Adaptive dimming and energy consumption optimization method, device and equipment for building lighting system based on digital twinning and medium

Abstract

The invention discloses an adaptive dimming and energy consumption optimization method and device for an architectural lighting system based on digital twinning, electronic equipment and a storage medium, and relates to the technical field of intelligent optimization, and the method comprises the steps: obtaining physical space data through a photon event sensor and an event camera; updating the digital twin dynamic photon flow field model based on compressed sensing, a radiation transfer equation and a bidirectional reflection distribution function; generating and screening an optimal dimming strategy by using a confrontation generation optimization architecture; personalized adaptation is realized through light environment fingerprint matching and transfer learning; based on a ternary closed-loop self-evolution mechanism, the model and architecture parameters are corrected, and the energy consumption and the lighting effect are dynamically optimized. The light environment modeling precision and the dimming adaptability are improved, the building lighting energy consumption is reduced, intelligent and energy-saving upgrading of the lighting system is achieved, and different user preferences and scene requirements are met.

Description

Technical Field

[0001] This invention relates to the field of intelligent optimization technology, and more specifically, to a method, apparatus, equipment, and medium for adaptive dimming and energy consumption optimization of a building lighting system based on digital twins. Background Technology

[0002] Currently, green and low-carbon development has become the core orientation of the construction industry. As an important component of building energy consumption, the energy-saving and intelligent upgrading of building lighting has become an industry consensus. At the same time, people's demands for comfort and personalization in building lighting are constantly increasing. Traditional lighting systems can no longer balance energy consumption control and lighting experience, and there is an urgent need for an efficient and precise adaptive dimming and energy consumption optimization solution.

[0003] Existing building lighting systems mostly use traditional illuminance sensors for point-based monitoring, which can only acquire local illumination data and cannot capture the spatiotemporal dynamic changes in the indoor light environment. Furthermore, they fail to effectively adapt to the real-time changes in the reflective properties of building interior surfaces, resulting in insufficient accuracy in light environment modeling. In terms of dimming strategy optimization, traditional algorithms lack physical constraints, are prone to getting trapped in local optima, and cannot comprehensively consider natural light prediction, population distribution, and user preferences to generate scientifically sound and reasonable dimming solutions.

[0004] Although digital twin and artificial intelligence technologies have been applied to the field of smart lighting, their integration still has shortcomings: a lack of dynamic modeling methods based on photon-level data makes precise synchronization between physical and digital spaces impossible; the optimization architecture lacks effective physical constraints and personalized adaptation mechanisms, resulting in poor strategy adaptability; and there is no closed-loop self-evolution capability, meaning system performance cannot continuously improve with the accumulation of operational data. Therefore, developing adaptive dimming and energy consumption optimization methods for building lighting based on digital twins to address these issues has become a key path to promoting the intelligent and energy-efficient upgrading of building lighting. Summary of the Invention

[0005] In order to overcome the problems of low modeling accuracy, lack of physical constraints in optimization, and lack of personalized adaptation and closed-loop self-evolution capabilities in existing technologies, this invention discloses a method, device, equipment and medium for adaptive dimming and energy consumption optimization of building lighting systems based on digital twins, which can effectively solve the above-mentioned technical problems.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: An adaptive dimming and energy consumption optimization method for building lighting systems based on digital twins, the method comprising: Photon event sensors and event cameras deployed within the building acquire photon data and surface reflection characteristics data of the physical space; Based on compressed sensing theory and radiative transfer equation, the photon data is sparsely sampled and photon trajectory reconstructed. Combined with the bidirectional reflection distribution function parameterization model and material reflection field database, the dynamic photon flow field model in the digital twin is updated. The generator network, based on an adversarial generative optimization architecture, takes the current photon flow field state, personnel distribution, natural light prediction, and user preferences as input and outputs candidate dimming strategies. By using a discriminator network with an adversarial generative optimization architecture and combining physical information neural network constraints, the authenticity and excellence of the candidate dimming strategies are scored, and the optimal dimming strategy is selected. Extract the optical environment fingerprint of the current photon flow field, perform similarity matching with the optical environment fingerprint database, and personalize the optimal dimming strategy through transfer learning; The optimal dimming strategy, after personalized adaptation, is sent to the building lighting actuator to control the operation of the lighting equipment; Based on actual operational data from physical space feedback, a ternary closed-loop self-evolution mechanism is used to perform parameter correction and self-evolution update on the dynamic photon flow field model and adversarial generative optimization architecture, thereby achieving dynamic optimization of energy consumption and lighting effects.

[0007] Preferably, the step of sparsely sampling and reconstructing photon trajectories from the photon data based on compressed sensing theory and radiative transfer equations, and updating the dynamic photon flow field model in the digital twin by combining a bidirectional reflectance distribution function parameterized model and a material reflectance field database, includes: The arrival time, incident angle, and wavelength distribution data of individual photons are collected by a photon event sensor, and sparse sampling is performed. Based on compressed sensing theory and combined with the radiative transfer equation as a physical constraint, the photon flow field distribution of the building space is reconstructed from sparsely sampled photon data. The changes in the reflectivity of the building's interior surfaces are captured by an event camera, and the optical properties of each surface in the digital twin are updated in real time using a bidirectional reflectance distribution function parameterized model. By calling the material reflection field database and using transfer learning to quickly match the reflection characteristics of newly emerging materials, the dynamic photon flow field model is updated, and the photon flux density, average photon free path, and anisotropy coefficient quantification index are calculated.

[0008] Preferably, the generator network with an adversarial generative optimization architecture takes the current photon flow field state, personnel distribution, natural light prediction, and user preferences as input, and outputs candidate dimming strategies, including: The generator network adopts a structure combining variational autoencoder and Transformer to capture the spatiotemporal correlation of photon flow field and related input parameters; The current photon flow field state, the distribution data of people in the building, the natural light prediction data for the future preset duration, and the user preference vector are obtained and input into the generator network. The generator network is trained based on the policy gradient algorithm and outputs candidate dimming strategies that include the brightness, color temperature, and beam angle parameters of each lighting device.

[0009] Preferably, the discriminator network using an adversarial generative optimization architecture, combined with physical information neural network constraints, scores the authenticity and superiority of the candidate dimming strategies to select the optimal dimming strategy, including: The discriminator network adopts a 3D convolutional neural network structure to extract the spatiotemporal features of the simulated photon flow field after the candidate dimming strategy is executed; The discriminator network introduces a physical information neural network as a constraint to ensure that the scoring process conforms to the radiative transfer equation and outputs the authenticity score and excellence score of the candidate dimming strategy. Based on the preset scoring threshold, candidate dimming strategies that meet both the authenticity and excellence scores are selected as the optimal dimming strategy; if there are multiple candidate strategies that meet the threshold, the strategy with the highest comprehensive score is selected as the optimal dimming strategy.

[0010] Preferably, the step of extracting the optical environment fingerprint of the current photon flow field, performing similarity matching with the optical environment fingerprint database, and personalizing the optimal dimming strategy through transfer learning includes: An autoencoder is used to reduce the dimensionality of the current photon flow field state and extract a light environment fingerprint that includes spatial illuminance distribution features, spectral distribution features, and dynamic change features. The similarity between the current ambient fingerprint and historical fingerprints in the ambient fingerprint database is calculated, and the similarity is measured using a dynamic time warping algorithm. The optimal strategy corresponding to the historical scene with the highest similarity is retrieved, and the historical strategy is transferred to the current scene through transfer learning. The optimal dimming strategy is then adjusted in a personalized way to suit the current user preferences.

[0011] Preferably, the actual operational data based on physical space feedback is used to perform parameter correction and self-evolutionary updates on the dynamic photon flow field model and the adversarial generative optimization architecture through a ternary closed-loop self-evolution mechanism, including: It receives actual lighting data, energy consumption data, and user feedback data from sensors in the physical space, compares them with simulated data in the digital twin, and uses the Kalman filter algorithm to correct the parameters of the dynamic photon flow field model. Based on actual operational data, the generator and discriminator networks of the adversarial generative optimization architecture are fine-tuned in the short term; the network is batch trained and the network weights are updated every morning using the accumulated operational data of the day. By using a federated learning framework, model parameters can be shared among multiple buildings, accelerating the system's self-evolution process and improving optimization results.

[0012] Preferably, the construction of the user preference vector includes: the digital twin control platform collecting the user's dimming operation history in different scenarios, extracting dimensional parameters such as preferred illuminance, preferred color temperature, glare sensitivity, and dynamic response speed, and using a latent semantic model to mine the implicit features of user preferences to construct the user preference vector.

[0013] Preferably, an adaptive dimming and energy consumption optimization device for a building lighting system based on digital twins includes at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to perform the adaptive dimming and energy consumption optimization method for a building lighting system based on digital twins as described above.

[0014] Preferably, an electronic device includes the aforementioned adaptive dimming and energy consumption optimization device for building lighting systems based on digital twins.

[0015] Preferably, a computer-readable storage medium stores computer-executable instructions for causing a computer to perform the adaptive dimming and energy consumption optimization method for a digital twin-based building lighting system as described above.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: Addressing the problem of low modeling accuracy, this invention acquires precise photon-level data and surface reflection characteristic data through photon event sensors and event cameras deployed within buildings. It then combines compressed sensing theory and the radiative transfer equation to reconstruct photon trajectories. Coupled with a bidirectional reflectance distribution function parameterized model and a material reflectance field database, it rapidly adapts to new materials and updates the dynamic photon flow field model through transfer learning. Simultaneously, it calculates quantitative indicators to assist in calibration, effectively overcoming the limitations of traditional point-based monitoring and improving the spatiotemporal accuracy of light environment modeling, thus providing support for dimming strategy optimization. Addressing the problem of optimization lacking physical constraints, this invention introduces physical information neural network constraints into the discriminator network within the adversarial generative optimization architecture, ensuring that the scoring process conforms to the radiative transfer equation. Combined with the structural design of the generator network and policy gradient training, it solves the problem of traditional optimization algorithms easily getting trapped in certain situations. To address the issue of local optima, the system ensures that the optimal dimming strategy is both realistic and superior, achieving a precise balance between energy consumption and lighting effect. To address the lack of personalized adaptation, an autoencoder extracts the light environment fingerprint, employs a dynamic time warping algorithm for similarity matching, and combines transfer learning to transfer historical best strategies. This, along with user preference vectors built based on latent semantic models, allows for adaptation to different user lighting preferences, achieving personalized dimming and improving user comfort. To address the lack of closed-loop self-evolution capabilities, a ternary closed-loop self-evolution mechanism is used. Physical space feedback data is used to correct model parameters via Kalman filtering. This involves short-term fine-tuning and long-term batch training of the adversarial generative architecture, combined with federated learning to achieve parameter sharing among multiple buildings. This allows the system to continuously self-evolve as operational data accumulates, improving dimming accuracy and energy efficiency in the long term, adapting to dynamic scene changes, and extending the system's adaptation lifecycle. Attached Figure Description

[0017] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other embodiments can be derived from the provided drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the steps of an adaptive dimming and energy consumption optimization method for a building lighting system based on digital twins; Figure 2 This is a structural diagram of an electronic device. Detailed Implementation

[0019] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions; It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.

[0020] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Example The embodiments of this application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of this application and are not intended to limit the scope of this application. Furthermore, it should be noted that, for ease of description, the accompanying drawings only show the parts related to the embodiments of this application, not all structures. Those skilled in the art, after reading this specification, should be able to conceive that any combination of technical features can constitute an optional implementation method, provided that the technical features do not contradict each other.

[0022] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship. In the description of this application, "multiple" means two or more, and "several" means one or more.

[0023] An adaptive dimming and energy consumption optimization method for building lighting systems based on digital twins, the method comprising: Photon event sensors and event cameras deployed within the building acquire photon data and surface reflection characteristics data of the physical space; Based on compressed sensing theory and radiative transfer equation, the photon data is sparsely sampled and photon trajectory reconstructed. Combined with the bidirectional reflection distribution function parameterization model and material reflection field database, the dynamic photon flow field model in the digital twin is updated. The generator network, based on an adversarial generative optimization architecture, takes the current photon flow field state, personnel distribution, natural light prediction, and user preferences as input and outputs candidate dimming strategies. By using a discriminator network with an adversarial generative optimization architecture and combining physical information neural network constraints, the authenticity and excellence of the candidate dimming strategies are scored, and the optimal dimming strategy is selected. Extract the optical environment fingerprint of the current photon flow field, perform similarity matching with the optical environment fingerprint database, and personalize the optimal dimming strategy through transfer learning; The optimal dimming strategy, after personalized adaptation, is sent to the building lighting actuator to control the operation of the lighting equipment; Based on actual operational data from physical space feedback, a ternary closed-loop self-evolution mechanism is used to perform parameter correction and self-evolution update on the dynamic photon flow field model and adversarial generative optimization architecture, thereby achieving dynamic optimization of energy consumption and lighting effects.

[0024] The process, based on compressed sensing theory and the radiative transfer equation, involves sparse sampling and photon trajectory reconstruction of the photon data. Combined with a bidirectional reflectance distribution function parameterized model and a material reflectance field database, the dynamic photon flow field model in the digital twin is updated, including: The arrival time, incident angle, and wavelength distribution data of individual photons are collected by a photon event sensor, and sparse sampling is performed. Based on compressed sensing theory and combined with the radiative transfer equation as a physical constraint, the photon flow field distribution of the building space is reconstructed from sparsely sampled photon data. The changes in the reflectivity of the building's interior surfaces are captured by an event camera, and the optical properties of each surface in the digital twin are updated in real time using a bidirectional reflectance distribution function parameterized model. By calling the material reflection field database and using transfer learning to quickly match the reflection characteristics of newly emerging materials, the dynamic photon flow field model is updated, and the photon flux density, average photon free path, and anisotropy coefficient quantification index are calculated.

[0025] The generator network, which employs an adversarial generative optimization architecture, takes into input the current photon flow field state, personnel distribution, natural light prediction, and user preferences, and outputs candidate dimming strategies, including: The generator network adopts a structure combining variational autoencoder and Transformer to capture the spatiotemporal correlation of photon flow field and related input parameters; The current photon flow field state, the distribution data of people in the building, the natural light prediction data for the future preset duration, and the user preference vector are obtained and input into the generator network. The generator network is trained based on the policy gradient algorithm and outputs candidate dimming strategies that include the brightness, color temperature, and beam angle parameters of each lighting device.

[0026] The discriminator network, employing an adversarial generative optimization architecture and combined with physical information neural network constraints, scores the authenticity and superiority of the candidate dimming strategies to select the optimal dimming strategy, including: The discriminator network adopts a 3D convolutional neural network structure to extract the spatiotemporal features of the simulated photon flow field after the candidate dimming strategy is executed; The discriminator network introduces a physical information neural network as a constraint to ensure that the scoring process conforms to the radiative transfer equation and outputs the authenticity score and excellence score of the candidate dimming strategy. Based on the preset scoring threshold, candidate dimming strategies that meet both the authenticity and excellence scores are selected as the optimal dimming strategy; if there are multiple candidate strategies that meet the threshold, the strategy with the highest comprehensive score is selected as the optimal dimming strategy.

[0027] The process of extracting the optical environment fingerprint of the current photon flow field, performing similarity matching with the optical environment fingerprint database, and then personalizing the optimal dimming strategy through transfer learning includes: An autoencoder is used to reduce the dimensionality of the current photon flow field state and extract a light environment fingerprint that includes spatial illuminance distribution features, spectral distribution features, and dynamic change features. The similarity between the current ambient fingerprint and historical fingerprints in the ambient fingerprint database is calculated, and the similarity is measured using a dynamic time warping algorithm. The optimal strategy corresponding to the historical scene with the highest similarity is retrieved, and the historical strategy is transferred to the current scene through transfer learning. The optimal dimming strategy is then adjusted in a personalized way to suit the current user preferences.

[0028] The actual operational data based on physical space feedback is used to perform parameter correction and self-evolution updates on the dynamic photon flow field model and the adversarial generative optimization architecture through a ternary closed-loop self-evolution mechanism, including: It receives actual lighting data, energy consumption data, and user feedback data from sensors in the physical space, compares them with simulated data in the digital twin, and uses the Kalman filter algorithm to correct the parameters of the dynamic photon flow field model. Based on actual operational data, the generator and discriminator networks of the adversarial generative optimization architecture are fine-tuned in the short term; the network is batch trained and the network weights are updated every morning using the accumulated operational data of the day. By using a federated learning framework, model parameters can be shared among multiple buildings, accelerating the system's self-evolution process and improving optimization results.

[0029] The construction of the user preference vector includes: the digital twin control platform collecting the user's dimming operation history in different scenarios, extracting dimensional parameters such as preferred illuminance, preferred color temperature, glare sensitivity, and dynamic response speed, and using a latent semantic model to mine the implicit features of user preferences to construct the user preference vector.

[0030] In the specific implementation process, such as Figure 1As shown, sensing devices are deployed to acquire physical space photon data and surface reflection characteristic data. Optionally, this step is the physical data input layer of the digital twin. The core is to deploy photon event sensors and event cameras inside the building to achieve photon-level fine perception of the light environment and real-time capture of the building surface reflection characteristics, providing raw data for subsequent digital twin model construction.

[0031] Photon event sensors are uniformly deployed along a grid pattern throughout the building's interior space, covering all illuminated areas to ensure comprehensive light environment perception. Event cameras are deployed in unobstructed key locations within the building to capture changes in the visual characteristics and reflectivity of various surfaces such as walls, floors, and furniture. The photon event sensors collect raw photon-level data, including arrival time, incident angle, and wavelength distribution of individual photons in the physical space, enabling microscopic perception of the light environment. The event cameras capture dynamic changes in the reflectivity of various surfaces within the building, reflecting the changing patterns of light reflection by the surface materials. The collected photon data and surface reflectivity data are transmitted in real-time to the digital twin control platform via IoT communication protocols such as Industrial Ethernet, ZigBee, and 5G. The data transmission frequency is adapted to the rate of change in the light environment, ensuring the real-time nature and effectiveness of the data.

[0032] Update the dynamic photon flow field model in the digital twin. Optionally, this step is the core modeling step of the digital twin. Based on compressed sensing theory and the radiative transfer equation, combined with the bidirectional reflectance distribution function (BRDF) parameterized model and material reflectance field database, the collected raw data is processed to achieve real-time updates of the dynamic photon flow field model, completing the accurate mapping of the physical light environment to the digital space. The specific implementation is as follows: Sparsely sample the photon-level data collected by the photon event sensor to reduce the data dimensionality while retaining effective light environment information, thereby reducing subsequent computational overhead. The sampling process follows the sparse representation rules of compressed sensing theory. Using the radiative transfer equation as a physical constraint, substitute the sparsely sampled photon data into the equation for solution, reconstruct the photon trajectory of the entire indoor area of ​​the building, restore the spatial distribution characteristics of the photon flow field, break through the limitations of traditional point-based monitoring by illuminance sensors, and achieve full-domain modeling of the light environment. Through the surface reflectance characteristic change data captured by the event camera, combined with the bidirectional reflectance distribution function (BRDF) parameterized model, update the optical characteristics of each surface of the building in the digital twin in real time, accurately characterizing the light reflection laws of different material surfaces. The system utilizes a pre-built material reflection field database (containing reflection characteristic parameters of common building materials such as wood, glass, tiles, and fabrics). When an unknown new material appears indoors, a transfer learning algorithm quickly matches its reflection characteristics, eliminating the need for remodeling and improving model update efficiency. After model update, three quantitative indicators—photon flux density, average photon free path, and anisotropy coefficient—are calculated as the basis for evaluating the accuracy of the photon flow field model. Simultaneously, they provide quantitative light environment state parameters for generating dimming strategies.

[0033] The generator network outputs candidate dimming strategies. Optionally, this step is the dimming strategy generation stage. Based on the generator network of the adversarial generative optimization architecture, multiple sets of candidate dimming strategies that meet basic lighting requirements are generated by inputting the light environment and building lighting-related influencing factors. The specific implementation is as follows: A hybrid structure combining variational autoencoder (VAE) and Transformer is used. The VAE completes feature extraction and dimensionality reduction of the input parameters, while the Transformer captures the spatiotemporal correlation of input parameters such as photon flow field and personnel distribution through a self-attention mechanism, adapting to the characteristics of the building light environment that dynamically changes with time and space. Four types of core parameters are input to the generator network: the current photon flow field state of the digital twin, personnel distribution data in the building (acquired by human body sensing sensors / video surveillance equipment), future preset duration natural light prediction data (generated by weather forecast and building lighting model), and user preference vector. The digital twin control platform collects users' dimming operation history in different scenarios, extracts parameters such as preferred illuminance, preferred color temperature, glare sensitivity, and dynamic response speed, and uses a latent semantic model to mine the implicit features of user preferences. After normalization, a multi-dimensional user preference vector is constructed to represent the user's personalized lighting needs. The generator network is trained based on a policy gradient algorithm, continuously optimizing network weights with the reward function of minimizing energy consumption and optimizing lighting effect. After training, multiple sets of candidate dimming strategies are output. Each strategy includes three control parameters: brightness, color temperature, and beam angle of each lighting device in the building, adapting to various types of smart lighting devices such as LED lights, panel lights, and spotlights.

[0034] The optimal dimming strategy is selected through a discriminator network. Optionally, this step is the dimming strategy selection stage. Based on an adversarial generative optimization architecture, the discriminator network, combined with Physical Information Neural Network (PINN) constraints, performs dual scoring on candidate dimming strategies for both realism and excellence, selecting the optimal dimming strategy that balances physical feasibility, energy consumption control, and lighting effect. The specific implementation is as follows: The discriminator network adopts a 3D Convolutional Neural Network (3DCNN) structure, which can effectively extract the spatiotemporal characteristics of the simulated photon flow field in the digital twin after the candidate dimming strategy is executed, while capturing the spatial distribution and temporal dynamic changes of the light environment. In the scoring process of the discriminator network, the Physical Information Neural Network is introduced as a hard constraint, and the radiative transfer equation is embedded in the network loss function to ensure that the scoring process strictly conforms to the physical laws of light propagation, avoids generating invalid dimming strategies that violate physical principles, and guarantees the realism of the strategy. The discriminator network outputs a authenticity score and an excellence score of 0-10 for each group of candidate dimming strategies. The authenticity score reflects the physical feasibility of the strategy, while the excellence score reflects the balance between energy consumption and lighting effect. The digital twin control platform presets a scoring threshold and selects candidate strategies that meet both scores. If there are multiple qualified strategies, the strategy with the highest comprehensive score (authenticity × 0.4 + excellence × 0.6) is selected as the optimal dimming strategy.

[0035] Light environment fingerprint matching and personalized adaptation of the optimal strategy. Optionally, this step is the personalized optimization stage of the dimming strategy. By extracting the light environment fingerprint, similarity matching, and transfer learning, the optimal dimming strategy is deeply combined with the current user preferences and scene features to achieve personalized adaptation of the strategy. The specific implementation is as follows: An autoencoder is used to reduce the dimensionality of the photon flow field state of the current digital twin, remove redundant information, and extract the core feature set containing spatial illuminance distribution features, spectral distribution features, and dynamic change features as the unique light environment fingerprint of the current light environment. The current light environment fingerprint is matched with the light environment fingerprint database of the digital twin control platform (which stores fingerprints of historical lighting scenes and corresponding optimal strategies). The Dynamic Time Warping (DTW) algorithm is used to measure the similarity to solve the problem of time dimension misalignment of fingerprints in different scenarios and improve the matching accuracy. The system retrieves the historical scene with the highest similarity to the current fingerprint from the fingerprint database and extracts the corresponding historical optimal dimming strategy. The parameters of the historical strategy are then transferred to the current scene using a transfer learning algorithm. Local parameter adjustments are made to the selected optimal dimming strategy to ensure that the strategy retains global optimality while accurately adapting to the current user's lighting preferences.

[0036] Strategy Distribution and Lighting Equipment Control. Optionally, this step is the physical execution phase of the dimming strategy. The personalized, optimal dimming strategy is distributed to the building lighting actuators to achieve precise control of the lighting equipment, completing the transition from digital space strategy generation to physical space equipment operation. The specific implementation is as follows: The digital twin control platform parses the personalized, optimal dimming strategy according to the partitions and numbers of the building's indoor lighting equipment, converting brightness, color temperature, and beam angle parameters into digital / analog control commands recognizable by the lighting actuators. These control commands are distributed to the corresponding building lighting actuators via communication protocols such as Modbus, Profinet, and Bluetooth. Upon receiving the commands, the actuators adjust the operating parameters of the lighting equipment in real time to achieve adaptive dimming. The delay between command distribution and equipment response is controlled within milliseconds to ensure real-time control. After the lighting equipment is running, the actuators feed back the actual operating parameters and energy consumption data of the equipment to the digital twin control platform in real time, providing actual physical space operating data for the three-dimensional closed-loop self-evolution.

[0037] The ternary closed-loop self-evolution mechanism enables dynamic optimization of the model and architecture. Optionally, this step is a continuous optimization phase of the system. Based on actual operational data from the physical space, the ternary closed-loop self-evolution mechanism is used to correct parameters and update the network of the dynamic photon flow field model and the adversarial generative optimization architecture, so that the system performance continuously improves with the accumulation of operational data. The specific implementation is as follows: The control platform receives actual lighting data, energy consumption data, and user feedback data from the physical space sensors, compares them with the simulated data of the digital twin, and uses the Kalman filter algorithm to correct the parameters of the photon flow field model in real time, reducing model prediction errors and achieving accurate synchronization between the digital twin and the physical space. Based on actual operational data, the generator and discriminator networks are trained and updated in two stages: first, short-term fine-tuning, which involves fine-tuning the network with small batch gradient descent based on real-time feedback data to adapt to short-term dynamic changes in the lighting environment; second, long-term batch training, which involves full batch training of the network using the accumulated operational data of the day during low-load periods of building lighting (such as early morning each day), updating the network weights, and improving the accuracy of strategy generation and selection. A federated learning framework is built, with digital twin control platforms of multiple buildings as clients and a cloud platform as a server. After each client completes model and network updates locally, it uploads the updated model parameters (without leaking the original data) to the cloud platform. The cloud platform aggregates and optimizes the parameters and then distributes them to each client. By sharing parameters among multiple buildings, the system's self-evolution process is accelerated, and the overall optimization effect is improved.

[0038] An adaptive dimming and energy consumption optimization device for a building lighting system based on digital twins includes at least one control processor (801) and a memory (802) for communicatively connecting to the at least one control processor (801); the memory (802) stores instructions that can be executed by the at least one control processor (801), the instructions being executed by the at least one control processor (801) to enable the at least one control processor (801) to perform the adaptive dimming and energy consumption optimization method for a building lighting system based on digital twins as described above.

[0039] The optimization device in this embodiment is a functional module combining hardware and software. It is the core carrier for implementing the above optimization method and is integrated into an electronic device. The specific implementation is as follows: The device includes at least one control processor (801) and a memory (802) that is communicatively connected to the control processor (801). The two interact with each other and transmit instructions through an internal bus, which is compatible with standards such as Universal Serial Bus and Industrial Internal Bus.

[0040] The memory (802) is a non-volatile storage medium that stores computer-executable instructions that can be executed by the control processor (801). It also stores various types of data required for the implementation of the method, such as photon data, material reflection field database, light environment fingerprint database, user operation history, model and network parameters. The instructions are modularly divided according to the seven steps of the method to form code segments that can be called independently and run collaboratively.

[0041] The control processor (801) adopts a high-performance microprocessor or digital signal processor (DSP). By running the executable instructions in the memory (802), it sequentially performs all operations such as data acquisition and parsing, digital model updating, dimming strategy generation / screening / personalized adaptation, strategy distribution, and model self-evolution. If there are multiple control processors (801), the computing tasks are distributed in a distributed computing manner to improve the system operating efficiency.

[0042] This device serves as a hub connecting physical sensing devices, lighting actuators, and a digital twin control platform, enabling bidirectional interaction between physical and digital data, computational operation of algorithm models, and generation and issuance of control commands.

[0043] An electronic device comprising the aforementioned adaptive dimming and energy consumption optimization device for building lighting systems based on digital twins.

[0044] The electronic device in this embodiment is a general-purpose intelligent control device, serving as the hardware carrier for implementing the aforementioned optimization method. It integrates the aforementioned adaptive dimming and energy consumption optimization device for building lighting systems based on digital twins, and its hardware architecture is as follows: Figure 2As shown, in addition to the core control processor (801) and memory (802), it also includes multiple peripheral functional modules (input devices (803), output devices (804), etc.). Each module works together to achieve complete lighting control functions. The specific implementation is as follows: The core module is the control processor (801) and memory (802) of the above-mentioned optimized device. It undertakes the calculation, control and data storage functions of the entire device and is the core of the electronic device. The sensing interface module is the connection interface between the electronic device and physical sensing devices such as photon event sensors, event cameras, and human body induction sensors. It is compatible with general industrial interfaces such as RS485 and Ethernet ports, realizes real-time reception and transmission of physical space data, and supports multiple devices to be connected at the same time. The actuator interface module is the connection interface between the electronic device and the building lighting actuator. It is compatible with multiple communication protocols such as Modbus, ZigBee, and Bluetooth, realizes the issuance of dimming control commands and the feedback reception of the operating status of lighting equipment. The communication module includes wired communication units (industrial Ethernet) and wireless communication units (5G / LoRa / ZigBee). It enables local communication between electronic devices and sensors / actuators, as well as remote communication with the cloud platform, providing communication support for sharing multiple building parameters in federated learning. The human-machine interface module includes a touchscreen, physical buttons, and a voice interaction unit, providing users with an operational interface. Users can submit lighting effect feedback, adjust lighting preferences, and manually intervene in dimming strategies through this module, while simultaneously displaying real-time system operating status, energy consumption data, and other information. The power supply module provides stable AC power to all modules of the electronic equipment and features overvoltage, overcurrent, and short-circuit protection, ensuring stable and long-term operation of the equipment in architectural lighting scenarios.

[0045] Overall workflow of the equipment: The electronic device receives physical space data through the sensor interface module, and the control processor (801) runs the instructions in the memory (802) to complete the digital twin model update, dimming strategy generation and optimization. The strategy is sent to the lighting actuator through the actuator interface module. At the same time, the actual operation data of the physical space is received through the communication module to complete the self-evolution update of the model and the network. The human-machine interaction module realizes the two-way interaction between the user and the system.

[0046] A computer-readable storage medium storing computer-executable instructions for causing a computer to perform the above-described adaptive dimming and energy consumption optimization method for a digital twin-based building lighting system.

[0047] In this embodiment, the computer-readable storage medium is a non-transitory, readable and writable storage medium storing computer-executable instructions. When these instructions are executed by the control processor (801) of the electronic device, all steps and preferred schemes of the above-mentioned adaptive dimming and energy consumption optimization method for building lighting systems based on digital twins can be implemented. The specific implementation is as follows: The storage media in this embodiment include, but are not limited to, solid-state drives (SSDs), hard disk drives (HDDs), USB flash drives, memory cards, read-only optical discs (CD-ROMs), phase-change memory (PRAMs), etc. Any non-transfer medium that can store computer-executable instructions and can be read by the processor of an electronic device can be used as the storage medium in this embodiment.

[0048] The computer-executable instructions in the storage medium are divided into seven code segments according to their functions: data acquisition module, model update module, strategy generation module, strategy filtering module, personalization adaptation module, strategy execution module, and self-evolution module. These segments correspond to the seven steps of the method. Each code segment communicates with the others through a standardized data interface and can run collaboratively or be called independently.

[0049] The control processor (801) of the electronic device reads and loads executable instructions from the storage medium into memory, and calls each code segment in sequence according to the process to complete the entire process operation from physical data acquisition to model self-evolution; the intermediate data generated during the instruction execution is temporarily stored in memory, and the execution result is written to the storage medium in real time to realize the persistent storage of data.

[0050] The executable instructions in the storage medium are cross-platform code that can run on various electronic device operating systems such as Windows, Linux, and embedded operating systems. It is compatible with different brands and models of intelligent control devices and has good portability and compatibility. The same or similar labels correspond to the same or similar parts; The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent. Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all implementation methods here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A method for adaptive dimming and energy consumption optimization of a building lighting system based on digital twins, characterized in that, The method includes: Photon event sensors and event cameras deployed within the building acquire photon data and surface reflection characteristics data of the physical space; Based on compressed sensing theory and radiative transfer equation, the photon data is sparsely sampled and photon trajectory reconstructed. Combined with the bidirectional reflection distribution function parameterization model and material reflection field database, the dynamic photon flow field model in the digital twin is updated. The generator network, based on an adversarial generative optimization architecture, takes the current photon flow field state, personnel distribution, natural light prediction, and user preferences as input and outputs candidate dimming strategies. By using a discriminator network with an adversarial generative optimization architecture and combining physical information neural network constraints, the authenticity and excellence of the candidate dimming strategies are scored, and the optimal dimming strategy is selected. Extract the optical environment fingerprint of the current photon flow field, perform similarity matching with the optical environment fingerprint database, and personalize the optimal dimming strategy through transfer learning; The optimal dimming strategy, after personalized adaptation, is sent to the building lighting actuator to control the operation of the lighting equipment; Based on actual operational data from physical space feedback, a ternary closed-loop self-evolution mechanism is used to perform parameter correction and self-evolution update on the dynamic photon flow field model and adversarial generative optimization architecture, thereby achieving dynamic optimization of energy consumption and lighting effects.

2. The optimization method according to claim 1, characterized in that, The process, based on compressed sensing theory and the radiative transfer equation, involves sparse sampling and photon trajectory reconstruction of the photon data. Combined with a bidirectional reflectance distribution function parameterized model and a material reflectance field database, the dynamic photon flow field model in the digital twin is updated, including: The arrival time, incident angle, and wavelength distribution data of individual photons are collected by a photon event sensor, and sparse sampling is performed. Based on compressed sensing theory and combined with the radiative transfer equation as a physical constraint, the photon flow field distribution of the building space is reconstructed from sparsely sampled photon data. The changes in the reflectivity of the building's interior surfaces are captured by an event camera, and the optical properties of each surface in the digital twin are updated in real time using a bidirectional reflectance distribution function parameterized model. By calling the material reflection field database and using transfer learning to quickly match the reflection characteristics of newly emerging materials, the dynamic photon flow field model is updated, and the photon flux density, average photon free path, and anisotropy coefficient quantification index are calculated.

3. The optimization method according to claim 1, characterized in that, The generator network, which employs an adversarial generative optimization architecture, takes into input the current photon flow field state, personnel distribution, natural light prediction, and user preferences, and outputs candidate dimming strategies, including: The generator network adopts a structure combining variational autoencoder and Transformer to capture the spatiotemporal correlation of photon flow field and related input parameters; The current photon flow field state, the distribution data of people in the building, the natural light prediction data for the future preset duration, and the user preference vector are obtained and input into the generator network. The generator network is trained based on the policy gradient algorithm and outputs candidate dimming strategies that include the brightness, color temperature, and beam angle parameters of each lighting device.

4. The optimization method according to claim 1, characterized in that, The discriminator network, employing an adversarial generative optimization architecture and combined with physical information neural network constraints, scores the authenticity and superiority of the candidate dimming strategies to select the optimal dimming strategy, including: The discriminator network adopts a 3D convolutional neural network structure to extract the spatiotemporal features of the simulated photon flow field after the candidate dimming strategy is executed; The discriminator network introduces a physical information neural network as a constraint to ensure that the scoring process conforms to the radiative transfer equation and outputs the authenticity score and excellence score of the candidate dimming strategy. Based on the preset scoring threshold, candidate dimming strategies that meet both the authenticity and excellence scores are selected as the optimal dimming strategy; if there are multiple candidate strategies that meet the threshold, the strategy with the highest comprehensive score is selected as the optimal dimming strategy.

5. The optimization method according to claim 1, characterized in that, The process of extracting the optical environment fingerprint of the current photon flow field, performing similarity matching with the optical environment fingerprint database, and then personalizing the optimal dimming strategy through transfer learning includes: An autoencoder is used to reduce the dimensionality of the current photon flow field state and extract a light environment fingerprint that includes spatial illuminance distribution features, spectral distribution features, and dynamic change features. The similarity between the current ambient fingerprint and historical fingerprints in the ambient fingerprint database is calculated, and the similarity is measured using a dynamic time warping algorithm. The optimal strategy corresponding to the historical scene with the highest similarity is retrieved, and the historical strategy is transferred to the current scene through transfer learning. The optimal dimming strategy is then adjusted in a personalized way to suit the current user preferences.

6. The optimization method according to claim 1, characterized in that, The actual operational data based on physical space feedback is used to perform parameter correction and self-evolution updates on the dynamic photon flow field model and adversarial generative optimization architecture through a ternary closed-loop self-evolution mechanism, including: It receives actual lighting data, energy consumption data, and user feedback data from sensors in the physical space, compares them with simulated data in the digital twin, and uses the Kalman filter algorithm to correct the parameters of the dynamic photon flow field model. Based on actual operational data, the generator and discriminator networks of the adversarial generative optimization architecture are fine-tuned in the short term; the network is batch trained and the network weights are updated every morning using the accumulated operational data of the day. By using a federated learning framework, model parameters can be shared among multiple buildings, accelerating the system's self-evolution process and improving optimization results.

7. The optimization method according to claim 1, characterized in that, The construction of the user preference vector includes: the digital twin control platform collecting the user's dimming operation history in different scenarios, extracting dimensional parameters such as preferred illuminance, preferred color temperature, glare sensitivity, and dynamic response speed, and using a latent semantic model to mine the implicit features of user preferences to construct the user preference vector.

8. A digital twin-based adaptive dimming and energy consumption optimization device for building lighting systems, characterized in that, It includes at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, which, when executed by the at least one control processor, enable the at least one control processor to perform the adaptive dimming and energy consumption optimization method for a building lighting system based on digital twins as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, Includes the adaptive dimming and energy consumption optimization device for building lighting systems based on digital twins as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the adaptive dimming and energy consumption optimization method for a building lighting system based on digital twins as described in any one of claims 1 to 7.

Images