LED intelligent illumination management system based on cloud platform

Abstract

The invention discloses an LED intelligent lighting management system based on a cloud platform, and relates to the technical field of intelligent control, a multi-source data collection module is arranged to collect lighting equipment data of a multi-protocol terminal through an edge computing node, and an over-lighting judgment module is arranged to judge the lighting equipment data according to an energy consumption data set in the equipment data set. Judging whether a local over-illumination phenomenon exists in a current illumination area or not, generating a judgment result, setting a control parameter generation module, if the over-illumination condition exists, generating an optimized illumination control parameter set by adopting a reinforcement learning algorithm, and if the over-illumination condition does not exist, generating the optimized illumination control parameter set by adopting a multi-target genetic algorithm; a dynamic adjustment module is arranged, the optimized illumination control parameter set is issued to the corresponding illumination equipment, and the light environment of the illumination equipment is dynamically adjusted; according to the invention, the energy consumption of illumination is reduced, and the comfort during illumination is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a cloud platform-based LED intelligent lighting management system. Background Art

[0002] Overlighting is a common yet often overlooked issue in intelligent LED lighting management. Overlighting refers to lighting that exceeds actual brightness requirements, resulting in energy waste and visual discomfort. Overlighting exists to varying degrees in office environments, which not only wastes energy but can also cause glare, visual fatigue, and other issues, impacting user productivity and health. When overlighting occurs, the system's primary goal is to reduce energy consumption and eliminate localized light imbalances. Under normal lighting conditions, the system must optimize energy consumption while maintaining comfort.

[0003] However, most current systems lack effective mechanisms for identifying over-lighting, relying solely on preset lighting schemes or simple time controls, and are unable to intelligently adjust to actual lighting conditions. Furthermore, existing systems typically employ a single control algorithm, failing to provide differentiated optimization strategies for different lighting conditions. This results in either excessive energy consumption or insufficient comfort in certain scenarios.

[0004] To this end, the present invention proposes a cloud platform-based LED intelligent lighting management system. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a cloud platform-based LED intelligent lighting management system that reduces lighting energy consumption and improves lighting comfort.

[0006] To achieve the above objectives, a cloud-based LED intelligent lighting management system is proposed, which includes a multi-source data collection module, an over-lighting judgment module, a control parameter generation module, and a dynamic adjustment module. The modules are electrically connected to each other.

[0007] The multi-source data collection module collects lighting device data from multi-protocol terminals through edge computing nodes. The lighting device data includes brightness data sets, color temperature data sets, and energy consumption data sets, which constitute the device data sets of the multi-protocol terminals. The device data sets are then sent to the lighting judgment module and the control parameter generation module.

[0008] an over-lighting judgment module, which judges whether there is a local over-lighting phenomenon in the current lighting area based on the energy consumption data set in the device data set, generates a judgment result, and sends the judgment result to the control parameter generation module;

[0009] a control parameter generation module, determining whether the judgment result indicates over-lighting; if so, dynamically optimizing the brightness and color temperature data in the device data set using a reinforcement learning algorithm to generate an optimized lighting control parameter set; if not, performing dual-objective optimization on the energy consumption and comfort data in the device data set using a multi-objective genetic algorithm to generate an optimized lighting control parameter set; and sending the optimized lighting control parameter set to the dynamic adjustment module;

[0010] The dynamic adjustment module sends the optimized lighting control parameter set to the corresponding lighting equipment to dynamically adjust the light environment of the lighting equipment.

[0011] The brightness dataset is collected as follows:

[0012] The edge computing node automatically identifies the type of communication protocol used by the connected lighting device through the protocol identification interface;

[0013] According to the identified protocol type, the edge computing node dynamically loads the corresponding protocol driver module to establish a communication connection with the lighting device; after the communication connection is established, the edge computing node sends a brightness data query instruction to the lighting device;

[0014] After receiving the brightness data query instruction, the lighting device returns the current brightness status data to the edge computing node.

[0015] Add standardized information such as lighting device ID, location information, and timestamp to each piece of brightness status data to form brightness data;

[0016] The time series of all brightness data sorted in chronological order is taken as the brightness dataset;

[0017] The color temperature dataset is collected as follows:

[0018] The edge computing node sends a color temperature data query instruction to the LED lighting device based on the established multi-protocol communication connection;

[0019] After receiving the color temperature data query command, the lighting device returns the current color temperature status data to the edge computing node;

[0020] Add the corresponding lighting device ID, location information, timestamp and other standardized information to the color temperature status data to form color temperature data;

[0021] The time series of all color temperature data sorted in chronological order is used as the color temperature dataset;

[0022] The energy consumption dataset is collected as follows:

[0023] Different collection strategies are used to collect energy consumption data based on the lighting equipment's energy consumption data acquisition capabilities. For lighting equipment with direct energy consumption measurement capabilities, the edge computing node directly queries energy consumption data including the device's real-time power, cumulative energy consumption, and energy efficiency indicators through the corresponding protocol. For lighting equipment without direct energy consumption measurement capabilities, the edge computing node estimates its energy consumption through an indirect method. The indirect method uses an energy consumption estimation model to calculate the theoretical energy consumption value based on the device's rated power, current brightness level, and operating time.

[0024] The time series of all energy consumption data sorted in chronological order is taken as the energy consumption dataset;

[0025] The device data set is composed as follows:

[0026] The brightness dataset, color temperature dataset, and energy consumption dataset are classified and integrated according to the device ID of the lighting device to generate a device dataset; the device dataset is stored in the form of a structured data table, which adopts a relational data structure, with each row representing a complete data record of a device corresponding to a time point, and each column representing a data field.

[0027] The determining whether there is a local over-lighting phenomenon in the current lighting area based on the energy consumption data set in the device data set includes the following steps:

[0028] Step 21: extracting the energy consumption dataset from the device dataset of the multi-protocol terminal as energy consumption distribution data of the current lighting area;

[0029] Step 22: Calculate the difference between the average energy consumption and the peak energy consumption of the current lighting area based on the energy consumption distribution data;

[0030] Step 23: Compare the difference value with a preset energy consumption threshold to determine whether there is a local over-lighting phenomenon in the current lighting area;

[0031] Step 24: If the difference value exceeds the energy consumption threshold, outputting a judgment result that a local over-lighting phenomenon exists; otherwise, outputting a judgment result that no local over-lighting phenomenon exists;

[0032] The step of dynamically optimizing the brightness and color temperature data in the device data set using a reinforcement learning algorithm to generate an optimized lighting control parameter set comprises the following steps:

[0033] Step 311: extracting a brightness dataset and a color temperature dataset from the device dataset as input data for a reinforcement learning algorithm;

[0034] Step 312: Pre-construct a deep Q-network-based reinforcement learning environment model suitable for parameter control of lighting equipment, defining a state space, an action space, and a reward function;

[0035] Step 313: Collecting device data sets and optimization parameters in the reward function within a preset time period as historical lighting data, and pre-training the deep Q network based on the historical lighting data to form an initial policy model;

[0036] Step 314: Deploy the initial policy model on the edge computing node, perform online reinforcement learning and policy update, and generate online learning results;

[0037] Step 315: Dynamically generate brightness and color temperature control parameters of the lighting device based on the online learning results;

[0038] The deep Q network uses a multi-layer perceptron structure, with the input layer corresponding to the state vector and the output layer corresponding to the Q-value estimates of each possible action. The hidden layer of the network uses the ReLU activation function.

[0039] The method of using a multi-objective genetic algorithm to perform dual-objective optimization on the energy consumption and comfort data in the device data set to generate an optimized lighting control parameter set includes the following steps:

[0040] Step 321: extracting energy consumption data and comfort-related data from the equipment data set to construct an input data set for a dual-objective optimization model;

[0041] Step 322: Based on the input data set, define an energy consumption objective function and a comfort objective function, and construct a dual-objective optimization model;

[0042] Step 323: Initialize the population of the multi-objective genetic algorithm for the dual-objective optimization model, set the algorithm parameters, and encode the individuals in the population;

[0043] Step 324: Iteratively perform selection, crossover, and mutation operations on the initialized population to generate a new population, and perform non-dominated sorting and crowding calculation;

[0044] Step 325: Evaluate the fitness of the individuals in the population and determine whether the algorithm has reached the termination condition. If so, output the Pareto optimal solution set; otherwise, return to step 324 to continue iteration.

[0045] Step 326: Select a lighting control parameter set solution from the Pareto optimal solution set according to the current system state and user preference;

[0046] Step 327: Convert the lighting control parameter set solution into standardized lighting control instructions to generate an optimized lighting control parameter set.

[0047] The method of dynamically adjusting the light environment of the lighting device is:

[0048] The edge computing node receives the optimized lighting control parameter set from the control parameter generation module, converts the lighting control parameter set into the corresponding protocol instructions of the lighting device according to the communication protocol type of the lighting device, and sends the protocol instructions to the corresponding lighting device through the multi-protocol communication interface. The S-type function is used to calculate the gradient curve of brightness and color temperature from the current value to the target brightness value and target color temperature value to dynamically control the lighting device.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] First, edge computing nodes collect lighting device data from multi-protocol terminals, including brightness, color temperature, and energy consumption information. Based on the energy consumption distribution data, the system then determines whether there is local overlighting. If overlighting is present, the system uses a reinforcement learning algorithm to dynamically optimize the brightness and color temperature data. This algorithm learns optimal strategies through interaction with the environment, adapting to complex and changing lighting needs. Under normal lighting conditions, the system uses a multi-objective genetic algorithm to optimize energy consumption and comfort data. By simulating natural evolution, this algorithm can simultaneously optimize multiple conflicting objectives. By identifying overlit areas and implementing differentiated optimization strategies, the system provides a suitable lighting environment for each area, meeting visual comfort requirements while reducing energy consumption. This reduces lighting energy consumption and improves comfort during lighting. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a module connection diagram of the cloud platform-based LED intelligent lighting management system;

[0052] Figure 2 This is an execution flow chart of the over-lighting judgment module;

[0053] Figure 3 The flowchart for generating the lighting control parameter set in the absence of over-lighting. DETAILED DESCRIPTION

[0054] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] In the lighting systems of large public spaces such as smart office buildings, commercial centers, medical institutions, and educational venues, traditional lighting systems often adopt a unified lighting solution, which cannot meet the differentiated lighting needs of different work areas and at different times, resulting in energy waste and poor user experience.

[0056] like Figure 1 As shown, a cloud platform-based LED intelligent lighting management system includes a multi-source data collection module, an over-lighting judgment module, a control parameter generation module, and a dynamic adjustment module; wherein each module is electrically connected;

[0057] The multi-source data collection module collects lighting device data from multi-protocol terminals through edge computing nodes. The lighting device data includes brightness data sets, color temperature data sets, and energy consumption data sets, which constitute the device data sets of the multi-protocol terminals. The device data sets are then sent to the lighting judgment module and the control parameter generation module.

[0058] an over-lighting judgment module, which judges whether there is a local over-lighting phenomenon in the current lighting area based on the energy consumption data set in the device data set, generates a judgment result, and sends the judgment result to the control parameter generation module;

[0059] Control parameter generation module, such as Figure 2 As shown, it is determined whether the judgment result contains an over-lighting phenomenon. If so, a reinforcement learning algorithm is used to dynamically optimize the brightness and color temperature data in the device data set to generate an optimized lighting control parameter set. If not, a multi-objective genetic algorithm is used to perform dual-objective optimization on the energy consumption and comfort data in the device data set to generate an optimized lighting control parameter set. The optimized lighting control parameter set is sent to a dynamic adjustment module.

[0060] The dynamic adjustment module sends the optimized lighting control parameter set to the corresponding lighting equipment to dynamically adjust the light environment of the lighting equipment.

[0061] First of all, it should be noted that the edge computing nodes generally refer to several micro servers with computing capabilities that are closest to the lighting equipment. The edge computing nodes are generally equipped with multiple communication protocol interfaces, including but not limited to DALI (Digital Addressable Lighting Interface), ZigBee, Bluetooth Mesh, WiFi, KNX and other common smart lighting protocols;

[0062] In an embodiment of the present invention, the brightness dataset is collected in the following manner:

[0063] The edge computing node automatically identifies the communication protocol used by connected lighting devices through a protocol identification interface. Specifically, the protocol identification interface sends a standardized protocol detection signal and then uses a pattern matching algorithm to determine the device's protocol type based on the characteristic patterns of the returned signal. This pattern matching algorithm, based on a pre-established protocol signature library, calculates the similarity between signal signatures and protocol templates in the signature library and selects the protocol type with the highest similarity as the identification result.

[0064] Based on the identified protocol type, the edge computing node dynamically loads the corresponding protocol driver module and establishes a communication connection with the lighting device. Once the communication connection is established, the edge computing node sends a brightness data query command to the lighting device. These brightness data query commands are encapsulated according to different protocols, but their core content is to request the current brightness status information of the device.

[0065] Subsequently, upon receiving the query command, the lighting device returns the current brightness status data to the edge computing node. The brightness status data can generally be expressed in percentage form, for example, 0% represents completely off and 100% represents maximum brightness. It can also be expressed in specific lumens (lm), depending on the functional characteristics of the device, which is not limited in this invention.

[0066] Then, standardized information such as lighting device ID, location information, and timestamp is added to each piece of brightness status data to form brightness data;

[0067] The time series of all brightness data sorted in chronological order is taken as the brightness dataset;

[0068] Furthermore, the color temperature dataset is collected as follows:

[0069] The edge computing node sends a color temperature data query instruction to the LED lighting device based on the established multi-protocol communication connection.

[0070] After receiving the query command, the lighting device returns the current color temperature status data to the edge computing node. The color temperature status data is usually expressed in Kelvin (K), indicating the color temperature value of the light source. In the field of smart lighting, the color temperature range is generally 2700K (warm white light) to 6500K (cold white light).

[0071] Then, the corresponding lighting device ID, location information, timestamp and other standardized information are added to the color temperature status data to form the color temperature data;

[0072] The time series of all color temperature data sorted in chronological order is used as the color temperature dataset;

[0073] In a further embodiment of the present invention, the color temperature data can also be correlated with lighting environment parameters for analysis. Specifically, the lighting environment parameters include but are not limited to outdoor natural light illumination, indoor activity type, and time factors. The purpose of the correlation analysis is to evaluate the matching degree between the current color temperature setting and environmental requirements, and to provide a decision basis for subsequent optimization control. Specifically, the correlation analysis uses a weighted scoring model, assigning different weights according to the importance of different environmental factors, and calculating the environmental adaptability score of the color temperature setting. The mathematical expression of the weighted scoring model is: ;

[0074] Where S represents the environmental adaptability score of the color temperature, Wi represents the weight of the i-th environmental factor, and Fi represents the fitness function value of the i-th environmental factor. The fitness function value is defined based on ergonomic research and lighting standards, and quantifies the relationship between environmental factors and ideal color temperature as a fitness value between 0 and 1. Finally, the environmental adaptability score is included in the color temperature data;

[0075] Furthermore, the energy consumption data set is collected in the following manner:

[0076] According to the energy consumption data acquisition capabilities of lighting equipment, different collection strategies are adopted to collect energy consumption data of lighting equipment;

[0077] The time series of all energy consumption data sorted in chronological order is taken as the energy consumption dataset;

[0078] Specifically, for advanced smart lighting devices with direct energy consumption measurement capabilities, the edge computing node directly queries the energy consumption data including the device's real-time power, cumulative energy consumption and energy efficiency indicators through the corresponding protocol; the energy efficiency indicators include but are not limited to: energy efficiency ratio (lm / W), energy consumption per unit area (kWh / m²), energy consumption per unit illuminance (kWh / lx), etc.

[0079] For common lighting devices that do not have direct energy consumption measurement capabilities, edge computing nodes use indirect methods to estimate their energy consumption. This indirect method uses an energy consumption estimation model to calculate theoretical energy consumption values ​​based on the device's rated power, current brightness level, and operating time. The basic form of the energy consumption estimation model is: E = P × B × T × C;

[0080] Where E represents the estimated energy consumption (kWh), P represents the rated power of the device (kW), B represents the brightness level coefficient (the ratio of power to brightness percentage, usually a nonlinear relationship), T represents the operating time (hours), and C represents the correction factor (accounting for factors such as device aging and ambient temperature). The brightness level coefficient B is obtained by querying a pre-established brightness-power mapping table. This mapping table is fitted based on a large amount of measured data and accurately reflects the actual power consumption at different brightness levels.

[0081] Secondly, for lighting areas that support smart meters or energy consumption sensors, the edge computing node can also collect regional energy consumption monitoring data through the corresponding communication interface. This regional energy consumption monitoring data provides a more accurate overall energy consumption situation and can be used to verify and calibrate the accumulated value of device-level energy consumption data.

[0082] Merging the brightness data set, color temperature data set, and energy consumption data set into a structured data table, and outputting a device data set for a multi-protocol terminal;

[0083] Specifically, the device data sets are merged as follows:

[0084] The brightness dataset, color temperature dataset, and energy consumption dataset are classified and integrated according to the device ID of the lighting device to generate a device dataset; the device dataset is stored in the form of a structured data table, which adopts a relational data structure, with each row representing a complete data record of a device corresponding to a time point, and each column representing a data field.

[0085] In the embodiment of the present invention, the data fields include at least three categories: (1) identification fields, such as device ID, timestamp, location information, etc.; (2) basic data fields, such as original or simply processed data such as brightness, color temperature, power, energy consumption, etc.; (3) analysis feature fields, such as intermediate analysis results such as energy efficiency indicators;

[0086] Furthermore, judging whether there is a local over-lighting phenomenon in the current lighting area based on the energy consumption data set in the device data set includes the following steps:

[0087] Step 21: extracting the energy consumption dataset from the device dataset of the multi-protocol terminal as energy consumption distribution data of the current lighting area;

[0088] Specifically, the energy consumption distribution data is extracted as follows:

[0089] First, energy consumption data of a preselected time period is selected from the device data set according to the device ID and timestamp, and then grouped according to spatial location information of the lighting devices, such as area numbers or coordinates.

[0090] Each group of energy consumption data forms a subset, representing the energy consumption distribution data of the area. For example, in the lighting system of an office area, the energy consumption data of the lighting equipment at each workstation will be extracted and classified into the corresponding workstation area subset.

[0091] Step 22: Calculate the difference between the average energy consumption and the peak energy consumption of the current lighting area based on the energy consumption distribution data;

[0092] Specifically, the mean energy consumption refers to the average energy consumption of all lighting devices in the current lighting area, reflecting the overall energy consumption level of the area. The peak energy consumption is the highest energy consumption of a single lighting device in the area, which may be caused by local overlighting. The difference value is used to quantify the degree of deviation between the mean energy consumption and the peak energy consumption, and the calculation formula is: ;

[0093] in, is the difference value, is the peak energy consumption, is the mean energy consumption;

[0094] The calculation process begins with a statistical analysis of each regional subset of the energy consumption distribution data. For each region, the mean and peak energy consumption of all lighting devices within it are calculated. The mean energy consumption is calculated using the arithmetic mean, while the peak energy consumption is calculated by traversing the energy consumption records of all devices in the region and selecting the maximum value. For example, in a conference room lighting system, if the mean energy consumption in a certain region is 50W and the peak energy consumption of a particular device reaches 80W, the difference is 30W.

[0095] In a further embodiment of the present invention, since the energy consumption of lighting equipment may fluctuate over time, a sliding window technique is employed to update the calculated results of the mean and peak energy consumption at fixed intervals, such as every minute, to ensure the real-time performance of the difference values. Furthermore, to reduce the impact of instantaneous fluctuations on the difference values, the energy consumption data can be smoothed, such as by using a moving average method.

[0096] Step 23: Compare the difference value with a preset energy consumption threshold to determine whether there is a local over-lighting phenomenon in the current lighting area;

[0097] Specifically, the preset energy consumption threshold is a critical value determined through experiments or historical data analysis, which is used to distinguish normal energy consumption from abnormal energy consumption. The local over-lighting phenomenon refers to the phenomenon that the energy consumption of a local area is significantly higher than the average level due to improper setting or failure of the lighting equipment. Therefore, the judgment logic is: if the difference value exceeds the energy consumption threshold, it is considered that there is a local over-lighting phenomenon; otherwise, it is considered that there is no local over-lighting phenomenon; the setting of the energy consumption threshold is based on the actual operating data of the lighting system. For example, through statistical analysis of historical energy consumption data, the upper limit of normal energy consumption fluctuations can be determined. Assuming that the energy consumption difference value of a certain office area usually does not exceed 20W, the energy consumption threshold is set to 20W. If the current difference value is 30W, it is judged that there is a local over-lighting phenomenon.

[0098] In a further embodiment of the present invention, the comparison process can also take into account region-specificity. Different regions may have different lighting requirements and equipment configurations, so energy consumption thresholds can be dynamically adjusted based on region type (e.g., office area, corridor, conference room). For example, the energy consumption threshold for a conference room may be higher than that for an office area because conference rooms have higher lighting requirements. In this embodiment of the present invention, energy consumption thresholds for each region are stored in a configuration table, enabling dynamic loading and updating.

[0099] Step 24: If the difference value exceeds the energy consumption threshold, outputting a judgment result that a local over-lighting phenomenon exists; otherwise, outputting a judgment result that no local over-lighting phenomenon exists;

[0100] Furthermore, the step of dynamically optimizing the brightness and color temperature data in the device data set using a reinforcement learning algorithm to generate an optimized lighting control parameter set includes the following steps:

[0101] Step 311: extracting a brightness dataset and a color temperature dataset from the device dataset as input data for a reinforcement learning algorithm;

[0102] Specifically, the brightness and color temperature datasets are extracted by first filtering out brightness and color temperature records from the device dataset for a recent period (e.g., the past 24 hours) based on device ID and timestamp. These records are then grouped according to the spatial location of the lighting devices to form a regional brightness and color temperature data matrix. Each row of this data matrix represents a point in time, and each column represents a lighting device. The matrix elements are the brightness or color temperature values ​​for the corresponding time and device, respectively.

[0103] It is understandable that the brightness and color temperature data matrix can intuitively reflect the spatiotemporal distribution characteristics of lighting parameters, facilitating state representation and feature extraction in subsequent reinforcement learning algorithms. In practical applications, in order to reduce the impact of data noise, the raw data can also be preprocessed, such as outlier detection and processing, data smoothing, etc.

[0104] Step 312: Pre-construct a deep Q-network-based reinforcement learning environment model suitable for parameter control of lighting equipment, defining a state space, an action space, and a reward function;

[0105] Specifically, in an embodiment of the present invention, the reinforcement learning environment model is a mathematical abstraction of a lighting system, and is used to simulate the process and results of lighting control decisions.

[0106] The state space in this embodiment is defined as a combination vector of the brightness distribution, color temperature distribution, and environmental factors such as natural light illumination and human activities in the current lighting area.

[0107] The action space in this embodiment is defined as a set of instructions for brightness adjustment instructions and color temperature adjustment instructions for each lighting device, including increasing, decreasing or maintaining the current brightness / color temperature value.

[0108] The reward function in this embodiment is designed as a weighted combination of energy consumption reduction and lighting quality improvement, for example, R = -w1 E1+w2 Q1; where R represents the reward value, E1 represents the change in energy consumption, Q1 represents the change in lighting quality, and w1 and w2 are the weight coefficients of energy consumption and lighting quality, respectively.

[0109] The lighting quality is comprehensively evaluated through indicators such as illumination uniformity and color temperature suitability.

[0110] In an embodiment of the present invention, the deep Q network adopts a multi-layer perceptron structure, where the input layer corresponds to the state vector and the output layer corresponds to the Q value estimate of each possible action. The hidden layer of the network uses the ReLU activation function to enhance the nonlinear expression ability of the model;

[0111] Step 313: Collecting device data sets and optimization parameters in the reward function within a preset time period as historical lighting data, and pre-training the deep Q network based on the historical lighting data to form an initial policy model;

[0112] In an embodiment of the present invention, the preset time period can be set to three months before the current time, and the optimization parameters are set to the energy consumption change and the lighting quality change;

[0113] The pre-training process is to use historical lighting control data to conduct preliminary training on the deep Q network before formally deploying the reinforcement learning algorithm to accelerate the subsequent online learning process;

[0114] Specifically, the method of forming the initial strategy model is:

[0115] Pre-training adopts supervised learning, taking the state-action pairs in historical data as training samples and the actual rewards as target values, and optimizing the network parameters by minimizing the mean square error between the predicted Q value and the target Q value.

[0116] The pre-training process is divided into three stages: data preparation, model training and verification.

[0117] In the data preparation stage, valid state-action-reward sequences are extracted from the historical database and standardized.

[0118] During the model training phase, the small batch gradient descent method is used to optimize the network parameters, and the learning rate adopts an adaptive adjustment strategy to balance the convergence speed and stability.

[0119] In the validation phase, an independent test dataset is used to evaluate the model performance to ensure that the pre-trained model has a certain generalization ability.

[0120] Through pre-training, the deep Q network can initially grasp the basic laws of lighting control;

[0121] Step 314: Deploy the initial policy model on the edge computing node, perform online reinforcement learning and policy update, and generate online learning results;

[0122] Specifically, generating online learning results includes the following steps:

[0123] First, the Deep Q Network enters the exploration phase, employing an ε-greedy strategy. That is, during each control cycle (e.g., 5 minutes), the edge computing node first obtains information about the current lighting environment, including brightness, color temperature, and energy consumption data for each lighting device. This information is then fed into the Deep Q Network, which randomly selects actions for exploration with a probability of ε, and selects the action with the highest current Q value with a probability of 1-ε.

[0124] The value of ε is initially set high, for example, 0.3, and gradually decreases as the learning process progresses, eventually stabilizing at a smaller value, for example, 0.05, to maintain adaptability to environmental changes. After each action, the edge computing node records the new state and reward obtained, forming a transfer sample and storing it in the experience replay buffer.

[0125] When the buffer accumulates a preset number of samples, the system randomly extracts small batches of samples from it to update the Q network parameters. The parameter update uses the temporal difference learning method to regularly synchronize the updated model parameters to the model structure of the deep Q network.

[0126] The final action selected by each ε-greedy strategy is the result of online learning;

[0127] Step 315: Dynamically generate brightness and color temperature control parameters of the lighting device based on the online learning results;

[0128] Specifically, by extracting the action selected by the deep Q network from the online learning results, an instruction set is extracted from the selected action, and the adjustment values ​​of brightness and color temperature in the brightness adjustment instruction and color temperature adjustment instruction in the instruction set are the brightness and color temperature control parameters.

[0129] It is understandable that reinforcement learning continuously learns the optimal strategy by interacting with the environment, can respond to environmental changes in real time, and is suitable for handling dynamically changing lighting needs. Because overlighting is a clear problem state, reinforcement learning can quickly find appropriate strategies to reduce overlighting through a reward and punishment mechanism.

[0130] Further, such as Figure 3 As shown, the use of a multi-objective genetic algorithm to perform dual-objective optimization on the energy consumption and comfort data in the device data set to generate an optimized lighting control parameter set includes the following steps:

[0131] Step 321: extracting energy consumption data and comfort-related data from the equipment data set to construct an input data set for a dual-objective optimization model;

[0132] Specifically, the input data set of the dual-objective optimization model is constructed as follows:

[0133] The energy consumption data of all lighting devices in the current time period are filtered out from the device data set, including power, cumulative energy consumption and energy efficiency indicators.

[0134] Extract data related to comfort, including brightness data, color temperature data, and environmental adaptability scores.

[0135] The above energy consumption data and comfort-related data are integrated according to device ID and timestamp to form the input data set of the structured dual-objective optimization model;

[0136] Step 322: Based on the input data set, define an energy consumption objective function and a comfort objective function, and construct a dual-objective optimization model;

[0137] Specifically, the energy consumption objective function is defined as follows: minimizing the total energy consumption of the lighting system is the optimization goal, and the mathematical expression is: ; where x is the decision variable vector corresponding to the input data set, including the brightness and color temperature setting values ​​of each lighting device, represents the energy consumption value of the e-th lighting device, n represents the total number of lighting devices, Represents the energy consumption objective function, that is, the normalized value of the total energy consumption of the system under a given control vector x. The smaller the better, represents the actual energy consumption value of the e-th lighting device under the control vector x. The energy consumption value can be a measured value or a theoretical value estimated based on device parameters and operating status, such as a theoretical value obtained by fitting a polynomial regression model;

[0138] In practical applications of the present invention, the energy consumption objective function may further consider time factors, such as the difference in peak and valley electricity prices. By introducing a time weight coefficient, it is encouraged to increase the lighting intensity during low electricity price periods and reduce the lighting intensity during peak electricity price periods, thereby further reducing the operating cost of the lighting system. The present invention is not limited here.

[0139] The definition of the comfort objective function is: maximizing the user's visual comfort is the optimization goal, and the mathematical expression is: ;in, It represents the overall visual comfort index of the system under a given control vector x. represents the comprehensive score of the user's visual experience for the e-th lighting device under the control vector x. The comfort score is a comprehensive indicator composed of three weighted sub-indicators: brightness suitability, color temperature suitability, and light environment uniformity. Brightness suitability assesses whether the lighting brightness meets the needs of a specific scene, color temperature suitability assesses whether the color temperature matches the environment and activity type, and light environment uniformity assesses the uniformity of the lighting distribution. The weights of these three sub-indicators can be dynamically adjusted according to the needs of different scenarios. For example, in an office environment, the weights of brightness suitability and light environment uniformity are higher, while in a leisure environment, the weight of color temperature suitability may be higher.

[0140] The dual-objective optimization model is constructed by combining the energy consumption objective function and the comfort objective function into a dual-objective optimization problem, i.e., simultaneously minimizing energy consumption and maximizing comfort. Since these two objectives often conflict—for example, reducing energy consumption may result in decreased comfort—it is necessary to find an appropriate balance between energy consumption and comfort, namely, a Pareto optimal solution set.

[0141] The mathematical expression of the dual-objective optimization model is: ;in, Represents a set of constraints, including brightness range constraints, color temperature range constraints, and lighting standard constraints for specific scenes. represents transpose, Represents the vector objective function in the dual-objective optimization model, which is used to combine two separate evaluation indicators and Combined into an overall optimization goal in a mathematical sense;

[0142] Step 323: Initialize the population of the multi-objective genetic algorithm for the dual-objective optimization model, set the algorithm parameters, and encode the individuals in the population;

[0143] Specifically, the population is initialized in the following manner: first, the population size is determined, which is set to 50-200 individuals in the embodiment of the present invention to balance computational efficiency and solution diversity.

[0144] A set of lighting control parameters is randomly generated for each individual, including the brightness and color temperature values ​​of each lighting device. This random generation process adheres to constraints to ensure the feasibility of the initial solution. For example, the brightness range is 0-100%, and the color temperature range is 2700K-6500K.

[0145] In a further embodiment, some individuals may be generated using heuristic rules, such as those based on small disturbances of the current lighting state or based on references to historical optimization results.

[0146] The algorithm parameters include: crossover probability (generally set to 0.7-0.9), mutation probability (generally set to 0.01-0.1), maximum number of iterations (generally set to 50-200 generations), and the size limit of the Pareto solution set. The specific values ​​of these algorithm parameters can be adjusted based on the problem scale and optimization effect. For example, a larger population size and more iterations may be required for complex, large-scale lighting systems; while for real-time control scenarios, the number of iterations may need to be limited to ensure the algorithm's responsiveness.

[0147] The population individuals are encoded using real number encoding, with each individual represented as a real vector whose dimensionality equals twice the number of lighting devices (each device corresponds to a brightness value and a color temperature value). For example, for a system with 10 lighting devices, each individual is a 20-dimensional real vector, where the first 10 elements represent the brightness value of each device, and the last 10 elements represent the color temperature value of each device. Compared to binary encoding, real number encoding is more suitable for optimization problems involving continuous variables, can more accurately represent lighting control parameter sets, and simplifies the implementation of genetic operations.

[0148] Step 324: Iteratively perform selection, crossover, and mutation operations on the initialized population to generate a new population, and perform non-dominated sorting and crowding calculation;

[0149] Specifically, the selection operation is performed using a selection strategy based on non-dominated sorting and crowding, namely the elite selection mechanism in NSGA-II (Non-dominated Sorting Genetic Algorithm II). First, the parent and child populations are merged. The merged population is then hierarchically sorted based on non-domination relationships to form multiple non-dominated fronts. The non-domination relationship states that if individual A is not inferior to individual B on any objective and is superior to individual B on at least one objective, then individual A dominates individual B. Individuals not dominated by any other individuals form the first non-dominated front. After removing these individuals from the first front, the remaining non-dominated individuals form the second non-dominated front, and so on. When selecting the next generation, individuals with higher non-dominated hierarchies are prioritized. If not all individuals from a given hierarchical level make it to the next generation, individuals with higher crowding (i.e., more isolated individuals in the objective space) are selected based on crowding to maintain population diversity.

[0150] The crossover operation is performed by randomly selecting two parent individuals from the current population and determining whether to perform a crossover based on the crossover probability. If crossover is performed, the simulated binary crossover (SBX) method is used to generate two offspring individuals. This simulated binary crossover method adjusts the degree of similarity between the offspring and the parent by controlling the distribution exponent parameter, thereby maintaining the excellent characteristics of the parent while generating new search directions. Specifically, for each pair of corresponding genes in the parent individuals (i.e., the brightness or color temperature values ​​of the same lighting device), a crossover factor is generated based on the probability distribution function. The offspring gene value is then calculated based on the crossover factor. For example, if the parent gene values ​​are x1 and x2, respectively, and the crossover factor is β, the offspring gene values ​​are 0.5×[(1+β)×x1+(1-β)×x2] and 0.5×[(1-β)×x1+(1+β)×x2].

[0151] The mutation operation is performed as follows: for each offspring individual, the decision is made based on the mutation probability whether to mutate each gene within it. If mutation is performed, a polynomial mutation method is used to generate perturbations near the current gene value to generate a new gene value. By controlling the distribution exponent parameter, the polynomial mutation method can adjust the intensity of the mutation, enabling both local search and maintaining a certain level of global exploration capability. Specifically, for a gene value xf, its mutated value is xf + δ, where δ is the perturbation generated according to the probability distribution function, and its magnitude is related to the gene's value range and the distribution exponent parameter. For example, if the gene's value range is [a, b], the maximum value of δ is min(xf - a, b - xf), ensuring that the mutated gene value remains within the valid range.

[0152] The non-dominated sort is calculated as follows: for each individual in the population, the set of individuals it dominates and the number of times it is dominated are calculated. Initially, individuals with a domination count of 0 constitute the first non-dominated frontier. Then, for each individual in the first frontier, the domination count of the individuals it dominates is reduced by 1. If the domination count becomes 0 after deduction, the individual enters the second non-dominated frontier. This process continues in this way until all individuals are assigned to a non-dominated frontier. The time complexity of the non-dominated sort is O(MN²), where M is the number of objective functions and N is the population size. To improve algorithm efficiency, a fast non-dominated sort algorithm can be used in practical implementations to reduce the time complexity to O(MN×logN) by optimizing the data structure and computational process.

[0153] The crowding degree is calculated by calculating the crowding degree of individuals in the same non-dominated front in the target space as a measure of diversity. Specifically, for each objective function, individuals are first sorted according to the objective function value. Then, the differences in the objective function values ​​between adjacent individuals are calculated and normalized before summing these differences to obtain a crowding degree index. A larger crowding degree index indicates a more isolated individual in the target space and a higher diversity. During the selection operation, when individuals need to be selected from the same non-dominated front, individuals with higher crowding degrees are prioritized to maintain the diversity and uniform distribution of the solution set.

[0154] Step 325: Evaluate the fitness of the individuals in the population and determine whether the algorithm has reached the termination condition. If so, output the Pareto optimal solution set; otherwise, return to step 324 to continue iteration.

[0155] Specifically, the fitness assessment method is as follows: for each individual, based on the lighting control parameter set represented by its encoding, the energy consumption objective function value and the comfort objective function value are calculated. The energy consumption objective function value is calculated based on the power model and operating time of the lighting equipment, while the comfort objective function value is calculated based on indicators such as brightness suitability, color temperature suitability, and light environment uniformity. During the calculation process, historical data in the device dataset can be used to establish a regression model to predict energy consumption and comfort performance under different control parameters, thereby improving the accuracy and efficiency of the assessment. For example, for energy consumption prediction, a brightness-power mapping model can be established; for comfort prediction, a comfort scoring model based on user feedback can be established.

[0156] The termination condition is determined by: the algorithm reaches a preset maximum number of iterations, the improvement in the Pareto front over multiple generations falls below a preset threshold, or the algorithm runtime exceeds a preset limit. The improvement in the Pareto front can be measured using metrics such as hypervolume or coverage, which comprehensively assess the quality and diversity of Pareto solutions. For example, the hypervolume metric calculates the volume of the region dominated by the Pareto solution in the target space; a larger metric indicates a higher-quality solution. The coverage metric calculates the proportion of dominance of the current generation's solution relative to the previous generation's solution; a stable metric indicates that the algorithm has essentially converged.

[0157] The Pareto optimal solution set is output by extracting all individuals on the first non-dominated front from the final population as the Pareto optimal solution set for the multi-objective optimization problem. Each solution in the Pareto optimal solution set represents a set of lighting control parameters that achieve a different balance between energy consumption and comfort. To facilitate subsequent decision-making, the Pareto optimal solution set can be post-processed, such as calculating a comprehensive score for each solution or selecting the most appropriate solution based on the current system state and user preferences. The comprehensive score U can be calculated using a weighted sum method: U = w3 × f1 + w4 × f2, where f1 and f2 are the normalized values ​​of the energy consumption and comfort objective functions, respectively, and w3 and w4 are corresponding weight coefficients, satisfying w3 + w4 = 1. The weight coefficients can be dynamically adjusted based on system operation policies. For example, the weight of energy consumption objectives can be increased during periods of energy shortage, while the weight of comfort objectives can be increased during important meetings or events.

[0158] Step 326: Selecting an appropriate lighting control parameter set solution from the Pareto optimal solution set according to the current system state and user preferences;

[0159] Specifically, the appropriate lighting control parameter set solution is selected as follows: First, based on the current system operation strategy, the weight coefficients of energy consumption and comfort targets are determined. This system operation strategy can be dynamically generated based on multiple dimensions of information, including time factors (e.g., working hours vs. non-working hours), spatial factors (e.g., important areas vs. general areas), activity factors (e.g., meetings vs. daily office work), and energy factors (e.g., peak power consumption vs. off-peak power consumption). For example, in important meeting areas during working hours, the comfort target might be weighted as high as 0.8; while in general areas during non-working hours, the energy consumption target might be weighted as high as 0.9.

[0160] Secondly, the comprehensive score of each solution in the Pareto optimal solution set is calculated, and the solution with the highest score is selected as the appropriate solution. In the process of normalizing the energy consumption objective function value and the comfort objective function value, the minimum-maximum normalization method is used to map the objective function values ​​to the interval [0,1] to eliminate the influence of different objective function dimensions.

[0161] The selected solution is then feasibility verified and fine-tuned. Feasibility verification involves checking whether the control parameters meet all constraints and are compatible with the current system state. Fine-tuning involves making subtle adjustments to the control parameters based on the actual system response characteristics to ensure a smooth transition and stable operation. For example, to prevent sudden changes in lighting brightness from causing discomfort to users, a brightness change rate limit can be set to ensure gradual brightness adjustment.

[0162] Step 327: Convert the appropriate lighting control parameter set solution into standardized lighting control instructions to generate an optimized lighting control parameter set;

[0163] Specifically, the standardized lighting control instructions are generated by first grouping the brightness and color temperature values ​​in the appropriate solution according to the lighting device ID to form a device-level control parameter set. This control parameter set includes parameters such as the target brightness value, target color temperature value, and transition time for each lighting device. The transition time defines the duration of the gradual transition from the current state to the target state and is generally set to 1-5 seconds to ensure smooth lighting changes.

[0164] It is understandable that energy consumption and comfort are two conflicting objectives. Therefore, a multi-objective genetic algorithm needs to be selected to find a suitable balance between the two conflicting objectives.

[0165] Furthermore, the method of dynamically adjusting the light environment of the lighting device is:

[0166] The edge computing node receives the optimized lighting control parameter set from the control parameter generation module, converts the lighting control parameter set into a lighting device-specific protocol instruction according to the communication protocol type of the lighting device, and sends the protocol instruction to the corresponding lighting device through the multi-protocol communication interface. The S-type function is used to calculate the gradient curve of brightness and color temperature from the current value to the target brightness value and target color temperature value to dynamically control the lighting device; the S-type function is, for example, the sigmoid function.

[0167] In addition, the parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0168] The above-described specific embodiments further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is merely a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

[0169] The above preset parameters or preset thresholds are all set by those skilled in the art according to actual conditions or obtained through large amounts of data simulation.

[0170] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A cloud platform-based LED intelligent lighting management system, characterized in that: It includes a multi-source data collection module, an over-lighting judgment module, a control parameter generation module, and a dynamic adjustment module; wherein each module is electrically connected; The multi-source data collection module collects lighting device data from multi-protocol terminals through edge computing nodes. The lighting device data includes brightness data sets, color temperature data sets, and energy consumption data sets, which constitute the device data sets of the multi-protocol terminals. The device data sets are then sent to the lighting judgment module and the control parameter generation module. an over-lighting judgment module, which judges whether there is a local over-lighting phenomenon in the current lighting area based on the energy consumption data set in the device data set, generates a judgment result, and sends the judgment result to the control parameter generation module; a control parameter generation module, determining whether the judgment result indicates over-lighting; if so, dynamically optimizing the brightness and color temperature data in the device data set using a reinforcement learning algorithm to generate an optimized lighting control parameter set; if not, performing dual-objective optimization on the energy consumption and comfort data in the device data set using a multi-objective genetic algorithm to generate an optimized lighting control parameter set; and sending the optimized lighting control parameter set to the dynamic adjustment module; The dynamic adjustment module sends the optimized lighting control parameter set to the corresponding lighting equipment to dynamically adjust the light environment of the lighting equipment.

2. The cloud platform-based LED intelligent lighting management system according to claim 1, characterized in that: The brightness dataset is collected as follows: The edge computing node automatically identifies the type of communication protocol used by the connected lighting device through the protocol identification interface; According to the identified protocol type, the edge computing node dynamically loads the corresponding protocol driver module to establish a communication connection with the lighting device; after the communication connection is established, the edge computing node sends a brightness data query instruction to the lighting device; After receiving the brightness data query instruction, the lighting device returns the current brightness status data to the edge computing node; Add standardized information to each piece of brightness status data to form brightness data; The time series of all brightness data sorted in chronological order is taken as the brightness dataset.

3. The cloud platform-based LED intelligent lighting management system according to claim 2, characterized in that: The color temperature dataset is collected as follows: The edge computing node sends a color temperature data query instruction to the LED lighting device based on the established multi-protocol communication connection; After receiving the color temperature data query command, the lighting device returns the current color temperature status data to the edge computing node; Adding standardized information to the color temperature status data to form color temperature data; The time series of all color temperature data sorted in chronological order is taken as the color temperature dataset.

4. The cloud platform-based LED intelligent lighting management system according to claim 3, characterized in that: The standardized information includes lighting device ID, location information and a timestamp.

5. The cloud platform-based LED intelligent lighting management system according to claim 4, characterized in that: The energy consumption dataset is collected as follows: Different collection strategies are used to collect energy consumption data of lighting equipment based on its energy consumption data acquisition capabilities. For lighting equipment with direct energy consumption measurement capabilities, edge computing nodes directly query energy consumption data including the equipment's real-time power, cumulative energy consumption, and energy efficiency indicators through the corresponding protocol. For lighting equipment without direct energy consumption measurement capabilities, edge computing nodes estimate their energy consumption through indirect methods. The time series of all energy consumption data sorted in chronological order is taken as the energy consumption dataset.

6. The cloud platform-based LED intelligent lighting management system according to claim 5, characterized in that: The indirect method uses an energy consumption estimation model to calculate theoretical energy consumption values ​​based on the rated power, current brightness level and operating time of the device.

7. The cloud platform-based LED intelligent lighting management system according to claim 6, characterized in that: The device data set is composed as follows: The brightness dataset, color temperature dataset, and energy consumption dataset are classified and integrated according to the device ID of the lighting device to generate a device dataset; the device dataset is stored in the form of a structured data table, which adopts a relational data structure, with each row representing a complete data record of a device corresponding to a time point, and each column representing a data field.

8. The cloud platform-based LED intelligent lighting management system according to claim 7, characterized in that: The determining whether there is a local over-lighting phenomenon in the current lighting area based on the energy consumption data set in the device data set includes the following steps: Step 21: extracting the energy consumption dataset from the device dataset of the multi-protocol terminal as energy consumption distribution data of the current lighting area; Step 22: Calculate the difference between the average energy consumption and the peak energy consumption of the current lighting area based on the energy consumption distribution data; Step 23: Compare the difference value with a preset energy consumption threshold to determine whether there is a local over-lighting phenomenon in the current lighting area; Step 24: If the difference value exceeds the energy consumption threshold, outputting a judgment result that there is a local over-lighting phenomenon; otherwise, outputting a judgment result that there is no local over-lighting phenomenon.

9. The cloud platform-based LED intelligent lighting management system according to claim 8, characterized in that: The step of dynamically optimizing the brightness and color temperature data in the device data set using a reinforcement learning algorithm to generate an optimized lighting control parameter set comprises the following steps: Step 311: extracting a brightness dataset and a color temperature dataset from the device dataset as input data for a reinforcement learning algorithm; Step 312: Pre-construct a deep Q-network-based reinforcement learning environment model suitable for parameter control of lighting equipment, defining a state space, an action space, and a reward function; Step 313: Collecting device data sets and optimization parameters in the reward function within a preset time period as historical lighting data, and pre-training the deep Q network based on the historical lighting data to form an initial policy model; Step 314: Deploy the initial policy model on the edge computing node, perform online reinforcement learning and policy update, and generate online learning results; Step 315: Based on the online learning results, dynamically generate brightness and color temperature control parameters of the lighting device.

10. The cloud platform-based LED intelligent lighting management system according to claim 9, characterized in that: The deep Q network adopts a multi-layer perceptron structure.

11. The cloud platform-based LED intelligent lighting management system according to claim 10, characterized in that: The method of using a multi-objective genetic algorithm to perform dual-objective optimization on the energy consumption and comfort data in the device data set to generate an optimized lighting control parameter set includes the following steps: Step 321: extracting energy consumption data and comfort-related data from the equipment data set to construct an input data set for a dual-objective optimization model; Step 322: Based on the input data set, define an energy consumption objective function and a comfort objective function, and construct a dual-objective optimization model; Step 323: Initialize the population of the multi-objective genetic algorithm for the dual-objective optimization model, set the algorithm parameters, and encode the individuals in the population; Step 324: Iteratively perform selection, crossover, and mutation operations on the initialized population to generate a new population, and perform non-dominated sorting and crowding calculation; Step 325: Evaluate the fitness of the individuals in the population and determine whether the algorithm has reached the termination condition. If so, output the Pareto optimal solution set; otherwise, return to step 324 to continue iteration. Step 326: Select a lighting control parameter set solution from the Pareto optimal solution set according to the current system state and user preference; Step 327: Convert the lighting control parameter set solution into standardized lighting control instructions to generate an optimized lighting control parameter set.

12. The cloud platform-based LED intelligent lighting management system according to claim 11, characterized in that: The method of dynamically adjusting the light environment of the lighting device is: The edge computing node receives the optimized lighting control parameter set from the control parameter generation module, converts the lighting control parameter set into the corresponding protocol instructions of the lighting device according to the communication protocol type of the lighting device, and sends the protocol instructions to the corresponding lighting device through the multi-protocol communication interface. The S-type function is used to calculate the gradient curve of brightness and color temperature from the current value to the target brightness value and target color temperature value to dynamically control the lighting device.

Images