector and sending them to the corresponding luminaire's driver module to execute actions such as brightness, color temperature, and switching timing. Furthermore, this operation can achieve comprehensive dynamic lighting response, including predictive start / stop, smooth transition, and mode adaptation.

[0095] For example, in a scenario of transitioning office space after an evening meeting, the intelligent-driven dynamic energy-saving lighting adjustment method of this embodiment can be as follows: The system detects that the user is about to end the video conference (activity type recognition), and at the same time, outdoor natural light rapidly decays; the multi-objective collaborative adjustment controller integrates factors such as the decreasing trend of perceived light comfort, the increase of fatigue index in the user comfort vector, and the approaching energy-saving period, and generates an adjustment decision vector containing the following components: the brightness adjustment is set to maintain 400 lux to support subsequent document processing; the color temperature adjustment factor gradually reduces the color temperature from 5000K to 3500K to adapt to the twilight environment; the switching timing offset is set to delay shutdown for 8 minutes after the meeting ends to cover possible short-term lingering; the dimming smoothness factor enables high-damping mode, so that the color temperature and brightness transition linearly within 5 minutes; the energy-saving mode migration factor switches to "evening energy-saving mode" to reduce the upper limit of illuminance in non-core areas. Based on this, the system performs multi-dimensional collaborative dimming, ensuring visual comfort for the user before leaving while avoiding continuous high brightness in vacant areas.

[0096] In one embodiment, sending the control decision vector to a multi-objective coordinated control controller for optimization includes:

[0097] Based on occupancy status, user activity type, and spatiotemporal distribution of outdoor natural light, workstation scene identification is performed, and scene identification results are established.

[0098] The search space is trimmed using scene recognition results, and the decision vector is adjusted and optimized based on the trimmed search space.

[0099] The workstation scene recognition process involves semantically classifying the current workstation usage context based on three inputs: occupancy status, user activity type, and the spatiotemporal distribution of outdoor natural light. This classification provides contextual semantic basis for search space pruning, improving optimization efficiency and result relevance. In this embodiment, workstation scene recognition can map multi-source sensory data into predefined scene labels using a rule engine or lightweight classification model. For example, workstation scene recognition may include, but is not limited to, one or more of the following: focused work scenarios, video conferencing scenarios, and temporary absence scenarios. The scene recognition result can be a structured semantic label output by the workstation scene recognition, representing the specific usage context of the current lighting control. This label can guide the optimization process of the adjustment decision vector to focus on a subset of feasible solutions matching the current scene. Furthermore, the scene recognition result is generated by the workstation scene recognition module and serves as a trigger condition for subsequent pruning logic. In an exemplary embodiment, the scene recognition result may include, but is not limited to, one or more of the following: high visual accuracy task labels, low-illuminance social labels, and unattended energy-saving labels.

[0100] The search space can be the original parameter space consisting of all possible values ​​of the decision vector, which can be used as the initial feasible region for a multi-objective optimization problem. In a specific embodiment, the search space can include, but is not limited to, one or more of the following: a full brightness-color temperature combination space, a region-independent control space, and a time-continuous dimming space. The pruned search space can be a reduced parameter space formed by excluding invalid or inefficient decision combinations that do not conform to the current usage context based on scene recognition results. This can be used to reduce the computational complexity of optimization and improve the quality and physical rationality of the solution. In this embodiment, the pruned search space achieves dimensionality reduction through hard or soft constraints. For example, the pruned search space can include, but is not limited to, one or more of the following: a meeting mode restricted space, an unattended minimalist space, and a natural light-dominated compensation space.

[0101] Adjustment decision vector optimization management can be a control process that performs multi-objective optimization within a specific search space to generate the optimal adjustment decision vector. This ensures the optimization process is efficient and the results meet the requirements of the current scenario. Furthermore, adjustment decision vector optimization management runs optimization algorithms in the pruned search space by invoking a multi-objective collaborative adjustment controller. In a specific embodiment, adjustment decision vector optimization management can include, but is not limited to, one or more of Pareto front search management, constraint satisfaction optimization scheduling, and dynamic weight adjustment mechanisms. Workstation scene identification based on occupancy status, user activity type, and outdoor natural light spatiotemporal distribution can be achieved by fusing three types of high-dimensional perception information and using preset logic or machine learning models to determine the typical usage scenario of the current workstation. Further, this operation can be implemented using decision tree rules (if occupancy = yes, activity = video conferencing, and natural light < 200 lux, then it is determined to be an "indoor video conferencing" scenario) or by training a lightweight neural network (inputting three types of feature vectors and outputting a scenario category probability distribution), thereby achieving abstraction from raw perception data to a high-level semantic scene, providing context for subsequent optimization.

[0102] Establishing scene recognition results can be achieved by solidifying the output of workstation scene recognition into structured labels and transmitting them to the search space trimming module, thereby forming scene context signals that can be understood by the optimization controller. Trimming the search space using scene recognition results can be done by dynamically removing or restricting inapplicable parameter combinations in the adjustment decision vector based on scene labels. Furthermore, this operation can be achieved by setting the upper limit of all lamp brightness to 0% in an "unmanned" scene to directly lock the energy-saving state, or by fixing the color temperature at 4000K±200K and allowing brightness adjustment only in the 300–500 lux range in a "video conferencing" scene. This can significantly reduce the scale of the optimization problem and avoid generating dimming strategies that are irrelevant or even conflicting with the current use. Optimizing the adjustment decision vector based on the trimmed search space can be achieved by initiating a multi-objective collaborative adjustment controller to perform optimization calculations within the reduced parameter space. Furthermore, this operation can be achieved by running the NSGA-II algorithm in the clipping space to evaluate only valid solutions, or by using a heuristic strategy to directly retrieve near-optimal solutions from the pre-stored scene-policy library and fine-tune them. This can improve the optimization speed and the practicality of the solutions, and enhance the system's response smoothness to complex lighting changes.

[0103] For example, in a video conference scenario on a cloudy afternoon, the intelligent-driven dynamic energy-saving adjustment method for sensing lighting in this embodiment can be as follows: The system detects that the workstation occupancy status is "occupied," the user activity type is "video conference," and the outdoor natural light spatiotemporal distribution shows that the cloud cover is thickening, causing the window illuminance to drop to 150 lux; the workstation scene recognition module integrates the three factors and outputs an "indoor video conference" scene label; this label triggers a search space clipping mechanism: prohibiting color temperatures below 3500K or above 4500K (to avoid color cast), limiting the brightness adjustment range to 400–600 lux (to meet the requirements for clear facial imaging), and turning off unnecessary decorative lighting; the multi-objective collaborative adjustment controller performs optimization within this clipped space, generating an adjustment decision vector under the premise of maintaining the stability of the user comfort vector, minimizing energy consumption, and matching the natural light color temperature, adjusting the main lighting panel to 500 lux / 4000K, and turning off the auxiliary lights; the entire process avoids trying unsuitable strategies such as high color temperature or low illuminance in the full parameter space, improving response speed and user experience.

[0104] In one embodiment, adjusting the decision vector optimization management based on the pruned search space includes:

[0105] In each iteration, after initializing the current adjustment decision vector, iterative optimization is performed using a lightweight gradient descent device.

[0106] Perform multiple rounds of iterative evaluation to generate continuous iterative evaluation results;

[0107] If the evaluation results of continuous iterations cannot meet the preset convergence threshold, then an auxiliary optimization instruction is generated.

[0108] The auxiliary optimization command calls the historical scenario energy-saving template for enhanced optimization.

[0109] The current adjustment decision vector can be an instance of an adjustment decision vector that serves as the starting point or intermediate state in a certain iteration. It can be used as an input variable of a lightweight gradient descent unit to calculate the gradient direction and update step size. In this embodiment, the current adjustment decision vector can be the optimization result of the previous round or a high-quality solution recommended by the historical scenario energy-saving template. For example, the current adjustment decision vector can be one or more of the following, including but not limited to the initial guess vector, intermediate iteration solutions, and local optimum candidates. The lightweight gradient descent unit can be a gradient optimization module with low computational overhead and fast convergence speed, specifically designed for the pruned low-dimensional search space. It can be used to improve the efficiency of a single iteration and adapt to the real-time dimming response requirements of office scenarios. Furthermore, the lightweight gradient descent unit can employ a simplified objective function gradient approximation or a first-order momentum mechanism to perform a local search in the adjustment decision vector parameter space. In an exemplary embodiment, the lightweight gradient descent unit can be one or more of the following, including but not limited to a stochastic coordinate descent unit, an adaptive step-size gradient unit, and a projected gradient optimizer.

[0110] In each iteration, the current adjustment decision vector is initialized. This can be done by setting the initial adjustment decision vector for this round of optimization, or by using the result from the previous round or the template recommendation value. Furthermore, this operation can be implemented by directly inheriting the output of the previous iteration or selecting a high-quality solution from a matched historical context energy-saving template, thus providing the starting point for the optimization algorithm and influencing the convergence speed and path. Iterative optimization using a lightweight gradient descent device can be achieved by updating the current adjustment decision vector along the negative gradient direction of the objective function within the pruned search space. Further, this operation can be implemented by using a fixed small step size for parameter fine-tuning to suit smooth target surfaces, or by using an adaptive learning rate to adjust the step size based on historical gradients to handle non-stationary target changes, thereby achieving efficient local search and reducing the computational load per iteration.

[0111] Multi-round iteration can be a process in the optimization management of adjustment decision vectors that repeatedly executes optimization steps until the termination condition is met. It can be used to gradually approach the Pareto optimal solution set and improve decision quality. In a specific embodiment, multi-round iteration can include, but is not limited to, one or more of coarse-grained initial exploration iteration, fine-tuning iteration, and convergence verification iteration. Executing multi-round iterations, recording evaluations, and generating continuous iterative evaluation results can involve multi-objective scoring of the adjustment decision vector output in each round and storing it in chronological order to form a sequence. Furthermore, this operation can be implemented by scoring and recording the solution in each round using the built-in evaluation function of the multi-objective collaborative adjustment controller, thereby providing a data basis for convergence judgment and supporting dynamic optimization strategy switching.

[0112] The continuous iterative evaluation results can be a time-series sequence of results obtained by quantitatively evaluating the adjustment decision vectors generated in each round of multiple iterations according to multi-objective collaborative indicators (comfort, energy consumption, natural light matching). This sequence can be used to determine whether the optimization process is stagnating or has stalled. In this embodiment, the continuous iterative evaluation results are scored and recorded by the built-in evaluation function of the multi-objective collaborative adjustment controller for each round. For example, the continuous iterative evaluation results can include, but are not limited to, one or more of the following: objective weighted scoring sequence, Pareto dominance relationship change sequence, constraint violation trend, etc. The preset convergence threshold can be a numerical or logical boundary for determining whether the continuous iterative evaluation results have reached an acceptable convergence state, and can be used as a basis for triggering the auxiliary optimization mechanism. In an exemplary embodiment, the preset convergence threshold can include, but is not limited to, one or more of the following: objective function change rate threshold, solution vector Euclidean distance threshold, evaluation indicator variance upper limit, etc.

[0113] Determining whether the continuous iterative evaluation results meet the preset convergence threshold can be achieved by calculating the statistical characteristics (such as variance and slope) of the evaluation sequence and comparing them with the threshold. Furthermore, this operation can be implemented through sliding window analysis or trend fitting, thereby identifying optimization stagnation states and deciding whether to activate auxiliary mechanisms. If the continuous iterative evaluation results fail to meet the preset convergence threshold, an auxiliary optimization instruction is generated. This can be triggered by generating a control signal when excessive fluctuations in the evaluation index or insufficient improvement are detected. Further, this operation can be dynamically generated by the optimization monitoring module based on the evaluation results, thereby enabling adaptive switching of optimization strategies and avoiding invalid loops. The auxiliary optimization instruction can be a control signal generated when the continuous iterative evaluation results fail to meet the preset convergence threshold, and can be used to activate an experience-based knowledge guidance mechanism. In this embodiment, the auxiliary optimization instruction is dynamically generated by the optimization monitoring module based on the evaluation results. For example, the auxiliary optimization instruction may include, but is not limited to, one or more of the following: template retrieval trigger signal, prior injection instruction, and optimization restart command.

[0114] Historical scenario energy-saving templates can be experiential knowledge base entries built upon verified Pareto optimal solution sets from similar past workstation scenarios (including occupancy status, activity type, and natural light distribution). These templates can provide high-quality initial solutions or guide search directions for the current optimization, accelerating convergence. Furthermore, historical scenario energy-saving templates can store a set of high-quality adjustment decision vectors corresponding to typical scenarios through long-term data clustering and scenario labeling. In a specific embodiment, historical scenario energy-saving templates may include, but are not limited to, one or more of the following: video conferencing energy-saving templates, cloudy day focused work templates, and window-side high natural light compensation templates. Using auxiliary optimization instructions to invoke historical scenario energy-saving templates for enhanced optimization can involve retrieving matching historical templates based on the characteristics of the current workstation scenario and integrating their high-quality solutions into the current optimization process. Furthermore, this operation can be achieved by directly setting the optimal solution in the template as the initial point for a new round of optimization, or by using the template solution set as an additional constraint or regularization term to guide the gradient direction, thereby accelerating convergence with the help of experiential knowledge and improving the practicality and robustness of the solution.

[0115] Reinforcement optimization can be an enhanced optimization process that integrates prior knowledge from historical context energy-saving templates on top of basic gradient optimization. It can be used to overcome local minima traps and improve the quality and robustness of solutions. In this embodiment, reinforcement optimization guides the search direction by injecting high-quality solutions from the template as seeds or constraints into the current optimization process. For example, reinforcement optimization may include, but is not limited to, one or more of the following: prior-guided optimization, template initialization optimization, and hybrid empirical-gradient optimization.

[0116] Taking the optimization oscillation caused by continuous alternating sunshine and cloudy weather as an example, the intelligent-driven sensing lighting dynamic energy-saving adjustment method in this embodiment can be as follows: The system encounters frequent cloud changes in the afternoon, and the spatiotemporal distribution of outdoor natural light fluctuates drastically; the lightweight gradient descent iterates for 5 consecutive rounds in the clipped search space, but the continuous iteration evaluation results show that the comfort and energy consumption indicators oscillate repeatedly and do not meet the preset convergence threshold; the system generates auxiliary optimization instructions, searches the historical scenario energy-saving template library, and matches the template "windowside workstation + document editing + sudden cloud change"; this template contains 3 sets of Pareto optimal adjustment decision vectors that have been verified to be effective under similar conditions in the past; the enhanced optimization module injects one of them as a new initial point into the optimization process, and the lightweight gradient descent converges quickly on this basis to generate a stable dimming strategy: the brightness is maintained at 450 lux ± 20 lux, the color temperature is locked at 4500K, and flickering adjustment caused by sudden changes in light is avoided; finally, a smooth transition and high-efficiency energy saving are achieved.

[0117] In one embodiment, the perception of light comfort is established using spatiotemporal distribution, including:

[0118] Feature extraction is performed using spatiotemporal distribution to establish a feature set. The extracted features include light intensity distribution, color temperature gradient, brightness fluctuation rate, incident azimuth and elevation angles, light spot uniformity, and natural light ratio.

[0119] Establish a comfort label mapping based on the existing office environment comfort dataset;

[0120] Based on the comfort label mapping, a sliding time window aggregation and matching of feature sets is performed to establish light comfort perception.

[0121] Spatiotemporal distribution is a joint dynamic distribution model of outdoor natural light in both temporal (e.g., diurnal variation, sudden weather changes) and spatial (e.g., building orientation, shading structure) dimensions. It can be used as the raw input for feature extraction, supporting the structured expression of multidimensional optical features. In this embodiment, spatiotemporal distribution can be obtained by reconstructing the outdoor natural light field, and its modeling process integrates diurnal trajectory illumination models, weather change response models, and building shading attenuation models. Illumination intensity distribution can be the numerical distribution of illuminance at different spatial locations, reflecting the uniformity of illumination on the work surface and the risk of local overexposure / darkness. For example, illumination intensity distribution may include one or more of the following: window edge illuminance gradient, desktop center illuminance, corner illuminance attenuation, etc. Color temperature gradient can be the rate and direction of color temperature change of natural light in spatial or temporal dimensions, influencing visual color consistency and physiological rhythm synchronization. In an exemplary embodiment, color temperature gradient may include vertical color temperature gradient, horizontal color temperature gradient, temporal color temperature drift, etc.

[0122] Brightness fluctuation rate can be the relative change in natural illuminance or brightness per unit time, and can be used to characterize illumination stability. High fluctuation rate can easily cause visual discomfort. Further, brightness fluctuation rate can include one or more of the following: cloud flickering fluctuations, leaf shading and shaking, and transient changes in building reflections. The incident azimuth angle can be the horizontal projection angle of sunlight relative to the building's due north direction, and can be used to determine the main path of natural light entering the room and glare risk areas. In a specific embodiment, the incident azimuth angle can include east-facing, south-facing, and west-facing incident angles. The elevation angle can be the angle between sunlight and the ground plane, and can be used to influence the penetration depth of natural light and the distribution pattern of indoor illuminance. For example, the elevation angle can include low elevation angles (sunrise and sunset), medium elevation angles (morning / afternoon), and high elevation angles (noon). Light spot uniformity can be the degree of consistency of illuminance within the light spot area formed by natural light on the work surface, and can be used to measure the presence of localized strong light or shadow interference. Further, light spot uniformity can include one or more of the following: edge attenuation uniformity, center-periphery contrast, and multi-source superposition uniformity.

[0123] The natural light percentage can be the proportion of total illuminance contributed by natural light, and can be used to assess the necessity and energy-saving potential of artificial lighting intervention. In one embodiment, the natural light percentage may include the percentage of full-window illumination, the percentage of partial shading, and the percentage of indirect diffuse illumination. The feature set can be a set of multidimensional structured parameters extracted from spatiotemporal distribution to characterize the light environment characteristics, and can be used to provide calculable input variables for comfort label mapping. In this embodiment, the feature set can be obtained by analyzing and quantizing spatiotemporal distribution data through signal processing or computer vision methods, and can be divided into static optical feature subsets, dynamic fluctuation feature subsets, spatial distribution feature subsets, etc.

[0124] Feature extraction using spatiotemporal distribution can involve applying signal analysis or image processing algorithms to the reconstructed spatiotemporal distribution data of outdoor natural light to extract predefined optical features. Furthermore, feature extraction using spatiotemporal distribution can be achieved by automatically extracting spatial features from light field images using convolutional neural networks, or by using Fourier transform to analyze the frequency domain characteristics of brightness fluctuations. This transforms the raw light field data into structured, interpretable multidimensional features, supporting subsequent comfort modeling. Establishing a feature set can involve organizing extracted features such as illuminance distribution and color temperature gradient into vectors or tensors of a unified format, thus forming standardized inputs that are easy to map to comfort labels. The office environment comfort dataset can be a labeled data set containing historical light environment parameters and corresponding user subjective comfort scores, which can be used to support the establishment of comfort label mappings, realizing the correlation between objective parameters and subjective feelings. In a specific embodiment, the office environment comfort dataset can be constructed through long-term field surveys, questionnaires, and synchronous sensor recordings, and may include laboratory control datasets, real office scenario datasets, and cross-seasonal mixed datasets.

[0125] Comfort label mapping can be a function or model that maps a feature set to discrete or continuous comfort levels, and can be used to transform multidimensional optical features into quantifiable comfort metrics that can be optimized. In this embodiment, comfort label mapping can be obtained by training a regression or classification model based on an office environment comfort dataset, which may include linear weighted mapping, neural network mapping, fuzzy rule mapping, etc. Establishing comfort label mapping based on an existing office environment comfort dataset can be achieved by using supervised learning methods to train a mapping model from feature set to comfort labels. Furthermore, this operation can be achieved by using a random forest regression model to fit continuous comfort scores, or by using a support vector machine to classify and model high / medium / low comfort levels, thereby realizing a quantitative correlation between objective lighting environment parameters and subjective comfort feelings.

[0126] A sliding time window is a mechanism for selecting data segments that move forward at a fixed length over a time series. It can be used to capture the temporal evolution of comfort perception and improve adaptability to dynamic lighting changes. For example, a sliding time window can include one or more of short-time response windows, medium-time smoothing windows, and long-time trend windows. Sliding time window clustering and matching of feature sets based on comfort label mapping can involve clustering or similarity matching of the feature set within the sliding time window, selecting the label closest to historical comfort samples. Furthermore, this operation can be achieved by calculating the feature mean within the window and then querying the nearest neighbor comfort sample, or by performing dynamic time warping (DTW) matching on the feature sequence within the window. This enhances the temporal continuity and dynamic adaptability of comfort perception and avoids misjudgments caused by transient noise.

[0127] Optical comfort perception can be a temporally continuous quantifiable representation of comfort based on feature sets, comfort label mappings, and the aggregation results of sliding time windows. It can be used to provide comfort inputs with temporal consistency and scene adaptability for multi-objective collaborative regulation. In an exemplary embodiment, optical comfort perception can match the optimal comfort label through feature clustering within a sliding window, outputting a comfort estimate for the current moment, which may include an instantaneous comfort snapshot, short-term comfort trends, and long-term comfort baselines. Establishing optical comfort perception can involve integrating the sliding window matching results to output the current comfort quantifiable value and its confidence level, thereby generating high-fidelity, smooth, and scene-adaptive comfort inputs to support subsequent multi-objective optimization.

[0128] For example, in an open-plan office environment under cloudy weather conditions, the intelligent-driven dynamic energy-saving adjustment method for sensing lighting in this embodiment can be as follows: The system reconstructs the spatiotemporal distribution of outdoor natural light, showing that it will experience three rapid alternations of cloud cover and light transmission within the next 10 minutes; the feature extraction module outputs high brightness fluctuation rate (>30% / min), low natural light ratio (<40%), and significant color temperature gradient (from 6500K to 4500K); the feature set is fed into a comfort label mapping model trained based on a historical office environment comfort dataset; a sliding time window (8 minutes in length) is used to cluster past feature sequences and match them with the "high risk of visual fatigue" label from the past; the light comfort sensing module outputs a current comfort score of 0.42 (out of 1.0), which is below the threshold of 0.6; this result is passed as input to a multi-objective collaborative adjustment controller, which, while ensuring that user comfort is maintained, pre-adjusts the brightness of LED lights and stabilizes the color temperature to 5000K to avoid insufficient illuminance and color temperature jumps caused by sudden changes in natural light, thus achieving a smooth transition.

[0129] In one embodiment, the energy consumption of the LED office workstation lighting system is dynamically adjusted based on the optimization result of the adjustment decision vector, including:

[0130] Establish a dataset of users' physiological states and configure the energy tolerance factor for each physiological state.

[0131] The user's physiological state dataset can be a collection of data composed of multidimensional physiological signals, used to characterize the user's current physical and mental state. It can provide the initial basis for identifying the user's physiological state type and support the dynamic configuration of the energy tolerance factor. In one specific embodiment, the user's physiological state dataset can continuously collect signals such as heart rate variability, skin conductance response, and eye movement frequency through a linked acquisition device, and perform time alignment and structured storage. For example, the user's physiological state dataset may include, but is not limited to, one or more subsets of data on focused states, visual fatigue states, and relaxation / recovery states. Physiological states can be discrete category labels of the user's current physical and mental operating mode derived from physiological signal clustering or classification, which can be used as an index for the energy tolerance factor, reflecting the user's sensitivity level to changes in lighting. Furthermore, the physiological states can be inferred in real time through a machine learning model, outputting state categories. In an exemplary embodiment, physiological states may include, but are not limited to, one or more subsets of high focus, visual fatigue, and low alertness.

[0132] The energy consumption tolerance factor can be a dimensionless weighted parameter associated with a specific physiological state, characterizing the acceptable range of lighting adjustment for the user. It can be used to dynamically adjust energy-saving intensity during energy consumption optimization, achieving a flexible balance between comfort and energy efficiency. In this embodiment, the energy consumption tolerance factor can be assigned a corresponding tolerance value to each type of physiological state based on historical comfort feedback or preset rules. For example, the energy consumption tolerance factor can employ a high tolerance factor (>1), a baseline tolerance factor (=1), a low tolerance factor (<1), etc. Establishing a user's physiological state dataset can involve continuously collecting and structurally storing multi-channel physiological signals from linked acquisition devices. Furthermore, establishing a user's physiological state dataset can be achieved through the above methods, thereby providing a data foundation for physiological state identification and energy consumption tolerance configuration.

[0133] Configuring energy consumption tolerance factors for each physiological state can be achieved by establishing a mapping relationship between predefined or learned tolerance values ​​and physiological state categories. Furthermore, configuring energy consumption tolerance factors for each physiological state can be achieved by training a tolerance regression model based on historical user comfort feedback data, or by using expert rules to set fixed tolerance values ​​for different states, thereby enabling energy-saving strategies to adaptively respond to user states.

[0134] An energy consumption optimization function is constructed using an energy consumption tolerance factor, and energy consumption optimization is performed based on the result of adjusting the decision vector optimization based on the energy consumption optimization function.

[0135] The energy consumption optimization function can be a reconstructed objective function for secondary optimization of the adjustment decision vector, derived by introducing an energy consumption tolerance factor. This function enables the energy-saving strategy to adapt to the user's current physiological state, increasing dimming intensity in a high-tolerance state and suppressing dimming intensity in a low-tolerance state. In a specific embodiment, the energy consumption optimization function can be multiplied or weighted by an energy consumption tolerance factor based on the original energy consumption minimization objective, forming a context-aware optimization objective. For example, the energy consumption optimization function can employ a multiplicative weighted energy consumption function, a piecewise linear tolerance function, or a state switching threshold function. The optimization result of the adjustment decision vector can be a set of lighting control parameters generated after the initial optimization by the multi-objective collaborative adjustment controller. This set can be used as the input basis for the energy consumption optimization function for secondary fine-tuning. In this embodiment, the optimization result of the adjustment decision vector is received and corrected by the energy consumption optimization function, generating an energy consumption optimization result that better suits the user's current physiological state.

[0136] The energy consumption optimization result can be the final control command obtained by combining the optimization result of the adjustment decision vector with the energy consumption tolerance factor for secondary optimization. This command can be used to drive the LED office workstation lighting system to perform context-aware dynamic dimming. Furthermore, the energy consumption optimization result can be used to obtain an updated adjustment decision vector by solving the energy consumption optimization function. In this embodiment, the energy consumption optimization result directly applies to the LED office workstation lighting system to achieve differentiated energy saving based on physiological states. Constructing the energy consumption optimization function using the energy consumption tolerance factor can be achieved by embedding the tolerance factor into the original energy consumption objective function to form a state-aware optimization objective. Furthermore, constructing the energy consumption optimization function using the energy consumption tolerance factor can be achieved by multiplying the energy consumption objective by the tolerance factor as a new objective, or by introducing a tolerance-related dimming amplitude upper limit into the constraints, thereby enabling the optimization process to have physiological state context awareness. Energy consumption optimization based on the adjustment decision vector optimization result can be achieved by using the adjustment decision vector optimization result as the initial value and solving for the optimal solution of the energy consumption optimization function. Furthermore, energy consumption optimization based on the adjustment decision vector optimization result of the energy consumption optimization function can be achieved in the above way, thereby further improving the energy-saving accuracy without destroying the original constraints of comfort and natural light integration.

[0137] Dynamic adjustment of LED office workstation lighting system based on energy consumption optimization results.

[0138] Dynamic adjustment of LED office workstation lighting systems based on energy consumption optimization results can be achieved by sending the final optimized control parameters to the LED luminaire execution unit. Furthermore, dynamic adjustment of LED office workstation lighting systems based on energy consumption optimization results can be realized through the above method, thereby achieving context-aware lighting control that provides energy on demand and adjusts lighting according to individual needs.

[0139] Taking a period of high-intensity focused work as an example, the intelligent-driven sensing lighting dynamic energy-saving adjustment method in this embodiment can be as follows: The system detects that the user is in a high-focus state through the linkage acquisition device (manifested as stable heart rate, low eye movement frequency, and weak skin conductance response), and the physiological state recognition module classifies it as a "high-focus state"; the energy consumption tolerance factor corresponding to this state is 1.3, indicating that the user is not sensitive to light fine-tuning; the energy consumption optimization function amplifies the energy-saving weight accordingly, reducing the brightness in the original adjustment decision vector from 60% to 45%, while maintaining the color temperature unchanged; the LED office workstation lighting system executes this energy consumption optimization result, reducing energy consumption without the user's notice; and when the system subsequently detects that the user has entered a visual fatigue state (increased blinking frequency and decreased gaze stability), the system automatically switches to a low tolerance factor of 0.7, suppressing further dimming and prioritizing the stability of illuminance, thereby dynamically balancing energy saving and comfort at different physiological stages.

[0140] In one embodiment, the linkage acquisition device includes a workstation camera, a sound acquisition device, and a wearable physiological data acquisition device.

[0141] The workstation camera can be an image sensing device deployed near the workstation to collect user visual behavior data. It can be used to support user activity type recognition and attention state judgment, providing visual behavior dimension data for constructing a user comfort vector. In this embodiment, the workstation camera can continuously capture user posture, head orientation, screen viewing time, and other behavioral characteristics through low resolution or privacy-preserving modes (such as skeleton extraction). Furthermore, the workstation camera can work in conjunction with a sound acquisition device and wearable physiological data acquisition devices to jointly constitute multimodal perception input. For example, the workstation camera can be one or more of the following, including but not limited to infrared depth cameras, wide-angle fisheye cameras, and embedded AI vision modules. The sound acquisition device can be an audio sensing device used to pick up acoustic signals from the surrounding environment of the workstation. It can be used to assist in determining the user's social or focused mode and enhance semantic understanding of lighting scene requirements. In an exemplary embodiment, the sound acquisition device can collect speech energy, spectral characteristics, and sound source direction through a microphone array, and combine this with an acoustic event detection algorithm to identify conversation, silence, or meeting status. Furthermore, the sound acquisition device can form a three-dimensional sensing structure of sound, vision, and physiology with the workstation camera and wearable physiological data acquisition device. In one specific embodiment, the sound acquisition device can employ a directional microphone, an omnidirectional microphone array, a noise-suppressing audio front-end, etc.

[0142] Wearable physiological data acquisition devices can be portable sensing terminals worn on a user's body for continuous monitoring of physiological indicators. They can provide objective physiological evidence reflecting internal states such as emotional arousal and fatigue levels, improving the accuracy of comfort modeling. In this embodiment, the wearable physiological data acquisition device can acquire signals related to the autonomic nervous system, such as heart rate, skin conductance, and body temperature, in real time through sensors such as photoplethysmography (PPG) and skin conductance. Furthermore, the wearable physiological data acquisition device, along with a workstation camera and a sound acquisition device, jointly outputs multi-dimensional user status data. For example, the wearable physiological data acquisition device can be one or more of the following, including but not limited to smart bracelets, ring-type physiological monitors, and patch-type bioelectric sensors. The linked acquisition device includes a workstation camera, a sound acquisition device, and the wearable physiological data acquisition device. These three types of devices can be used as specific implementations of the linked acquisition device, working collaboratively when the system is activated to acquire multimodal user status data. Furthermore, the linked data acquisition devices, including workstation cameras, sound acquisition devices, and wearable physiological data acquisition devices, can be realized by constructing a multi-dimensional perception system covering visual behavior, acoustic environment, and physiological response, thereby supporting the generation of high-fidelity user comfort vectors.

[0143] For example, in scenarios involving high-intensity focused work, the intelligent-driven dynamic energy-saving adjustment method for sensing lighting in this embodiment can be as follows: When a user wears a smart bracelet to write a document, the workstation camera detects that the user is staring at the screen for a long time and has a stable sitting posture, and the sound collector identifies that the environment is in a continuous silent state; the wearable device simultaneously displays a stable heart rate but a slight increase in skin conductance, indicating mild cognitive load; based on this, the system determines that the user is in a state of high focus, and sets the lighting requirement constraints to medium-high illuminance (450–500 lux) and cool color temperature (5000K), while increasing the weight of the "focus adaptation component" in the user comfort vector; when the cloud cover outside the window suddenly increases, causing the natural light to decay, the multi-objective collaborative adjustment controller prioritizes ensuring illuminance stability, only fine-tuning the artificial light source to maintain visual continuity, avoiding interruption of the focused state due to illuminance fluctuations, while controlling the total energy consumption within a reasonable range.

[0144] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing an intelligent-driven sensing lighting dynamic energy-saving adjustment program, wherein when the intelligent-driven sensing lighting dynamic energy-saving adjustment program is executed by a processor, it implements the steps of the intelligent-driven sensing lighting dynamic energy-saving adjustment method as described above.

[0145] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A smart-driven, sensing-based dynamic energy-saving adjustment method for lighting, characterized in that, The method includes: The system uses outdoor light sensors and indoor environmental sensors to perceive the ambient light field, reconstructs the spatiotemporal distribution of outdoor natural light, and uses the spatiotemporal distribution to establish a perception of light comfort. Extract the occupancy status and user activity type of the office workstations, and construct lighting requirement constraints based on the occupancy status and user activity type; Activate the linkage acquisition device, perform multi-dimensional data perception of the user based on the linkage acquisition device, and establish a user comfort vector using the multi-dimensional data perception results; After establishing the adjustment decision vector, the light comfort perception, the lighting demand constraint, and the user comfort vector are used as input data and sent to the multi-objective collaborative adjustment controller to perform adjustment decision vector optimization. The collaborative objectives of the multi-objective collaborative adjustment controller include user comfort maintenance objective, energy consumption minimization objective, and natural light fusion matching objective. The energy consumption of the LED office workstation lighting system is dynamically adjusted based on the optimization results of the adjustment decision vector.

2. The intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting as described in claim 1, characterized in that, The method of establishing light comfort perception using the spatiotemporal distribution includes: Perform a historical environment data call to obtain the historical environment dataset; After performing weather similarity clustering on the historical environmental dataset, abrupt change time points are predicted based on the time identifiers of the similarity clusters, and abrupt change node prediction result is established. Obtain the location data of the office workstation, perform online weather reading based on the location data, and establish online weather reading results; The light comfort perception is compensated by using the predicted results of the mutation nodes and the network weather reading results.

3. The intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting as described in claim 2, characterized in that, The method of compensating for light comfort perception using the mutation node prediction results and the network weather reading results includes: Temporal extraction of ambient light field perception is performed to establish a temporal distribution dataset of outdoor natural light spatiotemporal distribution; Using the aforementioned temporal distribution dataset, temporal prediction is performed to establish a perception of light comfort. Based on the predicted mutation node, conflict identification is performed for time-series prediction, and the conflict identification results are used to establish a first compensation feedback. The impact analysis of time-series forecasting is performed using the networked weather reading results, and a second compensation feedback is established. The first compensation feedback and the second compensation feedback are used to compensate for the perceived light comfort.

4. The intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting as described in claim 1, characterized in that, The adjustment decision vector includes brightness adjustment amount, color temperature adjustment factor, on / off timing offset, dimming smoothness factor, and energy-saving mode migration factor.

5. The intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting as described in claim 1, characterized in that, The step of sending the control decision vector to the multi-objective coordinated control controller for optimization includes: Based on the occupancy status, user activity type, and spatiotemporal distribution of outdoor natural light, workstation scene recognition is performed, and scene recognition results are established. The search space is trimmed using the scene recognition results, and the decision vector optimization management is performed based on the trimmed search space.

6. The intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting as described in claim 5, characterized in that, The optimization management of the decision vector based on the pruned search space includes: In each iteration, after initializing the current adjustment decision vector, iterative optimization is performed using a lightweight gradient descent device. Perform multiple rounds of iterative evaluation to generate continuous iterative evaluation results; If the continuous iterative evaluation results cannot meet the preset convergence threshold, then an auxiliary optimization instruction is generated; The auxiliary optimization command is used to invoke the historical scenario energy-saving template for enhanced optimization.

7. The intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting as described in claim 1, characterized in that, The method of establishing light comfort perception using the spatiotemporal distribution includes: Feature extraction is performed using the spatiotemporal distribution to establish a feature set. The extracted features include light intensity distribution, color temperature gradient, brightness fluctuation rate, incident azimuth and elevation angles, light spot uniformity, and natural light ratio. Establish a comfort label mapping based on the existing office environment comfort dataset; Based on the comfort label mapping, a sliding time window aggregation and matching of feature sets is performed to establish light comfort perception.

8. The intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting as described in claim 1, characterized in that, The dynamic energy consumption adjustment of the LED office workstation lighting system based on the optimization results of the adjustment decision vector includes: Establish a dataset of users' physiological states and configure the energy tolerance factor for each physiological state; An energy consumption optimization function is constructed using the energy consumption tolerance factor, and energy consumption optimization is performed based on the energy consumption optimization function to adjust the optimization results of the decision vector. Dynamic adjustment of LED office workstation lighting system based on energy consumption optimization results.

9. The intelligent-driven sensing-based dynamic energy-saving adjustment method for lighting as described in claim 1, characterized in that, The linked data acquisition equipment includes a workstation camera, a sound acquisition device, and a wearable physiological data acquisition device.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an intelligent-driven sensing lighting dynamic energy-saving adjustment program, which, when executed by a processor, implements the steps of the intelligent-driven sensing lighting dynamic energy-saving adjustment method as described in any one of claims 1 to 9.

Images