An underground parking lot intelligent lighting control system based on multi-objective real-time optimization

Abstract

The application discloses a kind of underground parking lot intelligent lighting control systems based on multi-objective real-time optimization, it is related to intelligent lighting control technical field, for the limitation of traditional lighting scheme, scheme includes: fast response layer, it is deployed in each lamp node, for real-time acquisition each node state data, through neighborhood negotiation and event trigger mechanism realizes millisecond level illumination response;Edge optimization layer, it is run in subarea control node, for building comprehensive optimization objective function, through submodular optimization algorithm and convex approximation method solve and output each lamp node optimal illumination intensity and power distribution parameter;Global learning layer, for the subarea operation data based on edge optimization layer, in combination with distributed reinforcement learning and federated learning technology, realize multi-objective weight dynamic optimization and strategy light weight, then issue to each subarea control node, support lighting control system long-term adaptive optimization and stable operation.The application is directed to underground parking lot.

Description

Technical Field

[0001] This invention relates to the field of intelligent lighting control technology, specifically to an intelligent lighting control system for underground parking lots based on multi-objective real-time optimization. Background Technology

[0002] Current energy-saving control systems for underground parking lot lighting largely rely on timed switching or simple zone-sensor triggering modes. These solutions exhibit several technical bottlenecks in practical applications: First, insufficient real-time response, leading to "black zones" due to trigger delays when vehicles pass quickly, threatening safety; second, a lack of flexibility in energy consumption control, with centralized architectures struggling to dynamically adjust dimming according to traffic flow, resulting in persistently high energy consumption during off-peak hours; third, limited lamp lifespan, with frequent start-stop cycles accelerating LED light decay due to thermal shock; fourth, poor visual comfort, with sudden brightness changes causing driver fatigue; and fifth, poor computational scalability, making it difficult for centralized AI algorithms to run in real-time at the edge. To address these issues, existing research often focuses on minimizing single-objective energy consumption, neglecting key performance aspects such as lighting continuity, LED thermal stability, and edge computing complexity. This leads to optimization solutions that are "one-sided"—either sacrificing safety for energy saving or failing to be implemented due to algorithmic complexity.

[0003] Therefore, there is an urgent need for a control method that breaks through the limitations of traditional centralized architecture and single-objective optimization, can run on edge nodes, has a hierarchical structure of "fast response-regional optimization-global learning", and supports "real-time trade-offs of multiple objectives under the constraint of safety and continuity", so as to achieve the synergistic optimization of safety, energy saving, comfort and stability of underground parking lot lighting. Summary of the Invention

[0004] This invention addresses the limitations of traditional solutions that rely on "single-objective optimization" and "centralized control" by providing an intelligent lighting control system for underground parking lots based on multi-objective real-time optimization.

[0005] The present invention provides an intelligent lighting control system for underground parking lots based on multi-objective real-time optimization, and the technical solution adopted to solve the above-mentioned technical problems is as follows:

[0006] An intelligent lighting control system for underground parking lots based on multi-objective real-time optimization, comprising:

[0007] The fast response layer is deployed on each lamp node to collect the status data of each lamp node in real time. Through neighborhood negotiation and event triggering mechanism, it achieves millisecond-level illumination response to ensure continuous and uninterrupted illumination along the path of people and vehicles. At the same time, it reports the collected lamp node status data and continuity indicators to the edge optimization layer.

[0008] The edge optimization layer operates on the zonal control nodes of each level of the underground parking lot. It takes the lamp node status data and continuity indicators uploaded by the fast response layer as input to construct a comprehensive optimization objective function. After solving the function through the sub-modulus optimization algorithm and the convex approximation method, it outputs the optimal illuminance and power allocation parameters of each lamp node, completes the precise control of lighting at the zonal scale, and reports the zonal operation data to the global learning layer.

[0009] The global learning layer, deployed on cloud servers or edge servers, is used to perform partition operation data reported by the edge optimization layer. Through a "periodic data aggregation-policy iteration-lightweight distribution" model, combined with distributed reinforcement learning and federated learning technologies, it achieves dynamic optimization of multi-objective weights and lightweight policy distribution, and then distributes the data to each partition control node to support the long-term adaptive optimization and stable operation of the lighting control system.

[0010] Optionally, the fast response layer involved is deployed on each independent luminaire node, and each luminaire node integrates hardware units and control logic; wherein: the hardware unit specifically includes a microcontroller unit, a light intensity sensor unit, a temperature sensor unit, a PWM dimming driver unit, and a short-range communication unit; the control logic achieves millisecond-level light response.

[0011] Further optional, each luminaire node Every 200ms, it sends a message to the neighboring lighting nodes. Broadcast state vector :

[0012] ,

[0013] in: The current illuminance of the luminaire represents the actual illuminance currently output by the luminaire node. Power represents the current real-time power consumption of the lighting node; The value is temperature, representing the real-time temperature of the lamp chip. The health status reflects the health level of the lighting fixture node hardware; The communication status reflects the quality of the communication link between the lighting node and its neighbors; The set of neighboring light fixture nodes, i.e., the set of light fixture nodes. A local lighting unit consisting of adjacent luminaires that communicate directly.

[0014] Optionally, the control logic integrated into the luminaire nodes involved is based on the illuminance information in the state vector broadcast by neighboring luminaire nodes, satisfying the following illuminance update rules:

[0015] ,

[0016] in: Indicates the lighting fixture node At any moment ; output light intensity; Indicates the lighting fixture node In the next cycle Light intensity; The update step size represents the update cycle of the illuminance intensity of the luminaire node; The negotiation coefficient is 0 < <1, determines the fusion ratio between its own light intensity and the average light intensity of the neighborhood; The number of neighboring light fixture nodes reflects the size of the neighborhood. The illuminance of neighboring luminaire nodes, i.e., the illuminance of each luminaire node within the neighborhood. At any moment The light intensity value is a reference benchmark for the light intensity in the neighborhood.

[0017] Alternatively, the fast-response layer may perform the following operations to achieve millisecond-level illumination response, ensuring continuous and uninterrupted illumination along the path of people and vehicles:

[0018] (1) The fast response layer calculates the path light track connectivity based on the current illumination intensity data obtained by updating the lamp nodes according to the rules. And set constraints to ensure the integrity of path lighting:

[0019] ,

[0020] in: The path light track connectivity describes a specific vehicle or pedestrian path. The relative degree of protection of light intensity at the weakest point; It is a set of paths or a set of lamp node indices for a single path; This is the index of the lighting node on the path; For path The first Individual lighting fixture node identification; For lighting fixture nodes The actual light intensity of the current output; The minimum safe light intensity threshold;

[0021] (2) Path optical track connectivity based on each path Constructing overall system continuity indicators The overall continuity index Provide a basis for judgment in the emergency response mechanism of the rapid response layer:

[0022] ,

[0023] in: Indicates the vehicle travel paths contained within a certain lighting area. The corresponding path optical track connectivity; For the first The weight of a path reflects the traffic flow density or frequency of use of that path;

[0024] (i) In certain extreme security-priority scenarios, the system adopts a minimum guarantee strategy, in which case the formula is used. This means that if any critical path fails to meet the standard, the system must take remedial measures immediately.

[0025] (ii) In general scenarios, a formula with normalized illuminance is used to compare the illuminance of each luminaire node with the minimum illuminance and sum and average them to comprehensively evaluate the uniformity and continuity of lighting throughout the entire area:

[0026] ,

[0027] in: This represents the total number of lighting fixture nodes counted within the region. This represents the current actual illumination intensity of each luminaire node, reflecting the actual illumination intensity received by that luminaire node. It is the minimum illuminance, which is the minimum illuminance among all luminaire nodes in the entire area;

[0028] (3) The fast response layer combines the illuminance data in the state vector of the luminaire node with the overall continuity index of the system, and triggers an emergency mechanism when it detects that the illuminance of a neighboring luminaire node does not conform to the preset change:

[0029] ,

[0030] in: The illuminance of a neighboring luminaire node represents the illuminance of a monitored neighboring luminaire node. Real-time light intensity or light intensity output; To ensure a safe light intensity threshold and guarantee that local areas reach a safe lighting level, this emergency mechanism must be completed within a preset period to achieve self-restoring lighting continuity.

[0031] Further, optionally, the comprehensive optimization objective function involved in the construction of the edge optimization layer. as follows:

[0032] ,

[0033] ,

[0034] ,

[0035] ,

[0036] ,

[0037] ,

[0038] ,

[0039] in: , , , , Indicates the weighting coefficient; Indicates system energy consumption. The t represents the real-time power of the i-th lamp node, and the t represents the running time. This represents the average lifespan loss rate of each lighting fixture node. For the number of nodes, For the first The lifespan loss rate of each lighting fixture node. For real-time temperature With real-time power Instantaneous lifespan loss, This represents the maximum reference loss of the luminaire node under design limits, used to normalize the lifespan loss so that the lifespan loss ratio is within [0,1]. Coefficients c, d, and e are used to adjust the contribution and influence of temperature and power, where c, d, and e are all greater than 0, and c + d + e = 1. Indicates the selected reference temperature. Indicates the selected reference power; Indicates visual comfort, Indicates the current light intensity. To update the step size, Indicates the light intensity at the previous moment; The computational complexity is used to quantify the overall computational cost of the algorithm model in the system. The model inference time reflects the time it takes for the algorithm model to process the input data and output the results, which is directly related to the real-time response capability of the system. The parameter scale refers to the total number of parameters to be optimized in the algorithm model, which directly determines the amount of storage space occupied by the model. Indicates thermal stability, This represents the rate of change of the temperature of the i-th lamp node over time.

[0040] Preferably, in terms of thermal stability Thermal dynamic constraints are introduced during the calculation process:

[0041] ,

[0042] in: The power-temperature rise coefficient is a proportionality factor that characterizes the temperature rise caused by power input. This represents the real-time power of the i-th lamp node; The heat dissipation attenuation coefficient characterizes the degree of influence of temperature difference on the heat dissipation rate. This represents the real-time temperature of the i-th lighting node; The ambient temperature is an external boundary condition and serves as a reference for heat dissipation at the luminaire nodes.

[0043] Alternatively, the global learning layer may perform the following operations:

[0044] (1) The global learning layer periodically receives partition operation data reported by each edge optimization layer, and optimizes the multi-objective weight vector through a distributed reinforcement learning algorithm. It also outputs dynamically adjusted weight parameters using a customized reward function:

[0045] ,

[0046] ,

[0047] ,

[0048] in: For multi-objective weight vectors; Energy consumption weighting coefficient; This is a weighting factor for the lifespan of the lighting fixture; This is a comfort-weighted coefficient. To calculate the complexity weighting coefficients; This is the thermal stability weighting coefficient;

[0049] This is a safety reward coefficient used to amplify safety indicators. Weight in the total reward; For continuous reward coefficients, amplify continuous indicators. Its influence; This is an energy consumption penalty factor used to reduce system energy consumption. It is considered a negative item in the reward; The thermal penalty coefficient is used to determine the thermal stability. It is considered a negative item in the reward; To calculate the complexity penalty coefficient, the computational complexity will be... It is considered a negative item in the reward;

[0050] (2) Using the weight optimization results as input, the global learning layer uses federated learning technology to reuse the optimization experience of mature scenarios through policy transfer, and then compresses the model volume through model distillation to finally generate a lightweight optimization policy;

[0051] (3) The lightweight optimization strategy is distributed to each partition control node, which not only supports the partition control node to independently run the solution process of the comprehensive optimization objective function in real time based on the strategy (by sub-modular optimization algorithm and convex approximation method), but also adapts to the addition of new partition control nodes or lighting nodes, and the replacement of faulty partition control nodes or lighting nodes, so as to ensure the long-term stable operation and continuous adaptive optimization of the lighting control system.

[0052] The intelligent lighting control system for underground parking lots based on multi-objective real-time optimization of the present invention has the following advantages compared with the prior art:

[0053] This invention employs a three-layer hierarchical architecture of "fast response layer - edge optimization layer - global learning layer," breaking through the limitations of traditional centralized control or single-objective optimization. It dynamically balances lighting safety, continuity, energy consumption, luminaire lifespan, visual comfort, edge computing complexity, and LED thermal stability. Under strict constraints of lighting safety and continuity, it can reduce energy consumption by 20%-30%, extend LED lifespan by 20%, and reduce temperature fluctuations by 35%, while supporting real-time operation of edge nodes. Through collaborative innovation in architecture, algorithm, and strategy, this system offers significant advantages over existing solutions in terms of safety, comfort, energy efficiency, and maintainability. Attached Figure Description

[0054] Appendix Figure 1 This is the system architecture diagram of the present invention. Detailed Implementation

[0055] To make the technical solution, the technical problem solved, and the technical effect of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with specific embodiments.

[0056] Example 1: Refer to Appendix Figure 1 This embodiment proposes an intelligent lighting control system for underground parking lots based on multi-objective real-time optimization, which includes a fast response layer, an edge optimization layer, and a global learning layer.

[0057] (i) Fast Response Layer: Deployed on each lamp node, it is used to collect the status data of each lamp node in real time. Through neighborhood negotiation and event triggering mechanism, it achieves millisecond-level illumination response to ensure continuous and uninterrupted illumination along the path of people and vehicles. At the same time, it reports the collected lamp node status data and continuity indicators to the edge optimization layer.

[0058] Specifically, the fast response layer is deployed on each independent luminaire node, and each luminaire node integrates hardware units and control logic; wherein: the hardware units specifically include microcontroller unit (MCU), light intensity sensor unit, temperature sensor unit, PWM dimming driver unit and short-range communication unit; the control logic realizes millisecond-level light response.

[0059] Each lighting fixture node Every 200ms, it sends a message to the neighboring lighting nodes. Broadcast state vector :

[0060] ,

[0061] in: The current illuminance of the luminaire represents the actual illuminance output by the luminaire node (unit: lux). It is a core control variable used for continuous calculation of illuminance. It is directly collected by the illuminance sensor unit built into the luminaire node, or derived from the PWM dimming signal and the luminaire illuminance calibration curve.

[0062] Power, representing the current real-time power consumption of the lighting node (unit: W), is used for energy consumption optimization objectives. The calculation is performed in real time through the current / voltage sampling circuit built into the power module (power = voltage × current).

[0063] The value is , representing the real-time temperature (°C) of the lamp chip, used for thermal stability constraints. With lifetime loss function Temperature data is collected directly through temperature sensor units integrated into the lighting fixture nodes (such as NTC thermistors and digital temperature chips).

[0064] The health status reflects the health level of the lighting node hardware (normalized value, range 0-1), where 1 indicates good condition and 0 indicates failure. It is used to determine whether to participate in collaborative computing or trigger emergency mechanisms. It is calculated based on historical data such as temperature drift, power attenuation, and light intensity deviation through a health assessment algorithm (such as a remaining lifetime prediction model).

[0065] The communication status reflects the quality of the communication link between the lamp node and its neighbors (normalized value, range 0-1). 1 indicates a stable link (low packet loss rate, high signal strength), and 0 indicates a failed link. It is used for the robustness and fault tolerance control logic of the edge optimization layer. It is calculated by weighting parameters such as the signal strength (RSSI), packet loss rate, and retransmission count of the communication module through a link quality assessment algorithm.

[0066] The set of neighboring light fixture nodes, i.e., the set of light fixture nodes. A local lighting unit consisting of adjacent luminaires that communicate directly (usually containing 3-5 luminaires, the exact number depending on the communication radius and deployment density).

[0067] The control logic for integrated lighting nodes is based on the state vectors broadcast by neighboring lighting nodes. The light intensity information meets the following light intensity update rules:

[0068] ,

[0069] in: The current illumination intensity represents the luminaire node. At any moment The output light intensity is the time-series value of the control variable;

[0070] The light intensity at the next moment represents the luminaire node. In the next cycle The light intensity is the result of the control update and is used for the next PWM dimming output.

[0071] The update step size represents the update cycle (200ms) of the luminaire node's illumination intensity, controlling the refresh rate;

[0072] The negotiation coefficient is a weighting parameter; 0 < <1, determines the fusion ratio of self-illuminance and neighborhood average illuminance. A larger value results in a faster response time but may cause fluctuations. A smaller value results in a smoother response but also indicates a lag.

[0073] The number of neighboring light fixture nodes reflects the size of the neighborhood.

[0074] The illuminance of neighboring luminaire nodes, i.e., the illuminance of each luminaire node within the neighborhood. At any moment The illumination intensity value is a reference benchmark for the illumination of the neighborhood and is used for spatial smoothness calculation.

[0075] The fast response layer performs the following operations to achieve millisecond-level lighting response, ensuring continuous and uninterrupted lighting along the path of people and vehicles:

[0076] (1) The fast response layer calculates the path light track connectivity based on the current illumination intensity data obtained by updating the lamp nodes according to the rules. And set constraints to ensure the integrity of path lighting:

[0077] ,

[0078] in: The path light track connectivity describes a specific vehicle or pedestrian path. The relative degree of light intensity guarantee at the weakest point; the larger the value, the better the path meets the minimum lighting guarantee.

[0079] It is a set of paths or a set of luminaire node indices for a single path (i.e., a set of the numbers of all luminaire nodes on the path).

[0080] This is the index of the lamp node on the path (an index variable used to iterate through each lamp node on the path);

[0081] For path The first The identifier of each lighting fixture node (the specific node number, corresponding to the first node in the path). (The unique identifier for each lamp)

[0082] For lighting fixture nodes The actual light intensity of the current output (unit: lux); The minimum safe light intensity threshold (the minimum light intensity standard to meet the safety of driving or pedestrians, in lux, such as 10-30 lux in road lighting);

[0083] (2) Path optical track connectivity based on each path Constructing overall system continuity indicators The overall continuity index Provide a basis for judgment in the emergency response mechanism of the rapid response layer:

[0084] ,

[0085] in: Indicates the vehicle travel paths contained within a certain lighting area. The corresponding path optical track connectivity; For the first The weight of a path reflects the traffic flow density or frequency of use of that path;

[0086] (i) In certain extreme security-priority scenarios, the system adopts a minimum guarantee strategy, in which case the formula is used. This means that if any critical path fails to meet the standard, the system must take remedial measures immediately.

[0087] (ii) In general scenarios, a formula with normalized illuminance is used to compare the illuminance of each luminaire node with the minimum illuminance and sum and average them to comprehensively evaluate the uniformity and continuity of lighting throughout the entire area:

[0088] ,

[0089] in: This represents the total number of lighting fixture nodes counted within the region. This represents the current actual illumination intensity of each luminaire node, reflecting the actual illumination intensity received by that luminaire node. It is the minimum illuminance, which is the minimum illuminance among all luminaire nodes in the entire area;

[0090] (3) The fast response layer combines the illuminance data in the state vector of the luminaire node with the overall continuity index of the system, and triggers an emergency mechanism when it detects that the illuminance of a neighboring luminaire node does not conform to the preset change:

[0091] ,

[0092] in: The illuminance of a neighboring luminaire node represents the illuminance of a monitored neighboring luminaire node. Real-time light intensity or light intensity output; To ensure a safe light intensity threshold and guarantee that local areas reach a safe lighting level, this emergency mechanism must be completed within ≤200ms to achieve self-restoring lighting continuity.

[0093] (ii) Edge optimization layer, which runs on the partition control nodes of each floor of the underground parking lot, is used to construct a comprehensive optimization objective function with the lamp node status data and continuity index uploaded by the fast response layer as input. After solving the objective function through the sub-modulus optimization algorithm and the convex approximation method, it outputs the optimal illuminance and power allocation parameters of each lamp node, completes the precise control of lighting at the partition scale, and reports the partition operation data to the global learning layer.

[0094] Comprehensive optimization objective function for edge optimization layer construction as follows:

[0095] ,

[0096] ,

[0097] ,

[0098] ,

[0099] ,

[0100] ,

[0101] ,

[0102] in: , , , , Indicates the weighting coefficient;

[0103] Indicates system energy consumption. The t represents the real-time power of the i-th lamp node, and the t represents the running time.

[0104] This represents the average lifespan loss rate of each lighting fixture node. The number of nodes; Let be the lifespan loss ratio of the luminaire node; For real-time temperature With real-time power Instantaneous lifespan loss; The maximum reference loss of the luminaire node under the design limit conditions is used to normalize the life loss so that the life loss ratio is [0,1]. The coefficients c, d, and e are used to adjust the contribution and influence of temperature and power. c, d, and e are all greater than 0, and c+d+e=1. Indicates the selected reference temperature. Indicates the selected reference power;

[0105] Indicates visual comfort, Indicates the current light intensity. To update the step size, Indicates the light intensity at the previous moment;

[0106] Computational complexity is used to quantify the overall computational overhead of the algorithm model in the system. Model inference time reflects the time it takes for the algorithm model to process input data and output results, which is directly related to the real-time response capability of the system. The shorter the time, the higher the time efficiency of the algorithm. Parameter scale refers to the total number of parameters to be optimized in the algorithm model (such as the weights and biases of the neural network), which directly determines the size of the storage space occupied by the model. The smaller the scale, the lower the demand for hardware storage resources.

[0107] Indicates thermal stability, This represents the rate of change of the temperature of the i-th lamp node over time.

[0108] Further in thermal stability Thermal dynamic constraints are introduced during the calculation process:

[0109] ,

[0110] in: The power-temperature rise coefficient is a proportionality factor that characterizes the temperature rise caused by power input. This represents the real-time power of the i-th lamp node; The heat dissipation attenuation coefficient characterizes the degree of influence of temperature difference on the heat dissipation rate. This represents the real-time temperature of the i-th lighting node; The ambient temperature is an external boundary condition and serves as a reference for heat dissipation at the luminaire nodes.

[0111] (iii) Global learning layer, deployed on cloud server or edge server, is used to achieve dynamic optimization of multi-objective weights and lightweight strategy based on the partition operation data reported by the edge optimization layer through the "periodic data aggregation-policy iteration-lightweight distribution" mode, combined with distributed reinforcement learning and federated learning technology, and then distributed to each partition control node to support the long-term adaptive optimization and stable operation of the lighting control system.

[0112] The global learning layer performs the following operations:

[0113] (1) The global learning layer periodically receives partition operation data reported by each edge optimization layer and optimizes the multi-objective weight vector through the distributed reinforcement learning (DDPG) algorithm. And with the help of a customized reward function Output the dynamically adjusted weight parameters:

[0114] ,

[0115] ,

[0116] ,

[0117] in: For multi-objective weight vectors; Energy consumption weighting coefficient; This is a weighting factor for the lifespan of the lighting fixture; This is a comfort-weighted coefficient. To calculate the complexity weighting coefficients; This is the thermal stability weighting coefficient;

[0118] This is a safety reward coefficient used to amplify safety indicators. Weight in the total reward; For continuous reward coefficients, amplify continuous indicators. Its influence; This is an energy consumption penalty factor used to reduce system energy consumption. It is considered a negative item in the reward; The thermal penalty coefficient is used to determine the thermal stability. It is considered a negative item in the reward; To calculate the complexity penalty coefficient, the computational complexity will be... It is considered a negative item in the reward;

[0119] (2) Using the weight optimization results as input, the global learning layer uses federated learning technology to reuse the optimization experience of mature scenarios through policy transfer, and then compresses the model volume through model distillation to finally generate a lightweight optimization policy;

[0120] (3) The lightweight optimization strategy is distributed to each partition control node, which not only supports the partition control node to independently run the solution process of the comprehensive optimization objective function in real time based on the strategy (by sub-modular optimization algorithm and convex approximation method), but also adapts to the addition of new partition control nodes or lighting nodes, and the replacement of faulty partition control nodes or lighting nodes, so as to ensure the long-term stable operation and continuous adaptive optimization of the lighting control system.

[0121] In summary, the intelligent lighting control system for underground parking lots based on multi-objective real-time optimization of this invention adopts a three-layer hierarchical architecture of "fast response layer - edge optimization layer - global learning layer," breaking through the limitations of traditional centralized control or single-objective optimization. It dynamically balances lighting safety, continuity, energy consumption, luminaire lifespan, visual comfort, edge computing complexity, and LED thermal stability. Under the premise of strictly ensuring lighting safety and continuity constraints, it can reduce energy consumption by 20%-30%, extend LED lifespan by 20%, and reduce temperature fluctuations by 35%, while supporting real-time operation of edge nodes. Through collaborative innovation of architecture, algorithm, and strategy, this system has significant advantages over existing solutions in terms of safety, comfort, energy saving, and maintainability.

[0122] The above specific examples illustrate the principles and implementation methods of the present invention in detail. These embodiments are merely for the purpose of helping to understand the core technical content of the present invention. Based on the above specific embodiments of the present invention, any improvements and modifications made to the present invention by those skilled in the art without departing from the principles of the present invention should fall within the patent protection scope of the present invention.

Claims

1. A multi-objective real-time optimization based intelligent lighting control system for underground parking, characterized in that, It comprises: A fast response layer deployed in each lamp node for real-time collection of state data of each lamp node, achieving millisecond-level light response through neighborhood negotiation and event triggering mechanism to ensure continuous light along the path of human and vehicle travel without fault, and uploading the collected lamp node state data and continuity indicators to the edge optimization layer; The edge optimization layer runs in the partition control node of each floor of the underground parking lot, and uses the lamp node state data and continuity indicators uploaded by the fast response layer as input to construct a comprehensive optimization objective function, which is solved by a sub-module optimization algorithm and a convex approximation method, and then outputs the optimal light intensity and power distribution parameters of each lamp node, completes the precise regulation and control of lighting at the partition scale, and uploads the partition operation data to the global learning layer; The global learning layer is deployed in a cloud server or an edge server, and uses the partition operation data reported by the edge optimization layer to achieve multi-objective weight dynamic optimization and strategy lightweight through a "periodic data aggregation-strategy iteration-lightweight distribution" mode, combined with distributed reinforcement learning and federated learning technology, and then issues the lightweight optimization strategy to each partition control node to support long-term adaptive optimization and stable operation of the lighting control system; The fast response layer performs the following operations to achieve millisecond-level light response to ensure continuous light along the path of human and vehicle travel without fault: (1) Fast response layer based on the current light intensity data obtained by the lamp node according to the updated rules, further calculate the path light track connectivity And set constraints to ensure the integrity of path lighting: ， wherein: is the path track connectivity, describing a specific path for a vehicle or pedestrian is the relative level of security of the light intensity at the weakest point on the path; is the set of light fixture node indices for the path set or single path; is the index of the light fixture node on the path; is the path is the first light fixture node identifier on the path; is the light fixture node is the actual light intensity of the current output of the light fixture node; is the minimum safe light intensity threshold; (2) Path light track connectivity degree based on each path , construct system overall continuity index , the overall continuity index Provide judgment basis for emergency mechanism of fast response layer: ， in: Indicates the vehicle travel paths contained within a certain lighting area. The corresponding path optical track connectivity; For the first The weight of a path reflects the traffic flow density or frequency of use of that path; (i) In certain extreme safety-first scenarios, the system adopts a minimal safeguard policy, in which case the formula is used, i.e. the system needs to take remedial action immediately if any critical path is not up to standard; (ii) In a general scenario, the formula in the form of light intensity normalization is used to compare and average the light intensity of each lamp node with the minimum light intensity, to comprehensively evaluate the uniformity and continuity of lighting in the entire area: ， Wherein: is the total number of the light nodes in the region; represents the current actual light intensity of each light node, reflecting the actual received light intensity of the light node; is the minimum light intensity, which is the minimum value of the light intensity of all light nodes in the entire region; (3) The fast response layer combines the light intensity data in the lamp node state vector and the system overall continuity indicator, and triggers the emergency mechanism when it detects that the light intensity of the neighbor lamp node does not meet the preset change: ， wherein: is the neighbor luminaire node light intensity, representing the real-time light intensity or light intensity output of a certain neighbor luminaire node being monitored; is the safety light intensity threshold, ensuring that the local area reaches a safe lighting level; this emergency mechanism needs to be completed within a preset period to achieve self-recovery lighting continuity.​ 2. The multi-objective real-time optimization based intelligent lighting control system for underground parking according to claim 1, wherein, The fast response layer is deployed in each independent lamp node, and each lamp node integrates hardware units and control logic; wherein: the hardware units specifically include a micro control unit, a light intensity sensor unit, a temperature sensor unit, a PWM dimming driver unit and a short-range communication unit; the control logic realizes millisecond-level light response.

3. The multi-objective real-time optimization based intelligent lighting control system for underground parking according to claim 2, wherein, Each luminaire node Every 200 ms to neighbor luminaire nodes Broadcast state vector : ， wherein: is the current light intensity of the luminaire, representing the actual light intensity output by the luminaire node at present; is the power, representing the real-time power consumption of the luminaire node at present; is the temperature, representing the real-time temperature of the luminaire chip; is the health status, reflecting the health degree of the hardware of the luminaire node; is the communication status, reflecting the quality of the communication link between the luminaire node and its neighbors; is the set of neighbor luminaire nodes, i.e. the local lighting unit composed of the neighboring luminaires directly communicating with the luminaire node .

4. The multi-objective real-time optimization based intelligent lighting control system for underground parking according to claim 3, wherein, The control logic integrated in the lamp node is based on the light intensity information in the state vector broadcast by the neighbor lamp node, and satisfies the following light intensity update rule: ， wherein: represents the output illumination intensity of the luminaire node at time ; represents the illumination intensity of the luminaire node at the next cycle ; is the update step, representing the luminaire node illumination intensity update cycle; is the negotiation coefficient, 0 <1, determining the fusion proportion of the self-illumination intensity and the average illumination intensity of the neighborhood; is the number of neighbor luminaire nodes, reflecting the scale of the neighborhood range; is the illumination intensity of the neighbor luminaire node, i.e., the illumination intensity value of each luminaire node in the neighborhood at time, which is the reference benchmark of the neighborhood illumination.

5. The multi-objective real-time optimization based intelligent lighting control system for underground parking according to claim 4, wherein, The integrated optimization objective function for the edge-optimized layer construction is as follows: As follows: ， ， ， ， ， ， ， in: , , , , Indicates the weighting coefficient; Indicates system energy consumption. The t represents the real-time power of the i-th lamp node, and the t represents the running time. This represents the average lifespan loss rate of each lighting fixture node. For the number of nodes, For the first The lifespan loss rate of each lighting fixture node. For real-time temperature With real-time power Instantaneous lifespan loss, This represents the maximum reference loss of the luminaire node under design limits, used to normalize the lifespan loss so that the lifespan loss ratio is within [0,1]. Coefficients c, d, and e are used to adjust the contribution and influence of temperature and power, where c, d, and e are all greater than 0, and c + d + e = 1. Indicates the selected reference temperature. Indicates the selected reference power; Indicates visual comfort, Indicates the current light intensity. To update the step size, Indicates the light intensity at the previous moment; The computational complexity is used to quantify the overall computational cost of the algorithm model in the system. The model inference time reflects the time it takes for the algorithm model to process the input data and output the results, which is directly related to the real-time response capability of the system. The parameter scale refers to the total number of parameters to be optimized in the algorithm model, which directly determines the amount of storage space occupied by the model. Indicates thermal stability, This represents the rate of change of the temperature of the i-th lamp node over time.

6. The multi-objective real-time optimization based intelligent lighting control system for underground parking according to claim 5, wherein, In thermal stability Introducing thermal dynamic constraints in the calculation process: ， wherein: is the power-temperature coefficient, representing the proportional coefficient of temperature rise caused by power input; represents the real-time power of the i-th luminaire node; is the heat dissipation attenuation coefficient, representing the degree of influence of temperature difference on heat dissipation rate; represents the real-time temperature of the i-th luminaire node; is the ambient temperature, which belongs to the external boundary condition and is the reference benchmark for heat dissipation of the luminaire node.

7. The multi-objective real-time optimization based intelligent lighting control system for underground parking according to claim 5, wherein, The global learning layer specifically performs the following operations: (1) The global learning layer periodically receives the partition operation data reported by each edge optimization layer, optimizes the multi-objective weight vector through a distributed reinforcement learning algorithm, and outputs the dynamically adjusted weight parameters with the help of a customized reward function: ​ ， ， ， wherein: is a multi-objective weight vector; is an energy consumption weight coefficient; is a luminaire lifetime weight coefficient; is a comfort weight coefficient; is a computational complexity weight coefficient; is a thermal stability weight coefficient; This is the safety reward coefficient, used to amplify the weight of safety indicators in the total reward; For continuous reward coefficients, amplify continuous indicators. Its influence; This is an energy consumption penalty factor used to reduce system energy consumption. It is considered a negative item in the reward; The thermal penalty coefficient is used to determine the thermal stability. It is considered a negative item in the reward; To calculate the complexity penalty coefficient, the computational complexity will be... It is considered a negative item in the reward; (2) With the weight optimization result as input, the global learning layer uses the federated learning technology to reuse the optimization experience of mature scenarios through strategy migration, and then compresses the model volume through model distillation, and finally generates a lightweight optimization strategy; (3) The lightweight optimization strategy is issued to each partition control node, which not only supports the partition control node to independently run the solving process of the comprehensive optimization objective function based on the strategy in real time, but also adapts to the addition of new partition control nodes or lamp nodes, the replacement of faulty partition control nodes or lamp nodes, and ensures the long-term stable operation and continuous adaptive optimization of the lighting control system.

Images